DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claims 1-23 are the currently pending claims hereby under examination.
Claim Interpretations
Claims 1, 2, and 17 recite an electronic hardware processor “configured to” execute stored instructions to determine that an intervention for hypotension has been implemented and determine a likelihood of effectiveness of the intervention. Under the broadest reasonable interpretation consistent with the specification, the recited processor and stored instructions require the system, upon execution of the instructions, to make the recited determinations from the claimed hemodynamic information. The claims do not additionally require the claimed system itself to administer the intervention, obtain post-intervention hemodynamic data, or compare pre-intervention and post-intervention measurements. The “likelihood of effectiveness” is interpreted as a predictive determination of expected responsiveness or effectiveness of the intervention.
Claims 1, 4, 13, 16, and 22 recite reducing overfitting of the reference feature pools by “removing one or more of the reference feature pools from remaining reference feature pools” and normalizing selected reference maximum features. Under the broadest reasonable interpretation consistent with the specification, the removal language is interpreted as excluding selected pooled feature contributions from further processing while retaining the other pooled contributions. This interpretation of the removal operation is consistent with Fig. 5 and ¶[0081]-[0083], which describe max pooling and dropout that temporarily zeros selected neurons during training. The listed order of the claim clauses does not, by itself, require the dropout/removal operation to occur before normalization. Although ¶[0084] describes an embodiment in which normalization layer 512 standardizes the outputs from dropout layer 510, that disclosed sequence does not impose a sequence on the claimed apparatus absent claim language requiring that relationship. The selected reference maximum features, however, must exist before those selected maxima can be normalized.
Claims 2, 11, and 17 recite “determine, based on the reference feature pools, normalized test maximum features from the plurality of test heart health parameters.” Under the broadest reasonable interpretation consistent with the specification, “based on the reference feature pools” requires a dependency between the reference feature pools and the determination of the normalized test maximum features, but does not require the reference feature pools themselves to be supplied as direct contemporaneous inputs to the test-feature determination. The limitation encompasses determining the normalized test maximum features using processing parameters learned or derived from the reference feature pools. This interpretation is also consistent with the Instant Application's description that dropout may be performed during training while, during inference of test data, neurons need not be dropped and their weights may instead be scaled (¶[0082]). That disclosure supports a distinction between training-stage feature processing and later test-data inference using processing established by training.
Claim 17 recites determining an HPI “based on” both the likelihood that the patient will experience hypotension and the likelihood of effectiveness of the intervention. The Examiner interprets this limitation as requiring the HPI to depend on both recited likelihood determinations. The claim does not require a particular mathematical formula, relative weighting, or specific proprietary implementation for combining the two likelihoods.
Claim Objections
Claims 1, 2, 9, 11, 15, 16, 20, and 23 are objected to because of the following informalities:
In claim 1, lines 14, 57, and 59, and claim 2, lines 12, 27, and 29: "the patient" lacks explicit antecedent basis, although based on the prior recitation of "one or more patients" and the test-signal context, the Examiner understands "the patient" to refer to the patient from whom the test analog hemodynamic sensor signal is received;
In claim 1, line 39, and claim 16, line 29: "the filtered sets of reference heart health parameters" lacks explicit antecedent basis, although based on the prior recitation of "the reference sets of heart health parameters" and the operation of applying a filter to those sets, the Examiner understands "the filtered sets of reference heart health parameters" to refer to those reference sets after filtering, and Applicant is requested to provide proper antecedent basis;
In claim 1, line 32, and claim 20, line 14: “diastolic peak” is inconsistent with the Instant Application's identification at ¶[0067] of the corresponding waveform extremum as the “diastolic minimum indicator 230”, which corresponds to minimum diastolic pressure, and its definition of pulmonary pulse pressure 232 as the difference between the minimum diastolic pressure and maximum systolic pressure; the Examiner therefore understands “diastolic peak” to refer to the diastolic pressure extremum/minimum, and Applicant is requested to amend the terminology for consistency; and
In claim 9, line 3: "that that the patient" appears to contain a typographical error and should be revised to "that the patient"; claim 11, line 30, claim 15, line 3, claim 16, line 45, and claim 23, line 3 have this same issue.
Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1, 4, 13, 16, and 22 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as failing to set forth the subject matter which the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the applicant regards as the invention.
Claim 1 recites "generate, based on the plurality of heart health parameters, one or more feature pools" in lines 50-51. Claim 4 recites "generating, based on the plurality of heart health parameters, one or more feature pools" in lines 8-9. Claim 13 recites "generating, based on the plurality of heart health parameters, one or more feature pools" in lines 8-9. Claim 16 recites "generate, based on the plurality of heart health parameters, one or more feature pools" in lines 37-38. Claim 22 recites "generating, based on the plurality of heart health parameters, one or more feature pools" in lines 8-9. Each of these claims or its incorporated parent claim separately identifies test heart health parameters and reference heart health parameters, but the claims do not identify which group is meant by "the plurality of heart health parameters". It is therefore unclear whether the feature pools are generated from the test heart health parameters, the reference heart health parameters, or another plurality. The Instant Application uses the unmodified phrase "the plurality of heart health parameters" in both contexts. Paragraph [0112] uses the phrase in describing test features used to determine a time to event and/or likelihood of effectiveness of an intervention, while ¶[0120], block 878, uses the same phrase in the reference-feature-pool flow after normalization of selected reference maximum features. Paragraph [0114] likewise repeats the disputed language. Thus, the disclosure itself does not establish whether the generic plurality in claim 1 refers to the test heart health parameters or the reference heart health parameters. For purposes of examination, the Examiner interprets "the plurality of heart health parameters" to mean the plurality of test heart health parameters because the claim immediately proceeds to determining normalized test maximum features from the test heart health parameters, and ¶[0112] supports that reading. However, the reference-feature usage at ¶[0120] leaves a second reasonable interpretation. Accordingly, claims 1, 4, 13, 16, and 22 are indefinite under 35 U.S.C. § 112(b).
The following is a quotation of 35 U.S.C. 112(d):
(d) REFERENCE IN DEPENDENT FORMS.—Subject to subsection (e), a claim in dependent form shall contain a reference to a claim previously set forth and then specify a further limitation of the subject matter claimed. A claim in dependent form shall be construed to incorporate by reference all the limitations of the claim to which it refers.
The following is a quotation of pre-AIA 35 U.S.C. 112, fourth paragraph:
Subject to the following paragraph [i.e., the fifth paragraph of pre-AIA 35 U.S.C. 112], a claim in dependent form shall contain a reference to a claim previously set forth and then specify a further limitation of the subject matter claimed. A claim in dependent form shall be construed to incorporate by reference all the limitations of the claim to which it refers.
Claim 6 is rejected under 35 U.S.C. 112(d) or pre-AIA 35 U.S.C. 112, 4th paragraph, as being of improper dependent form for failing to further limit the subject matter of the claim upon which it depends, or for failing to include all the limitations of the claim upon which it depends.
Claim 6 depends from claim 5, which depends from claim 2 and therefore incorporates all limitations of claim 2. Claim 2 recites "generate, based on the determined likelihood of effectiveness of the intervention for hypotension of the patient, data for displaying an alert indicating the likelihood of effectiveness of the intervention for hypotension" in lines 28-30. Claim 6 recites "generate, based on the determined likelihood of effectiveness of the intervention, data for displaying at least the alert indicating the likelihood of effectiveness of the intervention for hypotension" in lines 2-5. Claim 6 does not further limit the inherited requirement because claim 2 already requires the electronic hardware processor to execute instructions causing the system to generate data for displaying the alert based on the determined likelihood of intervention effectiveness. The phrase "at least the alert" does not further limit that inherited requirement, and the omission in claim 6 of language already required by claim 2 does not remove the inherited limitation. Accordingly, claim 6 fails to further limit the subject matter of claim 5 and does not satisfy 35 U.S.C. § 112(d).
Double Patenting
The nonstatutory double patenting rejection is based on a judicially created doctrine grounded in public policy (a policy reflected in the statute) so as to prevent the unjustified or improper timewise extension of the “right to exclude” granted by a patent and to prevent possible harassment by multiple assignees. A nonstatutory double patenting rejection is appropriate where the conflicting claims are not identical, but at least one examined application claim is not patentably distinct from the reference claim(s) because the examined application claim is either anticipated by, or would have been obvious over, the reference claim(s). See, e.g., In re Berg, 140 F.3d 1428, 46 USPQ2d 1226 (Fed. Cir. 1998); In re Goodman, 11 F.3d 1046, 29 USPQ2d 2010 (Fed. Cir. 1993); In re Longi, 759 F.2d 887, 225 USPQ 645 (Fed. Cir. 1985); In re Van Ornum, 686 F.2d 937, 214 USPQ 761 (CCPA 1982); In re Vogel, 422 F.2d 438, 164 USPQ 619 (CCPA 1970); In re Thorington, 418 F.2d 528, 163 USPQ 644 (CCPA 1969).
A timely filed terminal disclaimer in compliance with 37 CFR 1.321(c) or 1.321(d) may be used to overcome an actual or provisional rejection based on nonstatutory double patenting provided the reference application or patent either is shown to be commonly owned with the examined application, or claims an invention made as a result of activities undertaken within the scope of a joint research agreement. See MPEP § 717.02 for applications subject to examination under the first inventor to file provisions of the AIA as explained in MPEP § 2159. See MPEP § 2146 et seq. for applications not subject to examination under the first inventor to file provisions of the AIA . A terminal disclaimer must be signed in compliance with 37 CFR 1.321(b).
The filing of a terminal disclaimer by itself is not a complete reply to a nonstatutory double patenting (NSDP) rejection. A complete reply requires that the terminal disclaimer be accompanied by a reply requesting reconsideration of the prior Office action. Even where the NSDP rejection is provisional the reply must be complete. See MPEP § 804, subsection I.B.1. For a reply to a non-final Office action, see 37 CFR 1.111(a). For a reply to final Office action, see 37 CFR 1.113(c). A request for reconsideration while not provided for in 37 CFR 1.113(c) may be filed after final for consideration. See MPEP §§ 706.07(e) and 714.13.
The USPTO Internet website contains terminal disclaimer forms which may be used. Please visit www.uspto.gov/patent/patents-forms. The actual filing date of the application in which the form is filed determines what form (e.g., PTO/SB/25, PTO/SB/26, PTO/AIA /25, or PTO/AIA /26) should be used. A web-based eTerminal Disclaimer may be filled out completely online using web-screens. An eTerminal Disclaimer that meets all requirements is auto-processed and approved immediately upon submission. For more information about eTerminal Disclaimers, refer to www.uspto.gov/patents/apply/applying-online/eterminal-disclaimer.
Claim 1 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 1 of Smith et al. (U.S. Patent No. 12,672,822), hereinafter Reference or Ref, in view of Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes, Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, Al Hatib et al. (US 2018/0028076 A1), hereinafter Hatib'076, Al Hatib et al. (US 2018/0025290 A1), hereinafter Hatib'290, and Johnson et al. (US 2021/0275783 A1), hereinafter Johnson.
The analysis is as follows (please note, the bolded and underlined portions of the entries under the instant application (IA) are those portions of the IA claims that the Reference claim does not have or are different from in some form):
Instant Application (IA) claim language(bold + underlined = not in / different from Reference claim)
Reference - U.S. Patent No. 12,672,822(corresponding recited claim language)
Analysis (comparison and secondary prior art)
Claim 1: A hemodynamic sensor system configured to determine a likelihood of effectiveness of an intervention for hypotension, the system comprising:
Ref claim 1: “A system for monitoring arterial pressure of a patient and determining one or more scores that are predictive of responsiveness of the patient to a corresponding therapy”
Ref claim 1 already claims prospective scores predictive of patient responsiveness to corresponding therapies. Potes expressly identifies both fluid-responsiveness prediction and a hypotensive state as applications of the same arterial-pressure physiological-state model (Potes, ¶[0038]); applying the therapy-responsiveness determination in the hypotension context provides the claimed intervention-for-hypotension use. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 by applying Potes's known fluid-responsiveness prediction in the hypotension context, thereby providing the claimed intervention-for-hypotension use.
Claim 1: a hemodynamic sensor that produces analog hemodynamic sensor signals representative of arterial pressure signal waveforms of one or more patients;
Ref claim 1: “a hemodynamic sensor that produces sensed hemodynamic data representative of an arterial pressure waveform of the patient”
Ref claim 1 recites arterial-pressure sensing but not the analog implementation. Potes teaches analog arterial-pressure sensor data (¶[0043]); using that known sensor implementation supplies the recited analog signal form. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to use Potes's known analog arterial-pressure sensor implementation, thereby providing the recited analog signal form.
Claim 1: an analog-to-digital converter that converts the analog hemodynamic sensor signals to arterial pressure signal waveforms;
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Potes teaches ADC 44 converting analog hemodynamic sensor data to digital data representative of the arterial-pressure waveform (¶[0043], [0049]). Adding that known ADC permits the Reference's scoring processor to operate on digitized waveform data. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to include Potes's known ADC in the arterial-pressure sensing path so that the scoring processor operates on digitized waveform data.
Claim 1: a graphical user interface configured to display an alert indicating the determined likelihood of effectiveness of the intervention;
Ref claim 1: “a user interface” and “output, via the user interface, a representation of the first score that is predictive of responsiveness of the patient to the intravenous delivery of the saline fluid, a representation of the second score that is predictive of responsiveness of the patient to the intravenous delivery of the vasopressor, and a representation of the third score that is predictive of responsiveness of the patient to the intravenous delivery of the inotrope.”
Ref claim 1 already outputs therapy-responsiveness scores through a user interface, but does not recite the claimed graphical alert. Potes teaches a GUI and graphical sensory alarm displaying predicted probabilities and risk scores (¶[0048], [0056], [0058]); using that display technique for the Reference's responsiveness score provides the limitation. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to use Potes's known graphical probability and alarm presentation for the claimed responsiveness score.
Claim 1: a non-transitory memory having executable instructions and a deep learning model stored thereon;
Ref claim 1: “a system memory that stores hemodynamic therapy scoring software code”
Ref claim 1 recites stored scoring software but not a stored deep-learning model. Ko teaches a stored deep-learning program and neural-network structure, filter values, and activation functions (¶[0020]) and a 1D CNN processing arterial blood-pressure data (¶[0045], [0047]); implementing the numerical arterial-pressure scoring with that known model supplies the limitation. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to implement its arterial-pressure scoring with Ko's known 1D CNN architecture.
Claim 1: an electronic hardware processor in communication with the non-transitory memory and configured to execute the instructions to cause the system to at least:
Ref claim 1: “a hardware processor that is configured to execute the hemodynamic therapy scoring software code”
Ref claim 1 recites a system memory storing hemodynamic therapy scoring software code and a hardware processor configured to execute that stored code. The claimed processor-memory communication is therefore substantively present.
Claim 1: receive, from the hemodynamic sensor, a test analog hemodynamic sensor signal from the patient; convert, using the analog-to-digital converter, the test analog hemodynamic sensor signal to a test arterial pressure signal waveform;
Ref claim 1 derives scoring data “from the sensed hemodynamic data representative of the arterial pressure waveform of the patient.”
Ref claim 1 does not recite the additional limitation. Potes expressly receives analog arterial-pressure data from the patient and converts it through ADC 44 to digital waveform data (¶[0049]). Applying that known acquisition path to the Reference supplies the test-signal and conversion sequence. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to use Potes's known analog acquisition and ADC path for the patient arterial-pressure waveform.
Claim 1: extract from the test arterial pressure signal waveform a plurality of test heart health parameters from a set of available heart health parameters, wherein the set of available heart health parameters comprises: a mean arterial pressure (MAP); a stroke volume index (SVI); a hypotension prediction index (HPI); a systemic vascular resistance (SVR); a heart rate (HR); a cardiac output (CO); a time-based change in arterial pressure; a cardiac index (CI); a systemic vascular resistance index (SVRI); a normalized area of pulse pressure; an average distance between subsequent MAPs; an average distance between a systolic peak and a respective diastolic peak; and a stroke volume variation (SVV);
Ref claim 1: “derive, from the sensed hemodynamic data representative of the arterial pressure waveform of the patient, first magnitude data and first trend data of a first hemodynamic parameter selected from the group consisting of a stroke volume variation (SVV), a pulse pressure variation (PPV), a stroke volume (SV), and a stroke volume index (SVI)”; “derive, from the sensed hemodynamic data representative of the arterial pressure waveform of the patient, second magnitude data and second trend data of a second hemodynamic parameter selected from the group consisting of a systemic vascular resistance (SVR) and a systemic vascular resistance index (SVRI)”; and “derive, from the sensed hemodynamic data representative of the arterial pressure waveform of the patient, third magnitude data and third trend data of a third hemodynamic parameter selected from the group consisting of a maximal rate of rise of arterial pressure (dP/dtmax), a cardiac output (CO), and a cardiac index (CI).”
Ref claim 1 expressly claims SVI, SVV, SVR, SVRI, CO, and CI. Hatib'076 adds MAP and HPI (¶[0038]-[0042]), and Hatib'290 adds HR, time-based MAP change, and normalized arterial-pressure waveform area (¶[0042], [0063], Appendix A). Johnson teaches successive MAP variation over multiple beats/time and systolic-peak-to-diastolic-trough pressure difference (¶[0051], [0053]). Johnson does not expressly recite the claimed averages. Averaging the absolute successive-MAP differences and the per-beat systolic-to-diastolic pressure differences over the disclosed sampling window would provide representative multi-beat measures; under the adopted construction, the diastolic trough is the recited diastolic pressure extremum. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to include the cited known hemodynamic parameters in the available parameter set and to average the cited multi-beat pressure differences over the disclosed sampling window.
Claim 1: obtain a plurality of reference arterial pressure signal waveforms from a plurality of patients; extract from the plurality of reference arterial pressure signal waveforms a plurality of reference sets of heart health parameters, each of the reference sets of heart health parameters comprising corresponding heart health parameters from the set of available heart health parameters;
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Potes uses arterial-pressure waveforms from a critical-care patient population and feature-specific training and validation patient data (¶[0007], [0030]-[0031]). Applying that known multi-patient reference framework to the Reference's arterial-pressure scoring supplies corresponding reference parameter sets. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to use Potes's known multi-patient arterial-pressure reference framework with the claimed scoring process.
Claim 1: apply a filter to the reference sets of heart health parameters to generate a feature map associated with the filter; consolidate the filtered sets of reference heart health parameters into corresponding reference feature pools; select a reference maximum feature from each of the reference feature pools; reduce overfitting of the reference feature pools by removing one or more of the reference feature pools from remaining reference feature pools; normalize the selected reference maximum features from the reference feature pools; generate, based on the plurality of heart health parameters, one or more feature pools; determine normalized test maximum features from the plurality of test heart health parameters;
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Ko teaches convolution, max pooling, dropout to prevent overfitting, and batch normalization (¶[0045], [0047]), but does not expressly normalize the selected pooled maxima. Applying Ko's normalization to the selected reference and test pooled maxima would place corresponding retained feature values on a consistent numerical basis for the later patient-to-reference comparison. Applying Ko's dropout to the pooled reference representation supplies the recited overfitting reduction, while max pooling supplies the selected maxima. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to apply Ko's known convolution, max-pooling, dropout, and normalization operations to the reference and test feature data before the patient-to-reference comparison.
Claim 1: determine, based on a comparison between the normalized selected reference maximum features and the normalized test maximum features, that an intervention for hypotension has been implemented and the likelihood of effectiveness of the intervention for hypotension of the patient;
Ref claim 1: “output, via the user interface, a representation of the first score that is predictive of responsiveness of the patient to the intravenous delivery of the saline fluid, a representation of the second score that is predictive of responsiveness of the patient to the intravenous delivery of the vasopressor, and a representation of the third score that is predictive of responsiveness of the patient to the intravenous delivery of the inotrope.”
Ref claim 1 supplies therapy-specific responsiveness scores based on magnitude and trend data. Potes compares current-patient and reference-patient physiological features to identify whether a clinical intervention was administered (¶[0053]-[0055]) and expressly teaches tailoring the same arterial-pressure physiological-state model to predict fluid responsiveness (¶[0038]). Thus, using Ko's normalized reference/test comparison to supplement the corresponding Reference responsiveness score would add therapy-responsive physiological information suitable for the responsiveness determination while retaining the Reference's required magnitude/trend basis. The resulting effectiveness likelihood concerns the same detected intervention and is based on the claimed comparison. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to use the Potes and Ko patient-to-reference feature processing with the claimed therapy-responsiveness scoring to determine the implemented intervention and its expected effectiveness.
Claim 1: generate, based on the determined likelihood of effectiveness of the intervention for hypotension of the patient, data for displaying, via the graphical user interface, the alert indicating the likelihood of effectiveness of the intervention for hypotension.
Ref claim 1: “output, via the user interface, a representation of the first score that is predictive of responsiveness of the patient to the intravenous delivery of the saline fluid, a representation of the second score that is predictive of responsiveness of the patient to the intravenous delivery of the vasopressor, and a representation of the third score that is predictive of responsiveness of the patient to the intravenous delivery of the inotrope”
Ref claim 1 already outputs representations of its predictive responsiveness scores. Potes teaches communicating predicted probabilities and risk scores through a GUI and graphical alarm (¶[0056], [0058]); applying that known presentation technique to the Reference's responsiveness score provides the claimed alert form. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to present its responsiveness score using Potes's known graphical probability and alarm technique.
Claims 2-7 are rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 17 of Smith et al. (U.S. Patent No. 12,672,822), hereinafter Reference or Ref, in view of Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes, and Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko.
The analysis is as follows (please note, the bolded and underlined portions of the entries under the instant application (IA) are those portions of the IA claims that the Reference claim does not have or are different from in some form):
Instant Application (IA) claim language(bold + underlined = not in / different from Reference claim)
Reference - U.S. Patent No. 12,672,822(corresponding recited claim language)
Analysis (comparison and secondary prior art)
Claim 2: A hemodynamic sensor system configured to determine a likelihood of effectiveness of an intervention for hypotension, the system comprising:
Ref claim 17: “A system for monitoring arterial pressure of a patient and determining one or more scores that are predictive of responsiveness of the patient to a corresponding therapy”
Ref claim 17 already claims prospective scores predictive of responsiveness to saline, vasopressor, and inotrope therapies. Potes identifies both fluid responsiveness and a hypotensive state as applications of the same arterial-pressure physiological-state model (¶[0038]); applying the responsiveness determination in that context provides the claimed hypotension-intervention use. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 by applying Potes's known fluid-responsiveness prediction in the hypotension context, thereby providing the claimed intervention-for-hypotension use.
Claim 2: a hemodynamic sensor that produces analog hemodynamic sensor signals representative of arterial pressure signal waveforms of one or more patients; an analog-to-digital converter that converts the analog hemodynamic sensor signals to arterial pressure signal waveforms;
Ref claim 17: “a hemodynamic sensor that produces sensed hemodynamic data representative of an arterial pressure waveform of the patient”
Ref claim 17 recites arterial-pressure sensing but not the analog/ADC implementation. Potes teaches analog arterial-pressure sensor data and ADC conversion to digital waveform data (¶[0043], [0049]); using that known acquisition path supplies the added limitations. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 to use Potes's known analog arterial-pressure acquisition and ADC implementation.
Claim 2: a non-transitory memory having executable instructions and a deep learning model stored thereon;
Ref claim 17: “a system memory that stores hemodynamic therapy scoring software code”
Ref claim 17 does not recite the additional limitation. Ko teaches a stored deep-learning program and neural-network structure, filter values, and activation functions (¶[0020]) and a 1D CNN operating on arterial blood-pressure data (¶[0045], [0047]). Implementing the Reference's stored scoring software with that known deep-learning architecture supplies the limitation. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 to implement its stored arterial-pressure scoring software with Ko's known 1D CNN architecture.
Claim 2: an electronic hardware processor in communication with the non-transitory memory and configured to execute the instructions to cause the system to at least:
Ref claim 17: “a hardware processor that is configured to execute the hemodynamic therapy scoring software code to:”
Ref claim 17 directly claims a hardware processor configured to execute the hemodynamic therapy scoring software code stored in system memory. The processor limitation is substantively present.
Claim 2: receive, from the hemodynamic sensor, a test analog hemodynamic sensor signal from the patient; convert, using the analog-to-digital converter, the test analog hemodynamic sensor signal to a test arterial pressure signal waveform; extract from the test arterial pressure signal waveform a plurality of test heart health parameters from a set of available heart health parameters;
Ref claim 17: “derive, from the sensed hemodynamic data representative of the arterial pressure waveform of the patient: first magnitude data and first trend data of a first hemodynamic parameter indicative of a condition treatable by an intravenous delivery of a saline fluid to the patient; second magnitude data and second trend data of a second hemodynamic parameter indicative of a condition treatable by an intravenous delivery of a vasopressor to the patient; and third magnitude data and third trend data of a third hemodynamic parameter indicative of a condition treatable by an intravenous delivery of an inotrope to the patient.”
Ref claim 17 derives multiple therapy-related hemodynamic parameters from the patient's arterial-pressure waveform. Potes expressly receives analog patient waveform data, converts it with ADC 44, and extracts multiple hemodynamic parameters (¶[0049]-[0051]); using that known acquisition and feature-extraction implementation supplies the recited test processing. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 to use Potes's known analog acquisition, ADC, and feature-extraction implementation for the patient arterial-pressure waveform.
Claim 2: generate, using a plurality of reference arterial pressure signal waveforms from a plurality of patients, one or more filtered sets of reference heart health parameters associated with the plurality of reference arterial pressure signal waveforms; consolidate the filtered sets of reference heart health parameters into corresponding reference feature pools; determine, based on the reference feature pools, normalized test maximum features from the plurality of test heart health parameters;
No corresponding limitation is recited in Ref claim 17.
Ref claim 17 does not recite the additional limitation. Potes supplies reference-patient arterial-pressure waveforms and feature-specific reference data (¶[0007], [0030]-[0031]); Ko supplies convolution, max pooling, learned filter values, and batch normalization (¶[0039], [0044]-[0051]). Ko does not expressly normalize the selected pooled maxima. Applying the same normalization processing to the pooled reference/training representation and the selected test pooled maxima would control the scale of pooled feature values and maintain a consistent numerical representation between training and subsequent prediction. Max pooling supplies the selected maxima, and using processing learned from the reference feature pools makes the normalized test maxima determined based on those reference pools. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 to apply Potes's reference-patient framework and Ko's known convolution, max-pooling, and normalization processing to the reference and test feature data.
Claim 2: determine, based on the normalized test maximum features, that an intervention for hypotension has been implemented and the likelihood of effectiveness of the intervention for hypotension of the patient;
Ref claim 17: “output, via the user interface, a representation of the first score that is predictive of responsiveness of the patient to the intravenous delivery of the saline fluid, a representation of the second score that is predictive of responsiveness of the patient to the intravenous delivery of the vasopressor, and a representation of the third score that is predictive of responsiveness of the patient to the intravenous delivery of the inotrope”
Ref claim 17 supplies therapy-specific responsiveness scores based on magnitude and trend data. Potes identifies an administered intervention from patient/reference physiological features and expressly teaches tailoring its arterial-pressure physiological-state model to predict fluid responsiveness (¶[0038], [0053]-[0055]). Using Ko's normalized test maxima as the processed physiological representation for those determinations would provide therapy-responsive physiological information suitable for supplementing the corresponding Reference responsiveness score while retaining its magnitude/trend basis. The resulting effectiveness determination concerns the same detected intervention and is based on the normalized test features. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 to use the Potes and Ko processed patient-to-reference features for identifying the implemented intervention and determining the corresponding therapy-responsiveness likelihood.
Claim 2: generate, based on the determined likelihood of effectiveness of the intervention for hypotension of the patient, data for displaying an alert indicating the likelihood of effectiveness of the intervention for hypotension.
Ref claim 17: “output, via the user interface, a representation of the first score that is predictive of responsiveness of the patient to the intravenous delivery of the saline fluid, a representation of the second score that is predictive of responsiveness of the patient to the intravenous delivery of the vasopressor, and a representation of the third score that is predictive of responsiveness of the patient to the intravenous delivery of the inotrope”
Ref claim 17 already outputs representations of the responsiveness scores through the user interface. Potes teaches a sensory alarm, including visual display of predicted probabilities and risk scores (¶[0056], [0058]); applying that known alert technique to the responsiveness score provides the claimed displayed alert. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 to present its responsiveness score using Potes's known visual alarm and predicted-value display technique.
Claim 3: The hemodynamic sensor system of claim 2, wherein generating the one or more filtered sets of reference heart health parameters comprises: obtaining the plurality of reference arterial pressure signal waveforms from the plurality of patients; extracting from the plurality of reference arterial pressure signal waveforms a plurality of reference sets of heart health parameters, each of the reference sets of heart health parameters comprising corresponding heart health parameters from the set of available heart health parameters; and applying a filter to the reference sets of heart health parameters to generate a feature map associated with the filter.
No corresponding limitation is recited in Ref claim 17.
Ref claim 17 does not recite the additional limitation. Potes supplies the multi-patient reference waveforms and corresponding feature-specific reference parameter data (¶[0007], [0030]-[0031]). Ko applies learned weighted convolution filters and produces convolution-layer result values (¶[0045], [0047]); applying that operation to the reference parameter sets generates the recited filter-associated feature map. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 to apply Ko's learned convolution filters to Potes's reference-patient parameter sets, thereby generating the claimed filter-associated feature map.
Claim 4: The hemodynamic sensor system of claim 2, wherein determining the normalized test maximum features from the plurality of test heart health parameters comprises: selecting a reference maximum feature from each of the reference feature pools; reducing overfitting of the reference feature pools by removing one or more of the reference feature pools from remaining reference feature pools; normalizing the selected reference maximum features from the reference feature pools; and generating, based on the plurality of heart health parameters, one or more feature pools.
No corresponding limitation is recited in Ref claim 17.
Ref claim 17 does not recite the additional limitation. Ko teaches max pooling, dropout to prevent overfitting, and batch normalization (¶[0045], [0047]), but does not expressly normalize the selected pooled maxima. Applying dropout to the pooled reference representation supplies the recited removal for overfitting reduction, and applying Ko's normalization to the retained selected reference maxima places those values on the same numerical basis used for subsequent inference. Max pooling supplies the selected reference maxima and the test feature pools. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 to apply Ko's known max-pooling, dropout, and normalization operations to the pooled reference and test feature representations.
Claim 5: The hemodynamic sensor system of claim 2, further comprising a graphical user interface configured to display the alert indicating the determined likelihood of effectiveness of the intervention.
Ref claim 17: “a user interface”
Ref claim 17 does not recite the additional limitation. Potes teaches user interface 54 as a GUI and a visual sensory alarm displaying a risk score (¶[0048], [0058]). Configuring the Reference's user interface to use that known graphical alert technique for its responsiveness score provides the limitation. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 to use Potes's known graphical alert technique for the responsiveness score.
Claim 6: The hemodynamic sensor system of claim 5, wherein the electronic hardware processor is further configured to execute the instructions to cause the system to at least: generate, based on the determined likelihood of effectiveness of the intervention, data for displaying at least the alert indicating the likelihood of effectiveness of the intervention for hypotension.
Ref claim 17: “output, via the user interface, a representation of the first score that is predictive of responsiveness of the patient to the intravenous delivery of the saline fluid, a representation of the second score that is predictive of responsiveness of the patient to the intravenous delivery of the vasopressor, and a representation of the third score that is predictive of responsiveness of the patient to the intravenous delivery of the inotrope”
Ref claim 17 does not recite the additional limitation. Potes teaches generating and displaying predicted patient-state information through its existing display and alarm arrangement (¶[0056], [0058], [0064]). Applying that known presentation mechanism to the Reference's responsiveness score provides the recited processor-generated alert data. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 to use Potes's known display and alarm mechanism to generate the recited alert data for the responsiveness score.
Claim 7: The hemodynamic sensor system of claim 2, wherein the alert comprises at least one of a symbol, a numerical value, a visual design, a repeated indicator, or a highlight.
No corresponding limitation is recited in Ref claim 17.
Ref claim 17 does not recite the additional limitation. Potes teaches flashing and/or colored graphics and display of the numerical risk score (¶[0058]). Those disclosed forms provide at least the recited numerical value and visual-design alternatives. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 to display its alert using Potes's known numerical, flashing, or colored graphical forms.
Claims 8-9 are rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 17 of Smith et al. (U.S. Patent No. 12,672,822), hereinafter Reference or Ref, in view of Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes, Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, and Al Hatib et al. (US 2018/0008205 A1), hereinafter Hatib'205.
The analysis is as follows (please note, the bolded and underlined portions of the entries under the instant application (IA) are those portions of the IA claims that the Reference claim does not have or are different from in some form):
The Ref claim 17 + Potes + Ko analysis above supplies the limitations incorporated from instant claim 2. The additional limitations of claims 8-9 are addressed below.
Instant Application (IA) claim language(bold + underlined = not in / different from Reference claim)
Reference - U.S. Patent No. 12,672,822(corresponding recited claim language)
Analysis (comparison and secondary prior art)
Claim 8: The hemodynamic sensor system of claim 2, wherein the electronic hardware processor is further configured to execute the instructions to cause the system to at least: determine, based on the normalized test maximum features, that the patient will experience hypotension within a specified time with at least a target threshold confidence level.
No corresponding limitation is recited in Ref claim 17.
Ref claim 17 does not recite the additional limitation. Ko teaches training prediction processing for corresponding n-minute advance intervals, including 5-, 10-, and 15-minute examples, and predicts hypotension after the selected n-minute interval (¶[0041]-[0043], [0052]-[0053]). Hatib'205 teaches applying a predetermined percentage criterion, including an 85% threshold, to a future-hypotension risk score (¶[0021], [0074]-[0076]). Using processing trained for the selected interval and applying the percentage criterion provides the specified-time determination at the target confidence level. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 to apply Ko's interval-trained future-hypotension prediction together with Hatib'205's predetermined percentage criterion, thereby providing the specified-time determination at the target confidence level.
Claim 9: The hemodynamic sensor system of claim 8, wherein the electronic hardware processor is further configured to execute the instructions to cause the system to at least: generate, based on the determination that that the patient will experience hypotension within the specified time with at least the target threshold confidence level, data for displaying, via a graphical user interface, the alert indicating an expected time to event of hypotension within the patient.
No corresponding limitation is recited in Ref claim 17.
Ref claim 17 does not recite the additional limitation. Ko expressly notifies medical staff through a display of the occurrence of hypotension 'after n minutes' and permits different n-minute horizons (¶[0052]-[0053]). Potes already supplies a GUI for predicted patient-state information (¶[0048]); displaying Ko's expected event time through that GUI provides the limitation. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 to display Ko's expected event time through the graphical user interface supplied by Potes.
Claim 10 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 17 of Smith et al. (U.S. Patent No. 12,672,822), hereinafter Reference or Ref, in view of Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes, Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, Al Hatib et al. (US 2018/0008205 A1), hereinafter Hatib'205, and Gholami et al. (US 2020/0114077 A1), hereinafter Gholami.
The analysis is as follows (please note, the bolded and underlined portions of the entries under the instant application (IA) are those portions of the IA claims that the Reference claim does not have or are different from in some form):
The Ref claim 17 + Potes + Ko + Hatib'205 analysis above supplies the limitations incorporated from instant claims 2 and 8. The additional limitation of claim 10 is addressed below.
Instant Application (IA) claim language(bold + underlined = not in / different from Reference claim)
Reference - U.S. Patent No. 12,672,822(corresponding recited claim language)
Analysis (comparison and secondary prior art)
Claim 10: The hemodynamic sensor system of claim 8, further comprising an infusion pump, wherein the electronic hardware processor is further configured to execute the instructions to cause the system to at least: generate a control signal for the infusion pump to deliver an intravenous therapy to the patient based on the likelihood of effectiveness of the intervention for hypotension of the patient.
No corresponding limitation is recited in Ref claim 17.
Ref claim 17 does not recite the additional limitation. Gholami teaches closed-loop intravenous fluid administration using hemodynamic measurements, including fluid-responsiveness information, to compute infusion rates and send a command to an infusion pump (Abstract; ¶[0040], [0043], [0068], [0075]). Using the responsiveness likelihood already determined by the modified Reference system as an input to that pump controller provides automatic delivery when the therapy is expected to be effective. It would have been obvious to one of ordinary skill in the art to modify Ref claim 17 to use Gholami's known hemodynamic infusion-pump controller with the responsiveness likelihood already determined by the modified reference system.
Claims 11-14 are rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 1 of Al Hatib et al. (U.S. Patent No. 12,076,167), hereinafter Reference or Ref, in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko.
The analysis is as follows (please note, the bolded and underlined portions of the entries under the instant application (IA) are those portions of the IA claims that the Reference claim does not have or are different from in some form):
Instant Application (IA) claim language(bold + underlined = not in / different from Reference claim)
Reference - U.S. Patent No. 12,076,167(corresponding recited claim language)
Analysis (comparison and secondary prior art)
Claim 11: A hemodynamic sensor system configured to determine a likelihood that a patient will experience hypotension within a specified time with at least a target threshold confidence level, the system comprising:
Ref claim 1: “A system for monitoring of arterial pressure of a patient and providing a warning to medical personnel of a predicted future hypotensive event” and “determine, using the predictive weighting module, a risk score representing a probability of a future hypotension event for the patient based on a weighted combination of a plurality of hypotension profiling parameters including the one or more of the vital sign parameters, the differential parameters and the combinatorial parameters.”
Ref claim 1 already claims a future-hypotension probability and predetermined risk criterion. Ko teaches training prediction processing for corresponding n-minute advance intervals and generating an n-minute hypotension probability (¶[0041]-[0043], [0052]-[0053]). Using processing trained for the received interval and applying the Reference's criterion to that probability provides the specified-time, threshold-confidence determination. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to use Ko's interval-trained future-hypotension prediction while applying the predetermined risk criterion already recited by Ref claim 1.
Claim 11: a hemodynamic sensor that produces analog hemodynamic sensor signals representative of arterial pressure signal waveforms of patients; an analog-to-digital converter that converts the analog hemodynamic sensor signals to arterial pressure signal waveforms;
Ref claim 1: “a minimally invasive or non-invasive hemodynamic sensor that produces an analog hemodynamic sensor signal representative of an arterial pressure waveform of the patient” and “an analog-to-digital converter that converts the analog hemodynamic sensor signal to digital hemodynamic data.”
Ref claim 1 substantively recites the analog arterial-pressure sensor and ADC conversion. The plural wording of the instant sensor signals does not render the subject matter patentably distinct.
Claim 11: a non-transitory memory having executable instructions and a deep learning model stored thereon;
Ref claim 1: “a system memory that stores hypotension prediction software code including a predictive weighting module”
Ref claim 1 stores predictive software but does not recite the claimed deep-learning model. Ko teaches a stored deep-learning program and neural-network structure, filter values, and activation functions (¶[0020]) and a 1D CNN for arterial blood-pressure data (¶[0045], [0047]); implementing the Reference's predictive software with that known architecture supplies the limitation. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to implement its hypotension-prediction software with Ko's known 1D CNN architecture.
Claim 11: an electronic hardware processor in communication with the non-transitory memory and configured to execute the instructions to cause the system to at least:
Ref claim 1: “a hardware processor that is configured to execute the hypotension prediction software code”
Ref claim 1 recites system memory storing hypotension-prediction software code and a hardware processor configured to execute that stored code. The claimed processor-memory communication is therefore substantively present.
Claim 11: receive the specified time and
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Ko teaches training examples for different advance intervals, including 5, 10, and 15 minutes, and permits other values of n (¶[0041]-[0043], [0052]). Receiving the specified time selects prediction processing trained for the corresponding advance interval. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to receive the specified time and select Ko's prediction processing trained for the corresponding advance interval.
Claim 11: receive, from the hemodynamic sensor, a test analog hemodynamic sensor signal from the patient; convert, using the analog-to-digital converter, the test analog hemodynamic sensor signal to a test arterial pressure signal waveform; extract from the test arterial pressure signal waveform a plurality of test heart health parameters from a set of available heart health parameters;
Ref claim 1: “a minimally invasive or non-invasive hemodynamic sensor that produces an analog hemodynamic sensor signal representative of an arterial pressure waveform of the patient”; “an analog-to-digital converter that converts the analog hemodynamic sensor signal to digital hemodynamic data”; and “perform waveform analysis of the digital hemodynamic data to obtain vital sign parameters from the digital hemodynamic data representing the arterial pressure waveform.”
Ref claim 1 already senses the patient's analog arterial-pressure waveform, digitizes it, and performs waveform analysis to obtain vital-sign parameters. Treating the current patient's signal and derived parameters as the test signal and test parameters is the ordinary use of that claimed monitoring system.
Claim 11: generate, using a plurality of reference arterial pressure signal waveforms from a plurality of patients, one or more filtered sets of reference heart health parameters associated with the plurality of reference arterial pressure signal waveforms; consolidate the filtered sets of reference heart health parameters into corresponding reference feature pools; determine, based on the reference feature pools, normalized test maximum features from the plurality of test heart health parameters;
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Ko uses multi-patient arterial-pressure data for training and teaches convolution, max pooling, learned filter values, and batch normalization (¶[0039], [0044]-[0051]). Ko does not expressly normalize the selected pooled maxima. Applying the same normalization processing to the pooled reference/training representation and the selected test pooled maxima would control the scale of the pooled feature values and maintain a consistent numerical representation between training and subsequent prediction. Max pooling supplies the selected maxima, with the test processing based on parameters learned from the reference-patient feature pools. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to apply Ko's multi-patient CNN feature processing and normalization to the reference and test arterial-pressure feature data.
Claim 11: determine, based on the normalized test maximum features, the likelihood that the patient will experience hypotension within the specified time with at least the target threshold confidence level;
Ref claim 1: “determine, using the predictive weighting module, a risk score representing a probability of a future hypotension event for the patient based on a weighted combination of a plurality of hypotension profiling parameters including the one or more of the vital sign parameters, the differential parameters and the combinatorial parameters” and “invoke the sensory alarm to produce the sensory signal in response to the risk score satisfying a predetermined risk criterion.”
Ref claim 1 already determines a future-hypotension probability and applies a predetermined risk criterion. Ko supplies the normalized CNN feature representation and prediction processing trained for the selected n-minute interval (¶[0041]-[0047], [0052]-[0053]). Using those processed features with the interval-trained prediction and applying the Reference's criterion provides the limitation. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to use Ko's normalized interval-trained feature representation with the predetermined risk criterion already recited by Ref claim 1.
Claim 11: generate, based on the determination of the likelihood that that the patient will experience hypotension within the specified time with at least the target threshold confidence level, data for displaying an alert indicating an expected time to event of hypotension within the patient.
Ref claim 1: “a user interface that includes a sensory alarm that provides a sensory signal to warn medical personnel of the predicted future hypotensive event prior to the patient entering a hypotensive state” and “invoke the sensory alarm to produce the sensory signal in response to the risk score satisfying a predetermined risk criterion.”
Ref claim 1 already provides an alarm for the predicted future event. Ko expressly notifies medical staff through a display that hypotension is expected 'after n minutes' (¶[0052]-[0053]); adding that already-determined prediction time to the Reference's warning provides the expected-time alert. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to add Ko's already-determined prediction time to the claimed warning, thereby providing the expected-time alert.
Claim 12: The hemodynamic sensor system of claim 11, wherein generating the one or more filtered sets of reference heart health parameters comprises: obtaining the plurality of reference arterial pressure signal waveforms from the plurality of patients; extracting from the plurality of reference arterial pressure signal waveforms a plurality of reference sets of heart health parameters, each of the reference sets of heart health parameters comprising corresponding heart health parameters from the set of available heart health parameters; and applying a filter to the reference sets of heart health parameters to generate a feature map associated with the filter.
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Ko uses arterial blood-pressure data from a plurality of patients for training, inputs extracted data to the deep network, and applies weighted convolution filters (¶[0039], [0044]-[0047]). Applying those filters to the reference-patient parameter sets supplies the corresponding reference sets and filter-associated feature map. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to apply Ko's learned convolution filters to the multi-patient reference parameter sets, thereby providing the claimed filter-associated feature map.
Claim 13: The hemodynamic sensor system of claim 11, wherein determining the normalized test maximum features from the plurality of test heart health parameters comprises: selecting a reference maximum feature from each of the reference feature pools; reducing overfitting of the reference feature pools by removing one or more of the reference feature pools from remaining reference feature pools; normalizing the selected reference maximum features from the reference feature pools; and generating, based on the plurality of heart health parameters, one or more feature pools.
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Ko teaches max pooling, dropout to prevent overfitting, and batch normalization (¶[0045], [0047]), but does not expressly normalize the selected pooled maxima. Applying dropout to the pooled reference representation supplies the recited overfitting reduction, and applying Ko's normalization to the retained selected reference maxima provides normalized reference maxima on a consistent numerical basis for subsequent prediction. Max pooling supplies the selected maxima and test feature pools. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to apply Ko's max-pooling, dropout, and normalization operations to the reference and test feature representations.
Claim 14: The hemodynamic sensor system of claim 11, further comprising a graphical user interface configured to display the alert indicating an expected time to event of hypotension within the patient.
Ref claim 1: “a user interface that includes a sensory alarm that provides a sensory signal to warn medical personnel of the predicted future hypotensive event prior to the patient entering a hypotensive state”
Ref claim 1 recites a user interface that provides the hypotension warning. Configuring that interface as a graphical display for Ko's 'after n minutes' notification provides the claimed user-interface implementation and directly communicates the timing information taught by Ko (¶[0052]-[0053]). It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to present Ko's 'after n minutes' prediction through a graphical implementation of the user interface already recited by Ref claim 1.
Claim 15 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 5 of Al Hatib et al. (U.S. Patent No. 12,076,167), hereinafter Reference or Ref, through its dependence from claims 4 and 1, in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, Al Hatib et al. (US 2018/0192962 A1), hereinafter Hatib'962, and Gholami et al. (US 2020/0114077 A1), hereinafter Gholami.
The analysis is as follows (please note, the bolded and underlined portions of the entries under the instant application (IA) are those portions of the IA claims that the Reference claim does not have or are different from in some form):
Because selected Ref claim 5 inherits claims 4 and 1, the reference invention includes those ancestor limitations. The Ref claim 1 + Ko analysis above supplies the limitations incorporated from instant claim 11. Claim 5 additionally recites recommending a medical intervention for preventing the future hypotension event. The remaining infusion-pump delivery limitation of claim 15 is addressed below.
Instant Application (IA) claim language(bold + underlined = not in / different from Reference claim)
Reference - U.S. Patent No. 12,076,167(corresponding recited claim language)
Analysis (comparison and secondary prior art)
Claim 15: The hemodynamic sensor system of claim 11, further comprising an infusion pump configured to deliver an intravenous therapeutic agent to the patient based on the likelihood that that the patient will experience hypotension within the specified time with at least the target threshold confidence level.
Ref claim 5, via claims 4 and 1: “The system of claim 4, wherein the hardware processor is further configured to execute the hypotension prediction software code to recommend a medical intervention for preventing the future hypotension event.” Claim 4 further requires the processor to “identify a most probable cause of the future hypotension event.”
Ref claim 5 already links the predicted future hypotension event to a recommended preventive medical intervention, but does not recite automatically delivering an intravenous therapeutic agent with an infusion pump. Hatib'962 teaches automatically causing administration of saline or whole blood in response to detection of a predicted future hypotension event (¶[0021]). Gholami teaches intravenous fluid administration through an electronically controlled infusion pump driven by hemodynamic information (¶[0040], [0043], [0074]-[0076]). Using the threshold-satisfying future-hypotension prediction as the treatment trigger and implementing the preventive intervention with Gholami's pump would provide the recited delivery. It would have been obvious to one of ordinary skill in the art to modify Ref claim 5, through claims 4 and 1, to implement its recommended preventive intervention using Hatib'962's automatic administration teaching and Gholami's electronically controlled infusion pump.
Claim 16 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 1 of Al Hatib et al. (U.S. Patent No. 12,076,167), hereinafter Reference or Ref, in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, and Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes.
The analysis is as follows (please note, the bolded and underlined portions of the entries under the instant application (IA) are those portions of the IA claims that the Reference claim does not have or are different from in some form):
Instant Application (IA) claim language(bold + underlined = not in / different from Reference claim)
Reference - U.S. Patent No. 12,076,167(corresponding recited claim language)
Analysis (comparison and secondary prior art)
Claim 16: A hemodynamic sensor system configured to determine a likelihood that a patient will experience hypotension within a specified time with at least a target threshold confidence level, the system comprising:
Ref claim 1: “A system for monitoring of arterial pressure of a patient and providing a warning to medical personnel of a predicted future hypotensive event”; “determine, using the predictive weighting module, a risk score representing a probability of a future hypotension event for the patient based on a weighted combination of a plurality of hypotension profiling parameters including the one or more of the vital sign parameters, the differential parameters and the combinatorial parameters”; and “invoke the sensory alarm to produce the sensory signal in response to the risk score satisfying a predetermined risk criterion.”
Ref claim 1 claims a future-hypotension probability and threshold criterion. Ko teaches prediction processing trained for corresponding n-minute advance intervals and an n-minute probability output (¶[0041]-[0043], [0052]-[0053]). Applying the Reference's criterion to the probability generated by processing trained for the selected interval provides the specified-time, threshold-confidence determination. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to use Ko's interval-trained future-hypotension probability while applying the threshold criterion already recited by Ref claim 1.
Claim 16: a hemodynamic sensor that produces analog hemodynamic sensor signals representative of arterial pressure signal waveforms of one or more patients; an analog-to-digital converter that converts the analog hemodynamic sensor signals to arterial pressure signal waveforms;
Ref claim 1: “a minimally invasive or non-invasive hemodynamic sensor that produces an analog hemodynamic sensor signal representative of an arterial pressure waveform of the patient” and “an analog-to-digital converter that converts the analog hemodynamic sensor signal to digital hemodynamic data.”
Ref claim 1 substantively recites the analog arterial-pressure sensor and ADC conversion. The plural wording of the instant signals does not create patentable distinctness.
Claim 16: a graphical user interface configured to display an alert indicating an expected time to event of hypotension within the patient;
Ref claim 1: “a user interface that includes a sensory alarm that provides a sensory signal to warn medical personnel of the predicted future hypotensive event prior to the patient entering a hypotensive state”
Ref claim 1 already claims a user interface and sensory warning. Ko teaches displaying that hypotension is expected 'after n minutes' (¶[0052]-[0053]); presenting that timing information through a graphical implementation of the Reference's user interface provides the limitation. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to present Ko's expected event time through a graphical implementation of the user interface already recited by Ref claim 1.
Claim 16: a non-transitory memory having executable instructions and a deep learning model stored thereon;
Ref claim 1: “a system memory that stores hypotension prediction software code including a predictive weighting module”
Ref claim 1 does not recite the additional limitation. Ko teaches a stored deep-learning model and 1D CNN processing arterial blood-pressure data (¶[0020], [0045], [0047]). Implementing the Reference's predictive processing with that known architecture supplies the limitation. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to implement its hypotension-prediction processing with Ko's known stored 1D CNN architecture.
Claim 16: an electronic hardware processor in communication with the non-transitory memory and configured to execute the instructions to cause the system to at least:
Ref claim 1: “a hardware processor that is configured to execute the hypotension prediction software code to:”
Ref claim 1 directly claims a hardware processor configured to execute the hypotension-prediction software stored in system memory. The claimed processor-memory communication is therefore substantively present.
Claim 16: receive the specified time and
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Ko teaches training and prediction for different advance intervals and permits different values of n (¶[0041]-[0043], [0052]). Receiving the specified time selects prediction processing trained for the corresponding warning interval. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to receive the specified time and select Ko's prediction processing trained for the corresponding warning interval.
Claim 16: receive, from the hemodynamic sensor, a test analog hemodynamic sensor signal from the patient; convert, using the analog-to-digital converter, the test analog hemodynamic sensor signal to a test arterial pressure signal waveform; extract from the test arterial pressure signal waveform a plurality of test heart health parameters from a set of available heart health parameters;
Ref claim 1: “a minimally invasive or non-invasive hemodynamic sensor that produces an analog hemodynamic sensor signal representative of an arterial pressure waveform of the patient”; “an analog-to-digital converter that converts the analog hemodynamic sensor signal to digital hemodynamic data”; and “perform waveform analysis of the digital hemodynamic data to obtain vital sign parameters from the digital hemodynamic data representing the arterial pressure waveform.”
Ref claim 1 already senses the patient's arterial-pressure waveform, digitizes it, and obtains waveform-derived parameters. Treating the current-patient signal and parameters as test data is the ordinary operation of that claimed system.
Claim 16: obtain a plurality of reference arterial pressure signal waveforms from a plurality of patients; extract from the plurality of reference arterial pressure signal waveforms a plurality of reference sets of heart health parameters, each of the reference sets of heart health parameters comprising corresponding heart health parameters from the set of available heart health parameters;
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Ko uses arterial blood-pressure data from a plurality of patients for model development and training (¶[0039], [0044]); using those patient waveforms and extracted training parameters supplies the claimed reference waveforms and corresponding reference parameter sets. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to use Ko's known multi-patient arterial-pressure training framework as the claimed reference waveform and parameter-set source.
Claim 16: apply a filter to the reference sets of heart health parameters to generate a feature map associated with the filter; consolidate the filtered sets of reference heart health parameters into corresponding reference feature pools; select a reference maximum feature from each of the reference feature pools; reduce overfitting of the reference feature pools by removing one or more of the reference feature pools from remaining reference feature pools; normalize the selected reference maximum features from the reference feature pools; generate, based on the plurality of heart health parameters, one or more feature pools; determine normalized test maximum features from the plurality of test heart health parameters;
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Ko teaches convolution filters, max pooling, dropout to prevent overfitting, and batch normalization (¶[0045], [0047]), but does not expressly normalize selected pooled maxima. Applying dropout to the pooled reference representation reduces overfitting, and applying Ko's normalization to the selected reference and test pooled maxima places corresponding values on a consistent numerical basis for the later patient-to-reference comparison. Max pooling supplies the selected maxima and feature pools. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to apply Ko's convolution, max-pooling, dropout, and normalization operations to the reference and test feature data.
Claim 16: determine, based on a comparison between the normalized selected reference maximum features and the normalized test maximum features, the likelihood that the patient will experience hypotension within the specified time with at least the target threshold confidence level;
Ref claim 1: “determine, using the predictive weighting module, a risk score representing a probability of a future hypotension event for the patient based on a weighted combination of a plurality of hypotension profiling parameters including the one or more of the vital sign parameters, the differential parameters and the combinatorial parameters.”
Ref claim 1 supplies the future-hypotension prediction but not the claimed current-to-reference comparison. Potes compares current-patient and reference-patient features using a distance metric and permits adjustment of the selected feature and metric to improve prediction accuracy (¶[0037], [0053]-[0054]). Using Ko's normalized reference/test maxima as the compared values, with the hypotension predictor trained for the selected interval, supplies the claimed comparison basis for the Reference's prediction. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to use Potes's current-to-reference feature comparison with Ko's normalized feature values as the comparison basis for the future-hypotension prediction.
Claim 16: generate, based on the determination of the likelihood that that the patient will experience hypotension within the specified time with at least the target threshold confidence level, data for displaying, via the graphical user interface, the alert indicating the expected time to event of hypotension within the patient.
Ref claim 1: “a user interface that includes a sensory alarm that provides a sensory signal to warn medical personnel of the predicted future hypotensive event prior to the patient entering a hypotensive state” and “invoke the sensory alarm to produce the sensory signal in response to the risk score satisfying a predetermined risk criterion.”
Ref claim 1 already generates the future-hypotension alarm. Ko teaches displaying that the predicted event will occur after n minutes (¶[0052]-[0053]); displaying that timing information through the graphical implementation of the Reference's user interface provides the claimed expected-time alert. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to add Ko's already-determined prediction time to the future-hypotension alarm.
Claims 17-19 and 21-22 are rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 1 of Al Hatib et al. (U.S. Patent No. 11,317,820), hereinafter Reference or Ref, in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes, and Smith et al. (US 2021/0259629 A1), hereinafter Smith'629.
The analysis is as follows (please note, the bolded and underlined portions of the entries under the instant application (IA) are those portions of the IA claims that the Reference claim does not have or are different from in some form):
Instant Application (IA) claim language(bold + underlined = not in / different from Reference claim)
Reference - U.S. Patent No. 11,317,820(corresponding recited claim language)
Analysis (comparison and secondary prior art)
Claim 17: A hemodynamic sensor system configured to determine a likelihood that a patient will experience hypotension within a specified time with at least a target threshold confidence level and determine a likelihood of effectiveness of an intervention for the hypotension, the system comprising:
Ref claim 1: “A system for monitoring of arterial pressure of a patient and providing a warning to medical personnel of a predicted future hypotension event” and “determine a hypotension probability index (HPI) representing a probability of a future hypotension event for the patient based on the plurality of health parameters.”
Ref claim 1 already claims a future-hypotension HPI. Ko teaches prediction processing trained for corresponding n-minute advance intervals and an n-minute hypotension probability (¶[0041]-[0043], [0052]-[0053]); Smith'629 supplies numerical therapy-responsiveness scores (¶[0046]-[0053], [0081]-[0082]). Those teachings supply the specified-time confidence and therapy-effectiveness additions, with the detailed relationships addressed below. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 with Ko's interval-trained future-hypotension prediction and Smith'629's therapy-responsiveness scoring to provide the specified-time confidence and therapy-effectiveness determinations.
Claim 17: a hemodynamic sensor that produces analog hemodynamic sensor signals representative of arterial pressure signal waveforms of one or more patients; an analog-to-digital converter that converts the analog hemodynamic sensor signals to arterial pressure signal waveforms;
Ref claim 1: “a hemodynamic sensor that produces an analog hemodynamic sensor signal representative of an arterial pressure waveform of the patient” and “an analog-to-digital converter that converts the analog hemodynamic sensor signal to digital hemodynamic data.”
Ref claim 1 substantively recites the analog arterial-pressure sensor and ADC conversion. The plural wording of the instant signals does not create patentable distinctness.
Claim 17: a non-transitory memory having executable instructions and a deep learning model stored thereon;
Ref claim 1: “a system memory that stores hypotension prediction software code and a graphical user interface”
Ref claim 1 stores hypotension-prediction software and a graphical user interface but not a deep-learning model. Ko teaches a stored deep-learning program and a 1D CNN for arterial blood-pressure data (¶[0020], [0045], [0047]); using that known model supplies the limitation. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to implement its hypotension-prediction software with Ko's known stored 1D CNN architecture.
Claim 17: an electronic hardware processor in communication with the non-transitory memory and configured to execute the instructions to cause the system to at least:
Ref claim 1: “a hardware processor that is configured to: execute the hypotension prediction software code to:”
Ref claim 1 directly claims a hardware processor configured to execute the hypotension-prediction software stored in system memory. The processor limitation is substantively present.
Claim 17: receive the specified time;
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Ko teaches training examples for different future intervals and permits different values of n (¶[0041]-[0043], [0052]). Receiving the specified time selects prediction processing trained for the corresponding advance interval. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to receive the specified time and select Ko's prediction processing trained for the corresponding advance interval.
Claim 17: receive, from the hemodynamic sensor, a test analog hemodynamic sensor signal from the patient; convert, using the analog-to-digital converter, the test analog hemodynamic sensor signal to a test arterial pressure signal waveform; extract from the test arterial pressure signal waveform a plurality of test heart health parameters from a set of available heart health parameters;
Ref claim 1: “a hemodynamic sensor that produces an analog hemodynamic sensor signal representative of an arterial pressure waveform of the patient”; “an analog-to-digital converter that converts the analog hemodynamic sensor signal to digital hemodynamic data”; and “derive, from the digital hemodynamic data, a plurality of health parameters.”
Ref claim 1 already senses and digitizes the patient's arterial-pressure waveform and derives a plurality of health parameters. Treating the current-patient signal and derived parameters as test data is the ordinary operation of that claimed system.
Claim 17: generate, using a plurality of reference arterial pressure signal waveforms from a plurality of patients, one or more filtered sets of reference heart health parameters associated with the plurality of reference arterial pressure signal waveforms; consolidate the filtered sets of reference heart health parameters into corresponding reference feature pools; determine, based on the reference feature pools, normalized test maximum features from the plurality of test heart health parameters;
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Ko supplies multi-patient arterial-pressure training data, convolution, max pooling, learned filter values, and batch normalization (¶[0039], [0044]-[0051]), but does not expressly normalize the selected pooled maxima. Applying the same normalization processing to the pooled reference/training representation and the selected test pooled maxima would control the scale of the pooled feature values and maintain a consistent numerical representation between training and subsequent physiological-state prediction. Max pooling supplies the selected maxima, with the test processing based on parameters learned from the reference-patient feature pools. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to apply Ko's multi-patient CNN feature processing and normalization to the reference and test arterial-pressure feature data.
Claim 17: determine, based on the normalized test maximum features, the likelihood that the patient will experience hypotension within the specified time with at least the target threshold confidence level;
Ref claim 1: “determine a hypotension probability index (HPI) representing a probability of a future hypotension event for the patient based on the plurality of health parameters”
Ref claim 1 already determines an HPI representing future-hypotension probability. Ko supplies normalized CNN features and prediction processing trained for the selected n-minute interval (¶[0041]-[0047], [0052]-[0053]). Using those processed test features with the interval-trained predictor, and applying the Reference's predetermined risk criterion, provides the specified-time, threshold-confidence determination. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to use Ko's normalized interval-trained test features with the future-hypotension assessment already recited by Ref claim 1.
Claim 17: determine, based on the normalized test maximum features, that an intervention for hypotension has been implemented and the likelihood of effectiveness of the intervention for hypotension of the patient;
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Potes identifies administered interventions from patient/reference physiological features and expressly teaches tailoring its arterial-pressure physiological-state model to predict fluid responsiveness (¶[0038], [0053]-[0055]). Smith'629 supplies therapy-responsiveness scores based on arterial-pressure-derived magnitude and trend data (¶[0046]-[0053], [0081]-[0082]). Thus, using Ko's normalized test maxima as a processed representation of the same therapy-responsive physiological information would provide information suitable for the responsiveness determination while retaining Smith'629's magnitude/trend basis. The numerical effectiveness likelihood therefore concerns the same intervention identified as implemented and is based on the normalized test features. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to use Potes's intervention-state teaching, Smith'629's therapy-responsiveness scoring, and Ko's normalized test features to determine the implemented intervention and its expected effectiveness.
Claim 17: determine a hypotension prediction index (HPI) based on the determination of the likelihood that the patient will experience hypotension within the specified time with at least the target threshold confidence level and based on the likelihood of effectiveness of the intervention for hypotension of the patient;
Ref claim 1: “determine a hypotension probability index (HPI) representing a probability of a future hypotension event for the patient based on the plurality of health parameters”
Ref claim 1 already claims an HPI representing future-hypotension probability. Potes teaches using an intervention-related physiological-state probability as a factor in another current or future-event prediction (¶[0057]), and Smith'629 supplies the corresponding therapy-effectiveness likelihood. Using the predicted effectiveness of the detected hypotension intervention as a factor in the Reference's future-hypotension assessment would allow the continuing HPI risk to reflect whether the intervention is expected to improve the patient's hemodynamic condition, thereby making the HPI depend on both recited likelihoods. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to use the predicted effectiveness of the detected hypotension intervention as a factor in the future-hypotension HPI, as taught by Potes and Smith'629.
Claim 17: determine that the HPI exceeds a predetermined threshold; and generate, based on the determination that the HPI exceeds the predetermined threshold, an alert for display via a graphical user interface.
Ref claim 1: “a system memory that stores hypotension prediction software code and a graphical user interface”; “a display”; and “overlay an alarm screen as a pop-up on the display in response to the HPI of the patient satisfying a predetermined risk criterion.”
Ref claim 1 directly claims a graphical user interface and overlaying an alarm screen when the HPI satisfies a predetermined risk criterion. The instant threshold-and-alert limitation is therefore substantively present in Ref claim 1.
Claim 18: The hemodynamic sensor system of claim 17, further comprising the graphical user interface configured to display the alert indicating at least one of an expected time to event of hypotension within the patient or an indication of effectiveness of an intervention.
Ref claim 1 stores “a graphical user interface” and requires the processor to “overlay an alarm screen as a pop-up on the display” when the HPI satisfies the risk criterion.
Ref claim 1 does not recite the additional limitation. Ko teaches displaying that hypotension is expected 'after n minutes' (¶[0052]-[0053]). Adding that already-determined expected time to the Reference's graphical HPI alert satisfies at least the expected-time alternative. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to add Ko's already-determined expected event time to the graphical HPI alert.
Claim 19: The hemodynamic sensor system of claim 18, wherein the alert is displayed via the graphical user interface.
Ref claim 1: “a system memory that stores hypotension prediction software code and a graphical user interface” and “overlay an alarm screen as a pop-up on the display in response to the HPI of the patient satisfying a predetermined risk criterion.”
Ref claim 1 directly requires the graphical user interface to overlay the alarm screen on the display when the HPI satisfies the criterion. The dependent limitation is substantively recited by Ref claim 1.
Claim 21: The hemodynamic sensor system of claim 17, wherein generating the one or more filtered sets of reference heart health parameters comprises: obtaining the plurality of reference arterial pressure signal waveforms from the plurality of patients; extracting from the plurality of reference arterial pressure signal waveforms a plurality of reference sets of heart health parameters, each of the reference sets of heart health parameters comprising corresponding heart health parameters from the set of available heart health parameters; and applying a filter to the reference sets of heart health parameters to generate a feature map associated with the filter.
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Potes supplies multi-patient reference waveform data and corresponding feature-specific reference data (¶[0007], [0030]-[0031]). Ko supplies learned convolution filters and result values for arterial blood-pressure data (¶[0045], [0047]); applying those filters to the reference parameter sets provides the claimed feature map. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to apply Ko's learned convolution filters to Potes's multi-patient reference parameter sets, thereby providing the claimed feature map.
Claim 22: The hemodynamic sensor system of claim 17, wherein determining the normalized test maximum features from the plurality of test heart health parameters comprises: selecting a reference maximum feature from each of the reference feature pools; reducing overfitting of the reference feature pools by removing one or more of the reference feature pools from remaining reference feature pools; normalizing the selected reference maximum features from the reference feature pools; and generating, based on the plurality of heart health parameters, one or more feature pools.
No corresponding limitation is recited in Ref claim 1.
Ref claim 1 does not recite the additional limitation. Ko teaches max pooling, dropout to prevent overfitting, and batch normalization (¶[0045], [0047]), but does not expressly normalize the selected pooled maxima. Applying dropout to the pooled reference representation supplies the recited overfitting reduction, and applying Ko's normalization to the retained selected reference maxima provides normalized reference maxima on a consistent basis for subsequent prediction. Max pooling supplies the selected maxima and test feature pools. It would have been obvious to one of ordinary skill in the art to modify Ref claim 1 to apply Ko's max-pooling, dropout, and normalization operations to the pooled reference and test feature representations.
Claim 20 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 11 of Al Hatib et al. (U.S. Patent No. 11,317,820), hereinafter Reference or Ref, through its dependence from claims 10 and 1, in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes, Smith et al. (US 2021/0259629 A1), hereinafter Smith'629, Al Hatib et al. (US 2018/0025290 A1), hereinafter Hatib'290, and Johnson et al. (US 2021/0275783 A1), hereinafter Johnson.
The analysis is as follows (please note, the bolded and underlined portions of the entries under the instant application (IA) are those portions of the IA claims that the Reference claim does not have or are different from in some form):
Because selected Ref claim 11 inherits claims 10 and 1, the claimed reference invention includes those ancestor limitations. The Ko + Potes + Smith'629 analysis above supplies the additional dual-prediction limitations incorporated from instant claim 17. The added parameter-set limitation of claim 20 is addressed below.
Instant Application (IA) claim language(bold + underlined = not in / different from Reference claim)
Reference - U.S. Patent No. 11,317,820(corresponding recited claim language)
Analysis (comparison and secondary prior art)
Claim 20: The hemodynamic sensor system of claim 17, wherein the set of available heart health parameters comprises: a mean arterial pressure (MAP); a stroke volume index (SVI); a hypotension prediction index (HPI); a systemic vascular resistance (SVR); a heart rate (HR); a cardiac output (CO); a time-based change in arterial pressure; a cardiac index (CI); a systemic vascular resistance index (SVRI); a normalized area of pulse pressure; an average distance between subsequent MAPs; an average distance between a systolic peak and a respective diastolic peak; and a stroke volume variation (SVV).
Selected Ref claim 11, through claims 10 and 1, inherits claim 10: “the plurality of parameters presented by the selection screen include cardiac output, stroke volume, stroke volume variation, diastolic pressure, pulse rate, stroke volume index, systemic vascular resistance, mean arterial pressure, and hypotension probability index.” Selected Ref claim 11 further recites: “the plurality of parameters presented by the selection screen further include systemic vascular resistance index, cardiac index, and systolic pressure.”
Selected Ref claim 11, through claims 10 and 1, supplies MAP, SVI, HPI, SVR, CO, CI, SVRI, and SVV. Hatib'290 adds HR as a five-minute averaged parameter (Appendix A, item 9), time-based MAP change (¶[0042]), and normalized arterial-pressure waveform area (¶[0063]). Johnson teaches successive MAP variation over multiple beats/time and systolic-peak-to-diastolic-trough pressure difference (¶[0051], [0053]), but not the claimed averages. Averaging the absolute successive-MAP differences and the per-beat systolic-to-diastolic pressure differences over the sampling window provides representative multi-beat distance measures; under the adopted construction, the diastolic trough is the recited diastolic pressure extremum. It would have been obvious to one of ordinary skill in the art to modify selected Ref claim 11, through claims 10 and 1, to include the additional Hatib'290 and Johnson arterial-pressure-derived parameters and to average the disclosed multi-beat pressure differences over the sampling window.
Claim 23 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 6 of Al Hatib et al. (U.S. Patent No. 11,317,820), hereinafter Reference or Ref, through its dependence from claim 1, in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes, Smith et al. (US 2021/0259629 A1), hereinafter Smith'629, Al Hatib et al. (US 2018/0192962 A1), hereinafter Hatib'962, and Gholami et al. (US 2020/0114077 A1), hereinafter Gholami.
The analysis is as follows (please note, the bolded and underlined portions of the entries under the instant application (IA) are those portions of the IA claims that the Reference claim does not have or are different from in some form):
Because selected Ref claim 6 inherits claim 1, the reference invention includes the HPI/future-hypotension limitations of claim 1. The Ref claim 1 + Ko + Potes + Smith'629 analysis above supplies the additional dual-prediction limitations incorporated from instant claim 17. Claim 6 additionally recites that predictive health parameters assist the user in determining a medical intervention for preventing the future hypotension event. The remaining infusion-pump delivery limitation of claim 23 is addressed below.
Instant Application (IA) claim language(bold + underlined = not in / different from Reference claim)
Reference - U.S. Patent No. 11,317,820(corresponding recited claim language)
Analysis (comparison and secondary prior art)
Claim 23: The hemodynamic sensor system of claim 17, further comprising an infusion pump configured to deliver an intravenous therapeutic agent to the patient based on the likelihood that that the patient will experience hypotension within the specified time with at least the target threshold confidence level.
Ref claim 6, via claim 1: “The health monitoring unit of claim 1, wherein the HPI diagnostic screen displays predictive health parameters to assist the user in determining a medical intervention for preventing the future hypotension event.”
Ref claim 6 already links the predictive health parameters and future-hypotension HPI framework to determining a preventive medical intervention, but does not recite automatically delivering an intravenous therapeutic agent with an infusion pump. Hatib'962 teaches automatically causing administration of saline or whole blood in response to detection of a predicted future hypotension event (¶[0021]). Gholami teaches an electronically controlled infusion pump for intravenous fluid administration based on patient hemodynamic information (¶[0040], [0074]-[0076]). Using the threshold-satisfying future-hypotension prediction as the treatment trigger and implementing the preventive intervention with Gholami's pump would provide the recited delivery. It would have been obvious to one of ordinary skill in the art to modify Ref claim 6, through claim 1, to implement the preventive intervention using Hatib'962's automatic administration teaching and Gholami's electronically controlled infusion pump.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim 1 is rejected under 35 U.S.C. 103 as being unpatentable over Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes, and further in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, and further in view of Al Hatib et al. (US 2018/0028076 A1), hereinafter Hatib'076, and further in view of Al Hatib et al. (US 2018/0025290 A1), hereinafter Hatib'290, and further in view of Johnson et al. (US 2021/0275783 A1), hereinafter Johnson.
Regarding claim 1, Potes teaches a hemodynamic sensor that produces analog hemodynamic sensor signals representative of arterial pressure signal waveforms of one or more patients (Potes, ¶[0043], wherein hemodynamic sensor 34 provides hemodynamic data representative of the arterial pressure waveform of patient 36 to hemodynamic monitor 10 as an analog signal).
Potes further teaches an analog-to-digital converter that converts the analog hemodynamic sensor signals to arterial pressure signal waveforms (Potes, ¶[0043], [0049], wherein ADC 44 converts analog hemodynamic data to digital hemodynamic data representative of the arterial pressure waveform).
Potes further teaches an electronic hardware processor in communication with the non-transitory memory and configured to execute the instructions to cause the system to at least perform arterial-pressure waveform analysis and physiological-state prediction (Potes, ¶[0041]-[0042], [0046]-[0047]: hemodynamic monitor 10 includes system processor 40 and system memory 42, the memory stores waveform-analysis and physiological-state-prediction software code, processor 40 executes the code, and the memory may be a non-transitory computer-readable storage medium).
Potes further teaches receive, from the hemodynamic sensor, a test analog hemodynamic sensor signal from the patient (Potes, ¶[0049]: the hemodynamic monitor receives analog hemodynamic sensor data representative of the patient arterial-pressure waveform from hemodynamic sensor 34).
Potes further teaches convert, using the analog-to-digital converter, the test analog hemodynamic sensor signal to a test arterial pressure signal waveform (Potes, ¶[0049]: ADC 44 converts the received analog hemodynamic data to digital hemodynamic data representative of the arterial-pressure waveform).
Potes further teaches obtain a plurality of reference arterial pressure signal waveforms from a plurality of patients (Potes, ¶[0007]: the model-training system receives hemodynamic data representing arterial-pressure waveforms of a critical-care patient population and divides patient data into training and validation subsets, wherein those population waveforms provide the recited reference waveforms).
Potes further teaches extract from the plurality of reference arterial pressure signal waveforms a plurality of reference sets of heart health parameters (Potes, ¶[0007], [0030]-[0031]: the patient waveform data are transformed into physiological-state profiling parameters and divided into data segments “for each feature or hemodynamic parameter”, wherein each patient provides a corresponding parameter set).
Also, regarding claim 1, Potes does not fully teach a hemodynamic sensor system configured to determine a likelihood of effectiveness of an intervention for hypotension. Rather, Potes teaches that the physiological-state model can be “tailored to predict responsiveness to fluid delivery” and separately teaches that the same approach can be applied to other physiological states, including a “hypotensive” state (Potes, ¶[0038]). However, Potes does not expressly state that the disclosed fluid-responsiveness prediction is made for an intervention directed to hypotension.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Potes's physiological-state prediction model to apply its disclosed fluid-responsiveness prediction when the physiological state being evaluated is hypotension. The modification would have been technically feasible because Potes expressly identifies both fluid responsiveness and hypotensive states as applications of the same physiological-state model operating on arterial-pressure-derived profiling parameters. A person of ordinary skill would have been motivated to make the modification because Potes expressly contemplates tailoring the model to predict fluid responsiveness and applying the same modeling approach to hypotensive states (Potes, ¶[0038]). This would have constituted application of a known predictive use expressly contemplated by Potes to another expressly contemplated physiological state. A person of ordinary skill would have had a reasonable expectation of success because the modification would not require a different sensor signal or a different patient-comparison framework. The resulting Potes system would determine a predicted likelihood that a fluid intervention for hypotension will be effective for the patient.
Also, regarding claim 1, the modified Potes does not fully teach a graphical user interface configured to display an alert indicating the determined likelihood of effectiveness of the intervention. Rather, Potes teaches a graphical user interface that displays predicted physiological-state probabilities and teaches a sensory alarm that may include a graphical alert and display of a risk score (Potes, ¶[0042], [0048], [0056], [0058]). However, the modified Potes does not expressly teach configuring that graphical alert to indicate the fluid-responsiveness likelihood established above.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes to present the determined fluid-responsiveness likelihood through Potes's existing graphical display and alarm arrangement. The modification would have been technically feasible because the responsiveness likelihood is another numerical prediction generated by the same monitoring system, and Potes already displays predicted probabilities and risk scores through user interface 54. A person of ordinary skill would have been motivated to make the modification because Potes uses the graphical display and sensory alarm to communicate predicted patient-state information to medical personnel for clinical use (Potes, ¶[0056], [0058]). The modification would have constituted use of Potes's known display and alert technique for another prediction generated by the same system and would have provided the recited alert indicating the determined likelihood of effectiveness.
Also, regarding claim 1, the modified Potes does not fully teach a non-transitory memory having executable instructions and a deep learning model stored thereon. Rather, Potes teaches non-transitory system memory 42 storing executable waveform-analysis and physiological-state-prediction software code (Potes, ¶[0042], [0047]). However, Potes does not teach implementing the physiological-state prediction processing with a stored deep-learning model.
Ko teaches a deep-learning implementation for hypotension prediction in which a device has a dedicated program for a deep-learning model and a database stores “the structure of the deep neural network, filter values and activation functions” (Ko, ¶[0020]). Ko further teaches that a 1D CNN takes a one-dimensional vector or matrix as input and performs convolution and pooling, with the convolution operation using a weighted multidimensional matrix and the pooling stage including max pooling or mean pooling (Ko, ¶[0045]). Ko's disclosed 1D CNN includes seven convolution layers, seven max-pooling layers, eight dropout layers, batch normalization, and convolution layers receiving arterial blood-pressure data (Ko, ¶[0047]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes in view of Ko to implement feature processing for Potes's physiological-state prediction model with Ko's stored 1D CNN while retaining Potes's patient-to-reference similarity comparison as the decision framework. The modification would have been technically compatible because both systems numerically process arterial blood-pressure data, and Ko expressly provides arterial blood-pressure data to its convolution layers (Ko, ¶[0047]). Ko teaches convolution and pooling of one-dimensional inputs, including max pooling (Ko, ¶[0045]), while Potes teaches that the selected feature and other similarity-model parameters may be changed or adjusted to improve prediction accuracy (Potes, ¶[0037]). A person of ordinary skill would therefore have been motivated to apply Ko's known CNN feature-processing operations to the arterial-pressure-derived features used by Potes before Potes performs its existing similarity comparison. This would have constituted application of a known feature-processing technique to a related arterial-pressure prediction system ready for improvement. A person of ordinary skill would have had a reasonable expectation of success because the modification uses the same type of numerical arterial-pressure input and software processing and does not replace Potes's similarity-based prediction operation. The resulting system would include executable instructions and a stored deep-learning model in the non-transitory memory.
Also, regarding claim 1, the modified Potes does not fully teach extract from the test arterial pressure signal waveform a plurality of test heart health parameters from a set of available heart health parameters, wherein the set of available heart health parameters comprises: a mean arterial pressure (MAP); a stroke volume index (SVI); a hypotension prediction index (HPI); a systemic vascular resistance (SVR); a heart rate (HR); a cardiac output (CO); a time-based change in arterial pressure; a cardiac index (CI); a systemic vascular resistance index (SVRI); a normalized area of pulse pressure; an average distance between subsequent MAPs; an average distance between a systolic peak and a respective diastolic peak; and a stroke volume variation (SVV). Rather, Potes teaches extracting multiple arterial-pressure-derived physiological-state profiling parameters, including stroke volume, heart rate, MAP, and related hemodynamic features (Potes, ¶[0050]-[0052], [0061]). However, the modified Potes does not teach the complete recited set of available parameters. Claim 1 requires the set of available heart health parameters to include each recited parameter, but does not require every parameter in that available set to be among the plurality extracted in a given determination. Potes distinguishes between the larger set of hemodynamic parameters that can be determined and the subset selected for use, stating that “not all hemodynamic parameters determined are necessarily selected as physiological state profiling parameters 53” (Potes, ¶[0051]).
Hatib'076 teaches tracked health parameters including cardiac output (CO), stroke volume (SV), stroke volume variation (SVV), diastolic pressure (DIA), pulse rate (PR), stroke volume index (SVI), systemic vascular resistance (SVR), mean arterial pressure (MAP), HPI, systemic vascular resistance index (SVRI), cardiac index (CI), and systolic pressure (SYS) (Hatib'076, ¶[0039]). Hatib'076 further teaches that these are health parameters tracked by the health monitoring unit and identifies HPI expressly as one of the available health parameters (Hatib'076, ¶[0038]-[0042]).
Hatib'290 further teaches heart rate as a five-minute averaged parameter (Hatib'290, Appendix A, item 9), stroke volume variation and changes in MAP with respect to time (Hatib'290, ¶[0042]), ppAreaNor as the normalized area under the arterial-pressure waveform (Hatib'290, ¶[0063]), and pulsepres as pulse pressure (Hatib'290, ¶[0073]). Hatib'290 identifies the disclosed arterial-pressure-derived parameters as parameters predictive of future hypotension (Hatib'290, ¶[0034], [0039], [0054]-[0057]). Appendix A also applies 20-second and five-minute averaging criteria to the disclosed hypotension-predictive parameters.
Johnson teaches arterial-pressure variability metrics including “successive variation of mean arterial pressure over a given number of heartbeats or a given time period” (Johnson, ¶[0051]) and teaches pulse pressure as “the difference from the systolic peak to the diastolic trough” (Johnson, ¶[0053]). For purposes of examination, “diastolic peak” is interpreted as the diastolic pressure extremum/minimum. Johnson's diastolic trough therefore corresponds to the recited diastolic pressure extremum. Johnson does not expressly state that the successive MAP changes or per-beat systolic-to-diastolic pressure differences are averaged in the manner recited by claim 1.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes in view of Hatib'076, Hatib'290, and Johnson to include the foregoing known hemodynamic and arterial-pressure-derived measures among the available heart-health parameters processed by Potes's model. The references process hemodynamic parameters associated with arterial-pressure monitoring and prediction. Potes expressly teaches that feature selection can be changed or adjusted to improve prediction accuracy (Potes, ¶[0037]); Hatib'076 identifies HPI and the recited core hemodynamic measures as health parameters tracked in a hypotension-predictive system; and Hatib'290 identifies its additional arterial-pressure parameters as parameters predictive of future hypotension. Johnson identifies successive MAP variation over multiple heartbeats or a time period as a blood-pressure variability measure. A person of ordinary skill would have been motivated to calculate the absolute pressure-value difference between each pair of successive MAP values and average those absolute differences over Johnson's disclosed multi-heartbeat or time window to obtain a representative successive-MAP variability measure rather than rely on a single pair of MAP values. For the recited systolic-to-diastolic distance, Johnson identifies the underlying systolic-peak-to-diastolic-extremum pressure difference, while Hatib'290 establishes 20-second and five-minute averaging as sampling criteria for hypotension-predictive arterial-pressure parameters. A person of ordinary skill would therefore likewise have been motivated to average the per-beat systolic-to-diastolic pressure difference over such a sampling window to obtain a representative pressure-distance feature rather than rely on a single heartbeat. The modification would have constituted use and selection of known hypotension-related hemodynamic, arterial-pressure predictive, and variability features in Potes's existing feature-based model. The resulting available set would include the recited MAP, SVI, HPI, SVR, HR, CO, time-based arterial-pressure change, CI, SVRI, normalized arterial-pressure pulse-area measure, average successive-MAP pressure distance, averaged systolic-to-diastolic pressure distance, and SVV. Because Potes already extracts the same selected feature across its training and reference patients, adding those features to Potes's available feature set would also provide each of the reference sets of heart health parameters comprising corresponding heart health parameters from the set of available heart health parameters.
Also, regarding claim 1, the modified Potes does not fully teach apply a filter to the reference sets of heart health parameters to generate a feature map associated with the filter; consolidate the filtered sets of reference heart health parameters into corresponding reference feature pools; select a reference maximum feature from each of the reference feature pools; reduce overfitting of the reference feature pools by removing one or more of the reference feature pools from remaining reference feature pools; normalize the selected reference maximum features from the reference feature pools; generate, based on the plurality of heart health parameters, one or more feature pools; and determine normalized test maximum features from the plurality of test heart health parameters. Rather, Potes teaches corresponding patient and reference features and a similarity comparison, and the modified Potes includes Ko's CNN feature-processing technique as established above. However, the modified Potes does not expressly apply Ko's convolution, max-pooling, dropout, and normalization operations to produce the recited filtered, pooled, dropout-reduced, and normalized reference and test feature representations.
Ko teaches convolution operations using weighted filters, max pooling, dropout, and batch normalization. Ko teaches that a 1D CNN can perform convolution and pooling on a one-dimensional input, with max pooling included as a disclosed pooling method (Ko, ¶[0045]). Ko further teaches a 1D CNN having seven convolution layers, seven max-pooling layers, and eight dropout layers; max pooling reduces information by selecting the maximum value in a given area; dropout selects only a certain proportion of connections between successive layers to prevent overfitting; dropout layers can be placed in each convolution layer; and batch normalization can be applied to convolution-layer result values (Ko, ¶[0047]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes by applying Ko's CNN feature-processing operations to Potes's corresponding reference-patient and test-patient heart-health-parameter data before Potes performs its similarity comparison. The modification would have been technically feasible because the operations act on numerical feature values derived from the same arterial-pressure data already processed by Potes. Applying Ko's convolution operation to the reference parameter sets would generate filtered feature representations, wherein the result values generated through application of each learned convolution filter provide the filter-associated feature representation corresponding to the recited feature map. Applying max pooling would group the filtered values into pooling regions and select a maximum value from each region.
For purposes of examination, removing one or more of the reference feature pools from remaining reference feature pools is interpreted as excluding selected pooled feature contributions from further processing while retaining the other pooled contributions. The Instant Application describes the disclosed dropout operation as regularizing output data by randomly setting a fraction of neurons to zero so that those neurons are temporarily dropped from the network (Instant Application, ¶[0082]). Under that interpretation, applying Ko's dropout operation to the pooled reference-feature representation would exclude selected pooled feature contributions from further propagation while retaining the remaining contributions, thereby reducing overfitting for the purpose expressly taught by Ko (Ko, ¶[0047]).
For purposes of examination, generate, based on the plurality of heart health parameters, one or more feature pools is interpreted as referring to the plurality of test heart health parameters. Applying Ko's pooling operation to the test heart health parameters would therefore generate the recited test feature pools, and selecting the maxima from those pools would provide the test maximum features.
Ko teaches batch normalization of convolution-layer outputs but does not expressly teach normalizing the selected reference maximum features or the selected test maximum features. A person of ordinary skill would have been motivated to apply Ko's known normalization operation to those selected pooled maxima before Potes's numerical distance comparison so that corresponding reference and test feature values are supplied to the comparison on a common numerical basis rather than allowing differences in feature scale to affect the comparison. The claim does not require the dropout/removal operation to precede normalization merely because the clauses are listed in that order. Even if the listed claim language were interpreted to require the dropout/removal operation before normalization, it would have been obvious to apply Ko's normalization operation to the retained selected reference pooled maxima after dropout. Ko teaches both dropout for reducing overfitting and normalization of numerical CNN feature values within the same processing architecture (Ko, ¶[0047]). A person of ordinary skill would have been motivated to normalize the retained reference values actually supplied to subsequent prediction or comparison processing so that those values are provided on a consistent numerical basis. Arranging the known normalization operation after the dropout operation would additionally have constituted an obvious rearrangement of known processing steps yielding their expected functions, consistent with MPEP §2144.04(IV)(C) and In re Burhans, 154 F.2d 690, 69 USPQ 330 (CCPA 1946). Applying normalization to the selected reference and test maxima would provide the normalized selected reference maximum features and normalized test maximum features. The motivation is further supported by Potes's express teaching that feature selection, the similarity threshold, and the distance metric may be changed or adjusted to improve prediction accuracy (Potes, ¶[0037]).
Also, regarding claim 1, the modified Potes does not fully teach determine, based on a comparison between the normalized selected reference maximum features and the normalized test maximum features, that an intervention for hypotension has been implemented and the likelihood of effectiveness of the intervention for hypotension of the patient. Rather, Potes compares a patient's feature-specific data segments to corresponding reference-patient segments using a distance metric to identify similar reference data (Potes, ¶[0053]-[0054]) and then predicts a probability that the patient state is associated with a discrete state “indicating whether a particular clinical intervention has been administered” (Potes, ¶[0055]). The modified Potes further includes the fluid-responsiveness prediction and the normalized reference and test feature representations established above. However, the modified Potes does not expressly use those normalized maximum-feature representations as the inputs to Potes's distance comparison for both the intervention-state and responsiveness determinations.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes to use the normalized reference and test maximum-feature values produced by the incorporated Ko feature-processing pipeline as the corresponding feature values supplied to Potes's existing distance-based patient-to-reference comparison. The modification would have been technically feasible because the normalized values are processed representations of the same corresponding physiological features that Potes already compares. A person of ordinary skill would have been motivated to use those processed feature values because Potes expressly teaches that the selected feature, similarity threshold, and distance metric can be changed or adjusted to improve prediction accuracy (Potes, ¶[0037]). The modification does not replace Potes's principle of operation: Potes would still compare the current patient's corresponding physiological features with reference-patient features using a distance metric and use the similar reference data to predict a clinical state. Only the numerical representation of the compared features would be changed. Applying that same comparison framework to the fluid-responsiveness prediction established above would provide both the determination that the hypotension intervention has been implemented and the predicted likelihood that the intervention will be effective. A person of ordinary skill would have had a reasonable expectation of success because Potes expressly contemplates both intervention-state prediction and fluid-responsiveness prediction within the same physiological-state modeling framework.
The modified Potes further teaches generate, based on the determined likelihood of effectiveness of the intervention for hypotension of the patient, data for displaying, via the graphical user interface, the alert indicating the likelihood of effectiveness of the intervention for hypotension because, as established above, Potes displays predicted probabilities and risk scores through user interface 54 and the modified Potes presents the determined fluid-responsiveness likelihood through that existing graphical display and alarm arrangement (Potes, ¶[0056], [0058]).
Claims 2-7 are rejected under 35 U.S.C. 103 as being unpatentable over Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes, and further in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko.
Regarding claim 2, Potes teaches a hemodynamic sensor that produces analog hemodynamic sensor signals representative of arterial pressure signal waveforms of one or more patients (Potes, ¶[0043]: “hemodynamic sensor 34 provides the hemodynamic data representative of the arterial pressure waveform of patient 36 to hemodynamic monitor 10 as an analog signal”, wherein the analog arterial-pressure data correspond to the recited analog hemodynamic sensor signals).
Potes further teaches an analog-to-digital converter that converts the analog hemodynamic sensor signals to arterial pressure signal waveforms (Potes, ¶[0043]: the analog signal “is converted by ADC 44 to digital hemodynamic data representative of the arterial pressure waveform”; ¶[0049]: “ADC 44 converts the analog hemodynamic data to digital hemodynamic data representative of the arterial pressure waveform of the patient”, wherein ADC 44 performs the recited conversion).
Potes further teaches an electronic hardware processor in communication with the non-transitory memory and configured to execute the instructions to cause the system to at least perform arterial-pressure waveform analysis and physiological-state prediction (Potes, ¶[0041]: hemodynamic monitor 10 includes “system processor 40, system memory 42”; ¶[0042]: “system memory 42 stores waveform analysis software code 47 and physiological state prediction software code 48”; ¶[0046]: “System processor 40 is configured to execute waveform analysis software code 47” and physiological state prediction software code 48; ¶[0047]: “a computer-readable storage medium can include a non-transitory medium”, wherein processor 40 executes the stored instructions from non-transitory system memory 42).
Potes further teaches receive, from the hemodynamic sensor, a test analog hemodynamic sensor signal from the patient (Potes, ¶[0049]: “Hemodynamic sensor 34 provides the hemodynamic data (e.g., as analog sensor data), to hemodynamic monitor 10”, wherein the hemodynamic data are representative of the arterial-pressure waveform of patient 36).
Potes further teaches convert, using the analog-to-digital converter, the test analog hemodynamic sensor signal to a test arterial pressure signal waveform (Potes, ¶[0049]: “ADC 44 converts the analog hemodynamic data to digital hemodynamic data representative of the arterial pressure waveform of the patient”).
Potes further teaches extract from the test arterial pressure signal waveform a plurality of test heart health parameters from a set of available heart health parameters (Potes, ¶[0050]: “System processor 40 receives the arterial pressure waveform from patient 36 and extracts hemodynamic parameters, as well as differential and combinatorial parameters”; ¶[0051]: “multiple hemodynamic parameters can be determined using waveform analysis software code 47” and “not all hemodynamic parameters determined are necessarily selected as physiological state profiling parameters 53”, wherein Potes distinguishes the available determined parameters from the plurality selected for predictive processing).
Also, regarding claim 2, Potes does not fully teach a hemodynamic sensor system configured to determine a likelihood of effectiveness of an intervention for hypotension. Rather, Potes teaches that the physiological-state model can be “tailored to predict responsiveness to fluid delivery” and separately teaches that the same approach can be applied to other physiological states, including a “hypotensive” state (Potes, ¶[0038]). However, Potes does not expressly state that the disclosed fluid-responsiveness prediction is made for an intervention directed to hypotension.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Potes's physiological-state prediction model to apply its disclosed fluid-responsiveness prediction when the physiological state being evaluated is hypotension. The modification would have been technically feasible because Potes expressly identifies both fluid responsiveness and hypotensive states as applications of the same physiological-state model operating on arterial-pressure-derived profiling parameters. A person of ordinary skill would have been motivated to make the modification because Potes expressly contemplates tailoring the model to predict fluid responsiveness and applying the same modeling approach to hypotensive states (Potes, ¶[0038]). This would have constituted application of a known predictive use expressly contemplated by Potes to another expressly contemplated physiological state. A person of ordinary skill would have had a reasonable expectation of success because the modification would not require a different sensor signal or a different patient-comparison framework. The resulting Potes system would determine a predicted likelihood that a fluid intervention for hypotension will be effective for the patient.
Also, regarding claim 2, the modified Potes does not fully teach a non-transitory memory having executable instructions and a deep learning model stored thereon. Rather, Potes teaches non-transitory system memory 42 storing executable waveform-analysis and physiological-state-prediction software code (Potes, ¶[0042], [0047]). However, Potes does not teach implementing the physiological-state prediction processing with a stored deep-learning model.
Ko teaches a deep-learning implementation for hypotension prediction in which a device has a dedicated program for a deep-learning model and a database stores “the structure of the deep neural network, filter values and activation functions” (Ko, ¶[0020]). Ko further teaches that a 1D CNN takes a one-dimensional vector or matrix as input and performs convolution and pooling, with the convolution operation using a weighted multidimensional matrix and the pooling stage including max pooling or mean pooling (Ko, ¶[0045]). Ko's disclosed 1D CNN includes seven convolution layers, seven max-pooling layers, eight dropout layers, batch normalization, and convolution layers receiving arterial blood-pressure data (Ko, ¶[0047]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes in view of Ko to implement feature processing for Potes's physiological-state prediction model with Ko's stored 1D CNN while retaining Potes's patient-to-reference similarity comparison as the decision framework. The modification would have been technically compatible because both systems numerically process arterial blood-pressure data, and Ko expressly provides arterial blood-pressure data to its convolution layers (Ko, ¶[0047]). Ko teaches convolution and pooling of one-dimensional inputs, including max pooling (Ko, ¶[0045]), while Potes teaches that the selected feature and other similarity-model parameters may be changed or adjusted to improve prediction accuracy (Potes, ¶[0037]). A person of ordinary skill would therefore have been motivated to apply Ko's known CNN feature-processing operations to the arterial-pressure-derived features used by Potes before Potes performs its existing similarity comparison. This would have constituted application of a known feature-processing technique to a related arterial-pressure prediction system ready for improvement. A person of ordinary skill would have had a reasonable expectation of success because the modification uses the same type of numerical arterial-pressure input and software processing and does not replace Potes's similarity-based prediction operation. The resulting system would include executable instructions and a stored deep-learning model in the non-transitory memory.
Also, regarding claim 2, the modified Potes does not fully teach generate, using a plurality of reference arterial pressure signal waveforms from a plurality of patients, one or more filtered sets of reference heart health parameters associated with the plurality of reference arterial pressure signal waveforms; consolidate the filtered sets of reference heart health parameters into corresponding reference feature pools; determine, based on the reference feature pools, normalized test maximum features from the plurality of test heart health parameters. Rather, Potes teaches reference-patient arterial-pressure waveforms that are transformed into physiological-state profiling parameters and stored as feature-specific training and validation patient data segments (Potes, ¶[0007], [0030]-[0031]), but does not teach filtering those reference parameters with a CNN, consolidating the filtered reference parameters into feature pools, or determining normalized test maximum features based on those feature pools.
Ko teaches acquiring “arterial blood pressure data from each of the plurality of patients” for development of the prediction model (Ko, ¶[0039]) and using classified blood-pressure data as correct-answer data while inputting extracted data into the deep neural network (Ko, ¶[0044]). Ko further teaches convolution and max pooling, including selecting the maximum value in a given area, and batch normalization of convolution-layer result values (Ko, ¶[0045], [0047]). Ko repeatedly performs the training operations to machine-learn the network, thereby adjusting the network filter values (Ko, ¶[0049]), and then performs patient prediction using a network that has been machine-learned in advance (Ko, ¶[0050]-[0051]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes by applying Ko's CNN feature-processing operations to Potes's reference-patient and test-patient heart-health-parameter data before Potes performs its similarity comparison. Potes already represents the training and validation patient data as feature-specific physiological parameters, while Ko teaches convolution and pooling of one-dimensional numerical inputs. Applying Ko's weighted convolution to Potes's reference-patient parameter sets would generate filtered sets of reference heart-health parameters, and applying Ko's max pooling to those filtered feature values would consolidate them into corresponding reference feature pools. Those pooled reference-patient representations would participate in training the incorporated CNN, and Ko expressly teaches that repeated training adjusts the filter values subsequently used by the previously machine-learned network (Ko, ¶[0049]-[0051]). Applying Ko's max pooling to the test-patient feature values would group those values into pooling regions and select a maximum value from each region. Ko teaches batch normalization of convolution-layer result values, but does not expressly teach applying normalization to the selected test pooled maxima. A person of ordinary skill would have been motivated to apply Ko's known normalization operation to those selected test maxima before Potes's numerical patient-to-reference comparison so that the test feature values supplied to the comparison are provided on a consistent numerical basis. Applying normalization to the selected test pooled maxima would produce the recited normalized test maximum features. Thus, under the claim interpretation stated above, the test features are determined based on processing parameters learned or derived from the reference feature pools.
A person of ordinary skill would have been motivated to make the modification because Ko provides a known CNN feature-processing technique for multi-patient arterial-pressure data, while Potes expressly teaches that feature selection, the similarity threshold, and the distance metric can be changed or adjusted to improve prediction accuracy (Potes, ¶[0037]; Ko, ¶[0039], [0045], [0047]). The modification would have constituted application of a known feature-processing technique to Potes's existing reference/test arterial-pressure prediction framework. A person of ordinary skill would have had a reasonable expectation of success because the modification changes the numerical representation of the features supplied to Potes's comparison engine but does not change the underlying arterial-pressure data or Potes's patient-to-reference comparison principle.
Also, regarding claim 2, the modified Potes does not fully teach determine, based on the normalized test maximum features, that an intervention for hypotension has been implemented and the likelihood of effectiveness of the intervention for hypotension of the patient. Rather, Potes compares a patient's feature-specific data segments to corresponding reference-patient segments using a distance metric to identify similar reference data (Potes, ¶[0053]-[0054]) and then predicts a probability that the patient state is associated with a discrete state “indicating whether a particular clinical intervention has been administered” (Potes, ¶[0055]). The modified Potes further includes the fluid-responsiveness prediction and the normalized test maximum features established above. However, the modified Potes does not expressly use those normalized test maximum features as the current-patient feature values for both the intervention-state and responsiveness determinations.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes to use the normalized test maximum-feature values produced by the incorporated Ko feature-processing pipeline as the current-patient feature values supplied to Potes's existing distance-based patient-to-reference comparison. The modification would have been technically feasible because the normalized test values are processed representations of the same physiological features that Potes already compares with corresponding reference-patient features. A person of ordinary skill would have been motivated to use those processed feature values because Potes expressly teaches that the selected feature, similarity threshold, and distance metric can be changed or adjusted to improve prediction accuracy (Potes, ¶[0037]). The modification does not replace Potes's principle of operation: Potes would still compare the current patient's corresponding physiological features with reference-patient features using a distance metric and use the similar reference data to predict a clinical state. Only the numerical representation of the current-patient features would be changed. Applying that same comparison framework to the fluid-responsiveness prediction established above would provide both the determination that the hypotension intervention has been implemented and the predicted likelihood that the intervention will be effective based on the normalized test maximum features. A person of ordinary skill would have had a reasonable expectation of success because Potes expressly contemplates both intervention-state prediction and fluid-responsiveness prediction within the same physiological-state modeling framework.
Also, regarding claim 2, the modified Potes does not fully teach generate, based on the determined likelihood of effectiveness of the intervention for hypotension of the patient, data for displaying an alert indicating the likelihood of effectiveness of the intervention for hypotension. Rather, Potes displays predicted physiological-state probabilities associated with clinical interventions (Potes, ¶[0056], [0064]) and teaches sensory alarm 58, which may be a visual alarm and may include “display of the risk score via user interface 54 on display 12” (Potes, ¶[0058]). However, the modified Potes does not expressly generate the displayed alert data based on the fluid-responsiveness likelihood established above.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes to present the determined fluid-responsiveness likelihood through Potes's existing display and alarm arrangement. The modification would have been technically feasible because the responsiveness likelihood is another numerical prediction generated by the same monitoring system, and Potes already displays predicted probabilities and risk scores and invokes sensory alarm 58 through user interface 54. A person of ordinary skill would have been motivated to make the modification because Potes uses its display and alarm arrangement to communicate predicted patient-state information to medical personnel for clinical use (Potes, ¶[0056], [0058], [0064]). The modification would have constituted use of Potes's known display and alert technique for another prediction generated by the same system. The resulting system would generate, based on the determined likelihood of effectiveness of the intervention for hypotension of the patient, data for displaying an alert indicating that likelihood of effectiveness.
Regarding claim 3, the modified Potes teaches wherein generating the one or more filtered sets of reference heart health parameters comprises: obtaining the plurality of reference arterial pressure signal waveforms from the plurality of patients (Potes, ¶[0007]: the model-training system receives hemodynamic data representing arterial-pressure waveforms of a critical-care patient population and divides the patient data into training and validation patient subsets, wherein those population arterial-pressure waveforms provide the recited plurality of reference arterial pressure signal waveforms).
The modified Potes further teaches extracting from the plurality of reference arterial pressure signal waveforms a plurality of reference sets of heart health parameters, each of the reference sets of heart health parameters comprising corresponding heart health parameters from the set of available heart health parameters (Potes, ¶[0030]: the hemodynamic data associated with each validation patient and each training patient are divided into data segments “for each feature or hemodynamic parameter”, wherein corresponding feature-specific parameter data are extracted for the plurality of reference patients). Potes defines the physiological-state profiling parameters as features and permits a single or subset of the determined hemodynamic parameters to be selected for analysis (Potes, ¶[0011], [0030]). Potes further distinguishes the available determined parameters from those selected for predictive processing, stating that “not all hemodynamic parameters determined are necessarily selected as physiological state profiling parameters 53” (Potes, ¶[0051]). Thus, the corresponding parameters extracted for the reference patients are selected from the set of available heart health parameters established regarding claim 2.
The modified Potes further teaches applying a filter to the reference sets of heart health parameters to generate a feature map associated with the filter because the Ko CNN teaching incorporated regarding claim 2 applies learned convolution filters to Potes's reference-patient parameter sets. Ko teaches that a 1D CNN performs convolution on a one-dimensional vector or matrix using “a multidimensional matrix containing weights” (Ko, ¶[0045]) and further teaches convolution layers having 64, 32, or 128 filters of size 1 x 10 and “result values derived from each convolution layer” (Ko, ¶[0047]). The result values produced by application of each learned convolution filter necessarily form the feature representation, or feature map, associated with that filter. Thus, the modified Potes applies a filter to the reference sets of heart health parameters and generates a feature map associated with the applied filter.
Regarding claim 4, the modified Potes teaches wherein determining the normalized test maximum features from the plurality of test heart health parameters comprises: selecting a reference maximum feature from each of the reference feature pools because, as discussed regarding claim 2, the incorporated Ko CNN applies max pooling to the filtered reference feature values. Ko teaches that “max pooling layers can reduce the size of information by selecting the maximum value in a given area” (Ko, ¶[0047]). Thus, for each pooling region corresponding to a reference feature pool, the incorporated max-pooling operation selects the maximum feature value from that pool.
The modified Potes further teaches generating, based on the plurality of heart health parameters, one or more feature pools because the same max-pooling processing incorporated regarding claim 2 is applied to the test-patient heart-health parameters to group the test feature values into pooling regions before selection of the test maximum features. Under the test-parameter interpretation applied for examination, those pooling regions provide the recited feature pools generated from the plurality of test heart health parameters.
Also, regarding claim 4, the modified Potes does not fully teach reducing overfitting of the reference feature pools by removing one or more of the reference feature pools from remaining reference feature pools; and normalizing the selected reference maximum features from the reference feature pools. Rather, the modified Potes via Ko teaches dropout that selects only a proportion of connections between layers “to prevent overfitting” and teaches batch normalization of convolution-layer result values (Ko, ¶[0047]). The Instant Application's ¶[0082] supports treating dropout as the disclosed mechanism for excluding selected neural contributions from further processing. Under the interpretation stated above, Ko's dropout teaching therefore addresses the recited removal operation when applied to the pooled reference-feature representation because it prevents selected pooled feature contributions from propagating while retaining the remaining contributions, thereby reducing overfitting. Ko does not, however, expressly normalize the selected reference maximum features. It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes by applying Ko's dropout regularization to the pooled reference-feature representation. Under the interpretation stated above, removing one or more of the reference feature pools from remaining reference feature pools is interpreted as excluding selected pooled feature contributions from further processing while retaining the other pooled contributions. Ko's dropout operation would provide that result by preventing selected pooled feature contributions from propagating to the next processing layer while retaining the remaining contributions, thereby reducing overfitting for the purpose expressly taught by Ko.
A person of ordinary skill would further have been motivated to apply Ko's normalization operation to the selected reference maximum features before Potes's numerical distance comparison so that the compared reference feature values are supplied on a consistent numerical basis rather than allowing differences in feature scale to affect the comparison. Ko already teaches batch normalization as part of the same CNN processing architecture. The claim does not require the dropout/removal operation to precede normalization. Even if the listed claim language were interpreted to require the dropout/removal operation before normalization, it would have been obvious to apply Ko's normalization operation to the retained selected reference pooled maxima after dropout. Ko teaches both dropout for reducing overfitting and normalization of numerical CNN feature values within the same processing architecture (Ko, ¶[0047]). A person of ordinary skill would have been motivated to normalize the retained reference values actually supplied to subsequent prediction or comparison processing so that those values are provided on a consistent numerical basis. Arranging the known normalization operation after the dropout operation would additionally have constituted an obvious rearrangement of known processing steps yielding their expected functions, consistent with MPEP §2144.04(IV)(C) and In re Burhans, 154 F.2d 690, 69 USPQ 330 (CCPA 1946). Applying normalization to the selected pooled maxima would provide normalizing the selected reference maximum features from the reference feature pools.
Regarding claim 5, the modified Potes teaches a graphical user interface configured to display the alert indicating the determined likelihood of effectiveness of the intervention (Potes, ¶[0048]: “user interface 54 can take the form of a graphical user interface (GUI)”; ¶[0058]: sensory alarm 58 may be a visual alarm and may include “display of the risk score via user interface 54 on display 12”, wherein the graphical user interface displays the visual alert and associated predicted value). As discussed regarding claim 2, the modified Potes presents the determined fluid-responsiveness likelihood through this existing display and alarm arrangement. Accordingly, the graphical user interface of the modified Potes displays the alert indicating the determined likelihood of effectiveness of the intervention.
Regarding claim 6, the modified Potes teaches wherein the electronic hardware processor is further configured to execute the instructions to cause the system to at least: generate, based on the determined likelihood of effectiveness of the intervention, data for displaying at least the alert indicating the likelihood of effectiveness of the intervention for hypotension, as discussed regarding claim 2. In particular, Potes teaches sensory alarm 58 including “display of the risk score via user interface 54 on display 12” (Potes, ¶[0058]), and the modification regarding claim 2 applies that existing display and alarm arrangement to the determined likelihood of effectiveness of the hypotension intervention. Thus, the processor of the modified Potes generates, based on that determined likelihood of effectiveness, data for displaying the recited alert.
Regarding claim 7, Potes further teaches wherein the alert comprises at least one of a symbol, a numerical value, a visual design, a repeated indicator, or a highlight (Potes, ¶[0058]: sensory alarm 58 may be invoked as “flashing and/or colored graphics” shown by user interface 54 on display 12 and may include “display of the risk score via user interface 54 on display 12”, wherein the flashing and colored graphics provide the recited visual design and the displayed risk score provides the recited numerical value).
Claims 8-9 are rejected under 35 U.S.C. 103 as being unpatentable over Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes, and further in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, and further in view of Al Hatib et al. (US 2018/0008205 A1), hereinafter Hatib'205.
The modified Potes teaches the limitations of claim 2 as discussed above.
Regarding claim 8, the modified Potes does not fully teach wherein the electronic hardware processor is further configured to execute the instructions to cause the system to at least: determine, based on the normalized test maximum features, that the patient will experience hypotension within a specified time with at least a target threshold confidence level. Rather, as discussed regarding claim 2, the modified Potes generates normalized test maximum features from the patient's arterial-pressure-derived parameters using the incorporated Ko CNN processing. However, the modified Potes does not fully teach using those normalized test maximum features to determine that the patient will experience hypotension within a specified future time when the predicted probability satisfies a target confidence threshold.
Ko teaches calculating a patient's hypotension prediction value through a deep neural network and teaches that the predicted value may indicate “the probability or occurrence of hypotension in the patient after n minutes”, that the prediction may be expressed as “the probability of hypotension occurring in a certain percentage”, and that “n minutes later” may be set to various times (Ko, ¶[0052]). Ko further teaches determining whether hypotension will occur based on the hypotension prediction value (Ko, ¶[0053]).
Hatib'205 teaches determining a risk score “corresponding to the probability of a future hypotension event” and invoking an alarm when the risk score satisfies a predetermined risk criterion (Hatib'205, ¶[0021]). Hatib'205 further teaches that the predicted hypotension event may occur “approximately one to five minutes in the future, or up to approximately thirty minutes in the future” (Hatib'205, ¶[0021]). Hatib'205 additionally teaches converting the risk score to a percentage between zero and one hundred and expressly teaches that when the risk score exceeds “a threshold of eighty-five percent”, the predetermined risk criterion may be satisfied (Hatib'205, ¶[0074]-[0076]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes to use Ko's future-hypotension prediction based on the normalized test maximum features and to apply Hatib'205's predetermined percentage risk criterion to that prediction. The modification would have been technically feasible because Ko already produces a numerical probability of hypotension occurring after a selectable future time from the incorporated deep-learning processing, and Hatib'205 applies a numerical probability threshold to the same type of future-hypotension prediction. A person of ordinary skill would have been motivated to apply the threshold so that the system determines that the patient will experience hypotension only when the predicted future-hypotension probability reaches a selected confidence level warranting the determination. Hatib'205 expressly teaches that applying the predetermined risk criterion provides an early warning of a future hypotension event and enables timely intervention. A person of ordinary skill would have had a reasonable expectation of success because the modification applies a known numerical decision criterion to the numerical future-hypotension probability already generated by the modified system. The resulting system would determine, based on the normalized test maximum features, that the patient will experience hypotension within the specified prediction time when the predicted probability meets the target threshold confidence level.
Regarding claim 9, the modified Potes does not fully teach wherein the electronic hardware processor is further configured to execute the instructions to cause the system to at least: generate, based on the determination that the patient will experience hypotension within the specified time with at least the target threshold confidence level, data for displaying, via a graphical user interface, the alert indicating an expected time to event of hypotension within the patient. Rather, as discussed regarding claim 8, the modified Potes determines that the patient will experience hypotension within the specified future time when the predicted probability satisfies the target threshold confidence level. Potes further teaches user interface 54 implemented as a graphical user interface for presenting information on display 12 (Potes ¶[0048]). However, the modified Potes does not expressly teach displaying through that graphical user interface an alert indicating the expected time to the predicted hypotension event.
Ko teaches determining whether hypotension will occur based on the hypotension prediction value calculated by the deep-learning unit and notifying medical staff of “the occurrence of hypotension after n minutes through a display” (Ko, ¶[0053]). Ko further teaches that the prediction time may be set to various times (Ko, ¶[0052]). Thus, Ko expressly presents the future time associated with the predicted hypotension event through a display.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes to display Ko's expected time to the predicted hypotension event through Potes's existing graphical user interface. The modification would have been technically feasible because the modified Potes already determines the future-hypotension prediction and associated prediction time as discussed regarding claim 8, and Potes already provides a graphical user interface for displaying patient information. A person of ordinary skill would have been motivated to display the expected time to the predicted event because Ko expressly teaches notifying medical personnel that hypotension is expected “after n minutes”, thereby providing the timing information needed to respond to the predicted event. The resulting system would generate, based on the future-hypotension determination, data for displaying through the graphical user interface an alert indicating the expected time to the hypotension event.
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes, and further in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, and further in view of Al Hatib et al. (US 2018/0008205 A1), hereinafter Hatib'205, and further in view of Gholami et al. (US 2020/0114077 A1), hereinafter Gholami.
The modified Potes teaches the limitations of claim 2 and 8 as discussed above.
Regarding claim 10, the modified Potes does not fully teach further comprising an infusion pump, wherein the electronic hardware processor is further configured to execute the instructions to cause the system to at least: generate a control signal for the infusion pump to deliver an intravenous therapy to the patient based on the likelihood of effectiveness of the intervention for hypotension of the patient. Rather, as discussed regarding claim 2, the modified Potes determines a predicted likelihood that a fluid intervention for hypotension will be effective for the patient. However, the modified Potes does not teach using that predicted likelihood to control an infusion pump to deliver the therapy.
Gholami teaches a closed-loop fluid-resuscitation and cardiovascular-drug-administration system that receives hemodynamic measurements and computes appropriate infusion rates for administration through infusion pumps (Gholami, Abstract; ¶[0043]). Gholami teaches that fluid resuscitation is “intravenous administration of fluids” (Gholami, ¶[0040]) and that the fluid-resuscitation system uses static and dynamic indicators of fluid responsiveness to compute fluid-infusion rates (Gholami, ¶[0043]). Gholami further teaches that the fluid-resuscitation controller regulates a measure of fluid responsiveness to a desired value by adjusting the fluid infusion rate (Gholami, ¶[0068]). The higher-level controller sends a command to an infusion pump, which administers the fluid to the patient (Gholami, ¶[0075]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes to include Gholami's infusion pump and controller arrangement and to use Potes's predicted likelihood of responsiveness to fluid delivery in generating the command that controls delivery of the fluid therapy. The modification would have been technically feasible because Potes provides a predictive assessment of whether the patient will respond to fluid delivery, while Gholami uses patient-specific fluid-responsiveness information to determine whether and how fluid should be delivered by an infusion pump. Gholami expressly teaches using fluid-responsiveness measures in its fluid-resuscitation control framework and adjusting the infusion rate to regulate fluid responsiveness toward a desired value.
A person of ordinary skill would have been motivated to use Potes's predicted fluid-responsiveness likelihood as an additional decision input to Gholami's pump controller because the prediction directly indicates the expected benefit of the same fluid intervention that Gholami automatically administers. Doing so would permit the controller to issue the pump command when the patient's processed hemodynamic data indicate that fluid delivery is expected to be effective, consistent with Gholami's purpose of adapting fluid administration to the patient's fluid-responsiveness state and needs. Gholami also expressly teaches that its controller sends commands to an infusion pump that administers fluid to the patient. A person of ordinary skill would have had a reasonable expectation of success because the modification uses an additional hemodynamic-derived predictive quantity to inform the same software-controlled infusion decision that Gholami already performs. The resulting system would generate a control signal for the infusion pump to deliver intravenous fluid therapy based on the determined likelihood of effectiveness of the intervention for hypotension.
Claims 11-14 are rejected under 35 U.S.C. 103 as being unpatentable over Buddi et al. (WO 2022/164910 A1), hereinafter Buddi'910, and further in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko.
Regarding claim 11, Buddi'910 teaches a hemodynamic sensor system configured to determine a likelihood that a patient will experience hypotension within a specified time with at least a target threshold confidence level (Buddi'910, ¶[0020], [0041], [0044]: the post-induction predictive risk model determines a post-induction score representing the likelihood that the patient will experience a hypotensive event within a fifteen-minute period after beginning administration of anesthesia; the score may be normalized between 0 and 1 or 0 and 100, and sensory alarm 58 may be invoked when the score exceeds 0.8 or 80). Buddi'910 therefore teaches prediction of hypotension within a defined future period at a threshold likelihood, although its disclosed fifteen-minute period is measured from the beginning of anesthesia administration rather than received as a selectable prediction time.
Buddi'910 further teaches a hemodynamic sensor that produces analog hemodynamic sensor signals representative of arterial pressure signal waveforms of patients (Buddi'910, ¶[0028]-[0029], [0040]: the hemodynamic sensor senses data representative of the patient's arterial-pressure waveform and provides analog sensor data to hemodynamic monitor 10).
Buddi'910 further teaches an analog-to-digital converter that converts the analog hemodynamic sensor signals to arterial pressure signal waveforms (Buddi'910, ¶[0040]: ADC 44 converts the analog hemodynamic data to digital hemodynamic data representative of the arterial-pressure waveform).
Buddi'910 further teaches an electronic hardware processor in communication with the non-transitory memory and configured to execute the instructions to cause the system to at least perform arterial-pressure waveform analysis and future-hypotension prediction (Buddi'910, ¶[0022], [0037]-[0038], [0041]: hemodynamic monitor 10 includes system processor 40 and system memory 42, the memory stores post-induction hypotension prediction software code 48, processor 40 executes the code, and the memory may be a non-transitory computer-readable storage medium).
Buddi'910 further teaches receive, from the hemodynamic sensor, a test analog hemodynamic sensor signal from the patient (Buddi'910, ¶[0040]: hemodynamic sensor 34 provides analog hemodynamic sensor data representative of the patient's arterial-pressure waveform to hemodynamic monitor 10).
Buddi'910 further teaches convert, using the analog-to-digital converter, the test analog hemodynamic sensor signal to a test arterial pressure signal waveform (Buddi'910, ¶[0040]: ADC 44 converts the received analog hemodynamic data to digital hemodynamic data representative of the arterial-pressure waveform).
Buddi'910 further teaches extract from the test arterial pressure signal waveform a plurality of test heart health parameters from a set of available heart health parameters (Buddi'910, ¶[0047]-[0050]: system processor 40 extracts multiple hypotension parameters from arterial pressure waveform 60, and the post-induction risk score may be determined using one, a selected number, or all of the disclosed arterial-pressure-waveform features).
Also, regarding claim 11, Buddi'910 does not fully teach a non-transitory memory having executable instructions and a deep learning model stored thereon. Rather, Buddi'910 teaches non-transitory system memory storing executable post-induction hypotension prediction software and teaches that the predictive risk model may be an artificial neural network model (Buddi'910, ¶[0038], [0057], [0060]). However, Buddi'910 does not expressly teach implementing the prediction processing with the recited stored deep-learning model.
Ko teaches a deep-learning implementation for hypotension prediction in which a device has a dedicated program for a deep-learning model and a database stores “the structure of the deep neural network, filter values and activation functions” (Ko, ¶[0020]). Ko further teaches that a 1D CNN takes a one-dimensional vector or matrix as input and performs convolution and pooling, with the convolution operation using a weighted multidimensional matrix and the pooling stage including max pooling or mean pooling (Ko, ¶[0045]). Ko's disclosed 1D CNN includes seven convolution layers, seven max-pooling layers, eight dropout layers, batch normalization, and convolution layers receiving arterial blood-pressure data (Ko, ¶[0047]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified Buddi'910 in view of Ko to implement feature processing for Buddi'910's future-hypotension prediction model with Ko's stored 1D CNN while retaining Buddi'910's future-hypotension prediction as the decision framework. The modification would have been technically compatible because both systems numerically process arterial blood-pressure data, and Ko expressly provides arterial blood-pressure data to its convolution layers (Ko, ¶[0047]). Ko teaches convolution and pooling of one-dimensional inputs, including max pooling (Ko, ¶[0045]), while Buddi'910 teaches that its predictive risk model may be implemented using machine learning, including an artificial neural network, to identify parameters predictive of future hypotension (Buddi'910, ¶[0020], [0057]). A person of ordinary skill would therefore have been motivated to apply Ko's known CNN feature-processing operations to the arterial-pressure-derived features used by Buddi'910 before Buddi'910 performs its future-hypotension prediction. This would have constituted application of a known feature-processing technique to a related arterial-pressure prediction system ready for improvement. A person of ordinary skill would have had a reasonable expectation of success because the modification uses the same type of numerical arterial-pressure input and software processing and does not replace Buddi'910's future-hypotension prediction operation. The resulting system would include executable instructions and a stored deep-learning model in the non-transitory memory.
Also, regarding claim 11, the modified Buddi'910 does not fully teach receive the specified time. Rather, Buddi'910 uses a defined fifteen-minute prediction period measured from the beginning of anesthesia administration (Buddi'910, ¶[0020], [0044]). Thus, Buddi'910 provides a fixed clinically defined prediction horizon tied to induction, but does not teach receiving a selectable prediction time.
Ko teaches constructing training data for the prediction horizon by extracting input data n minutes before a hypotension event because the learning process is intended to predict whether hypotension will occur after n minutes (Ko, ¶[0041]). Ko gives 5-, 10-, and 15-minute training examples and teaches that other advance intervals may be used (Ko, ¶[0042]-[0043]). Ko further teaches that the resulting hypotension prediction may indicate “the probability or occurrence of hypotension in the patient after n minutes” and that “n minutes later” may be set to various times (Ko, ¶[0052]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Buddi'910 to receive a selected prediction time and use prediction processing trained for that selected advance-warning interval rather than being limited to Buddi'910's fixed fifteen-minute post-induction interval. Buddi'910 already performs future-hypotension prediction over a defined period, while Ko teaches preparing training inputs at selected intervals before the event and using the trained network to predict hypotension for the corresponding n-minute horizon. A person of ordinary skill would have been motivated to provide selectable clinically useful warning intervals, and would have had a reasonable expectation of success because Ko expressly teaches training and prediction for different advance intervals.
Also, regarding claim 11, the modified Buddi'910 does not fully teach generate, using a plurality of reference arterial pressure signal waveforms from a plurality of patients, one or more filtered sets of reference heart health parameters associated with the plurality of reference arterial pressure signal waveforms; consolidate the filtered sets of reference heart health parameters into corresponding reference feature pools; determine, based on the reference feature pools, normalized test maximum features from the plurality of test heart health parameters. Rather, Buddi'910 teaches receiving hemodynamic data including arterial-pressure waveforms from populations of positive and negative subjects, defining those data as training sets, extracting waveform features from the arterial-pressure waveforms, and transforming those waveform features into physiological parameters for training the predictive risk model (Buddi'910, ¶[0051]-[0057]). However, Buddi'910 does not teach filtering those reference parameters with a CNN, consolidating the filtered reference parameters into feature pools, or determining normalized test maximum features based on those feature pools.
Ko teaches acquiring “arterial blood pressure data from each of the plurality of patients” for development of the prediction model (Ko, ¶[0039]) and using classified blood-pressure data as correct-answer data while inputting extracted data into the deep neural network (Ko, ¶[0044]). Ko further teaches convolution, max pooling, and batch normalization (Ko, ¶[0045], [0047]). Ko repeatedly performs the training operations to machine-learn the network, thereby adjusting the network filter values (Ko, ¶[0049]), and then performs patient prediction using a network that has been machine-learned in advance (Ko, ¶[0050]-[0051]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Buddi'910 by applying Ko's CNN feature-processing operations to Buddi'910's reference-patient and test-patient heart-health-parameter data before Buddi'910 performs its future-hypotension prediction. Buddi'910 already represents the training-patient data as arterial-pressure-waveform-derived parameters, while Ko teaches convolution and pooling of one-dimensional numerical inputs. Applying Ko's weighted convolution to Buddi'910's reference-patient parameter sets would generate filtered reference feature representations, and applying Ko's max pooling would consolidate them into corresponding reference feature pools. Those pooled reference-patient representations would participate in training the incorporated CNN, and Ko expressly teaches that repeated training adjusts the filter values subsequently used by the previously machine-learned network (Ko, ¶[0049]-[0051]). Applying Ko's max pooling to the test-patient feature values would group those values into pooling regions and select a maximum value from each region. Ko teaches batch normalization of convolution-layer result values, but does not expressly teach applying normalization to the selected test pooled maxima. A person of ordinary skill would have been motivated to apply Ko's known normalization operation to those selected test maxima before the subsequent future-hypotension prediction so that the pooled feature values supplied to the predictive model use the normalization processing taught by Ko's CNN architecture. Applying normalization to the selected test pooled maxima would produce the recited normalized test maximum features.
As set forth in the Claim Interpretations above, determine, based on the reference feature pools, normalized test maximum features from the plurality of test heart health parameters does not require the reference feature pools to be supplied as direct contemporaneous inputs when the test features are determined. In the modified Buddi'910, the pooled reference-patient representations participate in training the CNN and thereby affect the learned filter values. Ko then applies that previously machine-learned network to the test-patient data (Ko, ¶[0049]-[0051]). The resulting normalized test maximum features are therefore determined using processing parameters learned or derived from the reference feature pools.
A person of ordinary skill would have been motivated to make the modification because Ko provides a known CNN feature-processing technique for multi-patient arterial-pressure data, while Buddi'910 expressly trains its future-hypotension model using arterial-pressure-waveform data from populations of positive and negative subjects and machine-learning techniques to minimize prediction error (Buddi'910, ¶[0051]-[0057]; Ko, ¶[0039], [0045], [0047]). The modification would have constituted application of a known feature-processing technique to Buddi'910's existing multi-patient arterial-pressure prediction framework. A person of ordinary skill would have had a reasonable expectation of success because the modification changes the numerical representation of the features supplied to Buddi'910's predictive model but does not change the underlying arterial-pressure data or Buddi'910's future-hypotension prediction principle.
The modified Buddi'910 further teaches determine, based on the normalized test maximum features, the likelihood that the patient will experience hypotension within the specified time with at least the target threshold confidence level because the normalized test maximum features produced by the incorporated Ko CNN processing are supplied as the test-patient feature representations used by Buddi'910's future-hypotension model. Buddi'910 determines a normalized post-induction score representing the likelihood of a future hypotensive event and teaches invoking sensory alarm 58 when the score exceeds 0.8 or 80 (Buddi'910, ¶[0041], [0044]). As modified above, the prediction is performed for the received specified time.
Also, regarding claim 11, the modified Buddi'910 does not fully teach generate, based on the determination of the likelihood that that the patient will experience hypotension within the specified time with at least the target threshold confidence level, data for displaying an alert indicating an expected time to event of hypotension within the patient. Rather, Buddi'910 teaches invoking sensory alarm 58 based on the future-hypotension score and warning of a hypotension event predicted to occur within fifteen minutes, but does not expressly display an expected time to the event.
Ko teaches determining whether hypotension will occur based on the hypotension prediction value and notifying medical staff of “the occurrence of hypotension after n minutes through a display” (Ko, ¶[0053]). Ko further teaches that the n-minute prediction horizon may be set to various times (Ko, ¶[0052]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Buddi'910 to display Ko's expected time to the predicted hypotension event with Buddi'910's existing future-hypotension alert. The modification would have been technically feasible because the modified Buddi'910 already determines the future-hypotension likelihood for the received prediction time and provides an alert when the likelihood satisfies the target threshold. A person of ordinary skill would have been motivated to display the expected time to the predicted event because Ko expressly teaches notifying medical personnel that hypotension is expected “after n minutes”, thereby providing timing information for responding to the predicted event. The resulting system would generate, based on the future-hypotension determination, data for displaying an alert indicating the expected time to the hypotension event.
Regarding claim 12, the modified Buddi'910 teaches wherein generating the one or more filtered sets of reference heart health parameters comprises: obtaining the plurality of reference arterial pressure signal waveforms from the plurality of patients (Buddi'910, ¶[0051]-[0053]: the predictive-risk-model training uses hemodynamic data from positive and negative subject populations, wherein the hemodynamic data include arterial-pressure-waveform data from the subjects).
The modified Buddi'910 further teaches extracting from the plurality of reference arterial pressure signal waveforms a plurality of reference sets of heart health parameters, each of the reference sets of heart health parameters comprising corresponding heart health parameters from the set of available heart health parameters (Buddi'910, ¶[0054]-[0057]: the arterial-pressure-waveform training data are processed to obtain waveform features and corresponding physiological parameters for use in training the predictive risk model). Thus, the reference-patient waveforms provide corresponding parameter data of the types used by the predictive model.
The modified Buddi'910 further teaches applying a filter to the reference sets of heart health parameters to generate a feature map associated with the filter because the Ko CNN teaching incorporated regarding claim 11 applies learned convolution filters to Buddi'910's reference-patient parameter sets. Ko teaches that a 1D CNN performs convolution on a one-dimensional vector or matrix using “a multidimensional matrix containing weights” (Ko, ¶[0045]) and further teaches convolution layers having 64, 32, or 128 filters of size 1 x 10 and “result values derived from each convolution layer” (Ko, ¶[0047]). The result values produced by application of each learned convolution filter necessarily form the feature representation, or feature map, associated with that filter. Thus, the modified Buddi'910 applies a filter to the reference sets of heart health parameters and generates a feature map associated with the applied filter.
Regarding claim 13, the modified Buddi'910 teaches wherein determining the normalized test maximum features from the plurality of test heart health parameters comprises: selecting a reference maximum feature from each of the reference feature pools because, as discussed regarding claim 11, the incorporated Ko CNN applies max pooling to the filtered reference feature values. Ko teaches that “max pooling layers can reduce the size of information by selecting the maximum value in a given area” (Ko, ¶[0047]). Thus, for each pooling region corresponding to a reference feature pool, the incorporated max-pooling operation selects the maximum feature value from that pool.
The modified Buddi'910 further teaches generating, based on the plurality of heart health parameters, one or more feature pools because the same max-pooling processing incorporated regarding claim 11 is applied to the test-patient heart-health parameters to group the test feature values into pooling regions before selection of the test maximum features. Under the test-parameter interpretation applied for examination, those pooling regions provide the recited feature pools generated from the plurality of test heart health parameters.
Also, regarding claim 13, the modified Buddi'910 does not fully teach reducing overfitting of the reference feature pools by removing one or more of the reference feature pools from remaining reference feature pools; and normalizing the selected reference maximum features from the reference feature pools. Rather, the modified Buddi'910 via Ko teaches dropout that selects only a proportion of connections between layers “to prevent overfitting” and teaches batch normalization of convolution-layer result values (Ko, ¶[0047]). The Instant Application's ¶[0082] supports treating dropout as the disclosed mechanism for excluding selected neural contributions from further processing. Under the interpretation stated above, Ko's dropout teaching therefore addresses the recited removal operation when applied to the pooled reference-feature representation because it prevents selected pooled feature contributions from propagating while retaining the remaining contributions, thereby reducing overfitting. Ko does not, however, expressly normalize the selected reference maximum features.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Buddi'910 by applying Ko's dropout regularization to the pooled reference-feature representation. Under the interpretation stated above, removing one or more of the reference feature pools from remaining reference feature pools is interpreted as excluding selected pooled feature contributions from further processing while retaining the other pooled contributions. Applying Ko's dropout operation to the pooled reference-feature representation would provide that result by excluding selected pooled feature contributions from further propagation while retaining the other pooled contributions, thereby reducing overfitting for the purpose expressly taught by Ko.
A person of ordinary skill would further have been motivated to apply Ko's normalization operation to the selected reference maximum features so that the pooled feature values supplied to subsequent CNN and prediction processing use the normalization convention already taught in Ko's CNN architecture. The claim does not require the dropout/removal operation to precede normalization. Even if the listed claim language were interpreted to require the dropout/removal operation before normalization, it would have been obvious to apply Ko's normalization operation to the retained selected reference pooled maxima after dropout. Ko teaches both dropout for reducing overfitting and normalization of numerical CNN feature values within the same processing architecture (Ko, ¶[0047]). A person of ordinary skill would have been motivated to normalize the retained reference values actually supplied to subsequent prediction or comparison processing so that those values are provided on a consistent numerical basis. Arranging the known normalization operation after the dropout operation would additionally have constituted an obvious rearrangement of known processing steps yielding their expected functions, consistent with MPEP §2144.04(IV)(C) and In re Burhans, 154 F.2d 690, 69 USPQ 330 (CCPA 1946). Applying Ko's normalization to the selected pooled maxima would provide normalizing the selected reference maximum features from the reference feature pools.
Regarding claim 14, the modified Buddi'910 teaches further comprising a graphical user interface configured to display the alert indicating an expected time to event of hypotension within the patient because Buddi'910 teaches that user interface 54 “can take the form of a graphical user interface (GUI)” presented at display 12 (Buddi'910, ¶[0039]) and that sensory alarm 58 is invoked through user interface 54 and can be presented as flashing or colored graphics on display 12 (Buddi'910, ¶[0044]). As discussed regarding claim 11, the modified Buddi'910 displays Ko's expected time to the predicted hypotension event with Buddi'910's existing future-hypotension alert. Accordingly, that alert indicating the expected time to the hypotension event is displayed through Buddi'910's graphical user interface.
Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Buddi et al. (WO 2022/164910 A1), hereinafter Buddi'910, and further in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, and further in view of Al Hatib et al. (US 2018/0192962 A1), hereinafter Hatib'962, and further in view of Gholami et al. (US 2020/0114077 A1), hereinafter Gholami.
The modified Buddi'910 teaches the limitations of claim 11 as discussed above.
Regarding claim 15, the modified Buddi'910 does not fully teach further comprising an infusion pump configured to deliver an intravenous therapeutic agent to the patient based on the likelihood that that the patient will experience hypotension within the specified time with at least the target threshold confidence level. Rather, as discussed regarding claim 11, the modified Buddi'910 determines the likelihood that the patient will experience hypotension within the specified future time with at least the target threshold confidence level. However, the modified Buddi'910 does not teach automatically administering an intravenous therapeutic agent in response to that future-hypotension determination or administering the therapeutic agent using an infusion pump.
Hatib'962 teaches identifying a most probable cause of a predicted future hypotension event and recommending a medical intervention for preventing the predicted event, including administration of saline or whole blood when low blood volume is identified (Hatib'962, ¶[0021]). Hatib'962 further teaches that “system 102 may automatically administer the medical intervention for preventing the future hypotension event” and “in response to detecting a future hypotension event”, may transmit a signal causing administration of saline or whole blood when low blood volume is identified (Hatib'962, ¶[0021]). Thus, Hatib'962 teaches automatically causing administration of a therapeutic agent based on the predicted future-hypotension condition.
Gholami teaches that fluid resuscitation is “intravenous administration of fluids” (Gholami, ¶[0040]) and teaches closed-loop fluid-resuscitation systems that use hemodynamic measurements to compute required fluid infusion rates (Gholami, ¶[0043]). Gholami further teaches a fluid-resuscitation controller that adjusts the fluid infusion rate based on a measure of fluid responsiveness (Gholami, ¶[0068]) and teaches sending infusion rates to an infusion pump that administers fluid to the patient (Gholami, ¶[0074]-[0076]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Buddi'910 in view of Hatib'962 and Gholami to automatically deliver intravenous fluid therapy through an infusion pump based on the future-hypotension likelihood determined by the modified Buddi'910. The modification would have been technically feasible because Hatib'962 expressly teaches automatically causing administration of saline or whole blood in response to detection of a predicted future hypotension event, while Gholami teaches the known pump-based implementation for intravenously administering fluid therapy in response to hemodynamic information. A person of ordinary skill would have been motivated to combine these teachings so that the advance hypotension prediction already generated by the modified Buddi'910 could be used to initiate timely fluid therapy intended to prevent or mitigate the predicted hypotensive event, consistent with Hatib'962's stated purpose of preventing the predicted future hypotension event and Gholami's closed-loop administration of fluid based on patient hemodynamics. A person of ordinary skill would have had a reasonable expectation of success because the modification uses Hatib'962's treatment-triggering prediction as an input to the type of electronically controlled fluid-infusion system already taught by Gholami. The resulting system would include an infusion pump configured to deliver an intravenous therapeutic agent to the patient based on the determined likelihood that the patient will experience hypotension within the specified time with at least the target threshold confidence level.
Claim 16 is rejected under 35 U.S.C. 103 as being unpatentable over Buddi et al. (WO 2022/164910 A1), hereinafter Buddi'910, and further in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, and further in view of Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes.
Regarding claim 16, Buddi'910 teaches a hemodynamic sensor system configured to determine a likelihood that a patient will experience hypotension within a specified time with at least a target threshold confidence level (Buddi'910, ¶[0020], [0041], [0044]: the predictive risk model determines a score representing the likelihood that the patient will experience a hypotensive event within a fifteen-minute period after beginning administration of anesthesia; the score may be normalized between 0 and 1 or 0 and 100, and sensory alarm 58 may be invoked when the score exceeds 0.8 or 80). Thus, Buddi'910 predicts hypotension within a defined future period when the predicted likelihood satisfies a threshold, although the disclosed fifteen-minute period is measured from the beginning of anesthesia administration.
Buddi'910 further teaches a hemodynamic sensor that produces analog hemodynamic sensor signals representative of arterial pressure signal waveforms of one or more patients and an analog-to-digital converter that converts the analog hemodynamic sensor signals to arterial pressure signal waveforms (Buddi'910, ¶[0040]: hemodynamic sensor 34 provides analog hemodynamic data representative of the arterial-pressure waveform and ADC 44 converts the analog data to digital hemodynamic data representative of the arterial-pressure waveform).
Buddi'910 further teaches an electronic hardware processor in communication with the non-transitory memory and configured to execute the instructions to cause the system to at least perform arterial-pressure-waveform analysis and future-hypotension prediction (Buddi'910, ¶[0022], [0037]-[0038], [0041]: hemodynamic monitor 10 includes system processor 40 and system memory 42, system memory 42 stores hypotension prediction software code, and processor 40 executes the stored code).
Buddi'910 further teaches receiving, from the hemodynamic sensor, a test analog hemodynamic sensor signal from the patient and converting, using the analog-to-digital converter, the test analog hemodynamic sensor signal to a test arterial pressure signal waveform (Buddi'910, ¶[0040]: hemodynamic sensor 34 supplies analog arterial-pressure-waveform data to the monitor and ADC 44 converts the data to digital arterial-pressure-waveform data).
Buddi'910 further teaches extract from the test arterial pressure signal waveform a plurality of test heart health parameters from a set of available heart health parameters (Buddi'910, ¶[0041], [0047]-[0050]: the hypotension prediction software obtains multiple hypotension profiling parameters from the arterial-pressure waveform for use in determining the future-hypotension risk score).
Buddi'910 further teaches obtain a plurality of reference arterial pressure signal waveforms from a plurality of patients and extract from the plurality of reference arterial pressure signal waveforms a plurality of reference sets of heart health parameters, each of the reference sets of heart health parameters comprising corresponding heart health parameters from the set of available heart health parameters (Buddi'910, ¶[0051]-[0057]: predictive-model training uses arterial-pressure-waveform data from positive and negative subject populations, extracts waveform features from those data, and transforms the waveform features into physiological parameters used to train the predictive model).
Also, regarding claim 16, Buddi'910 does not fully teach a graphical user interface configured to display an alert indicating an expected time to event of hypotension within the patient. Rather, Buddi'910 teaches a graphical user interface and teaches presenting sensory alarm 58 through user interface 54 on display 12, but does not expressly display the expected time to the predicted hypotension event (Buddi'910, ¶[0039], [0044]).
Ko teaches determining whether hypotension will occur based on a hypotension prediction value and notifying medical staff of “the occurrence of hypotension after n minutes through a display” (Ko, ¶[0053]). Ko further teaches that the n-minute prediction horizon may be set to various times (Ko, ¶[0052]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Buddi'910 in view of Ko to display through Buddi'910's existing graphical user interface the expected time to the predicted hypotension event. The modification would have been technically feasible because Buddi'910 already produces a future-hypotension prediction and presents an alert through its graphical user interface, while Ko provides the future-time information associated with the same type of hypotension prediction. A person of ordinary skill would have been motivated to make the modification because Ko expressly communicates that hypotension is expected after a specified number of minutes, thereby providing medical personnel with timing information for responding to the predicted event. The modification would have constituted use of a known display technique to present additional information associated with an existing prediction. The resulting graphical user interface would display an alert indicating the expected time to the predicted hypotension event.
Also, regarding claim 16, the modified Buddi'910 does not fully teach a non-transitory memory having executable instructions and a deep learning model stored thereon. Rather, Buddi'910 teaches non-transitory memory storing executable hypotension-prediction software and teaches that the predictive model may be an artificial neural network, but does not expressly teach the recited stored deep-learning model.
Ko teaches a dedicated deep-learning program and a database storing “the structure of the deep neural network, filter values and activation functions” (Ko, ¶[0020]). Ko further teaches a 1D CNN that performs convolution and pooling on one-dimensional inputs (Ko, ¶[0045]) and includes convolution layers, max-pooling layers, dropout layers, and batch normalization, with arterial blood-pressure data supplied to the convolution layers (Ko, ¶[0047]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Buddi'910 in view of Ko to implement Buddi'910's future-hypotension model using Ko's stored 1D CNN and to apply that CNN processing to Buddi'910's arterial-pressure-derived reference-patient and test-patient parameter data. The modification would have been technically feasible because Buddi'910 already teaches machine-learning and artificial-neural-network implementations for future-hypotension prediction, while Ko expressly applies a CNN to arterial blood-pressure data. A person of ordinary skill would have been motivated to make the modification to use Ko's known deep-learning feature-processing architecture for the same type of arterial-pressure prediction data. The modification would have constituted application of a known machine-learning technique to a similar predictive system ready for improvement. A person of ordinary skill would have had a reasonable expectation of success because both systems operate on numerical arterial-pressure data and use those data to predict hypotension.
Also, regarding claim 16, the modified Buddi'910 does not fully teach receive the specified time. Rather, Buddi'910 uses a defined fifteen-minute prediction period measured from the beginning of anesthesia administration, but does not teach receiving a selectable prediction time (Buddi'910, ¶[0020], [0044]).
Ko teaches constructing training data for a prediction horizon by extracting input data n minutes before a hypotension event because the learning process is intended to predict whether hypotension will occur after n minutes (Ko, ¶[0041]). Ko gives 5-, 10-, and 15-minute training examples and teaches that other advance intervals may be used (Ko, ¶[0042]-[0043]). Ko further teaches that the resulting hypotension prediction may indicate “the probability or occurrence of hypotension in the patient after n minutes” and that “n minutes later” may be set to various times (Ko, ¶[0052]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Buddi'910 in view of Ko to receive a selected prediction time and use prediction processing trained for that selected advance-warning interval. Buddi'910 already predicts hypotension over a defined future period, while Ko teaches preparing training inputs at selected intervals before the event and predicting hypotension for the corresponding n-minute horizon. A person of ordinary skill would have been motivated to permit selection of different clinically useful advance-warning intervals rather than limiting the system to Buddi'910's fixed fifteen-minute post-induction period, with a reasonable expectation of success because Ko expressly trains and operates the prediction network for such different advance intervals.
The modified Buddi'910 further teaches apply a filter to the reference sets of heart health parameters to generate a feature map associated with the filter because the incorporated Ko CNN performs convolution on a one-dimensional vector or matrix using “a multidimensional matrix containing weights” and includes convolution layers having multiple filters and corresponding convolution-layer result values (Ko, ¶[0045], [0047]). The result values generated by application of each convolution filter provide the feature map associated with that filter.
The modified Buddi'910 further teaches consolidate the filtered sets of reference heart health parameters into corresponding reference feature pools and select a reference maximum feature from each of the reference feature pools because Ko teaches max-pooling layers that reduce information by “selecting the maximum value in a given area” (Ko, ¶[0047]). Applying the incorporated max-pooling operation to the filtered reference feature values groups the values into pooling regions and selects a maximum feature value from each region.
The modified Buddi'910 further teaches generate, based on the plurality of heart health parameters, one or more feature pools because the incorporated Ko max-pooling processing is applied to the test-patient parameter data. Under the working interpretation applied for examination that “the plurality of heart health parameters” refers to the plurality of test heart health parameters, the max-pooling regions provide the recited test feature pools, and Ko's max pooling selects a maximum feature value from each pooling region.
Also, regarding claim 16, the modified Buddi'910 does not fully teach determine normalized test maximum features from the plurality of test heart health parameters. Rather, Ko teaches max pooling and batch normalization within the same CNN architecture, but does not expressly teach applying normalization to the selected test pooled maxima (Ko, ¶[0047]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Buddi'910 by applying Ko's known normalization operation to the selected test pooled maxima before subsequent prediction processing. Ko already teaches batch normalization within the incorporated CNN architecture, and applying that normalization to the selected pooled feature values would provide normalized representations for subsequent predictive processing. Applying normalization to the selected test pooled maxima would provide the recited normalized test maximum features.
Also, regarding claim 16, the modified Buddi'910 does not fully teach reduce overfitting of the reference feature pools by removing one or more of the reference feature pools from remaining reference feature pools; and normalize the selected reference maximum features from the reference feature pools. Rather, Ko teaches dropout that selects only a proportion of connections between layers “to prevent overfitting” and batch normalization of convolution-layer result values (Ko, ¶[0047]). The Instant Application's ¶[0082] supports treating dropout as the disclosed mechanism for excluding selected neural contributions from further processing. Under the interpretation stated above, Ko's dropout teaching therefore addresses the recited removal operation when applied to the pooled reference-feature representation because it prevents selected pooled feature contributions from propagating while retaining the remaining contributions, thereby reducing overfitting. Ko does not, however, expressly normalize the selected reference maximum features.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Buddi'910 by applying Ko's dropout regularization to the pooled reference-feature representation and applying Ko's normalization operation to the selected reference maximum features. Under the interpretation stated above, removing one or more of the reference feature pools from remaining reference feature pools is interpreted as excluding selected pooled feature contributions from further processing while retaining the other pooled contributions. Applying Ko's dropout operation to the pooled reference-feature representation would provide that result by excluding selected pooled feature contributions from further propagation while retaining the other pooled contributions, thereby reducing overfitting. A person of ordinary skill would have been motivated to normalize the selected reference maxima before the later patient-to-reference distance comparison so that corresponding feature values are compared on a consistent numerical basis. The claim does not require dropout/removal to precede normalization. Even if the listed claim language were interpreted to require the dropout/removal operation before normalization, it would have been obvious to apply Ko's normalization operation to the retained selected reference pooled maxima after dropout. Ko teaches both dropout for reducing overfitting and normalization of numerical CNN feature values within the same processing architecture (Ko, ¶[0047]). A person of ordinary skill would have been motivated to normalize the retained reference values actually supplied to subsequent prediction or comparison processing so that those values are provided on a consistent numerical basis. Arranging the known normalization operation after the dropout operation would additionally have constituted an obvious rearrangement of known processing steps yielding their expected functions, consistent with MPEP §2144.04(IV)(C) and In re Burhans, 154 F.2d 690, 69 USPQ 330 (CCPA 1946). These operations act on numerical arterial-pressure-derived features already used by the modified system and do not change the underlying prediction objective.
Also, regarding claim 16, the modified Buddi'910 does not fully teach determine, based on a comparison between the normalized selected reference maximum features and the normalized test maximum features, the likelihood that the patient will experience hypotension within the specified time with at least the target threshold confidence level. Rather, the modified Buddi'910 determines the future-hypotension likelihood using the processed arterial-pressure-derived features established above, including normalized reference and test feature representations. However, it does not teach comparing the normalized selected reference maximum features with the normalized test maximum features as part of that determination.
Potes teaches comparing current-patient feature-specific data segments with corresponding reference-patient data using a distance metric to identify similar reference data (Potes, ¶[0053]-[0054]). Potes further teaches that the selected feature, similarity threshold, and distance metric may be changed or adjusted to improve prediction accuracy (Potes, ¶[0037]). Thus, Potes teaches using a current-patient-to-reference-patient feature comparison in physiological-state prediction.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Buddi'910 in view of Potes to use the normalized selected reference maximum features and normalized test maximum features as the corresponding feature values in Potes's patient-to-reference distance comparison and to use the resulting comparison in determining Buddi'910's future-hypotension likelihood. The modification would have been technically feasible because the normalized reference and test maximum-feature values are processed representations of corresponding arterial-pressure-derived physiological features of the type used by both systems. A person of ordinary skill would have been motivated to make the modification because Potes teaches use of patient-to-reference feature similarity for physiological-state prediction and expressly teaches adjustment of the selected feature, similarity threshold, and distance metric to improve prediction accuracy (Potes, ¶[0037]). Buddi'910 likewise uses multi-patient arterial-pressure data to train a model for future-hypotension prediction. The modification would have constituted combining a known patient-to-reference comparison technique with a known future-hypotension prediction model operating on corresponding physiological features. The resulting future-hypotension likelihood would be based on the comparison between the normalized selected reference maximum features and the normalized test maximum features while retaining Buddi'910's specified prediction horizon and threshold criterion. A person of ordinary skill would have had a reasonable expectation of success because the modification supplies additional comparison-derived patient information to the prediction process without changing the underlying arterial-pressure measurements or Buddi'910's future-hypotension prediction objective.
The modified Buddi'910 further teaches generate, based on the determination of the likelihood that that the patient will experience hypotension within the specified time with at least the target threshold confidence level, data for displaying, via the graphical user interface, the alert indicating the expected time to event of hypotension within the patient. Buddi'910 invokes sensory alarm 58 based on the determined future-hypotension risk satisfying the threshold criterion, and as modified above in view of Ko, the graphical alert displays the expected n-minute time to the predicted hypotension event (Buddi'910, ¶[0044]; Ko, ¶[0052]-[0053]).
Claims 17-19 and 21-22 are rejected under 35 U.S.C. 103 as being unpatentable over Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes, and further in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, and further in view of Al Hatib et al. (US 2018/0028076 A1), hereinafter Hatib'076, and further in view of Smith et al. (US 2021/0259629 A1), hereinafter Smith'629.
Regarding claim 17, Potes teaches a hemodynamic sensor system including a hemodynamic sensor that produces analog hemodynamic sensor signals representative of arterial pressure signal waveforms of one or more patients (Potes, ¶[0043]: hemodynamic sensor 34 provides an analog hemodynamic signal representative of the arterial pressure waveform of patient 36).
Potes further teaches an analog-to-digital converter that converts the analog hemodynamic sensor signals to arterial pressure signal waveforms (Potes, ¶[0043], [0049]: ADC 44 converts the analog hemodynamic data to digital hemodynamic data representative of the arterial pressure waveform).
Potes further teaches an electronic hardware processor in communication with the non-transitory memory and configured to execute the instructions to cause the system to at least perform arterial-pressure waveform analysis and physiological-state prediction (Potes, ¶[0041]-[0042], [0046]-[0047]: hemodynamic monitor 10 includes system processor 40 and system memory 42, system memory 42 stores waveform-analysis and physiological-state-prediction software code, processor 40 executes the stored code, and the computer-readable storage medium may be non-transitory).
Potes further teaches receive, from the hemodynamic sensor, a test analog hemodynamic sensor signal from the patient and convert, using the analog-to-digital converter, the test analog hemodynamic sensor signal to a test arterial pressure signal waveform (Potes, ¶[0049]: hemodynamic sensor 34 provides analog hemodynamic sensor data representative of the arterial-pressure waveform and ADC 44 converts the analog data to digital arterial-pressure-waveform data).
Potes further teaches extract from the test arterial pressure signal waveform a plurality of test heart health parameters from a set of available heart health parameters (Potes, ¶[0011], [0030]-[0031], [0050]-[0051]: arterial-pressure-waveform data are transformed into physiological-state profiling parameters or features, and selected hemodynamic parameters are used for physiological-state prediction).
Also, regarding claim 17, Potes does not fully teach a non-transitory memory having executable instructions and a deep learning model stored thereon; generate, using a plurality of reference arterial pressure signal waveforms from a plurality of patients, one or more filtered sets of reference heart health parameters associated with the plurality of reference arterial pressure signal waveforms; consolidate the filtered sets of reference heart health parameters into corresponding reference feature pools; and determine, based on the reference feature pools, normalized test maximum features from the plurality of test heart health parameters. Rather, Potes teaches training and validation using arterial-pressure-waveform data from a plurality of patients, extracting physiological-state profiling parameters from the waveform data, and comparing current-patient feature data with corresponding reference-patient feature data (Potes, ¶[0007], [0030]-[0031], [0053]-[0054]). However, Potes does not teach the recited stored deep-learning model or the recited filtering, pooling, and normalization processing.
Ko teaches a deep-learning program and storage of “the structure of the deep neural network, filter values and activation functions” (Ko, ¶[0020]). Ko further teaches arterial blood-pressure data from a plurality of patients for development of the prediction model (Ko, ¶[0039]), input of extracted data into a deep neural network (Ko, ¶[0044]), and a 1D CNN using convolution, max pooling, and batch normalization (Ko, ¶[0045], [0047]). Ko repeatedly performs the training operations to machine-learn the network, thereby adjusting the filter values (Ko, ¶[0049]), and then performs patient prediction using a network that has been machine-learned in advance (Ko, ¶[0050]-[0051]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Potes in view of Ko to implement Potes's arterial-pressure feature processing with Ko's stored CNN and to apply Ko's convolution, pooling, and normalization operations to Potes's reference-patient and test-patient parameter data. Potes already represents arterial-pressure-waveform data as numerical physiological features, while Ko applies its CNN to numerical arterial blood-pressure data. Applying Ko's convolution filters to the reference-patient parameter sets would generate filtered reference feature representations, and applying Ko's max pooling would consolidate them into corresponding reference feature pools. Those pooled reference-patient representations would participate in training the CNN, and Ko expressly teaches that repeated training adjusts the filter values subsequently used by the previously machine-learned network (Ko, ¶[0049]-[0051]). Applying Ko's max pooling to the test-patient feature values would group those values into pooling regions and select a maximum value from each region. Ko teaches batch normalization of convolution-layer result values, but does not expressly teach applying normalization to the selected test pooled maxima. A person of ordinary skill would have been motivated to apply Ko's known normalization operation to those selected test maxima before Potes's subsequent physiological-state prediction and patient-to-reference comparison so that the processed test feature values are provided on a consistent numerical basis. Applying normalization to the selected test pooled maxima would produce the recited normalized test maximum features.
As set forth in the Claim Interpretations above, determine, based on the reference feature pools, normalized test maximum features from the plurality of test heart health parameters does not require the reference feature pools themselves to be supplied as direct contemporaneous inputs when the test features are determined. In the modified Potes, the pooled reference-patient representations participate in training the CNN and thereby affect the learned filter values. Ko then applies that previously machine-learned network to the test-patient data (Ko, ¶[0049]-[0051]). The normalized test maximum features are therefore determined using processing parameters learned or derived from the reference feature pools.
A person of ordinary skill would have been motivated to make the modification because Ko provides a known CNN feature-processing technique for multi-patient arterial-pressure data, while Potes teaches that feature selection, similarity threshold, and distance metric may be changed or adjusted to improve prediction accuracy (Potes, ¶[0037]; Ko, ¶[0039], [0045], [0047]). The modification would have constituted application of a known feature-processing technique to Potes's existing arterial-pressure physiological-state prediction framework. A person of ordinary skill would have had a reasonable expectation of success because the modification changes the numerical representation of the arterial-pressure-derived features without changing the underlying arterial-pressure data or Potes's physiological-state prediction framework.
Also, regarding claim 17, the modified Potes does not fully teach receive the specified time; determine, based on the normalized test maximum features, the likelihood that the patient will experience hypotension within the specified time with at least the target threshold confidence level. Rather, Potes teaches that its physiological-state modeling approach may be applied to a hypotensive state (Potes, ¶[0038]), but does not teach receiving a selectable future prediction time and determining a future-hypotension likelihood satisfying a target confidence threshold.
Ko teaches constructing training data for a selected advance interval by extracting input data n minutes before a hypotension event because the learning process is intended to predict hypotension after n minutes (Ko, ¶[0041]). Ko gives 5-, 10-, and 15-minute training examples and teaches that other advance intervals may be used (Ko, ¶[0042]-[0043]). Ko further teaches calculating a hypotension prediction value through the trained deep neural network, wherein the prediction may indicate “the probability or occurrence of hypotension in the patient after n minutes”, may be expressed as a percentage probability, and may use different values of n (Ko, ¶[0052]). Ko also teaches determining whether hypotension will occur based on the prediction value (Ko, ¶[0053]).
Hatib'076 teaches determining an HPI from health parameters and invoking a sensory alarm when the HPI satisfies predetermined risk criteria (Hatib'076, ¶[0025]). Hatib'076 further teaches that the associated future hypotension event may occur approximately one to five minutes in the future or up to approximately thirty minutes in the future (Hatib'076, ¶[0026]). Hatib'076 expressly identifies element 484 as “HPI 484” among the health parameters tracked by the monitoring system (Hatib'076, ¶[0039]) and further explains that the HPI acronym and the probability notation used to depict element 484 on the graphical user interface may be used interchangeably (Hatib'076, ¶[0042]). Hatib'076 teaches expressing that future-hypotension probability as a numerical value between zero and one hundred, wherein a value near one hundred indicates a high probability of an impending hypotension event, and teaches that exceeding a predetermined threshold of 85 may cause an alarm screen to appear substantially immediately (Hatib'076, ¶[0044], [0046]-[0047]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes in view of Ko and Hatib'076 to receive a selected prediction time, use prediction processing trained for that selected advance interval to determine a future-hypotension probability, and apply Hatib'076's predetermined percentage criterion to that probability. Ko expressly teaches both preparing training inputs at different advance intervals and producing an n-minute future-hypotension probability from the trained network, while Hatib'076 applies a numerical probability threshold to the same type of future-hypotension prediction. A person of ordinary skill would have been motivated to permit selection of clinically useful warning intervals and to apply the threshold so that a future event is identified when the probability reaches a level warranting clinical attention. The resulting system would determine, based on the normalized test maximum features, the likelihood that the patient will experience hypotension within the received specified time with at least the target threshold confidence level.
Also, regarding claim 17, the modified Potes does not fully teach determining, based on the normalized test maximum features, that an intervention for hypotension has been implemented and the likelihood of effectiveness of the intervention for hypotension of the patient. Rather, Potes compares the patient's feature-specific data with corresponding reference-patient data and predicts a probability that the patient's physiological state is associated with a discrete state “indicating whether a particular clinical intervention has been administered” (Potes, ¶[0053]-[0055]). Potes also teaches that its physiological-state model may be tailored to predict responsiveness to fluid delivery and may be applied to hypotensive states (Potes, ¶[0038]). However, the modified Potes does not expressly determine the recited likelihood of effectiveness of a hypotension intervention using the normalized test maximum features.
Smith'629 teaches determining from arterial-pressure-derived hemodynamic parameters a score that is “predictive of responsiveness of the patient to a therapy” (Smith'629, ¶[0046]-[0051]). Smith'629 expressly teaches numerical fluid, vasopressor, and inotrope therapy scores predictive of responsiveness to the corresponding therapies (Smith'629, ¶[0052]-[0053], [0081]-[0082]). The disclosed scores indicate the future responsiveness of the patient to the corresponding therapy and may range from zero to one hundred, with the numerical value corresponding to predicted responsiveness (Smith'629, ¶[0051]-[0053]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes in view of Smith'629 to express the responsiveness prediction for the intervention identified by Potes as having been administered as a numerical likelihood of effectiveness corresponding to that intervention. As established above, the modified Potes uses the normalized test maximum features as the processed patient-feature representation in its physiological-state prediction framework, and Potes expressly teaches tailoring that physiological-state model to predict responsiveness to fluid delivery (Potes, ¶[0038]). Thus, the modified Potes would use the normalized test maximum features in determining predicted responsiveness to the fluid intervention. Smith'629 further teaches expressing predicted future responsiveness to a corresponding therapy as a numerical score, including a fluid-therapy responsiveness score ranging from zero to one hundred (Smith'629, ¶[0051]-[0053], [0081]-[0082]). A person of ordinary skill would have been motivated to use Smith'629's numerical responsiveness-score format for Potes's fluid-responsiveness prediction because Smith'629 teaches that the score provides decision support for timely and effective treatment and indicates the patient's future responsiveness to the corresponding therapy. The resulting system would determine, based on the normalized test maximum features, that the hypotension intervention has been implemented and a numerical likelihood that the same intervention will be effective.
Also, regarding claim 17, the modified Potes does not fully teach determining a hypotension prediction index (HPI) based on the determination of the likelihood that the patient will experience hypotension within the specified time with at least the target threshold confidence level and based on the likelihood of effectiveness of the intervention for hypotension of the patient. Rather, the modified Potes provides the future-hypotension likelihood and intervention-effectiveness likelihood established above, but does not expressly use both determinations to produce an HPI.
Potes teaches that the predicted probability that a physiological state is associated with a clinical intervention may be supplied to another prediction algorithm and “utilized as a factor” in determining whether sensed hemodynamic data indicate another current or future physiological event (Potes, ¶[0057]). Potes's disclosed example applies this architecture to a nociception prediction, wherein the probability associated with the clinical intervention affects the probability of current or future nociception. Thus, the disclosed endpoint is nociception, but the teaching relied upon is the use of one intervention-related physiological-state probability as an input or factor affecting another current or future-event prediction.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes to use the determined likelihood of effectiveness of the same hypotension intervention determined to have been implemented as a factor affecting the future-hypotension likelihood and thereby determine an HPI based on both determinations. Potes expressly teaches the cross-model architecture of using one intervention-related physiological-state probability as a factor affecting another current or future-event probability, while Smith'629 teaches a numerical prediction of the patient's future responsiveness to the corresponding therapy. Accounting for the expected effectiveness of the implemented intervention in the continuing future-hypotension risk would permit the risk estimate to reflect whether that intervention is expected to improve the patient's hemodynamic condition. Under the interpretation applied for examination, the resulting adjusted future-hypotension risk constitutes the recited HPI because it depends on both the future-hypotension determination and the likelihood of effectiveness of the implemented intervention.
Also, regarding claim 17, the modified Potes does not fully teach determining that the HPI exceeds a predetermined threshold; and generate, based on the determination that the HPI exceeds the predetermined threshold, an alert for display via a graphical user interface. Rather, the modified Potes determines the HPI established above and Potes teaches a graphical user interface and sensory alarm for communicating predicted patient-state information (Potes, ¶[0048], [0056], [0058], [0064]), but does not expressly apply a predetermined threshold to that HPI.
Hatib'076 teaches determining an HPI from hemodynamic health parameters and teaches that the system processor executes a hypotension predictive graphical user interface to invoke a sensory alarm when the HPI satisfies predetermined risk criteria (Hatib'076, ¶[0025]). Hatib'076 further teaches overlaying an alarm screen as a pop-up on the display when the HPI satisfies the predetermined risk criteria (Hatib'076, ¶[0026]). Hatib'076 also teaches that the predetermined risk criteria may include the HPI exceeding a threshold of 85, which may cause the alarm screen to pop up substantially immediately (Hatib'076, ¶[0046]-[0047]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes to apply Hatib'076's predetermined HPI risk criterion to the HPI established above and to generate Hatib'076's graphical alert when that HPI exceeds the criterion. The modification would have been technically feasible because the HPI established above is a numerical future-hypotension risk quantity and Hatib'076 expressly applies a predetermined risk criterion to an HPI representing future-hypotension probability. A person of ordinary skill would have been motivated to make the modification to provide the known HPI-based warning taught by Hatib'076 when the future-hypotension risk reaches a level warranting clinical attention. A person of ordinary skill would have had a reasonable expectation of success because the modification applies Hatib'076's known numerical HPI threshold and graphical alarm arrangement to the numerical future-hypotension index already generated by the modified system. The resulting system would determine that the HPI exceeds a predetermined threshold and based on that determination, generate an alert for display via a graphical user interface.
Regarding claim 18, the modified Potes does not fully teach that the graphical user interface is configured to display the alert indicating at least one of an expected time to event of hypotension within the patient or an indication of effectiveness of an intervention. Rather, as discussed regarding claim 17, the modified Potes includes a graphical user interface configured to display the alert when the HPI exceeds the predetermined threshold. However, the modified Potes does not expressly configure that alert to indicate the expected time to the predicted hypotension event or the effectiveness of the intervention.
Ko teaches determining whether hypotension will occur based on the hypotension prediction value and notifying medical staff of “the occurrence of hypotension after n minutes through a display” (Ko, ¶[0053]). Ko further teaches that the n-minute prediction horizon may be set to various times (Ko, ¶[0052]). Thus, Ko teaches displaying the expected time to the predicted hypotension event.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes to configure the graphical user interface alert established regarding claim 17 to display Ko's expected time to the predicted hypotension event. The modification would have been technically feasible because the modified Potes already determines the future-hypotension likelihood for the selected prediction time and already generates a graphical alert when the HPI exceeds the predetermined threshold. A person of ordinary skill would have been motivated to include the expected time in the alert because Ko expressly teaches communicating that hypotension is expected “after n minutes”, thereby providing medical personnel with timing information concerning the predicted event. The modification would have constituted displaying an already-determined prediction parameter through the existing graphical alert rather than changing the underlying prediction process. A person of ordinary skill would have had a reasonable expectation of success because the modification merely adds the already-generated expected-time information to the existing graphical display. The resulting graphical user interface would display the alert indicating the expected time to the hypotension event, thereby satisfying the recited alternative.
Regarding claim 19, the modified Potes further teaches wherein the alert is displayed via the graphical user interface because Ko expressly teaches notifying medical staff of the predicted occurrence of hypotension after n minutes “through a display” (Ko, ¶[0053]). Hatib'076, as incorporated regarding claim 17, further teaches executing a hypotension predictive graphical user interface and overlaying the alarm screen as a pop-up when the HPI satisfies the predetermined risk criteria (Hatib'076, ¶[0025]-[0026]). As established regarding claim 18, the expected-time information is included in that alert. Thus, the expected-time alert is displayed via the graphical user interface.
Regarding claim 21, the modified Potes teaches wherein generating the one or more filtered sets of reference heart health parameters comprises: obtaining the plurality of reference arterial pressure signal waveforms from the plurality of patients (Potes, ¶[0007]: the model-training system receives hemodynamic data representing arterial-pressure waveforms from a critical-care patient population and divides the patient data into training and validation patient subsets).
The modified Potes further teaches extracting from the plurality of reference arterial pressure signal waveforms a plurality of reference sets of heart health parameters, each of the reference sets of heart health parameters comprising corresponding heart health parameters from the set of available heart health parameters (Potes, ¶[0030]: the hemodynamic data associated with each validation patient and each training patient are divided into data segments “for each feature or hemodynamic parameter”, thereby providing corresponding feature-specific parameter data for the plurality of reference patients). Potes further teaches that the determined hemodynamic parameters form a larger available set from which selected physiological-state profiling parameters may be used for prediction, stating that “not all hemodynamic parameters determined are necessarily selected as physiological state profiling parameters 53” (Potes, ¶[0051]).
The modified Potes further teaches applying a filter to the reference sets of heart health parameters to generate a feature map associated with the filter because, as discussed regarding claim 17, Ko's CNN convolution processing is applied to Potes's reference-patient parameter data. Ko teaches performing convolution on a one-dimensional vector or matrix using “a multidimensional matrix containing weights” (Ko, ¶[0045]) and teaches convolution layers having 64, 32, or 128 filters of size 1 x 10 and corresponding “result values derived from each convolution layer” (Ko, ¶[0047]). The result values generated through application of each convolution filter provide the filter-associated feature representation corresponding to the recited feature map.
Regarding claim 22, the modified Potes further teaches wherein determining the normalized test maximum features from the plurality of test heart health parameters comprises: selecting a reference maximum feature from each of the reference feature pools because the Ko CNN processing incorporated regarding claim 17 applies max pooling to the filtered reference feature values. Ko teaches that “max pooling layers can reduce the size of information by selecting the maximum value in a given area” (Ko, ¶[0047]). Thus, for each pooling region corresponding to a reference feature pool, the incorporated max-pooling operation selects the maximum feature value from that pool.
The modified Potes further teaches generating, based on the plurality of heart health parameters, one or more feature pools because the Ko max-pooling processing incorporated regarding claim 17 is also applied to the test-patient heart-health parameters, grouping the test feature values into pooling regions before selection of the test maximum features. For purposes of examination, “the plurality of heart health parameters” is interpreted as the plurality of test heart health parameters. Accordingly, the pooling regions generated from the test-patient parameters provide the recited one or more feature pools.
Also, regarding claim 22, the modified Potes does not fully teach reducing overfitting of the reference feature pools by removing one or more of the reference feature pools from remaining reference feature pools; and normalizing the selected reference maximum features from the reference feature pools. Rather, Ko teaches dropout that selects only a proportion of connections between layers “to prevent overfitting” and batch normalization of convolution-layer result values (Ko, ¶[0047]). The Instant Application's ¶[0082] supports treating dropout as the disclosed mechanism for excluding selected neural contributions from further processing. Under the interpretation stated above, Ko's dropout teaching therefore addresses the recited removal operation when applied to the pooled reference-feature representation because it prevents selected pooled feature contributions from propagating while retaining the remaining contributions, thereby reducing overfitting. Ko does not, however, expressly normalize the selected reference maximum features.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes by applying Ko's dropout regularization to the pooled reference-feature representation and applying Ko's normalization operation to the selected reference maximum features. Under the interpretation stated above, removing one or more of the reference feature pools from remaining reference feature pools is interpreted as excluding selected pooled feature contributions from further processing while retaining the other pooled contributions. Applying Ko's dropout operation to the pooled reference-feature representation would provide that result by excluding selected pooled feature contributions from further propagation while retaining the other pooled contributions, thereby reducing overfitting for the purpose expressly taught by Ko. A person of ordinary skill would have been motivated to normalize the selected pooled maxima so that the retained feature values supplied to subsequent prediction and comparison processing use the normalization convention already taught in Ko's CNN architecture. The claim does not require dropout/removal to precede normalization. Even if the listed claim language were interpreted to require the dropout/removal operation before normalization, it would have been obvious to apply Ko's normalization operation to the retained selected reference pooled maxima after dropout. Ko teaches both dropout for reducing overfitting and normalization of numerical CNN feature values within the same processing architecture (Ko, ¶[0047]). A person of ordinary skill would have been motivated to normalize the retained reference values actually supplied to subsequent prediction or comparison processing so that those values are provided on a consistent numerical basis. Arranging the known normalization operation after the dropout operation would additionally have constituted an obvious rearrangement of known processing steps yielding their expected functions, consistent with MPEP §2144.04(IV)(C) and In re Burhans, 154 F.2d 690, 69 USPQ 330 (CCPA 1946). The resulting processing would provide the recited overfitting reduction and normalized selected reference maximum features.
Claim 20 is rejected under 35 U.S.C. 103 as being unpatentable over Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes, and further in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, and further in view of Al Hatib et al. (US 2018/0028076 A1), hereinafter Hatib'076, and further in view of Smith et al. (US 2021/0259629 A1), hereinafter Smith'629, and further in view of Al Hatib et al. (US 2018/0025290 A1), hereinafter Hatib'290, and further in view of Johnson et al. (US 2021/0275783 A1), hereinafter Johnson.
The modified Potes teaches the limitations of claim 17 as discussed above.
Regarding claim 20, the modified Potes does not fully teach wherein the set of available heart health parameters comprises: a mean arterial pressure (MAP); a stroke volume index (SVI); a hypotension prediction index (HPI); a systemic vascular resistance (SVR); a heart rate (HR); a cardiac output (CO); a time-based change in arterial pressure; a cardiac index (CI); a systemic vascular resistance index (SVRI); a normalized area of pulse pressure; an average distance between subsequent MAPs; an average distance between a systolic peak and a respective diastolic peak; and a stroke volume variation (SVV). Rather, Potes teaches determining multiple arterial-pressure-derived physiological-state profiling parameters, including stroke volume, heart rate, MAP, and related hemodynamic features (Potes, ¶[0050]-[0052], [0061]). Potes further distinguishes between the parameters that can be determined and the subset selected for use, stating that “not all hemodynamic parameters determined are necessarily selected as physiological state profiling parameters 53” (Potes, ¶[0051]). However, the modified Potes does not teach the complete recited set of available heart health parameters.
Hatib'076 teaches a set of tracked health parameters including cardiac output (CO), stroke volume (SV), stroke volume variation (SVV), diastolic pressure, pulse rate (PR), stroke volume index (SVI), systemic vascular resistance (SVR), mean arterial pressure (MAP), HPI, systemic vascular resistance index (SVRI), cardiac index (CI), and systolic pressure (Hatib'076, ¶[0039]). Hatib'076 further teaches that these are health parameters being tracked by the health monitoring unit and that a user may select parameters from that available set for display (Hatib'076, ¶[0038]-[0042]). Hatib'290 further teaches heart rate as a five-minute averaged parameter (Hatib'290, Appendix A, item 9).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes by including Hatib'076's known hypotension-related hemodynamic parameters among the available parameters processed by Potes's system. The modification would have been technically feasible because both Potes and Hatib'076 derive and process hemodynamic parameters from arterial-pressure data. A person of ordinary skill would have been motivated to include Hatib'076's parameter set because Hatib'076 identifies those parameters as health parameters tracked in a system for evaluating and predicting hypotension, while Potes expressly permits selection of different arterial-pressure-derived features for physiological-state prediction. The resulting available parameter set would include MAP, SVI, HPI, SVR, HR, CO, CI, SVRI, and SVV.
Also, regarding claim 20, the modified Potes does not fully teach a time-based change in arterial pressure and a normalized area of pulse pressure. Rather, the modified Potes includes the arterial-pressure-derived parameter set established above, but does not teach these additional measures.
Hatib'290 teaches changes in MAP with respect to time as hypotension-predictive parameters (Hatib'290, ¶[0042]) and teaches ppAreaNor as a normalized area under the arterial-pressure waveform (Hatib'290, ¶[0063]). Hatib'290 identifies its disclosed arterial-pressure-derived parameters as parameters predictive of future hypotension (Hatib'290, ¶[0034], [0039], [0054]-[0057]). Thus, Hatib'290 teaches a time-based arterial-pressure-change measure and a normalized arterial-pressure pulse-area measure corresponding to the recited normalized area of pulse pressure.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes in view of Hatib'290 to include the disclosed time-based MAP-change parameter and normalized arterial-pressure pulse-area parameter among the available heart-health parameters. The modification would have been technically compatible because the additional parameters are calculated from the same type of arterial-pressure waveform already processed by Potes. A person of ordinary skill would have been motivated to include the parameters because Hatib'290 expressly identifies them for future-hypotension prediction, while Potes teaches adjusting the selected physiological-state features to improve prediction accuracy (Potes, ¶[0037]). The resulting available parameter set would further include a time-based change in arterial pressure and a normalized area of pulse pressure.
Also, regarding claim 20, the modified Potes does not fully teach an average distance between subsequent MAPs and an average distance between a systolic peak and a respective diastolic peak.
Johnson teaches calculating a blood-pressure variability metric using “successive variation of mean arterial pressure over a given number of heartbeats or a given time period” (Johnson, ¶[0051]). Johnson further teaches pulse pressure calculated as “the difference from the systolic peak to the diastolic trough” (Johnson, ¶[0053]). Johnson therefore identifies successive MAP changes across multiple heartbeats or a time period and the per-beat pressure-value distance between the systolic and diastolic extrema. For purposes of examination, the diastolic trough is the diastolic pressure extremum corresponding to the claim's recited “diastolic peak.” Johnson does not expressly state that either measure is averaged in the manner recited by claim 20.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes in view of Johnson to include these arterial-pressure variability measures among the available heart-health parameters. Johnson identifies successive variation of MAP over multiple heartbeats or a time period as a useful variability metric. A person of ordinary skill would have been motivated to calculate the absolute pressure-value difference between each pair of successive MAP values and average those absolute differences over Johnson's disclosed multi-heartbeat or time window to obtain a representative average successive-MAP distance rather than rely on a single pair of MAP values. A person of ordinary skill would likewise have been motivated to use Johnson's systolic-peak-to-diastolic-extremum pressure difference as the underlying per-beat distance and average that measure over the sampling periods already used for arterial-pressure parameters in the modified system to obtain a representative pressure-distance measure rather than rely on a single heartbeat. The resulting available parameter set would include the average distance between subsequent MAPs and the average distance between a systolic peak and a respective diastolic peak.
Claim 23 is rejected under 35 U.S.C. 103 as being unpatentable over Potes Blandon et al. (WO 2022/231890 A1), hereinafter Potes, and further in view of Ko et al. (KR 2021/0097514 A, English translation of record), hereinafter Ko, and further in view of Al Hatib et al. (US 2018/0028076 A1), hereinafter Hatib'076, and further in view of Smith et al. (US 2021/0259629 A1), hereinafter Smith'629, and further in view of Al Hatib et al. (US 2018/0192962 A1), hereinafter Hatib'962, and further in view of Gholami et al. (US 2020/0114077 A1), hereinafter Gholami.
The modified Potes teaches the limitations of claim 17 as discussed above.
Regarding claim 23, the modified Potes does not fully teach further comprising an infusion pump configured to deliver an intravenous therapeutic agent to the patient based on the likelihood that that the patient will experience hypotension within the specified time with at least the target threshold confidence level. Rather, as discussed regarding claim 17, the modified Potes determines the likelihood that the patient will experience hypotension within the specified future time with at least the target threshold confidence level. However, the modified Potes does not teach automatically administering a therapeutic agent in response to that threshold-satisfying future-hypotension determination or administering the therapeutic agent intravenously using an infusion pump.
Hatib'962 teaches identifying a most probable cause of a predicted future hypotension event and recommending a medical intervention for preventing the predicted event, including administration of saline or whole blood when low blood volume is identified (Hatib'962, ¶[0021]). Hatib'962 further teaches that “system 102 may automatically administer the medical intervention for preventing the future hypotension event” and that, “in response to detecting a future hypotension event”, the system may transmit a signal causing administration of saline or whole blood when low blood volume is identified (Hatib'962, ¶[0021]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes in view of Hatib'962 so that the determination that the future-hypotension likelihood satisfies the target threshold confidence level initiates administration of a therapeutic intervention for preventing the predicted hypotension event. The modification would have been technically feasible because the modified Potes already generates the threshold-satisfying future-hypotension determination used as the treatment trigger, while Hatib'962 expressly teaches automatically causing therapeutic administration in response to detection of a predicted future hypotension event. A person of ordinary skill would have been motivated to make the modification to permit treatment to begin before occurrence of the predicted hypotensive event, consistent with Hatib'962's stated purpose of preventing the future hypotension event. The resulting system would automatically initiate administration of a therapeutic agent based on the likelihood that the patient will experience hypotension within the specified time with at least the target threshold confidence level.
Also, regarding claim 23, the modified Potes does not fully teach an infusion pump configured to deliver an intravenous therapeutic agent to the patient. Rather, the modified Potes as established above automatically initiates therapeutic administration based on the future-hypotension determination, including administration of saline as taught by Hatib'962, but does not teach performing that intravenous administration using an infusion pump.
Gholami teaches that fluid resuscitation is “intravenous administration of fluids” (Gholami, ¶[0040]). Gholami further teaches a fully automated closed-loop fluid-resuscitation system in which an infusion rate computed from patient hemodynamic data is sent to an infusion pump (Gholami, ¶[0074]-[0075]). Gholami expressly teaches that the higher-level controller “sends a command to an infusion pump (external or built-in), which administers fluids or cardiovascular drugs to the patient” (Gholami, ¶[0075]). Gholami also teaches a fluid infusion pump in a combined closed-loop fluid and cardiovascular-drug administration system (Gholami, ¶[0076]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Potes in view of Gholami to implement the automatically initiated fluid therapy established above using Gholami's infusion-pump arrangement for intravenous fluid administration. The modification would have been technically feasible because Hatib'962 expressly identifies saline administration as a treatment for preventing the predicted future hypotension event, while Gholami expressly teaches intravenous fluid administration through an electronically controlled infusion pump. A person of ordinary skill would have been motivated to use Gholami's pump arrangement to provide controlled intravenous delivery of the automatically initiated fluid therapy. Gholami's fully automated closed-loop system already sends infusion commands to a pump based on patient hemodynamic information. A person of ordinary skill would have had a reasonable expectation of success because the modification uses a known infusion-pump delivery mechanism to administer the same type of fluid therapy already selected by the modified system. The resulting system would include an infusion pump configured to deliver an intravenous therapeutic agent to the patient based on the likelihood that the patient will experience hypotension within the specified time with at least the target threshold confidence level.
Conclusion
Any inquiry concerning this communication or earlier communications from the
examiner should be directed to AARON MERRIAM whose telephone number is (703) 756-
5938. The examiner can normally be reached M-F 7:00 am - 4:00 pm EST.
Examiner interviews are available via telephone, in-person, and video conferencing
using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is
encouraged to use the USPTO Automated Interview Request (AIR) at
http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s
supervisor, Jason Sims can be reached on (571)272-4867. The fax phone number for the
organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained
from Patent Center. Unpublished application information in Patent Center is available to
registered users. To file and manage patent submissions in Patent Center, visit:
https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for
more information about Patent Center and https://www.uspto.gov/patents/docx for
information about filing in DOCX format. For additional questions, contact the Electronic
Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO
Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/AARON MERRIAM/Examiner, Art Unit 3791
/MATTHEW KREMER/Primary Examiner, Art Unit 3791
Prosecution Insights