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 .
Applicant' s arguments, filed 9/3/2026, have been fully considered. The following rejections and/or objections are either reiterated or newly applied. They constitute the complete set presently being applied to the instant application.
Applicants have amended their claims, filed 9/3/2026, and therefore rejections newly made in the instant office action have been necessitated by amendment.
Claims 1-8, 11-13, 17, and 19-26 are the currently pending claims herby under examination. Claims 1-2, 5-8, 11-13, 17, and 19 were amended. Claims 9-10, 14-16, and 18 were canceled. Claims 20-26 are newly added.
Claim Objections
Claims 8 and 20-26 are objected to because of the following informalities:
In claim 8, lines 4 and 6 respectively: "a current threshold" and "the current threshold value" use inconsistent terminology for the same threshold and should be amended for consistency;
In claim 20, line 1: "a diagnosis of pain" and in claims 21-23, line 1 for each: "the diagnosis of pain" use terminology inconsistent with "the pain diagnosis" recited in claim 1 (lines 25-26) and should be amended to "the pain diagnosis" for consistency;
In claim 24, lines 1-2: "at least one time-series data" is grammatically incorrect and should be revised to identify the intended unit of time-series data, for example, "at least one set of time-series data";
In claim 25, line 6: "Euclidian" appears to contain a typographical error and should be revised to "Euclidean"; and
In claim 26, lines 1-2: “the band-pass filter consists of about 8 Hz to about 14 Hz” is inconsistent with the recitation of “band-pass filtering” in claim 6 and does not clearly state the frequency range over which the filtering is performed; Applicant is requested to revise the limitation to provide proper antecedent basis and clearly recite the intended filtering range, for example, “wherein the band-pass filtering is performed over a frequency range of about 8 Hz to about 14 Hz.”
Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claim 24 is rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention.
Claim 24 recites “wherein removing artifacts determines at least one time-series data having a highest variance” in lines 1-2. The application as originally filed has been reviewed for support for this limitation. Paragraph [0027] discloses artifact removal by reconstructing artifact-affected EEG data using a weighted average of neighboring electrode data, while ¶[0028] separately discloses common spatial pattern processing used for electrode selection by preselecting EEG electrodes that show the highest variance in their signal. Fig. 1a likewise depicts artifact removal at step 110 and common spatial pattern processing at separate subsequent step 112. The originally filed claims similarly recite highest-variance selection and artifact removal as separate operations. Neither these disclosures nor the application as a whole describes the presently claimed relationship in which the removing-artifacts operation determines at least one time series having a highest variance. The disclosure of artifact removal followed by a separate highest-variance selection step does not itself reasonably convey possession of artifact removal performing the claimed highest-variance determination. Accordingly, claim 24 lacks adequate written description under 35 U.S.C. § 112(a).
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 6, 19, and 22-26 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 6 recites "before the comparing, processing each of the plurality of time-series data to perform" in line 2. As amended, claim 1 recites two comparing operations: "comparing in real-time each of the plurality of time-series data with others of the time-series data" and "comparing the combined dataset to the database." It is therefore unclear which operation "the comparing" refers to, and a person of ordinary skill in the art would not be able to determine whether the recited cleaning, band-pass filtering, and artifact removal must precede the highest-variance comparison or need only precede the database comparison. The two readings differ in scope because the second permits the recited processing to be performed after the channel selection and detection operations. For purposes of examination, the Examiner interprets "the comparing" as the step of comparing in real-time each of the plurality of time-series data with others of the time-series data, consistent with Fig. 1a, in which preprocessing at 106 precedes common spatial pattern processing at 112. Accordingly, claim 6 is indefinite under 35 U.S.C. § 112(b).
Claims 24-26 are rejected by virtue of their dependence from claim 6.
Claim 19 recites "before the comparing, processing each of the plurality of time-series data to perform" in line 2. As amended, claim 13 recites two comparing operations: "comparing in real-time each of the plurality of time-series data with others of the time-series data" and "comparing the combined dataset to the database." It is therefore unclear which operation "the comparing" refers to, and a person of ordinary skill in the art would not be able to determine whether the recited cleaning, band-pass filtering, and artifact removal must precede the highest-variance comparison or need only precede the database comparison. The two readings differ in scope because the second permits the recited processing to be performed after the channel selection and detection operations. For purposes of examination, the Examiner interprets "the comparing" as the step of comparing in real-time each of the plurality of time-series data with others of the time-series data, consistent with Fig. 1a, in which preprocessing at 106 precedes common spatial pattern processing at 112. Accordingly, claim 19 is indefinite under 35 U.S.C. § 112(b).
Claim 22 recites “wherein the diagnosis of pain is 96% accurate” in line 1. The Instant Application at ¶[0039] defines accuracy based on true positives, true negatives, false positives, and false negatives, and ¶[0043] reports that the disclosed detection algorithm could classify pain patients from no-pain with 96% accuracy. For purposes of examination, the Examiner interprets the limitation to require 96% binary pain/no-pain classification accuracy calculated according to ¶[0039] over an evaluation dataset. However, neither claim 22 nor the Instant Application identifies the population, evaluation dataset, or evaluation conditions from which the true-positive, true-negative, false-positive, and false-negative values used to determine the claimed accuracy must be obtained. Because classification accuracy is determined from the outcomes in the evaluated set, the measured accuracy of the same diagnostic process may differ depending on the subjects or observations included in the evaluation. Accordingly, the claim does not provide an objective boundary for determining whether a particular diagnostic process satisfies the claimed 96% accuracy limitation. Therefore, claim 22 is indefinite under 35 U.S.C. § 112(b).
Claim 23 further recites “wherein the diagnosis of pain consists of low pain and high pain” in lines 1-2. Claim 21, from which claim 23 depends, recites that “the diagnosis of pain consists of no pain or pain.” It is unclear whether claim 23 is intended to further limit only the “pain” result of claim 21 into low-pain and high-pain classifications, such that the overall diagnosis remains no pain, low pain, or high pain, or whether claim 23 instead limits the entire diagnosis to only low pain and high pain, thereby excluding the no-pain result required by claim 21. This ambiguity arises because “the diagnosis of pain” is used to refer to the overall diagnostic determination while “pain” is also one of the possible results of that determination. For purposes of examination, the Examiner interprets claim 23 as further limiting the “pain” result of claim 21 into low pain and high pain while retaining the no-pain result. However, that interpretation is not compelled by the claim language. Accordingly, claim 23 is indefinite under 35 U.S.C. § 112(b).
Claim 23 recites “wherein a diagnosis of low pain is 100% accurate and a diagnosis of high pain is 100% accurate” in lines 2-3. The Instant Application at ¶[0039] defines accuracy as the proportion of true-positive and true-negative classifications among all evaluated classifications, and ¶[0043] states that the disclosed detection algorithm could classify high pain from low pain with 100% accuracy. The disclosure therefore describes an overall binary low-pain/high-pain classification accuracy, rather than separate accuracy measures for the low-pain and high-pain classes. For purposes of examination, the Examiner interprets claim 23 to require a binary low-pain/high-pain classification having 100% overall classification accuracy calculated according to ¶[0039] over an evaluation dataset. However, claim 23 does not identify the population, evaluation dataset, low-pain/high-pain classification boundary, or other evaluation conditions from which the classification outcomes are determined. Further, the claim separately recites that a diagnosis of low pain and a diagnosis of high pain are each 100% accurate, while the Instant Application defines only overall classification accuracy. Thus, it is unclear whether the separate 100% recitations require overall binary classification accuracy, class-specific sensitivity or specificity, positive predictive value, or another class-specific performance measure. Accordingly, claim 23 is indefinite under 35 U.S.C. § 112(b).
Claim 24 recites “wherein removing artifacts determines at least one time-series data having a highest variance” in lines 1-2. It is unclear whether the limitation requires the artifact-removal operation itself to determine the time-series data having the highest variance, or whether the limitation merely requires artifact removal to be followed by a separate determination of the time-series data having the highest variance. These interpretations materially differ because the first makes variance determination part of artifact removal, while the second merely specifies an order of separate preprocessing operations. The Instant Application at ¶[0027] describes artifact removal and ¶[0028] separately describes common spatial pattern processing used to identify EEG electrodes showing the highest variance, and Fig. 1a likewise shows artifact removal at step 110 followed by common spatial pattern processing at step 112. For purposes of examination, the Examiner interprets the limitation to require that the artifact-removal process include determining at least one EEG time series having the highest variance. However, this interpretation is not compelled by the claim language and disclosure because the disclosure instead presents artifact removal and highest-variance determination as separate operations. Accordingly, claim 24 is indefinite under 35 U.S.C. § 112(b).
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.
The instant application is a continuation-in-part of Application No. 15/257,019, now U.S. Patent No. 12,076,166 B2, hereto referred as Ref. The ordinary one-way test is applied.
Claims 1, 20, and 21 are rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 1 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 1: A method comprising at a computer system comprising a database and a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor:
Ref claim 1: “A method comprising at a computer system comprising a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor:”
Ref claim 1 recites the same computer implementation but does not require the database as part of the recited computer system. John teaches a pain-detection instrument including a database used by the processor to compare EEG-derived data with reference data (¶[0041]); including that database in the recited computer system is a predictable implementation of Ref claim 1’s pain-diagnosis process.
accessing a plurality of real-time time-series data based on electrophysiological signals of neural activity of a test subject, each electrophysiological signal obtained using one of a plurality of electro-encephalogram (EEG) sensors gathering raw EEG data representing a plurality of different frequencies, signals, patterns, and noise from the test subject; comparing in real-time each of the plurality of time-series data with others of the time-series data to determine those time-series data having a highest variance; based on the comparison, selecting in real-time, at the computer system, time-series data obtained from at least some sensors of the plurality of sensors that show the highest variance in their signal; detecting, at the computer system, a brain function pattern from the selected time-series data by at least one of: a spindle threshold analysis in the time domain, a power spectrum analysis in the frequency domain, and a wavelet analysis in the time-frequency domain;
Ref claim 1: “accessing a plurality of real-time time-series data based on electrophysiological signals of neural activity of a test subject, each electrophysiological signal obtained using one of a plurality of electro-encephalogram (EEG) sensors gathering raw EEG data representing a plurality of different frequencies, signals, patterns, and noise from the test subject; comparing in real-time each of the plurality of time-series data with others of the time series data to determine those time-series data having a highest variance; based on the comparison, selecting in real-time, at the computer system, time-series data from at least some sensors of the plurality of sensors that show the highest variance in their signal; detecting, at the computer system, a brain function pattern from the selected time-series data by at least one of: spindle threshold analysis in the time domain, a power spectrum analysis in the frequency domain, and a wavelet analysis in the time-frequency domain;”
Ref claim 1 substantively recites these EEG acquisition, highest-variance comparison/selection, and brain-function-pattern detection limitations. No secondary teaching is needed for these limitations.
combining the analyzed data into a combined dataset; comparing the combined dataset to the database and determining whether a pain diagnosis is available;
Ref claim 1: “determining whether the detected brain function pattern indicates an available pain diagnosis;”
Ref claim 1 determines whether the detected pattern indicates an available pain diagnosis but does not recite the combined-dataset/database comparison. John teaches selecting and combining qEEG features into a feature set and comparing EEG-derived data with database reference data to determine pain (¶¶[0022]-[0023], [0041], [0058]); using that known database comparison provides the recited diagnostic path.
based on the determination, when a pain diagnosis is not available, generating an instruction to generate a training dataset for a machine learning model and training the machine learning model using the training dataset;
Ref claim 1: “based on the determination, when a pain diagnosis is not available, generating an instruction to generate a training database for a machine learning model and training the machine learning model using the training database;”
Ref claim 1 already recites conditionally generating training data and training the machine-learning model. Reciting that collection as a training “dataset” rather than a training “database” does not provide a patentably distinct training operation.
determining whether the time-series data indicates a pain diagnosis based on a detected brain function pattern using the trained machine learning model; and outputting the indicated pain diagnosis in a human readable form, wherein the pain diagnosis comprises no pain or pain.
Ref claim 1: “determining whether the time-series data indicates a chronic pain diagnosis or an acute pain diagnosis based on the detected brain function pattern using the trained machine learning model; and outputting the indicated pain diagnosis in a human readable form.”
Ref claim 1 uses the trained model to determine chronic or acute pain. John teaches determining whether pain exists and outputting the pain result (¶¶[0041]-[0042], [0058]); using the classifier to return the binary pain/no-pain state is an obvious diagnostic-output variation.
Claim 20: The method of claim 1, wherein a diagnosis of pain comprises chronic pain or acute pain.
Ref claim 1: “determining whether the time-series data indicates a chronic pain diagnosis or an acute pain diagnosis based on the detected brain function pattern using the trained machine learning model; and outputting the indicated pain diagnosis in a human readable form.”
The additional chronic/acute limitation is already recited by Ref claim 1. John supplies only the inherited claim-1 differences identified above.
Claim 21: The method of claim 1, wherein the diagnosis of pain consists of no pain or pain.
Ref claim 1: “determining whether the time-series data indicates a chronic pain diagnosis or an acute pain diagnosis based on the detected brain function pattern using the trained machine learning model; and outputting the indicated pain diagnosis in a human readable form.”
Ref claim 1 does not limit the output to the binary no-pain/pain result. John expressly determines whether pain exists (¶[0058]); limiting the diagnostic outcome to those two states applies John’s binary pain determination to Ref claim 1 and provides the limitation.
Claim 2 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 2 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John, and further in view of Huupponen et al. (Huupponen, E., et al. "Optimization of Sigma Amplitude Threshold in Sleep Spindle Detection." J Sleep Res [Oxford UK], vol. 9, no. 4, December 2000, pp. 327-34), hereto referred as Huupponen.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 2: The method of claim 1, wherein the spindle threshold analysis in the time domain comprises: setting a threshold at a maximum value; reducing the threshold value repeatedly to obtain a current threshold value until the current threshold value is 0 microvolts; and detecting a spindle when a region of the data exceeds the current threshold value for at least about 0.5 seconds.
Ref claim 2, via claim 1: “wherein the spindle threshold analysis in the time domain comprises setting a threshold at a maximum value; reducing the threshold repeatedly until the threshold is at a minimum value; and detecting a spindle when a region of the data exceeds a current threshold value for at least a predetermined amount of time.”
Ref claim 2 already requires reducing the spindle threshold from a maximum to a minimum and detecting an above-threshold region for a predetermined time. Huupponen teaches amplitude-threshold optimization for spindle detection and that spindles shorter than 0.5 s are usually ignored (p. 327), with different candidate thresholds evaluated to select an optimum (pp. 330, 332). Selecting about 0.5 s therefore provides the known spindle-duration criterion. Ref claim 2 already requires reduction to a minimum threshold; using 0 microvolts as the lower endpoint of a nonnegative amplitude-threshold sweep is a predictable boundary implementation. John supplies the inherited claim-1 differences addressed above.
Claim 3 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 3 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John, and further in view of Huupponen et al. (Huupponen, E., et al. "Optimization of Sigma Amplitude Threshold in Sleep Spindle Detection." J Sleep Res [Oxford UK], vol. 9, no. 4, December 2000, pp. 327-34), hereto referred as Huupponen.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 3: The method of claim 2, wherein the power spectrum analysis in the frequency domain comprises performing power spectrum analysis using a Fourier transform or a fast Fourier transform.
Ref claim 3, via claims 2 and 1: “wherein the power spectrum analysis in the frequency domain comprises performing power spectrum analysis using a Fourier transform or a fast Fourier transform.”
The limitation added by claim 3 is already recited by Ref claim 3. John and Huupponen supply only the inherited claim-1 and claim-2 differences identified above.
Claim 4 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 4 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John, and further in view of Huupponen et al. (Huupponen, E., et al. "Optimization of Sigma Amplitude Threshold in Sleep Spindle Detection." J Sleep Res [Oxford UK], vol. 9, no. 4, December 2000, pp. 327-34), hereto referred as Huupponen.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 4: The method of claim 3, wherein the wavelet analysis in the time-frequency domain comprises performing wavelet analysis using a short time Fourier transform or a wavelet transform.
Ref claim 4, via claims 3, 2, and 1: “wherein the wavelet analysis in the time-frequency domain comprises performing wavelet analysis using a short time Fourier transform or a wavelet transform.”
The limitation added by claim 4 is already recited by Ref claim 4. John and Huupponen supply only the inherited claim-1 and claim-2 differences identified above.
Claim 5 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 5 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 5: The method of claim 1, further comprising: storing data related to the neural activity of the test subject in the database; receiving additional data representing at least one signal representing additional neural activity of the test subject or neural activity of another test subject; accessing the database using the additional data to determine whether a corresponding characterization of the neural activity of the test subject is found in the database; and when a corresponding characterization of the neural activity of the test subject is found in the database, outputting the corresponding characterization of the neural activity of the test subject found in the database.
Ref claim 5, via claim 1: “storing data related to the characterization of the neural activity in a database; receiving additional data representing at least one signal representing additional neural activity of the test subject or neural activity of another test subject; accessing the database using the additional data to determine whether a corresponding characterization of the neural activity is found in the database; and when a corresponding characterization of the neural activity is found in the database, outputting the corresponding characterization of the neural activity found in the database.”
Ref claim 5 substantively recites the same database storage, additional-neural-activity lookup, and corresponding-characterization output. The amendments identifying the neural activity as that of the test subject are antecedent/clarifying changes rather than a different database operation. John supplies the inherited claim-1 differences.
Claim 6 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 6 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 6: The method of claim 1, further comprising: before the comparing, processing each of the plurality of time-series data to perform: cleaning the raw EEG data from each EEG sensor to remove unwanted signals and noise to form cleaned EEG data from each sensor; band-pass filtering the cleaned EEG data from each sensor at the computer system by frequency filtering the cleaned EEG data from each sensor to analyze neural oscillation frequencies to obtain filtered EEG data from each sensor including neural oscillation frequencies; and removing, at the computer system, artifacts from the filtered EEG data from each sensor by averaging the filtered EEG data from each sensor with filtered EEG data from at least one other sensor.
Ref claim 6, via claim 1: “before the comparing, processing each of the plurality of time-series data to perform: cleaning the raw EEG data from each EEG sensor to remove unwanted signals and noise to form cleaned EEG data from each sensor; band-pass filtering the cleaned EEG data from each sensor at the computer system by frequency filtering the cleaned EEG data from each sensor to analyze neural oscillation frequencies to obtain a filtered EEG data from each sensor including neural oscillation frequencies; and removing, at the computer system, artifacts from the filtered EEG data from each sensor by averaging the filtered EEG data from each sensor with filtered EEG data from at least one other sensor.”
Ref claim 6 substantively recites the same cleaning, band-pass filtering, and artifact removal by averaging. The amendment from “a filtered EEG data” to “filtered EEG data” is grammatical only. John supplies the inherited claim-1 differences.
Claims 7, 8, and 11 are rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 10 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 7: A computer program product for monitoring neural activity, the computer program product comprising a non-transitory computer readable storage having program instructions embodied therewith, the program instructions executable by a computer, to cause the computer to perform a method comprising: at a computer system comprising a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor comprising receiving data representing at least one signal representing neural activity of a test subject and accessing a plurality of real-time time-series data based on electrophysiological signals obtained using a plurality of electro-encephalogram (EEG) sensors gathering raw EEG data representing a plurality of different frequencies, signals, patterns, and noise from the test subject;
Ref claim 10, via claims 9, 8, and 7: Ref claim 7 recites: “A computer program product for monitoring neural activity, the computer program product comprising a non-transitory computer readable storage having program instructions embodied therewith, the program instructions executable by a computer, to cause the computer to perform a method comprising: at the computer system comprising a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor, a digital signal representing neural activity of a test subject, accessing a plurality of real-time time-series data based on electrophysiological signals obtained using a plurality of electro-encephalogram (EEG) sensors gathering raw EEG data representing a plurality of different frequencies, signals, patterns, and noise from the test subject;”
The amended receiving/accessing syntax clarifies the computer operation but does not materially change the subject matter of Ref claim 10’s inherited claim-7 architecture.
comparing in real-time each of the plurality of time-series data with others of the time-series data to determine those time-series data having a highest variance; based on the comparison, selecting in real-time, at the computer system, time-series data obtained from at least some sensors of the plurality of sensors that show the highest variance in their signal;
Ref claim 10, via claim 7: “comparing in real-time each of the plurality of time-series data with others of the time series data to determine those time-series data having a highest variance; based on the comparison, selecting in real-time, at the computer system, time-series data from at least some sensors of the plurality of sensors that show the highest variance in their signal;”
Ref claim 10 inherits the same highest-variance comparison and selection from claim 7. No secondary teaching is needed.
detecting, at the computer system, a brain function pattern from the selected time-series data by analysis consisting of: a spindle threshold analysis in the time domain, a power spectrum analysis in the frequency domain, and a wavelet analysis in the time-frequency domain;
Ref claim 10, via claims 7-10: Ref claim 7 recites “by at least one of: spindle threshold analysis in the time domain, a power spectrum analysis in the frequency domain, and a wavelet analysis in the time-frequency domain”; Ref claim 8 recites “wherein the spindle threshold analysis in the time domain comprises” the threshold procedure; Ref claim 9 recites “wherein the power spectrum analysis in the frequency domain comprises performing power spectrum analysis using a Fourier transform or a fast Fourier transform”; and Ref claim 10 recites “wherein the wavelet analysis in the time-frequency domain comprises performing wavelet analysis using a short time Fourier transform or a wavelet transform.”
Ref claims 8, 9, and 10, in the selected dependency chain, further specify the spindle threshold, power spectrum, and wavelet analyses, respectively, and therefore at least suggest performing all three recited analyses. John teaches combining multiple qEEG feature sets to obtain a more sensitive set of features for pain assessment (¶[0023]). Performing all three analyses expressly identified in the Ref claim 10 chain would have been a predictable selection from the finite analysis modes recited in the reference claim.
combining the analysis into a single dataset; wherein the computer system further comprises a database; determining whether a comparison of the single dataset versus the database indicates an available pain diagnosis;
Ref claim 10, via claim 7: “determining whether the detected brain function pattern indicates an available pain diagnosis;” No combined-dataset/database limitation is recited in the claim-10 dependency chain.
Ref claim 10 does not recite the single-dataset/database comparison. John teaches combining selected qEEG features and comparing EEG-derived data with database reference data to determine pain (¶¶[0022]-[0023], [0041], [0058]); applying that known comparison to the Ref claim 10 program provides this limitation.
based on the determination, when a pain diagnosis is not available, generating an instruction to generate a training dataset for a machine learning model and training the machine learning model using the training dataset; determining whether the time-series data indicates a pain diagnosis based on a detected brain function pattern using the trained machine learning model; and outputting the indicated pain diagnosis in a human readable form, wherein the pain diagnosis comprises no pain or active pain.
Ref claim 10, via claim 7: “based on the determination, when a pain diagnosis is not available, generating an instruction to generate a training database for a machine learning model and training the machine learning model using the training database; determining whether the time-series data indicates a chronic pain diagnosis or an acute pain diagnosis based on the detected brain function pattern using the trained machine learning model; and outputting the indicated pain diagnosis in a human readable form.”
The conditional-training operation is already present in Ref claim 10 through claim 7. John teaches determining whether pain exists and outputting that result (¶¶[0041]-[0042], [0058]); using the resulting classifier for a no-pain/active-pain output is an obvious diagnostic-output variation. “Training dataset” versus “training database” does not create a patentably distinct training operation.
Claim 8: The computer program product of claim 7, wherein the spindle threshold analysis in the time domain comprises: setting a threshold at a maximum value; reducing the threshold repeatedly to obtain a current threshold until the threshold is at a minimum value; and detecting a spindle when a region of the data exceeds the current threshold value for at least a predetermined amount of time.
Ref claim 10, via claim 8: “wherein the spindle threshold analysis in the time domain comprises: setting a threshold at a maximum value; reducing the threshold repeatedly until the threshold is at a minimum value; and detecting a spindle when a region of the data exceeds a current threshold value for at least a predetermined amount of time.”
Ref claim 10 inherits the same threshold-analysis limitations from claim 8. The current-threshold wording is a clarification only. John supplies the inherited claim-7 differences.
Claim 11: The computer program product of claim 7, further comprising program instructions for: storing data related to the neural activity of the test subject in the database; receiving additional data representing at least one signal representing additional neural activity of the test subject or neural activity of another test subject; accessing the database using the additional data to determine whether a corresponding characterization of the neural activity of the test subject is found in the database; and when a corresponding characterization of the neural activity of the test subject is found in the database, outputting the corresponding characterization of the neural activity of the test subject found in the database.
No corresponding additional-data/database-characterization limitation is recited in Ref claim 10 or its dependency chain.
John teaches storing subject brain-wave data in a database, receiving further neural-activity data, comparing received data with stored reference/self-normative data, and outputting the corresponding pain-related characterization (¶¶[0015], [0019], [0041]-[0043], [0058]). Adding those known database operations to the Ref claim 10 program provides claim 11; John also supplies the inherited claim-7 differences.
Claim 12 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 12 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 12, via claim 7: detecting, at the computer system, a brain function pattern from the selected time-series data by analysis consisting of: a spindle threshold analysis in the time domain, a power spectrum analysis in the frequency domain, and a wavelet analysis in the time-frequency domain;
Ref claim 12, via claim 7: “detecting, at the computer system, a brain function pattern from the selected time-series data by at least one of: spindle threshold analysis in the time domain, a power spectrum analysis in the frequency domain, and a wavelet analysis in the time-frequency domain;”
Ref claim 12 already identifies the same three analysis modes as the finite claimed alternatives. John teaches selecting and combining multiple qEEG features into a feature set (¶¶[0022]-[0023]). Applying all three expressly recited analysis modes to obtain complementary time-, frequency-, and time-frequency information before combining their outputs is a predictable use of the analysis alternatives already identified in the reference claim.
Claim 12, via claim 7: combining the analysis into a single dataset; wherein the computer system further comprises a database; determining whether a comparison of the single dataset versus the database indicates an available pain diagnosis;
Ref claim 12, via claim 7: “determining whether the detected brain function pattern indicates an available pain diagnosis;” No single-dataset/database comparison is recited.
John teaches combining selected qEEG features and comparing EEG-derived data with database reference data to determine pain (¶¶[0022]-[0023], [0041], [0058]); applying that known comparison provides the inherited claim-7 limitation.
Claim 12: The computer program product of claim 7, further comprising: before the comparing, processing each of the plurality of time-series data to perform: cleaning the raw EEG data from each EEG sensor to remove unwanted signals and noise to form cleaned EEG data from each sensor; band-pass filtering the cleaned EEG data from each sensor at the computer system by frequency filtering the cleaned EEG data from each sensor to analyze neural oscillation frequencies to obtain filtered EEG data from each sensor including neural oscillation frequencies; and removing, at the computer system, artifacts from the filtered EEG data from each sensor by averaging the filtered EEG data from each sensor with filtered EEG data from at least one other sensor.
Ref claim 12, via claim 7: “before the comparing, processing each of the plurality of time-series data to perform: cleaning the raw EEG data from each EEG sensor to remove unwanted signals and noise to form cleaned EEG data from each sensor; band-pass filtering the cleaned EEG data from each sensor at the computer system by frequency filtering the cleaned EEG data from each sensor to analyze neural oscillation frequencies to obtain a filtered EEG data from each sensor including neural oscillation frequencies; and removing, at the computer system, artifacts from the filtered EEG data from each sensor by averaging the filtered EEG data from each sensor with filtered EEG data from at least one other sensor.”
The limitations added by claim 12 are substantively the same as Ref claim 12. The “filtered EEG data” amendment is grammatical only. John supplies the remaining inherited claim-7 differences.
Claim 13 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 18 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 13: A system for monitoring neural activity, the system comprising: a database, a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor to perform:
Ref claim 18, via claim 13: “A system for monitoring neural activity, the system comprising: a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor to perform:”
Ref claim 18 recites the same system architecture but not the database as a recited system component. John teaches the pain-detection instrument database used by the processor for EEG/reference-data comparison (¶[0041]); incorporating that database is a predictable implementation.
accessing a plurality of real-time time-series data based on electrophysiological signals of neural activity of a test subject, each electrophysiological signal obtained using one of a plurality of electro-encephalogram (EEG) sensors gathering raw EEG data representing a plurality of different frequencies, signals, patterns, and noise from the test subject; comparing in real-time each of the plurality of time-series data with others of the time-series data to determine those time-series data having a highest variance; based on the comparison, selecting in real-time, at a computer system, time-series data obtained from at least some sensors of the plurality of sensors that show the highest variance in their signal; detecting, at the computer system, a brain function pattern from the selected time-series data by at least one of: a spindle threshold analysis in the time domain, a power spectrum analysis in the frequency domain, and a wavelet analysis in the time-frequency domain;
Ref claim 18, via claim 13: “accessing a plurality of real-time time-series data based on electrophysiological signals of neural activity of a test subject, each electrophysiological signal obtained using one of a plurality of electro-encephalogram (EEG) sensors gathering raw EEG data representing a plurality of different frequencies, signals, patterns, and noise from the test subject; comparing in real-time each of the plurality of time-series data with others of the time series data to determine those time-series data having a highest variance; based on the comparison, selecting in real-time, at the computer system, time-series data from at least some sensors of the plurality of sensors that show the highest variance in their signal; detecting, at the computer system, a brain function pattern from the selected time-series data by at least one of: spindle threshold analysis in the time domain, a power spectrum analysis in the frequency domain, and a wavelet analysis in the time-frequency domain;”
These limitations are substantively recited by the Ref claim 18 dependency chain. No secondary teaching is needed.
combining the analyzed data into a combined dataset; comparing the combined dataset to the database and determining whether a pain diagnosis is available;
Ref claim 18, via claim 13: “determining whether the detected brain function pattern indicates an available pain diagnosis;”
Ref claim 18 does not recite the combined-dataset/database path. John teaches combining qEEG features and comparing EEG-derived data with database reference data to determine pain (¶¶[0022]-[0023], [0041], [0058]); applying that known comparison provides the limitation.
based on the determination, when a pain diagnosis is not available, generating an instruction to generate a training dataset for a machine learning model and training the machine learning model using the training dataset; determining whether the time-series data indicates a pain diagnosis based on a detected brain function pattern using the trained machine learning model; and outputting the indicated pain diagnosis in a human readable form;
Ref claim 18, via claim 13: “based on the determination, when a pain diagnosis is not available, generating an instruction to generate a training database for a machine learning model and training the machine learning model using the training database; determining whether the time-series data indicates a chronic pain diagnosis or an acute pain diagnosis based on the detected brain function pattern using the trained machine learning model; and outputting the indicated pain diagnosis in a human readable form.”
Ref claim 18 already contains the conditional machine-learning framework. Reciting a training “dataset” is not a patentably distinct training operation. John teaches determination of whether pain exists (¶[0058]), providing the more general pain-diagnosis output.
wherein the system is included in a portable or mobile computing device.
Ref claim 18: “wherein the processor, memory, and computer program instructions are included in a portable or mobile computing device.”
The portable/mobile limitation is expressly recited by Ref claim 18.
Claim 17 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 17 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 17, via its dependence from claim 13: combining the analyzed data into a combined dataset; comparing the combined dataset to the database and determining whether a pain diagnosis is available; based on the determination, when a pain diagnosis is not available, generating an instruction to generate a training dataset for a machine learning model and training the machine learning model using the training dataset; determining whether the time-series data indicates a pain diagnosis based on a detected brain function pattern using the trained machine learning model; and outputting the indicated pain diagnosis in a human readable form, wherein the system is included in a portable or mobile computing device.
Ref claim 17, via claim 13: “determining whether the detected brain function pattern indicates an available pain diagnosis; based on the determination, when a pain diagnosis is not available, generating an instruction to generate a training database for a machine learning model and training the machine learning model using the training database; determining whether the time-series data indicates a chronic pain diagnosis or an acute pain diagnosis based on the detected brain function pattern using the trained machine learning model; and outputting the indicated pain diagnosis in a human readable form.” No portable/mobile limitation is recited in the Ref claim 17 dependency chain.
John teaches combining qEEG features and comparing them with database reference data (¶¶[0022]-[0023], [0041], [0058]) and teaches a portable handheld pain-detection instrument (¶[0018]). Applying those known features supplies the inherited claim-13 differences.
Claim 17: The system of claim 13, further comprising computer program instructions for: storing data related to the neural activity of the test subject in the database; receiving additional data representing at least one signal representing additional neural activity of the test subject or neural activity of another test subject; accessing the database using the additional data to determine whether a corresponding characterization of the neural activity of the test subject is found in the database; and when a corresponding characterization of the neural activity of the test subject is found in the database, outputting the corresponding characterization of the neural activity of the test subject found in the database.
Ref claim 17, via claim 13: “further comprising computer program instructions for: storing data related to the characterization of the neural activity in a database; receiving additional data representing at least one signal representing additional neural activity of the test subject or neural activity of another test subject; accessing the database using the additional data to determine whether a corresponding characterization of the neural activity is found in the database; and when a corresponding characterization of the neural activity is found in the database, outputting the corresponding characterization of the neural activity found in the database.”
The database-characterization operations added by claim 17 are substantively the same as Ref claim 17. The amendments identifying the neural activity as that of the test subject are clarifying rather than a different operation.
Claim 19 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 19 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 19, via its dependence from claim 13: combining the analyzed data into a combined dataset; comparing the combined dataset to the database and determining whether a pain diagnosis is available; based on the determination, when a pain diagnosis is not available, generating an instruction to generate a training dataset for a machine learning model and training the machine learning model using the training dataset; determining whether the time-series data indicates a pain diagnosis based on a detected brain function pattern using the trained machine learning model; and outputting the indicated pain diagnosis in a human readable form, wherein the system is included in a portable or mobile computing device.
Ref claim 19, via claim 13: “determining whether the detected brain function pattern indicates an available pain diagnosis; based on the determination, when a pain diagnosis is not available, generating an instruction to generate a training database for a machine learning model and training the machine learning model using the training database; determining whether the time-series data indicates a chronic pain diagnosis or an acute pain diagnosis based on the detected brain function pattern using the trained machine learning model; and outputting the indicated pain diagnosis in a human readable form.” No portable/mobile limitation is recited in the Ref claim 19 dependency chain.
John teaches the combined qEEG/database comparison (¶¶[0022]-[0023], [0041], [0058]) and a portable handheld pain-detection instrument (¶[0018]); applying those known features supplies the inherited claim-13 differences.
Claim 19: The system of claim 13, further comprising: before the comparing, processing each of the plurality of time-series data to perform: cleaning the raw EEG data from each EEG sensor to remove unwanted signals and noise to form cleaned EEG data from each sensor; band-pass filtering the cleaned EEG data from each sensor at the computer system by frequency filtering the cleaned EEG data from each sensor to analyze neural oscillation frequencies to obtain filtered EEG data from each sensor including neural oscillation frequencies; and removing, at the computer system, artifacts from the filtered EEG data from each sensor by averaging the filtered EEG data from each sensor with filtered EEG data from at least one other sensor.
Ref claim 19, via claim 13: “before the comparing, processing each of the plurality of time-series data to perform: cleaning the raw EEG data from each EEG sensor to remove unwanted signals and noise to form cleaned EEG data from each sensor; band-pass filtering the cleaned EEG data from each sensor at the computer system by frequency filtering the cleaned EEG data from each sensor to analyze neural oscillation frequencies to obtain a filtered EEG data from each sensor including neural oscillation frequencies; and removing, at the computer system, artifacts from the filtered EEG data from each sensor by averaging the filtered EEG data from each sensor with filtered EEG data from at least one other sensor.”
The preprocessing limitations added by claim 19 are substantively the same as Ref claim 19. The “filtered EEG data” amendment is grammatical only. John supplies the inherited claim-13 differences.
Claim 22 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 1 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John, and further in view of Panavaranan et al. (Panavaranan, P. & Wongsawat, Y. (2013). EEG-based Pain Estimation via Fuzzy logic and polynomial kernel Support Vector Machine. The 2013 Biomedical Engineering International Conference (BMEiCON-2013)), hereto referred as Panavaranan.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 22: The method of claim 21, wherein the diagnosis of pain is 96% accurate.
Ref claim 1: “determining whether the detected brain function pattern indicates an available pain diagnosis; based on the determination, when a pain diagnosis is not available, generating an instruction to generate a training database for a machine learning model and training the machine learning model using the training database; determining whether the time-series data indicates a chronic pain diagnosis or an acute pain diagnosis based on the detected brain function pattern using the trained machine learning model; and outputting the indicated pain diagnosis in a human readable form.” No numerical classification-accuracy limitation is recited.
John supplies the inherited combined-dataset/database and binary pain/no-pain differences. Panavaranan teaches EEG pain-versus-rest classification using a polynomial-kernel SVM and reports 96.97% classification accuracy (p. 3, Figs. 5-6). Under the working interpretation stated in the § 112(b) analysis, using the higher-performing known classifier and obtaining the neighboring 96% empirical accuracy is a predictable performance result; the record identifies no technical criticality at exactly 96.00%.
Claim 23 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 1 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John, and further in view of Huang et al. (Huang, G., Xiao, P., Hung, Y.S., Iannetti, G.D., Zhang, Z.G., & Hu, L. (2013). A novel approach to predict subjective pain perception from single-trial laser-evoked potentials. NeuroImage, 81, 283-293), hereto referred as Huang, and further in view of Wager et al. (Wager, T.D., Atlas, L.Y., Lindquist, M.A., Roy, M., Woo, C.W., & Kross, E. (2013). An fMRI-Based Neurologic Signature of Physical Pain. New England Journal of Medicine, 368(15), 1388-1397), hereto referred as Wager.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 23: The method of claim 21, wherein the diagnosis of pain consists of low pain and high pain, wherein a diagnosis of low pain is 100% accurate and a diagnosis of high pain is 100% accurate.
Ref claim 1: “determining whether the time-series data indicates a chronic pain diagnosis or an acute pain diagnosis based on the detected brain function pattern using the trained machine learning model; and outputting the indicated pain diagnosis in a human readable form.” No low-pain/high-pain classification or numerical accuracy limitation is recited.
John supplies the inherited combined-dataset/database and binary pain/no-pain differences. Huang teaches electrophysiological classification into low-pain and high-pain states and evaluation by classification accuracy (pp. 287-289). Wager teaches neural-activity-based discrimination between two painful conditions with 100% sensitivity/specificity and corresponding decision accuracy (pp. 1391, 1393-1394). Under the working interpretation stated in the § 112(b) analysis, applying the known low/high classification and selecting sufficiently distinguishable conditions to obtain 100% binary decision accuracy is a predictable performance variation.
Claim 24 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 6 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John, and further in view of Nolan et al. (Nolan, H., Whelan, R., & Reilly, R.B. (2010). FASTER: Fully Automated Statistical Thresholding for EEG artifact Rejection. Journal of Neuroscience Methods, 192(1), 152-162), hereto referred as Nolan, and further in view of Alotaiby et al. (Alotaiby, T., El-Samie, F. E. A., Alshebeili, S. A., & Ahmad, I. (2015). A review of channel selection algorithms for EEG signal processing. EURASIP Journal on Advances in Signal Processing, 2015(1), Article 66), hereto referred as Alotaiby.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 24: The method of claim 6, wherein removing artifacts determines at least one time-series data having a highest variance.
Ref claim 6, via claim 1: Ref claim 1 recites “comparing in real-time each of the plurality of time-series data with others of the time series data to determine those time-series data having a highest variance”; Ref claim 6 recites “removing, at the computer system, artifacts from the filtered EEG data from each sensor by averaging the filtered EEG data from each sensor with filtered EEG data from at least one other sensor.”
John supplies the inherited claim-1 differences. Nolan teaches calculating variance for each EEG channel as part of channel-artifact detection and using outlying variance to identify contaminated channel data before removal/interpolation (pp. 154-156, Secs. 3.2 and 3.2.1). Alotaiby teaches comparing EEG-channel variances and selecting the channels having maximum variance (p. 6, Sec. 3.1.1). Ref claim 6 already includes both highest-variance comparison/selection and artifact removal; using Alotaiby’s maximum-variance comparison within Nolan’s variance-based artifact-detection stage predictably places the highest-variance determination within the artifact-removal operation, providing the recited relationship under the working claim interpretation.
Claim 25 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 6 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John, and further in view of Pizzagalli et al. (Diego A. Pizzagalli. “Electroencephalography and High-Density Electrophysiological Source Localization,” Handbook of Psychophysiology, 2007), hereto referred as Pizzagalli.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 25: The method of claim 6, wherein cleaning the raw EEG data comprises: determining EEG signals from a plurality of EEG sensors that are nearest neighbors; and generating a weighted average of the EEG signals from the plurality of nearest neighbor EEG sensors, wherein weights of each of the EEG signals are proportional to a Euclidian distance between the plurality of nearest neighbor EEG sensors.
Ref claim 6, via claim 1: “cleaning the raw EEG data from each EEG sensor to remove unwanted signals and noise to form cleaned EEG data from each sensor” and “removing, at the computer system, artifacts from the filtered EEG data from each sensor by averaging the filtered EEG data from each sensor with filtered EEG data from at least one other sensor.” No nearest-neighbor Euclidean-distance weighting during cleaning is recited.
John supplies the inherited claim-1 differences. Pizzagalli teaches linear nearest-neighbor interpolation in which corrupted EEG activity is reconstructed through a weighted average of neighboring electrodes, with weights proportional to Euclidean distance (p. 62, Sec. 4.1.2). Applying that known neighboring-electrode reconstruction during Ref claim 6’s cleaning/preprocessing provides corrected EEG data for the subsequent filtering and analysis and supplies the limitation.
Claim 26 is rejected on the ground of nonstatutory obviousness-type double patenting as being unpatentable over claim 6 of U.S. Patent No. 12,076,166 B2, hereto referred as Ref, in view of John et al. (US-20080249430-A1), hereto referred as John, and further in view of Zuckerman-Stark et al. (US-20130310660-A1), hereto referred as Zuckerman-Stark.
The analysis is as follows:
Instant Application Claim Language
Corresponding Reference-Claim Language
Concise Double-Patenting / Obviousness Analysis
Claim 26: The method of claim 6, wherein the band-pass filter consists of about 8 Hz to about 14 Hz.
Ref claim 6, via claim 1: “band-pass filtering the cleaned EEG data from each sensor at the computer system by frequency filtering the cleaned EEG data from each sensor to analyze neural oscillation frequencies to obtain a filtered EEG data from each sensor including neural oscillation frequencies.” No approximately 8-14 Hz range is recited.
John supplies the inherited claim-1 differences. Zuckerman-Stark teaches EEG frequency-band features for pain classification and expressly identifies the alpha band as 8-14 Hz (¶[0054], Table 1). Applying that known alpha-band range to Ref claim 6’s band-pass filtering isolates the alpha activity used for pain-related EEG analysis and provides the limitation under the working interpretation of claim 26.
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.
Claims 1, 5, 13, 17, 20-21 are rejected under 35 U.S.C. 103 as being unpatentable over John et al. (US-20080249430-A1), hereto referred as John, in view of Alotaiby et al. (Alotaiby, T., El-Samie, F. E. A., Alshebeili, S. A., & Ahmad, I. (2015). Α review of channel selection algorithms for EEG signal processing. EURASIP Journal on Advances in Signal Processing, 2015(1), Article 66), hereto referred as Alotaiby, and further in view of Krishnan et al. (US-20060184475-A1), hereto referred as Krishnan.
Regarding claim 1, John teaches A method comprising at a computer system comprising a database and a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor (John, ¶[0017]: “these LORETA images can be computed by an instrument that uses an appropriate array of electrodes placed upon the scalp, in accordance with the International 10/20 Electrode Placement System (Jasper 1958) and implements all computations from a set of digital amplifiers controlled and analyzed by a personal desktop or laptop computer”, ¶[0041]: “The digital signals are transmitted to a digital signal processor (DSP) 21 which may be included in or electrically coupled to a central processing unit (CPU) 25”, ¶[0041]: “the database 6, which may be stored locally in a memory in the PDI 16 (e.g., a nonvolatile memory) or remotely accessed via a network (e.g., the Internet, an intranet)”; John teaches a pain detecting instrument including database 6, a CPU, digital signal processor, and memory arrangement for storing data collected from the PDI, where the CPU/DSP performs the disclosed EEG processing and pain-detection operations, thereby teaching the claimed computer system under the broadest reasonable interpretation);
accessing a plurality of real-time time-series data based on electrophysiological signals of neural activity of a test subject, each electrophysiological signal obtained using one of a plurality of electro-encephalogram (EEG) sensors gathering raw EEG data representing a plurality of different frequencies, signals, patterns, and noise from the test subject (John, ¶[0019]: “The PDI 16 is coupled to electrode(s) 8 to receive electrical data corresponding to brain activity of the Subject 20”, ¶[0032]: “The PDI 16 may be configured to receive data from any number and/or type of biosensor electrodes”, ¶[0038]: “After artifact removal, the EEG may be subjected to spectral analysis using a signal processing technique such as, Fast Fourier Transform (FFT), wavelet analysis or fractal dimensional analysis”, ¶[0031]: “In the qEEG, the parameters may include, but are not limited to, power in a particular wide band frequency domain (e.g., delta, theta, alpha, beta, gamma), or in some particular frequency quantified using very narrow band (VNB) spectral analysis”; John teaches receiving EEG electrical brain-activity data from multiple biosensor electrodes, converting and processing the received electrode signals using an amplifier, ADC, filtering arrangement, CPU, and DSP, removing artifacts from the EEG before spectral analysis, and analyzing the EEG across multiple frequency bands, thereby teaching accessing EEG time-series signal data obtained from multiple EEG sensors and including frequency components and noise/artifacts before processing);
detecting, at the computer system, a brain function pattern from the selected time-series data by at least one of: a spindle threshold analysis in the time domain, a power spectrum analysis in the frequency domain, and a wavelet analysis in the time-frequency domain (John, ¶[0038]: “After artifact removal, the EEG may be subjected to spectral analysis using a signal processing technique such as, Fast Fourier Transform (FFT), wavelet analysis or fractal dimensional analysis”, ¶[0031]: “In the qEEG, the parameters may include, but are not limited to, power in a particular wide band frequency domain (e.g., delta, theta, alpha, beta, gamma), or in some particular frequency quantified using very narrow band (VNB) spectral analysis”; John teaches detecting pain-related brain-wave/qEEG patterns using FFT spectral analysis, frequency-domain power information, and wavelet analysis, thereby teaching at least power spectrum analysis in the frequency domain or wavelet analysis in the time-frequency domain. As discussed below, the modified John system performs this detection on the EEG channel data selected using Alotaiby’s maximum-variance channel-selection technique);
combining the analyzed data into a combined dataset; comparing the combined dataset to the database and determining whether a pain diagnosis is available (John, ¶[0022]: “a selected set of features is extracted from the digital data of acceptable quality” and “a selected set of features from the qEEG data which are particularly relevant to assessment of pain experienced are extracted from the overall data set obtained and compared to reference data”, wherein John further teaches computing univariate or multivariate data features for the extracted features; ¶[0023]: the qEEG feature sets may be used “in combination” as inputs to the pain-assessment classifier, and the results may be compared to data from the database to make determinations regarding the presence or absence, intensity, or type of pain; ¶[0058]: “the brain activity data is compared to the reference data to determine whether pain exists” and the brain-wave data are compared to reference data in database 6; John therefore teaches or at least suggests collecting analyzed qEEG feature data into a selected feature set used together for pain assessment and comparing the combined feature data to database reference data to determine whether a pain diagnosis is available);
outputting the indicated pain diagnosis in a human readable form (John, ¶[0042]: “the PDI 16 is coupled to a display for displaying brain wave data (i.e., EEG), the reference data and/or the pain index, etc. of the subject 20”; John, ¶[0058]: “The brain wave data is compared to the reference data in the database 6, and a value corresponding to the pain index (i.e., corresponding to a level of pain experienced by the subject 20) is returned and output by the I/O arrangement 24”; John teaches outputting the pain index and related brain-wave/reference information using an I/O arrangement and display); and
wherein the pain diagnosis comprises no pain or pain (John, ¶[0058]: “the brain activity data is compared to the reference data to determine whether pain exists”; John teaches determining the presence or absence of pain from the subject’s brain-wave data, thereby teaching a pain diagnosis comprising no pain or pain).
Also, regarding claim 1, John does not expressly teach comparing in real-time each of the plurality of time-series data with others of the time-series data to determine those time-series data having a highest variance; based on the comparison, selecting in real-time, at the computer system, time-series data obtained from at least some sensors of the plurality of sensors that show the highest variance in their signal. John teaches receiving EEG data from any number and/or type of biosensor electrodes (John, ¶[0019]: “The PDI 16 is coupled to electrode(s) 8 to receive electrical data corresponding to brain activity of the Subject 20”, ¶[0032]: “The PDI 16 may be configured to receive data from any number and/or type of biosensor electrodes”). However, John does not expressly teach selecting the EEG time-series data based on a comparison of channel variance.
Alotaiby is relied upon for this channel-selection feature. Alotaiby explains that EEG channel selection is used to reduce computational complexity, reduce overfitting from unnecessary channels, and reduce setup time while selecting relevant EEG channels for feature extraction (Alotaiby, p. 1, Abstract: “The main purpose of the channel selection process is threefold: (i) to reduce the computational complexity of any processing task performed on EEG signals by selecting the relevant channels and hence extracting the features of major importance, (ii) to reduce the amount of overfitting that may arise due to the utilization of unnecessary channels, for the purpose of improving the performance, and (iii) to reduce the setup time in some applications”). Alotaiby further teaches that channel selection may use EEG signal-processing tools including time-domain analysis, power spectral estimation, and wavelet transform (Alotaiby, p. 1, Abstract: “Signal processing tools such as time-domain analysis, power spectral estimation, and wavelet transform have been used for feature extraction and hence for channel selection in most of channel selection algorithms”). Most directly, Alotaiby teaches variance-based channel selection, disclosing that “the three channels with maximum variance are selected for signal classification” (Alotaiby, p. 6, Sec. 3.1.1). Thus, Alotaiby teaches comparing EEG channels based on variance and selecting the channels having maximum variance for EEG signal classification.
It would have been prima facie obvious before the effective filing date of the claimed invention to have modified John in view of Alotaiby to include Alotaiby’s variance-based channel selection as part of John’s computer-implemented EEG processing before John’s pain-related analysis. Because John’s system receives and processes EEG brain-activity data for pain assessment using the PDI/CPU/DSP arrangement, one skilled in the art would have understood that the variance-based channel selection would be implemented as part of the same computer-implemented EEG processing workflow before the diagnostic analysis, including for the accessed real-time EEG data. In the modified John system, Alotaiby’s maximum-variance channel-selection technique would have been performed on the EEG time-series data received from John’s multiple biosensor electrodes before John’s pain-related EEG analysis, such that the subsequent detection is performed using the selected maximum-variance EEG channel data. One skilled in the art would have been motivated to make the modification because John already receives EEG data from multiple biosensor electrodes for pain detection, while Alotaiby teaches that selecting relevant EEG channels reduces computational complexity, reduces overfitting, reduces setup time, and improves EEG classification performance. The modification would have been a predictable use of a known EEG signal-processing technique in John’s EEG-based medical diagnostic system because both references process EEG channel data for classification or detection tasks. The benefit of the modification would have been selecting more informative EEG channel data for John’s pain-detection classifier while reducing the amount of EEG data processed by John’s system.
Also, regarding claim 1, the modified John does not expressly teach based on the determination, when a pain diagnosis is not available, generating an instruction to generate a training dataset for a machine learning model and training the machine learning model using the training dataset. The modified John teaches applying classifier functions, including neural networks, to assess pain from EEG data (John, ¶[0025]: “Those skilled in the art will understand that, in other embodiments any or a combination of discriminant functions, cluster algorithms, neural networks and/or other classifier functions will be applied to the data for the purpose of assessing pain”). However, the modified John does not expressly teach conditionally generating or selecting training data and training a classifier when comparison of the current combined qEEG feature dataset to the existing pain database/classifier framework does not provide an available diagnosis for the current patient data.
Krishnan is relied upon for this conditional classifier-training feature. Krishnan is directed to medical decision support systems that address missing or unavailable patient-record information by assigning, constructing, and training classifiers based on the features available for the current patient record. Krishnan teaches that a classifier may be constructed for a new patient record, or only when an available classifier does not have the corresponding feature set for that patient record (Krishnan, ¶[0026]: “The classifier is constructed for each instance of a new patient record to be analyzed or only where another available classifier does not have a corresponding feature set”). Krishnan further teaches constructing the classifier from training data and training the classifier from a selected subset of that training data (Krishnan, ¶[0043]: “In another embodiment, the processor 12 obtains the classifier by constructing the classifier from training data”, “The processor 12 trains a classifier from the selected sub-set of the training data”). Krishnan explains that the selected features correspond to the features available for the current patient record, and that the processor constructs a classifier after selecting training-set patient-record information (Krishnan, ¶[0056]: “The selected features are the set or a sub-set of the features available for the current patient record”, ¶[0058]: “After selecting the training set patient record information, the processor constructs a classifier from the selected set of features from the training set patient records”). Thus, Krishnan teaches a medical decision-support technique in which training data are selected and a classifier is trained when the available classifier structure does not correspond to the available patient data.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John to incorporate Krishnan’s conditional medical classifier-training technique so that, when John’s pain database/classifier framework does not include a classifier or reference-data feature set corresponding to the current combined qEEG feature dataset, the system generates or selects a training-data subset and trains a machine learning classifier using that training data. In the further modified John system, the absence of a corresponding classifier or reference-data feature set in John’s pain-diagnosis framework would trigger Krishnan’s construction and training of a classifier from selected training data, thereby providing the claimed conditional training operation when comparison of the current combined qEEG feature dataset to the database does not provide an available pain diagnosis. One skilled in the art would have been motivated to make the modification because John and Krishnan are both directed to computer-implemented medical diagnostic systems that use patient data and classifier functions to produce diagnostic information. John teaches assessing pain using patient EEG data, reference data, and classifier functions, including neural networks. Krishnan teaches a known way to handle a patient record that is not adequately covered by an available classifier, namely constructing and training a classifier from selected training data corresponding to the available patient information. The modification would have been a predictable use of Krishnan’s medical classifier-training technique in John’s classifier-based pain-assessment framework. The benefit of the modification would have been improved diagnostic coverage for patient EEG feature data that are not adequately handled by John’s existing pain database or classifier framework.
Also, regarding claim 1, the modified John does not expressly teach determining whether the time-series data indicates a pain diagnosis based on a detected brain function pattern using the trained machine learning model.
Rather, the modified John teaches determining whether pain exists based on brain-wave/qEEG data (John, ¶[0058]: “the brain activity data is compared to the reference data to determine whether pain exists”). John also teaches that neural networks and other classifier functions may be applied to the data for assessing pain (John, ¶[0025]: “Those skilled in the art will understand that, in other embodiments any or a combination of discriminant functions, cluster algorithms, neural networks and/or other classifier functions will be applied to the data for the purpose of assessing pain”). However, the modified John does not expressly teach that the pain determination is made using the conditionally trained machine learning model as previously recited.
Krishnan teaches training a classifier from selected training data and applying the trained classifier to output diagnostic information for the current patient record (Krishnan, ¶[0043]: “The processor 12 trains a classifier from the selected sub-set of the training data”; Krishnan, ¶[0027]: “Applying the classifier determines a diagnosis. The classifier outputs a diagnosis, such as a conclusion, probability, location of concern or other information to assist with diagnosis”; Krishnan, ¶[0062]: “The classification is performed with neural network, filter, algorithm, or other now-known or later developed classifier or classification technique. The classifier is configured or trained for distinguishing between the desired States”). Thus, John teaches the pain determination and Krishnan teaches that a trained classifier is used to output the diagnostic classification.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John in view of Krishnan to use Krishnan’s trained classifier as the classifier in John’s pain determination. One skilled in the art would have been motivated to make the modification because John already contemplates classifier-based pain assessment, including neural networks and other classifier functions, and Krishnan provides a known medical decision-support technique for training and applying a classifier when the current patient data are not adequately handled by an existing classifier. The modification would have been a predictable use of Krishnan’s trained medical classifier within John’s pain-assessment framework. The benefit of the modification would have been enabling John’s system to determine a pain diagnosis using a classifier trained for the available patient data when the existing database/classifier framework does not provide an available diagnosis. Thus, in the further modified John system, John’s pain determination would have been made using the Krishnan-trained classifier when the classifier was constructed and trained for the available patient EEG data.
Regarding claim 5, the modified John teaches storing data related to the neural activity of the test subject in the database (John, ¶[0015]: “The database may further include self-normative data indicative of the subject’s brain wave activity in the presence and absence of pain” and “The database may further include medical histories and physical and/or neurological examination results for individuals in the population and/or the subject”; John, ¶[0041]: “the CPU 25 compares the signals output by the DSP 21 to the reference data stored in the database 6, which may be stored locally in a memory in the PDI 16 (e.g., a nonvolatile memory) or remotely accessed via a network (e.g., the Internet, an intranet)”; John teaches storing in database 6 data related to the test subject’s brain-wave activity, including self-normative brain-wave data for the subject);
receiving additional data representing at least one signal representing additional neural activity of the test subject or neural activity of another test subject (John, ¶[0019]: “The PDI 16 is coupled to electrode(s) 8 to receive electrical data corresponding to brain activity of the subject 20”; John, ¶[0032]: “The PDI 16 may be configured to receive data from any number and/or type of biosensor electrodes”; John, ¶[0057]: “the qEEG and/or the qSEP data of the subject 20 are used to generate data corresponding to the brain activity which corresponds to pain experienced by the subject 20”; John teaches receiving EEG/qEEG or qSEP data representing neural activity from a subject. Because John’s database includes population reference data from pluralities of individuals and the PDI is used to detect pain in subjects in clinical and non-clinical settings, John teaches or at least suggests receiving additional neural-activity data from the same subject or another subject for comparison to the database);
accessing the database using the additional data to determine whether a corresponding characterization of the neural activity of the test subject is found in the database (John, ¶[0015]: the subject’s EEG data may be compared to data in one or more databases, and the database may include self-normative data indicative of the subject’s brain-wave activity in the presence and absence of pain; John, ¶[0041]: “the CPU 25 compares the signals output by the DSP 21 to the reference data stored in the database 6, which may be stored locally in a memory in the PDI 16 (e.g., a nonvolatile memory) or remotely accessed via a network (e.g., the Internet, an intranet)”; John, ¶[0043]: “The qEEG analysis may reveal values of quantitative variables that exceed normative thresholds as identified in the database 6, indicative of the presence of pain”; John, ¶[0058]: “The brain wave data is compared to the reference data in the database 6, and a value corresponding to the pain index (i.e., corresponding to a level of pain experienced by the subject 20) is returned and output by the I/O arrangement 24”; John teaches accessing database 6 using received brain-wave data and comparing the data to stored reference data, including self-normative data for the subject, to determine a corresponding pain-related neural-activity characterization); and
when the corresponding characterization of the neural activity of the test subject is found in the database, outputting the corresponding characterization of the neural activity of the test subject found in the database (John, ¶[0015]: the database may include self-normative data indicative of the subject’s brain-wave activity in the presence and absence of pain; John, ¶[0041]: “As a result of the comparison, the CPU 25 outputs data indicating whether pain was detected and, if so, a magnitude of the pain (i.e., corresponding to the pain index)”; John, ¶[0042]: “the PDI 16 is coupled to a display for displaying brain wave data (i.e., EEG), the reference data and/or the pain index, etc. of the subject 20”; John, ¶[0058]: “The brain wave data is compared to the reference data in the database 6, and a value corresponding to the pain index (i.e., corresponding to a level of pain experienced by the subject 20) is returned and output by the I/O arrangement 24”; John teaches outputting the corresponding pain characterization obtained through comparison with database 6, including displaying the subject’s brain-wave data, corresponding reference data, and/or pain index in human-readable form).
Regarding claim 13, John teaches or at least suggests a system for monitoring neural activity, the system comprising: a database, a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor to perform (John, ¶[0017]: “these LORETA images can be computed by an instrument that uses an appropriate array of electrodes placed upon the scalp, in accordance with the International 10/20 Electrode Placement System (Jasper 1958) and implements all computations from a set of digital amplifiers controlled and analyzed by a personal desktop or laptop computer”, ¶[0041]: “The digital signals are transmitted to a digital signal processor (DSP) 21 which may be included in or electrically coupled to a central processing unit (CPU) 25”, ¶[0041]: “the database 6, which may be stored locally in a memory in the PDI 16 (e.g., a nonvolatile memory) or remotely accessed via a network (e.g., the Internet, an intranet)”; John teaches computer-based EEG pain-detection processing using a CPU/DSP arrangement, database, and memory, including computation and analysis of EEG data by a personal desktop or laptop computer. Under the broadest reasonable interpretation, John’s computer-implemented EEG processing, pain-assessment algorithm, database comparison, classifier functions, and output operations teach or at least suggest program instructions stored in computer-readable memory and executable by a processor to cause the system to perform the recited neural-activity monitoring operations);
accessing a plurality of real-time time-series data based on electrophysiological signals of neural activity of a test subject, each electrophysiological signal obtained using one of a plurality of electro-encephalogram (EEG) sensors gathering raw EEG data representing a plurality of different frequencies, signals, patterns, and noise from the test subject (John, ¶[0019]: “The PDI 16 is coupled to electrode(s) 8 to receive electrical data corresponding to brain activity of the Subject 20”, ¶[0032]: “The PDI 16 may be configured to receive data from any number and/or type of biosensor electrodes”, ¶[0038]: “After artifact removal, the EEG may be subjected to spectral analysis using a signal processing technique such as, Fast Fourier Transform (FFT), wavelet analysis or fractal dimensional analysis”, ¶[0031]: “In the qEEG, the parameters may include, but are not limited to, power in a particular wide band frequency domain (e.g., delta, theta, alpha, beta, gamma), or in some particular frequency quantified using very narrow band (VNB) spectral analysis”; John teaches receiving EEG electrical brain-activity data from multiple biosensor electrodes, converting and processing the received electrode signals using an amplifier, ADC, filtering arrangement, CPU, and DSP, removing artifacts from the EEG before spectral analysis, and analyzing the EEG across multiple frequency bands, thereby teaching accessing EEG time-series signal data obtained from multiple EEG sensors and including frequency components and noise/artifacts before processing);
detecting, at the computer system, a brain function pattern from the selected time-series data by at least one of: a spindle threshold analysis in the time domain, a power spectrum analysis in the frequency domain, and a wavelet analysis in the time-frequency domain (John, ¶[0038]: “After artifact removal, the EEG may be subjected to spectral analysis using a signal processing technique such as, Fast Fourier Transform (FFT), wavelet analysis or fractal dimensional analysis”, ¶[0031]: “In the qEEG, the parameters may include, but are not limited to, power in a particular wide band frequency domain (e.g., delta, theta, alpha, beta, gamma), or in some particular frequency quantified using very narrow band (VNB) spectral analysis”; John teaches detecting pain-related brain-wave/qEEG patterns using FFT spectral analysis, frequency-domain power information, and wavelet analysis, thereby teaching at least power spectrum analysis in the frequency domain or wavelet analysis in the time-frequency domain. As discussed below, the modified John system performs this detection on the EEG channel data selected using Alotaiby’s maximum-variance channel-selection technique);
combining the analyzed data into a combined dataset; comparing the combined dataset to the database and determining whether a pain diagnosis is available (John, ¶[0022]: “a selected set of features is extracted from the digital data of acceptable quality” and “a selected set of features from the qEEG data which are particularly relevant to assessment of pain experienced are extracted from the overall data set obtained and compared to reference data”, wherein John further teaches computing univariate or multivariate data features for the extracted features; John, ¶[0023]: the qEEG feature sets may be used alone or “in combination” as inputs to the pain-assessment classifier; John, ¶[0058]: “the brain activity data is compared to the reference data to determine whether pain exists” and the brain-wave data are compared to reference data in database 6; John therefore teaches or at least suggests collecting analyzed qEEG feature data into a selected feature set used together for pain assessment and comparing the selected feature data to database reference data to determine whether a pain diagnosis is available); and
outputting the indicated pain diagnosis in a human readable form (John, ¶[0042]: “the PDI 16 is coupled to a display for displaying brain wave data (i.e., EEG), the reference data and/or the pain index, etc. of the subject 20”; John, ¶[0058]: “The brain wave data is compared to the reference data in the database 6, and a value corresponding to the pain index (i.e., corresponding to a level of pain experienced by the subject 20) is returned and output by the I/O arrangement 24”; John teaches outputting the pain index and related brain-wave/reference information using an I/O arrangement and display).
Also, regarding claim 13, John does not expressly teach comparing in real-time each of the plurality of time-series data with others of the time-series data to determine those time-series data having a highest variance; based on the comparison, selecting in real-time, at a computer system, time-series data obtained from at least some sensors of the plurality of sensors that show the highest variance in their signal. John teaches receiving EEG data from any number and/or type of biosensor electrodes (John, ¶[0019]: “The PDI 16 is coupled to electrode(s) 8 to receive electrical data corresponding to brain activity of the Subject 20”, ¶[0032]: “The PDI 16 may be configured to receive data from any number and/or type of biosensor electrodes”). However, John does not expressly teach selecting the EEG time-series data based on a comparison of channel variance.
Alotaiby is relied upon for this channel-selection feature. Alotaiby explains that EEG channel selection is used to reduce computational complexity, reduce overfitting from unnecessary channels, and reduce setup time while selecting relevant EEG channels for feature extraction (Alotaiby, p. 1, Abstract: “The main purpose of the channel selection process is threefold: (i) to reduce the computational complexity of any processing task performed on EEG signals by selecting the relevant channels and hence extracting the features of major importance, (ii) to reduce the amount of overfitting that may arise due to the utilization of unnecessary channels, for the purpose of improving the performance, and (iii) to reduce the setup time in some applications”). Alotaiby further teaches that channel selection may use EEG signal-processing tools including time-domain analysis, power spectral estimation, and wavelet transform (Alotaiby, p. 1, Abstract: “Signal processing tools such as time-domain analysis, power spectral estimation, and wavelet transform have been used for feature extraction and hence for channel selection in most of channel selection algorithms”). Most directly, Alotaiby teaches variance-based channel selection, disclosing that “the three channels with maximum variance are selected for signal classification” (Alotaiby, p. 6, Sec. 3.1.1). Thus, Alotaiby teaches comparing EEG channels based on variance and selecting the channels having maximum variance for EEG signal classification.
It would have been prima facie obvious before the effective filing date of the claimed invention to have modified John in view of Alotaiby to include Alotaiby’s variance-based channel selection as part of John’s computer-implemented EEG processing before John’s pain-related analysis. Because John’s system receives and processes EEG brain-activity data for pain assessment using the PDI/CPU/DSP arrangement, one skilled in the art would have understood that the variance-based channel selection would be implemented as part of the same computer-implemented EEG processing workflow before the diagnostic analysis, including for the accessed real-time EEG data. In the modified John system, Alotaiby’s maximum-variance channel-selection technique would have been performed on the EEG time-series data received from John’s multiple biosensor electrodes before John’s pain-related EEG analysis, such that the subsequent detection is performed using the selected maximum-variance EEG channel data. One skilled in the art would have been motivated to make the modification because John already receives EEG data from multiple biosensor electrodes for pain detection, while Alotaiby teaches that selecting relevant EEG channels reduces computational complexity, reduces overfitting, reduces setup time, and improves EEG classification performance. The modification would have been a predictable use of a known EEG signal-processing technique in John’s EEG-based medical diagnostic system because both references process EEG channel data for classification or detection tasks. The benefit of the modification would have been selecting more informative EEG channel data for John’s pain-detection classifier while reducing the amount of EEG data processed by John’s system.
Also, regarding claim 13, the modified John does not expressly teach based on the determination, when a pain diagnosis is not available, generating an instruction to generate a training dataset for a machine learning model and training the machine learning model using the training dataset. The modified John teaches applying classifier functions, including neural networks, to assess pain from EEG data (John, ¶[0025]: “Those skilled in the art will understand that, in other embodiments any or a combination of discriminant functions, cluster algorithms, neural networks and/or other classifier functions will be applied to the data for the purpose of assessing pain”). However, the modified John does not expressly teach conditionally generating or selecting training data and training a classifier when comparison of the current combined qEEG feature dataset to the existing pain database/classifier framework does not provide an available diagnosis for the current patient data.
Krishnan is relied upon for this conditional classifier-training feature. Krishnan is directed to medical decision support systems that address missing or unavailable patient-record information by assigning, constructing, and training classifiers based on the features available for the current patient record. Krishnan teaches that a classifier may be constructed for a new patient record, or only when an available classifier does not have the corresponding feature set for that patient record (Krishnan, ¶[0026]: “The classifier is constructed for each instance of a new patient record to be analyzed or only where another available classifier does not have a corresponding feature set”). Krishnan further teaches constructing the classifier from training data and training the classifier from a selected subset of that training data (Krishnan, ¶[0043]: “In another embodiment, the processor 12 obtains the classifier by constructing the classifier from training data”, “The processor 12 trains a classifier from the selected sub-set of the training data”). Krishnan explains that the selected features correspond to the features available for the current patient record, and that the processor constructs a classifier after selecting training-set patient-record information (Krishnan, ¶[0056]: “The selected features are the set or a sub-set of the features available for the current patient record”, ¶[0058]: “After selecting the training set patient record information, the processor constructs a classifier from the selected set of features from the training set patient records”). Thus, Krishnan teaches a medical decision-support technique in which training data are selected and a classifier is trained when the available classifier structure does not correspond to the available patient data.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John to incorporate Krishnan’s conditional medical classifier-training technique so that, when John’s pain database/classifier framework does not include a classifier or reference-data feature set corresponding to the current combined qEEG feature dataset, the system generates or selects a training-data subset and trains a machine learning classifier using that training data. In the further modified John system, the absence of a corresponding classifier or reference-data feature set in John’s pain-diagnosis framework would trigger Krishnan’s construction and training of a classifier from selected training data, thereby providing the claimed conditional training operation when comparison of the current combined qEEG feature dataset to the database does not provide an available pain diagnosis. One skilled in the art would have been motivated to make the modification because John and Krishnan are both directed to computer-implemented medical diagnostic systems that use patient data and classifier functions to produce diagnostic information. John teaches assessing pain using patient EEG data, reference data, and classifier functions, including neural networks. Krishnan teaches a known way to handle a patient record that is not adequately covered by an available classifier, namely constructing and training a classifier from selected training data corresponding to the available patient information. The modification would have been a predictable use of Krishnan’s medical classifier-training technique in John’s classifier-based pain-assessment framework. The benefit of the modification would have been improved diagnostic coverage for patient EEG feature data that are not adequately handled by John’s existing pain database or classifier framework.
Also, regarding claim 13, the modified John does not expressly teach determining whether the time-series data indicates a pain diagnosis based on a detected brain function pattern using the trained machine learning model. Rather, the modified John teaches determining whether pain exists based on brain-wave/qEEG data (John, ¶[0058]: “the brain activity data is compared to the reference data to determine whether pain exists”). John also teaches that neural networks and other classifier functions may be applied to the data for assessing pain (John, ¶[0025]: “Those skilled in the art will understand that, in other embodiments any or a combination of discriminant functions, cluster algorithms, neural networks and/or other classifier functions will be applied to the data for the purpose of assessing pain”). However, the modified John does not expressly teach that the pain determination is made using the conditionally trained machine learning model as previously recited.
Krishnan teaches training a classifier from selected training data and applying the trained classifier to output diagnostic information for the current patient record (Krishnan, ¶[0043]: “The processor 12 trains a classifier from the selected sub-set of the training data”; Krishnan, ¶[0027]: “Applying the classifier determines a diagnosis. The classifier outputs a diagnosis, such as a conclusion, probability, location of concern or other information to assist with diagnosis”; Krishnan, ¶[0062]: “The classification is performed with neural network, filter, algorithm, or other now-known or later developed classifier or classification technique. The classifier is configured or trained for distinguishing between the desired States”). Thus, John teaches the pain determination and Krishnan teaches that a trained classifier is used to output the diagnostic classification.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John in view of Krishnan to use Krishnan’s trained classifier as the classifier in John’s pain determination. One skilled in the art would have been motivated to make the modification because John already contemplates classifier-based pain assessment, including neural networks and other classifier functions, and Krishnan provides a known medical decision-support technique for training and applying a classifier when the current patient data are not adequately handled by an existing classifier. The modification would have been a predictable use of Krishnan’s trained medical classifier within John’s pain-assessment framework. The benefit of the modification would have been enabling John’s system to determine a pain diagnosis using a classifier trained for the available patient data when the existing database/classifier framework does not provide an available diagnosis. Thus, in the further modified John system, John’s pain determination would have been made using the Krishnan-trained classifier when the classifier was constructed and trained for the available patient EEG data.
The modified John further teaches or at least suggests wherein the system is included in a portable or mobile computing device (John, ¶[0018]: “In the exemplary embodiment, the PDI 16 is implemented as a portable, handheld device for use in clinical and non-clinical settings”; John, ¶[0018]: “In an example of the latter case, an EMT or other medical personnel in the field may use the PDI 16 to detect and/or quantify pain experienced by a subject to, for example, determine an appropriate pain management strategy or, in conjunction with palpation, etc. assist in making a preliminary diagnosis”; John, ¶[0041]: “The digital signals are transmitted to a digital signal processor (DSP) 21 which may be included in or electrically coupled to a central processing unit (CPU) 25”; John, ¶[0041]: “the database 6, which may be stored locally in a memory in the PDI 16 (e.g., a nonvolatile memory) or remotely accessed via a network (e.g., the Internet, an intranet)”; John teaches that the pain detecting instrument is implemented as a portable, handheld device for use in clinical and non-clinical settings and further teaches that the PDI includes the processing and memory components used for the computer-implemented pain-detection functions).
Regarding claim 17, the modified John teaches or at least suggests computer program instructions for storing data related to the neural activity of the test subject in the database (John, ¶[0015]: “The database may further include self-normative data indicative of the subject’s brain wave activity in the presence and absence of pain” and “The database may further include medical histories and physical and/or neurological examination results for individuals in the population and/or the subject”; John, ¶[0041]: “the CPU 25 compares the signals output by the DSP 21 to the reference data stored in the database 6, which may be stored locally in a memory in the PDI 16 (e.g., a nonvolatile memory) or remotely accessed via a network (e.g., the Internet, an intranet)”; John teaches storing in database 6 data related to the test subject’s brain-wave activity, including self-normative brain-wave data for the subject);
receiving additional data representing at least one signal representing additional neural activity of the test subject or neural activity of another test subject (John, ¶[0019]: “The PDI 16 is coupled to electrode(s) 8 to receive electrical data corresponding to brain activity of the subject 20”; John, ¶[0032]: “The PDI 16 may be configured to receive data from any number and/or type of biosensor electrodes”; John, ¶[0057]: “the qEEG and/or the qSEP data of the subject 20 are used to generate data corresponding to the brain activity which corresponds to pain experienced by the subject 20”; John teaches receiving EEG/qEEG or qSEP data representing neural activity from a subject. Because John’s database includes population reference data from pluralities of individuals and the PDI is used to detect pain in subjects in clinical and non-clinical settings, John teaches or at least suggests receiving additional neural-activity data from the same subject or another subject for comparison to the database);
accessing the database using the additional data to determine whether a corresponding characterization of the neural activity of the test subject is found in the database (John, ¶[0015]: the subject’s EEG data may be compared to data in one or more databases, and the database may include self-normative data indicative of the subject’s brain-wave activity in the presence and absence of pain; John, ¶[0041]: “the CPU 25 compares the signals output by the DSP 21 to the reference data stored in the database 6, which may be stored locally in a memory in the PDI 16 (e.g., a nonvolatile memory) or remotely accessed via a network (e.g., the Internet, an intranet)”; John, ¶[0043]: “The qEEG analysis may reveal values of quantitative variables that exceed normative thresholds as identified in the database 6, indicative of the presence of pain”; John, ¶[0058]: “The brain wave data is compared to the reference data in the database 6, and a value corresponding to the pain index (i.e., corresponding to a level of pain experienced by the subject 20) is returned and output by the I/O arrangement 24”; John teaches accessing database 6 using received brain-wave data and comparing the data to stored reference data, including self-normative data for the subject, to determine a corresponding pain-related neural-activity characterization); and
when the corresponding characterization of the neural activity of the test subject is found in the database, outputting the corresponding characterization of the neural activity of the test subject found in the database (John, ¶[0015]: the database may include self-normative data indicative of the subject’s brain-wave activity in the presence and absence of pain; John, ¶[0041]: “As a result of the comparison, the CPU 25 outputs data indicating whether pain was detected and, if so, a magnitude of the pain (i.e., corresponding to the pain index)”; John, ¶[0042]: “the PDI 16 is coupled to a display for displaying brain wave data (i.e., EEG), the reference data and/or the pain index, etc. of the subject 20”; John, ¶[0058]: “The brain wave data is compared to the reference data in the database 6, and a value corresponding to the pain index (i.e., corresponding to a level of pain experienced by the subject 20) is returned and output by the I/O arrangement 24”; John teaches outputting the corresponding pain characterization obtained through comparison with database 6, including displaying the subject’s brain-wave data, corresponding reference data, and/or pain index in human-readable form).
Regarding claim 20, the modified John teaches wherein a diagnosis of pain comprises chronic pain or acute pain (John, ¶[0016]: "determining the presence and/or chronicity and/or intensity of pain experienced by the Subject as a function of the result data"; John, claim 3: "wherein the types of pain include chronic and acute"; John teaches determining the chronicity of pain from the subject's brain-wave data and expressly identifies chronic and acute as types of pain, thereby teaching a diagnosis of pain comprising chronic pain or acute pain).
Regarding claim 21, the modified John teaches or at least suggests wherein the diagnosis of pain consists of no pain or pain (John, ¶[0043]: “A first discriminant function A estimates a probability that pain is being experienced (A presence of pain) and a second discriminant function B estimates a probability that pain is not being experienced (B-absence of pain)”; ¶[0058]: “the brain activity data is compared to the reference data to determine whether pain exists,” and, only when pain is present, “the method may proceed to optional step 1210 where the pain is quantified”; John teaches a binary diagnostic determination of pain present or pain absent, with further pain quantification being optional, thereby teaching or at least suggesting a diagnosis limited to no pain or pain).
Claims 2-4, 7, 8, and 11 are rejected under 35 U.S.C. 103 as being unpatentable over John et al. (US-20080249430-A1), hereto referred as John, in view of Alotaiby et al. (Alotaiby, T., El-Samie, F. E. A., Alshebeili, S. A., & Ahmad, I. (2015). A review of channel selection algorithms for EEG signal processing. EURASIP Journal on Advances in Signal Processing, 2015(1), Article 66), hereto referred as Alotaiby, and further in view of Krishnan et al. (US-20060184475-A1), hereto referred as Krishnan, and further in view of Huupponen et al. (Huupponen, E., et al. "Optimization of Sigma Amplitude Threshold in Sleep Spindle Detection." J Sleep Res [Oxford UK], vol. 9, no. 4, December 2000, pp. 327-34), hereto referred as Huupponen.
The modified John teaches claim 1 as described above.
The modified John teaches claim 7 as described above.
Regarding claim 2, the modified John does not expressly teach wherein the spindle threshold analysis in the time domain comprises setting a threshold at a maximum value; reducing the threshold value repeatedly to obtain a current threshold value until the current threshold value is 0 microvolts; and detecting a spindle when a region of the data exceeds the current threshold value for at least about 0.5 seconds. Rather, the modified John teaches detecting pain-related brain-wave/qEEG patterns using FFT spectral analysis, frequency-domain power information, and wavelet analysis (John, ¶[0038]: “After artifact removal, the EEG may be subjected to spectral analysis using a signal processing technique such as, Fast Fourier Transform (FFT), wavelet analysis or fractal dimensional analysis”, ¶[0031]: “In the qEEG, the parameters may include, but are not limited to, power in a particular wide band frequency domain (e.g., delta, theta, alpha, beta, gamma), or in some particular frequency quantified using very narrow band (VNB) spectral analysis”). However, it does not expressly teach the particular spindle-threshold time-domain procedure recited in claim 2.
Huupponen is relied upon for threshold-based EEG spindle detection and threshold optimization. Huupponen teaches that sleep spindles are EEG waveforms and that “Spindle activity has been reported to reside between 10 and 16 Hz and spindles shorter than 0.5 s are usually ignored” (Huupponen, p. 327). Huupponen further teaches that spindle detectors commonly use a sigma amplitude threshold, stating that “Most of the methods use some form of sigma amplitude threshold” and that “[s]pindles are distinguished from background sigma activity using a threshold; a spindle is detected when the sigma amplitude exceeds the threshold value” (Huupponen, p. 327). Thus, Huupponen teaches the recited duration because spindles shorter than about 0.5 seconds are disregarded. Huupponen further teaches that, after the optimal threshold is determined, “[t]he recording was re-analysed with the level detector (using the optimal threshold from Step 3),” and that “[s]pindles were detected in 1-s intervals when the sigma amplitude x exceeded the optimal threshold value...” (Huupponen, p. 329).
Huupponen also teaches selecting or optimizing the amplitude threshold used in automatic spindle detection. Huupponen teaches that “[s]election of the threshold value is critical for the sensitivity of spindle detection” and that the use of the same fixed threshold value is problematic because “there is considerable intersubject variability in spindle amplitudes” and “[t]he spindle amplitudes may differ even between EEG channels of the same subject” (Huupponen, p. 327). Huupponen further teaches that “[a] valuable enhancement to a spindle detector using a threshold would be a method to automatically estimate the optimal value of sigma amplitude threshold to separate the spindles from background sigma activity” and that “[i]nstead of a fixed detection threshold, the optimal threshold is estimated for each individual to reduce interindividual and interchannel variability in EEG amplitudes” (Huupponen, p. 327). Thus, Huupponen teaches selecting or optimizing an amplitude threshold for automatic spindle detection based on the subject or recording rather than relying on a fixed threshold.
Huupponen further teaches evaluating different threshold values and selecting an optimal threshold from the amplitude values present in the recording. Huupponen teaches that the sigma amplitude is determined using the peak amplitude of each 1-s interval of the filtered signal (Huupponen, p. 328). Huupponen further teaches that the sigma amplitude range during nonspindle sleep and spindles differed across test recordings, including “3-25 μV during nonspindle sleep and 10-36 μV during spindles” in test recording 1 and “2.5-15 μV during nonspindle sleep and 7-16 μV during spindles” in test recording 4 (Huupponen, p. 330, Fig. 2 caption). Huupponen explains that ROC curves were determined to evaluate whether the thresholds were optimal, and that “[t]he ROC curves depict the true-positive rate as a function of the false-positive rate with different thresholds” (Huupponen, p. 330). Huupponen further states that “[t]hreshold estimation aims at selecting the optimal point on the ROC curve” and that “[t]he optimal threshold value to separate the spindles from background sigma activity ranged from 9.77 to 14.16 μV,” such that “[o]ne fixed value would not have been suitable for all the test recordings” (Huupponen, pp. 330, 332). Thus, Huupponen teaches evaluating different threshold values within the amplitude range and selecting an optimal threshold for detecting spindles.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John in view of Huupponen to include a spindle threshold analysis in the time domain comprising setting a threshold at a maximum amplitude value, evaluating candidate current threshold values by repeatedly reducing the threshold through the amplitude range until reaching 0 microvolts, and detecting a spindle when a region of EEG data exceeds a current threshold value for at least about 0.5 seconds. One skilled in the art would have been motivated to make the modification because John already teaches EEG-based pain assessment using brain-wave/qEEG data and known signal-processing techniques, while Huupponen teaches known automatic sleep spindle detection based on sigma amplitude thresholding, detecting a spindle when sigma amplitude exceeds the threshold value, disregarding spindle activity shorter than about 0.5 seconds, and evaluating different amplitude thresholds to optimize spindle detection. Because John’s pain assessment is based on detecting pain-related brain-wave/qEEG patterns from EEG data, one skilled in the art would have recognized Huupponen’s spindle-threshold technique as an applicable EEG pattern-detection technique for identifying spindle activity within the selected EEG time-series data before determining the pain diagnosis. John and Huupponen are analogous because both are directed to automated processing of EEG signal data to identify diagnostically or clinically relevant brain-wave patterns, and Huupponen’s spindle-threshold technique is reasonably pertinent to John’s problem of detecting pain-related EEG patterns using known signal-processing techniques. It would further have been obvious to implement Huupponen’s evaluation of different threshold values by setting the threshold initially at the maximum measured amplitude and repeatedly reducing the threshold through the full amplitude range to 0 microvolts. Huupponen expressly evaluates different thresholds applied to a peak-amplitude quantity reported in nonnegative microvolt values. Accordingly, 0 microvolts is the lower endpoint of the amplitude magnitude being thresholded, and traversing the finite range from the maximum amplitude to that endpoint would have been a predictable software implementation for evaluating candidate thresholds. The modification would have been a predictable use of a known EEG spindle-detection technique in John’s EEG-based pain-assessment framework. The benefit of the modification would have been providing John’s pain-detection system with an additional time-domain technique for detecting spindle activity from the selected EEG time-series data while permitting systematic evaluation of threshold values across recordings and subjects having different spindle amplitudes.
Regarding claim 3, the modified John teaches that the power spectrum analysis in the frequency domain comprises performing power spectrum analysis using a Fourier transform or a fast Fourier transform (John, ¶[0038]: “After artifact removal, the EEG may be subjected to spectral analysis using a signal processing technique such as, Fast Fourier Transform (FFT), wavelet analysis or fractal dimensional analysis”, ¶[0031]: “In the qEEG, the parameters may include, but are not limited to, power in a particular wide band frequency domain (e.g., delta, theta, alpha, beta, gamma), or in some particular frequency quantified using very narrow band (VNB) spectral analysis”; John teaches subjecting EEG data to spectral analysis using FFT and analyzing EEG power in frequency-domain bands, thereby teaching power spectrum analysis in the frequency domain using a fast Fourier transform).
Regarding claim 4, the modified John teaches that the wavelet analysis in the time-frequency domain comprises performing wavelet analysis using a short time Fourier transform or a wavelet transform (John, ¶[0038]: “After artifact removal, the EEG may be subjected to spectral analysis using a signal processing technique such as, Fast Fourier Transform (FFT), wavelet analysis or fractal dimensional analysis”; John expressly teaches subjecting EEG data to wavelet analysis after artifact removal).
Regarding claim 7, John teaches or at least suggests A computer program product for monitoring neural activity, the computer program product comprising a non-transitory computer readable storage having program instructions embodied therewith, the program instructions executable by a computer, to cause the computer to perform a method comprising: at a computer system comprising a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor (John, ¶[0017]: “these LORETA images can be computed by an instrument that uses an appropriate array of electrodes placed upon the scalp, in accordance with the International 10/20 Electrode Placement System (Jasper 1958) and implements all computations from a set of digital amplifiers controlled and analyzed by a personal desktop or laptop computer”, ¶[0041]: “The digital signals are transmitted to a digital signal processor (DSP) 21 which may be included in or electrically coupled to a central processing unit (CPU) 25”, ¶[0041]: “the database 6, which may be stored locally in a memory in the PDI 16 (e.g., a nonvolatile memory) or remotely accessed via a network (e.g., the Internet, an intranet)”; John teaches computer-based EEG pain-detection processing using a CPU/DSP arrangement and memory/database storage, including computation and analysis of EEG data by a personal desktop or laptop computer. Under the broadest reasonable interpretation, John’s computer-implemented EEG processing, pain-assessment algorithm, database comparison, classifier functions, and output operations teach or at least suggest program instructions stored in computer-readable memory and executable by a computer to cause the computer to perform the recited neural-activity monitoring method. To the extent John is argued not to expressly disclose the formal computer-program-product storage-medium format, Krishnan confirms that medical decision-support methods were conventionally implemented as “a computer readable storage media has stored therein data representing instructions executable by a programmed processor for addressing missing data in a medical decision support system” (Krishnan, ¶[0009]));
receiving data representing at least one signal representing neural activity of a test subject and accessing a plurality of real-time time-series data based on electrophysiological signals obtained using a plurality of electro-encephalogram (EEG) sensors gathering raw EEG data representing a plurality of different frequencies, signals, patterns, and noise from the test subject (John, ¶[0019]: “The PDI 16 is coupled to electrode(s) 8 to receive electrical data corresponding to brain activity of the Subject 20”, ¶[0032]: “The PDI 16 may be configured to receive data from any number and/or type of biosensor electrodes”, ¶[0038]: “After artifact removal, the EEG may be subjected to spectral analysis using a signal processing technique such as, Fast Fourier Transform (FFT), wavelet analysis or fractal dimensional analysis”, ¶[0031]: “In the qEEG, the parameters may include, but are not limited to, power in a particular wide band frequency domain (e.g., delta, theta, alpha, beta, gamma), or in some particular frequency quantified using very narrow band (VNB) spectral analysis”; John teaches receiving EEG electrical brain-activity data from multiple biosensor electrodes, converting and processing the received electrode signals using an amplifier, ADC, filtering arrangement, CPU, and DSP, removing artifacts from the EEG before spectral analysis, and analyzing the EEG across multiple frequency bands, thereby teaching receiving data representing neural activity and accessing EEG time-series signal data obtained from multiple EEG sensors and including frequency components and noise/artifacts before processing).
Also, regarding claim 7, John does not expressly teach comparing in real-time each of the plurality of time-series data with others of the time-series data to determine those time-series data having a highest variance; based on the comparison, selecting in real-time, at the computer system, time-series data obtained from at least some sensors of the plurality of sensors that show the highest variance in their signal. John teaches receiving EEG data from any number and/or type of biosensor electrodes (John, ¶[0019]: “The PDI 16 is coupled to electrode(s) 8 to receive electrical data corresponding to brain activity of the Subject 20”, ¶[0032]: “The PDI 16 may be configured to receive data from any number and/or type of biosensor electrodes”). However, John does not expressly teach selecting the EEG time-series data based on a comparison of channel variance.
Alotaiby is relied upon for this channel-selection feature. Alotaiby explains that EEG channel selection is used to reduce computational complexity, reduce overfitting from unnecessary channels, and reduce setup time while selecting relevant EEG channels for feature extraction (Alotaiby, p. 1, Abstract: “The main purpose of the channel selection process is threefold: (i) to reduce the computational complexity of any processing task performed on EEG signals by selecting the relevant channels and hence extracting the features of major importance, (ii) to reduce the amount of overfitting that may arise due to the utilization of unnecessary channels, for the purpose of improving the performance, and (iii) to reduce the setup time in some applications”). Alotaiby further teaches that channel selection may use EEG signal-processing tools including time-domain analysis, power spectral estimation, and wavelet transform (Alotaiby, p. 1, Abstract: “Signal processing tools such as time-domain analysis, power spectral estimation, and wavelet transform have been used for feature extraction and hence for channel selection in most of channel selection algorithms”). Most directly, Alotaiby teaches variance-based channel selection, disclosing that “the three channels with maximum variance are selected for signal classification” (Alotaiby, p. 6, Sec. 3.1.1). Thus, Alotaiby teaches comparing EEG channels based on variance and selecting the channels having maximum variance for EEG signal classification.
It would have been prima facie obvious before the effective filing date of the claimed invention to have modified John in view of Alotaiby to include Alotaiby’s variance-based channel selection as part of John’s computer-implemented EEG processing before John’s pain-related analysis. Because John’s system receives and processes EEG brain-activity data for pain assessment using the PDI/CPU/DSP arrangement, one skilled in the art would have understood that the variance-based channel selection would be implemented as part of the same computer-implemented EEG processing workflow before the diagnostic analysis, including for the accessed real-time EEG data. In the modified John system, Alotaiby’s maximum-variance channel-selection technique would have been performed on the EEG time-series data received from John’s multiple biosensor electrodes before John’s pain-related EEG analysis, such that the subsequent detection is performed using the selected maximum-variance EEG channel data. One skilled in the art would have been motivated to make the modification because John already receives EEG data from multiple biosensor electrodes for pain detection, while Alotaiby teaches that selecting relevant EEG channels reduces computational complexity, reduces overfitting, reduces setup time, and improves EEG classification performance. The modification would have been a predictable use of a known EEG signal-processing technique in John’s EEG-based medical diagnostic system because both references process EEG channel data for classification or detection tasks. The benefit of the modification would have been selecting more informative EEG channel data for John’s pain-detection classifier while reducing the amount of EEG data processed by John’s system.
Also, regarding claim 7, the modified John teaches a power spectrum analysis in the frequency domain, and a wavelet analysis in the time-frequency domain, but does not expressly teach detecting, at the computer system, a brain function pattern from the selected time-series data by analysis consisting of: a spindle threshold analysis in the time domain, a power spectrum analysis in the frequency domain, and a wavelet analysis in the time-frequency domain. John teaches detecting pain-related brain-wave/qEEG patterns using FFT spectral analysis, frequency-domain power information, and wavelet analysis (John, ¶[0038]: “After artifact removal, the EEG may be subjected to spectral analysis using a signal processing technique such as, Fast Fourier Transform (FFT), wavelet analysis or fractal dimensional analysis”, ¶[0031]: “In the qEEG, the parameters may include, but are not limited to, power in a particular wide band frequency domain (e.g., delta, theta, alpha, beta, gamma), or in some particular frequency quantified using very narrow band (VNB) spectral analysis”). However, the modified John does not expressly teach performing the recited spindle threshold analysis in addition to the power spectrum and wavelet analyses.
Huupponen is relied upon for the spindle threshold analysis. Huupponen teaches that sleep spindles are EEG waveforms and that “[m]ost of the methods use some form of sigma amplitude threshold” and “[s]pindles are distinguished from background sigma activity using a threshold; a spindle is detected when the sigma amplitude exceeds the threshold value” (Huupponen, p. 327). Huupponen further teaches determining sigma amplitude from successive 1-second intervals of filtered EEG data and detecting spindles using the amplitude threshold (Huupponen, pp. 328-329). Thus, Huupponen teaches a threshold-based time-domain spindle analysis of EEG data.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John in view of Huupponen to include Huupponen’s spindle threshold analysis together with John’s power spectrum and wavelet analyses. One skilled in the art would have been motivated to make the modification because John teaches extracting multiple qEEG features relevant to pain assessment and expressly teaches combining feature sets to obtain more sensitive sets of features for the pain-assessment classifier (John, ¶[0023]), while Huupponen teaches a known EEG analysis that identifies spindle activity using an amplitude threshold. The modification would have predictably provided an additional analyzed EEG feature for use in John’s multi-feature pain-assessment framework. Thus, in the further modified John, the recited analysis would consist of the spindle threshold, power spectrum, and wavelet analyses.
Also, regarding claim 7, the further modified John teaches or at least suggests combining the analysis into a single dataset; wherein the computer system further comprises a database; determining whether a comparison of the single dataset versus the database indicates an available pain diagnosis. John teaches that, after processing the subject’s EEG data, “a selected set of features is extracted from the digital data of acceptable quality” and that “a selected set of features from the qEEG data which are particularly relevant to assessment of pain experienced are extracted from the overall data set obtained and compared to reference data” (John, ¶[0022]). John further teaches computing univariate or multivariate data features for the extracted features (John, ¶[0022]). John teaches that feature sets may be used “in combination” to obtain a more sensitive set of features for use in the pain-assessment classifier (John, ¶[0023]). John further teaches database 6 containing the reference data used for pain assessment (John, ¶[0041]) and that “the brain activity data is compared to the reference data to determine whether pain exists” (John, ¶[0058]).
It would have been prima facie obvious to one of ordinary skill in the art to include the spindle-threshold result supplied by Huupponen together with the power-spectrum and wavelet-derived EEG features in John’s selected set of analyzed qEEG features. John expressly teaches combining qEEG feature sets to obtain a more sensitive set of features for pain assessment, and Huupponen’s spindle-threshold result would constitute another analyzed EEG feature available from the same EEG data. Combining the three analysis results into the single qEEG feature dataset used for John’s database comparison would therefore have been a predictable application of John’s express multi-feature approach and would have provided the benefit identified by John of producing a more sensitive feature set for the pain-assessment classifier. The further modified John therefore teaches or at least suggests combining the recited analyses into a single dataset and comparing that single dataset to the database to determine whether a pain diagnosis is available.
Also, regarding claim 7, the modified John does not expressly teach based on the determination, when a pain diagnosis is not available, generating an instruction to generate a training dataset for a machine learning model and training the machine learning model using the training dataset. The modified John teaches applying classifier functions, including neural networks, to assess pain from EEG data (John, ¶[0025]: “Those skilled in the art will understand that, in other embodiments any or a combination of discriminant functions, cluster algorithms, neural networks and/or other classifier functions will be applied to the data for the purpose of assessing pain”). However, the modified John does not expressly teach conditionally generating or selecting training data and training a classifier when comparison of the current single qEEG feature dataset to the existing pain database/classifier framework does not provide an available diagnosis for the current patient data.
Krishnan is relied upon for this conditional classifier-training feature. Krishnan is directed to medical decision support systems that address missing or unavailable patient-record information by assigning, constructing, and training classifiers based on the features available for the current patient record. Krishnan teaches that a classifier may be constructed for a new patient record, or only when an available classifier does not have the corresponding feature set for that patient record (Krishnan, ¶[0026]: “The classifier is constructed for each instance of a new patient record to be analyzed or only where another available classifier does not have a corresponding feature set”). Krishnan further teaches constructing the classifier from training data and training the classifier from a selected subset of that training data (Krishnan, ¶[0043]: “In another embodiment, the processor 12 obtains the classifier by constructing the classifier from training data”, “The processor 12 trains a classifier from the selected sub-set of the training data”). Krishnan explains that the selected features correspond to the features available for the current patient record, and that the processor constructs a classifier after selecting training-set patient-record information (Krishnan, ¶[0056]: “The selected features are the set or a sub-set of the features available for the current patient record”, ¶[0058]: “After selecting the training set patient record information, the processor constructs a classifier from the selected set of features from the training set patient records”). Thus, Krishnan teaches a medical decision-support technique in which training data are selected and a classifier is trained when the available classifier structure does not correspond to the available patient data.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John to incorporate Krishnan’s conditional medical classifier-training technique so that, when John’s pain database/classifier framework does not include a classifier or reference-data feature set corresponding to the current single qEEG feature dataset, the system generates or selects a training-data subset and trains a machine learning classifier using that training data. In the further modified John system, the absence of a corresponding classifier or reference-data feature set in John’s pain-diagnosis framework would trigger Krishnan’s construction and training of a classifier from selected training data, thereby providing the claimed conditional training operation when comparison of the current single qEEG feature dataset to the database does not provide an available pain diagnosis. One skilled in the art would have been motivated to make the modification because John and Krishnan are both directed to computer-implemented medical diagnostic systems that use patient data and classifier functions to produce diagnostic information. John teaches assessing pain using patient EEG data, reference data, and classifier functions, including neural networks. Krishnan teaches a known way to handle a patient record that is not adequately covered by an available classifier, namely constructing and training a classifier from selected training data corresponding to the available patient information. The modification would have been a predictable use of Krishnan’s medical classifier-training technique in John’s classifier-based pain-assessment framework. The benefit of the modification would have been improved diagnostic coverage for patient EEG feature data that are not adequately handled by John’s existing pain database or classifier framework.
Also, regarding claim 7, the modified John does not expressly teach determining whether the time-series data indicates a pain diagnosis based on a detected brain function pattern using the trained machine learning model.
Rather, the modified John teaches determining whether pain exists based on brain-wave/qEEG data (John, ¶[0058]: “the brain activity data is compared to the reference data to determine whether pain exists”). John also teaches that neural networks and other classifier functions may be applied to the data for assessing pain (John, ¶[0025]: “Those skilled in the art will understand that, in other embodiments any or a combination of discriminant functions, cluster algorithms, neural networks and/or other classifier functions will be applied to the data for the purpose of assessing pain”). However, the modified John does not expressly teach that the pain determination is made using the conditionally trained machine learning model as previously recited.
Krishnan teaches training a classifier from selected training data and applying the trained classifier to output diagnostic information for the current patient record (Krishnan, ¶[0043]: “The processor 12 trains a classifier from the selected sub-set of the training data”; Krishnan, ¶[0027]: “Applying the classifier determines a diagnosis. The classifier outputs a diagnosis, such as a conclusion, probability, location of concern or other information to assist with diagnosis”; Krishnan, ¶[0062]: “The classification is performed with neural network, filter, algorithm, or other now-known or later developed classifier or classification technique. The classifier is configured or trained for distinguishing between the desired States”). Thus, John teaches the pain determination and Krishnan teaches that a trained classifier is used to output the diagnostic classification.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John in view of Krishnan to use Krishnan’s trained classifier as the classifier in John’s pain determination. One skilled in the art would have been motivated to make the modification because John already contemplates classifier-based pain assessment, including neural networks and other classifier functions, and Krishnan provides a known medical decision-support technique for training and applying a classifier when the current patient data are not adequately handled by an existing classifier. The modification would have been a predictable use of Krishnan’s trained medical classifier within John’s pain-assessment framework. The benefit of the modification would have been enabling John’s system to determine a pain diagnosis using a classifier trained for the available patient data when the existing database/classifier framework does not provide an available diagnosis. Thus, in the further modified John system, John’s pain determination would have been made using the Krishnan-trained classifier when the classifier was constructed and trained for the available patient EEG data.
The modified John further teaches outputting the indicated pain diagnosis in a human readable form, wherein the pain diagnosis comprises no pain or active pain (John, ¶[0042]: “the PDI 16 is coupled to a display for displaying brain wave data (i.e., EEG), the reference data and/or the pain index, etc. of the subject 20”; John, ¶[0058]: “the brain activity data is compared to the reference data to determine whether pain exists” and “a value corresponding to the pain index (i.e., corresponding to a level of pain experienced by the subject 20) is returned and output by the I/O arrangement 24”; John teaches outputting a determination of whether pain is present or absent and a corresponding pain index, thereby teaching an indicated diagnosis of no pain or active pain in human-readable form).
Regarding claim 8, the modified John does not expressly teach that the spindle threshold analysis in the time domain comprises: setting a threshold at a maximum value; reducing the threshold repeatedly to obtain a current threshold until the threshold is at a minimum value; and detecting a spindle when a region of the data exceeds the current threshold value for at least a predetermined amount of time. Rather, the modified John teaches detecting pain-related brain-wave/qEEG patterns using FFT spectral analysis, frequency-domain power information, and wavelet analysis (John, ¶[0038]: “After artifact removal, the EEG may be subjected to spectral analysis using a signal processing technique such as, Fast Fourier Transform (FFT), wavelet analysis or fractal dimensional analysis”; ¶[0031]: “In the qEEG, the parameters may include, but are not limited to, power in a particular wide band frequency domain (e.g., delta, theta, alpha, beta, gamma), or in some particular frequency quantified using very narrow band (VNB) spectral analysis”). However, the modified John does not expressly teach the particular spindle-threshold time-domain procedure recited in claim 8.
Huupponen is relied upon for threshold-based EEG spindle detection and threshold optimization. Huupponen teaches that sleep spindles are EEG waveforms and that “Spindle activity has been reported to reside between 10 and 16 Hz and spindles shorter than 0.5 s are usually ignored” (Huupponen, p. 327). Huupponen further teaches that spindle detectors commonly use a sigma amplitude threshold, stating that “Most of the methods use some form of sigma amplitude threshold” and that “[s]pindles are distinguished from background sigma activity using a threshold; a spindle is detected when the sigma amplitude exceeds the threshold value” (Huupponen, p. 327). Thus, Huupponen teaches a predetermined duration aspect because spindles shorter than 0.5 seconds are usually ignored and the level detector detects spindles in 1-second intervals when the sigma amplitude exceeds the optimal threshold. Huupponen further teaches that, after the optimal threshold is determined, “[t]he recording was re-analysed with the level detector (using the optimal threshold from Step 3),” and that “[s]pindles were detected in 1-s intervals when the sigma amplitude x exceeded the optimal threshold value...” (Huupponen, p. 329).
Huupponen also teaches selecting or optimizing the amplitude threshold used in automatic spindle detection. Huupponen teaches that “[s]election of the threshold value is critical for the sensitivity of spindle detection” and that the use of the same fixed threshold value is problematic because “there is considerable intersubject variability in spindle amplitudes” and “[t]he spindle amplitudes may differ even between EEG channels of the same subject” (Huupponen, p. 327). Huupponen further teaches that “[a] valuable enhancement to a spindle detector using a threshold would be a method to automatically estimate the optimal value of sigma amplitude threshold to separate the spindles from background sigma activity” and that “[i]nstead of a fixed detection threshold, the optimal threshold is estimated for each individual to reduce interindividual and interchannel variability in EEG amplitudes” (Huupponen, p. 327). Thus, Huupponen teaches selecting or optimizing an amplitude threshold for automatic spindle detection based on the subject or recording rather than relying on a fixed threshold.
Huupponen further teaches evaluating different threshold values and selecting an optimal threshold from the amplitude values present in the recording. Huupponen teaches that the sigma amplitude range during nonspindle sleep and spindles differed across test recordings, including “3-25 μV during nonspindle sleep and 10-36 μV during spindles” in test recording 1 and “2.5-15 μV during nonspindle sleep and 7-16 μV during spindles” in test recording 4 (Huupponen, p. 330, Fig. 2 caption). Huupponen explains that ROC curves were determined to evaluate whether the thresholds were optimal, and that “[t]he ROC curves depict the true-positive rate as a function of the false-positive rate with different thresholds” (Huupponen, p. 330). Huupponen further states that “[t]hreshold estimation aims at selecting the optimal point on the ROC curve” and that “[t]he optimal threshold value to separate the spindles from background sigma activity ranged from 9.77 to 14.16 μV,” such that “[o]ne fixed value would not have been suitable for all the test recordings” (Huupponen, pp. 330, 332). Thus, Huupponen teaches evaluating different threshold values within amplitude ranges and selecting an optimal threshold for detecting spindles.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John in view of Huupponen to include a spindle threshold analysis in the time domain comprising setting a threshold at a high or maximum amplitude value, evaluating candidate current threshold values by repeatedly reducing the threshold toward a lower or minimum amplitude value, and detecting a spindle when a region of EEG data exceeds a current threshold value for at least a predetermined amount of time. One skilled in the art would have been motivated to make the modification because John already teaches EEG-based pain assessment using brain-wave/qEEG data and known signal-processing techniques, while Huupponen teaches known automatic sleep spindle detection based on sigma amplitude thresholding, detecting a spindle when sigma amplitude exceeds the threshold value, and ignoring spindles shorter than 0.5 seconds. Because John’s pain assessment is based on detecting pain-related brain-wave/qEEG patterns from EEG data, one skilled in the art would have recognized Huupponen’s spindle-threshold technique as an applicable EEG pattern-detection technique for identifying spindle activity within the selected EEG time-series data before determining the pain diagnosis. John and Huupponen are analogous because both are directed to automated processing of EEG signal data to identify diagnostically or clinically relevant brain-wave patterns, and Huupponen’s spindle-threshold technique is reasonably pertinent to John’s problem of detecting pain-related EEG patterns using known signal-processing techniques. It would further have been obvious to implement Huupponen’s evaluation of different threshold values by setting the threshold initially at a high or maximum amplitude value and repeatedly reducing the threshold toward a lower or minimum amplitude value because Huupponen teaches that threshold selection is critical to spindle-detection sensitivity, that one fixed threshold is unsuitable across recordings, that spindle and nonspindle sigma amplitudes occupy different amplitude ranges, and that ROC performance is evaluated using different thresholds to select an optimal threshold point. In view of those teachings, a software implementation that traverses the amplitude range from a maximum threshold toward a minimum threshold would have been one of a finite number of predictable ways to evaluate candidate threshold values and identify spindle events meeting the threshold and duration criteria. This is consistent with the Instant Application’s description of the threshold traversal as a typical threshold-setting method and the 10% step size as an example, without identifying a special technical effect tied to the maximum-to-minimum traversal direction or the particular decrement size (Instant Application, ¶[0031]). The modification would have been a predictable use of a known EEG spindle-detection technique in John’s EEG-based pain-assessment framework. The benefit of the modification would have been providing John’s pain-detection system with an additional time-domain technique for detecting spindle activity from the selected EEG time-series data while improving spindle detection sensitivity across recordings and subjects having different spindle amplitudes.
Regarding claim 11, the modified John teaches or at least suggests storing data related to the neural activity of the test subject in the database (John, ¶[0015]: “The database may further include self-normative data indicative of the subject’s brain wave activity in the presence and absence of pain” and “The database may further include medical histories and physical and/or neurological examination results for individuals in the population and/or the subject”; John, ¶[0041]: “the CPU 25 compares the signals output by the DSP 21 to the reference data stored in the database 6, which may be stored locally in a memory in the PDI 16 (e.g., a nonvolatile memory) or remotely accessed via a network (e.g., the Internet, an intranet)”; John teaches storing in database 6 data related to the test subject’s brain-wave activity, including self-normative brain-wave data for the subject);
receiving additional data representing at least one signal representing additional neural activity of the test subject or neural activity of another test subject (John, ¶[0019]: “The PDI 16 is coupled to electrode(s) 8 to receive electrical data corresponding to brain activity of the subject 20”; John, ¶[0032]: “The PDI 16 may be configured to receive data from any number and/or type of biosensor electrodes”; John, ¶[0057]: “the qEEG and/or the qSEP data of the subject 20 are used to generate data corresponding to the brain activity which corresponds to pain experienced by the subject 20”; John teaches receiving EEG/qEEG or qSEP data representing neural activity from a subject. Because John’s database includes population reference data from pluralities of individuals and the PDI is used to detect pain in subjects in clinical and non-clinical settings, John teaches or at least suggests receiving additional neural-activity data from the same subject or another subject for comparison to the database);
accessing the database using the additional data to determine whether a corresponding characterization of the neural activity of the test subject is found in the database (John, ¶[0015]: the subject’s EEG data may be compared to data in one or more databases, and the database may include self-normative data indicative of the subject’s brain-wave activity in the presence and absence of pain; John, ¶[0041]: “the CPU 25 compares the signals output by the DSP 21 to the reference data stored in the database 6, which may be stored locally in a memory in the PDI 16 (e.g., a nonvolatile memory) or remotely accessed via a network (e.g., the Internet, an intranet)”; John, ¶[0043]: “The qEEG analysis may reveal values of quantitative variables that exceed normative thresholds as identified in the database 6, indicative of the presence of pain”; John, ¶[0058]: “The brain wave data is compared to the reference data in the database 6, and a value corresponding to the pain index (i.e., corresponding to a level of pain experienced by the subject 20) is returned and output by the I/O arrangement 24”; John teaches accessing database 6 using received brain-wave data and comparing the data to stored reference data, including self-normative data for the subject, to determine a corresponding pain-related neural-activity characterization); and
when the corresponding characterization of the neural activity of the test subject is found in the database, outputting the corresponding characterization of the neural activity of the test subject found in the database (John, ¶[0015]: the database may include self-normative data indicative of the subject’s brain-wave activity in the presence and absence of pain; John, ¶[0041]: “As a result of the comparison, the CPU 25 outputs data indicating whether pain was detected and, if so, a magnitude of the pain (i.e., corresponding to the pain index)”; John, ¶[0042]: “the PDI 16 is coupled to a display for displaying brain wave data (i.e., EEG), the reference data and/or the pain index, etc. of the subject 20”; John, ¶[0058]: “The brain wave data is compared to the reference data in the database 6, and a value corresponding to the pain index (i.e., corresponding to a level of pain experienced by the subject 20) is returned and output by the I/O arrangement 24”; John teaches outputting the corresponding pain characterization obtained through comparison with database 6, including displaying the subject’s brain-wave data, corresponding reference data, and/or pain index in human-readable form).
Claims 6, 19, and 25 are rejected under 35 U.S.C. 103 as being unpatentable over John et al. (US-20080249430-A1), hereto referred as John, in view of Alotaiby et al. (Alotaiby, T., El-Samie, F. E. A., Alshebeili, S. A., & Ahmad, I. (2015). A review of channel selection algorithms for EEG signal processing. EURASIP Journal on Advances in Signal Processing, 2015(1), Article 66), hereto referred as Alotaiby, and further in view of Krishnan et al. (US-20060184475-A1), hereto referred as Krishnan, and further in view of Pizzagalli et al. (Diego A. Pizzagalli. “Electroencephalography and High-Density Electrophysiological Source Localization,” Handbook of Psychophysiology, 2007), hereto referred as Pizzagalli.
The modified John teaches claim 1 as described above.
The modified John teaches claim 13 as described above.
Regarding claim 6, the modified John teaches or at least suggests before the comparing, processing each of the plurality of time-series data to perform: cleaning the raw EEG data from each EEG sensor to remove unwanted signals and noise to form cleaned EEG data from each sensor (John, ¶[0032]: “The PDI 16 may be configured to receive data from any number and/or type of biosensor electrodes”; ¶[0045]: “the present invention utilizes algorithms to minimize any impact of such contaminating events”; ¶[0045]: “The EEG is filtered to discard segments of the electrical signals containing voltages larger than the threshold”; ¶[0045]: “After non-brain derived electrical activity has been removed from the recordings by these or similar denoising methods, the residual electrical signals may be assumed to be substantially artifact-free and are compiled to form a continuous, artifact-free sample of brain electrical activity to be used for further qEEG or qSEP processing”; John teaches receiving EEG data from multiple biosensor electrodes and removing artifact or non-brain-derived electrical activity from the EEG recordings before further qEEG processing, thereby teaching cleaning raw EEG data from the EEG sensors to remove unwanted artifact/noise signals and form substantially artifact-free cleaned EEG data);
band-pass filtering the cleaned EEG data from each sensor at the computer system by frequency filtering the cleaned EEG data from each sensor to analyze neural oscillation frequencies to obtain filtered EEG data from each sensor including neural oscillation frequencies (John, ¶[0053]: “the PDI 16 may utilize a digital comb filter to improve a signal-to-noise ratio (SNR) of the electrical signals gathered by the electrode(s) 8... In the digital comb filter, band pass frequencies are selected to form teeth of the comb using a phase variance parameter timed to correspond to the presence and absence of the pain events... As understood by those of skill in the art, optimal digital filtering may be an alternative to signal averaging for improving the SNR”; ¶[0038]: “After artifact removal, the EEG may be subjected to spectral analysis using a signal processing technique such as, Fast Fourier Transform (FFT), wavelet analysis or fractal dimensional analysis”; ¶[0038]: “The EEG power spectrum may be divided into wide bands (e.g., low delta (0.5-1.5 Hz), high delta (1.5-3.5 Hz), theta (3.5-7.5 Hz), low alpha (7.15-10 Hz), high alpha (10.0-12.5 Hz), beta (12.5-25 Hz), low gamma (25-35 Hz), high gamma (35-50 Hz) and ultra high (50-200 Hz)”; ¶[0039]: “Alternatively, the spectrum may be divided into very narrow bands (VNB) of approximately 0.39-0.50 Hz in width”; John teaches using a digital comb filter in which band-pass frequencies are selected, and further teaches post-artifact-removal spectral analysis in which the EEG power spectrum is divided into wide frequency bands and very narrow frequency bands, thereby teaching or at least suggesting frequency filtering the cleaned EEG data from the EEG sensors to analyze neural oscillation frequencies and obtain filtered EEG data including neural oscillation frequencies).
Also, regarding claim 6, the modified John does not expressly teach removing, at the computer system, artifacts from the filtered EEG data from each sensor by averaging the filtered EEG data from each sensor with filtered EEG data from at least one other sensor. John teaches removing artifacts and non-brain-derived electrical activity from EEG recordings using denoising methods (John, ¶[0045]: “After non-brain derived electrical activity has been removed from the recordings by these or similar denoising methods, the residual electrical signals may be assumed to be substantially artifact-free and are compiled to form a continuous, artifact-free sample of brain electrical activity to be used for further qEEG or qSEP processing”). However, John does not expressly teach removing artifacts by averaging filtered EEG data from a sensor with filtered EEG data from at least one other sensor.
Pizzagalli is relied upon for this EEG artifact-removal/interpolation feature. Pizzagalli teaches that corrupted EEG electrodes due to artifacts or technical malfunction should not simply be omitted because source localization relies on the scalp potential distribution, and that interpolation methods are used to address corrupted EEG signals (Pizzagalli, p. 62, 4.1.2: “The importance of spatial sampling for source localization is associated with the issue of how best to deal with electrodes with corrupted EEG signals due to excessive artifacts or technical malfunction”; p. 62, 4.1.2: “Consequently, simply omitting corrupted electrodes is not a feasible solution, and interpolation methods are needed”). Pizzagalli further teaches a linear nearest-neighbor interpolation method in which corrupted activity is reconstructed by averaging data from neighboring electrodes (Pizzagalli, p. 62, 4.1.2: “Broadly speaking, two interpolation methods have been utilized: linear (nearest neighbor) and spline interpolation methods”; p. 62, 4.1.2: “In the first method, corrupted activity is reconstructed through a weighted average using data from neighboring electrodes (the weights are proportional to the Euclidian distance between the electrodes)”). Thus, under the broadest reasonable interpretation, Pizzagalli teaches removing or correcting corrupted/artifact EEG activity from a sensor/electrode channel by reconstructing the corrupted activity through a weighted average using data from neighboring electrodes, thereby teaching removing artifacts from EEG data from a sensor by averaging that sensor’s EEG data with EEG data from at least one other sensor.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John in view of Pizzagalli to remove artifacts from filtered EEG data by reconstructing artifact-affected or corrupted electrode activity using a weighted average of data from neighboring electrodes. One skilled in the art would have been motivated to make the modification because John already teaches EEG-based pain assessment using multiple biosensor electrodes and denoising/artifact-removal methods before qEEG processing, while Pizzagalli teaches that simply omitting corrupted EEG electrodes is not feasible when scalp potential distribution is important and that corrupted EEG activity may be reconstructed using a weighted average of neighboring electrode data. Under the broadest reasonable interpretation, the claim does not require a common-average reference or uniform averaging of every electrode channel, and Pizzagalli’s reconstruction of corrupted electrode activity using a weighted average of neighboring electrode data teaches or at least suggests the claimed artifact-removal mechanism. John and Pizzagalli are analogous because both concern processing multi-electrode EEG data and improving the reliability of EEG signal analysis by addressing artifacts or corrupted electrode signals. The modification would have been a predictable use of a known EEG artifact-correction technique in John’s EEG-based pain-assessment system. Because John performs artifact removal and frequency-domain EEG processing before subsequent qEEG/pain analysis, one skilled in the art would have found it obvious to apply Pizzagalli’s neighboring-electrode reconstruction to the filtered EEG sensor data in the preprocessing workflow before Alotaiby’s channel selection and John’s later pain-related analysis. The benefit of the modification would have been preserving useful multi-electrode EEG data for John’s qEEG/pain-detection processing while reducing the impact of artifacts or corrupted electrode signals before Alotaiby’s variance-based channel selection and John’s subsequent pain-related EEG analysis.
Regarding claim 19, the modified John teaches or at least suggests before the comparing, processing each of the plurality of time-series data to perform: cleaning the raw EEG data from each EEG sensor to remove unwanted signals and noise to form cleaned EEG data from each sensor (John, ¶[0032]: “The PDI 16 may be configured to receive data from any number and/or type of biosensor electrodes”; ¶[0045]: “the present invention utilizes algorithms to minimize any impact of such contaminating events”; ¶[0045]: “The EEG is filtered to discard segments of the electrical signals containing voltages larger than the threshold”; ¶[0045]: “After non-brain derived electrical activity has been removed from the recordings by these or similar denoising methods, the residual electrical signals may be assumed to be substantially artifact-free and are compiled to form a continuous, artifact-free sample of brain electrical activity to be used for further qEEG or qSEP processing”; John teaches receiving EEG data from multiple biosensor electrodes and removing artifact or non-brain-derived electrical activity from the EEG recordings before further qEEG processing, thereby teaching cleaning raw EEG data from the EEG sensors to remove unwanted artifact/noise signals and form substantially artifact-free cleaned EEG data);
band-pass filtering the cleaned EEG data from each sensor at the computer system by frequency filtering the cleaned EEG data from each sensor to analyze neural oscillation frequencies to obtain filtered EEG data from each sensor including neural oscillation frequencies (John, ¶[0053]: “the PDI 16 may utilize a digital comb filter to improve a signal-to-noise ratio (SNR) of the electrical signals gathered by the electrode(s) 8... In the digital comb filter, band pass frequencies are selected to form teeth of the comb using a phase variance parameter timed to correspond to the presence and absence of the pain events... As understood by those of skill in the art, optimal digital filtering may be an alternative to signal averaging for improving the SNR”; ¶[0038]: “After artifact removal, the EEG may be subjected to spectral analysis using a signal processing technique such as, Fast Fourier Transform (FFT), wavelet analysis or fractal dimensional analysis”; ¶[0038]: “The EEG power spectrum may be divided into wide bands (e.g., low delta (0.5-1.5 Hz), high delta (1.5-3.5 Hz), theta (3.5-7.5 Hz), low alpha (7.15-10 Hz), high alpha (10.0-12.5 Hz), beta (12.5-25 Hz), low gamma (25-35 Hz), high gamma (35-50 Hz) and ultra high (50-200 Hz)”; ¶[0039]: “Alternatively, the spectrum may be divided into very narrow bands (VNB) of approximately 0.39-0.50 Hz in width”; John teaches using a digital comb filter in which band-pass frequencies are selected, and further teaches post-artifact-removal spectral analysis in which the EEG power spectrum is divided into wide frequency bands and very narrow frequency bands, thereby teaching or at least suggesting frequency filtering the cleaned EEG data from the EEG sensors to analyze neural oscillation frequencies and obtain filtered EEG data including neural oscillation frequencies).
Also, regarding claim 19, the modified John does not expressly teach removing, at the computer system, artifacts from the filtered EEG data from each sensor by averaging the filtered EEG data from each sensor with filtered EEG data from at least one other sensor. John teaches removing artifacts and non-brain-derived electrical activity from EEG recordings using denoising methods (John, ¶[0045]: “After non-brain derived electrical activity has been removed from the recordings by these or similar denoising methods, the residual electrical signals may be assumed to be substantially artifact-free and are compiled to form a continuous, artifact-free sample of brain electrical activity to be used for further qEEG or qSEP processing”). However, John does not expressly teach removing artifacts by averaging filtered EEG data from a sensor with filtered EEG data from at least one other sensor.
Pizzagalli is relied upon for this EEG artifact-removal/interpolation feature. Pizzagalli teaches that corrupted EEG electrodes due to artifacts or technical malfunction should not simply be omitted because source localization relies on the scalp potential distribution, and that interpolation methods are used to address corrupted EEG signals (Pizzagalli, p. 62, 4.1.2: “The importance of spatial sampling for source localization is associated with the issue of how best to deal with electrodes with corrupted EEG signals due to excessive artifacts or technical malfunction”; p. 62, 4.1.2: “Consequently, simply omitting corrupted electrodes is not a feasible solution, and interpolation methods are needed”). Pizzagalli further teaches a linear nearest-neighbor interpolation method in which corrupted activity is reconstructed by averaging data from neighboring electrodes (Pizzagalli, p. 62, 4.1.2: “Broadly speaking, two interpolation methods have been utilized: linear (nearest neighbor) and spline interpolation methods”; p. 62, 4.1.2: “In the first method, corrupted activity is reconstructed through a weighted average using data from neighboring electrodes (the weights are proportional to the Euclidian distance between the electrodes)”). Thus, under the broadest reasonable interpretation, Pizzagalli teaches removing or correcting corrupted/artifact EEG activity from a sensor/electrode channel by reconstructing the corrupted activity through a weighted average using data from neighboring electrodes, thereby teaching removing artifacts from EEG data from a sensor by averaging that sensor’s EEG data with EEG data from at least one other sensor.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John in view of Pizzagalli to remove artifacts from filtered EEG data by reconstructing artifact-affected or corrupted electrode activity using a weighted average of data from neighboring electrodes. One skilled in the art would have been motivated to make the modification because John already teaches EEG-based pain assessment using multiple biosensor electrodes and denoising/artifact-removal methods before qEEG processing, while Pizzagalli teaches that simply omitting corrupted EEG electrodes is not feasible when scalp potential distribution is important and that corrupted EEG activity may be reconstructed using a weighted average of neighboring electrode data. Under the broadest reasonable interpretation, the claim does not require a common-average reference or uniform averaging of every electrode channel, and Pizzagalli’s reconstruction of corrupted electrode activity using a weighted average of neighboring electrode data teaches or at least suggests the claimed artifact-removal mechanism. John and Pizzagalli are analogous because both concern processing multi-electrode EEG data and improving the reliability of EEG signal analysis by addressing artifacts or corrupted electrode signals. The modification would have been a predictable use of a known EEG artifact-correction technique in John’s EEG-based pain-assessment system. Because John performs artifact removal and frequency-domain EEG processing before subsequent qEEG/pain analysis, one skilled in the art would have found it obvious to apply Pizzagalli’s neighboring-electrode reconstruction to the filtered EEG sensor data in the preprocessing workflow before Alotaiby’s channel selection and John’s later pain-related analysis. The benefit of the modification would have been preserving useful multi-electrode EEG data for John’s qEEG/pain-detection processing while reducing the impact of artifacts or corrupted electrode signals before Alotaiby’s variance-based channel selection and John’s subsequent pain-related EEG analysis.
Regarding claim 25, the modified John further teaches or at least suggests wherein cleaning the raw EEG data comprises: determining EEG signals from a plurality of EEG sensors that are nearest neighbors; and generating a weighted average of the EEG signals from the plurality of nearest neighbor EEG sensors, wherein weights of each of the EEG signals are proportional to a Euclidian distance between the plurality of nearest neighbor EEG sensors. As discussed regarding claim 6, Pizzagalli teaches a “linear (nearest neighbor)” interpolation method in which “corrupted activity is reconstructed through a weighted average using data from neighboring electrodes (the weights are proportional to the Euclidian distance between the electrodes)” (Pizzagalli, p. 62, Sec. 4.1.2). Thus, the modified John teaches the additional limitations of claim 25.
It would have been prima facie obvious before the effective filing date of the claimed invention to have applied Pizzagalli's nearest-neighbor weighted-average reconstruction to the raw EEG data as part of John's cleaning operation because John removes non-brain-derived electrical activity to form a "continuous, artifact-free sample of brain electrical activity to be used for further qEEG or qSEP processing" (John, ¶[0045]), and Pizzagalli teaches that "simply omitting corrupted electrodes is not a feasible solution, and interpolation methods are needed" (Pizzagalli, p. 62, Sec. 4.1.2). Applying the reconstruction when the raw data are first cleaned would have predictably provided corrected electrode data to the subsequent filtering and analysis steps.
Claim 12 is rejected under 35 U.S.C. 103 as being unpatentable over John et al. (US-20080249430-A1), hereto referred as John, in view of Alotaiby et al. (Alotaiby, T., El-Samie, F. E. A., Alshebeili, S. A., & Ahmad, I. (2015). A review of channel selection algorithms for EEG signal processing. EURASIP Journal on Advances in Signal Processing, 2015(1), Article 66), hereto referred as Alotaiby, and further in view of Krishnan et al. (US-20060184475-A1), hereto referred as Krishnan, and further in view of Huupponen et al. (Huupponen, E., et al. "Optimization of Sigma Amplitude Threshold in Sleep Spindle Detection." J Sleep Res [Oxford UK], vol. 9, no. 4, December 2000, pp. 327-34), hereto referred as Huupponen, and further in view of Pizzagalli et al. (Diego A. Pizzagalli. “Electroencephalography and High-Density Electrophysiological Source Localization,” Handbook of Psychophysiology, 2007), hereto referred as Pizzagalli.
The modified John teaches claim 7 as described above.
Regarding claim 12, the modified John teaches or at least suggests before the comparing, processing each of the plurality of time-series data to perform: cleaning the raw EEG data from each EEG sensor to remove unwanted signals and noise to form cleaned EEG data from each sensor (John, ¶[0032]: “The PDI 16 may be configured to receive data from any number and/or type of biosensor electrodes”; ¶[0045]: “the present invention utilizes algorithms to minimize any impact of such contaminating events”; ¶[0045]: “The EEG is filtered to discard segments of the electrical signals containing voltages larger than the threshold”; ¶[0045]: “After non-brain derived electrical activity has been removed from the recordings by these or similar denoising methods, the residual electrical signals may be assumed to be substantially artifact-free and are compiled to form a continuous, artifact-free sample of brain electrical activity to be used for further qEEG or qSEP processing”; John teaches receiving EEG data from multiple biosensor electrodes and removing artifact or non-brain-derived electrical activity from the EEG recordings before further qEEG processing, thereby teaching cleaning raw EEG data from the EEG sensors to remove unwanted artifact/noise signals and form substantially artifact-free cleaned EEG data);
band-pass filtering the cleaned EEG data from each sensor at the computer system by frequency filtering the cleaned EEG data from each sensor to analyze neural oscillation frequencies to obtain filtered EEG data from each sensor including neural oscillation frequencies (John, ¶[0053]: “the PDI 16 may utilize a digital comb filter to improve a signal-to-noise ratio (SNR) of the electrical signals gathered by the electrode(s) 8... In the digital comb filter, band pass frequencies are selected to form teeth of the comb using a phase variance parameter timed to correspond to the presence and absence of the pain events... As understood by those of skill in the art, optimal digital filtering may be an alternative to signal averaging for improving the SNR”; ¶[0038]: “After artifact removal, the EEG may be subjected to spectral analysis using a signal processing technique such as, Fast Fourier Transform (FFT), wavelet analysis or fractal dimensional analysis”; ¶[0038]: “The EEG power spectrum may be divided into wide bands (e.g., low delta (0.5-1.5 Hz), high delta (1.5-3.5 Hz), theta (3.5-7.5 Hz), low alpha (7.15-10 Hz), high alpha (10.0-12.5 Hz), beta (12.5-25 Hz), low gamma (25-35 Hz), high gamma (35-50 Hz) and ultra high (50-200 Hz)”; ¶[0039]: “Alternatively, the spectrum may be divided into very narrow bands (VNB) of approximately 0.39-0.50 Hz in width”; John teaches using a digital comb filter in which band-pass frequencies are selected, and further teaches post-artifact-removal spectral analysis in which the EEG power spectrum is divided into wide frequency bands and very narrow frequency bands, thereby teaching or at least suggesting frequency filtering the cleaned EEG data from the EEG sensors to analyze neural oscillation frequencies and obtain filtered EEG data including neural oscillation frequencies).
Also, regarding claim 12, the modified John does not expressly teach removing, at the computer system, artifacts from the filtered EEG data from each sensor by averaging the filtered EEG data from each sensor with filtered EEG data from at least one other sensor. John teaches removing artifacts and non-brain-derived electrical activity from EEG recordings using denoising methods (John, ¶[0045]: “After non-brain derived electrical activity has been removed from the recordings by these or similar denoising methods, the residual electrical signals may be assumed to be substantially artifact-free and are compiled to form a continuous, artifact-free sample of brain electrical activity to be used for further qEEG or qSEP processing”). However, John does not expressly teach removing artifacts by averaging filtered EEG data from a sensor with filtered EEG data from at least one other sensor.
Pizzagalli is relied upon for this EEG artifact-removal/interpolation feature. Pizzagalli teaches that corrupted EEG electrodes due to artifacts or technical malfunction should not simply be omitted because source localization relies on the scalp potential distribution, and that interpolation methods are used to address corrupted EEG signals (Pizzagalli, p. 62, 4.1.2: “The importance of spatial sampling for source localization is associated with the issue of how best to deal with electrodes with corrupted EEG signals due to excessive artifacts or technical malfunction”; p. 62, 4.1.2: “Consequently, simply omitting corrupted electrodes is not a feasible solution, and interpolation methods are needed”). Pizzagalli further teaches a linear nearest-neighbor interpolation method in which corrupted activity is reconstructed by averaging data from neighboring electrodes (Pizzagalli, p. 62, 4.1.2: “Broadly speaking, two interpolation methods have been utilized: linear (nearest neighbor) and spline interpolation methods”; p. 62, 4.1.2: “In the first method, corrupted activity is reconstructed through a weighted average using data from neighboring electrodes (the weights are proportional to the Euclidian distance between the electrodes)”). Thus, under the broadest reasonable interpretation, Pizzagalli teaches removing or correcting corrupted/artifact EEG activity from a sensor/electrode channel by reconstructing the corrupted activity through a weighted average using data from neighboring electrodes, thereby teaching removing artifacts from EEG data from a sensor by averaging that sensor’s EEG data with EEG data from at least one other sensor.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John in view of Pizzagalli to remove artifacts from filtered EEG data by reconstructing artifact-affected or corrupted electrode activity using a weighted average of data from neighboring electrodes. One skilled in the art would have been motivated to make the modification because John already teaches EEG-based pain assessment using multiple biosensor electrodes and denoising/artifact-removal methods before qEEG processing, while Pizzagalli teaches that simply omitting corrupted EEG electrodes is not feasible when scalp potential distribution is important and that corrupted EEG activity may be reconstructed using a weighted average of neighboring electrode data. Under the broadest reasonable interpretation, the claim does not require a common-average reference or uniform averaging of every electrode channel, and Pizzagalli’s reconstruction of corrupted electrode activity using a weighted average of neighboring electrode data teaches or at least suggests the claimed artifact-removal mechanism. John and Pizzagalli are analogous because both concern processing multi-electrode EEG data and improving the reliability of EEG signal analysis by addressing artifacts or corrupted electrode signals. The modification would have been a predictable use of a known EEG artifact-correction technique in John’s EEG-based pain-assessment system. Because John performs artifact removal and frequency-domain EEG processing before subsequent qEEG/pain analysis, one skilled in the art would have found it obvious to apply Pizzagalli’s neighboring-electrode reconstruction to the filtered EEG sensor data in the preprocessing workflow before Alotaiby’s channel selection and John’s later pain-related analysis. The benefit of the modification would have been preserving useful multi-electrode EEG data for John’s qEEG/pain-detection processing while reducing the impact of artifacts or corrupted electrode signals before Alotaiby’s variance-based channel selection and John’s subsequent pain-related EEG analysis.
Claim 22 is rejected under 35 U.S.C. 103 as being unpatentable over John et al. (US-20080249430-A1), hereto referred as John, in view of Alotaiby et al. (Alotaiby, T., El-Samie, F. E. A., Alshebeili, S. A., & Ahmad, I. (2015). A review of channel selection algorithms for EEG signal processing. EURASIP Journal on Advances in Signal Processing, 2015(1), Article 66), hereto referred as Alotaiby, and further in view of Krishnan et al. (US-20060184475-A1), hereto referred as Krishnan, and further in view of Panavaranan et al. (Panavaranan, P. & Wongsawat, Y. (2013). EEG-based Pain Estimation via Fuzzy logic and polynomial kernel Support Vector Machine. The 2013 Biomedical Engineering International Conference (BMEiCON-2013)), hereto referred as Panavaranan.
The modified John teaches claim 1 as described above.
Regarding claim 22, the modified John does not expressly teach wherein the diagnosis of pain is 96% accurate.
Panavaranan is relied upon for the claimed pain-classification accuracy. Panavaranan teaches an EEG-based system for estimating pain level and classifying a pain state. Panavaranan records EEG from nine subjects during acute thermal pain stimulation using a 32-channel EEG system, processes the EEG using frequency-domain features including alpha- and beta-band power spectral density, and investigates support vector machine classifiers for separating a pain state from a resting state (Panavaranan, pp. 1-2). Panavaranan teaches that the linear SVM classified the resting and pain states with 73.33% accuracy, whereas the polynomial-kernel SVM provided significantly higher accuracy and expressly reports that “The polynomial SVM classification accuracy is 96.97 percent” (Panavaranan, p. 3, Figs. 5-6). Thus, Panavaranan teaches EEG-based binary classification between a pain state and a resting state with a measured classification accuracy of 96.97%.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John in view of Panavaranan to use Panavaranan’s polynomial-kernel SVM for binary EEG pain classification and thereby provide a pain-classification accuracy of approximately the recited performance level. John already teaches determining the presence or absence of pain from EEG brain activity and applying classifier functions, including neural networks and other classifiers, to assess pain. Panavaranan teaches the same general problem of classifying pain from EEG data and demonstrates that classifier selection materially affects the resulting accuracy, with the linear SVM producing 73.33% accuracy and the polynomial-kernel SVM producing 96.97% accuracy on the same pain/rest classification problem. One skilled in the art would therefore have been motivated to employ Panavaranan’s higher-performing polynomial-kernel SVM in John’s EEG-based pain-diagnosis system to improve classification accuracy.
Under the working interpretation stated above, the claimed 96% value is an empirically determined classification-performance result measured over an evaluation dataset. Panavaranan’s expressly reported 96.97% accuracy differs from the claimed 96% value by less than one percentage point and concerns the same type of binary EEG pain/rest classification. The record does not establish any technical criticality associated with precisely 96.00% accuracy as distinguished from the closely neighboring 96.97% accuracy demonstrated by Panavaranan. Moreover, Panavaranan itself demonstrates that classification accuracy varies with the classifier used for the same EEG classification problem. Accordingly, obtaining a binary EEG pain-classification accuracy of 96% would have been a predictable performance result within the ordinary variation and optimization of the known classifier arrangement rather than a patentably distinct result.
Claim 23 is rejected under 35 U.S.C. 103 as being unpatentable over John et al. (US-20080249430-A1), hereto referred as John, in view of Alotaiby et al. (Alotaiby, T., El-Samie, F. E. A., Alshebeili, S. A., & Ahmad, I. (2015). A review of channel selection algorithms for EEG signal processing. EURASIP Journal on Advances in Signal Processing, 2015(1), Article 66), hereto referred as Alotaiby, and further in view of Krishnan et al. (US-20060184475-A1), hereto referred as Krishnan, and further in view of Huang et al. (Huang, G., Xiao, P., Hung, Y.S., Iannetti, G.D., Zhang, Z.G., & Hu, L. (2013). A novel approach to predict subjective pain perception from single-trial laser-evoked potentials. NeuroImage, 81, 283-293), hereto referred as Huang, and further in view of Wager et al. (Wager, T.D., Atlas, L.Y., Lindquist, M.A., Roy, M., Woo, C.W., & Kross, E. (2013). An fMRI-Based Neurologic Signature of Physical Pain. New England Journal of Medicine, 368(15), 1388-1397), hereto referred as Wager.
The modified John teaches claim 1 as described above.
Regarding claim 23, the modified John does not expressly teach wherein the diagnosis of pain consists of low pain and high pain, wherein a diagnosis of low pain is 100% accurate and a diagnosis of high pain is 100% accurate.
Huang is relied upon for classification of electrophysiological pain data into low-pain and high-pain states. Huang teaches dividing pain-perception trials into two categories, low pain for VAS values below 5 and high pain for VAS values at least 5, extracting electrophysiological features from the trials, training a Naïve Bayes classifier using the training trials, and applying the classifier to predict low-pain or high-pain labels for test trials (Huang, p. 287). Huang further teaches evaluating the prediction using classification accuracy, defined as the percentage of accurate predictions, and expressly teaches that increasing the difference in VAS values between the classified pain categories increases classification accuracy (Huang, p. 288). Huang reports binary low-pain/high-pain classification accuracies of 86.3 ± 8.4% for within-individual prediction and 80.3 ± 8.5% for cross-individual prediction (Huang, p. 289, Table 1). Thus, Huang teaches using electrophysiological brain data and a machine-learning classifier to provide a binary diagnosis of lower and higher pain states and recognizes classification accuracy as a performance characteristic of the pain classifier.
Huang does not expressly teach 100% classification accuracy. Wager is relied upon for the known attainment of 100% decision accuracy in neural-activity-based discrimination of different pain intensities. Wager teaches using machine-learning analysis of brain activity to predict pain intensity and explains that, in its forced-choice discrimination test, sensitivity, specificity, positive predictive value, and decision accuracy are equal (Wager, p. 1391). Wager teaches that the neural signature response increased with increasing pain intensity and reports 100% sensitivity and specificity in forced-choice discrimination between painful conditions at 48.3°C and 47.3°C, with 15 trials performed for each condition (Wager, pp. 1393-1394, Fig. 2). Thus, Wager teaches that neural-activity-based discrimination between different pain-intensity conditions can achieve 100% decision accuracy over an evaluation dataset.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John in view of Huang and Wager to classify pain into lower-pain and higher-pain states and to provide a classifier having 100% classification accuracy over an evaluation dataset. Huang expressly teaches the claimed type of electrophysiological low-pain/high-pain classification, teaches evaluating that classification using accuracy, and teaches that classification accuracy increases as the difference between the pain categories increases. Wager further teaches that neural-activity-based discrimination of different pain-intensity conditions can attain 100% decision accuracy, including 100% discrimination between two painful temperature conditions. One skilled in the art seeking to implement John’s objective neural pain classifier would therefore have been motivated to apply Huang’s known low-pain/high-pain classification technique and to select classifier features and pain-intensity groupings that provide high classification accuracy, as expressly contemplated by Huang. In view of Wager’s demonstrated 100% decision accuracy for neural discrimination of sufficiently distinguishable pain-intensity conditions, obtaining 100% classification accuracy for the lower-pain and higher-pain conditions over an evaluation dataset would have been a predictable performance result of the known neural pain-classification techniques. The benefit would have been eliminating erroneous classification between the evaluated lower-pain and higher-pain states.
Claim 24 is rejected under 35 U.S.C. 103 as being unpatentable over John et al. (US-20080249430-A1), hereto referred as John, in view of Alotaiby et al. (Alotaiby, T., El-Samie, F. E. A., Alshebeili, S. A., & Ahmad, I. (2015). A review of channel selection algorithms for EEG signal processing. EURASIP Journal on Advances in Signal Processing, 2015(1), Article 66), hereto referred as Alotaiby, and further in view of Krishnan et al. (US-20060184475-A1), hereto referred as Krishnan, and further in view of Pizzagalli et al. (Diego A. Pizzagalli. “Electroencephalography and High-Density Electrophysiological Source Localization,” Handbook of Psychophysiology, 2007), hereto referred as Pizzagalli, and further in view of Nolan et al. (Nolan, H., Whelan, R., & Reilly, R.B. (2010). FASTER: Fully Automated Statistical Thresholding for EEG artifact Rejection. Journal of Neuroscience Methods, 192(1), 152-162), hereto referred as Nolan.
The modified John teaches claim 1 as described above.
Regarding claim 24, the modified John does not expressly teach wherein removing artifacts determines at least one time-series data having a highest variance, under the interpretation set forth above.
Nolan is relied upon for determining channel variance as part of EEG artifact detection and removal. Nolan teaches a fully automated EEG artifact-rejection process in which statistical parameters are calculated for the EEG data and outlying data are identified as contaminated. Nolan teaches that, for channel-artifact detection, variance is calculated for each EEG channel because a contaminated channel may have increased variance due to additive noise (Nolan, p. 154-156, Sec. 3.2). Nolan further teaches determining whether the calculated channel variance is an outlier, with a channel having an outlying variance being classified as contaminated, after which the contaminated channel is removed and its data reconstructed by interpolation from neighboring electrodes (Nolan, p. 154-156, Secs. 3.2). Thus, Nolan teaches that the artifact-removal process itself determines variance of the EEG time-series data in order to identify artifact-contaminated channel data.
Because Nolan's contamination criterion is a Z-score computed for each channel's variance relative to the distribution of variances across the channels, a channel flagged as having an outlying high variance is necessarily one of the channels having the highest variance among the plurality of channels (Nolan, p. 154, Sec. 3.2: "a channel whose variance had a Z-score of 3 would be deemed to be contaminated"; p. 156, Sec. 3.2.1.2). Nolan does not expressly require identifying the EEG time series having the highest variance. However, Alotaiby teaches estimating the variance of the available EEG channels and expressly teaches that “[t]he three channels with maximum variance are selected for signal classification” (Alotaiby, p. 6, Sec. 3.1.1). Thus, Alotaiby teaches comparing channel variances and determining the channel or channels having the maximum variance.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the artifact-removal processing of the modified John in view of Nolan and Alotaiby so that the channel variances determined during artifact detection are compared to identify at least the EEG time series having the highest variance before correction of artifact-affected channel data. Nolan teaches that increased channel variance can result from additive noise and uses channel variance as a parameter for identifying contaminated EEG channels. Alotaiby teaches the known comparison of EEG channel variances to identify the channel or channels having maximum variance. One skilled in the art therefore would have found it obvious to use the known maximum-variance comparison when implementing Nolan's variance-based artifact-detection process to identify the channel exhibiting the greatest variance as part of determining which EEG channel data warrant artifact correction. In the further modified system, artifact-affected EEG data identified using the variance-based determination would then be corrected using Pizzagalli's neighboring-electrode averaging technique as set forth regarding claim 6. The modification would have predictably allowed the artifact-removal process to identify EEG channel data exhibiting the greatest variance, including variance attributable to additive noise or contamination, before reconstructing the affected EEG data.
Claim 26 is rejected under 35 U.S.C. 103 as being unpatentable over John et al. (US-20080249430-A1), hereto referred as John, in view of Alotaiby et al. (Alotaiby, T., El-Samie, F. E. A., Alshebeili, S. A., & Ahmad, I. (2015). A review of channel selection algorithms for EEG signal processing. EURASIP Journal on Advances in Signal Processing, 2015(1), Article 66), hereto referred as Alotaiby, and further in view of Krishnan et al. (US-20060184475-A1), hereto referred as Krishnan, and further in view of Pizzagalli et al. (Diego A. Pizzagalli. “Electroencephalography and High-Density Electrophysiological Source Localization,” Handbook of Psychophysiology, 2007), hereto referred as Pizzagalli, and further in view of Zuckerman-Stark et al. (US-20130310660-A1), hereto referred as Zuckerman-Stark.
The modified John teaches claim 1 as described above.
Regarding claim 26, the modified John does not expressly teach wherein the band-pass filter consists of about 8 Hz to about 14 Hz. As discussed regarding claim 6, John teaches frequency filtering EEG data and further teaches analyzing EEG power within defined frequency bands, including low-alpha and high-alpha bands (John, ¶[0038]: low alpha of 7.15-10 Hz and high alpha of 10.0-12.5 Hz). John therefore teaches frequency-selective EEG processing directed to alpha-band activity, but does not expressly teach the recited approximately 8-14 Hz range.
Zuckerman-Stark teaches a system and method for pain classification using physiological signals, including EEG, and teaches using EEG frequency-band power as a feature for pain classification (Zuckerman-Stark, ¶¶[0046]-[0056]). In particular, Zuckerman-Stark identifies EEG/FEMG features including the power of the alpha, beta, gamma, delta, and theta frequency bands and expressly defines the alpha frequency range as 8-14 Hz (Zuckerman-Stark, p. 14, Table 1; ¶[0054]). Thus, Zuckerman-Stark teaches that 8-14 Hz is a known EEG alpha-band range used in an EEG-based pain-classification system.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified John in view of Zuckerman-Stark to perform the EEG frequency filtering over about 8 Hz to about 14 Hz when isolating alpha-band neural oscillations. John already teaches frequency-selective EEG processing and expressly analyzes low-alpha and high-alpha EEG activity in connection with pain assessment, while Zuckerman-Stark teaches using EEG alpha-band power as a feature for pain classification and identifies the alpha band as 8-14 Hz. One skilled in the art would have been motivated to use Zuckerman-Stark’s known 8-14 Hz alpha-band range in John’s EEG filtering so that the subsequent pain-related EEG analysis is performed on the alpha-band activity of interest. The modification would have been a predictable use of a known EEG alpha-band frequency range in John’s existing frequency-selective EEG processing.
Response to Arguments
Objections
Applicant's arguments filed 9/3/2026, pages 15-16, regarding the previous Objections of claims 1, 6-7, and 13 have been fully considered and are persuasive. The previous Objections have been withdrawn. However, there are new Objections as shown above.
Double Patenting
Applicant's arguments filed 9/3/2026, pages 16-17, regarding the previous statutory double patenting rejection of claims 1-19 have been fully considered and are persuasive. The previous statutory double patenting rejection have been withdrawn. However, there is a new nonstatutory obviousness-type double patenting rejection necessitated by amendment.
35 U.S.C. §112(b)
Applicant's arguments filed 9/3/2026, pages 17-18, regarding the previous 112(b) Rejections of claims 5, 7-12, and 17 have been fully considered and are persuasive. Claims 9 and 10 have been canceled, and the previous rejection concerning those claims is moot. The remainder of the previous 112(b) rejections have been withdrawn. However, there are new 112(b) rejections as shown above.
35 U.S.C. §103
Applicant's arguments filed 9/3/2026, pages 18-21, regarding the previous 103 Rejections of claims 1-19 have been fully considered. The present rejections have been revised as necessary to account for the amended claim language. Claims 9, 10, 14-16, and 18 have been canceled, and the previous rejections of those claims are moot. Claims 20-26 were newly added and were not subject to the previous rejections. Their present examination is set forth in the rejections above.
Applicant's Argument: Applicant argues that Alotaiby's maximum-variance channel-selection technique does not teach the real-time selection recited in independent claims 1, 7, and 13. Applicant notes that the variance calculation in Alotaiby, Sec. 3.1.1, is performed using training seizure data before the three channels having maximum variance are selected, and contends that this differs from comparing and selecting the time-series data having the highest variance in real time.
Examiner's Response: Applicant's argument is not persuasive. Applicant correctly notes that the particular maximum-variance example in Alotaiby, Sec. 3.1.1, calculates the variance of each channel using training seizure data before selecting the three channels having maximum variance (Alotaiby, pp. 5-6, Sec. 3.1.1: “The variance of ictal data of all available channels is estimated” and “the three channels with maximum variance are selected for signal classification”). The present rejection does not rely on this example as an express disclosure that the maximum-variance calculation itself is performed in real time. Rather, Alotaiby is relied upon for the known EEG channel-selection criterion of comparing channel variance and selecting channels having maximum variance.
John supplies the multi-channel EEG acquisition and computer-implemented pain-processing environment to which that selection criterion is applied. John teaches receiving EEG data from one or more EEG electrodes from a subject being analyzed for pain (John, ¶[0032]) and processing the digital EEG signals using a DSP and CPU, with the CPU comparing the processed signals to reference data to determine whether pain is detected (John, ¶[0041]). John further teaches comparing the subject's brain-activity data to reference data to determine whether pain exists (John, ¶[0058]). Thus, the rejection applies Alotaiby's known maximum-variance channel-selection criterion to the EEG channel data processed in John's system before John's subsequent pain-related analysis, such that the subsequent analysis is performed using the selected EEG channel data.
Alotaiby also provides a reason to perform channel selection in an operating EEG system. Alotaiby explains that reducing the number of EEG channels reduces computational complexity and that reduced computational complexity can provide “faster real-time response” in portable EEG systems (Alotaiby, p. 4). Alotaiby further discusses another channel-selection technique in which a channel-selection measure is produced by a “real-time EEG analysis for event detection (REACT) system” and used to reduce computational effort (Alotaiby, p. 6). Although that separate embodiment does not use maximum variance as its selection criterion, it confirms that EEG channel selection was known for use in connection with real-time EEG processing. Accordingly, the fact that the particular maximum-variance example of Sec. 3.1.1 uses training seizure data does not establish that the known maximum-variance channel-selection criterion would have been limited to offline processing or could not have been applied to the EEG data processed in John's system.
Further, claims 1, 7, and 13 do not require that the selected sensors have never previously been selected using training data or that the system operate without previously collected training data. The claims require that the recited comparison and selection be performed in real time on the plurality of time-series data. For the reasons set forth above and in the rejection, Applicant's identification of training data in Alotaiby's Sec. 3.1.1 embodiment does not establish error in the stated combination.
Applicant's Argument: Applicant argues that Alotaiby teaches away from using variance analysis or other statistical EEG channel-selection techniques for pain diagnosis because Alotaiby discusses seizure detection, sleep-state classification, motor-imagery classification, emotion classification, mental-task classification, and other EEG applications, but does not identify pain diagnosis as an application of its channel-selection techniques.
Examiner's Response: Applicant's argument is not persuasive. Alotaiby does not criticize, discredit, or discourage applying EEG channel-selection techniques to pain-related EEG processing. Instead, Alotaiby describes channel selection as a general EEG signal-processing technique used to reduce the number of channels presented for subsequent feature extraction and classification. Alotaiby identifies the purposes of channel selection as reducing computational complexity, reducing overfitting associated with unnecessary channels, and reducing setup time (Alotaiby, p. 1, Abstract). Alotaiby further states that channel selection may be performed using signal-processing tools including time-domain analysis, power spectral estimation, and wavelet transform (Alotaiby, p. 1, Abstract).
John independently supplies the pain-detection application. John teaches obtaining EEG signals from a subject being analyzed for the presence of pain (John, ¶[0032]), processing the EEG signals using a DSP and CPU and comparing the processed signals with reference data to determine whether pain is detected (John, ¶[0041]), and comparing brain-activity data with reference data to determine whether pain exists (John, ¶[0058]). The rejection therefore does not rely on Alotaiby to disclose pain diagnosis. Rather, Alotaiby's known EEG channel-selection technique is applied to John's EEG pain-detection system for the channel-selection benefits expressly identified by Alotaiby. Alotaiby's failure to list pain diagnosis among the particular applications surveyed does not constitute a teaching away from that combination.
Applicant's Argument: Applicant argues that the dependent claims are patentable because they incorporate the limitations of independent claims 1, 7, or 13 and therefore include the asserted deficiency concerning Alotaiby's maximum-variance channel selection.
Examiner's Response: Applicant's argument is not persuasive. The asserted deficiency in the rejection of independent claims 1, 7, and 13 has not been established for the reasons discussed above. Applicant's dependent-claim argument relies on the same asserted deficiency and therefore does not establish patentability of the dependent claims. The additional limitations of pending dependent claims 2-6, 8, 11, 12, 17, and 19 are addressed by the respective references and reasoning set forth in the present rejections. Claims 20-26 were newly added and were not subject to the previous rejections; their present rejections are separately set forth above.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
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
/ADAM J EISEMAN/Primary Examiner, Art Unit 3791