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 .
Status of the Claims
The status of the claims as of the response filed February 25, 2026, is as follows:
Claims 1, 4-5, 7, 10-19, 22-23, 25, 27-37, 40-41, 43, and 45-63 are pending, with claims 1, 19, and 37 being independent
Claims 2–3, 6, 8, 9, 20–21, 24, 26, 38–39, 42 and 44 are canceled
Applicant has amended claims 1, 4, 5, 7, 12, 15, 19, 22, 23, 25, 27, 29, 30, 33, 37, 40, 41, 43, 45, 48, and 51, and the amendments have been considered below.
Request for Continued Examination
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on February 25, 2026 has been entered.
Response to Amendment
The Examiner has addressed all Applicant arguments filed on February 25, 2026, for pending claims 1, 4, 5, 7, 10-19, 22, 23, 25, 27-37, 40, 41, 43, 45-63, on pages 1-11, regarding the rejection under 35 U.S.C. § 101.
Applicant argues that claims 1, 19, and 37, including feeding the feature vector matrix into one or more convolutional layers, learnable filters configured to focus attention on more than one leads, applying the learnable filters, pooling data, and outputting the status of the liver disease, wherein the status comprises an ACE score identifying a level of disease severity, do not recite a mental process because the human mind cannot practically perform simultaneous multi-lead CNN filtering, pooling, and ACE-score generation.
The Examiner respectfully disagrees as to Step 2A, Prong One. Applicant’s argument may show that the claim is not best characterized only as a mental process, but the amended claims still recite a mathematical concept. MPEP § 2106.04(a)(2) identifies mathematical concepts as mathematical relationships, mathematical formulas or equations, mathematical calculations, and states that if the identified limitations fall within an abstract-idea grouping, it is reasonable to conclude that the claim recites an abstract idea in Step 2A Prong One. Under BRI, claims 1, 19, and 37 recite mathematical processing of ECG voltage data by forming a feature vector matrix, applying CNN filters, pooling numerical layer data, and outputting an ACE score. The specification confirms this meaning: paragraph [0057] states Convolutional layers apply a convolution operation to the input; paragraph [0059] states each filter is convolved across the width and height of the input volume and computes the dot product between the entries of the filter and the input; and paragraph [0005] states the CNN processes digitized 12-lead ECGs and produces a numeric score between 0 and 1. The 2024 AI Guidance confirms that Step 2A, Prong One asks whether the claim sets forth or describes a judicial exception. Therefore, Applicant’s Prong One argument is not persuasive, and the § 101 rejection is maintained under Step 2A, Prong One on the mathematical-concept basis.
Applicant argues that claims 1, 19, and 37 integrate the abstract idea into a practical application because feeding the feature vector matrix into one or more convolutional layers, learnable filters configured to focus attention on more than one leads at a same time, and pooling data from the one or more convolutional layers into one or more connected layers allegedly improve computer-based diagnostics and solve the problem of detecting subtle ECG patterns.
The Examiner respectfully disagrees. MPEP § 2106.04(d)(1) requires both a disclosed technological improvement and claim language that reflects the improvement; a conclusory assertion of improvement is insufficient. MPEP § 2106.04(d)(1) states that the specification should be evaluated to determine if the disclosure provides sufficient details such that one of ordinary skill in the art would recognize the claimed invention as providing an improvement and that the claim itself reflects the disclosed improvement. Under BRI, the cited limitations define how the CNN mathematically processes ECG voltage data, not an improvement to the ECG device, processor, memory, or CNN technology itself. The specification confirms the claimed CNN processing is ordinary CNN operation: paragraph [0057] states Convolutional layers apply a convolution operation to the input; paragraph [0059] states each filter is convolved across the width and height of the input volume and computes the dot product between the entries of the filter and the input; and paragraph [0033] states CNNs analyze ECG voltage data to produce a disease-severity scoring system. These passages show useful diagnostic analysis, not improved computer functionality. Therefore, this argument is not persuasive, and the § 101 rejection is maintained under Step 2A, Prong Two.
Applicant argues that the claims are like Enfish because the CNN architecture allegedly improves how the computing system processes raw electrical voltage data, rather than merely using a computer as a tool.
The Examiner respectfully disagrees. Enfish applies when the claim improves computer functionality itself; MPEP § 2106 explains that eligibility is not based merely on a concrete result, utility, or adding a computer to an abstract idea. MPEP § 2106 states that eligibility should not be evaluated based on useful, concrete, and tangible result and that an abstract idea does not become nonabstract by limiting it to a particular field of use or technological environment. Here, the claim uses CNN components to analyze ECG data and output an ACE score. Paragraph [0005] states the CNN processes digitized 12-lead ECGs and produces a numeric score between 0 and 1, and paragraph [0033] states the CNN is used for producing a scoring system that correlates with the degree of disease severity. The claim improves the informational diagnostic result, not the operation of the computer. Therefore, the Enfish-based argument is not persuasive, and the rejection is maintained.
Applicant argues that claim 1 is like Example 47, claim 3 because the claimed CNN detects subclinical liver-disease severity in the same way Example 47 improves network intrusion detection.
The Examiner respectfully disagrees. Example 47 is fact-specific and states that examples are only intended to be illustrative and that all claims are analyzed for eligibility in accordance with their broadest reasonable interpretation. Example 47, claim 3 was eligible because the claim used the ANN result to perform real-time network remediation: dropping the one or more malicious network packets in real time and blocking future traffic from the source address. The Example 47 analysis states that the claim reflected the improvement by dropping potentially malicious packets and blocking future traffic from the source address. By contrast, claim 1 outputs an ACE score identifying a level of disease severity but does not recite treatment, device control, therapy modification, ECG-machine control, or another action that changes the technical system or patient care. Therefore, Example 47 does not support withdrawal, and the rejection is maintained.
Applicant argues that the specification sets forth an improvement in noninvasive medical diagnostics because cirrhosis-related ECG findings are subtle, non-specific, and not routinely used, and the claims reflect that improvement through the CNN feature-vector, filter, pooling, and ACE-score limitations.
The Examiner respectfully disagrees. The specification supports clinical usefulness, but clinical usefulness alone is not the Prong Two technological improvement required by MPEP § 2106.04(d)(1). Paragraph [0004] states ECG changes are subtle and non-specific and have not been incorporated into routine evaluation and prognostication of patients with cirrhosis. Paragraph [0033] states structural and metabolic changes may be reliably detected by a CNN and that the CNN produces a scoring system correlating with disease severity. These passages describe improved diagnostic information derived from ECG data, but the claims do not improve ECG acquisition, ECG hardware, CNN training technology, processor operation, memory structure, or a treatment process. Therefore, the claimed limitations do not integrate the mathematical analysis into a practical application, and the rejection is maintained.
Applicant argues that Ex parte Desjardins supports eligibility because the claims allegedly improve computer functionality and a technical field by using a specific CNN architecture to quantify subclinical disease severity.
The Examiner respectfully disagrees. Desjardins is distinguishable because the USPTO identified the claims there as reflecting improvements in AI technology itself. The USPTO notice states that Ex parte Desjardins, Appeal No. 2024-000567, was designated precedential and that the ARP reversed the § 101 rejection because the claims reflect improvements in artificial intelligence AI technology. Here, the pending claims do not improve the machine-learning model itself, reduce model storage, reduce model complexity, prevent catastrophic forgetting, or change training operation. The claims use CNN processing to generate a medical score from ECG data. Paragraph [0058] generally states a convolution operation may reduce free parameters and memory footprint, but the pending claims do not recite the kernel tiling, parameter-reduction, memory-footprint reduction, or model-training feature that provides that effect. Therefore, Desjardins does not make the present claims eligible, and the rejection is maintained.
Applicant argues that claims 1, 19, and 37 do not merely organize information because they configure a processor to use a specific architecture that resolves ECG data a human clinician cannot practically interpret.
The Examiner respectfully disagrees. The inability of a human to perform the calculation does not itself establish integration into a practical application. Under MPEP § 2106.05(g), data gathering and output activity that is incidental to the primary abstract analysis may be insignificant extra-solution activity; the section identifies a step of gathering data for use in a claimed process and an output report as examples. Here, the electrocardiograph and voltage-time data supply input data, and the ACE score is the output of the CNN mathematical analysis. The claims do not recite a further technical action using the score. Therefore, the additional elements do not impose a meaningful practical limit on the mathematical concept, and the rejection is maintained.
Applicant argues that claim 1 recites an inventive concept because feeding the feature vector matrix into one or more convolutional layers, learnable filters configured to focus attention on more than one leads at a same time, pooling data into one or more connected layers, and outputting an ACE score allegedly form a non-conventional, non-generic arrangement under BASCOM and Berkheimer.
The Examiner respectfully disagrees. Step 2B asks whether additional elements, individually or in combination, add significantly more than the judicial exception. MPEP § 2106 states that Step 2B evaluates whether additional elements of the claim provide an inventive concept, and MPEP § 2106.05(d) requires a factual determination only for an additional element or combination of additional elements alleged to be well-understood, routine, and conventional. Under the maintained BRI, the feature-vector matrix, convolution/filtering, pooling, and ACE-score output are the mathematical analysis itself, not additional elements beyond the exception. The record confirms this: paragraph [0057] states CNN hidden layers typically consist of convolutional layers, pooling layers, fully connected layers and normalization layers and paragraph [0059] states filters are convolved across the width and height of the input volume and compute the dot product between the entries of the filter and the input. Therefore, Applicant has not identified an additional element that supplies an inventive concept, and the rejection is maintained.
Applicant argues that no evidence shows the claimed simultaneous multi-lead attention, pooling, or their combination to produce an ACE score were well-understood, routine, or conventional.
The Examiner respectfully disagrees because the rejection does not rely on those CNN mathematical operations as the additional elements. The additional elements are the ECG data source/electrocardiograph and generic computing components. The specification expressly supports their routine nature: paragraph [0047] describes a standard 12-lead ECG where electrodes are placed and voltage is measured from twelve leads; paragraph [0084] states the computing node is in the form of a general-purpose computing device with processors, memory, and a bus; and paragraph [0095] states the storage medium is a tangible device that stores instructions. The ordered combination uses these components to acquire ECG data, execute CNN mathematical analysis, and output diagnostic information. It does not recite unconventional ECG hardware, improved processor architecture, improved memory structure, or a post-score treatment/control step. Therefore, Applicant’s Berkheimer argument is not persuasive, and the rejection is maintained.
Applicant argues that claim 1 differs from standard clinical ECG interpretation because it holistically analyzes raw voltage-time data to detect subclinical signals that clinicians would not identify by sequential waveform review.
The Examiner respectfully disagrees. Paragraph [0004] supports that ECG findings are subtle and non-specific and not used in routine prognostication, and paragraph [0005] supports that the CNN produces a numeric score between 0 and 1. These facts show improved diagnostic information, not an inventive concept in the additional elements. MPEP § 2106 cautions that eligibility is not based on a useful, concrete, and tangible result or mere use of a computer in a field of use. Therefore, the diagnostic usefulness of the ACE score does not add significantly more than the mathematical analysis, and the rejection is maintained.
Applicant argues that claims 19 and 37 overcome the § 101 rejection for the same reasons as claim 1.
The Examiner respectfully disagrees for the same reasons. Claim 19 adds an electrocardiograph, computing node, processor, and computer-readable storage medium, and claim 37 adds a computer-readable storage medium and processor. Paragraph [0084] identifies the computing node as a general-purpose computing device, and paragraph [0083] describes ordinary executable program modules. These elements store and execute the same mathematical CNN analysis and do not add an inventive technological application. Therefore, the rejection of claims 19 and 37 is maintained.
Applicant argues that the dependent claims overcome the § 101 rejection because they include the independent-claim elements and additional patent-eligible limitations.
The Examiner respectfully disagrees. The dependent claims add input data, data formatting, CNN/model details, training data, EHR storage or display, prediction, correlations, or clinical interpretation of the ACE score. These limitations narrow the same mathematical scoring analysis or report its result. They do not add an inventive concept beyond the judicial exception and generic data acquisition/computing environment. Therefore, the § 101 rejection of the dependent claims is maintained.
The Examiner has addressed all Applicant arguments filed on February 25, 2026, for pending claims 1, 4, 5, 7, 10-19, 22, 23, 25, 27-37, 40, 41, 43, 45-63, on pages 11-20, regarding the rejection under 35 U.S.C. § 103.
Applicant argues that claim 1, including providing the feature vector matrix to a pretrained learning system comprising a convolutional neural network and receiving from the convolutional neural network a status of liver disease... wherein the status comprises an ACE score identifying a level of disease severity, is not taught or suggested by Ma, Sang, and Clifford.
The Examiner respectfully disagrees for the reasons set forth in the detailed § 103 rejection above. Briefly, Ma is relied on for the ECG feature-vector/trained-model framework; Sang is relied on for the ECG-based liver-cirrhosis deep-learning prognosis model, including the CNN convolution, weighted-filter, pooling, and fully connected layer processing; and Clifford is relied on for the conventional 10-second/12-lead ECG configuration. Applicant’s argument is therefore not persuasive because it addresses the references as though each must independently teach the full amended limitation, rather than addressing the combined teachings relied on in the rejection. See MPEP 2145(V). The rejection of claim 1 is maintained. The same reasoning applies to independent claims 19 and 37 because Applicant relies on the same argument.
Applicant argues that the dependent claims are patentable for the same reasons argued for independent claims 1, 19, and 37, and further states that certain claims have been canceled.
The Examiner respectfully disagrees. Applicant presents no separate argument for the additional limitations of the dependent claims. Because the rejection of independent claims 1, 19, and 37 remains proper for the reasons set forth above and in the detailed § 103 rejection, Applicant’s dependent-claim argument is not separately persuasive.
Accordingly, the § 103 rejection of the remaining pending dependent claims under Ma, Sang, and Clifford is maintained.
Applicant also argues that claims 15, 33, and 51 are patentable over Ma, Sang, Clifford, and Zimmerman because they depend from independent claims 1, 19, and 37.
The Examiner respectfully disagrees. Applicant does not present a separate argument against Zimmerman or against the additional limitations of claims 15, 33, and 51. Because the rejection of the corresponding independent claims remains proper for the reasons set forth above, Applicant’s dependent-claim argument is not persuasive. Therefore, the § 103 rejection of claims 15, 33, and 51 over Ma, Sang, Clifford, and Zimmerman is maintained.
Applicant argues that Zimmerman does not cure the alleged deficiencies of Ma, Sang, and Clifford because Zimmerman does not teach the amended CNN/liver-disease/ACE-score limitation of independent claims 1, 19, and 37.
The Examiner respectfully disagrees. Zimmerman is not relied on to cure that limitation. As explained above and in the detailed § 103 rejection, Ma, Sang, and Clifford collectively teach or suggest the ECG feature-vector, CNN processing, liver-disease status, ACE-score severity, and 10-second/12-lead ECG features of independent claims 1, 19, and 37. Zimmerman is relied on only for the additional limitation recited in dependent claims 15, 33, and 51. Applicant’s argument is therefore not persuasive because it attacks Zimmerman for a feature supplied by the Ma/Sang/Clifford combination, not for the dependent-claim feature for which Zimmerman is applied.
Accordingly, the rejection of claims 15, 33, and 51 over Ma, Sang, Clifford, and Zimmerman is maintained.
Applicant argues that claims 5, 23, and 41 are patentable because Tang does not cure the alleged deficiencies of Ma, Sang, and Clifford.
The Examiner respectfully disagrees. Tang is not relied on to teach the amended CNN/liver-disease/ACE-score limitation of claims 1, 19, and 37. That limitation is addressed by Ma, Sang, and Clifford as explained above. Tang is relied on only for the additional dependent-claim limitation. Applicant presents no separate argument against Tang’s applied teaching for claims 5, 23, and 41. Therefore, the rejection is maintained.
Applicant argues that claims 12-13, 30-31, and 48-49 are patentable because Schreck does not cure the alleged deficiencies of Ma, Sang, Clifford, and Zimmerman.
The Examiner respectfully disagrees. Schreck is not relied on to teach the independent-claim CNN/liver-disease/ACE-score limitation. Ma, Sang, and Clifford are relied on for that limitation; Zimmerman and Schreck are applied for the additional dependent-claim limitations. Applicant does not separately address the Schreck-based limitations of claims 12-13, 30-31, and 48-49. Therefore, the rejection is maintained.
Applicant argues that claims 14, 32, and 50 are patentable because they depend from independent claims 1, 19, and 37.
The Examiner respectfully disagrees as to pending claims 14, 32 and 50. Applicant presents no separate argument for the additional limitations of those claims, and the rejection of the corresponding independent claims remains proper for the reasons stated above. Therefore, the rejection of claims 14, 32, and 50 is maintained.
Applicant argues that claims 17-18, 35-36, and 53-54 are patentable because Ioannou does not cure the alleged deficiencies of Ma, Sang, and Clifford.
The Examiner respectfully disagrees. Ioannou is not relied on to teach the amended independent-claim CNN/liver-disease/ACE-score limitation. That limitation is addressed by Ma, Sang, and Clifford. Ioannou is relied on only for the additional dependent-claim limitations. Applicant does not separately dispute Ioannou’s applied teaching for claims 17-18, 35-36, and 53-54. Therefore, the rejection is maintained.
Applicant argues that claims 56, 59, and 62 are patentable because Torregrosa does not cure the alleged deficiencies of Ma, Sang, and Clifford.
The Examiner respectfully disagrees. Torregrosa is not relied on for the independent-claim CNN/liver-disease/ACE-score limitation. Ma, Sang, and Clifford are relied on for that limitation, and Torregrosa is applied only for the additional dependent-claim limitations. Applicant provides no separate argument against Torregrosa’s applied teaching for claims 56, 59, and 62. Therefore, the rejection is maintained.
Applicant’s request for allowance is not persuasive because the pending § 103 rejections remain supported for the reasons stated above and in the detailed rejection. The canceled-claim rejections are moot; the remaining pending rejections are maintained.
Claim Rejections - 35 USC § 101
35 U.S.C. § 101 provides:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1, 4, 5, 7, 10-18, and 55-57; claims 19, 22, 23, 25, 27-36, and 58-60; and claims 37, 40, 41, 43, 45-54, and 61-63 are rejected under 35 U.S.C. § 101 because the claimed invention is directed to abstract ideas without additional elements that integrate the exception into a practical application or amount to significantly more than the exception.
Step 1:
Claims 1, 4, 5, 7, 10-18, and 55-57 are statutory process claims because claim 1 recites A method comprising acts of receiving voltage-time data, generating a feature vector matrix, providing the feature vector matrix to a pretrained learning system comprising a convolutional neural network, and receiving from the convolutional neural network a status of liver disease in the subject.
Claims 19, 22, 23, 25, 27-36, and 58-60 are statutory machine claims because claim 19 recites A system comprising: an electrocardiograph comprising a plurality of leads; and a computing node comprising a computer readable storage medium having program instructions embodied therewith and a processor of the computing node.
Claims 37, 40, 41, 43, 45-54, and 61-63 are statutory manufacture claims because claim 37 recites A computer program product comprising a computer readable storage medium having program instructions embodied therewith. Paragraph [0095] states The computer readable storage medium can be a tangible device and is not to be construed as being transitory signals per se. See MPEP § 2106.03.
Accordingly, the pending claims satisfy Step 1, and the analysis proceeds to Step 2A.
Step 2A, Prong One:
Prong One asks whether the claims recite a judicial exception. Under BRI, independent claims 1, 19, and 37 recite mathematical calculations because the claims organize ECG voltage values into a feature vector matrix, apply CNN convolution/filtering/pooling operations, and output an ACE severity score.
Claim 1 recites:
A method comprising:
receiving voltage-time data of a subject, the voltage-time data comprising voltage data of a plurality of leads of an electrocardiograph;
generating a feature vector matrix from the voltage-time data wherein the feature vector matrix corresponds to a time segment of an ECG signal comprising a contiguous subset of the voltage-time data, the time segment having a duration selected from a plurality of predefined durations including 10 seconds of 12 leads, 2 seconds of 12 leads, 10 seconds of six limb leads, and 10 seconds of a single lead;
providing the feature vector matrix to a pretrained learning system comprising a convolutional neural network; and
receiving from the convolutional neural network a status of liver disease in the subject by:
feeding the feature vector matrix into one or more convolutional layers of the convolutional neural network, wherein each convolutional layer of the one or more convolutional layers comprises learnable filters configured to focus attention on more than one leads of the plurality of leads at a same time;
applying the learnable filters to the feature vector matrix to identify features in the feature vector matrix;
pooling data from the one or more convolutional layers into one or more connected layers of the convolutional neural network; and
outputting the status of the liver disease, wherein the status comprises an ACE score identifying a level of disease severity.
Note: Bold language identifies the additional element considered outside the mathematical concept. Non-bold language identifies the abstract idea for Prong One.
Under BRI, the non-bolded limitations perform the mathematical analysis in four connected steps: first, ECG voltage-time data is arranged into a feature vector matrix; second, the matrix is processed by CNN convolutional layers; third, learnable filters identify numerical signal features and pooling reduces or summarizes the layer data; fourth, the connected layers output an ACE score identifying liver-disease severity.
The specification confirms that these are model calculations. Paragraph [0057] states a convolutional neural network (CNN) is a class of feed-forward artificial neural networks, The hidden layers of a CNN typically consist of convolutional layers, pooling layers, fully connected layers and normalization layers, and Convolutional layers apply a convolution operation to the input, passing the result to the next layer. Paragraph [0059] states A convolutional layer’s parameters consist of a set of learnable filters (or kernels), each filter is convolved across the width and height of the input volume, and the network learns filters that activate when it detects some specific type of feature. Paragraph [0075] identifies the output as a continuous value between 0 and 1 indicating the estimated likelihood of cirrhosis on each ECG.
The dependent claims remain within the same abstract idea. Claims 4, 5, 22, 23, 40, and 41 add demographic or genomic input data. Claims 7, 15, 17, 18, 25, 33, 35, 36, 43, 51, 53, and 54 define the CNN architecture or training data. Claims 10-14, 16, 27-32, 34, 45-50, and 52 define data source, matrix format, display, storage, or prediction output. Claims 55-63 define ACE-score correlations or classifications. These limitations narrow the mathematical analysis but do not remove it.
Accordingly, the pending claims recite abstract ideas, and the analysis proceeds to Step 2A, Prong Two.
Step 2A, Prong Two:
Prong Two asks whether the additional elements apply the exception in a meaningful practical way. The additional elements do not do so.
The additional elements are the ECG data source recited in claims 1, 19, and 37, the electrocardiograph recited in claim 19, the computing node recited in claim 19, the computer-readable storage medium recited in claims 19 and 37, and the processor recited in claims 19 and 37.
The ECG data source and electrocardiograph do not integrate the exception. Claim 1 recites voltage-time data comprising voltage data of a plurality of leads of an electrocardiograph. Claim 19 recites an electrocardiograph comprising a plurality of leads and program instructions for receiving voltage-time data of a subject from an echocardiograph. Claim 37 recites program instructions for receiving voltage-time data of a subject from an echocardiograph.
Paragraph [0047] states ten electrodes are placed on the patient's limbs and on the surface of the chest, measured from twelve different angles (leads), and recorded over a period of time. These elements supply ECG data for the mathematical analysis; the claims do not improve ECG measurement, electrode operation, or signal acquisition. See MPEP §§ 2106.05(a), 2106.05(c), and 2106.05(g).
The computing node does not integrate the exception. Claim 19 recites a computing node, and paragraph [0084] states computer system/server 12 in computing node 10 is shown in the form of a general-purpose computing device. The node executes the CNN analysis and returns the ACE score, but the claims do not improve operation of the computing node. See MPEP §§ 2106.05(a), 2106.05(f), and 2106.05(h).
The computer readable storage medium does not integrate the exception. Claims 19 and 37 recite a computer readable storage medium storing program instructions. Paragraph [0095] states The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. Storing instructions for the mathematical analysis does not improve storage technology, transform an article, or provide a treatment step. See MPEP §§ 2106.05(a), 2106.05(c), and 2106.05(f).
The processor does not integrate the exception. Claims 19 and 37 recite a processor executing the instructions. Paragraph [0084] identifies one or more processors or processing units 16. The processor executes the abstract ECG-data analysis; the claims do not improve processor operation or recite a changed processor architecture. See MPEP §§ 2106.05(a) and 2106.05(f).
The additional elements in combination do not integrate the exception. The ECG source supplies input data, and the computing components store and execute instructions for the feature-matrix and CNN analysis. Paragraph [0005] states that the CNN processes digitized 12-lead ECGs and produces a numeric score between 0 and 1 and that the ACE score is associated with markers of liver disease severity and mirrors the progression and resolution of liver disease. Paragraph [0076] states The ACE score is also associated with liver disease severity as determined by the MELD-Na as well as individual laboratory markers. These statements show clinical usefulness of the score, not improvement to ECG hardware, ECG acquisition, computer operation, storage technology, or processor functionality. The claims also do not recite administering treatment, altering therapy, controlling a medical device, or changing patient care based on the ACE score. See MPEP §§ 2106.04(d)(2), 2106.05(f), and 2106.05(h).
The dependent claims do not add integration. Claims 4, 5, 12-14, 22, 23, 30-32, 40, 41, and 48-50 add input data or matrix format. Claims 7, 15, 17, 18, 25, 33, 35, 36, 43, 51, 53, 54, 57, 60, and 63 further define the CNN, training data, or classification result. Claims 10, 11, 16, 27-29, 34, 45-47, 52, and 55-62 recite EHR source, storage, display, prediction, correlation, or transplant-trend output. These limitations collect, format, report, or clinically interpret data and do not add a technical application beyond the mathematical scoring. See MPEP §§ 2106.05(g) and 2106.05(h).
Accordingly, the pending claims do not integrate the mathematical concept into a practical application, and the analysis proceeds to Step 2B.
Step 2B:
Step 2B asks whether the additional elements in the pending claims add significantly more than the exception. They do not.
For independent claims 1, 19, and 37, the same additional elements are evaluated here: the ECG data source, the electrocardiograph recited in claim 19, the computing node recited in claim 19, the computer readable storage medium recited in claims 19 and 37, and the processor recited in claims 19 and 37. See MPEP §§ 2106.05(d) and 2106.07(a), subsection III.
The ECG data source in claims 1, 19, and 37, and the electrocardiograph in claim 19, do not add significantly more. Paragraph [0047] states ten electrodes are placed on the patient's limbs and on the surface of the chest, measured from twelve different angles (leads), and recorded over a period of time. The claims use these elements to acquire ECG data and do not recite unconventional ECG hardware, electrode placement, or signal acquisition.
The computing node in claim 19 does not add significantly more. Paragraph [0084] states computer system/server 12 in computing node 10 is shown in the form of a general-purpose computing device and includes one or more processors or processing units 16, a system memory 28, and a bus 18. Paragraph [0083] states program modules may include routines, programs, objects, components, logic, data structures. The computing node performs ordinary computer functions of receiving data, executing instructions, processing data, and outputting results. See MPEP §§ 2106.05(d), 2106.05(f), and 2106.07(a).
The computer readable storage medium in claims 19 and 37 does not add significantly more. Paragraph [0095] states The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The claims use the medium to store instructions for the mathematical analysis, which does not add an inventive technological application. See MPEP §§ 2106.05(d) and 2106.05(f).
The processor in claims 19 and 37 does not add significantly more. Paragraph [0084] identifies one or more processors or processing units 16. The claims use the processor to execute the mathematical analysis, which does not add a technological feature beyond computer implementation. See MPEP §§ 2106.05(d) and 2106.05(f).
The additional elements in independent claims 1, 19, and 37, considered in combination, do not add significantly more. Paragraph [0057] states The hidden layers of a CNN typically consist of convolutional layers, pooling layers, fully connected layers and normalization layers. Paragraph [0074] states The CNN’s architecture includes 9 convolutional blocks and two fully connected blocks, and was trained from random initial values using the Adam optimizer, Both a Res-Net architecture (1, 2) and vanilla CNN (3, 4) were tested, and a simple CNN was selected. The ordered combination uses ECG acquisition and generic computing components to perform and report CNN mathematical analysis, rather than adding an inventive technological application.
Dependent claims 4, 5, 7, 10-18, 22, 23, 25, 27-36, 40, 41, 43, 45-54, and 55-63 do not add significantly more. They add input data, data formatting, model details, training data, EHR storage or display, prediction, correlations, or clinical interpretation of the ACE score. These limitations define the data, model, or output of the same mathematical analysis and do not add an inventive technological application.
Accordingly, claims 1, 4, 5, 7, 10-18, 19, 22, 23, 25, 27-37, 40, 41, 43, 45-63 do not add significantly more than the abstract idea and are patent ineligible under 35 U.S.C. § 101.
Claim Rejections - 35 USC § 103
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1, 4, 10-11, 16, 55, and 57; 19, 22, 27-29, 34, 37, 40, 45-47, 52, 58, 60, 61, and 63 are rejected under 35 U.S.C. § 103 as being unpatentable over Ma US20210022633A1 in combination with KR-20210097510 – Sang and Clifford-NPL. See PTO-892.
Claim 1. MA teaches, A method comprising:
receiving voltage-time data of a subject, the voltage-time data comprising voltage data of a plurality of leads of an electrocardiograph; (Ma, paragraphs [0039], [0053-0054], [0062], [0063], [0102]).
MA discloses a "signal obtaining module 410" configured to "obtain an ECG signal", potentially from an "ECG Machine 110" which uses "electrodes" to "record electrical activities of the heart over a period of time".
generating a feature vector from the voltage-time data (Ma, paragraphs [0005], [0012], [0055], [0064]-[0097], [0039], [0062-0066]).
Ma discloses a "feature extracting module 420" that obtains a "feature vector" comprising specific features (TCSC, L-Z complexity, EMD complex number, sample entropy, wavelet transform energy) "of the ECG signal", detailing the generation process for each feature from the ECG signal data. Ma also describe the measure of electrical activities of the heart over a period of time using electrodes in segment like for example eight-second of the original signal.
providing the feature vector matrix to a pretrained learning system; (MA, paragraphs [0004], [0006], [0011], [0013], [0056]-[0057], [0087], [0100], [0102], [0111]-[0113], abstract, fig.7).
MA describes determining an arrhythmia type "based on the trained prediction model and the feature vector", requiring the feature vector to be provided (inputted ) to the "trained prediction model", which is a learning system.
receiving from the a status of ; Ma, [0102]: the processor 220 may input the ECG signal and / or the feature vector thereof into the trained prediction model…The output of the trained prediction model may include a predicted arrhythmia type of the ECG signal…)
and outputting the status of the , (Ma, [0102]: the processor 220 may input the ECG signal and / or the feature vector thereof into the trained prediction model…The output of the trained prediction model may include a predicted arrhythmia type of the ECG signal…)
(Ma, paragraphs [0004], [0010], [0018], [0020], [0057], [0102]). (Ma discloses obtaining ("determining" ) the "output of the trained prediction model" which includes a "predicted arrhythmia type". This output provides information concerning the subject's medical condition status (arrhythmia type, e.g., VF, VT, SVT ).
MA, however, does not explicitly disclose determining a status of liver disease comprises an ACE score identifying a level of disease severity using its method, as its focus is solely on cardiac arrhythmia types.
But Sang describes the direct link between liver disease and ECG analysis, stating " a liver cirrhosis patient using deep learning-based electrocardiogram analysis data , page 2 and calculate a predicted value (ACE score) of the prognosis of a liver cirrhosis patient , page 4 that could be considered as survival with in 3 months.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine Ma with Sang since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Additional, both relate electrocardiogram (ECG) signals / cardiac electrophysiology to the assessment of a patient's health condition and the use of deep learning models [Ma, par. 0006-0007, Sang, page 2 and 4].
A person of ordinary skill in the art, knowing Ma's method for using ECG feature vectors and a trained model to determine a health status (arrhythmia type ) and being aware from Sang, would have been motivated to apply Ma's known analysis method to ECG data to determine the status of liver disease using a score of severity to resolve data variability that depending on the reading physician and do not reflect the patient’s current vital sign, Sang page 2,4.
Ma discloses a "feature extracting module 420" that obtains a "feature vector" comprising specific features (TCSC, L-Z complexity, EMD complex number, sample entropy, wavelet transform energy) "of the ECG signal", detailing the generation process for each feature from the ECG signal data. Ma also describe the measure of electrical activities of the heart over a period of time using electrodes in segment like for example eight-second of the original signal.
MA, however, does not explicitly disclose the specificity of configuration of leads and times describe in claim 1 second limitation.
However, Clifford explicitly describe “12 ECG leads, using the first 10 seconds, Clifford, page 6 method 2.7”
It is obvious to combine for example a POSITA with the interest to describe an specify segment leads plus time configuration not explicitly describe in MA, would consider to include 12 ECG leads, using the first 10 seconds according to method of Clifford because it is a conventional configuration to ensure compatibility with recognized clinical standards and to optimize the analysis.
Obvious Rational per amended limitations:
feeding the feature vector matrix into one or more convolutional layers of the convolutional neural network, wherein each convolutional layer of the one or more convolutional layers comprises learnable filters configured to focus attention on more than one leads of the plurality of leads at a same time; applying the learnable filters to the feature vector matrix to identify features in the feature vector matrix; pooling data from the one or more convolutional layers into one or more connected layers of the convolutional neural network;
Under the broadest reasonable interpretation, the above limitation requires ECG-derived vector or matrix data input to a CNN, convolutional layers having weighted filters that extract ECG features, pooling of convolutional-layer information, transfer to a connected layer. The phrase focus attention on more than one lead ... at a same time is interpreted as convolutional filters operating across the ECG input data dimension that includes multiple ECG leads or channels, not as requiring a separate neural-network attention mechanism.
Ma teaches the base ECG-feature-to-trained-model process because Ma inputs ECG feature information into a trained prediction model and outputs a medical prediction. Ma 0102 discloses: the processor 220 may input the ECG signal and / or the feature vector thereof into the trained prediction model, and Ma 0102 further discloses: The output of the trained prediction model may include a predicted arrhythmia type of the ECG signal. Ma also teaches that the classifier may be a machine-learning model because Ma 0110 discloses: the classifier model may include a random forest model, a BP neural network model, a support vector machine SVM model, a wavelet neural network model, a clustering model, or the like, or a combination thereof. Thus, Ma provides the primary ECG feature-vector trained-prediction framework, but Ma does not teach convolutional layers of a CNN, weighted CNN filters operating on ECG matrix data, pooling into a fully connected layer.
Sang teaches the missing CNN and liver-disease prediction features. Sang teaches on p. 5/7 that the first deep neural network may be a one-dimensional convolutional neural network CNN or a recurrent neural network RNN. Sang also teaches on p. 4/7 that A one-dimensional CNN Convolutional Neural Network receives a one-dimensional vector or matrix as an input, and a convolution step or pooling step can be performed before the general ANN step, and that In the convolution step, a multidimensional matrix with weights and a convolution operation are performed. Sang further teaches weighted-filter feature extraction on p. 4/7 because a one-dimensional filter with different weights for each frequency band for each time section is applied to the image data to extract features for frequency more precisely than 2D CNN.
Sang also teaches the claimed CNN layer sequence. On p. 4/7, Sang discloses that, when the first deep neural network is a one-dimensional CNN, the first deep neural network includes 11 convolutional layers, one fully-connected layer, It may include four max-pooling layers and a dropout layer. Sang further discloses on p. 4/7 that The convolutional layer may receive electrocardiogram data 1 x 3000 x L and may have 64 or 32 filters with a size of 1 x 16, that Convolutional layers are a layer that performs convolution on the input ECG signal, that the max pooling layer can reduce the size of information by selecting the maximum value in a given region, and that A fully connected layer can do transformations that connect both neurons in one layer and neurons in the next layer. These disclosures teach feeding ECG vector or matrix data into CNN convolutional layers, applying weighted convolutional filters to extract features, pooling the convolutional output, and passing the pooled information to a fully connected layer.
Sang further teaches the liver-disease output. On p. 4/7, Sang discloses that the second deep learning unit 140 may calculate a predicted value of the prognosis of a liver cirrhosis patient through the second deep neural network, and that the predictive value may indicate survival or death within three months. Sang also states on p. 4/7 that the invention improves prognosis prediction by using an ensemble deep learning network with electrocardiogram data, liver function evaluation data, and other data. Thus, Sang teaches outputting a liver-cirrhosis prognosis, which reasonably reads on the claimed status of the liver disease under BRI.
A POSITA would have combined Ma with Sang by implementing Ma’s ECG trained-prediction framework with Sang’s one-dimensional CNN because Ma teaches providing ECG feature data to a trained model, while Sang teaches the missing CNN processing path. Thus, applying Sang’s known CNN ECG feature-extraction architecture to Ma’s trained ECG prediction model would have predictably resulted in feeding the ECG feature vector matrix into CNN convolutional layers, applying weighted learnable filters to identify ECG features, using filters that operate on the lead dimension under Applicant’s own BRI of focus attention, pooling the convolutional output, and passing the pooled data into a fully connected layer. This is the predictable use of a known CNN technique to improve a similar ECG-based medical-prediction method.
Claim 19. MA teaches, A system comprising:
an electrocardiograph comprising a plurality of leads; (MA, paras [0021], [0038-0039], FIGS. 6A-6D ). MA discloses an "ECG Machine" for recording the "electrical activities of the heart over a period of time", meeting the "electrocardiograph" element.
a computing node comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor of the computing node to cause the processor to perform a method comprising: (MA, [0038-0041], [0047-0049], FIG. 2)
MA discloses a "server 120" and a general "computing device 200" explicitly containing a "processor" and memory ("ROM 230, or a random access memory (RAM) 240" ), fulfilling the BRI of a computing node.
generating a feature vector matrix from the voltage-time data (Ma, paragraphs [0005], [0012], [0055], [0064]-[0097], [0039], [0062-0066]).
Ma discloses a "feature extracting module 420" that obtains a "feature vector" comprising specific features (TCSC, L-Z complexity, EMD complex number, sample entropy, wavelet transform energy) "of the ECG signal", detailing the generation process for each feature from the ECG signal data. Ma also describe the measure of electrical activities of the heart over a period of time using electrodes in segment like for example eight-second of the original signal.
providing the feature vector matrix to a pretrained learning system (MA, paragraphs [0004], [0006], [0013], [0056]-[0057], [0100], [0102], [0111]-[0113], abstract, refer to claim 1).
MA describes determining an arrhythmia type "based on the trained prediction model and the feature vector", requiring the feature vector to be provided (inputted ) to the "trained prediction model", which is a learning system.
receiving from the comprising a convolutional neural network; a status of by: feeding the feature vector matrix into one or more convolutional layers of the convolutional neural network, wherein each convolutional layer of the one or more convolutional layers comprises learnable filters configured to focus attention on more than one leads of the plurality of leads at a same time; applying the learnable filters to the feature vector matrix to identify features in the feature vector matrix; pooling data from the one or more convolutional layers into one or more connected layers of the convolutional neural network; and outputting the status of the liver disease, (refer to claim 1)
(Ma, paragraphs [0004], [0010], [0018], [0020], [0057], [0102]). (Ma discloses obtaining ("determining" ) the "output of the trained prediction model" which includes a "predicted arrhythmia type". This output provides information concerning the subject's medical condition status (arrhythmia type, e.g., VF, VT, SVT ).
MA, however, does not explicitly disclose determining a status of liver disease comprises an ACE score identifying a level of disease severity using its method, as its focus is solely on cardiac arrhythmia types.
But Sang describes the direct link between liver disease and ECG analysis, stating " a liver cirrhosis patient using deep learning-based electrocardiogram analysis data , page 2 and calculate a predicted value (ACE score) of the prognosis of a liver cirrhosis patient , page 4 that could be considered as survival within 3 months.
It would have been obvious to combine Ma with Sang because both relate electrocardiogram (ECG) signals / cardiac electrophysiology to the assessment of a patient's health condition and the use of deep learning models [Ma, par. 0006-0007, Sang, page 2 and 4].
A person of ordinary skill in the art, knowing Ma's method for using ECG feature vectors and a trained model to determine a health status (arrhythmia type ) and being aware from Sang, would have been motivated to apply Ma's known analysis method to ECG data to determine the status of liver disease using a score of severity to resolve data variability that depending on the reading physician and do not reflect the patient’s current vital sign, Sang page 2,4.
Ma discloses a "feature extracting module 420" that obtains a "feature vector" comprising specific features (TCSC, L-Z complexity, EMD complex number, sample entropy, wavelet transform energy) "of the ECG signal", detailing the generation process for each feature from the ECG signal data. Ma also describe the measure of electrical activities of the heart over a period of time using electrodes in segment like for example eight-second of the original signal.
MA, however, does not explicitly disclose the specificity of configuration of leads and times describe in claim 1 second limitation.
However, Clifford explicitly describe “12 ECG leads, using the first 10 seconds, Clifford, page 6 method 2.7”
It is obvious to combine for example a POSITA with the interest to describe an specify segment leads plus time configuration not explicitly describe in MA, would consider to include 12 ECG leads, using the first 10 seconds according to method of Clifford because it is a conventional configuration to ensure compatibility with recognized clinical standards and to optimize the analysis.
Claim 37. Ma teaches, A computer program product for detection of
a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to perform a method comprising: (MA, paras [0018], [0042], [0049], [0101], Claim 20),
MA discloses a "non-transitory computer-readable medium" storing a "set of instructions" or "program instructions" that are "executed by at least one processor" or "to be executed by the processor 220" to "directs the at least one processor to: obtain an electrocardiogram signal; obtain a feature vector... obtain a trained prediction model; and determine an arrhythmia type" or "perform exemplary methods".
generating a feature vector matrix from the voltage-time data wherein the feature vector matrix corresponds to a time segment of an ECG signal comprising a contiguous subset of the voltage-time data, the time segment having a duration selected from a plurality of predefined durations including 10 seconds of 12 leads, 2 seconds of 12 leads, 10 seconds of six limb leads, and 10 seconds of a single lead; (Ma, paragraphs [0005], [0012], [0055], [0064]-[0097], [0039], [0062-0066]).
Ma discloses a "feature extracting module 420" that obtains a "feature vector" comprising specific features (TCSC, L-Z complexity, EMD complex number, sample entropy, wavelet transform energy) "of the ECG signal", detailing the generation process for each feature from the ECG signal data. Ma also describe the measure of electrical activities of the heart over a period of time using electrodes in segment like for example eight-second of the original signal.
providing the feature vector matrix to a pretrained learning system comprising a convolutional neural network; (MA, paragraphs [0004], [0006], [0013], [0056]-[0057], [0100], [0102], [0111]-[0113], abstract).
MA describes determining an arrhythmia type "based on the trained prediction model and the feature vector", requiring the feature vector to be provided (inputted ) to the "trained prediction model", which is a learning system.
receiving from the convolutional neural network a status of liver disease in the subject, by: feeding the feature vector matrix into one or more convolutional layers of the convolutional neural network, wherein each convolutional layer of the one or more convolutional layers comprises learnable filters configured to focus attention on more than one leads of the plurality of leads at a same time; applying the learnable filters to the feature vector matrix to identify features in the feature vector matrix; pooling data from the one or more convolutional layers into one or more connected layers of the convolutional neural network; and outputting the status of the liver disease (Refer to claim 1)
wherein the status comprises an ACE score identifying a level of disease severity (Ma, paragraphs [0004], [0010], [0018], [0020], [0057], [0102]). (Ma discloses obtaining ("determining" ) the "output of the trained prediction model" which includes a "predicted arrhythmia type". This output provides information concerning the subject's medical condition status (arrhythmia type, e.g., VF, VT, SVT ).
MA, however, does not explicitly disclose determining a status of liver disease comprises an ACE score identifying a level of disease severity using its method, as its focus is solely on cardiac arrhythmia types.
But Sang describes the direct link between liver disease and ECG analysis, stating " a liver cirrhosis patient using deep learning-based electrocardiogram analysis data , page 2 and calculate a predicted value (ACE score) of the prognosis of a liver cirrhosis patient , page 4 that could be considered as survival within 3 months.
It would have been obvious to combine Ma with Sang because both relate electrocardiogram (ECG) signals / cardiac electrophysiology to the assessment of a patient's health condition and the use of deep learning models [Ma, par. 0006-0007, Sang, page 2 and 4].
A person of ordinary skill in the art, knowing Ma's method for using ECG feature vectors and a trained model to determine a health status (arrhythmia type ) and being aware from Sang, would have been motivated to apply Ma's known analysis method to ECG data to determine the status of liver disease using a score of severity to resolve data variability that depending on the reading physician and do not reflect the patient’s current vital sign, Sang page 2,4.
Ma discloses a "feature extracting module 420" that obtains a "feature vector" comprising specific features (TCSC, L-Z complexity, EMD complex number, sample entropy, wavelet transform energy) "of the ECG signal", detailing the generation process for each feature from the ECG signal data. Ma also describe the measure of electrical activities of the heart over a period of time using electrodes in segment like for example eight-second of the original signal.
MA, however, does not explicitly disclose the specificity of configuration of leads and times describe in claim 1 second limitation.
However, Clifford explicitly describe “12 ECG leads, using the first 10 seconds, Clifford, page 6 method 2.7”
It is obvious to combine for example a POSITA with the interest to describe an specify segment leads plus time configuration not explicitly describe in MA, would consider to include 12 ECG leads, using the first 10 seconds according to method of Clifford because it is a conventional configuration to ensure compatibility with recognized clinical standards and to optimize the analysis. .
Claim 4. Ma in combination with Sang and Clifford teaches, The method of claim 1, further comprising:
receiving demographic information of the subject, wherein generating the feature vector matrix comprises adding the demographic information to the feature vector matrix.
Ma teaches generating a "feature vector of the electrocardiogram signal" ([0004], [0064]) based on the "electrocardiogram signal" ([0004], [0062]), which is derived solely from the ECG voltage data, for use with a "trained prediction model" ([0004]).
However, Ma does not disclose the steps of "receiving demographic information of the subject" or "adding the demographic information to the feature vector".
Sang teaches receiving demographic data, " other data may include demographic data and clinical data "can be fed into the first deep neural network using the electrocardiogram data as an input value, and the electrocardiogram analysis data, liver function evaluation data, and other data. Refer Sang, page 2-3
A person of ordinary skill in the art would be motivated to modify Ma's feature vector generation to include receiving demographic information and adding it to the feature vector, as taught by Sang , because demographic factors like age and sex are known to correlate with cardiac health and disease risk. Incorporating this standard patient information, as explicitly shown by Sang , into Ma's feature set would be expected to refine the input to Ma's prediction model ([Ma, [0004]]) and potentially improve its accuracy or provide more personalized risk assessments, representing a predictable modification to enhance the model's performance.
Claim 10. Ma in combination with Sang and Clifford teaches, The method of claim 1, further comprising:
providing the status to an electronic health record system for storage in a health record associated with the subject. , (Ma, 0040, 0042, 0044-0045, 0103)
Ma discloses "sending analysis result" (providing the status) to system components like a "storage device" holding "medical records" or a "hospital monitoring center" (electronic health record system.
Claim 11. Ma in combination with Sang and Clifford teaches, The method of claim 1, further comprising:
providing the status to a computing node for display to a user. (Ma, 0045, 0046, 0103)
Ma discloses "sending analysis result" (providing the status under BRI) to a "user terminal," "mobile terminal," or "ECG Machine" (computing node under BRI) to "present information ... to a user" (patient/medical staff).
Claim 16. Ma in combination with Sang and Clifford teaches, The method of claim 1, further comprising generating a liver disease outcome prediction for the subject.
(Sang, page 2)
Note: The following claims 22, 27-29, 34, 40, 45-47 and 52 are rejected with the same analysis above because they are very similar to claims 1, 4, 10-11, 16.
Claim 55. . Ma in combination with Sang and Clifford teaches, The method of claim 1, wherein the ACE score is correlated with a MELD-Na score and a plurality of individual laboratory markers. (Sang, page. 2-4)
Sang describes a method to increase the accuracy of predicting liver cirrhosis prognosis by functionally correlating a patient's electrocardiogram analysis data with both a liver disease severity score and multiple lab results. The method uses a deep neural network that takes in electrocardiogram analysis data (an ECG-derived score), liver function evaluation data, and other data to calculate a single predictive value for the patient's prognosis. Sang explicitly identifies that the liver function evaluation data may include a Meld score... a Meld-Na, thereby teaching the correlation of an ECG-derived score with a MELD-Na score. Furthermore, the method incorporates other data which includes a plurality of individual laboratory markers such as AST, ALT... bilirubin... creatinine, international normalized ratio (INR). By combining the ECG-derived score with both the MELD-Na score and these numerous lab markers within a single computational model to produce a unified output,
Claim 57. The method of claim 1, wherein the pretrained learning system classifies electrocardiogram signals from subjects with compensated cirrhosis as indicative of cirrhosis.
Ma developed a system classifying electrocardiogram (ECG) signals to identify arrhythmia types, but it did not disclose the capability to indicate cirrhosis in compensated patients. This limitation is noted as Ma's outputs were restricted to arrhythmia diagnoses. (MA, Fig. 5, 0008-0009, 0063-0065)
Sang's work addresses the classification of ECGs for cirrhosis indication, particularly in a compensated context, by using deep neural networks and incorporating cirrhosis-state information like MELD score and Child-Pugh score. This approach aims for early detection and measurement of cirrhosis severity. (Sang, page 3)
Combining Ma's ECG classification pipeline with Sang's cirrhosis-specific labeling and compensated/decompensated schema would be obvious to a skilled person. Both references share the purpose of automated inference from ECGs using machine learning to generate clinically relevant labels.
Integrating Sang’s cirrhosis-specific labeling and compensated/decompensated context into Ma’s pretrained ECG classifier would predictably confer Sang’s stated benefit improved cirrhosis prediction and earlier detection.
Note: Claims 58, 60, 61 and 63 are rejected with the analysis of claims 55 and 57 for being very similar.
Claims 7, 25, and 43 are rejected under 35 U.S.C. § 103 as being unpatentable over Ma US20210022633A1 in view of KR20210097510A Sang and Clifford-NPL, and further in view of Ribeiro et al, refer to PTO-892-U-RCE.
Claim 7. Ma in combination with Sang and Clifford teaches, The method of claim 1, wherein the convolutional neural network comprises at least one residual connection.
Ma teaches ECG-derived data provided to a trained prediction model because Ma discloses inputting ECG signal and/or feature-vector information into a trained prediction model and outputting a medical prediction. Thus, Ma supplies the ECG feature-vector/trained-model framework, but does not teach the convolutional neural network comprises at least one residual connection.
Ribeiro teaches the missing residual-CNN feature because Ribeiro discloses a CNN for automatic 12-lead ECG diagnosis using residual blocks and skip connections. Ribeiro states that we used a convolutional neural network similar to the residual network, but adapted to unidimensional signals, and further explains that this architecture allows DNNs to be efficiently trained by including skip connections. Ribeiro also states that the network consists of a convolutional layer Conv followed by four residual blocks with two convolutional layers per block, and that Max Pooling and convolutional layers with filter length 1 1x1 Conv are included in the skip connections to make the dimensions match those from the signals in the main branch. These disclosures teach the residual connection in the CNN itself because the skip connections are part of the convolutional residual-block architecture used for ECG classification. The extracted quotes are from Ribeiro, Neural network architecture and training section.
A person of ordinary skill in the art would have combined Ma with Ribeiro before January 7, 2022 by implementing Ma’s trained ECG prediction model with Ribeiro’s residual CNN architecture. Ma teaches the base ECG prediction framework because the processor 220 may input the ECG signal and / or the feature vector thereof into the trained prediction model, and the output of the trained prediction model may include a predicted arrhythmia type of the ECG signal (Ma, [0102]). Ribeiro teaches the compatible ECG CNN architecture because we used a convolutional neural network similar to the residual network... adapted to unidimensional signals and this architecture allows DNNs to be efficiently trained by including skip connections (Ribeiro, Neural network architecture and training).
The modification would have been a predictable substitution of Ribeiro’s known ECG residual CNN for Ma’s trained ECG prediction model because both references process ECG data with trained models to output ECG-based diagnostic classifications. Ribeiro expressly uses ECG recordings as neural-network input, teaches a convolutional layer Conv followed by four residual blocks with two convolutional layers per block, and includes Max Pooling... and convolutional layers with filter length 1 1x1 Conv... in the skip connections (Ribeiro, Neural network architecture and training). Thus, using Ribeiro’s residual/skip-connection CNN in Ma would have predictably provided Ma’s trained ECG model with a known ECG-CNN architecture that improves DNN training while preserving Ma’s purpose of classifying ECG signals.
Note: Claims 25 and 43 are rejected for the same reasons because they recite the corresponding system and computer-program-product residual-connection limitations.
Claim(s) 15, 33, and 51 are rejected under 35 U.S.C. 103 as being unpatentable over MA US-20210022633A1 in combination of KR-20210097510 – Sang and Clifford-NPL refer to PTO-892 in further view of Noah Zimmerman US20220384045.
Claim 15. Ma in combination with Sang and Clifford teaches, The method of claim 1, wherein the convolutional neural network comprises at least nine convolutional blocks and (Sang, page 4)
Ma teaches using a "trained prediction model" ([0004], [0012]) with features extracted from an ECG signal to determine arrhythmia type and does not explicitly disclose that the trained prediction model comprises a convolutional neural network (CNN). However, Sang teaches using a one-dimensional Convolutional Neural Network (CNN) is used as the "first deep neural network" to "calculate the electrocardiogram analysis data" by processing "electrocardiogram data" as input. This involves applying a "one-dimensional filter with different weights" to extract features from the ECG signal, enabling analysis and potentially contributing to liver cirrhosis prognosis prediction. Refer to Sang, page 4
Additional Sang describe "11 convolutional layers" and only "one fully-connected layer," not two, to process raw ECG signals. This architecture is used because the convolutional layers are designed to extract key features from the one-dimensional ECG data, while the single fully-connected layer serves to combine these features and transform them into a final output for analysis.
However, Zimmerman discloses at least two (in fact four) fully connected blocks and more than nine convolutional blocks when added the sub blocks. (Zimmerman, paragraph 0141, 0148-0149, fig. 4B)
It would have been obvious to one of ordinary skill in the art to implement Ma's "trained prediction model" ([Ma 0004,]) using a CNN architecture as taught by Zimmerman (0140-0141 and 0209) Both references are in the field of using computational models to analyze ECG data. An ordinary person in the art could be motivated to include CNN taught by Zimmerman since are efficient computing models which have shown their strengths in solving hard problems in artificial intelligence (Zimmerman, paragraph 0139) and just a simple substitution of one know element for another.
Note: Claims 33, and 51 are rejected with claim 15 to be very similar.
Claim(s) 5, 23, 41 are rejected under 35 U.S.C. 103 as being unpatentable over MA US20210022633A1 in combination of KR-20210097510 – Sang and Clifford-NPL refer to PTO-892 and in further view of Tang US20180046773A1.
Claim 5. Ma in combination with Sang and Clifford teaches, The method of claim 1,
Ma taught generating a feature vector paragraphs 0004, 0064 and does not disclose receiving genomic information of the subject and the adding the genomic information to the feature vector. But Tang use gene patent data input as for example DNA and RNA in paragraph 0050.
A person of ordinary skill in the art would have been motivated to include genomic information with ECG-derived features because Tang teaches gene input data such as DNA or RNA for medical prediction, and the Ma/Sang combination already uses ECG-derived data to generate a liver-disease status. Adding genomic information would have predictably expanded the patient-specific feature set used for disease-status prediction.
Note: Claims 23 and 41 are rejected with the same analysis above to be very similar to claim 5.
Claim(s) 12-13, 30-31, 48-49 are rejected under 35 U.S.C. 103 as being unpatentable over MA US-20210022633A1 in combination of KR-20210097510 – Sang, Clifford-NPL refer to PTO-892 U, and in further view of Schreck US6901285B2.
Claim 12. Ma in combination with Sang and Clifford teaches, The method of claim 1, wherein the feature vector matrix (Ma, 0055, 0063), Ma discloses a "feature extracting module 420" to "obtain a feature vector of the ECG signal" and does not discloses comprises a Matrix having a plurality of rows and a plurality of columns, the plurality of rows corresponding to a temporal dimension and the plurality of columns corresponding to a spatial dimension
However, Schreck disclose in Column 6, lines 15 -21 and Column 4 lines 15-21 “a typical 3- dimensional spatial”, “The sequence of digitized voltage-time measurements forms a Matrix [V], which is a 3×M Matrix, where M is the number of measurements in time, as indicated in block 104. Typically, 300 sequential time measurements are taken.” that according to figure 7, has different planes X, Y and Z that are analogous to plurality of rows and columns corresponding a spatial (lead) dimension.
It is obvious to combine Ma in combination with Sang with Schreck because both references digitize identical ECG inputs for automated analysis; adopting Schreck’s lead × time Matrix is a routine data-layout choice that enables Ma’s existing classifier to operate on a structured tensor and, per Schreck, facilitates subsequent linear algebra operations (“multiplying the … Matrix” ). A person of ordinary skill would be motivated by the computational convenience and the explicit instruction in Schreck to store ECG data in that Matrix form.
Claim 13. Ma in combination with Sang and Clifford teaches, The method of claim 12, wherein each of the plurality of rows correspond to one of the plurality of leads and each of the plurality of columns corresponds to a timestamp. (Schreck, column 4, lines 1-21)
Schreck discloses a Matrix “wherein each of the plurality of rows correspond to one of the plurality of leads and each of the plurality of columns corresponds to a timestamp.
Note: Claims 30-31, 48-49 are rejected with the same analysis above for being very similar.
Claim(s) 14, 32, and 50 are rejected under 35 U.S.C. 103 as being unpatentable over MA US-20210022633A1 in combination of KR-20210097510 – Sang and Clifford-NPL refer to PTO-892 U. In further view of Noah Zimmerman US20220384045, Plus Schreck US6901285B2.
Claim 14. Ma in combination with Sang, Clifford and Zimmerman in further view with Schreck teaches, The method of claim 12, wherein the temporal dimension has a and Schreck teach the feature-vector matrix of claim 12, but do not expressly disclose a 500Hz temporal resolution.
However, Zimmerman mentioned ECGs sampled at either 250 Hz or 500 Hz paragraph 0251. A POSITA would therefore be motivated to integrate this sampling resolution into Ma’s workflow to ensure compatibility with existing ECG repositories and analytic models.
Note: Claims 32 and 50 are rejected with the analysis above to being very similar with claim 14.
Claim(s) 17-18, 35-36, 53-54 are rejected under 35 U.S.C. 103 as being unpatentable over MA US20210022633A1 in combination of KR-20210097510 – Sang and Clifford -NPL in further view of Loannou- NPL. Refer to PTO-892.
Claim 17. Ma in combination with Sang and Clifford teaches, The method of claim 1, wherein the pretrained learning system is trained by receiving a training set of voltage-time data MA teaches training an ECG-based classifier by “obtaining training samples including a plurality of arrhythmia types of historical electrocardiogram signals” paragraph 0006 and does not disclose using a cirrhosis patient cohort; its training samples are limited to arrhythmia-labeled ECGs, not liver-disease cases.
However, Loannou (JAMA Net Open 2020) discloses assembling a large cirrhosis cohort for prognostic modeling: “This prognostic study included 48 151 patients with hepatitis C virus (HCV) related cirrhosis in the national Veterans Health Administration …” selecting and grouping a cirrhosis patient population for Machine-learning training. Both MA and Loannou share the general concept of training predictive models on disease-specific patient data. It would have been obvious to adapt MA’s ECG-training pipeline to the cirrhosis cohort taught by Loannou simply substituting arrhythmia ECGs with ECGs collected from their cirrhosis patients based on the know correlation ECG and cirrhosis.
Claim 18. Ma in combination with Sang and Clifford teaches in further view of Loannou teaches, The method of claim 17, wherein the training set of voltage-time data is from one of a retrospective cohort subset or a prospective cohort subset. (Loannou, “This prognostic study included 48,151 patients with hepatitis C virus (HCV)–related cirrhosis … who had at least 3 years of follow-up after the diagnosis of cirrhosis.”)
Loannou, use retrospective cohort subset who had at least 3 years of follow-up after the diagnosis of cirrhosis.
Note: Claims 35-36 and 53-54 are rejected with the above analysis for being very similar to claims 17-18.
Claim(s) 56, 59 and 62 are rejected under 35 U.S.C. 103 as being unpatentable over MA US20210022633A1 in combination of KR-20210097510 – Sang and Clifford -NPL in further view of Torregrosa - NPL. Refer to PTO-892.
Claim 56. Ma in combination with Sang and Clifford teaches, The method of claim 1, wherein trends in the ACE score
Ma in combination with Sang and Clifford teaches claim 1, however does not specify in liver transplantation data progression.
Torregrosa teaches the missing element, describing longitudinal pre-/post-transplant changes demonstrating resolution of cirrhosis-induced cardiac abnormalities, including Fifteen cirrhotics were reevaluated 6–12 months after transplantation and Liver transplantation reverses these alterations, normalization of systolic response, significant increases in cardiac index. This directly maps to trends, before and after a liver transplantation reflect a progression and resolution of cirrhosis and supplies the longitudinal element absent in Sang. Refer to abstract in Torregrosa
It would have been obvious to a POSITA to combine the MA + Sang + Clifford framework with Torregrosa because MA supplies a deployed ECG ingestion/feature/model pipeline across clinical devices obtain an electrocardiogram signal, obtain a feature vector, obtain a trained prediction model supporting system-level status generation and downstream integration, while Clifford standardizes multi-lead configuration and segment duration for comparable inputs (full 12-lead generation via individual Dower transforms using the first 10 seconds), and Sang ties the ECG analysis to liver-disease severity by fusing ECG-derived data with MELD-Na and laboratory markers. Torregrosa then adds the missing longitudinal validation pre/post-transplant normalization providing explicit trend evidence that the score reflects disease course rather than medication effects.
Adding Torregrosa’s longitudinal pre-/post-transplant evidence would have predictably supported the use of ECG-derived liver-disease scoring to track disease course before and after transplantation.
Note: Claims 59 and 62 it is rejected with the same analysis above for being very similar to claim 56.
Relevant Prior Art:
US 20210401376 A1
[0012] FIG. 4 shows a model architecture, in an embodiment. The input is a sequence of 256 Hz ECG signal. The convolutional neural networks (CNNs) include two main blocks, inception block and pre-activation residual block (Resblock). The CNNs are applied to extract spatial features from the ECG signal by filters and moving windows. The extracted features are passed into long short-term memory (LSTMs) layer, where t indicates timestep and h indicates the hidden cells which pass the information from one timestep to the next timestep. The output of the model is a sequence of probabilities of presence arousal.
[0036] The inception module 206 contains a plurality of machine-learning or statistical models that process the ECG sequence 202 in parallel. Each machine-learning model applies one or more filters, or kernels, to the ECG sequence 202 to obtain one or more corresponding feature maps. Each filter has a different combination of size (i.e., the number of sequential values of the ECG sequence 202 to which the filter is applied) and weights (or kernel coefficients). The inception module 206 also includes a concatenator 208 that concatenates all of the feature maps from all of the machine-learning models to create a channel array 210.
[0037] In embodiments, the machine-learning models are artificial neural networks. For example, the machine-learning models may be convolutional neural networks (CNNs) 204, as shown in FIG. 2. Specifically, the inception module 206 includes a first CNN 204(1) that outputs feature maps via a first channel 205(1), a second CNN 204(2) that outputs additional feature maps via a second channel 205(2), and a third CNN 204(3) that outputs even more feature maps via a third channel 205(3). The inception module 206 may include more or fewer CNNs 204, and corresponding channels 205, than shown in FIG. 2 without departing from the scope hereof. The CNNs 204 convolve the ECG sequence 202 at different timescales. For example, the CNNs 204(1), 204(2), and 204(3) may have filter sizes of 11, 15, and 23 data points, respectively. For f.sub.s=256 Hz, these filter sizes correspond to timescales of 43 ms, 59 ms, and 90 ms, respectively. However, the CNNs 204 may use any combination of filter sizes without departing from the scope hereof. Each CNN 204 may have a single layer or multiple layers. One or more of the CNNs 204 may include a pooling layer, such as a max pooling layer. Alternatively or additionally, a pooling layer may be used as one of the plurality of machine-learning models.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JOSHUA DAMIAN RUIZ whose telephone number is (571)272-0409. The examiner can normally be reached 0800-1800.
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, Shahid Merchant can be reached at (571) 270-1360. 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.
/JOSHUA DAMIAN RUIZ/Examiner, Art Unit 3684
/Shahid Merchant/Supervisory Patent Examiner, Art Unit 3684
Prosecution Insights