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 Claims
This action is in reply to an application filed on 05/10/2025. Claims 1-11 are currently pending and have been examined.
Priority
This application is the U.S. national stage of PCT/KR2023/018074, filed November 10, 2023, which claims priority to KR 10-2022-0150322, filed November 11, 2022, and KR 10-2023-0154684, filed November 9, 2023. The application publishes in the family as WO 2024/101954 A1. The earliest effective filing date available to any claim is accordingly November 11, 2022, and every reference applied below published before that date.
Claim Objections
Claim 11 is objected to as a substantial duplicate of claim 1. Claim 11 recites, word for word, the same preamble and the same four steps as claim 1: “A method performed by an electronic device including at least one processor, the method comprising: obtaining biometric-data of a user; generating first information about a first disease of the user by inputting the obtained biometric-data to a pre-trained first neural network model; generating second information about the first disease of the user by inputting the obtained biometric-data to a pre-trained second neural network model; and generating a diagnosis result for the user regarding a first category of the first disease based on the first and second information.” The two claims are identical in scope. A patent should not contain two claims of the same scope. See MPEP 608.01(m). Applicant is required to cancel one of claims 1 and 11, or to amend claim 11 so that it differs in scope from claim 1.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
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 1-10 are rejected under 35 U.S.C. § 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 — Statutory Category
Claims 1-9 and 11 recite a method and fall within the process category. Claim 10 recites an electronic device and falls within the machine category. The analysis proceeds to Step 2A.
Step 2A Prong One — The claims recite a judicial exception
Stripped of the hardware, claim 1 recites: look at a patient’s physiological measurement; form one judgment about whether the patient has a disease; form a second judgment about that same disease; and arrive at a diagnosis from the two judgments together. That is a mental process. Observing data, evaluating it and forming a diagnostic judgment are acts that can be performed in the human mind, or by a physician with pen and paper, and have been performed that way for as long as the electrocardiogram has existed. See MPEP 2106.04(a)(2)(III).
The specification says as much. It describes the conventional practice as “measuring an electrocardiogram, displaying a measured electrocardiogram signal in the form of a graph, and determining whether the presence or absence of myocardial infarction and ischemic heart disease in a patient’s heart based on the graph.” A cardiologist reading that graph performs the very determination claim 1 recites; the claim differs in that a neural network performs it instead.
Claims 2, 3 and 4 add a second exception. They recite adjusting a first probability value by applying a weight corresponding to a second probability value, and comparing the adjusted value to reference values to select among three outcomes. Those are mathematical relationships and calculations. See MPEP 2106.04(a)(2)(I). Claims 2-4 therefore recite both a mental process and a mathematical concept.
Step 2A Prong Two — The exception is not integrated into a practical application
The additional elements beyond the exception are: an electronic device; at least one processor; at least one memory; a communication interface (claim 10); and a pre-trained first neural network model and a pre-trained second neural network model.
The hardware is recited generically and performs its ordinary functions — a processor executes, a memory stores, an interface communicates. Reciting generic computer components to carry out an exception is no more than an instruction to apply it using a computer as a tool, which does not integrate. See MPEP 2106.05(f); Alice Corp. v. CLS Bank Int’l, 573 U.S. 208, 223-24 (2014).
The neural network models are recited only by what they produce, never by how they are built or trained. The claims require that each model be “pre-trained” and that biometric data be input to it, and nothing more: no architecture, no layer structure, no training objective, no feature engineering. Claim 8 comes closest, but it specifies the labelling of the training data rather than any structure of the model. A neural network recited as a black box that emits a classification is a tool the exception is applied with, not an improvement to the tool. See MPEP 2106.05(a); Enfish, LLC v. Microsoft Corp., 822 F.3d 1327, 1336 (Fed. Cir. 2016).
Nor is the obtaining step more than data gathering. Receiving an electrocardiogram so that it can be analyzed is necessary antecedent activity and is insignificant extra-solution activity. See MPEP 2106.05(g).
The improvement the specification identifies is diagnostic accuracy — a better answer to a medical question. An improvement in the accuracy of an abstract determination is an improvement to the exception itself, not to any technology. See MPEP 2106.05(a).
Step 2A Prong Two — The dependent claims
Claims 2, 3 and 4 supply no additional element. They recite the mathematics of the exception itself — weighting one probability by another, and comparing the result to cut-offs — stated in more detail than claim 1 states it. Narrowing an abstract idea by reciting more of its own steps does not integrate that idea into a practical application. See MPEP 2106.04(a)(2)(I).
Claim 5 specifies the content of the two judgments, requiring that the first be the presence or absence of the disease and the second be its type. The content of information is part of the exception rather than an additional element. Specifying what a diagnostician concludes says nothing about how any device operates.
Claim 6 requires that the two models be extracted from among a plurality based on the first disease and the category of the diagnosis result. Choosing which yardstick to apply to which question is itself part of the abstract diagnostic process. The claim moreover recites the selection functionally, by what it accomplishes rather than by any mechanism that accomplishes it, so it adds no structure capable of integrating the exception.
Claim 7 confines the method to electrocardiogram data, to myocardial infarction or ischemic heart disease, and to ST elevation or non-ST elevation myocardial infarction as the type. Those recitations limit the exception to a particular technological environment and a particular disease. Limiting an abstract idea to a field of use does not integrate it. See MPEP 2106.05(h); Bilski v. Kappos, 561 U.S. 593, 612 (2010).
Claim 8 describes how the training data were annotated, requiring that a single electrocardiogram signal carry two labels assigned on different criteria. Annotating data according to two criteria is a choice about how information is organised. As claimed it adds no structure to either model and no function to the machine; the claim reaches the trained models only through the data they were shown, never through what they are.
Claim 9 requires that the two generating steps be performed in parallel. That is an instruction about the order, or the absence of order, in which two computations occur. The two steps have no data dependency — each takes the same biometric data, and neither consumes the other’s output — so performing them concurrently is simply what a general-purpose computer does with independent work. Reciting it amounts to applying the exception on a computer. See MPEP 2106.05(f).
Step 2B — No inventive concept
Considered individually and as an ordered combination, the additional elements do not amount to significantly more than the exception.
Individually, each element performs its ordinary and expected function. As to the neural network models specifically, applicant’s own specification supplies the evidence that inputting an electrocardiogram to a neural network model to obtain a diagnostic result was already conventional: “recently, with the development of deep learning technology, there has been increasing interest in a method of obtaining an electrocardiogram signal analysis result only by inputting the graph of the electrocardiogram signal to a neural network model through the application of deep learning technology to the medical field.” That is applicant’s own statement that the technique was well-understood, routine and conventional before the filing date. See Berkheimer v. HP Inc., 881 F.3d 1360, 1369 (Fed. Cir. 2018); MPEP 2106.05(d)(I)(2).
The prior art of record corroborates it independently. Strodthoff (2020) benchmarks convolutional networks on twelve-lead electrocardiograms across diagnostic superclasses including myocardial infarction and their subclasses. Cho (2020) — a paper by two of the present inventors — reports a deep-learning algorithm detecting myocardial infarction from electrocardiography. Both predate the critical date.
As an ordered combination, the sequence recited — obtain a measurement, evaluate it two ways, combine the evaluations into a conclusion — is the only order in which the underlying diagnostic reasoning can proceed. An ordering that merely follows the logic of the idea supplies no inventive concept. See OIP Technologies, Inc. v. Amazon.com, Inc., 788 F.3d 1359, 1363 (Fed. Cir. 2015). Claims 1-11 are accordingly not patent eligible.
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, 2, 5, 6. 8, 10, and 11 are rejected under 35 U.S.C. 103 as being unpatentable over Khosousi, et al. (US 2020/0205745 A1) in view of J. Wehrmann, R. Cerri & R. C. Barros, “Hierarchical Multi-Label Classification Networks,” Proceedings of the 35th International Conference on Machine Learning (ICML), PMLR 80:5225-5234 (2018).
With regards to claim 1, Khosousi teaches a method performed by an electronic device including at least one processor, the method comprising: obtaining biometric-data of a user (¶0023: “receiving, by a processor, a biophysical signal data set of a subject acquired from one or more channels of one or more sensors.” The acquired cardiac signal is the biometric data, and at ¶0011 it expressly includes an electrocardiogram: “cardiac signals that may be acquired via conventional electrocardiogram (ECG/EKG) equipment, bipolar wide-band biopotential (cardiac) signals that may be acquired from other equipment such as those described herein.”); generating first information about a first disease of the user by inputting the obtained biometric-data to a pre- trained first neural network model (¶0023: “determining, by the processor, a value (e.g., risk/likelihood, binary indication) indicative of presence or absence of cardiac disease or condition (e.g., coronary arterial disease ...) by directly inputting the pre-processed data set to one or more neural networks ... or ensemble(s) thereof.”); generating second information about the first disease of the user by inputting the obtained biometric-data to a pre-trained second neural network model (¶0029: “determining, by the processor, one or more location values indicative of presence or absence of cardiac disease or condition at a given coronary artery by inputting ... the pre-processed data set ... to one or more second neural networks ... trained with ... subjects labeled ... with a diagnosis of presence or absence of coronary artery disease located at a coronary artery selected from the group consisting of a left main artery (LMA), a proximal left circumflex artery (Prox LCX) ...” The same pre-processed data set is input to both networks).
Khosousi does not explicitly teach …and generating a diagnosis result for the user regarding a first category of the first disease based on the first and second information. Wehrmann teaches …and generating a diagnosis result for the user regarding a first category of the first disease based on the first and second information (§2.1, Eq. (6): “P_F = β(P^1_L ⊕ P^2_L, ..., P^|H|_L) + (1 − β)P_G,” every position of P_F being a function of both the level-specific output and the global output; and §3: “the final predictions (binary vector indicating the presence or absence of each class) are generated after thresholding those probabilities.”). For the motivation to combine, both are directed to machine-learned classification of a condition and its subordinate category from physiological data, and Wehrmann addresses precisely the step Khosousi leaves open. See MPEP 2141.01(a).
Khosousi produces a disease-level value and a category-level value from one acquisition but reports the category value alone, the disease value serving only to gate whether the second network runs. Wehrmann’s answer to that situation is to weight the two into one vector and threshold it, and it states the benefit in its own terms: a purely global approach is “less likely to capture local information from the hierarchy, eventually underfitting,” while a purely local approach suffers “the well-known error-propagation problem.” Applying that weighting to the two values Khosousi already computes is the use of a known technique for its known purpose, with the predictable result of a single category-level call informed by both. See MPEP 2143(I)(G).
Khosousi further teaches from claim 10, An electronic device, comprising: a communication interface; memory configured to store pre-trained first and second neural network models; and at least one processor configured to (¶0065: “FIG. 11 shows an exemplary computing environment in which example embodiments and aspects may be implemented.” The system, at ¶0048, is “an assessment system coupled, directly or indirectly, to said device. The assessment system includes one or more processors; and a memory having instructions stored thereon, wherein execution of the instruction by the one or more processors cause the one or more processors to perform any one of the above-recited method,” and the communication interface is “a system comprising: a device configured to acquire phase-gradient biophysical signals ...; and an assessment system coupled, directly or indirectly, to said device.” The memory necessarily stores the pre-trained models: a trained neural network is a set of learned parameters, and a processor cannot apply parameters that are not held in memory. Khosousi applies pre-trained networks throughout. See MPEP 2112).
Claim 10 recites the device that performs the steps of claim 1 and claim 11 is the exact same wording of claim 1. Therefore, these claims are rejected for the same reasons as claim 1.
With regards to claim 2, Wehrmann teaches the method of claim 1, wherein generating the diagnosis result comprises: adjusting a first probability value included in the first information based on a second probability value included in the second information (§2.1, Eq. (6): “P_F = β(P^1_L ⊕ P^2_L, ..., P^|H|_L) + (1 − β)P_G.” The first probability value is adjusted on the basis of the second. Claim 2 recites no particular arithmetic for the adjustment), and generating a diagnosis result for the user regarding the first disease of the first category based on the first probability value (§3: “The outputs of HMCN and baseline algorithms are probability values for each class. Hence, the final predictions (binary vector indicating the presence or absence of each class) are generated after thresholding those probabilities.” The quantity thresholded is P_F, the combined value of Eq. (6), and it is thresholded at every class position, the category-level class included. The category-level result is therefore read off the adjusted value, which is what this limb requires). The motivation to combine Wehrmann with Khosousi is the same as applied in claim 1.
With regards to claim 5, Khosousi teaches the method of claim 1, wherein the first information is information about a presence or absence of the first disease, and the second information is information about a first type of the first disease (¶¶0023 and 0029: the first network yields a value “indicative of presence or absence of cardiac disease or condition” and the second yields location values for that disease “at a given coronary artery ... selected from the group consisting of a left main artery (LMA), a proximal left circumflex artery (Prox LCX) ...”).
With regards to claim 6, Khosousi teaches the method of claim 1, comprising extracting first and second neural network models from among a plurality of neural network models (¶¶0023 and 0029: the system holds more than one network and draws on each for its own purpose. The disease-level determination is made by “one or more neural networks ... or ensemble(s) thereof” and the category-level determination by “one or more second neural networks ... or ensemble(s) thereof.” Those are a plurality of neural network models, and the method applies the first for the presence determination and the second for the location determination. Khosousi also describes the act in terms at ¶0064: FIG. 10 is “a process to select a neural network model (e.g., a deep neural network model, a convolutional neural network model, etc.) configured to non-invasively assess presence or non-presence of coronary artery disease or a condition in a person.”) based on the first disease and the category of the diagnosis result (¶0030: “comparing, by the processor, the value ... indicative of the presence of cardiac disease or condition to a threshold value, wherein the step of determining the one or more location values ... is performed based on the comparison.” The second network is run because the first disease was found present and because a location-level answer is wanted. That is selection on the disease and on the category of the result).
With regards to claim 8, Khosousi teaches the method of claim 1, wherein: the first and second neural network models are pre- trained based on training data in which a plurality of labels set based on different criteria are assigned for a same electrocardiogram signal; and the plurality of labels comprises a first-category label corresponding to the presence or absence of the first disease and a second-category label corresponding to the first type of the first disease (¶¶0027, 0029 and 0037. The first-category label is the presence label: “the label for presence of coronary artery disease comprises a Gensini-based score.” The second-category label is the location label: subjects are “labeled (e.g., binary labels) with a diagnosis of presence or absence of coronary artery disease located at a coronary artery selected from the group consisting of a left main artery (LMA), a proximal left circumflex artery (Prox LCX) ...,” and the localization array “comprise a plurality of elements each corresponding to a label indicative of presence or non-presence of the cardiac disease or condition at a given location in the coronary artery.” Presence and type, the two criteria claim 8 names.).
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Khosousi, et al. (US 2020/0205745 A1) in view of J. Wehrmann, R. Cerri & R. C. Barros, “Hierarchical Multi-Label Classification Networks,” Proceedings of the 35th International Conference on Machine Learning (ICML), PMLR 80:5225-5234 (2018) in further view of A. Björkelund, M. Ohlsson, J. Lundager Forberg, A. Mokhtari, P. Olsson de Capretz, U. Ekelund & J. Björk, “Machine learning compared with rule-in/rule-out algorithms and logistic regression to predict acute myocardial infarction based on troponin T concentrations,” J. Am. Coll. Emerg. Physicians Open 2021;2(2):e12363.
With regards to claim 4, Khosousi does not explicitly teach the method of claim 2, wherein generating the diagnosis result comprises: generating a first diagnosis result corresponding to the first disease of the first category when the first probability value is equal to or larger than a first value, generating a second diagnosis result corresponding to the first disease of the first category when the first probability value is smaller than the first value and equal to or larger than a second value, and generating a third diagnosis result corresponding to the first disease of the first category when the first probability value is smaller than the second value. Björkelund teaches the method of claim 2, wherein generating the diagnosis result comprises: generating a first diagnosis result corresponding to the first disease of the first category when the first probability value is equal to or larger than a first value, generating a second diagnosis result corresponding to the first disease of the first category when the first probability value is smaller than the first value and equal to or larger than a second value, and generating a third diagnosis result corresponding to the first disease of the first category when the first probability value is smaller than the second value (partitions the predicted probability of acute myocardial infarction produced by an artificial neural network using two cut-offs into three graded results. At Results: “For ANN, the derived probability thresholds were ≤ 0.02164 for rule-out and ≥ 0.1278 for rule-in.” A probability at or above the rule-in value yields the rule-in (high risk) result, a probability at or above the rule-out value but below the rule-in value yields the intermediate result, and a probability below the rule-out value yields the rule-out (low risk) result. All three are results regarding acute myocardial infarction. That is the first value, the second value and the three diagnosis results). For the motivation to combine, Björkelund is directed to machine-learning diagnosis of a cardiac condition in the same clinical setting as Khosousi, and addresses the step left open once a model probability has been computed: what disposition to report. It is analogous art. See MPEP 2141.01(a). It states the reason to adopt its three-way partition. Of the guideline algorithms it replaces it observes that they “do not provide any probabilistic assessment of the risk” and that they “still leave a substantial proportion in an intermediate group that cannot be accurately classified as low or high risk.” Applying two cut-offs to the probability produced by the combination of Rejection 1 gives the clinician a rule-in, rule-out or intermediate disposition instead of a bare binary label, with the predictable benefit Björkelund reports. See MPEP 2143(I)(G).
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Khosousi, et al. (US 2020/0205745 A1) in view of J. Wehrmann, R. Cerri & R. C. Barros, “Hierarchical Multi-Label Classification Networks,” Proceedings of the 35th International Conference on Machine Learning (ICML), PMLR 80:5225-5234 (2018) in further view of H.Y. Choi et al., “Diagnostic Accuracy of the Deep Learning Model for the Detection of ST Elevation Myocardial Infarction on Electrocardiogram,” J. Pers. Med. 2022, 12(3), 336.
With regards to claim 7, Khosousi teaches the method of claim 5, wherein: the biometric-data comprises an electrocardiogram signal (¶0011: “cardiac signals that may be acquired via conventional electrocardiogram (ECG/EKG) equipment, bipolar wide-band biopotential (cardiac) signals that may be acquired from other equipment such as those described herein.”); the first disease comprises any one of myocardial infarction and ischemic heart disease (¶0023, whose condition is “cardiac disease or condition (e.g., coronary arterial disease, pulmonary hypertension, pulmonary arterial hypertension, left heart failure, right heart failure, and abnormal left-ventricular end diastolic pressure (LVEDP)).” Coronary arterial disease is ischemic heart disease).
Khosousi does not explicitly teach …and the first type comprises any one of ST elevation myocardial infarction (STEMI) and non-ST elevation myocardial infarction (NSTEMI). Choi teaches … and the first type comprises any one of ST elevation myocardial infarction (STEMI) and non-ST elevation myocardial infarction (NSTEMI) (Abstract, is “to measure the diagnostic accuracy of the deep learning model (DLM) for ST-elevation myocardial infarction (STEMI) on a 12-lead electrocardiogram (ECG) according to culprit artery sorts,” and “the DLM was trained with STEMI and normal sinus rhythm ECG for external validation.” Its reported result, also at the Abstract, is that “the AUROC for overall STEMI was 0.998 (0.996–0.999) with SEN 97.4% (95.7–100) and SPE 99.2% (98.1–99.4),” and it concludes that “DLM showed high diagnostic accuracy for STEMI detection, regardless of the type of culprit artery.” The architecture is shown at Figure 1, captioned “Model architecture of deep learning model. ECG, electrocardiogram; STEMI, ST elevation myocardial infarction; DNN, deep neural networks.” For the motivation to combine, Choi is directed to the same task as the second network of Rejection 1 — identifying which kind of ischemic event an electrocardiogram shows — in a modality Khosousi expressly contemplates, and reports that the task is performed accurately by a deep neural network. It is analogous art. See MPEP 2141.01(a). Khosousi’s second network already returns a category of the disease, expressed as the coronary artery involved; substituting the ST elevation category Choi identifies for the arterial category Khosousi names is the simple substitution of one known category scheme for another to obtain a predictable result, and a skilled artisan would make it because STEMI is the distinction that governs immediate treatment. See MPEP 2143(I)(B).
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Khosousi, et al. (US 2020/0205745 A1) in view of J. Wehrmann, R. Cerri & R. C. Barros, “Hierarchical Multi-Label Classification Networks,” Proceedings of the 35th International Conference on Machine Learning (ICML), PMLR 80:5225-5234 (2018) in further view of Rim, et al. (US 2020/0160999 A1).
With regards to claim 9, Khosousi does not explicitly teach the method of claim 1, wherein generating the first information about the first disease and generating the second information about the first disease are performed in parallel. Rim teaches the method of claim 1, wherein generating the first information about the first disease and generating the second information about the first disease are performed in parallel (¶0277: “a plurality of diagnosis assistance neural networks that perform prediction on different labels may be used in parallel.” ¶0278: the parallel system “may train a plurality of neural network models for obtaining a plurality of diagnosis assistance information and obtain the plurality of diagnosis assistance information using the trained plurality of neural network models.” ¶0279 gives the two-model case in the claim’s own shape: the system “may train, on the basis of fundus images, a first neural network model that obtains a first diagnosis assistance information related to the presence of an eye disease of a patient and a second neural network model that obtains a second diagnosis assistance information related to the presence of a systemic disease of the patient.”). For the motivation to combine, the two networks above have no data dependency: each receives the same pre-processed data set and neither consumes the other’s output. Khosousi runs them in sequence only because its disease-level value gates the second, a constraint the claim does not impose. Rim is directed to the same problem of obtaining several items of diagnosis assistance information about one patient from one captured signal, and answers it by running a model per diagnostic label in parallel, at ¶0277: “a plurality of diagnosis assistance neural networks that perform prediction on different labels may be used in parallel.” Removing the gate and running Khosousi’s two networks concurrently, as Rim teaches, is the use of a known technique for its known purpose and makes both outputs available at once. It is analogous art, both being directed to neural-network diagnosis from patient data. See MPEP 2141.01(a) and 2143(I)(G)
Allowable Subject Matter
Claim 3 is objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims, and rewritten to overcome 101 rejections.
Claim 3 recites “applying a weight corresponding to the second probability value,” and conditions the adjustment on a comparison of the second value with a reference value, maintaining the first value when the test fails. Wehrmann, §2.1, Eq. (6) scales the first value and adds a contribution from the second, and Kim uses a second model’s output to “generate a confidence score of the result/output of a primary MLM.” The conditional test is NOT disclosed. Wehrmann’s combination is unconditional and its weight β is a fixed hyperparameter, not a weight corresponding to the second probability value; neither reference compares the second value against a reference value, nor maintains the first value unchanged when the test fails.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Lee, et al. (US 2014/0101080 A1) which discloses an apparatus and a method for diagnosis are provided. The apparatus for diagnosis lesion include: a model generation unit configured to categorize learning data into one or more categories and to generate one or more categorized diagnostic models based on the categorized learning data, a model selection unit configured to select one or more diagnostic model for diagnosing a lesion from the categorized diagnostic models, and a diagnosis unit configured to diagnose the lesion based on image data of the lesion and the selected one or more diagnostic model.
Kim, et al. (US 2021/0117977 A1) which discloses a processing system including at least one processor may obtain at least one of a first machine learning model or a second machine learning model, deploy the at least one of the first machine learning model or the second machine learning model to a plurality of trained machine learning models for operating in parallel with respect to a same prediction task, obtain at least one data set, apply the at least one data set to the plurality of trained machine learning models, obtain the first result of the first machine learning model and a second result of the second machine learning model in accordance with the applying, store the first result of the first machine learning model and the second result of the second machine learning model, and provide an output in accordance with at least one of the first result or the second result.
N. Strodthoff, P. Wagner, T. Schaeffter & W. Samek, “Deep Learning for ECG Analysis: Benchmarks and Insights from PTB-XL,” arXiv:2004.13701v1 (28 Apr. 2020) which discloses electrocardiography is a very common, non invasive diagnostic procedure and its interpretation is increasingly supported by automatic interpretation algorithms. The progress in the field of automatic ECG interpretation has up to now been hampered by a lack of appropriate datasets for training as well as a lack of well-defined evaluation procedures to ensure comparability of different algorithms. To alleviate these issues, we put forward first benchmarking results for the recently published, freely accessible PTB-XL dataset, covering a variety of tasks from different ECG statement prediction tasks over age and gender prediction to signal quality assessment. We find that convolutional neural networks, in particular resnet- and inception-based architectures, show the strongest performance across all tasks outperforming feature-based algorithms by a large margin. These results are complemented by deeper insights into the classification algorithm in terms of hidden stratification, model uncertainty and an exploratory interpretability analysis. We also put forward benchmarking results for the ICBEB2018 challenge ECG dataset and discuss prospects of transfer learning using classifiers pretrained on PTB-XL. With this resource, we aim to establish the PTB-XL dataset as a resource for structured benchmarking of ECG analysis algorithms and encourage other researchers in the field to join these efforts.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Joey Burgess whose telephone number is (571)270-5547. The examiner can normally be reached Monday through Friday 9-6.
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, Kambiz Abdi can be reached on 571-272-6702 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.
/JOSEPH D BURGESS/ Primary Examiner, Art Unit 3685
Prosecution Insights