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 .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on Apr. 21, 2026 is in compliance with the
provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the
examiner.
Response to Amendment
The amendment filed May 26, 2026 has been entered. Claims 11-30 remain pending in the application. Applicant’s amendments to the Specification, Drawings, and Claims have overcome each and every objection, and 112 rejection previously set forth in the Non-Final Office Action mailed April 20, 2026.
Response to Arguments
Drawings:
Applicant amended drawings and addressed all previous objections and the previous objections have been withdrawn.
Specification:
Applicant amended specification and addressed all previous objections and the previous objections have been withdrawn.
Claim Objections:
Applicant amended claims and addressed all previous objections and the previous objections have been withdrawn.
35 USC § 112:
Applicant amended claims and addressed all previous 35 USC 112 rejections and the previous rejections have been withdrawn.
35 USC § 101:
Applicant's arguments on pages 16-19 filed May 26, 2026 have been fully considered but they are not persuasive.
On page 17 of Applicant’s response, applicant argues that the claims do not recite an abstract idea. Applicant argues that the step of obtaining electrogram data from an ECG for a patient is not a mental step which could practically be performed in the human mind and is a data acquisition step which is performed by a computer. Examiner respectfully disagrees. Under the broadest reasonable interpretation of this limitation, this limitation can be done by a human obtaining an already acquired ECG of the patient and observing it. It is further noted that the rejection already states that “other than reciting” computer components, a person would be capable of performing the step of obtaining ECG data.
Applicant’s argument that the steps of applying the encoded electrogram data to the trained Bayesian-based model to diagnose a particular type of arrhythmia and encoding data from the ECG for the patient by clustering each segment of the electrogram data could not practically be performed in the human mind has been fully considered but is not found persuasive in view of new grounds of rejection under 35 U.S.C. 101 set forth in this rejection necessitated by amended claim limitations. Claims now recite steps of training, encoding and applying are based on Bayesian-based model. As indicated in the rejection below, the specification discloses Bayesian model on pages 24 and 25 as a model based on mathematical formulas. Since claims now recite Bayesian model, which is clearly described as a mathematical model, claims are now reciting a mathematical concept.
On pages 18-19 of Applicant’s response, applicant argues that the claims improve machine learning in ECG systems and improves diagnostics. Applicant points to the specification to argue that the device can diagnose more accurately than previous technologies and has the ability of single-sample or small-sample learning, and can identify and judge rare or unseen arrhythmias and therefore improves the technology. Applicant elaborates that instead of using a deep learning model trained on various features which struggles with weak arrhythmias, the claimed invention assigns ECG segments into clusters of similar segments, encodes the ECG segments by cluster number, and applies the encoded ECG segments to a Bayesian-based dynamic programming algorithm which results in an improved model that generates more accurate results. First, it is noted that while ECG data is being used as part of the evaluation using Bayesian-based model, ECG parts are not part of the claimed inventions. Further, MPEP 2106.05(a)(II) states the following:
However, it is important to keep in mind that an improvement in the abstract idea itself (e.g. a recited fundamental economic concept) is not an improvement in technology.
In this case, the stated improvement is in the use of Bayesian-based dynamic programming algorithm for more accurate diagnosis and it appears that the improvement is in the abstract idea, which would not be considered an improvement in technology.
35 USC § 103:
Applicant’s arguments, see pg. 19-21, filed May 26, 2026, with respect to Sannino and Goldberg, have overcome the previous 35 USC 103 rejections from the May 26, 2026 Office Action. Examiner agrees with applicant that Sannino, Goldberg, and Guerrero alone or in combination do not disclose all of the limitations including a Bayesian-based model wherein each segment includes one of a P-wave, a P-Q interval, a QRS complex, an S-T interval, a T wave, or a T-P interval. However, during the interview on Apr. 27, 2026, reference to the prior 35 USC 103 rejection for claim 15 was discussed as indicated on the interview summary mailed Apr. 29, 2026 and Guerrero does teach the various segments. In response to applicant's arguments against the references individually, one cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references. See In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981); In re Merck & Co., 800 F.2d 1091, 231 USPQ 375 (Fed. Cir. 1986). Although it is true that Sannino, Goldberg, and Guerrero do not use Bayesian-based models, Sannino does not have assigned cluster numbers or particular segments of interest, Goldberg also does not segment into the particular claimed segments, and Guerrero does not cluster the segments, Sannino could be modified with reasonable expectation of success. As taught by Guerrero, segmenting the heart into a P-wave, a P-Q interval, a QRS complex, an S-T interval, a T wave, or a T-P interval is beneficial for showing different functions of the heart which can then be used to show different arrythmias depending on where the issues arise. Further, Goldberg teaches the benefit of clustering to properly diagnose arrythmia since they are dangerous medical conditions and proper detection is important.
As Sannino, Goldberg, and Guerrero do not teach a Bayesian-based model, a new 35 USC 103 rejection below is being applied using Sannino, Goldberg, and Guerrero in light of the amendments to the claim.
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 11-30 are rejected under 35 U.S.C. 101 because the claimed invention is directed to the abstract ideas of “obtaining electrogram data from an ECG for a patient”, “encoding the electrogram data from the ECG for the patient by clustering each segment of the electrogram data for the patient according to the cluster number assigned to the cluster of the plurality of clusters specific to a particular type of each segment”, and “applying the encoded electrogram data from the ECG for the patient to the trained Bayesian-based model to diagnose a particular type of arrhythmia for the patient” without significantly more.
Step 1
Claims 11-18 recite a method, claims 19-24 recite a machine, and claims 25-30 recite an article of manufacture, and therefore fall within the statutory categories.
Step 2A, Prong 1
Claims 11, 19, and 25 recites the limitations of obtaining electrogram data from an ECG for a patient; encoding the electrogram data from the ECG for the patient by clustering each segment of the electrogram data for the patient according to the cluster number assigned to the cluster of the plurality of clusters specific to a particular type of each segment; and applying the encoded electrogram data from the ECG for the patient to the trained Bayesian-based model to diagnose a particular type of arrhythmia for the patient. The limitations, as drafted, are a process that, under its broadest reasonable interpretation, covers mental process and mathematical concept but for the recitation of “by one or more processors”, which is a computer processor, “non-transitory computer-readable medium storing instructions” in claims 19 and 25, and “a computing device” in claim 19. That is, other than reciting the “one or more processors”, “non-transitory computer-readable medium storing instructions”, and “a computing device” nothing in the claims precludes the steps from practically being performed in the human mind. MPEP 2106.04(a)(2)(I) states that "The mathematical concepts grouping is defined as mathematical relationships, mathematical formulas or equations, and mathematical calculations"" and MPEP 2106.04(a)(2)(III) states that the courts consider a mental process (thinking) that "can be performed in the human mind, or by a human using a pen and paper" to be an abstract idea. For example, aside from the “training a neural network”, “processor”, “non-transitory computer-readable medium storing instructions”, and “a computing device” language, the claim encompasses inspecting electrogram data from an ECG of a patient, clustering the ECG data by each segment of the data, and comparing this data to previously acquired ECG data from a plurality of patients to decide which type of arrhythmia is present using clustering with Bayesian-based model. The specification discloses Bayesian model on pages 24 and 25 as a model based on mathematical formulas. As such, these limitations are nothing more than a mental processs of a person looking at ECG data and preforming a mathematical process using ECG data to determine arrhythmia.
Step 2A, Prong 2
The claim recites additional elements: “processor”, “non-transitory computer-readable medium storing instructions”, and “a computing device” to perform the abstract steps.
The “processor”, “non-transitory computer-readable medium storing instructions”, and “a computing device” read on a computer implemented system and are recited at a high level of generality, i.e., as a generic processor, performing a generic computer function of processing data and displaying that data. This generic limitation is no more than mere instructions to apply the exception using generic computer components. Accordingly, this additional limitation does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea.
Step 2B
As discussed with respect to Step 2A Prong Two, the additional elements in the claim amount to no more than mere instructions to apply the exception using a generic computer component. The same analysis applies here in 2B, i.e., mere instructions to apply an exception on a generic computer cannot integrate a judicial except into a practical application at Step 2A or provide an inventive concept in Step 2B.
Under 2019 PEG, a conclusion that an additional element is insignificant extra-solution activity in Step 2A should be re-evaluated in Step 2B to determine if it is more than what is well-understood, routine, conventional activity in the field. The specification in pg. 16 ¶1-2 and pg. 29 does not provide any indication that the computer processor is anything other than a generic, off-the-shelf computer component. Court decisions cited in MPEP 2106.05(d)(II) indicate that computer‐implemented processes not to be significantly more than an abstract idea (and thus ineligible) where the claim as a whole amounts to nothing more than generic computer functions merely used to implement an abstract idea, such as an idea that could be done by a human analog (i.e., by hand or by merely thinking). Accordingly, a conclusion that the generic computer functions merely being used to implement an abstract idea is well-understood, routine, conventional activity is supported under Berkheimer Option 2.
Dependent claims 12-13, 15, 17-18, 20-21, 23, 26-27, and 29 further limits the process of obtaining electrogram data from an ECG for a patient, encoding the electrogram data from the ECG for the patient by clustering each segment of the electrogram data, and applying the encoded electrogram data from the ECG for the patient to the trained Bayesian-based model to diagnose a particular type of arrhythmia for the patient. Therefore, these claims further limit the abstract idea already indicated in independent claims 11, 19, and 25 and they are ineligible for the same reasons provided for claims 11, 19, and 25 above.
For these reasons, there is no inventive concept in the claims and thus they are ineligible.
Dependent claims 14, 22, and 27 further limit the step of clustering the ECG and add additional abstract limitations of “separately counting a distribution of lengths for each type of segment”, “calculating an average length of each type of segment”, “determining a number of control points and a number of indexes for each type of segment according to the average length”, and “determining a set of features for each segment according to the average length, the number of control points, and the number of indexes for the type of segment”. The claim encompasses counting lengths, calculating an average, determining a number from this average, and determining features based off these values. Calculating an average is a mathematical process. These limitations are nothing more than a person looking at ECG data, preforming a mathematical process, and determining numbers and features based on their observations. This is a mental process. For these reasons, there is no inventive concept in the claims and thus they are ineligible.
Dependent claims 16, 24, and 30 add the abstract limitation of “classifying the feature vector for the ECG using naive Bayes.” According to geeksforgeeks.org, “Naive Bayes is a machine learning classification algorithm that predicts the category of a data point using probability” (geeksforgeeks.org, Feb. 27, 2026). Therefore, this limitation is a mathematical process and there is no inventive concept in the claims and thus they are ineligible.
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.
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.
Claims 11, 19, and 25 are rejected under 35 U.S.C. 103 as being unpatentable over Sannino et al. (Future Generation Computer Systems, 2018, as provided on the IDS submitted Apr. 4, 2023, hereinafter referred to as “Sannino”), Guerrero et al. (US 20020138013 A1, published Sept. 26, 2002, hereinafter referred to as “Guerrero”), Goldberg et al. (US 20220068483 A1, published Mar. 3, 2022, hereinafter referred to as “Goldberg”), and Tran (US 20070276270 A1, published Nov. 29, 2007, hereinafter referred to as “Tran”).
Regarding claims 11, 19, and 25, Sannino teaches a method for analyzing an electrocardiogram (ECG) to detect arrhythmia using machine learning techniques (“In this paper, we have proposed an approach based on a Deep Neural Network (DNN) for the automatic classification of abnormal ECG beats, differentiated from normal ones.” in section 5 pg. 454), the method comprising: training, by one or more processors (machine learning models inherently operate on a system with a processor), a model to detect arrhythmia using (“the final dataset was composed of a total of 4576 items, 2288 representing the normal beats class (N) and 2288 representing the abnormal beats class (A). This dataset, thanks to the information contained in subid in each item, was divided into the training (DS1) and test (DS2) set… More details about the item distribution over the testing and training sets are reported in Table 1” in section 3.3 pg. 452), for a plurality of ECGs for a plurality of patients (“In this work, we have used the MIT-BIH Arrhythmia Database, a dataset of standard test material used since 1980 in innumerable scientific works for the evaluation of arrhythmia detectors and classifiers. It was compiled by collecting 24-hour ECG recordings from 47 subjects” section 3.1 pg. 448), (i) encoded electrogram data from the ECG indicating characteristics of each segment of a heartbeat (“Starting from the MIT-BIH Arrhythmia Database we have created a new dataset by applying to the ECG signals the following processing steps… Signal Segmentation: to divide the ECG signal into single beats that will be classified as normal or abnormal.” section 3.2 pg. 448 and “Each item i of the new dataset consists of: i = subid; samplen |n=1:50; PreRR; PostRR; LocalRR; GlobalRR; class; where: subid identifies the subject; sample1 , ... , sample50 represent the beat samples; PreRR; PostRR; LocalRR; GlobalRR are the extracted attributes defined in the previous section; and class represents the class, encoded as 1 for normal beats and 2 for abnormal ones.” in section 3.3 pg. 451), (ii) an indication of whether the patient has arrhythmia (“Signal Segmentation: to divide the ECG signal into single beats that will be classified as normal or abnormal.” section 3.2 pg. 448) and, for a subset of the plurality of patients that have arrhythmia, (iii) an indication of a type of arrhythmia of a plurality of types of arrhythmia (“Of these 84,615 items, 66,750 represent normal beats (N), 2288 represent abnormal beats (premature ventricular contractions (V), supra-ventricular premature beats (S), or a fusion of ventricular and normal beats (F))” in 3.3 pg. 451).
Sannino does not disclose wherein the model is Bayesian-based, wherein the electrogram data is encoded by clustering each segment into a cluster of a plurality of clusters each assigned a cluster number, wherein each segment includes one of: a P-wave, a P-Q interval, a QRS complex, an S-T interval, a T wave, or a T-P interval; and obtaining, by the one or more processors, electrogram data from an ECG for a patient; encoding, by the one or more processors, the electrogram data from the ECG for the patient by clustering each segment of the electrogram data for the patient according to the cluster number assigned to the cluster of the plurality of clusters specific to a particular type of each segment; and applying, by the one or more processors, the encoded electrogram data from the ECG for the patient to the trained Bayesian-based model to diagnose a particular type of arrhythmia for the patient and the claim 19 and 25 limitations of a computing device for analyzing and identifying electrocardiogram, the computer device comprising a non-transitory computer-readable medium storing instructions thereon, that when executed by the one or more processes, cause the computing device to perform the steps above.
Guerrero’s invention relates to improved methods and systems for analysis of dynamic electrocardiograms and other similar waves of biological origin with the purpose of facilitating improved diagnosis of pathological states in human and veterinary medicine (¶[0002]). Referring to FIG. 1, there is shown an exemplary Holter electrocardiogram. The P wave is the ECG representation of the atrial depolarization which cause its contraction. PQ is the segment between the P and the Q; it represents the delay of the electrical wave of depolarization at the atrioventricular node to allow the contraction of the atria and fill the ventricles before the latter depolarize and expel blood into the body. Ta (a microvolt shift in the PQ not present in this figure) is due to abnormal atrial repolarization caused by ischemia. The QRS is the ECG representation of ventricular depolarization which cause ventricular contraction. The ST segment represents the initial repolarization of the ventricles. The ascending limb of the T wave represents epicardial (outer surface of the ventricle) repolarization which changes into endocardial (inner surface of the ventricle) and mesocardial repolarization at the apex of the upright T wave. Ventricular repolarization is complete when the T wave returns to the isoelectric line. Several different morphologies of the T wave are associated with non-homogeneous repolarization, a sign of myocardial cell hypoxygenation and risk for lethal arrhythmia. TP is the isoelectric segment between the offset of the T and the onset of the P waves. TP must be considered as the isoelectric line when Ta is present. The second beat is a premature depolarization characterized by abnormal QRS and T morphology as well as greater voltage and duration than the normal beats (¶[0024]).
Therefore, it would have been obvious to a person having ordinary skill in the art at the time of filing to have the segments include one of a P wave, P-Q interval, QRS complex, S-T interval, T wave, or T-P interval as taught by Geurrero in the method and device of Sannino since each of these segments represent a different function of the heart and therefore can show information related to different arrythmias.
Sannino and Guerrero do not disclose wherein the model is Bayesian-based, wherein the electrogram data is encoded by clustering each segment into a cluster of a plurality of clusters each assigned a cluster number; and obtaining, by the one or more processors, electrogram data from an ECG for a patient; encoding, by the one or more processors, the electrogram data from the ECG for the patient by clustering each segment of the electrogram data for the patient according to the cluster number assigned to the cluster of the plurality of clusters specific to a particular type of each segment; and applying, by the one or more processors, the encoded electrogram data from the ECG for the patient to the trained Bayesian-based model to diagnose a particular type of arrhythmia for the patient and the claim 19 and 25 limitations of a computing device for analyzing and identifying electrocardiogram, the computer device comprising a non-transitory computer-readable medium storing instructions thereon, that when executed by the one or more processes, cause the computing device to perform the steps above.
Goldberg’s invention relates to methods for cardiac mapping, specifically in relation to clustering. As described in reference to FIG. 1 and FIG. 2, the system 100, 200 can process 161, 220 biometric data associated with a patient. The system 100, 200 can receive 163, 260 the biometric data from measurements performed on a patient 125 during an EP procedure. The system 100 may obtain ECGs (¶[0031]). The classification engine 101 employed by the system 100 herein manipulates and evaluates the ECG data to produce improved tissue data that enables more accurate diagnosis, images, scans, and/or maps for treating an abnormal heartbeat or arrhythmia (¶[0049]). The system's 300 (Fig. 3) components 320, 340, 360, may be employed by one or more software modules executable by one or more processors of the system 100, 200, described above with reference to FIG.1 and FIG.2. The data processor 320 may receive as an input biometric data 310 containing data collected from a patient during a medical procedure. The data processor 320 may process the received biometric data 310, for example, to improve the discriminatory properties of these data. The data processor 320 may feed the evaluation engine 340 with beat segments 335.1-335.M extracted from ECG data of the biometric data 310. Each beat segment (e.g., including ECG signals 335.m) is representative of the arrhythmia type it was produced from (¶[0054]). The evaluation engine 340 may classify the beat segments 330 into clusters (e.g., cluster A 352 and cluster B 354). Classification of a beat segment into a cluster may be based on the segment's beat characteristics and the output 350 of the evaluation engine 340 may be segments that are classified into clusters based on the segments' beat characteristics (¶[0055]). Fig.’s 4 and 6 show additional clustering methods. In step 730 of Fig. 7, a model is trained to predict a type of arrhythmia when presented with biometric data of a new patient where the received training dataset 710 may include biometric data measured by the system, 100 or 200 (¶[0065]-[0066]). The evaluation engine 340, may train a machine learning model 810 of Fig. 8 and may apply the trained model to predict the classification of an arrhythmia condition in a patient. Algorithms disclosed herein may be applied to train models based on a training dataset 820 (obtained by various modalities from patients experiencing an arrhythmia condition) and corresponding classification of the patients' arrhythmia condition (¶[0073]). The functions noted in the blocks may occur out of the order noted in the Figures (¶[0074]). The methods described herein may be implemented in a computer program, software, or firmware incorporated in a computer-readable medium for execution by a computer or processor (¶[0075]).
Therefore, it would have been obvious to a person having ordinary skill in the art at the time of filing to encode the electrogram data through clustering each segment of the heartbeat, and obtain electrogram data from an ECG for a patient and apply the encoded electrogram data from the ECG for the patient to the trained machine learning model to diagnose a particular type of arrhythmia for the patient as taught by Goldberg in the method and device of Sannino and Guerrero in order to diagnose arrhythmia during a procedure since they are common and dangerous medical conditions (¶[0018]). Further, clustering allows visualization of data associated with segments in each cluster (¶[0055]). It also would have been obvious to have a computing device for analyzing and identifying electrocardiogram, the computer device comprising a non-transitory computer-readable medium storing instructions accompanying the processor as taught by Goldberg in the device of Sannino and Guerrero in order to provide a system for the machine learning algorithm to operate on.
Sannino, Guerrero, and Goldberg do not disclose the clusters are assigned a cluster number, however, Goldberg taught labelling the clusters based on letters (A and B in ¶[0055]). Labelling the clusters based on numbers is a design choice that produces the same functionality as assigning the clusters a letter and therefore would have been obvious to a person having ordinary skill in the art to use.
Sannino, Guerrero, and Goldberg do not teach wherein the model is Bayesian-based.
Tran’s invention relates generally to methods and systems for monitoring a person with nodes forming a wireless mesh to detect heart related issues. A predictive model, including time series models such as those employing autoregression analysis and other standard time series methods, dynamic Bayesian networks and Continuous Time Bayesian Networks, or temporal Bayesian-network representation and reasoning methodology, is built, and then the model, in conjunction with a specific query makes target inferences (¶[0192]). Bayesian networks provide not only a graphical, easily interpretable alternative language for expressing background knowledge, but they also provide an inference mechanism; that is, the probability of arbitrary events can be calculated from the model. Intuitively, given a Bayesian network, the task of mining interesting unexpected patterns can be rephrased as discovering item sets in the data which are much more--or much less--frequent than the background knowledge suggests. These cases are provided to a learning and inference subsystem, which constructs a Bayesian network that is tailored for a target prediction. The Bayesian network is used to build a cumulative distribution over events of interest (¶[0193]).
Therefore, it would have been obvious to a person having ordinary skill in the art at the time of filing to use a Bayesian-based model as taught by Tran in the method and device of Sannino, Guerrero, and Goldberg since these models are easily interpretable and provide data and patterns that are more difficult to uncover.
Claims 12, 14, 16, 20, 22, 24, 26, 28 and 30 are rejected under 35 U.S.C. 103 as being unpatentable over Sannino, Guerrero, Goldberg, and Tran (hereinafter referred to as “modified Sannino”), as applied to claims 11, 19, and 25 above, in even further view of Iguazio (What is a feature vector?, Iguazio, May 2022).
Modified Sannino teaches the method of claim 11 and the device of claims 19 and 25.
Regarding claims 12, 20, and 26, Sannino also teaches wherein obtaining the electrogram data from the ECG for the patient includes: locating, by the one or more processors, feature points in the electrogram data according to P-QRS-T characteristic waves for N heartbeats (“Starting from the MIT-BIH Arrhythmia Database we have created a new dataset by applying to the ECG signals the following processing steps, summarized in Fig. 4 … Peak Detection: to determine the positions of all peaks of interest found in the ECG signal. The p-wave and t-wave positions are useful for the signal segmentation. Instead, the r-peak detection is essential for the temporal features extraction.” In section 3.2 pg. 448); segmenting, by the one or more processors, the feature points into segments for each of the N heartbeats (“Signal Segmentation: to divide the ECG signal into single beats that will be classified as normal or abnormal.” In section 3.2 pg. 448).
Sannino does not disclose wherein encoding the electrogram data from the ECG for the patient by clustering each segment of the electrogram data includes: clustering, by the one or more processors, each segment into the cluster of the plurality of clusters using k-means clustering, wherein each segment is clustered according to the cluster number assigned to the cluster for the segment; the method further comprising: splicing, by the one or more processors, encoded data for each segment to generate a feature vector for the N heartbeats for the ECG; and wherein applying the encoded electrogram data from the ECG for the patient to the trained Bayesian-based model includes applying the feature vector for the ECG to the trained Bayesian-based model to diagnose the particular type of arrhythmia for the patient.
Goldberg teaches k-mean clustering in Fig. 6. The method 600 may be used to classify beat segments into clusters based on the segments’ beat characteristics (¶[0063]). Classification of the data points in a two-dimensional spaced according to a K-means clustering algorithm is demonstrated. As shown, the data points are to be classified into two clusters (K=2)—cluster A and cluster B 610. The process of assigning data points into clusters may repeat until a maximum iteration number is reached or until there is no change in data points assignments into clusters (¶[0065]).
Therefore, it would have been obvious to a person having ordinary skill in the art at the time of filing to cluster the segments using k-means as taught by Goldberg in the method and device of Sannino in order to pre-define the number of clusters.
Modified Sannino does not disclose the method further comprising: splicing, by the one or more processors, encoded data for each segment to generate a feature vector for the N heartbeats for the ECG; and wherein applying the encoded electrogram data from the ECG for the patient to the trained Bayesian-based model includes applying the feature vector for the ECG to the trained Bayesian-based model to diagnose the particular type of arrhythmia for the patient.
Iguazio teaches that feature vectors represent features used by machine learning models in multi-dimensional numerical values. As machine learning models can only deal with numerical values, converting any necessary features into feature vectors is crucial. Machine learning models can only deal with quantitative data. As such, we must always convert features of observed phenomena into numerical values and feed them into a machine learning model in the same order. In short, we must represent features in feature vectors (pg. 1).
Therefore, it would have been obvious to a person having ordinary skill in the art at the time of filing to generate feature vectors from the encoded data and apply the feature vector to the trained Bayesian-based model as taught by Iguazio in the device of modified Sannino because machine learning models require numerical values to work and this reduces the amount of noise being fed into the system.
Modified Sannino and Iguazio teach the method of claim 12 and the device of claims 20 and 26.
Regarding claims 14, 22, and 28, Goldberg also teaches wherein clustering each segment includes clustering the segment into one of the plurality of clusters by comparing the determined set of features for the segment to a set of features for each of the plurality of clusters (“The evaluation engine 340 of Fig. 3 may classify the beat segments 330 into clusters (e.g., cluster A 352 and cluster B 354). Classification of a beat segment into a cluster may be based on the segment's beat characteristics and the output 350 of the evaluation engine 340 may be segments that are classified into clusters based on the segments' beat characteristics” (¶[0055]).
Although Goldberg describes mathematically operations done to the data, Modified Sannino and Iguazio do not disclose the method and device further comprising: separately counting, by the one or more processors, a distribution of lengths for each type of segment; calculating, by the one or more processors, an average length of each type of segment; determining, by the one or more processors, a number of control points and a number of indexes for each type of segment according to the average length; and determining, by the one or more processors, a set of features for each segment according to the average length, the number of control points, and the number of indexes for the type of segment.
It would be obvious to a person having ordinary skill in the art at the time of filing to count a distribution of lengths and average the lengths and then determine the control points and indexes from the average because these are methods to organize data into a more meaningful format and allows for better analysis of patterns and trends. It would be obvious to then use these factors when determining the features of the segments since there is a reduction of noise and the data will be more representative of the whole.
Regarding claims 16, 24, and 30, Sannino teaches further comprising: classifying, by the one or more processors, the feature vector (taught by Iguazio in claim 12 above) for the ECG using naive Bayes (“To demonstrate the quality of the proposed DNN in terms of accuracy, sensitivity, and specificity, we carried out a precise comparison by taking into account eleven other well-known classifiers. For each category, we selected the most representative algorithms, in terms of classification performance. We considered Naive Bayes from among the Bayesian techniques” in section 4, pg. 453).
Claims 13, 21, and 27 are rejected under 35 U.S.C. 103 as being unpatentable over modified Sannino and Iguazio, as applied to claims 12, 20, and 26 above, and in even further view of Patil et al. (US 20180206751 A1, published July 26, 2018, hereinafter referred to as “Patil”).
Modified Sannino and Iguazio teach the method of claim 12 and the device of claims 20 and 26.
Regarding claims 13, 21, and 27, Sannino also teaches wherein locating the feature points includes: determining, by the one or more processors, a position of an R peak using an R peak detection algorithm (“Peak Detection: to determine the positions of all peaks of interest found in the ECG signal. The p-wave and t-wave positions are useful for the signal segmentation. Instead, the r-peak detection is essential for the temporal features extraction.” In section 3.2 pg. 448); the method for analyzing an ECG further comprising: decomposing, by the one or more processors, the electrogram data into the N heartbeats according to the feature points (“Signal Segmentation: to divide the ECG signal into single beats that will be classified as normal or abnormal.” In section 3.2 pg. 448).
Guerrero also teaches dividing, by the one or more processors, each heartbeat of the N heartbeats into the 6 segments according to the feature points.
Referring now to FIG. 1, there is shown an exemplary Holter electrocardiogram. The P wave is the ECG representation of the atrial depolarization which cause its contraction. PQ is the segment between the P and the Q; it represents the delay of the electrical wave of depolarization at the atrioventricular node to allow the contraction of the atria and fill the ventricles before the latter depolarize and expel blood into the body. Ta (a microvolt shift in the PQ not present in this figure) is due to abnormal atrial repolarization caused by ischemia. The QRS is the ECG representation of ventricular depolarization which cause ventricular contraction. The ST segment represents the initial repolarization of the ventricles. The ascending limb of the T wave represents epicardial (outer surface of the ventricle) repolarization which changes into endocardial (inner surface of the ventricle) and mesocardial repolarization at the apex of the upright T wave. Ventricular repolarization is complete when the T wave returns to the isoelectric line. Several different morphologies of the T wave are associated with non-homogeneous repolarization, a sign of myocardial cell hypoxygenation and risk for lethal arrhythmia. TP is the isoelectric segment between the offset of the T and the onset of the P waves. TP must be considered as the isoelectric line when Ta is present. The second beat is a premature depolarization characterized by abnormal QRS and T morphology as well as greater voltage and duration than the normal beats (¶[0024]).
Therefore, it would have been obvious to a person having ordinary skill in the art at the time of filing to divide the heartbeats into 6 segments according to the feature points, wherein the segments include a P wave, P-Q interval, QRS complex, S-T interval, T wave, and T-P interval as taught by Geurrero in the method and device of modified Sannino, Iguazio, Patil, and Wang since each of these segments represent a different function of the heart and therefore can show information related to different arrythmias.
Modified Sannino and Iguazio do not disclose using, by the one or more processors, a moving window to iteratively query both sides of the R peak; determining, by the one or more processors, a minimum value in the moving window positioned to the left of the R peak as a position of a Q peak; determining, by the one or more processors, a minimum value in the moving window positioned to the right of the R peak as a position of an S peak; performing, by the one or more processors, time window traversal from a QRS complex according to the Q, R, and S peaks to identify a P wave and a T wave, including: calculating, by the one or more processors, a maximum distance between a starting point and an ending point of an auxiliary line segment at each point in windows to the left and right of the QRS complex according to a boundary detection algorithm based on a local distance transformation to find the starting point and the ending point of the P wave and T wave; establishing, by the one or more processors, a first time window before a Q start of the QRS complex, and a second time window after an S end of the QRS complex; detecting, by the one or more processors, a P peak within the first time window using the R peak detection algorithm; and detecting, by the one or more processors, a T peak within the second time window using the R peak detection algorithm.
Patil’s invention relates to Electrocardiography (ECG) data. More particularly, to auto labeling of activity on ECG data indicating the activity performed by the subject while obtaining the ECG (¶[0001]). The processing unit (Fig. 1, 102) is also provided for identifying the fiducial points on the ECG signal (Fig. 2, 203). Subsequent to applying the adaptive segment pass filter, the fiducial points P, Q, R, S and T points are identified by the processing unit (102) in a given segment of the ECG signal. The R peak being more prominent in the ECG is initially identified using the frequency domain approach (¶[0031]). The first lowest minima on the left and right side of R peak forms the Q and S point, the time window of the search of Q and S with respect to R is empirically defined (¶[0034]). In order to determine P wave, a time window may be set prior to the beginning of QRS complex fiducial and QRS onset. The time window that approximately contains P wave is set heuristically and extended from QRS onset to the beginning of heartbeat. The beat begin fiducial point can be determined by searching of first isoelectric sample prior to the start of atrial deflection. For the detection of P waves, delineator computes the slope threshold, which is obtained by using the first derivative approach (¶[0035]). The zero crossing of the differentiated signal maps to the P peak and succeeding the lowest crest on the differentiated signal maps to the P offset. Fragment analysis approach is followed in determination of T peak by considering the segment in the ECG signal after S point (ST segment in FIG. 4). Fragment analysis is performed (204) on the ECG signal after identifying the fiducial points, by the processing unit (102). R-R deviation, inversion of T peak, distance between fiducial points (QRS, P-QRS etc.) are computed. Other features that may be extracted from the fragment analysis are time interval from R peak to P off-set and P peak, time interval from R peak to S, and time interval from S to J point (¶[0036]-[0038]).
Therefore, it would have been obvious to a person having ordinary skill in the art at the time of filing to find the minimum value to the left of the R wave as the Q point and to the right the S point and set up a time window to the left of the QRS to find the P wave and to the right of the QRS to find the T wave as taught by Patil in the method and device of modified Sannino and Iguazio in order to automatically labeling of activity of a subject on electrocardiography (ECG) data (¶[0013]).
Modified Sannino, Iguazio, and Patil do not disclose performing, by the one or more processors, time window traversal from a QRS complex according to the Q, R, and S peaks to identify a P wave and a T wave, including: calculating, by the one or more processors, a maximum distance between a starting point and an ending point of an auxiliary line segment at each point in windows to the left and right of the QRS complex according to a boundary detection algorithm based on a local distance transformation to find the starting point and the ending point of the P wave and T wave.
Wang’s study relates to Automatic detection of P-wave in an electrocardiogram (ECG) using an improved method based on local distance transform. Auxiliary algorithms have also been introduced to improve the precision of local distance transform, such as recognition of horizontal segment and rising or declining segment; then, feedback comes to narrow or widen the P wave searching area for the need of local distance transform (pg. 2197). In each heartbeat, the QRS complex is used as a reference for P waves, P wave searching area is identified and extended from the R peak backwardly to one third of the previous RR interval, which makes sure the presence of P waves in this area. Local distance transform mainly refers to calculate maximum distance point from each of point on signal curve to the line which connects the start and end points of auxiliary segment including the extracted feature points, the point is identified as signal curve feature points, that is the onset and end of signal curve. Fig. 3 is P wave boundary detection process based on local distance transform (pg. 2198). In theory, the end of auxiliary segment ought to locate the peak point of signal peak form, but waveform is difficult to represent prefect signal peak form actually. Adopting peak point as endpoint of distance transform exists bigger error for not ideal P wave, so we search eligible maximum value in P wave peak area, which to ensure accuracy of transform. The study determines the P wave onset and the P wave end (pg. 2199). Simplicity and efficiency of the algorithms make it suitable for transplanting to wearable medical devices whose processing ability is weak (pg. 2200).
Therefore, it would have been obvious to a person having ordinary skill in the art at the time of filing to calculate a maximum distance between a starting point and an ending point of an auxiliary line segment at each point in windows to the left and right of the QRS complex according to a boundary detection algorithm based on a local distance transformation to find the starting point and the ending point of the P wave and T wave as taught by Wang in the method and device of modified Sannino, Iguazio, and Patil in order to create a simple and efficient algorithm to detect the start and end points of the P wave. Further, it would have been obvious to apply this to the T wave by changing the window of interest since the T wave is a stronger signal.
Claims 15, 23, and 29 are rejected under 35 U.S.C. 103 as being unpatentable over modified Sannino and Iguazio, as applied to claims 12, 20, and 26 above, as evidenced by Cross Validated (2020).
Regarding claims 15, 23, and 29, Goldberg also teaches wherein clustering each segment using k-means clustering includes: configuring, by the one or more processors, a set of clusters for a particular type of the segment; assigning, by the one or more processors, a number to each cluster in the set of clusters (“The evaluation engine 340 of Fig. 3 may classify the beat segments 330 into clusters (e.g., cluster A 352 and cluster B 354). Classification of a beat segment into a cluster may be based on the segment's beat characteristics and the output 350 of the evaluation engine 340 may be segments that are classified into clusters based on the segments' beat characteristics” (¶[0055]); computing, by the one or more processors, distances between the segment and cluster centroids for the set of clusters using Euclidean distance (K-means implicitly uses Euclidean distances – Cross Validated pg. 1); clustering, by the one or more processors, the segment into one of the set of clusters according to the distances; and assigning, by the one or more processors, a cluster number to the segment corresponding to the cluster assigned to the segment (Fig. 6 “two initial centroids are determined for each cluster, set apart from each other. The centroids are denoted by hollow circles 620, 630. Then, in a first iteration, the data points may be classified into two clusters based on their similarity to each centroid 640. Thus, data points to the left of the boundary line may be assigned to cluster A 642 and data points to the right of the boundary line may be classified to cluster B 644. Next, the centroids of the clusters are computed 650 based on the data points within each cluster 652, 654. In a second iteration, the data points may again be reclassified into two clusters based on their similarity to each of the centroids 660. Thus, data points to the left of the boundary line may be assigned to cluster A 662 and data points to the right of the boundary line may be assigned to cluster B 664. Next, the centroids of the clusters are recomputed 670 based on the data points within each cluster 672, 674. The process of assigning data points into clusters 640, 660 and recomputing the clusters' centroids 650, 670 may repeat until a maximum iteration number is reached or until there is no change in data points assignments into clusters.” In ¶[0064]).
Claim 17 is rejected under 35 U.S.C. 103 as being unpatentable over modified Sannino as applied to claim 11 above, in further view of Breed et al. (US 20020059022 A1, published May 16, 2002, hereinafter referred to as “Breed”).
Modified Sannino teaches the method of claim 11.
Regarding claim 17, Sannino also teaches wherein training the Bayesian-based model includes: extracting, by the one or more processors, ECG data samples for the plurality of patients; filtering, by the one or more processors, the ECG data samples to obtain a training data set and a testing data set (“the final dataset was composed of a total of 4576 items, 2288 representing the normal beats class (N) and 2288 representing the abnormal beats class (A). This dataset, thanks to the information contained in subid in each item, was divided into the training (DS1) and test (DS2) set… More details about the item distribution over the testing and training sets are reported in Table 1” in section 3.3 pg. 452 and “In this work, we have used the MIT-BIH Arrhythmia Database, a dataset of standard test material used since 1980 in innumerable scientific works for the evaluation of arrhythmia detectors and classifiers. It was compiled by collecting 24-hour ECG recordings from 47 subjects” section 3.1 pg. 448); and performing, by the one or more processors, data distribution (“Of these 84,615 items, 66,750 represent normal beats (N), 2288 represent abnormal beats (premature ventricular contractions (V), supra-ventricular premature beats (S), or a fusion of ventricular and normal beats (F)), and 14,828 represent unclassifiable beats (Q). Due to their non-classification, we removed from the dataset the last 14,828 items. Additionally, this was necessary in order to balance the dataset due to the fact that the classes were imbalanced, namely we had too many normal beats (N) compared to abnormal ones (V, Sand F). In fact, in these cases, conventional algorithms are often biased towards the majority class because their loss functions attempt to optimize quantities such as error rate, not taking into consideration the data distribution.” In section 3.3 pg. 451).
Modified Sannino does not disclose performing, by the one or more processors, class similarity analysis on the training data set and the testing data set to improve a discrimination degree of the electrogram data in the training data set and the testing data set.
Breed’s invention relates to the field of determining the occupancy state of the vehicle which entails sensing, detecting, monitoring and/or identifying various objects, and parts thereof, which are located within the passenger compartment of the vehicle. This is accomplished in part through a specific placement of transducers of the system, the use of a pattern recognition system, possibly a trained neural network and combinations of neural networks called modular neural, voting or ensemble neural networks, and/or a novel analysis of the signals from the transducers (¶[0005]). Since it is very easy to take large amounts data and yet large databases require considerably longer training time for a neural network, a test of the variability of the database can be made using a neural network. If, for example, after removing half of the data in the database, the performance of a trained neural network against the validation database does not decrease, then the system designer suspects that the training database contains a large amount of redundant data. Techniques such as similarity analysis can then be used to remove data that is virtually indistinguishable from other data. Since it is important to have a varied database, it is undesirable generally to have duplicate or essentially duplicate vectors in the database since the presence of such vectors can bias the system and drive the system more toward memorization and away from generalization (¶[0260]).
Therefore, it would have been obvious to a person having ordinary skill in the art at the time of filing to perform a similarity analysis on the training and testing data sets as taught by Breed in the method of modified Sannino in order to increase the performance of the algorithm and remove redundant factors therefore improving a discrimination degree.
Claim 18 is rejected under 35 U.S.C. 103 as being unpatentable over modified Sannino as applied to claim 11 above, in further view of Cao et al. (US 20210369131 A1, published Dec. 2, 2021, hereinafter referred to as “Cao”), and in even further view of Huang et al. (ICECEE, 2015, hereinafter referred to as “Huang”).
Modified Sannino teaches the method of claim 11.
Regarding claim 18, Sannino teaches wherein training the Bayesian-based model using the electrogram data for the plurality of patients includes: obtaining, by the one or more processors, ECG data samples for the plurality of patients (“In this work, we have used the MIT-BIH Arrhythmia Database, a dataset of standard test material used since 1980 in innumerable scientific works for the evaluation of arrhythmia detectors and classifiers. It was compiled by collecting 24-hour ECG recordings from 47 subjects” section 3.1 pg. 448).
Modified Sannino does not disclose for each of the ECG data samples: filtering, by the one or more processors, an electromyography (EMG) signal from the ECG data sample using a Butterworth low-pass filter; filtering, by the one or more processors, a power frequency interference signal from the ECG data sample using a 50 Hz finite impulse response notch filter with a Kaiser window function; and filtering, by the one or more processors, ECG baseline drift from the ECG data sample using an infinite impulse response zero-phase shift digital filter; and training, by the one or more processors, the Bayesian-based model using electrogram data from the filtered ECG data samples.
Cao’s invention relates to the technical field of data processing, and more particularly, to an electrocardiogram information ambulatory monitoring method and an ambulatory monitoring system. In the system, high frequency, low-frequency noise interference and baseline drift are eliminated by filtering to improve the accuracy of artificial intelligence analysis (¶[0104]). Differences in the lead, sample frequency and transmission data format used by different electrocardiogram devices may be eliminated, and the high frequency, low-frequency noise interference and baseline drift may be removed by digital signal filtering (¶[0105]). The digital signal filtering may adopt a high-pass filter, low-pass filter and median filtering respectively to eliminate power line interferences, electromyogram interferences and baseline drift interferences, so as to avoid the impact on subsequent analysis (¶[0106]). More specifically, a low-pass, high-pass Butterworth filter may be used for zero-phase shift filtering to eliminate the baseline drift and high-frequency noise interference, and to retain effective electrocardiogram signals. The median filtering may replace an amplitude of a sequence in a center of a window with a median of voltage amplitudes of data points in a sliding window of a preset length of time, and a low-frequency baseline drift may be eliminated (¶[0107]).
Therefore, it would have been obvious to a person having ordinary skill in the art at the time of filing to filter an EMG signal from the ECG data using a Butterworth low-pass filter and filter ECG baseline drift using an infinite impulse response zero-phase shift digital filter (in this case a Butterworth filter) as taught by Cao in the method of modified Sannino in order to improve the accuracy of the artificial intelligence analysis and to remove noise interference. The filtered samples could then be input into the training dataset.
Modified Sannino and Cao do not disclose filtering, by the one or more processors, a power frequency interference signal from the ECG data sample using a 50 Hz finite impulse response notch filter with a Kaiser window function.
Huang teaches that the ECG signal amplitude is very low, very susceptible to interference from the external environment, and the 50Hz power frequency interference is the main interference source of ECG signal. In the solution of many of the most commonly used method, is simple, is to design a fixed center frequency segment stop filter is a notch filter to eliminate the power line interference (pg. 453). This paper uses the window function method to design 50Hz filters. The first requirement of window function method to choose a proper ideal low-pass filter, because the impulse response is non causal and infinite impulse response, to intercept it using optimization window window function, so as to obtain the linear phase FIR filter and causality. In a number of window function, Kaiser window is the window function is close to the optimal window structure, it can adjust the filter parameters according to the different, so using the Kaiser window function in the filter design (pg. 454).
Therefore, it would have been obvious to a person having ordinary skill in the art at the time of filing to filter the power frequency interference signal from the ECG using a 50 Hz finite impulse response notch with a Kaiser window function as taught by Huang in the method of modified Sannino and Cao because ECG signal amplitude is very susceptible to interference from the external environment and since 50 Hz is the main interference source, this filtering method is successful at removing that noise.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Emily N Cirulnick whose telephone number is (571)272-9734. The examiner can normally be reached M-Th 8-5:30 and every other F 8-4:30ET.
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, Unsu Jung can be reached at (571) 272-8506. 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.
/E.N.C./ Patent Examiner, Art Unit 3792
/UNSU JUNG/ Supervisory Patent Examiner, Art Unit 3792
Prosecution Insights