Method and system for analyzing cardiac rhythm - Patents.com

Abstract

The present invention relates to a method (100) and system for analyzing electrocardiogram segments previously acquired from a cardiac device.

Description

[Technical field]

[0001] FIELD OF THEINVENTION The present invention relates to the field of electrophysiological signal analysis, and in particular to a system and method for analyzing electrocardiogram episodes obtained from a cardiac monitoring device to distinguish true positive episodes involving abnormal cardiac rhythms from false positive episodes involving normal cardiac rhythms. [Background technology]

[0002] 2. Background of the Invention An electrocardiogram (ECG or EKG) records the electrical signals from the heart. The waves in an ECG signal are referred to by letters.

[0003] An insertable or implantable cardiac monitor is a small cardiac monitoring device that continuously measures ECG signals over an extended period of time (e.g., up to several years). Some of these insertable or implantable cardiac monitors are further configured to continuously analyze the measured ECG signals to identify abnormal heart rhythms, also called episodes, and record only the episodes. The insertable or implantable cardiac monitor wirelessly transmits the recorded episodes (i.e., segments of the measured ECG signals) to an Internet-connected transmitter or other connected device, such as a mobile phone, which transmits the ECG signals over the Internet to a medical professional. Those medical professionals can identify cardiac episodes by observing or printing the ECG signals using a software platform and making a subjective decision based on their training and experience. In addition to automatic episode transmission, the insertable or implantable cardiac monitor is programmed to transmit an alert (with a potentially abnormal ECG signal) to the medical professional.

[0004] Conventional platforms for observing ECG signals received from an insertable or implantable cardiac monitor have several drawbacks. First, insertable or implantable cardiac monitors are designed to be sensitive and output an alert in response to every abnormal rhythm episode. As a result, insertable or implantable cardiac monitors output a significant number of false positive alerts. Requiring physicians to review and evaluate false positive alerts places an unnecessary burden on physicians, adds costs to the healthcare system, and can induce additional stress on the individual patients being monitored.

[0005] Second, insertable or implantable cardiac monitors from all major manufacturers output data onto their own device-specific platforms, therefore, medical professionals who treat patients with insertable or implantable cardiac monitors from different manufacturers must learn and use multiple different platforms.

[0006] Finally, conventional platforms are not configured to allow for software analysis of the ECG signal. Instead, conventional platforms typically output information to an electronic medical record system and a report to a physician who then observes and evaluates. The physician is then expected to make a subjective decision based on their training and experience. Summary of the Invention

[0007] overview The present invention thus relates to a computer-implemented method for analyzing electrocardiographic episodes previously acquired from a cardiac-connected device to distinguish true positive episodes involving abnormal cardiac rhythms from false positive episodes involving normal cardiac rhythms, said method comprising: - receiving the episodes, each episode including at least one segment of an electrocardiogram signal; - For each segment in one episode, o utilizing at least one algorithm to identify R waves within the segment and using said R waves to calculate at least one feature of the segment; o segmenting said segment into at least one sub-segment or into at least two overlapping sub-segments; For each sub-segment, utilizing R waves identified within the subsegment to calculate at least one feature of the subsegment; providing at least one feature of the subsegment and at least one feature of the segment as inputs to a machine learning algorithm and obtaining a score as output, the machine learning algorithm being configured to output the score; - classifying an episode using the scores obtained for each subsegment within the episode to distinguish true positive episodes from false positive episodes; - outputting the true positive episode; Includes.

[0008] Advantageously, the method of the present invention can reduce the number of false positive episodes reviewed by healthcare providers. Indeed, the combination of global information about the episode, obtained as feature(s) of the segment(s), with more local information, obtained as feature(s) of the subsegment(s), improves the efficiency of rejecting false positive episodes.

[0009] The method may output for all analyzed episodes their classification, i.e. true positive or false positive. Furthermore, the machine learning algorithm may be configured to classify abnormal cardiac rhythms into different classes associated with specific pathologies.

[0010] According to one embodiment, the score is a score vector, and the score vector obtained for each sub-segment within an episode is used to classify the episode in order to distinguish true positive episodes from false positive episodes. Advantageously, the information regarding the classification of abnormal heart rhythms is then provided to the score vector for each sub-segment, which is used to determine which abnormal heart rhythms are present therein for a true positive episode. This provides the user with more useful information. [Means for solving the problem]

[0011] According to one embodiment, the method includes providing the true positive episodes to a remote monitoring platform to augment the information available at the remote monitoring platform itself and ultimately make the output of the method available to healthcare providers across the platform. The term "remote" refers to the monitoring platform not being in the vicinity of the subject carrying the cardiac connectivity device.

[0012] According to one embodiment, for each segment within an episode, the at least one feature of the segment that is calculated is at least one of the following: morphological and / or rhythmic features. Morphological features are advantageously extracted for a segment, since they provide an analysis of the morphology of the P-waves and QRS complexes along the entire segment, and can capture irregularities that are characteristic of prolonged episodes (notably AT / AF and ventricular tachycardia). Rhythm features are advantageously extracted for a segment, since they provide an analysis of the R-peaks statistically, capturing patterns of R-peak intervals. Such patterns can be very useful to discriminate between abnormalities of prolongation (normal rhythms with AT / AF or ventricular / atrial premature beats).

[0013] According to one embodiment, for each subsegment, the subsegment features calculated are at least one of the following: rhythm features, variability features, neural network features and / or spectral features. Advantageously, variability features are extracted for the subsegments only, since the analysis of the variability of the signal in different time frames allows the quantification of the amount of the signal that can be explained by cardiac or non-cardiac origin (artifacts). The results from the analysis of the frequency of the cardiac signal characterize the regularity of the signal, so the spectral features obtained for the subsegments are particularly advantageous and can distinguish between low frequency regular signals (normal rhythm), low frequency irregular signals (various anomalies) and high frequency signals (artifacts).

[0014] The present invention also relates to a system for analyzing electrocardiographic episodes previously obtained from a cardiac device to distinguish true positive episodes involving abnormal cardiac rhythms from false positive episodes involving normal cardiac rhythms, said system comprising: - at least one input adapted to receive the episodes, each episode comprising at least one segment of an electrocardiogram signal; - For each segment in one episode, ● utilizing at least one algorithm to identify R waves within the segment and utilizing the R waves to calculate at least one feature of the segment; ● segmenting the segment into at least two overlapping sub-segments; ● For each sub-segment, o utilizing identified R-waves contained in said subsegment for calculating at least one feature of said subsegment; providing at least one feature of the subsegment and at least one feature of the segment as inputs to a machine learning algorithm, the machine learning algorithm being configured to output a score vector; at least one processor configured for; - classifying an episode using the score vector obtained for each subsegment within the episode to distinguish true positive episodes from false positive episodes; - at least one output adapted to provide said true positive episodes; Includes.

[0015] According to one embodiment, for each segment within an episode, the at least one feature of the segment is at least one of the following: morphological feature and / or rhythmic feature.

[0016] According to one embodiment, for each sub-segment, the computed features of the sub-segment are at least one of the following: rhythmic features, variance features and / or spectral features.

[0017] According to one embodiment, morphological features are statistics calculated based on the shape of the ECG signal, and rhythm features are statistics calculated based on the duration between R waves. These statistics may be calculated directly for specific features (e.g., minimum, maximum, median, standard deviation, or more complex features). However, statistics in this specification also refer to some categorical features (e.g., type of p-wave, which may be positive, negative, or unknown) computed from past statistics.

[0018] According to one embodiment, the processor is further configured to input each subsegment of the episode into a neural network to extract at least one neural network feature of the subsegment as an output of the neural network, said neural network feature being one of the subsegment features provided as input to a machine learning algorithm. In this embodiment, for each subsegment, the computed feature of the subsegment is at least one of the following: rhythmic features, variability features, neural network features and / or spectral features.

[0019] According to one embodiment, the neural network is a convolutional neural network.

[0020] According to one embodiment, the neural network is at least one of the following: a convolutional neural network, a deep belief neural network, and a recurrent neural network, or a combination thereof. The use of a convolutional neural network can capture interesting patterns in different periods of a signal, while a recurrent neural network can capture information about the overall progression of a signal, and a deep belief network is advantageous to train on an unlabeled data training set.

[0021] According to one embodiment, in addition to or instead of the neural network, the processor is further configured to compute at least one feature of each sub-segment of an episode by inputting the sub-segment into a transformer and extract at least one transformer feature as an output of the transformer. The transformer is a deep learning model that employs an attention mechanism to weight the influence of different parts of the input data. Advantageously, the transformer is able to capture long-range interactions between signals and to associate sub-segments of one episode that are far apart.

[0022] According to one embodiment, the at least one algorithm for identifying R waves is selected from the following list: an XQRS detection algorithm, a stationary wavelet transform process and / or an optimized knowledge base (OKB) detection algorithm.

[0023] According to one embodiment, the processor is configured to utilize a combination algorithm configured to identify R-waves in a segment using at least two algorithms and combine the R-waves obtained from the at least two algorithms to obtain a corrected R-wave.

[0024] According to one embodiment, R-waves in an episode are identified using at least two algorithms, and at least one rhythmic feature of the segments and subsegments is calculated using the R-waves obtained from each of the at least two algorithms. Of note, at least one rhythmic feature of the segments and subsegments is calculated using the corrected R-waves obtained from the combined algorithm. This embodiment can advantageously reduce errors in estimating the location of the R-waves, and therefore obtain more accurate assessments of the segment and subsegment features.

[0025] According to one embodiment, a machine learning algorithm is trained on a dataset comprising a plurality of annotated episodes, the dataset comprising representative episodes of abnormal heart rhythms.

[0026] According to one embodiment, the dataset of annotated episodes includes episodes associated with asystole, bradycardia, atrial fibrillation, atrial tachycardia, ventricular tachycardia, and other abnormalities and / or artifacts.

[0027] According to one embodiment, the machine learning algorithm is the XGBoost algorithm.

[0028] According to one embodiment, the input is further configured to receive episodes from multiple cardiac devices from multiple manufacturers.

[0029] According to one embodiment, the processor is configured to normalize the episodes received from the multiple cardiac devices.

[0030] According to one embodiment, the system is included in a remote monitoring platform.

[0031] The invention further relates to a non-transitory computer-readable storage medium comprising instructions which, when the program is executed by a computer, cause the computer to carry out the steps of the method according to any one of the previous embodiments herein.

[0032] The present invention further relates to a computer program comprising instructions that, when said program is executed by a computer, cause said computer to carry out a method for analyzing electrocardiographic episodes previously obtained from a cardiac connection device according to any one of the methods according to any one of the previous embodiments herein.

[0033] The present invention relates to a computer-implemented method for identifying abnormal heart rhythms, the method comprising: - receiving an electrocardiogram (ECG) signal from a cardiac device, the ECG signal including a series of R waves; - Identifying R waves in the ECG signal; - Calculating morphological features of the ECG signal; - calculating rhythm features of the ECG signal; - segmenting the ECG signal into a series of overlapping segments; - Calculating the variation features of each segment; - Computing rhythmic features of each segment; - Computing the spectral features of each segment; - training a machine learning algorithm on a dataset of annotated ECG segments that contain ECG segments that are indicative of abnormal heart rhythms; - classifying each segment by a machine algorithm based on the variability features of the segment, the rhythm features of the segment, the spectral features of the segment, the morphological features of the ECG signal, and the rhythm features of the ECG signal; and - classifying the ECG signal based on the classification of the segments; Includes.

[0034] According to one embodiment, the morphological features of the ECG signal are statistics calculated based on the shape of the ECG signal.

[0035] According to one embodiment, the rhythmic features are statistics calculated based on the periods between waves.

[0036] According to one embodiment, R-waves in the ECG signal are identified using a plurality of methods, rhythm features of the ECG signal are calculated for each of the plurality of methods, and rhythm features of each segment are identified for each of the plurality of methods.

[0037] According to one embodiment, the methods include an XQRS detection method, a stationary wavelet transform process, or an optimized knowledge-based (OKB) detection method.

[0038] According to one embodiment, the dataset of annotated ECG segments includes ECG segments indicative of asystole, bradycardia, atrial fibrillation or tachycardia, ventricular tachycardia or fibrillation, or artifact.

[0039] According to one embodiment, the machine learning algorithm is the XGBoost algorithm.

[0040] According to one embodiment, the method further includes receiving ECG signals from multiple cardiac devices from multiple manufacturers.

[0041] According to one embodiment, the method further includes normalizing the ECG signals received from the multiple cardiac devices.

[0042] According to one embodiment, the method further includes providing a platform for observing the received ECG signal and classifying the received ECG signal.

[0043] definition In the present invention, the following terms have the following meanings:

[0044] - "episode" refers to a portion of an electrocardiogram signal having a finite duration identified and recorded by the manufacturer of the cardiac device used to measure the electrocardiogram signal itself. In practice, the manufacturer may implement an identification method configured to perform a preliminary analysis on the measured electrocardiogram signal in order to identify, among the measured signals, a portion of the electrocardiogram signal related to an abnormal cardiac rhythm and then record it. If the patient activates the recording, the portion of the electrocardiogram signal may also be recorded. In addition to the portion of the electrocardiogram signal, the episode further includes the date (time and date) of the recording and the type of recording (i.e., the patient or the identification method or the episode that was activated). The episodes recorded from a cardiac device essentially depend on the cardiac device itself (i.e., the quality of the acquired signal) and the identification method of the manufacturer (i.e., the accuracy of the distinction) and may therefore differ between devices of one manufacturer and between different devices of different manufacturers.

[0045] - "Cardiac (connected) device" refers to a device configured to measure an electrocardiogram signal, perform at least one preliminary analysis of the ECG signal, and transmit said episodes to an external receiver in order to detect an episode by an identification method. Said device may be, for example, an implantable loop recorder, a portable electrocardiogram telemetry, an insertable cardiac monitor, a pacemaker, an implantable cardioverter defibrillator (ICD), a cardiac resynchronization therapy (CRT) device, and the like.

[0046] - "Abnormal heart rhythm" refers to any physiological abnormality that may be identifiable in a cardiac signal. For example, in the present invention, the following abnormalities may be identified, but are not limited to: "sinoatrial block, paralysis or arrest", "atrial fibrillation", "atrial fibrillation or flutter", "atrial flutter", "atrial tachycardia", "junctional tachycardia", "supraventricular tachycardia", "sinus tachycardia", "ventricular tachycardia", "pacemaker", "ventricular extrasystole", "atrial extrasystole", "first degree atrioventricular block (AVB)", "second degree atrioventricular block (Mobitz type I)", "second degree atrioventricular block (Mobitz type II)", "third degree atrioventricular block", "Wolf-Parkinson-White syndrome", "left bundle branch block", "right bundle branch block", "intraventricular conduction delay", "left ventricular hypertrophy", "right ventricular hypertrophy", "acute myocardial infarction", "old myocardial infarction", "ischemia", "hyperkalemia", "hypokalemia", "Brugada", "QTc prolongation", and others.

[0047] - The term "processor" should not be construed as being limited to hardware capable of executing software, but refers in a general way to a processing device, which may include, for example, a computer, a microprocessor, an integrated circuit, or a programmable logic device (PLD). The processor may encompass one or more graphics processing units (GPUs), whether utilized for computer graphics and image processing or other functions. In addition, instructions and / or data capable of implementing the relevant functionality and / or resulting functionality may be stored on any processor-readable medium, such as, for example, an integrated circuit, a hard disk, an optical disk such as a CD (compact disk), a DVD (digital versatile disk), a RAM (random access memory), or a ROM (read only memory). The instructions may be stored in hardware, software, firmware, or any combination thereof, among others.

[0048] - "Machine Learning Algorithm (ML)" refers to a computer algorithm that, in a traditional manner, improves automatically through experience based on training data that is capable of adjusting the parameters of a computer model through the reduction of the gap between the output of a prediction extracted from the training data and the output of an evaluation computed by the computer model.

[0049] - "Dataset" refers to a collection of data used to build an ML mathematical model to make a data-driven prediction or decision. In supervised learning (i.e., inference functions from known input-output examples in the form of labeled training data), three types of ML databases (also called ML sets) are typically dedicated to three different tasks: training, i.e., fitting parameters, validation, i.e., tuning ML hyperparameters (which are parameters used to control the learning process), and validation, i.e., checking independently that the newer model provides satisfactory results with the training dataset that was used to build the mathematical model.

[0050] - "Neural network or artificial neural network (ANN)" refers to a category of ML that contains nodes (called neurons) and connections between the neurons modeled by weights. For each neuron, the output is a function of the input or set of inputs by an activation function. Neurons are generally organized into layers, such that neurons in one layer connect only to neurons in the previous and next layers.

[0051] "QRS" or "QRS complex" refers to a deflection in an electrocardiogram tracing that represents ventricular activity of the heart. The QRS complex generally includes a Q wave, an R wave, and an S wave, occurring in rapid succession.

[0052] - "False positive" refers to an error in bivariate classification where the test result erroneously indicates the presence of disease, such as an abnormal rhythm, within an episode when no abnormal rhythm is present, and "false negative" is the opposite error where the test result erroneously fails to indicate the presence of disease when an abnormal rhythm is present. These are two types of errors in bivariate testing, in contrast to the two correct results, "true positive" and "true negative."

[0053] - "Remote monitoring platform" refers to any system for data management configured to receive and store and / or analyze data received from at least one cardiac device corresponding to at least one patient. In one example, the platform may generally receive data from one or more cardiac devices of a patient to manage the care of those patients, including receiving data, reports and information from one or more such devices to enable a healthcare provider to monitor, document and report on the health status of the patient. Such data may be received at the remote monitoring platform through many sources, including the corresponding device, device programming equipment, and reported from the patient or manufacturer, or from a third party that receives the data from the device. In this example, the remote monitoring platform may provide one or more interfaces to allow healthcare providers or other users of the platform to manage the receipt of device data.

[0054] Detailed Description The following detailed description will be better understood when read in conjunction with the drawings. For illustrative purposes, a computer-implemented method and system for analyzing electrocardiogram episodes is shown in a preferred embodiment. However, it should be understood that the present invention is not limited to the precise configuration, structure, features, embodiments, and aspects shown. The drawings are not drawn to scale and do not limit the scope of the claims to the depicted embodiments. Therefore, when features recited in the appended claims follow reference signs, it should be understood that such signs are included only to enhance the comprehension of the claims, and do not limit the scope of the claims.

[0055] Features and advantages of the invention will become apparent from the following description of embodiments of the system, which description is given solely by way of example and with reference to the accompanying drawings, in which: [Brief description of the drawings]

[0056] [Figure 1] 2 is a flow chart illustrating the main steps of the method of the present invention according to one embodiment. [Diagram 2] 2 is a flow chart illustrating the main steps of the method of the present invention according to one embodiment. [Diagram 3] 2 is a flow chart illustrating the successive steps performed in a system for analyzing electrocardiographic episodes previously obtained from the cardiac device of FIG. 1. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0057] Although various embodiments have been described and illustrated, the detailed description should not be construed as limiting thereto. Various modifications may be made to the embodiments by those skilled in the art without departing from the true scope of the disclosure as defined in the claims.

[0058] To overcome these and other shortcomings of conventional cardiac monitoring platforms, a device diagnostic platform is provided that receives and normalizes ECG signals from any cardiac device (e.g., implantable loop recorders, pacemakers, defibrillators, etc.) produced by any manufacturer, thus providing medical professionals with a single platform for monitoring all of their patients with cardiac implants.

[0059] The portion of the electrocardiogram signal that is an episode may be a segment of a few seconds to 15 minutes if the anomaly observed in the heart rhythm is of short duration (e.g., the criteria for detecting the anomaly is valid at one single point). However, some of the observed anomalies may have a relatively long duration and not present any significant variation between their occurrences. In these cases, a cardiac device identifying one of these abnormal heart rhythms may be configured to record as an episode a first segment of the ECG that corresponds to the onset of the abnormal heart rhythm and a second segment of the ECG that spans the end of the abnormal heart rhythm. Recording the entire episode would consume too much memory space. According to the manufacturing design, the cardiac device may also record more than two segments in one episode.

[0060] The disclosed computer-implemented method and system also includes a machine learning algorithm that reviews episodes received from the cardiac device to identify true abnormal and true normal cardiac rhythms, thereby eliminating the need for a physician to manually review false positive alerts output by the cardiac device. The machine learning algorithm utilizes a rule-based process to review a significant number of episodes received from the cardiac device that previously required subjective review by a physician. An exemplary method of identifying abnormal cardiac rhythms to distinguish true positive episodes containing abnormal cardiac rhythms from false positive episodes containing normal cardiac rhythms is described below.

[0061] A recorded episode is received from a cardiac device, for example an insertable or implantable cardiac monitor. The recorded episode may have been previously transmitted from the cardiac device to a receiving device configured to store the episode in a medical database. The method may thus receive episodes stored in the medical database.

[0062] An ECG segment (referred to herein as an "episode") received from a cardiac device is typically between approximately 9 seconds and approximately 5 minutes in length.

[0063] As shown in Fig. 1, the method 100 includes a first step 101 in which R-waves are identified in each segment of an episode. According to one embodiment, the R-waves in the segments are identified using at least one algorithm. The at least one algorithm for identifying R-waves may be selected from the following list: XQRS detection algorithm, a stationary wavelet transform process and / or an optimized knowledge base (OKB) detection algorithm.

[0064] In the XQRS detection algorithm, the ECG segment is bandpass filtered between 5 and 20 Hz to obtain a filtered ECG segment. A moving wave integration (MWI) with a Ricker wavelet is applied to the filtered ECG segment to preserve the square of the integrated signal. A calibration is performed to initialize the running parameters of noise and QRS amplitude, the QRS detection threshold, and the most recent RR interval. If calibration is not possible, default parameters are used. At each local maximum in the MWI signal, the XQRS detection method determines if the local maximum is a QRS complex. To be classified as a QRS, the QRS must come after a refractory period and cross the QRS detection threshold, and if it is close enough to the previous QRS, it must not be classified as a T wave. If successfully classified, the current detection threshold and heart rate parameters are updated. If not classified as a QRS, the local maximum is classified as a noise peak and the running parameters are updated. The local maximum of the QRS complex corresponds to the location of the peak of the R wave (i.e., the R peak) of the QRS complex. For each new QRS detected, a list of RR intervals is computed to calculate the time difference between each successive R peak. When a new maximum is reached, XQRS computes the duration between this maximum and the last identified R peak. If this duration is less than 1.66 times the most recent RR interval (calculated as the duration between the last identified R peak and the previous one), a classification of this maximum as either QRS or not is performed. If no QRS is detected within 1.66 times the most recent RR interval. If not, a back search QRS detection is performed on previous peaks using a lower QRS detection threshold before classifying the maximum.

[0065] In the stationary wavelet transform process, the two-dimensional wavelet transform of the ECG segment is computed, a threshold is applied to the level 2 detailed coefficients of the wavelet transform based on the mean and standard deviation, and the process is repeated for the threshold coefficients in fixed length windows with no overlap. One QRS complex is located at the maximum coefficient of each window with a fixed non-zero threshold coefficient, and each QRS complex that is within the maximum allowed RR interval range is merged and the location of each QRS complex is corrected to place it at a local maximum of the ECG segment before computing the two-dimensional wavelet transform. As with the previous algorithms, knowing the correct location of the QRS complex allows the location of the corresponding R peak to be determined.

[0066] In the OKB detection method, an ECG segment is bandpass filtered using a 3D Butter bandpass filter, the filtered signal is squared, the QRS and beat moving averages of the squared signal are computed (in QRS length and beat length windows respectively), corresponding blocks are created using the beat moving average and binarized by thresholding the QRS moving average, and for each corresponding block, the QRS complex is placed at the local maximum of the original ECG segment of said block. As with the previous algorithms, knowing the exact location of the QRS complex allows the location of the corresponding R peak to be determined.

[0067] The R peaks obtained from each of the three algorithms may be used to calculate the RR interval within each ECG segment.

[0068] According to one embodiment, R waves within an episode are identified using at least two of the algorithms listed above.

[0069] Alternatively, R waves may be identified using a novel combination algorithm developed by the inventors that employs an XQRS detection algorithm, a stationary wavelet transform process, and an optimized knowledge-based (OKB) detection algorithm.

[0070] In this embodiment, the R-peaks identified by the previous three methods are repositioned at the extremes. R-peaks detected by at least two of the three algorithms are used in this embodiment. A list of RR interval extensions that are at least x times longer than the RR interval median is created (where x is an empirically determined float). The location of a hypothetical auxiliary R-peak is inferred for each of those RR interval extensions in the list (by filling them with the RR distance median from past R-peaks). The ratio between the average gradient (respectively amplitude value) in these hypothetical R-peak zones and the average gradient (respectively amplitude value) of the surrounding area is computed for each new hypothetical R-peak. R-peaks with a ratio greater than an empirically defined threshold are retained. These last four steps are repeated with an adaptive interval median (based on the moving median of the RR interval). At each iteration, the current result is recombined with the past result. Then, the RR interval median is computed based on the location of the new R-peak (old estimate + new estimate) and the process is repeated. This iteration improves the detection of RR peaks when R peaks containing multiple gaps are lost by previous methods. The iteration is performed until it stops adding new R peaks. Advantageously, this embodiment allows the R peaks to be relocated more accurately, improving the discrimination ability of the method of the present invention.

[0071] The location of the R-wave peak may be used to calculate at least one feature of each segment of an episode (step 102 of the method).

[0072] According to one embodiment, at least one morphological feature and / or at least one rhythmic feature is calculated for each segment of each episode. The morphological feature may be a statistic calculated based on the shape of the ECG signal. The rhythmic feature may be a statistic calculated based on the period between waves (e.g. the period between each R peak). Clustering is an example of a rhythmic feature.

[0073] The morphological features may include QRS features and P-wave features. To determine the QRS features, QRS rhythms are extracted at each R-peak (based on two predefined delays, i.e., the delay before and the delay after the R-peak), the QRS rhythm median is computed, each QRS rhythm is assigned a mean and maximum distance to the QRS median, QRS rhythms with mean and maximum distances smaller than predefined thresholds are selected, the onset and end of the QRS peak are defined for each of those representative QRS rhythms (based on extreme value analysis), and the following features are defined: median QRS peak width, median QR delay, median RS delay, median QR delay, standard deviation of QS delay, and maximum QS delay.

[0074] To determine P-wave features, the R peak that defines the most representative QRS complex is identified, a rhythm that is approximately a PR rhythm is extracted at each R peak (based on the two delays from the R peak), the signal medians of all PR rhythms are computed, the P wave (assumed to be the peak with the greatest amplitude) is placed at the signal median, the prominence and area of ​​that wave are identified, and the average distance of all PR rhythms from the signal median is identified. The P wave (assumed to be the peak with the greatest amplitude) is also placed at all of the PR rhythms, and the standard deviation of the PR delays is identified.

[0075] Rhythm features may be calculated for each segment using the R peaks detected by each of the four rhythm extraction algorithms described above (XQRS detection algorithm, stationary wavelet transform process, OKB detection algorithm, and a combination algorithm developed by the inventors). The following features may be calculated for each group of R peaks detected in each segment for each algorithm: mean, median, minimum, maximum, and standard deviation of R-R interval duration; mean, median, and standard deviation of absolute variation of R-R interval duration; and sample entropy of R-R interval duration using vectors of length 2 and Chebyshev distance.

[0076] In each of the four rhythm extraction algorithms, the R peak location array may be converted into a three-dimensional vector. The first dimension is the R-R interval from the first R peak to the second to last R peak, the second dimension is the R-R interval from the second R peak to the last R peak, and the third dimension is from the first peak position to the third to last R peak position, adjusted by a normalization factor. These three-dimensional vectors are grouped into clusters by a clustering algorithm such as the DBScan algorithm. In the clustering, different algorithms may be used, such as DBScan, K-means, MeanShift, Spectral Clustering, Birch or Ward. The clusters are grouped into regular clusters where the first two dimensions are close (i.e., the R-R intervals at times n and n-1 are close) and irregular clusters where they are far apart. Based on the RR intervals and the clustering, several statistics are calculated, including the number of clusters identified, the clustering score, the percentage of unclassified rhythms (not included in any cluster) among all rhythms, the percentage of rhythms in regular clusters among all rhythms, the standard deviation of rhythm fluctuations in regular clusters, the mean and standard deviation of the difference between the first two dimensions of rhythms in regular clusters, the ratio between the mean period of the fastest cluster and the mean period of the slowest cluster, and the overlap time between these clusters.

[0077] In one embodiment, at least one Transformer feature is computed for each subsegment of each episode in addition to at least one rhythmic, variance and / or spectral feature, and each subsegment of the episode may be provided as an input to a Transformer to extract at least one Transformer feature as an output of the Transformer.

[0078] In one embodiment, at least one neural network feature is calculated for each subsegment of each episode in addition to at least one rhythmic, variance and / or spectral feature. Each subsegment of the episode may be provided as an input to a neural network in order to extract at least one neural network feature as an output of the neural network. Advantageously, the neural network is used to identify signal patterns that are not documented in the literature.

[0079] In one advantageous embodiment, the neural network is trained together with XGBoost using unannotated ECG data to exploit patterns identifiable in a larger ECG dataset to improve performance.

[0080] In one embodiment, the neural network is an architecture that combines at least two of a convolutional neural network, a deep belief neural network, a recurrent neural network, or a cited neural network.

[0081] The neural network may be a convolutional neural network. A convolutional neural network is a type of neural network that exploits continuity in ECG data. The convolutional neural network may be trained and validated on a subset of the set trained with XGBoost. In one advantageous embodiment, the convolutional neural network is jointly trained with XGBoost using unannotated ECG data to exploit patterns identifiable in a larger ECG dataset to improve performance.

[0082] In one embodiment, the processor is also configured to input each segment of the episode into the neural network and extract at least one neural network feature of the segment as an output of the neural network, in this embodiment, for each segment, the computed feature of the segment is at least one of the following: rhythmic features, morphological features and / or neural network features.

[0083] Each segment in each episode may be segmented into at least two subsegments of equal duration. In this step 103, the method utilizes a sliding window to identify overlapping subsegments of fixed duration. For example, each segment may be segmented into 10 second subsegments starting 1 second apart.

[0084] The method may further include the step 104 of using the R waves identified in each sub-segment to calculate at least one feature of said sub-segment.

[0085] Among others, variability features, rhythm features, neural network features and spectral features may be calculated for each sub-segment. Variability features may be quantiles of the rolling deviation of the signal, which may be computed in windows of different durations. Rhythmic features that may be calculated for each of the four RR interval arrays described above may include the mean, median, minimum, maximum and standard deviation of the RR interval duration; the mean, median and standard deviation of the absolute variation of the RR interval duration; and the sample entropy of the RR interval duration (using length vector2 and Chebyshev distance). Spectral features may be spectral characteristics of the signal based on the Fast Fourier Transform (FFT) of the signal filtered by a bandpass filter. The spectral features may include the fundamental frequency of the signal, the value of the FFT at its fundamental frequency, and the power ratio of the fundamental frequency, harmonics and the total FFT (calculated over multiple harmonics and frequency bandwidths).

[0086] Each episode may be analyzed based on the subsegment features (variation features, rhythm features, neural network features and spectral features) and the segment features (morphological features and rhythm features) from which the subsegments were extracted.

[0087] The systems and methods described herein may be configured to classify each episode as an indication of asystole, bradycardia, atrial fibrillation or atrial tachycardia (AT / AF), ventricular tachycardia (VT), artifact, or normal heart rhythm. Asystole (or primary arrest) is the absence of any ventricular contraction for a minimum period (e.g., a minimum period corresponding to a configurable asystole interval). Bradycardia is a slow ventricular rate (e.g., a ventricular rate less than a configurable bradycardia rate with a minimum period of 4 beats). Ventricular tachycardia may be at least one of the following: a tachycardia originating within the ventricle, or a non-sustained ventricular tachycardia. Atrial tachycardia / atrial fibrillation (AT / AF) is at least one of the following: atrial tachycardia (ectopic), atrial flutter, or atrial fibrillation. An artifact is the presence of non-cardiac noise. In the absence of any of the five aforementioned abnormal heart rhythms, the system classifies the episode as a normal heart rhythm.

[0088] According to one embodiment, the method comprises a step 105, for each sub-segment, providing the sub-segment features and the segment features to which the sub-segment belongs as input to a machine learning algorithm, the machine learning algorithm being configured to output a score vector. In one example, the features of each sub-segment and the features of the corresponding segment may be concatenated into a vector of flattened features that are input to the machine learning algorithm.

[0089] The method may include a step 106 of obtaining as output a score vector for each of the sub-segments segmented into the segment(s) of the episode.

[0090] The machine learning algorithm may be trained on a dataset that includes a plurality of annotated episodes, the dataset including representative episodes of abnormal and normal cardiac rhythms. The annotated episodes of the dataset allow for supervised training of the machine learning architecture. The dataset may also include unannotated episodes that are used in other types of training, such as unsupervised or semi-supervised training approaches.

[0091] According to one embodiment, the machine learning algorithm includes a chain of at least two machine learning algorithms.

[0092] In a further step 107, the method may utilize the score vector obtained for each sub-segment in an episode to classify the episode in order to distinguish true positive episodes from false positive episodes.

[0093] In one exemplary embodiment, to classify each episode, the method utilizes a classifier chain of five machine learning algorithms (e.g., the XGBoost algorithm) as the machine learning algorithm, each of which identifies whether the segment is an indicator of one of the five aforementioned abnormal cardiac rhythms. More specifically, a first XGBoost instance may be trained with a dataset of annotated samples to identify samples as being or not indicative of asystole; a second XGBoost instance may be trained with a dataset of annotated samples to identify samples as being or not indicative of bradycardia; a third XGBoost instance may be trained with a dataset of annotated samples to identify samples as being or not indicative of atrial fibrillation or atrial tachycardia; a fourth XGBoost instance may be trained with a dataset of annotated samples to identify samples as being or not indicative of ventricular tachycardia; and a fifth XGBoost instance may be trained with a dataset of annotated samples to identify samples as being or not indicative of artifact.

[0094] If the five machine learning algorithms do not identify any of the aforementioned abnormal cardiac rhythms in the score vector, the subsegment is classified as having a normal cardiac rhythm.

[0095] Each episode is classified based on the classification of sub-segments within the segment(s) of the episode. For example, if an episode includes a sub-segment classified as artifact and another sub-segment classified as atrial fibrillation or atrial tachycardia, the episode is classified as artifact and atrial fibrillation or atrial tachycardia. In this case, the episode is a true positive episode that should be reviewed by the healthcare provider. If all of the sub-segments within the episode are classified as normal heart rhythm, the episode is classified as normal heart rhythm. These types of events are cardiac device false positives that are normal events but are erroneously labeled as abnormal, creating excessive undesirable information recorded and transmitted by the cardiac device, so that the healthcare provider does not need to review them. In fact, relevant information of the patient's clinical status may not be obtained by these erroneously labeled normal events.

[0096] In a second exemplary embodiment, the machine learning algorithm is a classifier chain of six machine learning algorithms trained to classify episodes into at least one of six classes or into none of these six classes. Different decision tree based algorithms may be used in the chain, such as XGBoost, LightGBM, AdaBoost or Random Forest. In one embodiment, the machine learning algorithm is XGBoost, which advantageously offers the best compromise between the best possible performance and the time of training required to obtain the parameters that reach these best possible performances.

[0097] In one example, the classifier chain includes six XGBoosts. Each of the machine learning algorithms in the chain is trained with a dataset of annotated episodes to identify episodes as at least one of the following classifications: asystole, bradycardia, atrial fibrillation or atrial tachycardia (AT / AF), ventricular tachycardia or ventricular fibrillation (VT / VF), artifact, and indicators of normal or abnormal heart rhythm. Each of the machine learning algorithms in the chain is trained as a classifier chain, such that the output of each algorithm is part of the input of every next algorithm. This advantageously improves the classification efficiency of the chain. The classification performed by the chain on each subsegment provides as output a six-dimensional score vector. After classification, the score vector may be populated with "1" or "0", where "1" corresponds to the attribution of a unique label by the corresponding machine learning algorithm of the chain and "0" corresponds to the absence of one unique label. The five coefficients associated with abnormal heart rhythm classifications (i.e., all classifications other than normal or abnormal heart rhythm classifications) in each score vector obtained for each subsegment may then be combined at the episode level using logical OR with each of the score vectors obtained for an episode. The coefficients associated with the classification "normal or abnormal heart rhythm" in each score vector obtained for each subsegment may then be combined at the episode level with logical AND. This example provides an episode score vector of six coefficients including Booleans, where five coefficients (e.g., the first five) are set to "1" whenever at least one of the subsegments in the episode is classified into one of the abnormal heart rhythm classifications, and one coefficient (e.g., the last one) is set to "1" if all of the subsegments in the episode are labeled as "normal heart rhythm". This episode score vector is finally converted into an output configured to identify the episode as a normal heart rhythm, i.e., a false positive, or an abnormal heart rhythm, i.e., a true positive.Further, in addition to information that the episode is a true positive episode, the output may also include an indication of at least one abnormal cardiac rhythm classification with which the subsegment of the episode is associated.

[0098] In one embodiment, the method receives as input episodes from multiple cardiac devices from multiple manufacturers. Advantageously, the method may include normalizing the episodes received from the multiple cardiac devices, thereby removing the mean and variance of the input signals to make them similar to other signals.

[0099] The platform may be configured to provide a medical professional with the ability to view the ECG episodes, the decisions made by the cardiac devices, and the decisions made by the methods / systems of the present invention. If the episode is classified as normal heart rhythm or artifact, the platform may be configured to refrain from outputting an alert to the medical professional. By classifying ECG episodes as described above, the disclosed system reduces the number of false positive alerts that must be reviewed by a physician.

[0100] The embodiments disclosed herein include various operations described herein. As discussed above, the operations may be performed by hardware components and / or embodied in machine-executable instructions and used to cause a general-purpose or special-purpose processor programmed with the instructions to perform the operations. Alternatively, the operations may be performed by a combination of hardware, software, and / or firmware.

[0101] Performance of one or more operations described herein may be distributed among one or more processors and deployed across multiple machines as well as being present on one machine. In some embodiments, one or more processors or processor-implemented modules may be located in a single geographic location (e.g., in a residential environment, a work environment, or a server farm). In other embodiments, one or more processors or processor-implemented modules may be distributed across multiple geographic locations.

[0102] As shown in FIG. 3, the present invention also relates to a system 1 for analyzing electrocardiogram episodes previously obtained from a cardiac device to distinguish true positive episodes involving abnormal cardiac rhythms from false positive episodes involving normal cardiac rhythms, the system including at least one processor and all necessary circuitry and / or storage media to implement the methods described herein above.

[0103] The system may be implemented by a server (i.e., a remote monitoring platform) that receives data (i.e., episodes, etc.) from at least one cardiac device corresponding to at least one patient. Communication of the data to the remote monitoring platform may be performed over a communication network, such as the Internet.

[0104] The present invention further relates to a computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to perform the steps of a method for analyzing electrocardiographic episodes previously obtained from a cardiac connection device as described herein above.

[0105] The computer program implementing the method of the embodiment of the present invention can be generally distributed to users by distributed computer readable storage media, including but not limited to SD cards, external storage devices, microchips, flash memory devices, portable hard drives and software websites. The computer program can be copied from the distributed media to a hard disk or similar intermediate storage medium. The computer program can be executed by loading computer instructions from either the distributed media or these intermediate storage media into the computer's execution memory to configure the computer to operate according to the method of the present invention. All these operations are well known to those skilled in the art of computer systems.

[0106] The instructions or software that control the processor or computer implementing the hardware components and perform the methods described above, as well as any associated data, data files, and data structures, are recorded, stored, or fixed on one or more non-transitory computer-readable storage media. Examples of non-transitory computer-readable storage media include read-only memory (ROM), random-access memory (RAM), flash memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk, and any device known to those skilled in the art that can store the instructions or software and any associated data, data files, and data structures in a temporary manner and provide the instructions or software and any associated data, data files, and data structures to the processor or computer so that the processor or computer can execute the instructions. In one embodiment, the instructions or software, and any associated data, data files, and data structures, are distributed across network-coupled computer systems such that the instructions or software, and any associated data, data files, and data structures are stored, accessed, and executed by processors or computers in a distributed fashion.

[0107] The present invention further relates to a computer program product for analyzing electrocardiogram episodes previously obtained from a cardiac connection device, the computer program product comprising instructions that, when the program is executed by a computer, cause the computer to perform the steps of the method according to any one of the embodiments described hereinbefore.

[0108] A computer program product implementing the methods described above may be described as computer programs, code segments, instructions, or any combination thereof, for individually or collectively instructing or configuring a processor or computer to operate as a mechanical or special purpose computer to perform the operations performed by the hardware components. In one example, the computer program product includes machine code that is directly executed by the processor or computer, such as machine code generated by a compiler. In another example, the computer program product includes higher level code that is executed by the processor or computer using an interpreter. A programmer skilled in the art can readily write instructions or software based on the block diagrams and flow charts illustrated in the drawings and corresponding descriptions herein to disclose algorithms for performing the operations of the methods described above.

Claims

1. 1. A computer-implemented method (100) for analyzing electrocardiogram episodes previously obtained from a cardiac-connected device to assist in distinguishing true positive episodes involving abnormal cardiac rhythms from false positive episodes involving normal cardiac rhythms, the computer-implemented method being performed automatically by a computer directed by computer software, the computer-implemented method comprising: receiving said episodes, each episode comprising at least one segment of an electrocardiogram signal; - for each segment within an episode, - identifying R-waves in said segment using at least one algorithm (101) and using said R-waves to calculate at least one feature of said segment (102); Segmenting (103) said segment into at least two sub-segments; For each sub-segment, utilizing R-waves identified within said sub-segment to calculate at least one characteristic of said sub-segment (104); providing (105) at least one feature of said sub-segment and at least one feature of said segment as input to a machine learning algorithm and obtaining (106) a score vector as output, said machine learning algorithm being configured to output said score vector; classifying (107) an episode using the score vector obtained for each subsegment within said episode in order to distinguish true positive episodes from false positive episodes; - outputting a classification result including at least said true positive episodes; 4. A computer-implemented method comprising:

2. 2. The computer-implemented method of claim 1, wherein for each segment within an episode, the at least one computed feature of the segment is at least one of the following: morphological features and / or rhythmic features.

3. 3. A computer-implemented method according to claim 1, wherein for each subsegment, the computed features of the subsegment are at least one of the following: rhythmic features, variance features and / or spectral features.

4. The computer-implemented method of claim 1 , further comprising providing the true positive episodes to a remote monitoring platform.

5. 1. A system (1) for analyzing electrocardiographic episodes previously obtained from a cardiac device to distinguish true positive episodes involving abnormal cardiac rhythms from false positive episodes involving normal cardiac rhythms, the system comprising: at least one input adapted to receive said episodes, each episode comprising at least one segment of an electrocardiogram signal; at least one processor, For each segment in an episode, - identifying (101) R-waves in said segment using at least one algorithm and using (102) said R-waves to calculate at least one feature of said segment; - segmenting (103) said segment into at least two sub-segments; - for each sub-segment, - using (104) the identified R-waves contained in said sub-segment for calculating at least one characteristic of said sub-segment; providing (105) at least one feature of said subsegment and at least one feature of said segment as input to a machine learning algorithm, said machine learning algorithm being configured to output a score vector (106); classifying (107) an episode using the score vector obtained for each subsegment within said episode in order to distinguish true positive episodes from false positive episodes; At least one processor configured to: at least one output adapted to provide said true positive episodes; Including, the system.

6. The system of claim 5 , wherein for each segment in an episode, at least one feature of the segment is at least one of the following: morphological features and / or rhythmic features.

7. 7. The system of claim 6, wherein the morphological features are statistics calculated based on the shape of the ECG signal and the rhythm features are statistics calculated based on the period between R waves.

8. The system of claim 5 , wherein the processor is configured to segment the segment into at least two overlapping sub-segments.

9. The system of claim 5 , wherein for each subsegment, the computed features of the subsegment are at least one of the following: rhythmic features, variance features and spectral features.

10. 10. The system of claim 9, wherein the processor is further configured to compute at least one feature for each subsegment of the episode by inputting the subsegment into a transformer and extracting at least one transformer feature as an output of the transformer.

11. 10. The system of claim 9, wherein the processor is further configured to compute at least one feature for each subsegment of the episode by inputting the subsegment into a neural network and extracting at least one neural network feature as an output of the neural network.

12. 12. The system of claim 11, wherein the neural network is at least one of the following: a convolutional neural network, a deep belief neural network, and a recurrent neural network.

13. 13. The system of any one of claims 5 to 12, wherein at least one algorithm for identifying R-waves is selected from the following list: an XQRS detection algorithm, a stationary wavelet transform process and / or an optimized knowledge-based (OKB) detection algorithm.

14. 14. The system of claim 5, wherein the processor is further configured to utilize a combination algorithm configured to identify R-waves in the segment using at least two algorithms and to combine R-waves obtained from the at least two algorithms.

15. 15. The system of claim 6, wherein R-waves within the episode are identified using at least two algorithms, and at least one rhythm feature of the segments and / or subsegments is calculated using R-waves obtained from each of the at least two algorithms.

16. The system of claim 5 , wherein the machine learning algorithm is an XGBoost algorithm.

17. 17. The system of claim 5, wherein the machine learning algorithm is trained on a dataset comprising a plurality of annotated episodes, the dataset comprising representative episodes of abnormal cardiac rhythms.

18. 18. The system of claim 5, wherein the dataset of annotated episodes comprises episodes associated with asystole, bradycardia, atrial fibrillation, atrial tachycardia, ventricular tachycardia and / or artifacts.

19. 19. The system of claim 5, wherein the input is further configured to receive episodes from multiple cardiac devices from multiple manufacturers.

20. 20. The system of claim 19, wherein the processor is further configured to normalize episodes received from the multiple cardiac devices.

21. 20. A system according to any one of claims 5 to 19, comprised in a remote monitoring platform.

22. A non-transitory computer readable storage medium comprising instructions that, when executed by a computer, cause the computer to perform a method for analyzing electrocardiographic episodes previously obtained from a cardiac connection device according to any one of claims 1 to 4.

23. A computer program comprising instructions which, when executed by a computer, cause the computer to carry out a method for analysing electrocardiographic episodes previously obtained from a cardiac connection device according to any one of claims 1 to 4.

Images