System, device, and method for cardiac diagnosis and / or monitoring

Abstract

To perform cardiac diagnosis and / or arrhythmia monitoring.SOLUTION: An external heart monitoring device may include: a plurality of ECG electrodes to sense surface ECG activity; ECG processing circuitry to process the surface ECG activity to provide at least one ECG signal; a non-transitory computer-readable medium comprising a rhythm change classifier comprising at least one neural network; and at least one processor to receive the ECG signal(s), detect with the rhythm change classifier time data corresponding to a predetermined rhythm change in the ECG signal(s), determine based on the detected time data at least one ECG signal portion corresponding to the predetermined rhythm change, and transmit the at least one determined ECG signal portion to a remote computer system.SELECTED DRAWING: Figure 1A

Description

[Technical Field]

[0001] [CROSS-REFERENCE TO RELATED APPLICATIONS] This application claims the benefit of U.S. Provisional Patent Application No. 62 / 927,428, filed October 29, 2019, the disclosure of which is incorporated herein by reference in its entirety.

[0002] Embodiments of the present disclosure are directed to systems, devices, and methods for cardiac diagnosis and / or monitoring, and more particularly, for arrhythmia monitoring using a trained classifier that includes at least one neural network. [Background technology]

[0003] There are a wide variety of electronic and mechanical devices for monitoring and / or treating a patient's medical condition. In some instances, medical devices such as cardiac monitors or defibrillators may be surgically implanted or externally connected to a patient, depending on the underlying medical condition being monitored and / or treated. In some cases, physicians may use medical devices alone or in combination with drug therapy to treat conditions such as cardiac arrhythmias.

[0004] Such patients can include those with heart failure, e.g., congestive heart failure (CHF). CHF is a medical condition in which the heart's pumping ability is inadequate to meet the body's demands. Generally, many disease processes can reduce the heart's pumping ability, leading to congestive heart failure. Symptoms of congestive heart failure vary but can include fatigue, decreased exercise capacity, shortness of breath, and swelling (edema). The diagnosis of congestive heart failure is based on a personal medical history, a thorough physical examination, and selected laboratory tests.

[0005] Additionally or alternatively, patients may suffer from cardiac arrhythmias. Ventricular fibrillation is one of the most fatal cardiac arrhythmias, occurring when normal, regular electrical impulses are replaced by irregular, rapid impulses, causing the heart muscle to stop contracting normally and begin quivering. If normal cardiac contractions are not restored, normal blood flow ceases, potentially leading to organ damage or death within minutes. Because patients are unaware of the impending fibrillation, they often die before they can obtain necessary medical assistance. Other cardiac arrhythmias include an excessively slow heart rate, known as bradycardia, and an excessively fast heart rate, known as tachycardia. Cardiac arrest can occur when various cardiac arrhythmias, such as ventricular fibrillation, ventricular tachycardia, pulseless electrical activity (PEA), and asystole (the heart stops all electrical activity), prevent the heart from delivering sufficient blood flow to the brain and other vital organs to sustain life.

[0006] Cardiac arrest and other cardiac health issues are the leading causes of death worldwide. Various resuscitation efforts are performed to maintain the body's circulatory and respiratory systems during cardiac arrest and attempt to save the patient's life. The earlier these efforts are initiated, the higher the patient's chances of survival. Implantable cardioverter-defibrillators (ICDs) or external defibrillators (e.g., manual defibrillators or automated external defibrillators (AEDs)) have significantly improved the ability to treat these life-threatening conditions. These devices work by applying corrective electrical pulses directly to the patient's heart. Ventricular fibrillation or ventricular tachycardia can be treated with an implantable or external defibrillator, for example, by delivering a therapeutic shock to the heart to restore a normal rhythm. To treat conditions such as bradycardia, an implantable or external pacing device can pace the patient's heart until normal cardiac electrical activity returns. External pacemakers, defibrillators, and other medical monitors are designed for ambulatory and / or long-term use and have further improved the ability to timely detect and treat life-threatening medical conditions. Examples of external cardiac monitoring and / or treatment devices include cardiac monitors, the ZOLL LifeVest® wearable cardioverter-defibrillator available from ZOLL Medical Corporation, and the AED Plus, also available from ZOLL Medical Corporation.

[0007] Certain cardiac monitoring and / or therapy devices may communicate patient-related biometric data (e.g., ECG signal samples and / or ECG signal portions / strips) to a cloud (e.g., a remote server, a cardiac monitoring facility, and / or the like) for review (e.g., by a technician, a prescriber / treating physician, and / or the like). However, such communication may require a large bandwidth (e.g., all measured biometric data constantly streaming in real time, and / or the like), may include a large amount of biometric data not related to cardiac events (e.g., biometric data related to normal sinus rhythm (NSR) and / or the like), and may require a human reviewer to review the large amount of data without guidance as to which portions of it are related to cardiac events. Additionally or alternatively, it may be difficult to accurately determine whether the human reviewer correctly identified and / or annotated cardiac events in the biometric data. For example, technicians may have different levels of experience and expertise, and failing to identify a cardiac event and / or incorrectly identifying a cardiac event may result in harm to the patient (e.g., if needed treatment is not provided and / or unnecessary treatment is provided, respectively). Summary of some embodiments

[0008] Embodiments of the present disclosure include an arrhythmia monitoring system. In some embodiments, the arrhythmia monitoring system may include an external cardiac monitoring device for a patient. The external cardiac monitoring device may include a plurality of ECG electrodes that sense surface electrocardiogram (ECG) activity of the patient, an ECG processing circuit that processes the surface ECG activity of the patient to provide at least one ECG signal for the patient on at least one ECG channel, a non-transitory computer-readable medium containing a rhythm change classifier, and at least one processor operatively connected to the at least one ECG channel and the non-transitory computer-readable medium. In some embodiments, the rhythm change classifier may include at least one neural network trained based on a historical collection of multiple ECG signal portions with known rhythm change information. In some embodiments, the at least one processor may receive the at least one ECG signal received via the at least one ECG channel, detect time data corresponding to a predetermined rhythm change in the at least one ECG signal using the rhythm change classifier, the time data including at least one of a start time, a time interval, or any combination thereof, determine at least one ECG signal portion associated with the detected time data that corresponds to the predetermined rhythm change in the at least one ECG signal based on the detected time data, and transmit the at least one determined ECG signal portion to a remote computer system.

[0009] In some embodiments, the predetermined rhythm change may be associated with an arrhythmia.

[0010] In some embodiments, the non-transitory computer-readable medium may comprise at least one of a memory, a programmable circuit board, a field programmable gate array, an integrated circuit, or any combination thereof. In some embodiments, the at least one neural network may comprise at least one of a convolutional neural network, a recurrent neural network, an attention network, a fully connected neural network, or any combination thereof.

[0011] In some embodiments, the at least one neural network may include at least one convolutional neural network including multiple convolutional layers. Additionally or alternatively, the multiple convolutional layers may include at least seven and no more than ten convolutional layers. Additionally or alternatively, the at least one convolutional neural network may further include an input layer and an output layer.

[0012] In some embodiments, the at least one ECG signal may comprise a plurality of ECG signal samples. Additionally or alternatively, the input layer may comprise at least one node for each ECG signal sample of the plurality of ECG signal samples. In some embodiments, an output of the output layer may comprise an index of the time data corresponding to the rhythm change.

[0013] In some embodiments, the at least one ECG signal portion may include an ECG signal portion having a duration greater than or equal to 15 seconds and less than or equal to 120 seconds. For example, the ECG signal portion may have a duration greater than or equal to 15 seconds and less than or equal to 60 seconds.

[0014] In some embodiments, the at least one ECG signal may comprise a plurality of ECG signal samples. Additionally or alternatively, the plurality of ECG signal samples may be sampled at a rate greater than or equal to 100 Hz and less than or equal to 500 Hz.

[0015] In some embodiments, the plurality of ECG signal samples may be sampled at a rate greater than 100 Hz and less than 500 Hz.

[0016] In some embodiments, the at least one ECG channel may include multiple ECG channels. Additionally or alternatively, the at least one ECG signal may include at least one individual ECG signal associated with each individual ECG channel of the multiple ECG channels. In some embodiments, the multiple ECG channels may include a first ECG channel and a second ECG channel. Additionally or alternatively, the at least one ECG signal may include a first individual ECG signal associated with the first ECG channel and a second individual ECG signal associated with the second ECG channel. In some embodiments, the first individual ECG signal may be orthogonal to the second individual ECG signal.

[0017] In some embodiments, the at least one neural network may have multiple Siamese branches. Additionally or alternatively, each individual Siamese branch of the multiple Siamese branches may be associated with an individual ECG channel of the multiple ECG channels. In some embodiments, the at least one neural network may further have at least one additional layer connected to the multiple Siamese branches. In some embodiments, each Siamese branch of the multiple Siamese branches may have multiple convolutional layers. Additionally or alternatively, the dimensions of each of the multiple convolutional layers of each individual Siamese branch may be the same as the dimensions of each of the multiple convolutional layers of each other Siamese branch.

[0018] In some embodiments, the processor may further detect the predetermined rhythm change based on the at least one ECG signal using the rhythm change classifier.

[0019] In some embodiments, the device may further include at least one sensor and associated sensor circuitry that senses non-ECG biometric data of the patient. Additionally or alternatively, detecting the predetermined rhythm change may be further based on the non-ECG biometric data of the patient. In some embodiments, the at least one sensor may include at least one of an accelerometer, a heart sound detector, or a combination thereof. Additionally or alternatively, the non-ECG biometric data may include at least one of acceleration data, heart sound data, or any combination thereof.

[0020] In some embodiments, detecting the predetermined rhythm change may be further based on at least one baseline ECG signal portion of the patient. Additionally or alternatively, detecting the predetermined rhythm change may be further based on at least one reference vector of the patient.

[0021] In some embodiments, detecting the predetermined rhythm change may be further based on at least one calibration measurement of the patient. Additionally or alternatively, the at least one calibration measurement may be based on at least one second ECG signal from second surface ECG activity sensed by a second plurality of ECG electrodes. Additionally or alternatively, the second plurality of ECG electrodes may be independent of the plurality of ECG electrodes of the external cardiac monitoring device.

[0022] In some embodiments, detecting the predetermined rhythm change may be further based on at least one previous ECG signal portion.

[0023] In some embodiments, a gateway device may be included. Additionally or alternatively, transmitting the at least one determined ECG signal portion to the remote computer system may include transmitting the at least one determined ECG signal portion from the external cardiac monitoring device to the gateway device. Additionally or alternatively, the gateway device may receive the at least one determined ECG signal portion from the external cardiac monitoring device and transmit the at least one determined ECG signal portion to the remote server.

[0024] In some embodiments, the remote computer system may be in communication with the external cardiac monitoring device. Additionally or alternatively, the remote computer system may receive the at least one determined ECG signal portion from the external cardiac monitoring device and / or analyze the at least one determined ECG signal portion to classify an arrhythmia type for the rhythm change in the at least one ECG signal. In some embodiments, the arrhythmia type may include at least one of a heart rate change, atrial fibrillation, flutter, supraventricular tachycardia, ventricular tachycardia, pause, atrioventricular block, ventricular fibrillation, bigeminy, triphemism, ventricular ectopic beat, bradycardia, tachycardia, a morphological change in the at least one ECG signal, or any combination thereof.

[0025] In some embodiments, the remote computer system includes an arrhythmia type classifier, the arrhythmia type classifier including at least one second neural network trained based on a second historical collection of a second plurality of ECG signal portions having known arrhythmia type information. Additionally or alternatively, analyzing the at least one determined ECG signal portion may include detecting the arrhythmia type associated with the rhythm change based on the at least one determined ECG signal portion using the arrhythmia type classifier.

[0026] In some embodiments, the remote computer system may further transmit at least one message associated with the at least one determined ECG signal portion and an arrhythmia type associated with the rhythm change to a computing device associated with a technician.

[0027] In some embodiments, the remote computer system may further analyze the at least one determined ECG signal portion to identify rare arrhythmias for the rhythm changes in the at least one ECG signal.

[0028] In some embodiments, the processor may further use the rhythm change classifier to determine a confidence score associated with the predetermined rhythm change based on the at least one ECG signal.

[0029] In some embodiments, the processor may further transmit at least one second ECG signal portion of the at least one ECG signal to the remote computer system. Additionally or alternatively, the at least one second ECG signal portion may be independent of the detected time data corresponding to the predetermined rhythm change in the at least one ECG signal. In some embodiments, the processor may further randomly determine the at least one second ECG signal portion. In some embodiments, the processor may further use the rhythm change classifier to determine a first confidence score associated with the predetermined rhythm change based on the at least one ECG signal, the first confidence score being greater than a first threshold. Additionally or alternatively, the processor may further use the rhythm change classifier to detect second time data corresponding to a potential rhythm change in the at least one ECG signal. Additionally or alternatively, the processor may further determine, using the rhythm change classifier, a second confidence score associated with the potential rhythm change based on the at least one ECG signal, the second confidence score being less than the first threshold and greater than a second threshold. Additionally or alternatively, the processor may further determine, based on the detected second temporal data, a portion of the at least one second ECG signal associated with the detected second temporal data that corresponds to the potential rhythm change in the at least one ECG signal.

[0030] In some embodiments, the remote computer system may further send at least one message related to the at least one second ECG signal portion to a computing device associated with the technician and / or receive annotation data related to at least one annotation for the at least one second ECG signal portion from the computing device associated with the technician.

[0031] In some embodiments, the remote computer system may further transmit the at least one annotation for the at least one second ECG signal portion to the external cardiac monitoring device. Additionally or alternatively, the processor may further retrain the rhythm change classifier based on the at least one second ECG signal portion and the annotation data.

[0032] In some embodiments, the remote computer system may further add the at least one second ECG signal portion to the historical collection of the plurality of ECG signal portions. Additionally or alternatively, the known rhythm change information for the at least one second ECG signal portion may include at least a portion of the annotation data. Additionally or alternatively, the remote computer system may further train an updated rhythm change classifier based on the historical collection of the plurality of ECG signal portions with the known rhythm change information. The remote computer system may further transmit the updated rhythm change classifier to the external cardiac monitoring device. Additionally or alternatively, the processor may further replace the rhythm change classifier with the updated rhythm change classifier.

[0033] In some embodiments, the remote computer system may include an arrhythmia type classifier having at least one second neural network trained based on a second historical collection of a second plurality of ECG signal portions with known arrhythmia type information. Additionally or alternatively, the remote computer system may further add the at least one second ECG signal portion to the second historical collection of the second plurality of ECG signal portions. Additionally or alternatively, the known arrhythmia type information for the at least one second ECG signal portion may include at least a portion of the annotation data. Additionally or alternatively, the remote computer system may further retrain the arrhythmia type classifier based on the second historical collection of the second plurality of ECG signal portions with known arrhythmia type information.

[0034] In some embodiments, the at least one determined ECG signal portion may include a plurality of determined ECG signal portions. Additionally or alternatively, the remote computer system may be in communication with the external cardiac monitoring device. Additionally or alternatively, the remote computer system may receive the plurality of determined ECG signal portions from the external cardiac monitoring device and / or analyze each individual determined ECG signal portion of the plurality of determined ECG signal portions to classify an individual class for each individual determined ECG signal portion. Additionally or alternatively, the class for at least two individual determined ECG signal portions may include a first class. In some embodiments, the remote computer system may further transmit at least one message related to the at least two individual determined ECG signal portions and the first class to a computing device associated with a technician.

[0035] In some embodiments, the processor may further detect at least one of a peak number and a heart rate based on the at least one ECG signal using the rhythm change classifier. Additionally or alternatively, the processor may determine that the detected at least one of the peak number and the heart rate is greater than a first threshold value for the patient and less than a second threshold value for the patient. Additionally or alternatively, the second threshold value for the patient may be less than the first threshold value for the patient. Additionally or alternatively, the processor may further detect that the predetermined rhythm change based on the at least one of the peak number and the heart rate is greater than the first threshold value for the patient or the second threshold value for the patient.

[0036] In some embodiments, the historical collection of the plurality of ECG signal portions may include a second plurality of ECG signal portions of at least one second ECG signal based on second surface ECG activity sensed by a second plurality of ECG electrodes. Additionally or alternatively, the second plurality of ECG electrodes may be separate from the plurality of ECG electrodes of the external cardiac monitoring device.

[0037] In some embodiments, the historical collection of the plurality of ECG signal portions may include a first ECG signal portion associated with a first time and a second ECG signal portion associated with a second time after the first time. Additionally or alternatively, the rhythm change classifier may be trained by predicting a predicted ECG signal portion associated with the second time based on the first ECG signal portion using the rhythm change classifier, determining at least one error value based on the predicted ECG signal portion and the second ECG signal portion, and / or training the rhythm change classifier based on the at least one error value. In some embodiments, the at least one error value may include one of a prediction error or a control loss.

[0038] In some embodiments, the historical collection of the plurality of ECG signal portions may include a first ECG signal portion associated with a first time and a second ECG signal portion associated with a second time. Additionally or alternatively, the rhythm change classifier may be trained by predicting, with the rhythm change classifier, a predicted time associated with the second ECG signal portion based on the first ECG signal portion and the second ECG signal portion, determining at least one error value based on the predicted time and the second time, and / or training the rhythm change classifier based on the at least one error value.

[0039] Embodiments of the present disclosure may include an arrhythmia detection system. In some embodiments, the arrhythmia detection system may include a non-transitory computer-readable medium having an arrhythmia type classifier including at least one neural network trained based on a historical collection of multiple ECG signal portions with known arrhythmia type information and at least one processor operably connected to the non-transitory computer-readable medium. In some embodiments, the arrhythmia type classifier may include at least one neural network trained based on a historical collection of multiple ECG signal portions with known arrhythmia type information. Additionally or alternatively, the at least one processor may receive at least one ECG signal and annotation data associated with at least one annotation for each of the at least one ECG signal, detect arrhythmia types in the at least one ECG signal and time data associated with the detected arrhythmia type using an arrhythmia type classifier, determine at least one ECG signal portion associated with the detected arrhythmia type in the at least one ECG signal based on the time data, determine a likelihood score for the at least one annotation based on the detected arrhythmia type, generate at least one message based on the likelihood score for the at least one determined ECG signal portion and the at least one annotation, and / or transmit the at least one message associated with the at least one determined ECG signal portion. In some embodiments, the at least one message may indicate at least one of recommending annotating the at least one determined ECG signal portion based on the detected arrhythmia type or recommending re-evaluating the annotation data associated with the at least one determined ECG signal portion based on the likelihood score.

[0040] In some embodiments, the at least one processor may further determine that the likelihood score is less than a threshold. Additionally or alternatively, generating the at least one message may include generating the at least one message indicating a recommendation to re-evaluate the annotation data associated with the at least one determined ECG signal portion based on the determining that the likelihood score is less than the threshold.

[0041] In some embodiments, the annotation data may be received from a first computing device associated with a technician. Additionally or alternatively, the at least one message may be sent to a second computing device associated with a supervisor of the technician.

[0042] In some embodiments, the at least one neural network may comprise at least one of a deep neural network, a convolutional neural network, a recurrent neural network, an attention network, a fully connected neural network, or any combination thereof.

[0043] In some embodiments, the known arrhythmia type information may include a plurality of annotations. Additionally or alternatively, each annotation of the plurality of annotations may be associated with a respective ECG signal portion of the plurality of ECG signal portions. Additionally or alternatively, the arrhythmia type classifier may be trained based on the plurality of ECG signals and the plurality of annotations.

[0044] In some embodiments, the plurality of annotations may be from a plurality of technicians. Additionally or alternatively, each annotation of the plurality of annotations may be associated with an individual technician of the plurality of technicians and an individual ECG signal portion of the plurality of ECG signal portions. Additionally or alternatively, the arrhythmia type classifier for a first technician of the plurality of technicians may be trained based on a subset of the plurality of ECG signals and the plurality of annotations associated with at least one other technician of the plurality of technicians different from the first technician. In some embodiments, each annotation may be associated with at least one arrhythmia type associated with the individual ECG signal portion.

[0045] In some embodiments, the historical collection of the plurality of ECG signal portions may include a first plurality of ECG signal portions associated with at least one first ECG electrode and a second plurality of ECG signal portions associated with at least one second ECG electrode. Additionally or alternatively, each individual ECG signal portion of the second plurality of ECG signal portions may correspond to an individual ECG signal portion of the first plurality of ECG signal portions. Additionally or alternatively, the known arrhythmia type information may include a plurality of annotations, and / or each individual annotation of the plurality of annotations may be associated with an individual ECG signal portion of the first plurality of ECG signal portions. Additionally or alternatively, the arrhythmia type classifier may be trained by: predicting a predicted arrhythmia type for each individual ECG signal portion of the second plurality of ECG signal portions using the arrhythmia type classifier; determining at least one error value based on the predicted arrhythmia type and the individual annotation of the plurality of annotations associated with the individual ECG signal portion of the first plurality of ECG signal portions that corresponds to the individual ECG signal portion of the second plurality of ECG signal portions; and training the arrhythmia type classifier based on the at least one error value.

[0046] In some embodiments, the historical collection of the plurality of ECG signal portions may include a first plurality of ECG signal portions of at least one first ECG signal based on first surface ECG activity sensed by at least one first ECG electrode and a second plurality of ECG signal portions of at least one second ECG signal based on second surface ECG activity sensed by at least one second ECG electrode. Additionally or alternatively, the at least one second ECG electrode may be independent of the at least one first ECG electrode. Additionally or alternatively, each ECG signal portion of the first plurality of ECG signal portions may be combined with an individual ECG signal portion of the second plurality of ECG signal portions to form a plurality of estimated ECG signal portions. Additionally or alternatively, the known arrhythmia type information may include a plurality of annotations, and / or each individual annotation of the plurality of annotations may be associated with an individual estimated ECG signal portion of the plurality of estimated ECG signal portions.

[0047] In some embodiments, at least some of the plurality of ECG signal portions of the historical acquisition may be time warped to form a plurality of warped ECG signal portions.

[0048] In some embodiments, at least some of the ECG signal portions of the historical collection may be time warped to form signal portions that are at least one of filtered, inverted, or a combination thereof.

[0049] In some embodiments, the at least one noise signal portion may be combined with at least some of the plurality of ECG signal portions of the historical collection.

[0050] In some embodiments, at least a portion of the plurality of ECG signal portions of the historical collection may be style transferred.

[0051] An embodiment of the present disclosure may include an arrhythmia monitoring system. In some embodiments, the arrhythmia monitoring system may include an external cardiac monitoring device for a patient and a gateway device. The external cardiac monitoring device may include a plurality of ECG electrodes that sense the patient's surface electrocardiogram (ECG) activity, an ECG processing circuit that processes the patient's surface ECG activity to provide at least one ECG signal for the patient on at least one ECG channel, and at least one first processor operatively connected to the at least one ECG channel. The at least first processor may receive the ECG signals received via the ECG channel and transmit the ECG signals (e.g., to a gateway device). The gateway device may include a non-transitory computer-readable medium containing a rhythm change classifier and at least one second processor operatively connected to the non-transitory computer-readable medium. The rhythm change classifier may include at least one neural network trained based on a historical collection of ECG signal portions with known rhythm change information. The at least one second processor may receive the at least one ECG signal from the external cardiac monitoring device, detect time data corresponding to a predetermined rhythm change in the at least one ECG signal using the rhythm change classifier, the time data including at least one of a start time, a time interval, or any combination thereof, determine at least one ECG signal portion associated with the detected time data that corresponds to the predetermined rhythm change in the at least one ECG signal based on the detected time data, and transmit the at least one determined ECG signal portion to a remote computer system.

[0052] Some embodiments of the present disclosure may include an arrhythmia monitoring system, device, or method according to any and / or another embodiment shown, described, and / or disclosed herein.

[0053] It is understood that all combinations of the above concepts, and additional concepts described in more detail below (where such concepts are not mutually inconsistent), are contemplated as being part of the inventive subject matter described herein. In particular, all combinations of claimed subject matter appearing at the end of this disclosure are contemplated as being part of the inventive subject matter described herein. It is also understood that the terms expressly used herein, and which may be incorporated by reference into any disclosure, are to be given the meaning most consistent with the specific concepts disclosed herein. [Brief explanation of the drawings]

[0054] Those skilled in the art will understand that the drawings are primarily for illustrative purposes and that the drawings are not intended to limit the scope of the inventive subject matter described herein. The drawings are not necessarily to scale, and in some instances, various aspects of the inventive subject matter described herein may be shown exaggerated or enlarged in the drawings to facilitate an understanding of different features. In the drawings, like reference characters generally refer to like features (e.g., functionally similar and / or structurally similar elements).

[0055] [Figure 1A] FIG. 1 illustrates an exemplary block diagram of an environment for cardiac diagnostics and / or arrhythmia monitoring according to some embodiments. [Figure 1B] FIG. 1 illustrates an exemplary block diagram of an environment for cardiac diagnostics and / or arrhythmia monitoring according to some embodiments. [Figure 1C] FIG. 1 illustrates an exemplary block diagram of an environment for cardiac diagnostics and / or arrhythmia monitoring according to some embodiments.

[0056] [Figure 2A] FIG. 1 illustrates an exemplary block diagram of a system architecture for cardiac diagnostics and / or arrhythmia monitoring according to some embodiments. [Figure 2B]FIG. 1 illustrates an exemplary block diagram of a system architecture for cardiac diagnostics and / or arrhythmia monitoring according to some embodiments. [Figure 2C] FIG. 1 illustrates an exemplary block diagram of a system architecture for cardiac diagnostics and / or arrhythmia monitoring according to some embodiments.

[0057] [Figure 3A] FIG. 1 illustrates an exemplary swimlane diagram of communication flow for an exemplary process for cardiac diagnostics and / or arrhythmia monitoring according to some embodiments. [Figure 3B] FIG. 1 illustrates an exemplary swimlane diagram of communication flow for an exemplary process for cardiac diagnostics and / or arrhythmia monitoring according to some embodiments. [Figure 3C] FIG. 1 illustrates an exemplary swimlane diagram of communication flow for an exemplary process for cardiac diagnostics and / or arrhythmia monitoring according to some embodiments.

[0058] [Figure 4A] FIG. 1 illustrates an exemplary flowchart of a process for cardiac diagnosis and / or arrhythmia monitoring according to some embodiments. [Figure 4B] FIG. 1 illustrates an exemplary flowchart of a process for cardiac diagnosis and / or arrhythmia monitoring according to some embodiments. [Figure 4C] FIG. 1 illustrates an exemplary flowchart of a process for cardiac diagnosis and / or arrhythmia monitoring according to some embodiments.

[0059] [Figure 5A] FIG. 1 illustrates an exemplary diagram of a neural network of an exemplary rhythm change classifier in accordance with some embodiments.

[0060] [Figure 5B] FIG. 1 illustrates an exemplary diagram of a neural network of an arrhythmia type classifier according to some embodiments.

[0061] [Figure 6A] FIG. 1 illustrates exemplary ECG signal portions in accordance with some embodiments. [Figure 6B] FIG. 1 illustrates exemplary ECG signal portions in accordance with some embodiments. [Figure 6C] FIG. 1 illustrates exemplary ECG signal portions in accordance with some embodiments. [Figure 6D] FIG. 1 illustrates exemplary ECG signal portions in accordance with some embodiments. [Figure 6E] FIG. 1 illustrates exemplary ECG signal portions in accordance with some embodiments.

[0062] [Figure 7] FIG. 1 illustrates an exemplary block diagram of components of one or more computing devices on which the processes described herein may be implemented, according to some embodiments.

[0063] [Figure 8] FIG. 1 illustrates an exemplary schematic diagram of measurement and transmission of physiological data obtained via a body-wearable sensor (e.g., an external cardiac monitoring device) disclosed herein, in accordance with some embodiments.

[0064] [Figure 9A] 1A-1C illustrate an exemplary sensor (e.g., of an external cardiac monitoring device) disclosed herein and a patch that holds the sensor in proximity to the body, according to some embodiments. [Figure 9B] 1A-1C illustrate an exemplary sensor (e.g., of an external cardiac monitoring device) disclosed herein and a patch that holds the sensor in proximity to the body, according to some embodiments. [Figure 9C] 1A-1C illustrate an exemplary sensor (e.g., of an external cardiac monitoring device) disclosed herein and a patch that holds the sensor in proximity to the body, according to some embodiments. [Figure 9D]1A-1C illustrate an exemplary sensor (e.g., of an external cardiac monitoring device) disclosed herein, a patch that holds the sensor in proximity to the body, and attachment of the patch containing the sensor to a patient's skin, according to some embodiments. [Figure 9E] 1A-1C illustrate an exemplary sensor (e.g., of an external cardiac monitoring device) disclosed herein, a patch that holds the sensor in proximity to the body, and attachment of the patch containing the sensor to a patient's skin, according to some embodiments.

[0065] [Figure 10A] 1A-1C illustrate exemplary front views of sensors (e.g., of an external cardiac monitoring device) disclosed herein, according to some embodiments. [Figure 10B] 1A-1C illustrate exemplary rear views of sensors (e.g., of an external cardiac monitoring device) disclosed herein, according to some embodiments. [Figure 10C] 1A-1C illustrate exemplary exploded views of sensors (e.g., of an external cardiac monitoring device) disclosed herein, according to some embodiments.

[0066] [Figure 11A] FIG. 1 shows an example illustration of a device electronics architecture for measuring and transmitting patient physiological data (e.g., biometric data) in accordance with some embodiments.

[0067] [Figure 11B] FIG. 1 illustrates a block diagram of an exemplary architecture of a radio frequency (RF) module in accordance with some embodiments.

[0068] [Figure 11C] FIG. 1 illustrates a block diagram of another exemplary architecture of an RF module in accordance with some embodiments.

[0069] [Figure 12]1 illustrates an exemplary external, ambulatory, patient-wearable medical device (e.g., an external cardiac monitoring device) according to some embodiments.

[0070] [Figure 13] 1 illustrates an exemplary component-level diagram of a medical device (eg, an external cardiac monitoring device) according to some embodiments. DETAILED DESCRIPTION OF THE INVENTION

[0071] The present disclosure relates to systems, devices, and methods for cardiac diagnosis and / or arrhythmia monitoring, including heart failure status monitoring. For example, one or more trained classifiers, each including at least one neural network, may be used in a cardiac monitoring device (e.g., an external, wearable, and / or similar cardiac monitoring device), a computer system, and / or the like to detect (e.g., identify and / or perform the like) rhythm variations, arrhythmia types, and / or the like based on at least one ECG signal or portion thereof. Exemplary usage scenarios include use of the systems, devices, and methods for cardiac diagnosis and / or arrhythmia monitoring in the context of mobile cardiac telemetry, cardiac Holter monitoring (including extended cardiac Holter monitoring), wearable cardiac monitors, wearable defibrillators, wearable cardioverter-defibrillators, and other such ambulatory cardiac monitoring and / or therapy systems.

[0072] In some embodiments, the arrhythmia monitoring system may include an external cardiac monitoring device for the patient. For example, the external cardiac monitoring device may include multiple electrocardiogram (ECG) electrodes for sensing the patient's surface ECG activity. The ECG processing circuit may process the patient's surface ECG activity to provide at least one ECG signal to the patient in at least one ECG channel. The rhythm change classifier may be implemented in a non-transitory computer-readable medium (e.g., a memory, a programmable circuit board, a field-programmable gate array, an integrated circuit, any combination thereof, and / or the like). The rhythm change classifier may include at least one neural network trained based on a historical collection of multiple ECG signal portions with known rhythm change information. Further, at least one processor may be operatively connected to the ECG channel and the non-transitory computer-readable medium. The processor may receive the ECG signal via the ECG channel. The processor may use the rhythm change classifier to detect time data corresponding to rhythm changes (e.g., predetermined rhythm changes, and / or the like) in the ECG signal. For example, the time data may include a start time, a time interval, any combination thereof, and / or the like. The processor may also determine, based on the detected time data, at least one ECG signal portion associated with the detected time data that corresponds to a predetermined rhythm change in the ECG signal. The processor may also transmit the determined ECG signal portion to a remote computer system (e.g., a remote server, a cardiac monitoring facility, and / or the like).

[0073] For example, in such an embodiment, the rhythm change classifier (e.g., a neural network and / or the like) may detect (e.g., identify and / or the like) rhythm changes without classifying a particular rhythm type. For illustrative purposes, the rhythm change classifier may detect when the cardiac rhythm changes from normal sinus rhythm (NSR) to atrial fibrillation (AFIB), from AFIB to NSR, from AFIB to atrial flutter (AFL), from one form to another, and / or the like. Furthermore, each time a rhythm change is detected (e.g., identified and / or the like), a portion of the ECG signal (e.g., an ECG strip and / or the like) including the detected rhythm change may be transmitted to a remote computer system (e.g., bearing an indicator, message, marking, and / or the like indicating the detected rhythm change). Additionally or alternatively, the rhythm change classifier may determine a confidence score associated with the detected rhythm change (e.g., output a confidence score associated with the probability that the detected rhythm change is actually a rhythm change).

[0074] In some embodiments, the rhythm change classifier may also use (e.g., receive as input to a neural network, and / or the like) additional sensor-related data (e.g., non-ECG biometric data from at least one sensor), including, but not limited to, accelerometer data, heart sound data, electromagnetic waves (e.g., radio frequency (RF) waves, and / or the like) scattered and / or reflected from internal tissue, any combination thereof, and / or the like. Such additional non-ECG biometric data may further improve accuracy, reliability, and / or the like associated with rhythm change detection.

[0075] In some embodiments, the rhythm change classifier may also use (e.g., receive as input to a neural network, compare, and / or the like) data related to a baseline ECG signal, a reference vector, calibration measurements, previous measurements, and / or the like for the patient. For example, such baseline signals and / or additional measurements may be obtained using high-precision equipment in a clinic, may be obtained by an external cardiac monitoring device at known rest periods, and / or the like. The use of such patient-specific baseline signals and / or additional measurements may further improve the accuracy, reliability, and / or the like associated with rhythm change detection.

[0076] In some embodiments, using a neural network trained to detect (e.g., identify and / or the like) rhythm changes and transmit ECG signal portions only if a rhythm change is detected may allow for reduced power consumption, suitable for, for example, wearable, external devices having a relatively small power source (e.g., battery, and / or the like). Furthermore, the rhythm change classifier (e.g., its neural network) may be implemented using low-power hardware (e.g., programmable circuit boards, field-programmable gate arrays, integrated circuits, and / or the like), which may further reduce power consumption.

[0077] In some embodiments, the classifier and / or its neural network logic may be distributed among multiple devices (e.g., external cardiac monitoring devices, gateway devices, and / or remote computer systems), e.g., to optimize power consumption, reduce the amount of transmitted data, improve accuracy of rhythm change detection, and / or address other constraints. For example, the rhythm change classifier may be implemented (e.g., fully, partially, and / or similarly) on a gateway device (e.g., a smartphone, tablet, laptop computer, and / or similar mobile computing device) that may receive ECG signal data and / or other sensor data from the external cardiac monitoring device (e.g., via low-power wireless transmissions such as Bluetooth®, Bluetooth Low Energy (BLE), and / or the like). Additionally or alternatively, a second classifier (e.g., a larger, more accurate rhythm change classifier, arrhythmia type classifier, and / or the like) may be implemented (e.g., fully, partially, and / or similarly) on a remote computer system.

[0078] In some embodiments, additional ECG signal portions may be sampled (e.g., by a technician, and / or the like) and transmitted from the external cardiac monitoring device to a remote computer system for annotation (e.g., randomly, at predetermined intervals based on a confidence score below an upper threshold but above a lower threshold, and / or the like) to test the performance of a classifier (e.g., a rhythm change classifier, an arrhythmia type classifier, and / or the like), expand the training dataset (e.g., added to a historical collection of ECG signal portions, and / or the like), retrain the classifier, and / or the like.

[0079] In some embodiments, ECG signal portions transmitted from the external cardiac monitoring device to the remote computer system may be categorized (e.g., bucketed and / or the like), grouped, and / or the like. Furthermore, ECG signal portions in the same category / group may be presented together to a user (e.g., a technician and / or the like). This may enhance human review (e.g., processing, querying, annotation, and / or the like) of such ECG signal portions.

[0080] In some embodiments, the arrhythmia detection system may include at least one non-transitory computer-readable medium (e.g., a memory, a programmable circuit board, a field-programmable gate array, an integrated circuit, any combination thereof, and / or the like) to which at least one processor may be operatively connected. The arrhythmia type classifier may be implemented on the non-transitory computer-readable medium. The arrhythmia type classifier may include at least one neural network trained based on a historical collection of multiple ECG signal portions with known arrhythmia type information. The processor may receive annotation data associated with at least one ECG signal and at least one annotation for each ECG signal. For example, the annotation data may be received from a computing device associated with a technician. The processor may use the arrhythmia type classifier to detect arrhythmia types in the ECG signal and time data associated with the detected arrhythmia type. The time data may include at least one of a start time, a time interval, or any combination thereof. The processor may determine at least one ECG signal portion associated with the detected arrhythmia type in the ECG signal based on the time data. The processor may determine a likelihood score for the annotation based on the detected arrhythmia type. The processor may generate at least one message based on the determined ECG signal portion and the likelihood score for the annotation. For example, the message may indicate at least one of: recommending annotating the determined ECG signal portion based on the detected arrhythmia type; recommending re-evaluating annotation data associated with the determined ECG signal portion based on the likelihood score; and / or the like. The processor may, for example, transmit the message associated with the ECG signal portion to a technician's computing device.

[0081] For example, in such embodiments, technicians may have different levels of experience, expertise, and / or the like. A system using such an arrhythmia type classifier may identify arrhythmia inconsistencies, anomalies, missed events, and / or the like. Additionally or alternatively, such a system may route (e.g., communicate, transmit, and / or the like) identified ECG signal portions to the technician, another more senior technician (e.g., the technician's supervisor), and / or the like for further review (e.g., processing, querying, annotation, and / or the like).

[0082] 1A-1C, which illustrate an example block diagram of an environment 100 in which the systems, products, and / or methods described herein may be implemented. As shown in FIGS. 1A-1C, the environment 100 may include a cardiac monitoring device 102, a remote computer system 104, a data repository 106, a technician device 108, a supervisor device 110, and / or a gateway device 129.

[0083] Cardiac monitoring device 102 may comprise one or more devices capable of receiving information from and / or communicating information to (e.g., directly and / or indirectly via wired and / or wireless networks and / or other suitable communication technologies) remote computer system 104, gateway device 129, data repository 106, technician device 108, and / or supervisor device 110. In some embodiments, cardiac monitoring device 102 may be an external cardiac monitoring device as described herein (e.g., wearable on a patient, externally connected to a patient, and / or the like). In some embodiments, cardiac monitoring device 102 may include ECG electrodes 122 (e.g., multiple ECG electrodes sensing a patient's surface ECG activity), ECG processing circuitry 124 (e.g., ECG processing circuitry processing a patient's surface ECG activity to provide at least one ECG signal for the patient in at least one ECG channel), a rhythm change classifier 112 (e.g., implemented on at least one non-transitory computer-readable medium), a processor 126 (e.g., at least one processor operably connected to the at least one ECG channel and the non-transitory computer-readable medium), an ECG sensor 128 (e.g., at least one sensor and associated sensor circuitry sensing the patient's non-ECG biometric data), and / or the like, as described herein. In some embodiments, rhythm change classifier 112 may include at least one neural network trained based on historical collections of multiple ECG signal portions 116 with known rhythm change information 118, as described herein. In some embodiments, as described herein, processor 126 may receive at least one ECG signal via at least one ECG channel (e.g., from ECG processing circuit 124). Additionally or alternatively, processor 126 may detect (e.g., using rhythm change classifier 112) temporal data corresponding to predetermined rhythm changes in the ECG signal, as described herein.In some embodiments, the time data may include at least one of a start time, a time interval, any combination thereof, and / or the like, as described herein. Additionally or alternatively, the processor 126 may determine (e.g., based on the detected time data) at least one ECG signal portion associated with the detected time data that corresponds to a predetermined rhythm change in the at least one ECG signal, as described herein. Additionally or alternatively, the processor 126 may transmit the at least one determined ECG signal portion to the remote computer system 104, as described herein.

[0084] The remote computer system 104 may comprise one or more devices capable of receiving information from and / or communicating information to the cardiac monitoring device 102, the gateway device 129, the data repository 106, the technician device 108, and / or the supervisor device 110 (e.g., directly and / or indirectly via a wired and / or wireless network and / or any other suitable communication technology). In some embodiments, the remote computer system 104 may comprise a server, a group of servers, and / or other similar devices, as described herein. Additionally or alternatively, the remote computer system 104 may comprise at least one other computing device, separate from or including the server and / or group of servers, such as a portable and / or handheld device (e.g., a computer, laptop, personal digital assistant (PDA), smartphone, tablet, and / or the like), desktop computer, and / or other similar device, as described herein. In some embodiments, the remote computer system 104 may be associated with a cardiac monitoring facility, and / or the like, as described herein. Additionally or alternatively, the cardiac monitoring facility may be associated with a provider (e.g., manufacturer, distributor, and / or the like) of the cardiac monitoring device 102, as described herein. In some embodiments, the remote computer system 104 may include at least one classifier (e.g., rhythm change classifier 112, arrhythmia type classifier 114, and / or the like), as described herein. Additionally or alternatively, each classifier (e.g., rhythm change classifier 112, arrhythmia type classifier 114, and / or the like) may be implemented by at least one non-transitory computer-readable medium.In some embodiments, the remote computer system 104 may comprise at least one processor operatively connected to a non-transitory computer-readable medium, as described herein. In some embodiments, the remote computer system 104 may communicate with a data repository 106 that is local or remote to the remote computer system 104. In some embodiments, the remote computer system 104 may receive information from, store information in, communicate information to, or retrieve information stored in the data repository 106.

[0085] In some embodiments, the remote computer system 104 (e.g., its processor) may receive at least one determined ECG signal portion from the (external) cardiac monitoring device 102, analyze the determined ECG signal portion to classify an arrhythmia type relative to rhythm changes in the ECG signal, and / or the like, as described herein. For example, the arrhythmia type classifier 114 may include at least one (second) neural network trained based on a (second) historical collection of a (second) plurality of ECG signal portions 116 with known arrhythmia type information 120, as described herein. In some embodiments, the arrhythmia type may include at least one of a heart rate change, atrial fibrillation, flutter, supraventricular tachycardia, ventricular tachycardia, pauses, atrioventricular block, ventricular fibrillation, bigeminy, trigeminy, ventricular ectopic beats, bradycardia, tachycardia, morphological changes in at least one ECG signal, any combination thereof, and / or the like, as described herein.

[0086] In some embodiments, the remote computer system 104 (e.g., its processor) may receive annotation data associated with at least one ECG signal and at least one annotation for each ECG signal, as described herein. Additionally or alternatively, the remote computer system 104 (e.g., its processor) may detect arrhythmia types in the ECG signals (e.g., using the arrhythmia type classifier 114) and time data associated with the detected arrhythmia types, as described herein. In some embodiments, the time data may comprise at least one of a start time, a time interval, any combination thereof, and / or the like, as described herein. In some embodiments, the remote computer system 104 (e.g., its processor) may determine at least one ECG signal portion (e.g., based on the time data) associated with the detected arrhythmia type in the ECG signal, as described herein. In some embodiments, the remote computer system 104 (e.g., its processor) may determine a likelihood score for the annotation based on the detected arrhythmia type, as described herein. In some embodiments, the remote computer system 104 (e.g., its processor) may generate at least one message based on the determined ECG signal portion and the likelihood score for the annotation, as described herein. Additionally or alternatively, the message may indicate at least one of: recommending annotating the determined ECG signal portion based on the detected arrhythmia type, as described herein; or recommending re-evaluating the annotation data associated with the determined ECG signal portion based on the likelihood score; and / or the like. In some embodiments, the remote computer system 104 (e.g., its processor) may transmit a message related to the determined ECG signal portion (e.g., to the technician device 108, the supervisor device 110, and / or the like).

[0087] Data repository 106 may comprise one or more devices that can receive information from and / or communicate information to cardiac monitoring device 102, remote computer system 104, gateway device 129, technician device 108, and / or supervisor device 110 (e.g., directly and / or indirectly via wired and / or wireless networks and / or any other suitable communication technology). In some embodiments, data repository 106 may comprise a server, a group of servers, and / or other similar devices, as described herein. In some embodiments, data repository 106 may be associated with a cardiac monitoring facility, as described herein, and / or the like. Additionally or alternatively, a cardiac monitoring facility, as described herein, may be associated with a provider (e.g., a manufacturer, distributor, and / or the like) of cardiac monitoring device 102. In some embodiments, data repository 106 is part of remote computer system 104. Additionally or alternatively, the data repository 106 may be local or remote to the remote computer system 104. In some embodiments, the remote computer system 104 may receive information from the data repository 106, store information in the data repository 106, communicate information to the data repository 106, or retrieve information stored in the data repository 106.

[0088] In some embodiments, the data repository 106 may comprise at least one historical collection of the plurality of ECG signal portions 116, as described herein. Additionally or alternatively, the data repository 106 may comprise known rhythm change information 118 for at least some historical collection of the plurality of ECG signal portions 116 (e.g., a first plurality of ECG signal portions, and / or the like), as described herein. Additionally or alternatively, the data repository 106 may comprise known arrhythmia type information 120 for at least some historical collection of the plurality of ECG signal portions 116 (e.g., a second plurality of ECG signal portions, and / or the like), as described herein. In some embodiments, the data repository 106 may comprise patient-specific information, such as, for example, a baseline ECG and / or other baseline physiological information, a patient-specific reference ECG and / or other reference physiological information, patient-specific calibration data, and / or the like. For example, prior to initial deployment to a patient, the cardiac monitoring device 102 may be attached to the patient and / or used to determine initial patient-specific information (e.g., baseline ECG and / or physiological data, and / or the like). Such patient-specific information may be designated as baseline data for the patient. Additionally or alternatively, certain patient-specific data may be designated as reference data for particular analyses, as described herein. In some embodiments, the data repository 106 may include patient-specific classifiers (e.g., patient-specific rhythm change classifiers, patient-specific arrhythmia type classifiers, and / or the like) that may be trained based on such baseline and / or reference data, as described herein.

[0089] Technician device 108 may comprise one or more technician-related devices that can receive information from and / or communicate information to cardiac monitoring device 102, remote computer system 104, gateway device 129, data repository 106, and / or supervisor device 110 (e.g., directly and / or indirectly via wired and / or wireless networks and / or any other suitable communication technology). In some embodiments, technician device 108 may comprise at least one computing device, such as a portable and / or handheld device (e.g., a computer, laptop, personal digital assistant (PDA), smartphone, tablet, and / or the like), desktop computer, and / or other similar device, as described herein. In some embodiments, technician device 108 may be associated with a cardiac monitoring facility, and / or the like, as described herein. Additionally or alternatively, the cardiac monitoring facility may be associated with a provider (e.g., a manufacturer, distributor, and / or the like) of cardiac monitoring device 102, as described herein. In some embodiments, the technician device 108 is part of the remote computer system 104. Additionally or alternatively, the technician device 108 may be local or remote to the remote computer system 104. In some embodiments, the technician device 108 may include at least one input component (e.g., a touch panel display, a keyboard, a keypad, a mouse, buttons, switches, a microphone, a camera, and / or the like) that allows the technician device 108 to receive information via user input and / or the like. Additionally or alternatively, the technician device 108 may include at least one output component (e.g., a display, a touch screen, a speaker, and / or the like) that provides output information from the technician device 108.In some embodiments, the technician device 108 may receive messages (e.g., annotations, arrhythmia type identification, rhythm variation identification, classification, and / or recommendations regarding the like) related to the ECG signal and / or portions thereof (e.g., from the remote computer system 104, the cardiac monitoring device 102, the supervisor device 110, and / or the like) as described herein. Additionally or alternatively, the technician device 108 may communicate (e.g., to the remote computer system 104, the cardiac monitoring device 102, the supervisor device 110, and / or the like) annotation data related to at least one annotation for the individual ECG signal and / or portions thereof as described herein.

[0090] The supervisor device 110 may comprise a device associated with a supervisor (e.g., a supervisor of at least one technician, and / or the like) that can receive information from and / or communicate information to the cardiac monitoring devices 102, the remote computer system 104, the gateway device 129, the data repository 106, and / or the technician device 108 (e.g., directly and / or indirectly via wired and / or wireless networks and / or any other suitable communication technology). In some embodiments, the supervisor device 110 may comprise at least one computing device, such as a portable and / or handheld device (e.g., a computer, a laptop, a personal digital assistant (PDA), a smartphone, a tablet, and / or the like), a desktop computer and / or other similar device, and / or the like, as described herein. In some embodiments, the supervisor device 110 may be associated with a cardiac monitoring facility and / or the like, as described herein. Additionally or alternatively, the cardiac monitoring facility may be associated with a provider (e.g., manufacturer, distributor, and / or the like) of the cardiac monitoring device 102, as described herein. In some embodiments, the supervisor device 110 may be part of the remote computer system 104. Additionally or alternatively, the supervisor device 110 may be local or remote to the remote computer system 104. In some embodiments, the supervisor device 110 may include at least one input component (e.g., a touch panel display, keyboard, keypad, mouse, buttons, switches, microphone, camera, and / or the like) that allows the supervisor device 110 to receive information via user input and / or the like. Additionally or alternatively, the supervisor device 110 may include at least one output component (e.g., a display, touch screen, speaker, and / or the like) that provides output information from the supervisor device 110.In some embodiments, the supervisor device 110 may receive messages (e.g., annotations, arrhythmia type identification, rhythm variation identification, classification, and / or recommendations regarding the like) related to the ECG signals and / or portions thereof (e.g., from the remote computer system 104, the cardiac monitoring device 102, the technician device 108, and / or the like) as described herein. Additionally or alternatively, the supervisor device 110 may communicate (e.g., to the remote computer system 104, the cardiac monitoring device 102, the technician device 108, and / or the like) annotation data related to at least one annotation for the individual ECG signals and / or portions thereof as described herein.

[0091] Gateway device 129 may comprise one or more devices capable of receiving information from and / or communicating information to cardiac monitoring device 102, remote computer system 104, data repository 106, technician device 108, and / or supervisor device 110 (e.g., directly and / or indirectly via wired and / or wireless networks and / or any other suitable communication technology). In some embodiments, gateway device 129 may comprise at least one computing device such as a portable and / or handheld device (e.g., a computer, laptop, personal digital assistant (PDA), smartphone, tablet, and / or the like), desktop computer, and / or other similar device, as described herein. In some embodiments, gateway device 129 may be associated with a patient, e.g., an individual patient associated with (e.g., connected to, implanted in, and / or the like) cardiac monitoring device 102. In some embodiments, gateway device 129 may be associated with a cardiac monitoring facility, as described herein, and / or the like. Additionally or alternatively, the cardiac monitoring facility may be associated with a provider (e.g., manufacturer, distributor, and / or the like) of cardiac monitoring device 102, as described herein. In some embodiments, gateway device 129 may be local or remote to remote cardiac monitoring system 102. Additionally or alternatively, gateway device 129 may be local or remote to remote computer system 104. In some embodiments, gateway device 129 may include at least one input component (e.g., a touch panel display, keyboard, keypad, mouse, buttons, switches, microphone, camera, and / or the like) that allows gateway device 129 to receive information via user input and / or the like.Additionally or alternatively, gateway device 129 may include at least one output component (e.g., a display, touchscreen, speaker, and / or the like) that provides output information from gateway device 129. In some embodiments, gateway device 129 may receive biometric data (e.g., ECG signals, ECG signal portions, non-ECG biometric data, and / or the like) from cardiac monitoring device 102 and / or the like, as described herein. Additionally or alternatively, gateway device 129 may communicate biometric data (e.g., ECG signals, ECG signal portions, non-ECG biometric data, and / or the like) to remote computer system 104 and / or the like, as described herein. In some embodiments, rhythm change classifier 112 may be implemented (e.g., completely, partially, or the like) by a non-transitory computer-readable medium of gateway device 129 (e.g., independently of, instead of, or in addition to cardiac monitoring device 102). Additionally or alternatively, the processor of gateway device 129 may receive at least one ECG signal (e.g., from cardiac monitoring device 102) and / or detect time data corresponding to a predetermined rhythm change in the ECG signal (e.g., using rhythm change classifier 112) as described herein. In some embodiments, the time data may include at least one of a start time, a time interval, any combination thereof, and / or the like as described herein. Additionally or alternatively, the processor of gateway device 129 may determine (e.g., based on the detected time data) at least one ECG signal portion associated with the detected time data corresponding to a predetermined rhythm change in the at least one ECG signal and / or transmit the at least one determined ECG signal portion to remote computer system 104 as described herein.

[0092] In some embodiments, the cardiac monitoring device 102, the remote computer system 104, the data repository 106, the technician device 108, the supervisor device 110, and / or the gateway device 129 may be connected by one or more networks. The networks may comprise one or more wired and / or wireless networks. For example, the network may include a cellular network (e.g., a long-term evolution (LTE) network, a third generation (3G) network, a fourth generation (4G) network, a code division multiple access (CDMA) network, and / or the like), a public land mobile network (PLMN), a local area network (LAN), a wide area network (WAN), a metropolitan area network (MAN), a telephone network (e.g., the public switched telephone network (PSTN)), a private network, a virtual private network (VPN), a local network, an ad hoc network, an intranet, the Internet, an optical fiber-based network, a cloud computing network, and / or the like, and / or a combination of these or other types of networks.

[0093] Reference is now made to FIG. 2A , which illustrates an exemplary block diagram of a system architecture 200a for arrhythmia monitoring according to some embodiments. In addition to system components, the system architecture 200a also illustrates data flow between the system components. As illustrated in FIG. 2A , the system architecture 200a may include a cardiac monitoring device 202a, a remote computer system 204a, a data repository 206a, a technician device 208a, and / or a gateway device 229a. In some embodiments, the cardiac monitoring device 202a may be the same as or similar to the cardiac monitoring device 102. In some embodiments, the remote computer system 204a may be the same as or similar to the remote computer system 104 (e.g., one or more devices in the remote computer system 104). In some embodiments, the data repository 206a may be the same as or similar to the data repository 106 (e.g., one or more devices in the data repository 106). In some embodiments, the technician device 208a may be the same as or similar to the technician device 108. In some embodiments, gateway device 229a may be the same as or similar to gateway device 129.

[0094] 2A , at 230a, the remote computer system 204a may receive (e.g., retrieve, search, send requests and / or queries to the data repository 206a for communication, and / or the like) a historical collection of multiple ECG signal portions and associated information (e.g., known rhythm change information, known arrhythmia type information, and / or the like) as described herein, for example, from the data repository 206a. In some embodiments, the remote computer system 204a may train at least one neural network of at least one classifier (e.g., a rhythm change classifier, an arrhythmia type classifier, and / or the like) based on the historical collection of multiple ECG signal portions and associated information (e.g., each of the known rhythm change information, known arrhythmia type information, and / or the like) as described herein.

[0095] In some embodiments, the rhythm change classifier may comprise at least one neural network as described herein. Additionally or alternatively, the at least one neural network may comprise at least one of a convolutional neural network, a recurrent neural network, an attention network, a fully-connected neural network, any combination thereof, and / or the like. For example, the neural network may include at least one convolutional neural network having multiple convolutional layers. In some embodiments, the convolutional neural network may have 5 to 40 convolutional layers (e.g., at least 5 convolutional layers and up to 40 convolutional layers). For example, the convolutional neural network may have 7 to 10 convolutional layers (e.g., at least 7 and no more than 10 convolutional layers). In some embodiments, each convolutional layer may include at least one convolutional node (e.g., multiple convolutional nodes). In some embodiments, the convolutional neural network may further include an input layer and an output layer. For example, the ECG signal may include a plurality of ECG signal samples. Additionally or alternatively, the input layer may include at least one node for each ECG signal sample of the plurality of ECG signal samples (or a subset of the plurality of ECG signal samples, e.g., associated with a predetermined time period, a buffer size for the ECG signal samples, and / or the like). Additionally or alternatively, the input layer may further include at least one input for non-ECG biometric data associated with each sensor of the cardiac monitoring device 202a, as described herein. In some embodiments, the output of the output layer may include an indication of time data corresponding to the rhythm change. Additionally or alternatively, the output of the output layer may include a confidence score, as described herein. In some embodiments, the neural network may include a plurality of Siamese branches (e.g., each individual Siamese branch associated with an individual ECG channel), as described herein.

[0096] In some embodiments, the remote computer system 204a may train the rhythm change classifier. For example, the remote computer system 204a may train the rhythm change classifier by using the rhythm change classifier to generate predicted rhythm change information (e.g., data related to predicted rhythm changes (e.g., probabilities, confidence scores, and / or the like), data related to the absence of predicted rhythm changes (e.g., probabilities, confidence scores, and / or the like), and / or the like) for each ECG signal portion of a historical collection of multiple ECG signal portions (or a first plurality of ECG signal portions), determining at least one error value based on the predicted rhythm change information and the known rhythm change information, and updating the rhythm change classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss.

[0097] In some embodiments, the historical collection of multiple ECG signal portions may include a first ECG signal portion associated with a first time and a second ECG signal portion associated with a second time after the first time. Additionally or alternatively, the remote computer system 204a may train the rhythm change classifier by generating, using the rhythm change classifier, a predicted ECG signal portion associated with the second time based on the first ECG signal portion, determining at least one error value based on the predicted ECG signal portion and the second ECG signal portion, and updating the rhythm change classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss.

[0098] In some embodiments, the historical collection of multiple ECG signal portions may include a first ECG signal portion associated with a first time and a second ECG signal portion associated with a second time. Additionally or alternatively, the remote computer system 204a may train the rhythm change classifier by using the rhythm change classifier to generate a predicted time associated with the second ECG signal portion based on the first ECG signal portion and the second ECG signal, determining at least one error value based on the predicted time and the second time, and updating the rhythm change classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss.

[0099] In some embodiments, there may be an insufficient number of ECG signal portions in a historical collection of multiple ECG signal portions with known rhythm change information to train a rhythm change classifier to perform a desired task (e.g., detecting and / or identifying at least one predetermined rhythm change). Additionally or alternatively, there may be sufficient data (e.g., past ECG signal portions and / or the like) to train the rhythm change classifier to perform another task (e.g., that may be related in some way to the target task). In some embodiments, the rhythm change classifier may be trained to perform another task (e.g., counting R peaks based on ECG signals, determining heart rate, and / or the like). Additionally or alternatively, the rhythm change classifier may be applied to perform the target task. For example, in some embodiments, the rhythm change classifier may be retrained using a limited amount of ECG signal portions in a historical collection of multiple ECG signal portions with known rhythm change information and / or the like. Additionally or alternatively, the rhythm change classifier may be used to perform another task (e.g., counting R peaks based on ECG signals, determining a heart rate, and / or the like), and its output may be applied to the target task. For example, a processor (e.g., of cardiac monitoring device 202a and / or gateway device 229a) may use the rhythm change classifier to detect at least one peak count or heart rate based on at least one ECG signal. Additionally or alternatively, the processor (e.g., of cardiac monitoring device 202a and / or gateway device 229a) may determine that the detected at least one peak count or heart rate exceeds a first threshold value for the patient (e.g., a tachycardia onset threshold) or is below a second threshold value for the patient (e.g., a bradycardia onset threshold) (e.g., where the second threshold value for the patient may be less than the first threshold value for the patient).Additionally or alternatively, the processor (e.g., of the cardiac monitoring device 202a and / or gateway device 229a) may detect a predetermined rhythm change based on at least one peak number or heart rate exceeding a first threshold value for the patient (e.g., a tachycardia onset threshold) or falling below a second threshold value for the patient (e.g., a bradycardia onset threshold).

[0100] In some embodiments, there may be an insufficient number of ECG signal portions associated with (e.g., sensed from and / or related by the same) the plurality of ECG electrodes of cardiac monitoring device 202a in a historical collection of ECG signal portions with known rhythm change information to train a rhythm change classifier for ECG signals received from the plurality of ECG electrodes of cardiac monitoring device 202a. Additionally or alternatively, there may be sufficient data (e.g., past ECG signal portions, and / or the like) associated with (e.g., sensed from and / or related by the same) a second plurality of ECG electrodes independent of the plurality of ECG electrodes of cardiac monitoring device 202a (e.g., electrodes of an ECG device separate from cardiac monitoring device 202a, such as a 12-lead ECG sensor, a separate external and / or wearable cardiac monitoring device, and / or the like) to train a rhythm change classifier based on the second plurality of ECG electrodes. In some embodiments, the rhythm change classifier may be trained based on ECG signal portions associated with the second plurality of ECG electrodes (e.g., sensed from the second plurality of ECG electrodes and / or the like). Additionally or alternatively, the rhythm change classifier may then be applied to detect predetermined rhythm changes based on the plurality of ECG electrodes of cardiac monitoring device 202a. In some embodiments, remote computer system 204a may determine (e.g., calculate and / or perform the like) a transformation (e.g., vector projection and / or the like) of the ECG signal portions associated with the second plurality of ECG electrodes onto the plurality of ECG electrodes of cardiac monitoring device 202a and use the transformation of the ECG signal portions to train the rhythm change classifier as if the ECG signal portions were associated with (e.g., sensed from and / or the like) the plurality of ECG electrodes of cardiac monitoring device 202a.

[0101] In some embodiments, the remote computer system 204a may train the arrhythmia type classifier by: predicting, using the arrhythmia type classifier, a predicted arrhythmia type for each individual ECG signal portion of a historical collection of the plurality of ECG signal portions (or a second plurality of ECG signal portions); determining at least one error value based on the predicted arrhythmia type and known arrhythmia type information (e.g., individual annotations associated with known arrhythmia types for each individual ECG signal portion); and training the arrhythmia type classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss.

[0102] In some embodiments, the arrhythmia type classifier may comprise at least one neural network (e.g., at least one second neural network) as described herein. Additionally or alternatively, the at least one (second) neural network may comprise at least one of a deep neural network, a convolutional neural network, a recurrent neural network, an attention network, a fully connected neural network, any combination thereof, and / or the like. For example, the neural network may comprise at least one convolutional neural network having multiple convolutional layers. In some embodiments, the convolutional neural network may have 5 to 40 convolutional layers (e.g., at least 5 convolutional layers and up to 40 convolutional layers). For example, the convolutional neural network may have 7 to 10 convolutional layers (e.g., at least 7 and no more than 10 convolutional layers). In some embodiments, each convolutional layer may include at least one convolutional node (e.g., multiple convolutional nodes). In some embodiments, the convolutional neural network may further have an input layer and an output layer. For example, the ECG signal may include a plurality of ECG signal samples. Additionally or alternatively, the input layer may include at least one node for each ECG signal sample of the plurality of ECG signal samples (or a subset of the plurality of ECG signal samples, e.g., associated with a predetermined time period, a buffer size for the ECG signal samples, and / or the like). Additionally or alternatively, the input layer may further include at least one input for non-ECG biometric data associated with a sensor (e.g., of the cardiac monitoring device 202a, and / or the like), as described herein. In some embodiments, the output of the output layer may include an indication of the time data corresponding to the arrhythmia type. Additionally or alternatively, the output of the output layer may include a confidence score, a likelihood score, and / or the like, as described herein.In some embodiments, the neural network may include multiple Siamese branches (eg, each individual Siamese branch associated with an individual ECG channel) as described herein.

[0103] 2A , at 232a, remote computer system 204a may communicate the trained rhythm change classifier (or weights thereof) to cardiac monitoring device 202a and / or gateway device 229a, as described herein. In some embodiments, after training, weights corresponding to the trained rhythm change classifier may be communicated to cardiac monitoring device 202a and / or gateway device 229a. Additionally or alternatively, a copy of the trained rhythm change classifier (or weights thereof) may be downloaded from remote computer system 204a and / or installed (e.g., uploaded to, written to, configured on, and / or the like) on at least one non-transitory computer-readable medium (e.g., memory, programmable circuit board, field programmable gate array (FPGA), integrated circuit, any combination thereof, and / or the like) that may be installed on and / or part of cardiac monitoring device 202a and / or gateway device 229a.

[0104] In some embodiments, the cardiac monitoring device 202a may be an external cardiac monitoring device for the patient, as described herein. For example, the (external) cardiac monitoring device 202a may include a plurality of ECG electrodes that sense the patient's surface ECG activity. Additionally or alternatively, the (external) cardiac monitoring device 202a may include ECG processing circuitry that processes the patient's surface ECG activity to provide at least one ECG signal for the patient on at least one ECG channel.

[0105] In some embodiments, the (external) cardiac monitoring device 202a may comprise a non-transitory computer-readable medium (e.g., a memory, a programmable circuit board, a field-programmable gate array, an integrated circuit, any combination thereof, and / or the like) having (e.g., implementing, embodying, storing, and / or the like) a trained rhythm change classifier as described herein (which may include, e.g., at least one neural network trained based on historical collections of multiple ECG signal portions with known rhythm change information). Additionally or alternatively, the (external) cardiac monitoring device 202a may comprise at least one processor operably connected to the ECG channel and / or the non-transitory computer-readable medium.

[0106] In some embodiments, gateway device 229a may comprise a non-transitory computer-readable medium (e.g., memory, a programmable circuit board, a field-programmable gate array, any combination thereof, and / or the like) having (e.g., implementing, embodying, storing, and / or the like) a trained rhythm change classifier as described herein (which may include, e.g., at least one neural network trained based on a historical collection of multiple ECG signal portions with known rhythm change information). Additionally or alternatively, gateway device 229a may comprise at least one processor operably connected to the non-transitory computer-readable medium.

[0107] In some embodiments, the cardiac monitoring device 202a (e.g., its processor) may receive ECG signals via an ECG channel. Additionally or alternatively, the cardiac monitoring device 202a (e.g., its processor) may use a rhythm change classifier to detect time data corresponding to a predetermined rhythm change in at least one ECG signal. For example, the predetermined rhythm change may be associated with an arrhythmia (e.g., a change in heart rate, atrial fibrillation, flutter, supraventricular tachycardia, ventricular tachycardia, pauses, atrioventricular block, ventricular fibrillation, bigeminy, trigeminy, ventricular ectopic beats, bradycardia, tachycardia, a change in morphology of at least one ECG signal, any combination thereof, and / or the like). Additionally or alternatively, the time data may include at least one of a start time, a time interval, any combination thereof, and / or the like. In some embodiments, the cardiac monitoring device 202a (e.g., its processor) may determine, based on the detected time data, a portion of at least one ECG signal associated with the detected time data that corresponds to the predetermined rhythm change in the ECG signal.

[0108] In some embodiments, the cardiac monitoring device 202a (e.g., its processor) may receive ECG signals via an ECG channel. Additionally or alternatively, the cardiac monitoring device 202a (e.g., its processor) may transmit the ECG signals to the gateway device 229a. In some embodiments, the gateway device 229a (e.g., its processor) may use a rhythm change classifier to detect temporal data corresponding to predetermined rhythm changes in the ECG signals. For example, the predetermined rhythm changes may be associated with arrhythmias (e.g., changes in heart rate, atrial fibrillation, flutter, supraventricular tachycardia, ventricular tachycardia, pauses, atrioventricular block, ventricular fibrillation, bigeminy, trigeminy, ventricular ectopic beats, bradycardia, tachycardia, changes in the morphology of at least one ECG signal, any combination thereof, and / or the like) as described herein. Additionally or alternatively, the temporal data may include at least one of a start time, a time interval, any combination thereof, and / or the like. In some embodiments, the gateway device 229a (e.g., its processor) may determine, based on the detected time data, at least one ECG signal portion associated with the detected time data that corresponds to a predetermined rhythm change in the ECG signal.

[0109] In some embodiments, an ECG signal portion may include a duration greater than or equal to 15 seconds and less than or equal to 120 seconds (e.g., based on a predetermined time interval, a buffer size for ECG signal samples, and / or the like). For example, the ECG signal portion may have a duration greater than or equal to 15 seconds and less than or equal to 60 seconds. In some embodiments, an ECG signal portion may include a duration between 15 seconds and 180 seconds. In some embodiments, an ECG signal portion may include a duration between 15 seconds and 240 seconds. In some embodiments, an ECG signal portion may include a duration between 15 seconds and 480 seconds. In some embodiments, an ECG signal portion may include a duration between 15 seconds and 1000 seconds. In some embodiments, the duration of an ECG signal portion may be specified by user input via a user interface control (e.g., cardiac monitoring device 202a, gateway device 229a, and / or the like).

[0110] In some embodiments, the ECG signal comprises a plurality of ECG signal samples. For example, the ECG signal samples may be sampled at a rate between 10 Hz and 1000 Hz (e.g., greater than or equal to 10 Hz and less than or equal to 1000 Hz), between 100 Hz and 500 Hz (e.g., greater than or equal to 100 Hz and less than or equal to 500 Hz), and / or the like.

[0111] In some embodiments, the ECG channel may include multiple ECG channels (e.g., channels of a standard 12-lead ECG system, a subset thereof, and / or the like). Additionally or alternatively, the ECG signal may include at least one individual ECG signal associated with each individual ECG channel. In some embodiments, the multiple ECG channels may include a first ECG channel and a second ECG channel. Additionally or alternatively, the ECG signal may include a first individual ECG signal associated with the first ECG channel and a second individual ECG signal associated with the second ECG channel. In some embodiments, the first individual ECG signal may be substantially orthogonal to the second individual ECG signal. For example, the first ECG channel may be associated with a first electrode positioned proximate the front of the patient and a second electrode positioned proximate the patient's back (e.g., a front-to-back (FB) channel). Additionally or alternatively, the second ECG channel may be associated with a third electrode positioned proximate a first side of the patient and a fourth electrode positioned proximate a second side of the patient (e.g., a side-to-side (SS) channel). Additionally or alternatively, the FB channel may be orthogonal to the SS channel (e.g., a first hypothesized line connecting the first and second electrodes is substantially (e.g., approximately and / or similarly) orthogonal (e.g., perpendicular and / or similar) to a second hypothesized line connecting the third and fourth electrodes, and the second hypothesized line may include a component (e.g., a vector component that may be calculated and / or obtained by the same, a vector projection, and / or the like) that is orthogonal to the first hypothesized line, and / or the like).

[0112] In some embodiments, the neural network of the rhythm change classifier (e.g., cardiac monitoring device 202a and / or gateway device 229a) may include multiple Siamese branches. Additionally or alternatively, each individual Siamese branch may be associated with an individual ECG channel (e.g., of the multiple ECG channels). In some embodiments, the neural network of the rhythm change classifier may further include at least one additional layer connected to the multiple Siamese branches. In some non-limiting embodiments, each Siamese branch of the multiple Siamese branches may include multiple convolutional layers. Additionally or alternatively, the dimensions of each of the multiple convolutional layers of each individual Siamese branch may be the same as the dimensions of each of the multiple convolutional layers of each other Siamese branch (e.g., the convolutional layers of all Siamese branches may have the same dimensions).

[0113] In some embodiments, cardiac monitoring device 202a and / or gateway device 229a (e.g., its processor) may further detect the predetermined rhythm change based on the at least one ECG signal (e.g., using a trained rhythm change classifier). In some embodiments, cardiac monitoring device 202a may further include at least one sensor and associated sensor circuitry that senses the patient's non-ECG biometric data (which, in some embodiments, may be communicated to gateway device 229a). Additionally or alternatively, detecting the predetermined rhythm change may be further based on the patient's non-ECG biometric data (e.g., the patient's non-ECG biometric data may be input into a neural network of the rhythm change classifier, combined with the output of the rhythm change classifier, and / or the like). In some embodiments, the at least one sensor may include at least one of an accelerometer, a heart sound detector, a receiver for electromagnetic waves (e.g., RF) (e.g., an antenna and / or the like), any combination thereof, and / or the like. Additionally or alternatively, the non-ECG biometric data may include at least one of acceleration data, heart sound data, electromagnetic waves (e.g., RF) scattered and / or reflected from internal tissue, any combination thereof, and / or the like.

[0114] In some embodiments, detecting the predetermined rhythm change may be further based on at least one baseline ECG signal portion of the patient. For example, the baseline ECG signal portion may be obtained (e.g., measured, recorded, stored, and / or the like) using high-precision ECG measurements (e.g., from an ECG device separate from cardiac monitoring device 202a, such as a 12-lead ECG sensor and / or the like) during a patient visit to a clinic. Additionally or alternatively, the baseline ECG signal portion may be obtained (e.g., measured, recorded, stored, and / or the like) using cardiac monitoring device 202a (and / or gateway device 229a) during known and / or expected rest periods (e.g., overnight hours and / or similar times). Additionally or alternatively, the baseline ECG signal portion may be obtained (e.g., measured, recorded, stored, and / or the like) using cardiac monitoring device 202a (and / or gateway device 229a) during normal sinus rhythm periods.

[0115] In some embodiments, detecting the predetermined rhythm change may be further based on at least one calibration measurement of the patient. For example, the calibration measurement may be based on at least one second ECG signal from second surface ECG activity sensed by a second plurality of ECG electrodes, which may be independent of the plurality of ECG electrodes of the (external) cardiac monitoring device 202a. For example, the calibration measurement may be obtained (e.g., measured, recorded, stored, and / or the like) using high-precision ECG measurements (e.g., from an ECG device separate from the cardiac monitoring device 202a, such as a 12-lead ECG sensor) during a patient visit to a clinic.

[0116] In some embodiments, detecting the predetermined rhythm change may be further based on at least one reference vector of the patient. For example, the reference vector may include a vector determined (e.g., calculated and / or obtained by the like) based on a set of multiple past measurements (e.g., past ECG signal segments, and / or the like). In some embodiments, the cardiac monitoring device 202a (and / or gateway device 229a) may use a reference extraction neural network to determine the reference vector based on past measurements (e.g., past ECG signal segments, and / or the like).

[0117] In some embodiments, detecting the predetermined rhythm change may be further based on at least one previous ECG signal portion. For example, at least one vector may be determined (e.g., calculated, and / or the like) based on a set of multiple previous ECG signal portions (e.g., from at least one previous time period). In some embodiments, multiple vectors may be determined (e.g., calculated, and / or the like) based on a set of multiple previous ECG signal portions from each of multiple previous time periods (e.g., a previous day, a previous week, a previous month, and / or the like). In some embodiments, each vector may be determined (e.g., by the cardiac monitoring device 202a and / or the gateway device 229a) using a reference extraction neural network based on past measurements (e.g., past ECG signal portions, and / or the like).

[0118] In some embodiments, gateway device 229a may facilitate communication between cardiac monitoring device 202a and remote computer system 204a, as described herein. For example, remote computer system 204a may communicate a trained rhythm change classifier to gateway device 229a. Additionally or alternatively, gateway device 229a may store the trained rhythm change classifier and / or communicate the trained rhythm change classifier to cardiac monitoring device 202a. In some embodiments, gateway device 229a may store the trained rhythm change classifier (and / or its corresponding weights). Additionally or alternatively, gateway device 229a may use the trained rhythm change classifier to perform rhythm change classification, as described herein. For example, the gateway device 229a may receive (e.g., continuously, periodically, and / or the like) ECG signals from the cardiac monitoring device 202a, process the ECG signals from the cardiac monitoring device 202a based on a trained rhythm change classifier, and / or communicate (e.g., transmit, and / or the like) determined / identified ECG signal portions that correspond to predetermined rhythm changes to the remote computer system 204a.

[0119] In some embodiments, the processor (e.g., of the cardiac monitoring device 202a) may further determine (e.g., using a rhythm change classifier) ​​a confidence score associated with a predetermined rhythm change based on the at least one ECG signal. For example, the output of at least one neural network of the rhythm change classifier may include a confidence score (e.g., a probability that the determined ECG signal portion includes a predetermined rhythm change, and / or the like). In some embodiments, the rhythm change classifier may include multiple neural networks, each of which may output a predicted value associated with at least one predetermined rhythm change. Additionally or alternatively, such outputs may be combined (e.g., aggregated and / or the like) and / or a confidence score may be calculated based on such outputs.

[0120] 2A , at 234a, cardiac monitoring device 202a and / or gateway device 229a (e.g., a processor thereof) may communicate (e.g., transmit, and / or the like) the determined ECG signal portions to remote computer system 204a as described herein. Additionally or alternatively, cardiac monitoring device 202a and / or gateway device 229a (e.g., a processor thereof) may detect and / or communicate indicia (e.g., flags, indicators, confidence scores, marks, metadata, time data, and / or the like) associated with predetermined rhythm changes detected (e.g., identified and / or obtained by the like) in the ECG signal portions.

[0121] In some embodiments, the processor (e.g., of cardiac monitoring device 202a and / or gateway device 229a) may further communicate (e.g., transmit and / or the like) at least one second ECG signal portion of the ECG signal to remote computer system 204a. Additionally or alternatively, the second ECG signal portion may be independent of the detected time data corresponding to the predetermined rhythm change in the ECG signal. In some embodiments, the processor (e.g., of cardiac monitoring device 202a and / or gateway device 229a) may randomly determine the second ECG signal portion (e.g., randomly sample the ECG signal to determine the second ECG signal portion). In some embodiments, the processor (e.g., of cardiac monitoring device 202a and / or gateway device 229a) may use a rhythm change classifier to determine a first confidence score associated with the predetermined rhythm change based on the ECG signal, and the first confidence score may be above a first threshold. Additionally or alternatively, the processor (e.g., of cardiac monitoring device 202a and / or gateway device 229a) may use a rhythm change classifier to detect second time data corresponding to a potential rhythm change in the ECG signal (e.g., the second time data may include at least one of a second start time, a second time interval, any combination thereof, and / or the like). Additionally or alternatively, the processor (e.g., of cardiac monitoring device 202a and / or gateway device 229a) may use a rhythm change classifier to determine a second confidence score associated with the potential rhythm change based on the ECG signal, where the second confidence score may be below a first threshold and above a second threshold. Additionally or alternatively, the processor (e.g., of cardiac monitoring device 202a and / or gateway device 229a) may determine a second ECG signal portion associated with the detected second time data that corresponds to a potential rhythm change in the ECG signal based on the detected second time data.

[0122] In some embodiments, gateway device 229a may facilitate communication between cardiac monitoring device 202a and remote computer system 204a, as described herein. For example, transmitting the determined ECG signal portions to remote computer system 204a may include cardiac monitoring device 202a communicating (e.g., transmitting, and / or the like) the determined ECG signal portions to gateway device 229a. Additionally or alternatively, gateway device 229a may receive the determined ECG signal portions from cardiac monitoring device 202a and / or communicate the determined ECG signal portions to remote server 204a.

[0123] In some embodiments, the remote computing system 204a may receive the determined ECG signal portion (e.g., from the cardiac monitoring device 202a and / or the gateway device 229a). Additionally or alternatively, the remote computing system 204a may analyze the determined ECG signal portion to classify an arrhythmia type for the rhythm change in the ECG signal. In some embodiments, the arrhythmia type may include at least one of a change in heart rate, atrial fibrillation, flutter, supraventricular tachycardia, ventricular tachycardia, pauses, atrioventricular block, ventricular fibrillation, bigeminy, trigeminy, ventricular ectopic beats, bradycardia, tachycardia, a change in the morphology of at least one ECG signal, any combination thereof, and / or the like.

[0124] In some embodiments, the remote computer system 204a may include an arrhythmia type classifier (e.g., having at least one (second) neural network trained based on a (second) historical collection of a (second) plurality of ECG signal portions with known arrhythmia type information) as described herein. Additionally or alternatively, analyzing the at least one determined ECG signal portion may include the remote computer system 204a detecting, using the arrhythmia type classifier, an arrhythmia type associated with the rhythm change based on the determined ECG signal portion.

[0125] In some embodiments, the onset of bradycardia may include a patient's heart rate decreasing below a first threshold value for the patient. For example, the first threshold value may include a value of at least 20 beats per minute (BPM) and up to 100 BPM, a value of at least 30 BPM and up to 100 BPM, and / or similar values, calculated over a predetermined interval (e.g., 16 beats, and / or similar number of beats). For example, a default first threshold value may include 40 BPM, and the first threshold value may be adjusted for each individual patient (e.g., by a prescriber, treating physician, and / or the like). In some embodiments, the offset of bradycardia may include a patient's heart rate increasing above a second threshold value for the patient. For example, the second threshold value may include a value of at least 20 beats per minute (BPM) and up to 100 BPM, calculated over a predetermined interval (e.g., 16 beats, and / or similar number of beats). For example, the default second threshold may include 45 BPM, and the second threshold may be adjusted for each individual patient (e.g., by a prescriber, treating physician, and / or the like). In some embodiments, the patient must maintain bradycardia for a first selected period before a rhythm change is reported (e.g., before an ECG signal portion is classified as bradycardia, before any message associated with the ECG signal portion is communicated, and / or the like). For example, the first selected period may include a value of at least 0 seconds and at most 600 seconds, at least 15 seconds and at most 600 seconds, and / or the like. For example, the default first period may include 30 seconds, and the first period may be adjusted for each individual patient (e.g., by a prescriber, treating physician, and / or the like).

[0126] In some embodiments, the onset of tachycardia may include the patient's heart rate increasing above a third threshold for the patient. For example, the third threshold may include a value of at least 100 BPM and a maximum of 250 BPM, a value of at least 100 BPM and a maximum of 249 BPM, and / or similar values, and may be calculated over a predetermined interval (e.g., 16 beats, and / or similar number of beats). For example, a default third threshold may include 1300 BPM, and the third threshold may be adjusted for each individual patient (e.g., by a prescriber, treating physician, and / or the like). In some embodiments, the offset of tachycardia may include the patient's heart rate decreasing below a fourth threshold for the patient. For example, the fourth threshold may include a value of at least 100 BPM and a maximum of 250 BPM, a value of at least 100 BPM and a maximum of 249 BPM, and / or similar values, and may be calculated over a predetermined interval (e.g., 16 beats, and / or similar number of beats). For example, a default fourth threshold may include 110 BPM, and the fourth threshold may be adjusted for each individual patient (e.g., by a prescriber, treating physician, and / or the like). In some embodiments, the patient must maintain a tachycardia for a second selected period before a rhythm change is reported (e.g., before an ECG signal portion is classified as tachycardia, before any message associated with the ECG signal portion is communicated, and / or the like). For example, the second selected period may include a value of at least 0 seconds and at most 600 seconds, at least 15 seconds and at most 600 seconds, and / or the like. For example, the default second period may include 30 seconds, and the second period may be adjusted for each individual patient (e.g., by a prescriber, treating physician, and / or the like).

[0127] In some embodiments, the onset of atrial fibrillation may include the patient's trembling or irregular heartbeat. In some embodiments, the patient must maintain atrial fibrillation for a third selected period before a rhythm change is reported (e.g., before an ECG signal portion is classified as atrial fibrillation, before any message associated with the ECG signal portion is communicated, and / or the like). For example, the third selected period may include at least 0 minutes and up to 60 minutes, and / or the like. For example, a default third period may include 5 minutes, and the third period may be adjusted for each individual patient (e.g., by a prescriber, treating physician, and / or the like).

[0128] In some embodiments, the onset of a cardiac pause may include a prolonged RR interval, signifying an interruption of ventricular depolarization. In some embodiments, the patient must maintain a cardiac pause for a fourth selected period before a rhythm change is reported (e.g., before an ECG signal portion is classified as a cardiac pause, before any message associated with the ECG signal portion is communicated, and / or the like). For example, the fourth selected period may include at least 1500 milliseconds (ms) and up to 15000 ms, and / or the like. For example, a default fourth period may include 3000 ms, and the fourth period may be adjusted for each individual patient (e.g., by a prescriber, treating physician, and / or the like).

[0129] In some embodiments, the onset of a bradycardia rate change may include a decrease in the patient's heart rate below a first threshold by a first predetermined value. In some embodiments, a rhythm change may be reported (e.g., classifying an ECG signal portion as a bradycardia rate change, communicating a message associated with the ECG signal portion, and / or the like) whenever the patient's heart rate decreases by at least an integer multiple of the first predetermined value. For example, the first predetermined value may include a value of at least 0 BPM and up to 100 BPM, and / or similar values. For example, a default first predetermined value may be 5 BPM, and the first predetermined value may be adjusted for each individual patient (e.g., by a prescriber, treating physician, and / or the like).

[0130] In some embodiments, the onset of a tachycardia rate change may include an increase in the patient's heart rate above a third threshold by at least a second predetermined value. In some embodiments, a rhythm change may be reported (e.g., classifying an ECG signal portion as a tachycardia rate change, communicating a message associated with the ECG signal portion, and / or the like) whenever the patient's heart rate increases by at least an integer multiple of the second predetermined value. For example, the second predetermined value may include a value of at least 0 BPM and at most 250 BPM, and / or similar values. For example, a default second predetermined value may be 10 BPM, and the second predetermined value may be adjusted for each individual patient (e.g., by a prescriber, treating physician, and / or the like).

[0131] In some embodiments, the remote computer system 204a may analyze the determined ECG signal portions to identify at least one arrhythmia associated with the rhythm change in the at least one ECG signal. In some embodiments, the arrhythmia may be one or more rare arrhythmias for which the remote computer system 204a may have been previously trained. Additionally or alternatively, the remote computer system 204a may use any suitable signal processing technique (e.g., separate from or including an arrhythmia type classifier as described herein) to identify the rare arrhythmia. For example, a small portion of the historical collection of such ECG signal portions may be associated with rare arrhythmias for which the arrhythmia type classifier may not have been adequately trained to classify such rare arrhythmias. Additionally or alternatively, the remote computer system 204a may use signal processing techniques, predetermined rules, and / or the like to identify such rare arrhythmias.

[0132] In some embodiments, the determined ECG signal portion may include multiple determined ECG signal portions. Additionally or alternatively, the remote computer system 204a may receive multiple determined ECG signal portions from the cardiac monitoring device 202a and / or the gateway device 229a, as described herein. In some embodiments, the remote computer system 204a may analyze each individual determined ECG signal portion to classify an individual class for each individual determined ECG signal portion. Additionally or alternatively, the class of at least two individual determined ECG signal portions may include a first class (e.g., at least two ECG signal portions may belong to the same class / group). In some embodiments, analyzing each individual determined ECG signal portion may include determining (e.g., calculating, and / or the like) a vector for each individual determined ECG signal portion. Additionally or alternatively, the vectors may be classified into classes (e.g., groups, clusters, and / or the like) based on similarities between individual vectors (e.g., vector distance, clustering, and / or the like).

[0133] 2A, the remote computer system 204a may communicate (e.g., transmit and / or perform the like) at 236a at least one message related to the determined ECG signal portion and / or arrhythmia type related to the rhythm change, as described herein. For example, a message may be communicated from the remote computer system 204a to the technician device 208a, as described herein.

[0134] In some embodiments, the remote computer system 204a may transmit to the technician device 208a at least one message related to the second ECG signal portion (e.g., a randomly determined second ECG signal portion as described herein, a second ECG signal portion determined to have a confidence score below a first threshold and above a second threshold, and / or the like).

[0135] In some embodiments, the remote computer system 204a may transmit to the technician device 208a at least one message related to the at least two individually determined ECG signal portions and the first class. For example, by grouping similar ECG signal portions together in the same class (e.g., the first class), a technician using the technician device 208a may review the at least two individually determined ECG signal portions more efficiently, quickly, and / or similarly.

[0136] 2A , at 238a, the remote computer system 204a may receive annotation data associated with at least one annotation from the technician device 208a, as described herein. For example, the annotation data may be communicated from the technician device 208a to the remote computer system 204a, as described herein. Additionally or alternatively, portions of the ECG signal associated with such annotations may be communicated using the annotation data, as described herein.

[0137] In some embodiments, the remote computer system 204a may receive (e.g., from the technician device 208a) annotation data related to at least one annotation for a second ECG signal portion (e.g., a randomly determined second ECG signal portion, a second ECG signal portion determined to have a confidence score below a first threshold and above a second threshold, and / or the like, as described herein). In some embodiments, the remote computer system 204a may retrain the rhythm change classifier (and / or train an updated rhythm change classifier) ​​based on a historical collection of multiple ECG signal portions with known rhythm change information, second ECG signal portions, and associated annotation data. In some embodiments, the remote computer system 204a may retrain the arrhythmia type classifier based on a historical collection of multiple ECG signal portions with known rhythm change information, second ECG signal portions, and associated annotation data.

[0138] 2A, at 240a, the remote computer system 204a may communicate (e.g., send, write, and / or the like) annotation data to the data repository 206a, as described herein. Additionally or alternatively, the ECG signal portion associated with the annotation may be communicated with the annotation data, as described herein.

[0139] In some embodiments, annotation data and ECG signal portions associated with such annotations may be added to a historical collection of multiple ECG signal portions. For example, the annotation data may be saved as known rhythm change information and / or known arrhythmia type information for the ECG signal portion associated with the annotation. In some embodiments, the remote computer system 204a may retrain the rhythm change classifier (and / or train an updated rhythm change classifier) ​​based on a historical collection of multiple ECG signal portions with known rhythm change information (which may include ECG signal portions and / or annotation data associated therewith). In some embodiments, the remote computer system 204a may retrain the arrhythmia type classifier based on a historical collection of multiple ECG signal portions with known arrhythmia type information (which may include ECG signal portions and / or annotation data associated therewith).

[0140] In some embodiments, the remote computer system 204a may add annotation data related to at least one annotation for a second ECG signal portion (e.g., a randomly determined second ECG signal portion, a second ECG signal portion determined to have a confidence score below a first threshold and above a second threshold, and / or the like, as described herein) to the historical collection of multiple ECG signal portions in the data repository 206a. In some embodiments, the remote computer system 204a may retrain the rhythm change classifier (and / or train an updated rhythm change classifier) ​​based on the historical collection of multiple ECG signal portions with known rhythm change information (which may include the second ECG signal portion and / or annotation data associated therewith). In some embodiments, the remote computer system 204a may retrain the arrhythmia type classifier based on the historical collection of multiple ECG signal portions with known arrhythmia type information (which may include the second ECG signal portion and / or annotation data associated therewith).

[0141] 2A , at 242a, remote computer system 204a may communicate the retrained rhythm change classifier (and / or trained, updated rhythm change classifier) ​​to cardiac monitoring device 202a and / or gateway device 229a as described herein. Additionally or alternatively, a copy of the retrained rhythm change classifier (and / or trained, updated rhythm change classifier) ​​may be downloaded from remote computer system 204a and / or installed (e.g., uploaded to, written to, configured on, and / or the like) on at least one non-transitory computer-readable medium (e.g., memory, programmable circuit board, FPGA, integrated circuit, any combination thereof, and / or the like) that may be installed on and / or part of cardiac monitoring device 202a and / or gateway device 229a.

[0142] Reference is now made to FIG. 2B , which illustrates an exemplary block diagram of a system architecture 200b for arrhythmia monitoring according to some embodiments. In addition to system components, the system architecture 200b also illustrates data flow between the system components. As shown in FIG. 2B , the system architecture 200b may include a cardiac monitoring device 202b, a remote computer system 204b, a data repository 206b, a technician device 208b, and / or a gateway device 229b. In some embodiments, the cardiac monitoring device 202b may be the same as or similar to the cardiac monitoring device 102, the cardiac monitoring device 202a, and / or the like. In some embodiments, the remote computer system 204b may be the same as or similar to the remote computer system 104 (e.g., one or more devices of the remote computer system 104), the remote computer system 204a (e.g., one or more devices of the remote computer system 204a), and / or the like. In some embodiments, data repository 206b may be the same as or similar to data repository 106 (e.g., one or more devices of data repository 106), data repository 206a (e.g., one or more devices of data repository 206a), and / or the like. In some embodiments, technician device 208b may be the same as or similar to technician device 108, technician device 208a, and / or the like. In some embodiments, gateway device 229b may be the same as or similar to gateway device 129.

[0143] 2B , at 230b, the remote computer system 204b may receive (e.g., retrieve, search, send requests and / or queries to the data repository 206b for communication, and / or the like) a historical collection of multiple ECG signal portions and information associated therewith (e.g., known rhythm variation information, known arrhythmia type information, and / or the like), for example, from the data repository 206b. In some embodiments, the remote computer system 204b may train at least one neural network of at least one classifier (e.g., an arrhythmia type classifier, and / or the like) based on the historical collection of multiple ECG signal portions and information associated therewith (e.g., known arrhythmia type information, and / or the like), as described herein.

[0144] In some embodiments, the arrhythmia type classifier may comprise at least one neural network (e.g., at least one second neural network) as described herein. Additionally or alternatively, the at least one (second) neural network may comprise at least one of a deep neural network, a convolutional neural network, a recurrent neural network, an attention network, a fully connected neural network, any combination thereof, and / or the like. For example, the neural network may include at least one convolutional neural network having multiple convolutional layers. In some embodiments, the convolutional neural network may have 5 to 40 convolutional layers (e.g., at least 5 convolutional layers and up to 40 convolutional layers). For example, the convolutional neural network may have 7 to 10 convolutional layers (e.g., at least 7 and no more than 10 convolutional layers). In some embodiments, each convolutional layer may include at least one convolutional node (e.g., multiple convolutional nodes). In some embodiments, the convolutional neural network may further have an input layer and an output layer. For example, the ECG signal may include a plurality of ECG signal samples. Additionally or alternatively, the input layer may include at least one node for each ECG signal sample of the plurality of ECG signal samples (or a subset of the plurality of ECG signal samples, e.g., associated with a predetermined time period, a buffer size for the ECG signal samples, and / or the like). Additionally or alternatively, the input layer may further include at least one input for non-ECG biometric data associated with a sensor (e.g., of the cardiac monitoring device 202 and / or the like), as described herein. In some embodiments, the output of the output layer may include an indication of the time data corresponding to the arrhythmia type. Additionally or alternatively, the output of the output layer may include a confidence score, a likelihood score, and / or the like, as described herein.In some embodiments, the neural network may include multiple Siamese branches (eg, each individual Siamese branch associated with an individual ECG channel) as described herein.

[0145] In some embodiments, the remote computer system 204b may train the arrhythmia type classifier by using the arrhythmia type classifier to generate a predicted arrhythmia type for each individual ECG signal portion of a historical collection of the plurality of ECG signal portions (or a second plurality of ECG signal portions), determining at least one error value based on the predicted arrhythmia type and known arrhythmia type information (e.g., individual annotations associated with known arrhythmia types for each individual ECG signal portion), and updating the arrhythmia type classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a contrast loss.

[0146] 2B , at 232b, remote computer system 204b may communicate the historical collection of multiple ECG signal portions and information associated therewith (e.g., known rhythm change information, and / or the like) to cardiac monitoring device 202b and / or gateway device 229b, as described herein. Additionally or alternatively, cardiac monitoring device 202b and / or gateway device 229b may train a rhythm change classifier, which may be implemented by at least one non-transitory computer-readable medium (e.g., memory, programmable circuit board, field programmable gate array (FPGA), integrated circuit, any combination thereof, and / or the like) that may be installed on and / or part of cardiac monitoring device 202b and / or gateway device 229b, as described herein.

[0147] In some embodiments, cardiac monitoring device 202b may be an external cardiac monitoring device for a patient, as described herein. For example, (external) cardiac monitoring device 202b may include a plurality of ECG electrodes that sense the patient's surface ECG activity. Additionally or alternatively, (external) cardiac monitoring device 202b may include ECG processing circuitry that processes the patient's surface ECG activity to provide at least one ECG signal for the patient on at least one ECG channel.

[0148] In some embodiments, the (external) cardiac monitoring device 202b may include a non-transitory computer-readable medium (e.g., a memory, a programmable circuit board, a field-programmable gate array, any combination thereof, and / or the like) having (e.g., implementing, embodying, storing, and / or the like) a rhythm change classifier as described herein (which may include, e.g., at least one neural network). Additionally or alternatively, the (external) cardiac monitoring device 202b may include at least one processor operably connected to the ECG channel and the non-transitory computer-readable medium.

[0149] In some embodiments, gateway device 229b may comprise a non-transitory computer-readable medium (e.g., memory, a programmable circuit board, a field-programmable gate array, any combination thereof, and / or the like) having (e.g., implementing, embodying, storing, and / or the like) a rhythm change classifier as described herein (which may, e.g., include at least one neural network). Additionally or alternatively, gateway device 229b may comprise at least one processor operably coupled to the non-transitory computer-readable medium.

[0150] In some embodiments, the rhythm change classifier may comprise at least one neural network, as described herein. Additionally or alternatively, the at least one neural network may comprise at least one of a convolutional neural network, a recurrent neural network, an attention network, a fully-connected neural network, any combination thereof, and / or the like, as described herein. For example, the neural network may include at least one convolutional neural network having multiple convolutional layers. In some embodiments, the convolutional neural network may have 5 to 40 convolutional layers (e.g., at least 5 convolutional layers and up to 40 convolutional layers). For example, the convolutional neural network may have 7 to 10 convolutional layers (e.g., at least 7 and no more than 10 convolutional layers). In some embodiments, each convolutional layer may include at least one convolutional node (e.g., multiple convolutional nodes). In some embodiments, the convolutional neural network may further include an input layer and an output layer. For example, the ECG signal may include a plurality of ECG signal samples. Additionally or alternatively, the input layer may include at least one node for each ECG signal sample of the plurality of ECG signal samples (or a subset of the plurality of ECG signal samples, e.g., associated with a predetermined time period, a buffer size for the ECG signal samples, and / or the like). Additionally or alternatively, the input layer may further include at least one input for non-ECG biometric data associated with each sensor of the cardiac monitoring device 202b, as described herein. In some embodiments, the output of the output layer may include an indication of time data corresponding to the rhythm change. Additionally or alternatively, the output of the output layer may include a confidence score, as described herein. In some embodiments, the neural network may include a plurality of Siamese branches (e.g., each individual Siamese branch associated with an individual ECG channel), as described herein.

[0151] In some embodiments, cardiac monitoring device 202b and / or gateway device 229b may train the rhythm change classifier by using the rhythm change classifier to generate predicted rhythm change information (e.g., data related to predicted rhythm changes (e.g., probabilities, confidence scores, and / or the like), data related to the absence of predicted rhythm changes (e.g., probabilities, confidence scores, and / or the like), and / or the like) for each ECG signal portion of a historical collection of multiple ECG signal portions, determining at least one error value based on the predicted rhythm change information and the known rhythm change information, and updating the rhythm change classifier based on the error value (e.g., updating its weight, and / or the like) (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss.

[0152] In some embodiments, the historical collection of multiple ECG signal portions may include a first ECG signal portion associated with a first time and a second ECG signal portion associated with a second time after the first time. Additionally or alternatively, cardiac monitoring device 202b and / or gateway device 229b may train the rhythm change classifier by generating, using the rhythm change classifier, a predicted ECG signal portion associated with the second time based on the first ECG signal portion, determining at least one error value based on the predicted ECG signal portion and the second ECG signal portion, and updating the rhythm change classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss.

[0153] In some embodiments, the historical collection of the plurality of ECG signal portions may include a first ECG signal portion associated with a first time and a second ECG signal portion associated with a second time. Additionally or alternatively, cardiac monitoring device 202b and / or gateway device 229b may train a rhythm change classifier by: generating, with a rhythm change classifier, a predicted time associated with the second ECG signal portion based on the first ECG signal portion and the second ECG signal; determining at least one error value based on the predicted time and the second time; and updating the rhythm change classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss.

[0154] In some embodiments, there may be an insufficient number of ECG signal portions in a historical collection of multiple ECG signal portions with known rhythm change information to train a rhythm change classifier to perform a desired task (e.g., detecting and / or identifying at least one predetermined rhythm change). Additionally or alternatively, there may be sufficient data (e.g., past ECG signal portions and / or the like) to train the rhythm change classifier to perform another task (e.g., that may be related in some way to the target task). In some embodiments, the rhythm change classifier may be trained to perform another task (e.g., counting R peaks based on ECG signals, determining heart rate, and / or the like). Additionally or alternatively, the rhythm change classifier may be applied to subsequently perform the target task. For example, in some embodiments, the rhythm change classifier may be retrained using a limited amount of ECG signal portions in a historical collection of multiple ECG signal portions with known rhythm change information and / or the like. Additionally or alternatively, the rhythm change classifier may be used to perform another task (e.g., counting R peaks based on ECG signals, determining a heart rate, and / or the like), and its output may be applied to the target task. For example, a processor (e.g., of cardiac monitoring device 202b and / or gateway device 229b) may use the rhythm change classifier to detect at least one peak count or heart rate based on at least one ECG signal. Additionally or alternatively, the processor (e.g., of cardiac monitoring device 202b and / or gateway device 229b) may determine that the detected at least one peak count or heart rate exceeds a first threshold value for the patient (e.g., a tachycardia onset threshold) or is below a second threshold value for the patient (e.g., a bradycardia onset threshold) (e.g., where the second threshold value for the patient may be less than the first threshold value for the patient).Additionally or alternatively, the processor (e.g., of the cardiac monitoring device 202b and / or the gateway device 229b) may detect a predetermined rhythm change based on at least one peak number or heart rate exceeding a first threshold value for the patient (e.g., a tachycardia onset threshold) or falling below a second threshold value for the patient (e.g., a bradycardia onset threshold).

[0155] In some embodiments, there may be an insufficient number of ECG signal portions associated with (e.g., sensed from and / or related by the same) the plurality of ECG electrodes of cardiac monitoring device 202b in a historical collection of ECG signal portions with known rhythm change information to train a rhythm change classifier for ECG signals received from the plurality of ECG electrodes of cardiac monitoring device 202a. Additionally or alternatively, there may be sufficient data (e.g., past ECG signal portions, and / or the like) associated with (e.g., sensed from and / or related by the same) the second plurality of ECG electrodes independent from the plurality of ECG electrodes of cardiac monitoring device 202a (e.g., electrodes of an ECG device separate from cardiac monitoring device 202b, such as a 12-lead ECG sensor, a separate external and / or wearable cardiac monitoring device, and / or the like) to train a rhythm change classifier based on the second plurality of ECG electrodes. In some embodiments, the rhythm change classifier may be trained based on ECG signal portions associated with the second plurality of ECG electrodes (e.g., sensed from the second plurality of ECG electrodes and / or the like). Additionally or alternatively, the rhythm change classifier may then be applied to detect predetermined rhythm changes based on the plurality of ECG electrodes of cardiac monitoring device 202b. In some embodiments, cardiac monitoring device 202b and / or gateway device 229b may determine (e.g., calculate and / or perform the like) a transformation (e.g., vector projection and / or the like) of the ECG signal portions associated with the second plurality of ECG electrodes onto the plurality of ECG electrodes of cardiac monitoring device 202b and use the transformation of the ECG signal portions to train the rhythm change classifier as if the ECG signal portions were associated with (e.g., sensed from and / or the like) the plurality of ECG electrodes of cardiac monitoring device 202b.

[0156] In some embodiments, cardiac monitoring device 202b (e.g., its processor) may receive ECG signals via an ECG channel as described herein. Additionally or alternatively, cardiac monitoring device 202b (e.g., its processor) may detect temporal data corresponding to predetermined rhythm changes in the ECG signal using a rhythm change classifier as described herein. In some embodiments, cardiac monitoring device 202a (e.g., its processor) may determine, based on the detected temporal data, at least one ECG signal portion associated with the detected temporal data that corresponds to the predetermined rhythm change in the ECG signal as described herein.

[0157] In some embodiments, cardiac monitoring device 202b (e.g., its processor) may receive ECG signals via an ECG channel as described herein. Additionally or alternatively, cardiac monitoring device 202b (e.g., its processor) may transmit ECG signals to gateway device 229b. In some embodiments, gateway device 229b may detect temporal data corresponding to predetermined rhythm changes in the ECG signal using a rhythm change classifier as described herein. In some embodiments, gateway device 229b (e.g., its processor) may determine, based on the detected temporal data, at least one ECG signal portion associated with the detected temporal data that corresponds to the predetermined rhythm change in the ECG signal as described herein.

[0158] In some embodiments, the neural network of the rhythm change classifier (eg, cardiac monitoring device 202b and / or gateway device 229b) may include multiple Siamese branches, as described herein.

[0159] In some embodiments, cardiac monitoring device 202b and / or gateway device 229b (e.g., a processor thereof) may further detect the predetermined rhythm change based on at least one ECG signal (e.g., using a trained rhythm change classifier) ​​as described herein. In some embodiments, cardiac monitoring device 202b may further include at least one sensor and associated sensor circuitry that senses the patient's non-ECG biometric data (which, in some embodiments, may be communicated to gateway device 229b) as described herein. Additionally or alternatively, detecting the predetermined rhythm change may be further based on the patient's non-ECG biometric data as described herein (e.g., the patient's non-ECG biometric data may be input into a neural network of the rhythm change classifier, combined with the output of the rhythm change classifier, and / or the like).

[0160] In some embodiments, detecting the predetermined rhythm change may be further based on at least one baseline ECG signal portion of the patient, as described herein.

[0161] In some embodiments, detecting the predetermined rhythm change may be further based on at least one calibration measurement of the patient, as described herein.

[0162] In some embodiments, detecting the predetermined rhythm change may be further based on at least one reference vector of the patient, as described herein.

[0163] In some embodiments, detecting the predetermined rhythm change may be further based on at least one previous ECG signal portion, as described herein.

[0164] In some embodiments, the gateway device 229b may facilitate communication between the cardiac monitoring device 202b and the remote computer system 204b as described herein.

[0165] In some embodiments, the processor (e.g., of the cardiac monitoring device 202b and / or the gateway device 229b) may further determine (e.g., using a rhythm change classifier) ​​a confidence score associated with a given rhythm change based on at least one ECG signal, as described herein.

[0166] 2B , at 234b, cardiac monitoring device 202b and / or gateway device 229b (e.g., a processor thereof) may communicate (e.g., transmit and / or the like) the determined ECG signal portions to remote computer system 204b as described herein. Additionally or alternatively, cardiac monitoring device 202b and / or gateway device 229b (e.g., a processor thereof) may communicate indicia (e.g., flags, indicators, confidence scores, marks, metadata, time data, and / or the like) associated with predetermined rhythm changes detected (e.g., identified and / or obtained by the like) in the ECG signal portions as described herein.

[0167] In some embodiments, the processor (e.g., of cardiac monitoring device 202b and / or gateway device 229b) may further communicate (e.g., transmit and / or the like) at least one second ECG signal portion of the ECG signal to remote computer system 204b, as described herein. Additionally or alternatively, the second ECG signal portion may be independent of the detected time data corresponding to the predetermined rhythm change in the ECG signal, as described herein.

[0168] In some embodiments, the gateway device 229b may facilitate communication between the cardiac monitoring device 202b and the remote computer system 204b as described herein.

[0169] In some embodiments, the remote computing system 204b may receive the determined ECG signal portions (e.g., from the cardiac monitoring device 202b and / or the gateway device 229b) as described herein. Additionally or alternatively, the remote computing system 204b may analyze the determined ECG signal portions to classify an arrhythmia type for the rhythm changes in the ECG signal as described herein. For example, the remote computer system 204b may include an arrhythmia type classifier (e.g., having at least one (second) neural network trained based on a (second) historical collection of a (second) plurality of ECG signal portions with known arrhythmia type information) as described herein.

[0170] In some embodiments, the remote computer system 204b may analyze the determined ECG signal portion to identify at least one arrhythmia associated with a rhythm change in the at least one ECG signal, as described herein. In some embodiments, the arrhythmia may be one or more rare arrhythmias for which the remote computer system 204a may have been previously trained. Additionally or alternatively, the remote computer system 204a may use any suitable signal processing technique (e.g., separate from or including an arrhythmia type classifier as described herein) to identify rare arrhythmias, as described herein.

[0171] In some embodiments, the determined ECG signal portion may include multiple determined ECG signal portions, as described herein. Additionally or alternatively, the remote computer system 204b may receive multiple determined ECG signal portions from the cardiac monitoring device 202b and / or the gateway device 229b, as described herein. In some embodiments, the remote computer system 204b may analyze each individual determined ECG signal portion to classify an individual class for each individual determined ECG signal portion, as described herein.

[0172] 2B, the remote computer system 204b may communicate (e.g., transmit and / or perform the like) at 236b at least one message related to the determined ECG signal portion and / or the arrhythmia type related to the rhythm change, as described herein. For example, a message may be communicated from the remote computer system 204b to the technician device 208b, as described herein.

[0173] In some embodiments, the remote computer system 204b may transmit at least one message related to the second ECG signal portion (e.g., a randomly determined second ECG signal portion as described herein, a second ECG signal portion determined to have a confidence score below a first threshold and above a second threshold, and / or the like) to the technician device 208b, as described herein.

[0174] In some embodiments, the remote computer system 204b may transmit to the technician device 208b at least two individually determined ECG signal portions and at least one message related to the first class, as described herein.

[0175] 2B , at 238b, the remote computer system 204b may receive annotation data associated with at least one annotation from the technician device 208a, as described herein. For example, the annotation data may be communicated from the technician device 208b to the remote computer system 204b, as described herein. Additionally or alternatively, portions of the ECG signal associated with such annotations may be communicated using the annotation data, as described herein.

[0176] In some embodiments, the remote computer system 204b may receive (e.g., from the technician device 208b) annotation data related to at least one annotation for a second ECG signal portion (e.g., a randomly determined second ECG signal portion as described herein, a second ECG signal portion determined to have a confidence score below a first threshold and above a second threshold, and / or the like).

[0177] In some embodiments, the remote computer system 204b may retrain the arrhythmia type classifier based on a historical collection of multiple ECG signal portions with known rhythm variation information, the second ECG signal portion, and associated annotation data, as described herein.

[0178] 2B, at 240b, the remote computer system 204b may communicate (e.g., send, write, and / or the like) the annotation data to the data repository 206b, as described herein. Additionally or alternatively, the ECG signal portion associated with the annotation may be communicated with the annotation data, as described herein.

[0179] In some embodiments, annotation data and ECG signal portions associated with such annotations may be added to a historical collection of multiple ECG signal portions, as described herein. In some embodiments, the remote computer system 204b may retrain the arrhythmia type classifier based on a historical collection of multiple ECG signal portions with known arrhythmia type information (which may include ECG signal portions and / or annotation data associated therewith).

[0180] In some embodiments, the remote computer system 204b may add annotation data related to at least one annotation for a second ECG signal portion (e.g., a randomly determined second ECG signal portion, a second ECG signal portion determined to have a confidence score below a first threshold and above a second threshold, and / or the like, as described herein) to the historical collection of multiple ECG signal portions in the data repository 206b. In some embodiments, the remote computer system 204b may retrain the arrhythmia type classifier based on the historical collection of multiple ECG signal portions with known arrhythmia type information (which may include the second ECG signal portion and / or annotation data associated therewith), as described herein.

[0181] 2B , at 242b, remote computer system 204b may communicate (e.g., transmit, and / or the like) annotation data associated with at least one annotation for the second ECG signal portion to cardiac monitoring device 202b and / or gateway device 229b, as described herein. Additionally or alternatively, cardiac monitoring device 202b and / or gateway device 229b (e.g., a processor thereof) may retrain the rhythm change classifier (and / or train an updated rhythm change classifier) ​​based on the known rhythm change information, the second ECG signal portion, and a historical collection of multiple ECG signal portions with associated annotation data, as described herein.

[0182] Reference is now made to FIG. 2C , which illustrates an exemplary block diagram of a system architecture 200c for arrhythmia monitoring according to some embodiments. In addition to system components, the system architecture 200c also illustrates data flow between the system components. As shown in FIG. 2C , the system architecture 200c may include a remote computer system 204c, a data repository 206c, a technician device 208c, and / or a supervisor device 210c. In some embodiments, the remote computer system 204c may be the same as or similar to the remote computer system 104 (e.g., one or more devices of the remote computer system 104), the remote computer system 204a (e.g., one or more devices of the remote computer system 204a), the remote computer system 204b (e.g., one or more devices of the remote computer system 204b), and / or the like. In some embodiments, data repository 206c may be the same as or similar to data repository 106 (e.g., one or more devices of data repository 106), data repository 206a (e.g., one or more devices of data repository 206a), data repository 206b (e.g., one or more devices of data repository 206b), and / or the like. In some embodiments, technician device 208c may be the same as or similar to technician device 108, technician device 208a, technician device 208b, and / or the like. In some embodiments, supervisor device 210c may be the same as or similar to supervisor device 110 and / or the like.

[0183] 2C , at 230c, the remote computer system 204c may receive (e.g., retrieve, search, send requests and / or queries to the data repository 206c for communication, and / or the like) a historical collection of multiple ECG signal portions and associated information (e.g., known arrhythmia type information, and / or the like) as described herein, for example, from the data repository 206c. In some embodiments, the remote computer system 204c may train at least one neural network of at least one classifier (e.g., an arrhythmia type classifier, and / or the like) based on the historical collection of multiple ECG signal portions and associated information (e.g., known arrhythmia type information, and / or the like) as described herein.

[0184] In some embodiments, the arrhythmia type classifier may comprise at least one neural network (e.g., at least one second neural network) as described herein. Additionally or alternatively, the at least one (second) neural network may comprise at least one of a deep neural network, a convolutional neural network, a recurrent neural network, an attention network, a fully connected neural network, any combination thereof, and / or the like. For example, the neural network may comprise at least one convolutional neural network having multiple convolutional layers. In some embodiments, the convolutional neural network may have 5 to 40 convolutional layers (e.g., at least 5 convolutional layers and up to 40 convolutional layers). For example, the convolutional neural network may have 7 to 10 convolutional layers (e.g., at least 7 and no more than 10 convolutional layers). In some embodiments, each convolutional layer may include at least one convolutional node (e.g., multiple convolutional nodes). In some embodiments, the convolutional neural network may further have an input layer and an output layer. For example, the ECG signal may include a plurality of ECG signal samples. Additionally or alternatively, the input layer may include at least one node for each ECG signal sample of the plurality of ECG signal samples (or a subset of the plurality of ECG signal samples, e.g., associated with a predetermined time period, a buffer size for the ECG signal samples, and / or the like). Additionally or alternatively, the input layer may further include at least one input for non-ECG biometric data associated with a sensor (e.g., of the cardiac monitoring device 102 and / or the like), as described herein. In some embodiments, the output of the output layer may include an indication of the time data corresponding to the arrhythmia type. Additionally or alternatively, the output of the output layer may include a confidence score, a likelihood score, and / or the like, as described herein.In some embodiments, the neural network may include multiple Siamese branches (eg, each individual Siamese branch associated with an individual ECG channel) as described herein.

[0185] In some embodiments, the remote computer system 204c may train the arrhythmia type classifier by: predicting, using the arrhythmia type classifier, a predicted arrhythmia type for each individual ECG signal portion of a historical collection of the plurality of ECG signal portions (or a second plurality of ECG signal portions); determining at least one error value based on the predicted arrhythmia type and known arrhythmia type information (e.g., individual annotations associated with known arrhythmia types for each individual ECG signal portion); and training the arrhythmia type classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a contrast loss.

[0186] In some embodiments, the known arrhythmia type information may include multiple annotations. For example, each annotation may be associated with a separate ECG signal portion of the multiple ECG signal portions. In some embodiments, the remote computer system 204c may train an arrhythmia type classifier based on the multiple ECG signals and the multiple annotations, as described herein.

[0187] In some embodiments, the multiple annotations may be from multiple technicians (e.g., multiple technician devices 208c and / or the like). Additionally or alternatively, each annotation of the multiple annotations may be associated with an individual technician of the multiple technicians and / or an individual ECG signal portion of the multiple ECG signal portions. In some embodiments, the arrhythmia type classifier may be trained separately for each technician. For example, for a first technician of the multiple technicians, the arrhythmia type classifier (e.g., remote computer system 204c) may be trained based on a subset of the multiple ECG signals and multiple annotations associated with at least one other technician of the multiple technicians that is different from the first technician (e.g., training the arrhythmia type classifier for the first technician only based on annotations from the other technicians).

[0188] In some embodiments, each annotation may be associated with one possible arrhythmia type of the individual ECG signal and / or portion thereof (e.g., a label associated with the possible arrhythmia type, a text string identifying at least one possible arrhythmia type, and / or the like).

[0189] In some embodiments, there may be an insufficient number of ECG signal portions associated with (e.g., sensed from and / or related by the same) at least one second ECG electrode in a historical collection of multiple ECG signal portions with known arrhythmia type information (e.g., annotation labels and / or the like) to train an arrhythmia type classifier for ECG signals received from the second ECG electrodes. Additionally or alternatively, there may be sufficient data (e.g., past ECG signal portions and / or the like) associated with (e.g., sensed from and / or related by the same) at least one first ECG electrode independent of the second ECG electrodes (e.g., an electrode of an ECG device separate from the second ECG electrodes, such as a 12-lead electrocardiogram sensor, a separate external and / or wearable cardiac monitoring device, and / or the like) to train a rhythm change classifier based on the second ECG electrodes. In some embodiments, the known arrhythmia type information may include a plurality of annotations, each of which may be associated with an individual ECG signal portion of the first plurality of ECG signal portions associated with the first ECG electrode. In some embodiments, each individual ECG signal portion of the second plurality of ECG signal portions associated with the second ECG electrode may correspond to an individual ECG signal portion of the first plurality of ECG signal portions. In some embodiments, the remote computer system 204c may train the arrhythmia type classifier by predicting a predicted arrhythmia type for each individual ECG signal portion of the second plurality of ECG signal portions using the arrhythmia type classifier, determining at least one error value (e.g., based on backpropagation and / or the like) based on the predicted arrhythmia type and an individual annotation of the plurality of annotations associated with the individual ECG signal portion of the first plurality of ECG signal portions that corresponds to the individual ECG signal portion of the second plurality of ECG signal portions, and training the arrhythmia type classifier (e.g., updating a weight and / or the like) based on the at least one error value.

[0190] In some embodiments, ECG signal portions associated with multiple electrodes are combined (e.g., by vector addition, vector projection, transformation, and / or the like) to form estimated ECG signal portions that may be familiar and / or suitable for review by a human user (e.g., a technician and / or the like). For example, the historical collection of multiple ECG signal portions may include a first plurality of ECG signal portions of at least one first ECG signal based on first surface ECG activity sensed by at least one first ECG electrode and a second plurality of ECG signal portions of at least one second ECG signal based on second surface ECG activity sensed by at least one second ECG electrode. Additionally or alternatively, the at least one second ECG electrode may be independent of the at least one first ECG electrode. In some embodiments, each ECG signal portion of the first plurality of ECG signal portions may be combined (e.g., by vector addition, vector projection, transformation, and / or the like) with an individual ECG signal portion of the second plurality of ECG signal portions to form a plurality of estimated ECG signal portions (e.g., by remote computer system 204c). In some embodiments, the known arrhythmia type information may include multiple annotations, where each individual annotation may be associated with a separate estimated ECG signal portion of the multiple estimated ECG signal portions.

[0191] In some embodiments, at least some of the multiple ECG signal portions of the historical collection may be time warped (e.g., time dilated and / or the like) to form multiple warped ECG signal portions (e.g., by the remote computer system 204c using any suitable signal processing techniques for time warping, time dilation, and / or the like).

[0192] In some embodiments, at least some of the ECG signal portions of the historical collection may be filtered, inverted, any combination thereof, and / or the like (eg, by the remote computer system 204c).

[0193] In some embodiments, the at least one noise signal portion may be combined (eg, by a remote computer system 204c) with at least some of the ECG signal portions of the historical collection.

[0194] In some embodiments, at least a portion of the plurality of ECG signal portions of the history collection may be style-transferred (e.g., by a remote computer system 204c). For example, the remote computer system 204c may search for ECG signals that share low-level features with one reference signal and high-level features with a second reference signal (e.g., the second reference signal may be associated with a rare type / class of arrhythmia and / or the like). Additionally or alternatively, the low- and / or high-level features may be the output of a pre-trained classification network having respective low- and / or high-level features.

[0195] As shown in FIG. 2C, at 232c, the remote computer system 204c may receive annotation data associated with at least one ECG signal and at least one annotation for each ECG signal, as described herein.

[0196] In some embodiments, the remote computer system 204c may detect arrhythmia types in the ECG signal using an arrhythmia type classifier, as described herein, and temporal data associated with the detected arrhythmia type. For example, the temporal data may include at least one of a start time, a time interval, any combination thereof, and / or the like, as described herein. In some embodiments, the remote computer system 204c may determine at least one ECG signal portion associated with the detected arrhythmia type in the ECG signal based on the temporal data, as described herein.

[0197] In some embodiments, the remote computer system 204c may determine a likelihood score for the annotation based on the detected arrhythmia type. For example, the output of at least one neural network of the arrhythmia type classifier may include a confidence score (e.g., probability and / or the like) associated with each possible arrhythmia type, as described herein. For example, the arrhythmia type determined by the arrhythmia type classifier may be the arrhythmia type (e.g., probability and / or the like) with the highest confidence score. Additionally or alternatively, each annotation may be associated with a possible arrhythmia type (e.g., a label associated with the possible arrhythmia type, a text string identifying at least one possible arrhythmia type, and / or the like). In some embodiments, the likelihood score for each annotation may be a confidence score (e.g., determined by the arrhythmia type classifier for the possible arrhythmia type associated with such annotation). In some embodiments, the arrhythmia type classifier may comprise multiple neural networks, and each such neural network may output a confidence score associated with at least one possible arrhythmia type.

[0198] In some embodiments, the remote computer system 204c may generate at least one message based on the likelihood score for the at least one determined ECG signal portion and the at least one annotation, as described herein. For example, the message may indicate at least one of: recommending annotating the at least one determined ECG signal portion based on the detected arrhythmia type; or recommending re-evaluating annotation data associated with the at least one determined ECG signal portion based on the likelihood score; and / or the like.

[0199] In some embodiments, the remote computer system 204c may determine that the likelihood score is below a threshold. Additionally or alternatively, generating the message may include the remote computer system 204c generating at least one message indicating a recommendation to re-evaluate the annotation data associated with the at least one determined ECG signal portion based on determining that the likelihood score is below a threshold, as described herein.

[0200] As shown in FIG. 2C, at 234c, the remote computer system 204c may transmit at least a portion of a message related to at least one determined ECG signal portion to the technician device 208c, as described herein.

[0201] As shown in FIG. 2C, at 236c, the remote computer system 204c may transmit at least a portion of a message related to at least one determined ECG signal portion to the supervisor device 210c, as described herein.

[0202] Reference is now made to FIG. 3A , which illustrates an exemplary swimlane diagram of a process 300a for arrhythmia monitoring according to some embodiments. In addition to system components, process 300a also illustrates communication flow between the system components. As illustrated in FIG. 3A , in some embodiments, cardiac monitoring device 302a may be the same as or similar to cardiac monitoring device 102, cardiac monitoring devices 202a and / or 202b, and / or the like. In some embodiments, remote computer system 304a may be the same as or similar to remote computer system 104, remote computer systems 204a, 204b, and / or 204c, and / or the like. In some embodiments, data repository 306a may be the same as or similar to data repository 106, data repositories 206a, 206b, and / or 206c, and / or the like. In some embodiments, technician device 308a may be the same as or similar to technician device 108, technician devices 202a, 202b, and / or 202c, and / or the like. In some embodiments, gateway device 329a may be the same as or similar to gateway device 129, gateway devices 229a and / or 229b, and / or the like.

[0203] As shown in FIG. 3A , at 310a, the remote computer system 304a may receive (e.g., obtain, retrieve, send requests and / or queries to the data repository 206a for communication, and / or the like) a historical collection of multiple ECG signal portions and associated information (e.g., known rhythm variation information, known arrhythmia type information, and / or the like) from, for example, the data repository 306a, as described herein.

[0204] As shown in FIG. 3A, at 310a, the remote computer system 304a may train at least one neural network of at least one classifier (e.g., a rhythm change classifier, an arrhythmia type classifier, and / or the like) based on a historical collection of multiple ECG signal portions and information associated therewith (e.g., each of known rhythm change information, known arrhythmia type information, and / or the like), as described herein.

[0205] In some embodiments, the rhythm change classifier may comprise at least one neural network, as described herein. For example, the at least one neural network may comprise at least one of a convolutional neural network, a recurrent neural network, an attention network, a fully connected neural network, any combination thereof, and / or the like, as described herein.

[0206] In some embodiments, the remote computer system 304a may train the rhythm change classifier by predicting predicted rhythm change information (e.g., data related to predicted rhythm changes (e.g., probabilities, confidence scores, and / or the like), data related to the absence of predicted rhythm changes (e.g., probabilities, confidence scores, and / or the like), and / or the like) for each ECG signal portion of a historical collection of multiple ECG signal portions using a rhythm change classifier as described herein, determining at least one error value based on the predicted rhythm change information and the known rhythm change information, and training the rhythm change classifier based on the error value (e.g., updating its weights and / or the like) (e.g., using backpropagation and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss, as described herein.

[0207] In some embodiments, the historical collection of multiple ECG signal portions may include a first ECG signal portion associated with a first time and a second ECG signal portion associated with a second time after the first time. Additionally or alternatively, the remote computer system 304a may train the rhythm change classifier by predicting a predicted ECG signal portion associated with the second time based on the first ECG signal portion using a rhythm change classifier as described herein, determining at least one error value based on the predicted ECG signal portion and the second ECG signal portion, and training the rhythm change classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss, as described herein.

[0208] In some embodiments, the historical collection of the plurality of ECG signal portions may include a first ECG signal portion associated with a first time and a second ECG signal portion associated with a second time. Additionally or alternatively, the remote computer system 304a may train a rhythm change classifier by predicting a predicted time associated with the second ECG signal portion based on the first ECG signal portion and the second ECG signal using a rhythm change classifier as described herein, determining at least one error value based on the predicted time and the second time, and training the rhythm change classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss, as described herein.

[0209] In some embodiments, the remote computer system 304a may train the arrhythmia type classifier by predicting a predicted arrhythmia type for each individual ECG signal portion of a historical collection of a plurality of ECG signal portions (or a second plurality of ECG signal portions) using an arrhythmia type classifier as described herein, determining at least one error value based on the predicted arrhythmia type and known arrhythmia type information (e.g., individual annotations associated with known arrhythmia types for each individual ECG signal portion), and training the arrhythmia type classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss, as described herein.

[0210] In some embodiments, there may be an insufficient number of ECG signal portions in a historical collection of multiple ECG signal portions with known rhythm change information to train a rhythm change classifier to perform a desired task (e.g., detecting and / or identifying at least one predetermined rhythm change). Additionally or alternatively, there may be sufficient data (e.g., past ECG signal portions and / or the like) to train the rhythm change classifier to perform another task (e.g., that may be related in some way to the target task). In some embodiments, the rhythm change classifier may be trained to perform another task (e.g., counting R peaks based on ECG signals, determining heart rate, and / or the like). Additionally or alternatively, the rhythm change classifier may be applied to perform the target task as described herein. For example, in some embodiments, the rhythm change classifier may be retrained using a limited amount of ECG signal portions in a historical collection of multiple ECG signal portions with known rhythm change information and / or the like as described herein. Additionally or alternatively, the rhythm change classifier may be used to perform another task (e.g., counting R peaks based on ECG signals, determining a heart rate, and / or the like) as described herein, and its output may be applied to the target task. For example, a processor (e.g., of the cardiac monitoring device 302a and / or the gateway device 329a) may use the rhythm change classifier to detect at least one peak count or heart rate based on at least one ECG signal. Additionally or alternatively, the processor (e.g., of the cardiac monitoring device 302a and / or the gateway device 329a) may determine that the detected at least one peak count or heart rate exceeds a first threshold value for the patient (e.g., a tachycardia onset threshold) or is below a second threshold value for the patient (e.g., a bradycardia onset threshold) (e.g., where the second threshold value for the patient may be less than the first threshold value for the patient).Additionally or alternatively, the processor (e.g., of the cardiac monitoring device 302a and / or the gateway device 329a) may detect the predetermined rhythm change based on at least one peak number or heart rate exceeding a first threshold value for the patient (e.g., a tachycardia onset threshold) or falling below a second threshold value for the patient (e.g., a bradycardia onset threshold).

[0211] In some embodiments, there may be an insufficient number of ECG signal portions associated with (e.g., sensed from and / or related by the same) the plurality of ECG electrodes of cardiac monitoring device 302a in a historical collection of ECG signal portions with known rhythm change information to train a rhythm change classifier for ECG signals received from the plurality of ECG electrodes of cardiac monitoring device 202a. Additionally or alternatively, there may be sufficient data (e.g., past ECG signal portions, and / or the like) associated with (e.g., sensed from and / or related by the same) a second plurality of ECG electrodes independent from the plurality of ECG electrodes of cardiac monitoring device 302a (e.g., electrodes of an ECG device separate from cardiac monitoring device 302b, such as a 12-lead ECG sensor, a separate external and / or wearable cardiac monitoring device, and / or the like) to train a rhythm change classifier based on the second plurality of ECG electrodes. In some embodiments, the rhythm change classifier may be trained based on ECG signal portions associated with the second plurality of ECG electrodes (e.g., sensed from the second plurality of ECG electrodes and / or the like), as described herein. Additionally or alternatively, the rhythm change classifier may then be applied to detect predetermined rhythm changes based on the plurality of ECG electrodes of cardiac monitoring device 302a, as described herein. In some embodiments, remote computer system 204a may determine (e.g., calculate and / or perform the like) a transformation (e.g., vector projection and / or the like) of the ECG signal portions associated with the second plurality of ECG electrodes onto the plurality of ECG electrodes of cardiac monitoring device 302a, and may use the transformation of the ECG signal portions to train the rhythm change classifier as if the ECG signal portions were associated with (e.g., sensed from and / or the like) the plurality of ECG electrodes of cardiac monitoring device 302a.

[0212] In some embodiments, the arrhythmia type classifier may comprise at least one neural network (e.g., at least one second neural network) as described herein. Additionally or alternatively, as described herein, the at least one (second) neural network may comprise at least one of a deep neural network, a convolutional neural network, a recurrent neural network, an attention network, a fully connected neural network, any combination thereof, and / or the like.

[0213] 3A , at 314a, remote computer system 304a may communicate the trained rhythm change classifier (or weights thereof) to cardiac monitoring device 302a and / or gateway device 329a, as described herein. In some embodiments, after training, weights corresponding to the trained rhythm change classifier may be communicated to cardiac monitoring device 302a and / or gateway device 329a. Additionally or alternatively, a copy of the trained rhythm change classifier (or weights thereof) may be downloaded from remote computer system 304a and / or installed (e.g., uploaded to, written to, configured on, and / or the like) on at least one non-transitory computer-readable medium (e.g., memory, programmable circuit board, field programmable gate array (FPGA), integrated circuit, any combination thereof, and / or the like) that may be installed on and / or part of cardiac monitoring device 302a and / or gateway device 329a.

[0214] In some embodiments, the cardiac monitoring device 302a may be an external cardiac monitoring device for the patient, as described herein. For example, the (external) cardiac monitoring device 302a may include a plurality of ECG electrodes that sense the patient's surface ECG activity, as described herein. For example, the (external) cardiac monitoring device 302a may include ECG processing circuitry that processes the patient's surface ECG activity to provide at least one ECG signal for the patient on at least one ECG channel.

[0215] In some embodiments, the (external) cardiac monitoring device 302a may comprise a non-transitory computer-readable medium (e.g., a memory, a programmable circuit board, a field-programmable gate array, any combination thereof, and / or the like) having (e.g., implementing, embodying, storing, and / or the like) a trained rhythm change classifier as described herein (which may include, e.g., at least one neural network trained based on historical collections of multiple ECG signal portions with known rhythm change information). Additionally or alternatively, the (external) cardiac monitoring device 302a may comprise at least one processor operably connected to the ECG channel and the non-transitory computer-readable medium.

[0216] In some embodiments, the gateway device 329a may comprise a non-transitory computer-readable medium (e.g., a memory, a programmable circuit board, a field-programmable gate array, any combination thereof, and / or the like) having (e.g., implementing, embodying, storing, and / or the like) a trained rhythm change classifier as described herein (which may include, e.g., at least one neural network trained based on a historical collection of multiple ECG signal portions with known rhythm change information). Additionally or alternatively, the gateway device 329a may comprise at least one processor operably connected to the non-transitory computer-readable medium.

[0217] In some embodiments, the gateway device 329a may facilitate communication between the cardiac monitoring device 202a and the remote computer system 204a as described herein.

[0218] As shown in FIG. 3A, at 316a, the cardiac monitoring device 302a and / or the gateway device 329a (e.g., its processor) may receive an ECG signal. In some embodiments, cardiac monitoring device 302a may receive ECG signals via an ECG channel as described herein. Additionally or alternatively, cardiac monitoring device 302a may communicate (e.g., transmit) ECG signals to gateway device 329a and / or gateway device 329a may receive ECG signals from cardiac monitoring device 302a as described herein.

[0219] 3A , at 318a, the cardiac monitoring device 302a and / or the gateway device 329a (e.g., a processor thereof) may detect temporal data corresponding to a predetermined rhythm change in at least one ECG signal using a rhythm change classifier, as described herein. For example, the predetermined rhythm change may be associated with an arrhythmia, as described herein (e.g., a change in heart rate, atrial fibrillation, flutter, supraventricular tachycardia, ventricular tachycardia, pauses, atrioventricular block, ventricular fibrillation, bigeminy, triplegia, ventricular ectopic beats, bradycardia, tachycardia, a change in morphology of at least one ECG signal, any combination thereof, and / or the like). Additionally or alternatively, as described herein, the temporal data may include at least one of a start time, a time interval, any combination thereof, and / or the like.

[0220] As shown in FIG. 3A, at 320a, the cardiac monitoring device 302a and / or the gateway device 329a (e.g., a processor thereof) may determine (e.g., based on the detected time data) at least one ECG signal portion associated with the detected time data that corresponds to a predetermined rhythm change in the ECG signal, as described herein.

[0221] In some embodiments, the ECG channel may include multiple ECG channels, as described herein. Additionally or alternatively, the ECG signal may include at least one individual ECG signal associated with each individual ECG channel, as described herein. In some embodiments, the multiple ECG channels may include a first ECG channel and a second ECG channel, as described herein. Additionally or alternatively, the ECG signal may include a first individual ECG signal associated with the first ECG channel and a second individual ECG signal associated with the second ECG channel, as described herein. In some embodiments, the first individual ECG signal may be substantially orthogonal to the second individual ECG signal, as described herein.

[0222] In some embodiments, the neural network of the rhythm change classifier (e.g., cardiac monitoring device 302a and / or gateway device 329a) may include multiple Siamese branches as described herein. Additionally or alternatively, each individual Siamese branch may be associated with an individual ECG channel (e.g., of the multiple ECG channels) as described herein.

[0223] In some embodiments, cardiac monitoring device 302a and / or gateway device 329a (e.g., its processor) may further detect the predetermined rhythm change based on at least one ECG signal (e.g., using a trained rhythm change classifier). In some embodiments, cardiac monitoring device 302a may further include at least one sensor and associated sensor circuitry that senses the patient's non-ECG biometric data (which, in some embodiments, may be communicated to gateway device 329a), as described herein. Additionally or alternatively, detecting the predetermined rhythm change may be further based on the patient's non-ECG biometric data, as described herein (e.g., the patient's non-ECG biometric data may be input into a neural network of the rhythm change classifier, combined with the output of the rhythm change classifier, and / or the like).

[0224] In some embodiments, detecting the predetermined rhythm change may be further based on at least one baseline ECG signal portion of the patient, as described herein.

[0225] In some embodiments, detecting the predetermined rhythm change may be further based on at least one calibration measurement of the patient, as described herein.

[0226] In some embodiments, detecting the predetermined rhythm change may be further based on at least one reference vector of the patient, as described herein.

[0227] In some embodiments, detecting the predetermined rhythm change may be further based on at least one previous ECG signal portion, as described herein.

[0228] In some embodiments, the processor (e.g., of the cardiac monitoring device 302a and / or the gateway device 329a) may further determine (e.g., using a rhythm change classifier) ​​a confidence score associated with a given rhythm change based on at least one ECG signal, as described herein.

[0229] 3A, at 322a, cardiac monitoring device 302a and / or gateway device 329a (e.g., a processor thereof) may communicate (e.g., transmit, and / or the like) the determined ECG signal portions to remote computer system 304a as described herein. Additionally or alternatively, cardiac monitoring device 302a and / or gateway device 329a (e.g., a processor thereof) may detect and / or communicate indicia (e.g., flags, indicators, confidence scores, marks, metadata, time data, and / or the like) associated with predetermined rhythm changes detected (e.g., identified and / or obtained by the like) in the ECG signal portions.

[0230] In some embodiments, the processor (e.g., of the cardiac monitoring device 202a and / or the gateway device 329a) may further communicate (e.g., transmit and / or the like) at least one second ECG signal portion of the ECG signal to the remote computer system 304a, as described herein. Additionally or alternatively, the second ECG signal portion may be independent of the detected time data corresponding to the predetermined rhythm change in the ECG signal, as described herein.

[0231] In some embodiments, the gateway device 329a may facilitate communication between the cardiac monitoring device 302a and the remote computer system 304a, as described herein. For example, transmitting the determined ECG signal portions to the remote computer system 204a may include the cardiac monitoring device 202a communicating (e.g., transmitting, and / or the like) the determined ECG signal portions to the gateway device 329a, as described herein.

[0232] In some embodiments, the remote computing system 304a may receive the determined ECG signal portions (eg, from the cardiac monitoring device 302a and / or the gateway device 329a).

[0233] 3A , at 324a, the remote computing system 304a may analyze the determined ECG signal portion to classify an arrhythmia type for the rhythm change in the ECG signal, as described herein. In some embodiments, the arrhythmia type may include at least one of a change in heart rate, atrial fibrillation, flutter, supraventricular tachycardia, ventricular tachycardia, pauses, atrioventricular block, ventricular fibrillation, bigeminy, triplegia, ventricular ectopic beats, bradycardia, tachycardia, a change in the morphology of at least one ECG signal, any combination thereof, and / or the like, as described herein.

[0234] In some embodiments, the remote computer system 304a may include an arrhythmia type classifier (e.g., having at least one (second) neural network trained based on a (second) historical collection of a (second) plurality of ECG signal portions with known arrhythmia type information) as described herein. Additionally or alternatively, analyzing the at least one determined ECG signal portion may include the remote computer system 304a detecting an arrhythmia type associated with the rhythm change based on the determined ECG signal portion using the arrhythmia type classifier as described herein.

[0235] In some embodiments, the remote computer system 304a may analyze the determined ECG signal portion to identify at least one arrhythmia associated with a rhythm change in the at least one ECG signal, as described herein. In some embodiments, the arrhythmia may be one or more rare arrhythmias for which the remote computer system 304a may have been previously trained. Additionally or alternatively, the remote computer system 304a may use any suitable signal processing technique (e.g., separate from or including an arrhythmia type classifier as described herein) to identify rare arrhythmias, as described herein.

[0236] In some embodiments, the determined ECG signal portion may include multiple determined ECG signal portions. Additionally or alternatively, the remote computer system 304a may receive multiple determined ECG signal portions from the cardiac monitoring device 302a and / or the gateway device 329a, as described herein. In some embodiments, the remote computer system 304a may analyze each individual determined ECG signal portion to classify an individual class for each individual determined ECG signal portion, as described herein. Additionally or alternatively, the class of at least two individual determined ECG signal portions may include a first class, as described herein (e.g., at least two ECG signal portions may belong to the same class / group).

[0237] 3A, the remote computer system 304a may communicate (e.g., transmit and / or perform the like) at 326a at least one message related to the determined ECG signal portion and / or the arrhythmia type related to the rhythm change, as described herein. For example, a message may be communicated from the remote computer system 304a to the technician device 308a, as described herein.

[0238] In some embodiments, the remote computer system 304a may transmit at least one message related to the second ECG signal portion (e.g., a randomly determined second ECG signal portion as described herein, a second ECG signal portion determined to have a confidence score below or above a first threshold and above a second threshold, and / or the like) to the technician device 308a, as described herein.

[0239] In some embodiments, the remote computer system 304a may transmit to the technician device 308a at least two individually determined ECG signal portions and at least one message related to the first class, as described herein.

[0240] As shown in FIG. 3A, at 328a, the technician device 308b may receive at least one annotation associated with the ECG signal portion from a user (eg, a technician and / or the like) via, for example, input.

[0241] 3A , at 330a, the remote computer system 304a may receive annotation data associated with the annotations from the technician device 308a, as described herein. For example, the annotation data may be communicated from the technician device 308a to the remote computer system 304a, as described herein. Additionally or alternatively, portions of the ECG signal associated with such annotations may be communicated using the annotation data, as described herein.

[0242] In some embodiments, the remote computer system 304a may receive (e.g., from the technician device 308a) annotation data related to annotations for second ECG signal portions (e.g., randomly determined second ECG signal portions, determined to have a confidence score below a first threshold and above a second threshold, and / or the like, as described herein).

[0243] 3A , at 332a, the remote computer system 304a may retrain the rhythm change classifier (and / or train an updated rhythm change classifier) ​​based on a historical collection of multiple ECG signal segments with known rhythm change information, second ECG signal segments, and associated annotation data, as described herein. In some embodiments, the remote computer system 304a may retrain the arrhythmia type classifier based on a historical collection of multiple ECG signal segments with known rhythm change information, second ECG signal segments, and associated annotation data.

[0244] 3A, at 334a, the remote computer system 304a may communicate (e.g., send, write, and / or the like) the annotation data to the data repository 306a, as described herein. Additionally or alternatively, the ECG signal portion associated with the annotation may be communicated with the annotation data, as described herein.

[0245] In some embodiments, annotation data and ECG signal portions associated with such annotations may be added to a historical collection of multiple ECG signal portions. For example, the annotation data may be saved as known rhythm change information and / or known arrhythmia type information for the ECG signal portion associated with the annotation. In some embodiments, the remote computer system 304a may retrain the rhythm change classifier (and / or train an updated rhythm change classifier) ​​based on a historical collection of multiple ECG signal portions with known rhythm change information (which may include ECG signal portions and / or annotation data associated therewith). In some embodiments, the remote computer system 304a may retrain the arrhythmia type classifier based on a historical collection of multiple ECG signal portions with known arrhythmia type information (which may include ECG signal portions and / or annotation data associated therewith).

[0246] In some embodiments, the remote computer system 304a may add annotation data related to at least one annotation for a second ECG signal portion (e.g., a randomly determined second ECG signal portion, a second ECG signal portion determined to have a confidence score below a first threshold and above a second threshold, and / or the like, as described herein) to the historical collection of multiple ECG signal portions in the data repository 306a. In some embodiments, the remote computer system 304a may retrain the rhythm change classifier (and / or train an updated rhythm change classifier) ​​based on the historical collection of multiple ECG signal portions with known rhythm change information (which may include the second ECG signal portion and / or annotation data associated therewith). In some embodiments, the remote computer system 304a may retrain the arrhythmia type classifier based on the historical collection of multiple ECG signal portions with known arrhythmia type information (which may include the second ECG signal portion and / or annotation data associated therewith).

[0247] 3A , at 336a, remote computer system 304a may communicate the retrained rhythm change classifier (and / or trained, updated rhythm change classifier) ​​and / or its weighting to cardiac monitoring device 302a and / or gateway device 329a, as described herein. Additionally or alternatively, a copy of the retrained rhythm change classifier (and / or trained, updated rhythm change classifier) ​​and / or its weighting may be downloaded from remote computer system 304a and / or installed (e.g., uploaded to, written to, configured on, and / or the like) on at least one non-transitory computer-readable medium (e.g., memory, programmable circuit board, FPGA, integrated circuit, any combination thereof, and / or the like) that may be installed on and / or part of cardiac monitoring device 302a and / or gateway device 329a.

[0248] Referring now to FIG. 3B, FIG. 3B illustrates an exemplary swimlane diagram of a process 300b for arrhythmia monitoring according to some embodiments. In addition to the system components, process 300b illustrates the communication flow between the system components. As shown in FIG. 3B, in some embodiments, cardiac monitoring device 302b may be the same as or similar to cardiac monitoring device 102, cardiac monitoring devices 202a and / or 202b, cardiac monitoring device 302a, and / or the like. In some embodiments, remote computer system 304b may be the same as or similar to remote computer system 104, remote computer systems 204a, 204b, and / or 204c, remote computer system 304a, and / or the like. In some embodiments, data repository 306b may be the same as or similar to data repository 106, data repositories 206a, 206b, and / or 206c, data repository 306a, and / or the like. In some embodiments, technician device 308b may be the same as or similar to technician device 108, technician devices 208a, 208b and / or 208c, technician device 308a, and / or the like. In some embodiments, gateway device 329b may be the same as or similar to gateway device 129, gateway devices 229a and / or 229b, gateway device 329a, and / or the like.

[0249] As shown in FIG. 3B , at 310b, the remote computer system 304b may receive (e.g., obtain, retrieve, send requests and / or queries to the data repository 306b, and / or the like) a historical collection of multiple ECG signal portions and associated information (e.g., known rhythm change information, known arrhythmia type information, and / or the like) from, for example, the data repository 306b, as described herein.

[0250] As shown in FIG. 3B, at 312b, the remote computer system 304b may train at least one neural network of at least one classifier (e.g., an arrhythmia type classifier, and / or the like) based on a historical collection of multiple ECG signal portions and information associated therewith (e.g., known arrhythmia type information, and / or the like), as described herein.

[0251] In some embodiments, the arrhythmia type classifier may comprise at least one neural network (e.g., at least one second neural network) as described herein. Additionally or alternatively, as described herein, the at least one (second) neural network may comprise at least one of a deep neural network, a convolutional neural network, a recurrent neural network, an attention network, a fully-connected neural network, any combination thereof, and / or the like. In some embodiments, the neural network may include multiple Siamese branches (e.g., individual Siamese branches associated with individual ECG channels) as described herein.

[0252] In some embodiments, the remote computer system 304b may train the arrhythmia type classifier by: predicting, using the arrhythmia type classifier, a predicted arrhythmia type for each individual ECG signal portion of a historical collection of the plurality of ECG signal portions (or a second plurality of ECG signal portions); determining at least one error value based on the predicted arrhythmia type and known arrhythmia type information (e.g., individual annotations associated with known arrhythmia types for each individual ECG signal portion); and training the arrhythmia type classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss.

[0253] As shown in FIG. 3B, at 314b, the remote computer system 304b may communicate the historical collection of multiple ECG signal portions and information associated therewith (e.g., known rhythm change information, and / or the like) to the cardiac monitoring device 302b and / or the gateway device 329b, as described herein.

[0254] In some embodiments, the gateway device 329b may facilitate communication between the cardiac monitoring device 302b and the remote computer system 304b as described herein.

[0255] As shown in FIG. 3B, at 316b, cardiac monitoring device 302b and / or gateway device 329b may train a rhythm change classifier, which may be implemented by at least one non-transitory computer-readable medium (e.g., memory, programmable circuit board, field programmable gate array (FPGA), integrated circuit, any combination thereof, and / or the like) installed on and / or part of cardiac monitoring device 302b and / or gateway device 329b, as described herein.

[0256] In some embodiments, cardiac monitoring device 302b may be an external cardiac monitoring device for the patient, as described herein. For example, (external) cardiac monitoring device 302b may include a plurality of ECG electrodes that sense the patient's surface ECG activity. Additionally or alternatively, (external) cardiac monitoring device 302b may include ECG processing circuitry that processes the patient's surface ECG activity to provide at least one ECG signal for the patient on at least one ECG channel.

[0257] In some embodiments, the (external) cardiac monitoring device 302b may comprise a non-transitory computer-readable medium (e.g., a memory, a programmable circuit board, a field-programmable gate array, any combination thereof, and / or the like) having (e.g., implementing, embodying, storing, and / or the like) a rhythm change classifier as described herein (which may, e.g., include at least one neural network). Additionally or alternatively, the (external) cardiac monitoring device 302b may comprise at least one processor operably connected to the ECG channel and the non-transitory computer-readable medium.

[0258] In some embodiments, gateway device 329b may comprise a non-transitory computer-readable medium (e.g., memory, a programmable circuit board, a field-programmable gate array, any combination thereof, and / or the like) having (e.g., implementing, embodying, storing, and / or the like) a trained rhythm change classifier as described herein (which may, for example, include at least one neural network trained based on a historical collection of multiple ECG signal portions with known rhythm change information). Additionally or alternatively, gateway device 329b may comprise at least one processor operably connected to the non-transitory computer-readable medium.

[0259] In some embodiments, the rhythm change classifier may comprise at least one neural network, as described herein. Additionally or alternatively, the at least one neural network may comprise at least one of a convolutional neural network, a recurrent neural network, an attention network, a fully connected neural network, any combination thereof, and / or the like, as described herein. In some embodiments, the neural network may include multiple Siamese branches, as described herein (e.g., individual Siamese branches associated with individual ECG channels).

[0260] In some embodiments, cardiac monitoring device 302b and / or gateway device 329b may train a rhythm change classifier by predicting predicted rhythm change information (e.g., data related to predicted rhythm changes (e.g., probabilities, confidence scores, and / or the like), data related to the absence of predicted rhythm changes (e.g., probabilities, confidence scores, and / or the like), and / or the like) for each ECG signal portion of a historical collection of multiple ECG signal portions using a rhythm change classifier as described herein, determining at least one error value based on the predicted rhythm change information and the known rhythm change information, and training the rhythm change classifier based on the error value (e.g., updating its weights and / or the like) (e.g., using backpropagation and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss as described herein.

[0261] In some embodiments, the historical collection of multiple ECG signal portions may include a first ECG signal portion associated with a first time and a second ECG signal portion associated with a second time after the first time, as described herein. Additionally or alternatively, cardiac monitoring device 302b and / or gateway device 329b may train the rhythm change classifier by predicting a predicted ECG signal portion associated with a second time based on the first ECG signal portion using a rhythm change classifier, as described herein, determining at least one error value based on the predicted ECG signal portion and the second ECG signal portion, and training the rhythm change classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss, as described herein.

[0262] In some embodiments, the historical collection of multiple ECG signal portions may include a first ECG signal portion associated with a first time and a second ECG signal portion associated with a second time, as described herein. Additionally or alternatively, cardiac monitoring device 302b and / or gateway device 329b may train a rhythm change classifier by predicting a predicted time associated with the second ECG signal portion based on the first time and the second time using a rhythm change classifier, as described herein, determining at least one error value based on the predicted time and the second time, and training the rhythm change classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss, as described herein.

[0263] In some embodiments, there may be an insufficient number of ECG signal portions in a historical collection of multiple ECG signal portions with known rhythm change information to train a rhythm change classifier to perform a desired task (e.g., detecting and / or identifying at least one predetermined rhythm change), as described herein. Additionally or alternatively, there may be sufficient data (e.g., past ECG signal portions and / or the like) to train a rhythm change classifier to perform another task (e.g., that may be related in some way to the target task), as described herein. In some embodiments, a rhythm change classifier may be trained to perform another task (e.g., counting R peaks based on an ECG signal, determining heart rate, and / or the like), as described herein. Additionally or alternatively, a rhythm change classifier may be applied to perform a target task, as described herein. For example, in some embodiments, a rhythm change classifier may be retrained using a limited amount of ECG signal portions in a historical collection of multiple ECG signal portions with known rhythm change information, as described herein, and / or the like. Additionally or alternatively, the rhythm change classifier may be used to perform another task (e.g., counting R peaks based on ECG signals, determining heart rate, and / or the like) and its output may be applied to the target task, as described herein. For example, a processor (e.g., of cardiac monitoring device 302b and / or gateway device 329b) may use a rhythm change classifier to detect at least one peak count or heart rate based on at least one ECG signal, as described herein.Additionally or alternatively, as described herein, the processor (e.g., of the cardiac monitoring device 302b and / or the gateway device 329b) may determine that at least one detected peak count or heart rate exceeds a first threshold value for the patient (e.g., a tachycardia onset threshold) or falls below a second threshold value for the patient (e.g., a bradycardia onset threshold) (e.g., where the second threshold value for the patient may be less than the first threshold value for the patient).

[0264] In some embodiments, there may be an insufficient number of ECG signal portions associated with (e.g., sensed from and / or related by the same) the plurality of ECG electrodes of cardiac monitoring device 302a in a historical collection of ECG signal portions with known rhythm change information to train a rhythm change classifier for ECG signals received from the plurality of ECG electrodes of cardiac monitoring device 302b, as described herein. Additionally or alternatively, there may be sufficient data (e.g., past ECG signal portions, and / or the like) associated with (e.g., sensed from and / or related by the same) a second plurality of ECG electrodes independent from the plurality of ECG electrodes of cardiac monitoring device 302a (e.g., electrodes of an ECG device separate from cardiac monitoring device 302b, such as a 12-lead ECG sensor, a separate external and / or wearable cardiac monitoring device, and / or the like) to train a rhythm change classifier based on the second plurality of ECG electrodes, as described herein. In some embodiments, as described herein, a rhythm change classifier may be trained based on ECG signal portions associated with the second plurality of ECG electrodes (e.g., sensed from the second plurality of ECG electrodes and / or the like). Additionally or alternatively, as described herein, the rhythm change classifier may then be applied to detect predetermined rhythm changes based on the plurality of ECG electrodes of cardiac monitoring device 302b. In some embodiments, as described herein, cardiac monitoring device 302b and / or gateway device 329b may determine (e.g., calculate and / or perform the like) a transformation (e.g., vector projection and / or the like) of the ECG signal portions associated with the second plurality of ECG electrodes onto the plurality of ECG electrodes of cardiac monitoring device 302b and use the transformation of the ECG signal portions to train a rhythm change classifier as if the ECG signal portions were associated with (e.g., sensed from and / or the like) the plurality of ECG electrodes of cardiac monitoring device 302b.

[0265] 3B , at 318b, cardiac monitoring device 302b and / or gateway device 329b (e.g., a processor thereof) may receive ECG signals. In some embodiments, cardiac monitoring device 302a may receive ECG signals via an ECG channel as described herein. Additionally or alternatively, cardiac monitoring device 302a may communicate (e.g., transmit) ECG signals to gateway device 329a and / or gateway device 329a may receive ECG signals from cardiac monitoring device 302a as described herein.

[0266] As shown in FIG. 3B, at 320b, the cardiac monitoring device 302b and / or the gateway device 329b (e.g., its processor) may use a rhythm change classifier as described herein to detect time data corresponding to predetermined rhythm changes in the ECG signal.

[0267] In some embodiments, the neural network of the rhythm change classifier (eg, cardiac monitoring device 302b and / or gateway device 329b) may include multiple Siamese branches, as described herein.

[0268] In some embodiments, cardiac monitoring device 302b and / or gateway device 329b (e.g., its processor) may further detect the predetermined rhythm change based on at least one ECG signal (e.g., using a trained rhythm change classifier) ​​as described herein. In some embodiments, cardiac monitoring device 302b may further include at least one sensor and associated sensor circuitry that senses the patient's non-ECG biometric data (which, in some embodiments, may be communicated to gateway device 329b) as described herein. Additionally or alternatively, detecting the predetermined rhythm change may be further based on the patient's non-ECG biometric data as described herein (e.g., the patient's non-ECG biometric data may be input into a neural network of the rhythm change classifier, combined with the output of the rhythm change classifier, and / or the like).

[0269] In some embodiments, detecting the predetermined rhythm change may be further based on at least one baseline ECG signal portion of the patient, as described herein.

[0270] In some embodiments, detecting the predetermined rhythm change may be further based on at least one calibration measurement of the patient, as described herein.

[0271] In some embodiments, detecting the predetermined rhythm change may be further based on at least one reference vector of the patient, as described herein.

[0272] In some embodiments, detecting the predetermined rhythm change may be further based on at least one previous ECG signal portion, as described herein.

[0273] In some embodiments, the processor (e.g., of the cardiac monitoring device 302b and / or the gateway device 329b) may further determine (e.g., using a rhythm change classifier) ​​a confidence score associated with a given rhythm change based on at least one ECG signal, as described herein.

[0274] As shown in FIG. 3B, at 322b, the cardiac monitoring device 302b and / or the gateway device 329b (e.g., a processor thereof) may determine (e.g., based on the detected time data) at least one ECG signal portion associated with the detected time data that corresponds to a predetermined rhythm change in the ECG signal, as described herein.

[0275] 3B , at 324b, cardiac monitoring device 302b and / or gateway device 329b (e.g., a processor thereof) may communicate (e.g., transmit, and / or the like) the determined ECG signal portions to remote computer system 304b as described herein. Additionally or alternatively, cardiac monitoring device 302b and / or gateway device 329b (e.g., a processor thereof) may detect and / or communicate indicia (e.g., flags, indicators, confidence scores, marks, metadata, time data, and / or the like) associated with the predetermined rhythm changes detected (e.g., identified and / or obtained by the like) in the ECG signal portions as described herein.

[0276] In some embodiments, the processor (e.g., of cardiac monitoring device 302b and / or gateway device 329b) may further communicate (e.g., transmit and / or the like) at least one second ECG signal portion of the ECG signal to remote computer system 304b, as described herein. Additionally or alternatively, the second ECG signal portion may be independent of the detected time data corresponding to the predetermined rhythm change in the ECG signal, as described herein.

[0277] In some embodiments, the gateway device 329b may facilitate communication between the cardiac monitoring device 302b and the remote computer system 304b as described herein.

[0278] In some embodiments, the remote computing system 304b may receive the determined ECG signal portions (eg, from the cardiac monitoring device 302b and / or the gateway device 329b) as described herein.

[0279] 3B, at 326b, the remote computing system 304b may analyze the determined ECG signal portions to classify an arrhythmia type for the rhythm change in the ECG signal as described herein. For example, the remote computing system 304b may include an arrhythmia type classifier (e.g., having at least one (second) neural network trained based on a (second) historical collection of a (second) plurality of ECG signal portions with known arrhythmia type information) as described herein.

[0280] In some embodiments, the remote computer system 304b may analyze the determined ECG signal portion to identify at least one arrhythmia associated with a rhythm change in the at least one ECG signal, as described herein. In some embodiments, the arrhythmia may be one or more rare arrhythmias for which the remote computer system 204a may have been previously trained. Additionally or alternatively, the remote computer system 204a may use any suitable signal processing technique (e.g., separate from or including an arrhythmia type classifier as described herein) to identify rare arrhythmias, as described herein.

[0281] In some embodiments, the determined ECG signal portion may include multiple determined ECG signal portions, as described herein. Additionally or alternatively, the remote computer system 304b may receive multiple determined ECG signal portions from the cardiac monitoring device 302b and / or the gateway device 329b, as described herein. In some embodiments, the remote computer system 304b may analyze each individual determined ECG signal portion to classify an individual class for each individual determined ECG signal portion, as described herein.

[0282] 3B, the remote computer system 304b may communicate (e.g., transmit and / or perform the like) at 328b at least one message related to the determined ECG signal portion and / or the arrhythmia type related to the rhythm change, as described herein. For example, a message may be communicated from the remote computer system 304b to the technician device 308b, as described herein.

[0283] In some embodiments, the remote computer system 304b may transmit at least one message related to the second ECG signal portion (e.g., a randomly determined second ECG signal portion as described herein, a second ECG signal portion determined to have a confidence score below a first threshold and above a second threshold, and / or the like) to the technician device 308b, as described herein.

[0284] In some embodiments, the remote computer system 304b may transmit to the technician device 308b at least two individually determined ECG signal portions and at least one message related to the first class, as described herein.

[0285] As shown in FIG. 3B, at 330b, the technician device 308b may receive at least one annotation associated with the ECG signal portion from a user (eg, a technician and / or the like) via, for example, input.

[0286] 3B, at 332b, the remote computer system 304b may receive annotation data associated with at least one annotation from the technician device 308a, as described herein. For example, the annotation data may be communicated from the technician device 308b to the remote computer system 304b, as described herein. Additionally or alternatively, portions of the ECG signal associated with such annotations may be communicated using the annotation data, as described herein.

[0287] In some embodiments, the remote computer system 304b may receive (e.g., from the technician device 308b) annotation data related to at least one annotation for a second ECG signal portion (e.g., a randomly determined second ECG signal portion, a second ECG signal portion determined to have a confidence score below a first threshold and above a second threshold, and / or the like, as described herein).

[0288] As shown in FIG. 3B, at 334b, the remote computer system 304b may retrain the arrhythmia type classifier based on the known rhythm variation information, the historical collection of multiple ECG signal portions with the second ECG signal portion and the associated annotation data, as described herein.

[0289] 3B, at 336b, the remote computer system 304b may communicate (e.g., send, write, and / or the like) the annotation data to the data repository 306b, as described herein. Additionally or alternatively, the ECG signal portion associated with the annotation may be communicated with the annotation data, as described herein.

[0290] In some embodiments, annotation data and ECG signal portions associated with such annotations may be added to a historical collection of multiple ECG signal portions, as described herein. In some embodiments, the remote computer system 304b may retrain the arrhythmia type classifier based on a historical collection of multiple ECG signal portions with known arrhythmia type information (which may include ECG signal portions and / or annotation data associated therewith).

[0291] In some embodiments, the remote computer system 304b may add annotation data related to at least one annotation for a second ECG signal portion (e.g., a randomly determined second ECG signal portion, a second ECG signal portion determined to have a confidence score below a first threshold and above a second threshold, and / or the like, as described herein) to the historical collection of multiple ECG signal portions in the data repository 306b. In some embodiments, the remote computer system 304b may retrain the arrhythmia type classifier based on the historical collection of multiple ECG signal portions with known arrhythmia type information (which may include the second ECG signal portion and / or annotation data associated therewith), as described herein.

[0292] 3B , at 338b, remote computer system 304b may communicate (e.g., transmit, and / or the like) annotation data associated with at least one annotation for the second ECG signal portion to cardiac monitoring device 302b and / or gateway device 329b, as described herein. Additionally or alternatively, cardiac monitoring device 302b and / or gateway device 329b (e.g., a processor thereof) may retrain the rhythm change classifier (and / or train an updated rhythm change classifier) ​​based on the known rhythm change information, the second ECG signal portion, and a historical collection of multiple ECG signal portions with associated annotation data, as described herein.

[0293] Reference is now made to FIG. 3C , which illustrates an exemplary swimlane diagram of a process 300c for arrhythmia monitoring according to some embodiments. In addition to the system components, the process 300c also illustrates the communication flow between the system components. As shown in FIG. 3C , in some embodiments, the remote computer system 304c may be the same as or similar to the remote computer system 104, the remote computer systems 204a, 204b, and / or 204c, the remote computer systems 304a and / or 304b, and / or the like. In some embodiments, the data repository 306c may be the same as or similar to the data repository 106, the data repositories 206a, 206b, and / or 206c, the data repositories 306a and / or 306b, and / or the like. In some embodiments, the technician device 308c may be the same as or similar to the technician device 108, technician devices 208a, 208b and / or 208c, technician devices 308a and / or 308b, and / or the like. In some embodiments, the supervisor device 310c may be the same as or similar to the supervisor device 110, supervisor device 210c, and / or the like.

[0294] As shown in FIG. 3C, at 312c, the remote computer system 304c may receive (e.g., obtain, retrieve, send requests and / or queries to the data repository 206c, and / or the like) a historical collection of multiple ECG signal portions and associated information (e.g., known arrhythmia type information, and / or the like) from, for example, the data repository 306c, as described herein.

[0295] As shown in FIG. 3C, at 314c, the remote computer system 304c may train at least one neural network of at least one classifier (e.g., an arrhythmia type classifier, and / or the like) based on a historical collection of multiple ECG signal portions and information associated therewith (e.g., known arrhythmia type information, and / or the like), as described herein.

[0296] In some embodiments, the arrhythmia type classifier may comprise at least one neural network (e.g., at least one second neural network) as described herein. Additionally or alternatively, as described herein, the at least one (second) neural network may comprise at least one of a deep neural network, a convolutional neural network, a recurrent neural network, an attention network, a fully-connected neural network, any combination thereof, and / or the like. In some embodiments, the neural network may include multiple Siamese branches (e.g., individual Siamese branches associated with individual ECG channels) as described herein.

[0297] In some embodiments, as described herein, the remote computer system 304c may train the arrhythmia type classifier by predicting, using the arrhythmia type classifier, a predicted arrhythmia type for each individual ECG signal portion of a historical collection of the plurality of ECG signal portions (or a second plurality of ECG signal portions), determining at least one error value based on the predicted arrhythmia type and known arrhythmia type information (e.g., individual annotations associated with known arrhythmia types for each individual ECG signal portion), and training (e.g., updating its weight, and / or the like) the arrhythmia type classifier based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, as described herein, the error value may include one of a prediction error or a control loss.

[0298] In some embodiments, the known arrhythmia type information may include multiple annotations, as described herein. For example, each annotation may be associated with a separate ECG signal portion of the multiple ECG signal portions, as described herein. In some embodiments, the remote computer system 304c may train the arrhythmia type classifier based on the multiple ECG signals and the multiple annotations, as described herein.

[0299] In some embodiments, as described herein, the plurality of annotations may be from multiple technicians (e.g., multiple technician devices 308c and / or the like). Additionally or alternatively, as described herein, each annotation of the plurality of annotations may be associated with an individual technician of the plurality of technicians and / or an individual ECG signal portion of the plurality of ECG signal portions. In some embodiments, as described herein, the arrhythmia type classifier may be trained separately for each technician. For example, for a first technician of the plurality of technicians, an arrhythmia type classifier (e.g., remote computer system 304c) may be trained based on a subset of the plurality of ECG signals and a plurality of annotations associated with at least one other technician of the plurality of technicians who is different from the first technician, as described herein.

[0300] In some embodiments, as described herein, each annotation may be associated with one possible arrhythmia type of the individual ECG signal and / or portion thereof (e.g., a label associated with the possible arrhythmia type, a text string identifying at least one possible arrhythmia type, and / or the like).

[0301] In some embodiments, there may be an insufficient number of ECG signal portions associated with (e.g., sensed from and / or related by the same) at least one second ECG electrode in a historical collection of multiple ECG signal portions with known arrhythmia type information (e.g., annotation labels and / or the like) to train an arrhythmia type classifier for ECG signals received from the second ECG electrodes, as described herein. Additionally or alternatively, there may be sufficient data (e.g., past ECG signal portions and / or the like) associated with (e.g., sensed from and / or related by the same) at least one first ECG electrode independent of the second ECG electrodes (e.g., an electrode of an ECG device separate from the second ECG electrodes, such as a 12-lead electrocardiogram sensor, a separate external and / or wearable cardiac monitoring device, and / or the like) to train a rhythm change classifier based on the second ECG electrodes, as described herein. In some embodiments, as described herein, the known arrhythmia type information may include a plurality of annotations, each of which may be associated with an individual ECG signal portion of a first plurality of ECG signal portions associated with a first ECG electrode. In some embodiments, as described herein, each individual ECG signal portion of a second plurality of ECG signal portions associated with a second ECG electrode may correspond to an individual ECG signal portion of the first plurality of ECG signal portions.In some embodiments, as described herein, the remote computer system 304c may train the arrhythmia type classifier by predicting a predicted arrhythmia type for each individual ECG signal portion of the second plurality of ECG signal portions using the arrhythmia type classifier, determining at least one error value (e.g., based on backpropagation, and / or the like) based on the predicted arrhythmia type and an individual annotation of the plurality of annotations associated with the individual ECG signal portion of the first plurality of ECG signal portions that corresponds to the individual ECG signal portion of the second plurality of ECG signal portions, and training the arrhythmia type classifier (e.g., updating a weight, and / or the like) based on the at least one error value.

[0302] In some embodiments, ECG signal portions associated with multiple electrodes are combined (e.g., by vector addition, vector projection, transformation, and / or the like) to form estimated ECG signal portions that may be familiar and / or suitable for review by a human user (e.g., a technician and / or the like), as described herein. For example, the historical collection of multiple ECG signal portions may include a first plurality of ECG signal portions of at least one first ECG signal based on first surface ECG activity sensed by at least one first ECG electrode, and a second plurality of ECG signal portions of at least one second ECG signal based on second surface ECG activity sensed by at least one second ECG electrode, as described herein. In some embodiments, as described herein, each ECG signal portion of the first plurality of ECG signal portions may be combined (e.g., by vector addition, vector projection, transformation, and / or the like) with an individual ECG signal portion of the second plurality of ECG signal portions to form a plurality of estimated ECG signal portions (e.g., by remote computer system 304c). In some embodiments, the known arrhythmia type information may include multiple annotations, as described herein. Additionally or alternatively, each individual annotation may be associated with a separate estimated ECG signal portion of the multiple estimated ECG signal portions, as described herein.

[0303] In some embodiments, as described herein, at least some of the multiple ECG signal portions of the historical collection may be time warped (e.g., time dilated and / or the like) to form multiple warped ECG signal portions (e.g., by the remote computer system 304c using any suitable signal processing techniques for time warping, time dilation, and / or the like).

[0304] In some embodiments, at least some of the ECG signal portions of the historical collection may be filtered, inverted, any combination thereof, and / or the like (e.g., by the remote computer system 304c) as described herein.

[0305] In some embodiments, the at least one noise signal portion may be combined (eg, by a remote computer system 304c) with at least some of the multiple ECG signal portions of the historical collection, as described herein.

[0306] In some embodiments, at least a portion of the plurality of ECG signal portions of the historical collection may be style transferred (eg, by a remote computer system 304c) as described herein.

[0307] As shown in FIG. 3C, at 316c, the remote computer system 304c may receive annotation data associated with at least one ECG signal and at least one annotation for each ECG signal, as described herein.

[0308] 3C , at 318c, the remote computer system 304c may detect arrhythmia types in the ECG signal and temporal data associated with the detected arrhythmia types using an arrhythmia type classifier, as described herein. For example, the temporal data may include at least one of a start time, a time interval, any combination thereof, and / or the like, as described herein. In some embodiments, the remote computer system 304c may determine at least one ECG signal portion associated with the detected arrhythmia type in the ECG signal based on the temporal data, as described herein.

[0309] In some embodiments, the remote computer system 304c may determine a likelihood score for the annotation based on the detected arrhythmia type. For example, the output of at least one neural network of the arrhythmia type classifier may include a confidence score (e.g., probability and / or the like) associated with each possible arrhythmia type, as described herein. For example, the arrhythmia type determined by the arrhythmia type classifier may be the arrhythmia type (e.g., probability and / or the like) with the highest confidence score. Additionally or alternatively, each annotation may be associated with a possible arrhythmia type (e.g., a label associated with the possible arrhythmia type, a text string identifying at least one possible arrhythmia type, and / or the like). In some embodiments, the likelihood score for each annotation may be a confidence score (e.g., determined by the arrhythmia type classifier for the possible arrhythmia type associated with such annotation). In some embodiments, the arrhythmia type classifier may comprise multiple neural networks, and each such neural network may output a confidence score associated with at least one possible arrhythmia type.

[0310] 3C , at 320c, the remote computer system 304c may generate at least one message based on the likelihood score for the at least one determined ECG signal portion and the at least one annotation, as described herein. For example, the message may indicate at least one of: recommending annotating the at least one determined ECG signal portion based on the detected arrhythmia type; or recommending re-evaluating annotation data associated with the at least one determined ECG signal portion based on the likelihood score; and / or the like.

[0311] In some embodiments, the remote computer system 304c may determine that the likelihood score is below a threshold. Additionally or alternatively, generating the message may include the remote computer system 304c generating at least one message indicating a recommendation to re-evaluate the annotation data associated with the at least one determined ECG signal portion based on determining that the likelihood score is below a threshold, as described herein.

[0312] As shown in FIG. 3C, at 322c, the remote computer system 304c may transmit at least a portion of a message related to at least one determined ECG signal portion to the technician device 308c, as described herein.

[0313] As shown in FIG. 3C, at 324c, the remote computer system 204c may transmit at least a portion of a message related to the at least one determined ECG signal portion to the supervisor device 310c, as described herein.

[0314] 4A , which shows an example flowchart of a process 400 a for arrhythmia monitoring, according to some embodiments. In some embodiments, one or more stages of process 400 a may be performed (e.g., fully, partially, or the like) by cardiac monitoring device 102. In some non-limiting embodiments, one or more stages of process 400 a may be performed (e.g., fully, partially, or the like) by another system, device, group of systems, or device separate from or included with cardiac monitoring device 102, such as remote computer system 104, data repository 106, technician device 108, gateway device 129, and / or the like.

[0315] 4A , in step 402a, a rhythm change classifier may be received and / or installed. For example, the rhythm change classifier (and / or its weights) may be received in cardiac monitoring device 102 and / or gateway device 129 (e.g., from remote computer system 104, data repository 106, and / or the like) and / or installed in cardiac monitoring device 102 and / or gateway device 129 (e.g., on a non-transitory computer-readable medium of cardiac monitoring device 102 and / or gateway device 129), as described herein. For example, the rhythm change classifier may comprise at least one neural network trained based on a historical collection of multiple ECG signal portions with known rhythm change information, as described herein.

[0316] In some embodiments, the remote computer system 104 may train the rhythm change classifier as described herein. Additionally or alternatively, the remote computer system 104 may communicate the trained rhythm change classifier to the cardiac monitoring device 102 and / or the gateway device 129 as described herein.

[0317] In some embodiments, the gateway device 129 may enable communication between the cardiac monitoring device 102 and the remote computer system 104 as described herein.

[0318] 4A , at stage 404a, at least one ECG signal may be received. For example, cardiac monitoring device 102 and / or gateway device 129 (e.g., a processor thereof) may receive the at least one ECG signal. In some embodiments, cardiac monitoring device 102 may receive the ECG signal via at least one ECG channel as described herein. Additionally or alternatively, cardiac monitoring device 102 may communicate (e.g., transmit) the ECG signal to gateway device 129 and / or gateway device 129 may receive the ECG signal from cardiac monitoring device 102 as described herein.

[0319] As shown in FIG. 4A, at step 406a, at least one predetermined rhythm change and / or its time may be detected. For example, cardiac monitoring device 102 and / or gateway device 129 (e.g., its processor) may detect temporal data corresponding to a predetermined rhythm change in at least one ECG signal using a rhythm change classifier as described herein. Additionally or alternatively, cardiac monitoring device 102 and / or gateway device 129 (e.g., its processor) may detect a predetermined rhythm change based on at least one ECG signal (using a rhythm change classifier) ​​as described herein.

[0320] In some embodiments, cardiac monitoring device 102 and / or gateway device 129 (e.g., its processor) may detect the predetermined rhythm change based on at least one ECG signal (e.g., using a trained rhythm change classifier). In some embodiments, cardiac monitoring device 102 may further include at least one sensor and associated sensor circuitry that senses the patient's non-ECG biometric data (which, in some embodiments, may be communicated to gateway device 129), as described herein. Additionally or alternatively, detecting the predetermined rhythm change may be further based on the patient's non-ECG biometric data, as described herein (e.g., the patient's non-ECG biometric data may be input into a neural network of the rhythm change classifier, combined with the output of the rhythm change classifier, and / or the like).

[0321] In some embodiments, detecting the predetermined rhythm change may be further based on at least one baseline ECG signal portion of the patient, as described herein.

[0322] In some embodiments, detecting the predetermined rhythm change may be further based on at least one calibration measurement of the patient, as described herein.

[0323] In some embodiments, detecting the predetermined rhythm change may be further based on at least one reference vector of the patient, as described herein.

[0324] In some embodiments, detecting the predetermined rhythm change may be further based on at least one previous ECG signal portion, as described herein.

[0325] In some embodiments, the cardiac monitoring device 102 and / or the gateway device 129 (e.g., its processor) may further determine (e.g., using a rhythm change classifier) ​​a confidence score associated with a given rhythm change based on at least one ECG signal, as described herein.

[0326] 4A , in stage 408a, at least one ECG signal portion may be determined (e.g., based on the detected time data, the detected predetermined rhythm change, and / or the like) as described herein. For example, the cardiac monitoring device 102 and / or the gateway device 129 (e.g., a processor thereof) may determine (e.g., based on the detected time data) at least one ECG signal portion associated with the detected time data that corresponds to the predetermined rhythm change in the ECG signal.

[0327] 4A , at stage 410a, the ECG signal portion may be transmitted. For example, cardiac monitoring device 102 and / or gateway device 129 may communicate (e.g., transmit, and / or the like) the determined ECG signal portion to remote computer system 104 as described herein. Additionally or alternatively, cardiac monitoring device 102 and / or gateway device 129 (e.g., a processor thereof) may detect and / or communicate indicia (e.g., flags, indicators, confidence scores, marks, metadata, time data, and / or the like) associated with the predetermined rhythm change detected (e.g., identified and / or obtained by the like) in the ECG signal portion.

[0328] In some embodiments, the cardiac monitoring device 102 and / or the gateway device 129 (e.g., its processor) may further communicate (e.g., transmit and / or the like) at least one second ECG signal portion of the ECG signal to the remote computer system 104, as described herein. Additionally or alternatively, the second ECG signal portion may be independent of the detected time data corresponding to the predetermined rhythm change in the ECG signal, as described herein.

[0329] In some embodiments, the gateway device 129 may enable communication between the cardiac monitoring device 102 and the remote computer system 104 as described herein.

[0330] In some embodiments, the remote computing system 104 may receive the determined ECG signal portion (eg, from the cardiac monitoring device 102 and / or the gateway device 129).

[0331] 4A, in step 412a, a retrained (e.g., updated and / or the like) rhythm change classifier may be received. For example, cardiac monitoring device 102 and / or gateway device 129 may receive a retrained (e.g., updated and / or the like) rhythm change classifier from remote computer system 104.

[0332] In some embodiments, the remote computing system 104 may analyze the determined ECG signal portions to classify arrhythmia type for rhythm changes in the ECG signal as described herein. For example, the remote computing system 204a may include an arrhythmia type classifier (e.g., having at least one (second) neural network trained based on a (second) historical collection of a (second) plurality of ECG signal portions with known arrhythmia type information) as described herein.

[0333] In some embodiments, the remote computer system 104 may communicate (e.g., transmit and / or the like) at least one message related to the determined ECG signal portion and / or the arrhythmia type related to the rhythm change, as described herein. For example, a message may be communicated from the remote computer system 104 to the technician device 108, as described herein.

[0334] In some embodiments, the remote computer system 104 may transmit at least one message related to the second ECG signal portion (e.g., a randomly determined second ECG signal portion as described herein, a second ECG signal portion determined to have a confidence score below a first threshold and above a second threshold, and / or the like) to the technician device 108, as described herein.

[0335] In some embodiments, the technician device 108 may receive at least one annotation associated with the ECG signal portion from a user (e.g., a technician and / or the like), for example, via input. Additionally or alternatively, the remote computer system 104 may receive annotation data associated with the annotation from the technician device 108, as described herein.

[0336] In some embodiments, the remote computer system 104 may retrain the rhythm change classifier (and / or train an updated rhythm change classifier) ​​based on a historical collection of multiple ECG signal portions with known rhythm change information, the second ECG signal portion, and its associated annotation data, as described herein (e.g., before and / or after adding the second ECG signal portion and its associated annotation data to the historical collection, as described herein).

[0337] In some embodiments, remote computer system 104 may communicate the retrained rhythm change classifier (and / or trained, updated rhythm change classifier) ​​to cardiac monitoring device 102 and / or gateway device 129 as described herein. Additionally or alternatively, a copy of the retrained rhythm change classifier (and / or trained, updated rhythm change classifier) ​​may be downloaded from remote computer system 104 and / or installed (e.g., uploaded to, written to, configured on, and / or the like) on at least one non-transitory computer-readable medium (e.g., memory, programmable circuit board, FPGA, integrated circuit, any combination thereof, and / or the like) that may be installed on and / or part of cardiac monitoring device 102 and / or gateway device 129.

[0338] In some non-limiting embodiments, process 400a may include repeating at least some steps (e.g., steps 404a-410a, 404a-412a, and / or similar steps). For example, at least some such steps may be repeated continuously, periodically, and / or similarly. For example, ECG signals may be continuously received (404a). Additionally or alternatively, the received ECG signals may be continuously analyzed using a rhythm change classifier. For example, predetermined rhythm changes and / or associated temporal data may be detected (406a) as frequently as rhythm changes occur in the ECG signal. Additionally or alternatively, ECG signal portions may be determined (408a) and / or transmitted (410a) as frequently as rhythm changes and / or associated temporal data may be detected. For example, the rhythm change classifier may be retrained (412a) periodically, continuously, and / or similarly.

[0339] 4B , which shows an example flowchart of a process 400b for arrhythmia monitoring, according to some embodiments. In some embodiments, one or more stages of process 400b may be performed (e.g., fully, partially, or the like) by cardiac monitoring device 102. In some non-limiting embodiments, one or more stages of process 400b may be performed (e.g., fully, partially, or the like) by another system, device, group of systems, or device separate from or included with cardiac monitoring device 102, such as remote computer system 104, data repository 106, technician device 108, gateway device 129, and / or the like.

[0340] 4B , in stage 402b, a historical collection of multiple ECG signal portions and information associated therewith (e.g., known rhythm change information and / or the like) may be received. For example, cardiac monitoring device 102 and / or gateway device 129 may receive the historical collection of multiple ECG signal portions and / or information associated therewith from remote computer system 104, data repository 106, and / or the like, as described herein.

[0341] 4B, in stage 404b, a rhythm change classifier may be trained. For example, cardiac monitoring device 102 and / or gateway device 129 may train a rhythm change classifier, which may be implemented by at least one non-transitory computer-readable medium (e.g., memory, a programmable circuit board, a field programmable gate array (FPGA), an integrated circuit, any combination thereof, and / or the like) installed on and / or part of cardiac monitoring device 102 and / or gateway device 129, as described herein.

[0342] In some embodiments, cardiac monitoring device 102 may be an external cardiac monitoring device for the patient, as described herein.

[0343] In some embodiments, the rhythm change classifier may comprise at least one neural network, as described herein. Additionally or alternatively, the at least one neural network may comprise at least one of a convolutional neural network, a recurrent neural network, an attention network, a fully connected neural network, any combination thereof, and / or the like, as described herein. In some embodiments, the neural network may include multiple Siamese branches, as described herein (e.g., individual Siamese branches associated with individual ECG channels).

[0344] In some embodiments, cardiac monitoring device 102 and / or gateway device 129 may use a rhythm change classifier to train the rhythm change classifier by generating predicted rhythm change information (e.g., data related to predicted rhythm changes (e.g., probabilities, confidence scores, and / or the like), data related to the absence of predicted rhythm changes (e.g., probabilities, confidence scores, and / or the like), and / or the like) for each ECG signal portion of a historical collection of multiple ECG signal portions, determining at least one error value based on the predicted rhythm change information and the known rhythm change information, and updating the rhythm change classifier based on the error value (e.g., updating its weight, and / or the like) (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss, as described herein.

[0345] In some embodiments, the historical collection of multiple ECG signal portions may include a first ECG signal portion associated with a first time and a second ECG signal portion associated with a second time after the first time, as described herein. Additionally or alternatively, the cardiac monitoring device 102 and / or the gateway device 129 may train the rhythm change classifier by predicting a predicted ECG signal portion associated with the second time based on the first ECG signal portion using a rhythm change classifier, as described herein, determining at least one error value based on the predicted ECG signal portion and the second ECG signal portion, and training the rhythm change classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss, as described herein.

[0346] In some embodiments, the historical collection of multiple ECG signal portions may include a first ECG signal portion associated with a first time and a second ECG signal portion associated with a second time, as described herein. Additionally or alternatively, the cardiac monitoring device 102 and / or the gateway device 129 may train the rhythm change classifier by predicting a predicted time associated with the second ECG signal portion based on the first ECG signal portion and the second ECG signal using the rhythm change classifier, as described herein, determining at least one error value based on the predicted time and the second time, and training the rhythm change classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss, as described herein.

[0347] In some embodiments, there may be an insufficient number of ECG signal portions in a historical collection of multiple ECG signal portions with known rhythm change information to train a rhythm change classifier to perform a desired task (e.g., detecting and / or identifying at least one predetermined rhythm change), as described herein. Additionally or alternatively, there may be sufficient data (e.g., past ECG signal portions and / or the like) to train a rhythm change classifier to perform another task (e.g., that may be related in some way to the target task), as described herein. In some embodiments, a rhythm change classifier may be trained to perform another task (e.g., counting R peaks based on an ECG signal, determining heart rate, and / or the like), as described herein. Additionally or alternatively, a rhythm change classifier may be applied to perform a target task, as described herein. For example, in some embodiments, a rhythm change classifier may be retrained using a limited amount of ECG signal portions in a historical collection of multiple ECG signal portions with known rhythm change information, as described herein, and / or the like. Additionally or alternatively, the rhythm change classifier may be used to perform another task (e.g., counting R peaks based on ECG signals, determining heart rate, and / or the like) and its output may be applied to the target task, as described herein. For example, as described herein, a processor (e.g., of cardiac monitoring device 102 and / or gateway device 129) may use a rhythm change classifier to detect at least one peak count or heart rate based on at least one ECG signal.Additionally or alternatively, as described herein, the processor (e.g., of the cardiac monitoring device 102 and / or gateway device 129) may determine that at least one detected peak count or heart rate exceeds a first threshold value for the patient (e.g., a tachycardia onset threshold) or falls below a second threshold value for the patient (e.g., a bradycardia onset threshold) (e.g., where the second threshold value for the patient may be less than the first threshold value for the patient).

[0348] In some embodiments, there may be an insufficient number of ECG signal portions associated with (e.g., sensed from and / or related by the same) the plurality of ECG electrodes of cardiac monitoring device 302a in a historical collection of ECG signal portions with known rhythm change information to train a rhythm change classifier, as described herein, for ECG signals received from the plurality of ECG electrodes of cardiac monitoring device 102. Additionally or alternatively, there may be sufficient data (e.g., past ECG signal portions, and / or the like) associated with (e.g., sensed from and / or related by the same) a second plurality of ECG electrodes independent of the plurality of ECG electrodes of cardiac monitoring device 102 (e.g., electrodes of an ECG device separate from cardiac monitoring device 102, such as a 12-lead ECG sensor, a separate external and / or wearable cardiac monitoring device, and / or the like) to train a rhythm change classifier, as described herein, based on the second plurality of ECG electrodes. In some embodiments, the rhythm change classifier may be trained based on ECG signal portions associated with (e.g., sensed from and / or the like) the second plurality of ECG electrodes, as described herein. Additionally or alternatively, the rhythm change classifier may then be applied to detect predetermined rhythm changes based on the plurality of ECG electrodes of cardiac monitoring device 102, as described herein. In some embodiments, cardiac monitoring device 102 and / or gateway device 129 may determine (e.g., calculate and / or perform the like) a transformation (e.g., vector projection and / or the like) of the ECG signal portions associated with the second plurality of ECG electrodes onto the plurality of ECG electrodes of cardiac monitoring device 102, as described herein, and may use the transformation of the ECG signal portions to train the rhythm change classifier as if the ECG signal portions were associated with (e.g., sensed from and / or the like) the plurality of ECG electrodes of cardiac monitoring device 102.

[0349] 4B, in stage 406b, at least one ECG signal may be received as described herein. For example, the cardiac monitoring device 102 and / or the gateway device 129 (e.g., a processor thereof) may receive the ECG signal. In some embodiments, cardiac monitoring device 102 may receive ECG signals via an ECG channel as described herein. Additionally or alternatively, cardiac monitoring device 102 may communicate (e.g., transmit) ECG signals to gateway device 129 and / or gateway device 129 may receive ECG signals from cardiac monitoring device 102 as described herein.

[0350] As shown in FIG. 4B, at step 408b, at least one predetermined rhythm change and / or its time may be detected. For example, cardiac monitoring device 102 and / or gateway device 129 (e.g., its processor) may detect temporal data corresponding to a predetermined rhythm change in at least one ECG signal using a rhythm change classifier as described herein. Additionally or alternatively, cardiac monitoring device 102 and / or gateway device 129 (e.g., its processor) may detect a predetermined rhythm change based on at least one ECG signal (using a rhythm change classifier) ​​as described herein.

[0351] In some embodiments, cardiac monitoring device 102 and / or gateway device 129 (e.g., its processor) may detect the predetermined rhythm change based on at least one ECG signal (e.g., using a trained rhythm change classifier). In some embodiments, cardiac monitoring device 102 may further include at least one sensor and associated sensor circuitry that senses the patient's non-ECG biometric data (which, in some embodiments, may be communicated to gateway device 129), as described herein. Additionally or alternatively, detecting the predetermined rhythm change may be further based on the patient's non-ECG biometric data, as described herein (e.g., the patient's non-ECG biometric data may be input into a neural network of the rhythm change classifier, combined with the output of the rhythm change classifier, and / or the like).

[0352] In some embodiments, detecting the predetermined rhythm change may be further based on at least one baseline ECG signal portion of the patient, as described herein.

[0353] In some embodiments, detecting the predetermined rhythm change may be further based on at least one calibration measurement of the patient, as described herein.

[0354] In some embodiments, detecting the predetermined rhythm change may be further based on at least one reference vector of the patient, as described herein.

[0355] In some embodiments, detecting the predetermined rhythm change may be further based on at least one previous ECG signal portion, as described herein.

[0356] In some embodiments, the cardiac monitoring device 102 and / or the gateway device 129 (e.g., its processor) may further determine (e.g., using a rhythm change classifier) ​​a confidence score associated with a given rhythm change based on at least one ECG signal, as described herein.

[0357] 4B , in stage 410b, at least one ECG signal portion may be determined (e.g., based on the detected time data, the detected predetermined rhythm change, and / or the like) as described herein. For example, the cardiac monitoring device 102 and / or the gateway device 129 (e.g., a processor thereof) may determine (e.g., based on the detected time data) at least one ECG signal portion associated with the detected time data that corresponds to the predetermined rhythm change in the ECG signal.

[0358] 4B , in stage 412b, the ECG signal portion may be transmitted. For example, cardiac monitoring device 102 and / or gateway device 129 may communicate (e.g., transmit, and / or the like) the determined ECG signal portion to remote computer system 104 as described herein. Additionally or alternatively, cardiac monitoring device 102 and / or gateway device 129 (e.g., a processor thereof) may detect and / or communicate indicia (e.g., flags, indicators, confidence scores, marks, metadata, time data, and / or the like) associated with the predetermined rhythm change detected (e.g., identified and / or obtained by the like) in the ECG signal portion.

[0359] In some embodiments, the cardiac monitoring device 102 and / or the gateway device 129 (e.g., its processor) may further communicate (e.g., transmit and / or the like) at least one second ECG signal portion of the ECG signal to the remote computer system 104, as described herein. Additionally or alternatively, the second ECG signal portion may be independent of the detected time data corresponding to the predetermined rhythm change in the ECG signal, as described herein.

[0360] In some embodiments, the gateway device 129 may enable communication between the cardiac monitoring device 102 and the remote computer system 104 as described herein.

[0361] In some embodiments, the remote computing system 104 may receive the determined ECG signal portion (eg, from the cardiac monitoring device 102 and / or the gateway device 129).

[0362] 4B , in stage 414b, annotation data associated with the second ECG signal portion and / or at least one annotation for the second ECG signal portion may be received, as described herein. For example, cardiac monitoring device 102 and / or gateway device 129 may receive annotation data associated with the second ECG signal portion and / or at least one annotation for the second ECG signal portion from remote computer system 104, as described herein.

[0363] In some embodiments, the remote computing system 104 may analyze the determined ECG signal portions to classify arrhythmia type for rhythm changes in the ECG signal as described herein. For example, the remote computing system 204a may include an arrhythmia type classifier (e.g., having at least one (second) neural network trained based on a (second) historical collection of a (second) plurality of ECG signal portions with known arrhythmia type information) as described herein.

[0364] In some embodiments, the remote computer system 104 may communicate (e.g., transmit and / or the like) at least one message related to the determined ECG signal portion and / or the arrhythmia type related to the rhythm change, as described herein. For example, a message may be communicated from the remote computer system 104 to the technician device 108, as described herein.

[0365] In some embodiments, the remote computer system 104 may transmit at least one message related to the second ECG signal portion (e.g., a randomly determined second ECG signal portion as described herein, a second ECG signal portion determined to have a confidence score below a first threshold and above a second threshold, and / or the like) to the technician device 108, as described herein.

[0366] In some embodiments, the technician device 108 may receive at least one annotation associated with the ECG signal portion from a user (e.g., a technician and / or the like), for example, via input. Additionally or alternatively, the remote computer system 104 may receive annotation data associated with the annotation from the technician device 108, as described herein.

[0367] In some embodiments, the remote computer system 104 may communicate annotation data associated with the second ECG signal portion and / or at least one annotation for the second ECG signal portion to the cardiac monitoring device 102 and / or the gateway device 129 as described herein.

[0368] 4B, in step 416b, the rhythm change classifier may be retrained. For example, the cardiac monitoring device 102 and / or the gateway device 129 may retrain the rhythm change classifier (and / or train an updated rhythm change classifier) ​​based on a historical collection of multiple ECG signal portions with known rhythm change information, the second ECG signal portion, and the associated annotation data, as described herein.

[0369] In some non-limiting embodiments, process 400b may include repeating at least some steps (e.g., steps 406b-412b, 406b-416b, and / or similar steps). For example, at least some such steps may be repeated continuously, periodically, and / or similarly. For example, ECG signals may be continuously received (406b). Additionally or alternatively, the received ECG signals may be continuously analyzed using a rhythm change classifier. For example, predetermined rhythm changes and / or associated temporal data may be detected (408b) as frequently as rhythm changes occur in the ECG signal. Additionally or alternatively, ECG signal portions may be determined (410b) and / or transmitted (412b) as frequently as rhythm changes and / or associated temporal data may be detected. For example, the annotations and / or their associated ECG signal portions may be received (414b) and / or the rhythm change classifier may be retrained (416b) periodically, continuously, and / or the like.

[0370] 4C , which shows an example flowchart of a process 400 c for arrhythmia monitoring, according to some embodiments. In some embodiments, one or more stages of process 400 c may be performed (e.g., fully, partially, or the like) by remote server 104. In some non-limiting embodiments, one or more stages of process 400 c may be performed (e.g., fully, partially, or the like) by another system, device, group of systems, or group of devices separate from or included with remote server 104, such as cardiac monitoring device 102, data repository 106, technician device 108, gateway device 129, and / or the like.

[0371] 4C , at stage 402a, a historical collection of multiple ECG signal portions and information associated therewith (e.g., known arrhythmia type information and / or the like) may be received. For example, the remote computer system 104 may receive (e.g., obtain, retrieve, send requests and / or queries to be communicated to the data repository 106, and / or the like) the historical collection of multiple ECG signal portions and information associated therewith (e.g., known arrhythmia type information and / or the like) as described herein from, for example, the data repository 106.

[0372] 4C , in step 404c, an arrhythmia type classifier may be trained as described herein. For example, the remote computer system 104 may train at least one neural network of at least one classifier (e.g., an arrhythmia type classifier, and / or the like) based on a historical collection of multiple ECG signal portions and information associated therewith (e.g., known arrhythmia type information, and / or the like) as described herein.

[0373] In some embodiments, the arrhythmia type classifier may comprise at least one neural network (e.g., at least one second neural network) as described herein. Additionally or alternatively, as described herein, the at least one (second) neural network may comprise at least one of a deep neural network, a convolutional neural network, a recurrent neural network, an attention network, a fully-connected neural network, any combination thereof, and / or the like. In some embodiments, the neural network may include multiple Siamese branches (e.g., individual Siamese branches associated with individual ECG channels) as described herein.

[0374] In some embodiments, the remote computer system 104 may train the arrhythmia type classifier by predicting a predicted arrhythmia type for each individual ECG signal portion of a historical collection of a plurality of ECG signal portions (or a second plurality of ECG signal portions) using the arrhythmia type classifier as described herein, determining at least one error value based on the predicted arrhythmia type and known arrhythmia type information (e.g., individual annotations associated with known arrhythmia types for each individual ECG signal portion), and training the arrhythmia type classifier (e.g., updating its weight, and / or the like) based on the error value (e.g., using backpropagation, and / or the like). In some embodiments, the error value may include one of a prediction error or a control loss, as described herein.

[0375] In some embodiments, the known arrhythmia type information may include multiple annotations, as described herein. For example, each annotation may be associated with a separate ECG signal portion of the multiple ECG signal portions, as described herein. In some embodiments, the remote computer system 104 may train an arrhythmia type classifier based on the multiple ECG signals and the multiple annotations, as described herein.

[0376] In some embodiments, as described herein, the plurality of annotations may be from multiple technicians (e.g., multiple technician devices 108 and / or the like). Additionally or alternatively, as described herein, each annotation of the plurality of annotations may be associated with an individual technician of the plurality of technicians and / or an individual ECG signal portion of the plurality of ECG signal portions. In some embodiments, as described herein, the arrhythmia type classifier may be trained separately for each technician. For example, for a first technician of the plurality of technicians, an arrhythmia type classifier (e.g., remote computer system 104) may be trained based on a subset of the plurality of ECG signals and a plurality of annotations associated with at least one other technician of the plurality of technicians who is different from the first technician, as described herein.

[0377] In some embodiments, as described herein, each annotation may be associated with one possible arrhythmia type of the individual ECG signal and / or portion thereof (e.g., a label associated with the possible arrhythmia type, a text string identifying at least one possible arrhythmia type, and / or the like).

[0378] In some embodiments, there may be an insufficient number of ECG signal portions associated with (e.g., sensed from and / or related by the same) at least one second ECG electrode in a historical collection of multiple ECG signal portions with known arrhythmia type information (e.g., annotation labels and / or the like) to train an arrhythmia type classifier for ECG signals received from the second ECG electrodes, as described herein. Additionally or alternatively, there may be sufficient data (e.g., past ECG signal portions and / or the like) associated with (e.g., sensed from and / or related by the same) at least one first ECG electrode independent of the second ECG electrodes (e.g., an electrode of an ECG device separate from the second ECG electrodes, such as a 12-lead electrocardiogram sensor, a separate external and / or wearable cardiac monitoring device, and / or the like) to train a rhythm change classifier based on the second ECG electrodes, as described herein. In some embodiments, as described herein, the known arrhythmia type information may include a plurality of annotations, each of which may be associated with an individual ECG signal portion of a first plurality of ECG signal portions associated with a first ECG electrode. In some embodiments, as described herein, each individual ECG signal portion of a second plurality of ECG signal portions associated with a second ECG electrode may correspond to an individual ECG signal portion of the first plurality of ECG signal portions.In some embodiments, as described herein, the remote computer system 104 may train the arrhythmia type classifier by predicting a predicted arrhythmia type for each individual ECG signal portion of the second plurality of ECG signal portions using the arrhythmia type classifier, determining at least one error value (e.g., based on backpropagation, and / or the like) based on the predicted arrhythmia type and an individual annotation of the plurality of annotations associated with the individual ECG signal portion of the first plurality of ECG signal portions that corresponds to the individual ECG signal portion of the second plurality of ECG signal portions, and training the arrhythmia type classifier (e.g., updating a weight, and / or the like) based on the at least one error value.

[0379] In some embodiments, ECG signal portions associated with multiple electrodes are combined (e.g., by vector summation, vector projection, transformation, and / or the like) to form estimated ECG signal portions that may be familiar and / or suitable for review by a human user (e.g., a technician and / or the like), as described herein. For example, the historical collection of multiple ECG signal portions may include a first plurality of ECG signal portions of at least one first ECG signal based on first surface ECG activity sensed by at least one first ECG electrode, and a second plurality of ECG signal portions of at least one second ECG signal based on second surface ECG activity sensed by at least one second ECG electrode, as described herein. In some embodiments, each ECG signal portion of the first plurality of ECG signal portions may be combined (e.g., by vector summation, vector projection, transformation, and / or the like) with an individual ECG signal portion of the second plurality of ECG signal portions to form multiple estimated ECG signal portions (e.g., by remote computer system 104), as described herein. In some embodiments, the known arrhythmia type information may include multiple annotations, as described herein. Additionally or alternatively, each individual annotation may be associated with a separate estimated ECG signal portion of the multiple estimated ECG signal portions, as described herein.

[0380] In some embodiments, as described herein, at least some of the multiple ECG signal portions of the historical collection may be time warped (e.g., time dilated and / or the like) to form multiple warped ECG signal portions (e.g., by the remote computer system 104 using any suitable signal processing technique for time warping, time dilation, and / or the like).

[0381] In some embodiments, at least some of the ECG signal portions of the historical collection may be filtered, inverted, any combination thereof, and / or the like (e.g., by the remote computer system 104) as described herein.

[0382] In some embodiments, the at least one noise signal portion may be combined (eg, by a remote computer system 104) with at least some of the multiple ECG signal portions of the historical collection, as described herein.

[0383] In some embodiments, at least a portion of the plurality of ECG signal portions of the historical collection may be style transferred (eg, by a remote computer system 104) as described herein.

[0384] 4C, in step 406c, annotation data associated with at least one ECG signal and at least one annotation for each ECG signal may be received. For example, as described herein, the remote computer system 104 may receive (e.g., from the technician device 108) annotation data associated with at least one ECG signal and at least one annotation for each ECG signal.

[0385] 4C , at step 408c, the arrhythmia type may be detected (e.g., classified) as described herein. For example, the remote computer system 104 may use an arrhythmia type classifier as described herein to detect arrhythmia types in the ECG signal and time data associated with the detected arrhythmia types. For example, the time data may include at least one of a start time, a time interval, any combination thereof, and / or the like as described herein. In some embodiments, the remote computer system 104 may determine at least one ECG signal portion associated with the detected arrhythmia type in the ECG signal based on the time data as described herein.

[0386] In some embodiments, the remote computer system 104 may determine a likelihood score for the annotation based on the detected arrhythmia type. For example, the output of at least one neural network of the arrhythmia type classifier may include a confidence score (e.g., probability and / or the like) associated with each possible arrhythmia type, as described herein. For example, the arrhythmia type determined by the arrhythmia type classifier may be the arrhythmia type (e.g., probability and / or the like) with the highest confidence score. Additionally or alternatively, each annotation may be associated with a possible arrhythmia type (e.g., a label associated with the possible arrhythmia type, a text string identifying at least one possible arrhythmia type, and / or the like). In some embodiments, the likelihood score for each annotation may be a confidence score (e.g., determined by the arrhythmia type classifier for the possible arrhythmia type associated with such annotation). In some embodiments, the arrhythmia type classifier may comprise multiple neural networks, and each such neural network may output a confidence score associated with at least one possible arrhythmia type.

[0387] 4C , at stage 410c, at least one message may be generated (e.g., based on the determined ECG signal portion, the likelihood score, and / or the like). For example, the remote computer system 104 may generate the at least one message based on the likelihood score for the at least one determined ECG signal portion and the at least one annotation, as described herein. For example, the message may indicate at least one of: recommending annotating the at least one determined ECG signal portion based on the detected arrhythmia type; or recommending re-evaluating annotation data associated with the at least one determined ECG signal portion based on the likelihood score; and / or the like.

[0388] In some embodiments, the remote computer system 104 may determine that the likelihood score is below a threshold. Additionally or alternatively, generating the message may include the remote computer system 104 generating at least one message indicating a recommendation to re-evaluate the annotation data associated with the at least one determined ECG signal portion based on determining that the likelihood score is below a threshold, as described herein.

[0389] 4C, in step 412c, a message may be transmitted. For example, the remote computer system 104 may transmit at least a portion of a message associated with the at least one determined ECG signal portion to the technician device 108, as described herein. Additionally or alternatively, the remote computer system 104 may transmit at least a portion of a message associated with the at least one determined ECG signal portion to the supervisor device 110, as described herein.

[0390] In some non-limiting embodiments, process 400c may include repeating at least some steps (e.g., steps 406c-412c, 402c-412c, and / or similar steps). For example, at least some such steps may be repeated continuously, periodically, and / or similarly. For example, ECG signals and / or annotations associated therewith may be continuously received (406c), periodically and / or similarly as frequently as a technician provides such annotations. Additionally or alternatively, arrhythmia types may be classified (408c) and / or messages may be continuously generated (410c) and / or transmitted (412c) periodically and / or similarly as frequently as the ECG signals and / or annotations are received. For example, a historical collection of ECG signal portions may be continuously, periodically, and / or similarly updated (e.g., by adding new ECG signal portions and / or known arrhythmia type information, and / or the like). Additionally or alternatively, the arrhythmia type classifier may be continuously (re)trained (404c) periodically and / or similarly as frequently as updating the historical collection of ECG signal portions.

[0391] Referring now to FIG. 5A, FIG. 5A shows an exemplary diagram of a neural network 500a of an exemplary rhythm change classifier. As shown in FIG. 5A, the neural network 500a may comprise at least one input layer 502a, as described herein. Additionally or alternatively, the neural network 500a may comprise at least one output layer 508a, as described herein. In some embodiments, the neural network 500a may comprise at least one hidden layer (e.g., a first hidden layer 504a, a last hidden layer 506a, and / or the like), as described herein. In some embodiments, the hidden layer (e.g., a first hidden layer 504a, a last hidden layer 506a, and / or the like), as described herein, may comprise multiple convolutional layers.

[0392] Referring now to FIG. 5B, FIG. 5B illustrates an exemplary diagram neural network 500b of an exemplary arrhythmia type classifier. As shown in FIG. 5B, as described herein, neural network 500b may comprise at least one input layer 502b. Additionally or alternatively, as described herein, neural network 500b may comprise at least one output layer 508b. In some embodiments, as described herein, neural network 500b may comprise at least one hidden layer (e.g., first hidden layer 504b, last hidden layer 506b, and / or the like). In some embodiments, as described herein, hidden layers (e.g., first hidden layer 504b, last hidden layer 506b, and / or the like) may comprise multiple convolutional layers.

[0393] 6A-6E, which illustrate exemplary ECG signal portions. As shown in FIG. 6A, a first ECG signal portion 600a may comprise a first ECG signal from a first channel 602a and a second ECG signal from a second channel 604a. In some embodiments, the first ECG signal portion 600a may indicate normal sinus rhythm (NSR) in both channels.

[0394] 6B, the second ECG signal portion 600b may comprise a portion of the first ECG signal from the first channel 602b and the second ECG signal from the second channel 604b. In some embodiments, the second ECG signal portion 600b may show rhythm changes in both channels between 18 and 19 seconds.

[0395] 6C, the third ECG signal portion 600c may comprise a portion of the first ECG signal from the first channel 602c and the second ECG signal from the second channel 604c. In some embodiments, the third ECG signal portion 600c may indicate a therapy treatment (e.g., a 150 J therapeutic shock) between 62 and 63 seconds on both channels.

[0396] 6D, the fourth ECG signal portion 600d may comprise portions of the first ECG signal from the first channel 602d and the second ECG signal from the second channel 604d. In some embodiments, the fourth ECG signal portion 600d may show rhythm changes within a highlight period 606d in both channels.

[0397] 6E, the fifth ECG signal portion 600e may comprise a portion of the ECG signal from channel 604e. In some embodiments, the fifth ECG signal portion 600e may show a rhythm change between 93 seconds and 94 seconds.

[0398] 7, which is a diagram of example components of a device 700. The device 700 may correspond to one or more of the cardiac monitoring device 102, the remote computer system 104, the data repository 106, the technician device 108, the supervisor device 110, and / or the gateway device 129. In some non-limiting embodiments, the cardiac monitoring device 102, the remote computer system 104, the data repository 106, the technician device 108, the supervisor device 110, and / or the gateway device 129 may comprise at least one of the devices 700 and / or at least one component of the devices 700. As shown in FIG. 7, the device 700 may comprise a bus 702, a processor 704, a memory 706, a storage component 708, an input component 710, an output component 712, and a communication interface 714.

[0399] The bus 702 may comprise components that permit communication among the components of the device 700. In some non-limiting embodiments, the processor 704 may be implemented in hardware, firmware, or a combination of hardware and software. For example, the processor 704 may include a processor (e.g., a central processing unit (CPU), a graphics processing unit (GPU), an accelerated processing unit (APU), and / or the like), a microprocessor, a digital signal processor (DSP), and / or any processing component (e.g., a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), and / or the like), and / or the like that can be programmed to perform a function. The memory 706 may include random access memory (RAM), read-only memory (ROM), and / or another type of dynamic or static storage device (e.g., flash memory, magnetic memory, optical memory, and / or the like) that stores information and / or instructions for use by the processor 704.

[0400] The storage component 708 may store information and / or software related to the operation and use of the device 700. For example, the storage component 708 may include a hard disk (e.g., a magnetic disk, optical disk, magneto-optical disk, solid-state disk, and / or the like), a compact disk (CD), a digital versatile disk (DVD), a floppy disk, a cartridge, a magnetic tape, and / or another type of computer-readable medium along with a corresponding drive.

[0401] The input components 710 may include components (e.g., a touch panel display, a keyboard, a keypad, a mouse, buttons, switches, a microphone, a camera, and / or the like) that allow the device 700 to receive information via user input and / or the like. Additionally or alternatively, the input components 710 may include sensors for detecting information (e.g., a global positioning system (GPS) component, an accelerometer, a gyroscope, an actuator, and / or the like). The output components 712 may include components (e.g., a display, a speaker, one or more light-emitting diodes (LEDs), and / or the like) that provide output information from the device 700.

[0402] The communication interface 714 may include transceiver-like components (e.g., a transceiver, a separate receiver and transmitter, and / or the like) that enable the device 700 to communicate with other devices, such as via a wired connection, a wireless connection, or a combination of wired and wireless connections. The communication interface 714 may allow the device 700 to receive information from other devices and / or provide information to other devices. For example, the communication interface 714 may include an Ethernet interface, an optical interface, a coaxial interface, an infrared interface, a radio frequency (RF) interface, a universal serial bus (USB) interface, a Wi-Fi interface, a Bluetooth interface, a Zigbee interface, a cellular network interface, and / or the like.

[0403] The device 700 may perform one or more processes described herein. The device 700 may perform these processes based on the processor 704 executing software instructions stored by a computer-readable medium, such as the memory 706 and / or the storage component 708. A computer-readable medium (e.g., a non-transitory computer-readable medium) is defined herein as a non-transitory memory device. A non-transitory memory device includes memory space located within a single physical storage device and memory space distributed across multiple physical storage devices.

[0404] Software instructions may be read into memory 706 and / or storage component 708 from another computer-readable medium or from another device via communications interface 714. When executed, the software instructions stored in memory 706 and / or storage component 708 can cause processor 704 to perform one or more operations described herein. Additionally or alternatively, hardwired circuitry may be used in place of, or in combination with, software instructions to perform one or more operations described herein. Thus, the embodiments described herein are not limited to any specific combination of hardware circuitry and software.

[0405] The number and arrangement of components shown in Figure 7 are provided by way of example. In some non-limiting embodiments, device 700 may include additional, fewer, different, or differently arranged components than those shown in Figure 7. Additionally or alternatively, a set of components (e.g., one or more components) of device 700 may perform one or more functions described as being performed by another set of components of device 700.

[0406] Referring now to FIG. 8, FIG. 8 illustrates an exemplary cardiac monitoring device, e.g., arrhythmia and bodily fluid monitoring system, including a physiological monitoring device 810, hereinafter “sensor,” and a wearable patch 860 that positions the sensor on or near the surface of a body (e.g., a patient). Additionally, the system may include a portable data transmission device (gateway) 830 that can continuously transmit data acquired by the sensor 810 to a remote computer system (e.g., one or more servers 850) for processing and / or analysis. Thus, for example, the gateway device 830 may transmit data received from the sensor 810 to the server 850 with little or no delay or latency. Thus, in the context of data transmission between the device 810 and the server 850, “continuous” in this disclosure includes continuous (without interruption) or near-continuous, i.e., within one minute of a device measurement completion and / or device event occurrence. Continuity may also be achieved by repeated successive burst transmissions, such as high-speed transmissions. Similarly, according to the present disclosure, the term "immediately" includes occurring or taking place at once or nearly immediately, i.e., within one minute, after the completion of the measurement and / or occurrence of the event that occurred on the device.

[0407] Additionally, in the context of physiological data acquisition by device 810, "continuous" also includes the clinically continuous uninterrupted collection of sensor data, such as ECG data and / or accelerometer data. In this case, short interruptions in data acquisition of up to one second several times per hour, or longer interruptions of several minutes several times per day, may be considered "continuous." With regard to latency resulting from a continuous approach as described herein, this relates to an overall response time budget that may equate to an overall response time (e.g., the time from when an event is detected to when a notification about that event is issued) of between about five and about fifteen minutes. Thus, transmission and reception latency is measured in minutes.

[0408] Furthermore, the wearable devices described herein are configured for extended and / or prolonged use or wear by, or attachment or connection to, a patient. For example, a device such as described herein may be capable of being used or worn by, or attached or connected to, a patient for, e.g., up to 24 hours or more (e.g., for weeks, months, or years) without substantial interruption. In some examples, such a device can be removed for a period of time before resuming use, wear, attachment, or connection to a patient, e.g., to replace batteries, perform technical service, update device software or firmware, and / or take a shower or engage in other activities, without departing from the scope of the examples described herein.

[0409] In some embodiments, transmission of data / signals 820 between the sensors 810 and the gateway device 830 may be one-way (e.g., from the sensors 810 to the gateway device 830), or the transmission may be bidirectional. Similarly, transmission of data / signals 840 between the gateway device 830 and the server 850 may be one-way (e.g., from the gateway device 830 to the server 850) or bidirectional. The system may also include a charger (not shown) for powering the system's electronics.

[0410] In some embodiments, the sensor 810 continuously monitors, records, and transmits physiological data about the wearer of the sensor 810 to the gateway device 830. In particular, the sensor 810 may not stop monitoring and / or recording further data while transmitting already acquired data to the gateway device 830. In other words, in some embodiments, both the monitoring / recording process and the transmitting process occur simultaneously, or at least nearly simultaneously.

[0411] As another example, if the sensor 810 suspends monitoring and / or recording of further data while transmitting already acquired data to the gateway device 830, the sensor 810 may subsequently resume monitoring and / or recording of further data before all of the already acquired data has been transmitted to the gateway device 830. In other words, the period of suspension for monitoring and / or recording may be small compared to the time required to transmit the already acquired data (e.g., including values ​​between about 0% and about 80%, about 0% and about 60%, about 0% and about 40%, about 0% and about 20%, about 0% and about 10%, about 0% and about 5%, and sub-ranges therebetween), facilitating near-continuous monitoring and / or recording of further data during transmission of the already acquired physiological data. For example, in one particular scenario, if the duration of the measurement time is about 2 minutes, the period of suspension or suspension of subsequent monitoring and / or recording of measurement data may range from a few milliseconds to about 1 minute. Examples of reasons for pausing or suspending such data include allowing for the completion of certain data integrity tests and / or other online tests of previously acquired data, as described in more detail below. If there is a problem with the previously measured data, the sensor 810 may notify the patient and / or a remote technician of the problem so that appropriate adjustments can be made.

[0412] In some embodiments, the bandwidth of the link 820 between the sensor 810 and the gateway device 830 may be larger, and in some embodiments significantly larger, than the bandwidth of the acquired data transmitted over the link 820 (e.g., burst transmission). Such embodiments ameliorate issues that may arise from link interruptions, periods of reduced / absent reception, etc. In some embodiments, when transmission resumes after an interruption, the resumption may be in the form of last-in, first-out (LIFO). The gateway device 830 may be configured to operate in a store-and-forward mode, where data received from the sensor 810 is first stored in the gateway device's onboard memory and then forwarded to an external server. This mode is useful, for example, when the link with the server is temporarily unavailable. In some embodiments, the gateway device 830 functions as a pipeline, allowing data from the sensor 810 to pass immediately to the server. In a further example, data from the sensor may be compressed using data compression techniques to reduce memory requirements as well as transmission time and power consumption.

[0413] In some embodiments, the sensor 810 may be configured to monitor, record, and transmit some data in a continuous or near-continuous manner, while monitoring, recording, and transmitting some other data in a non-continuous manner (e.g., periodic, aperiodically, etc.), as described above. For example, the sensor 810 may be configured to record and transmit ECG data continuously or near-continuously, while the radio frequency (RF)-based measurements and / or transmissions may be periodic. For example, the ECG data may be transmitted to the gateway device 830 (and thereafter the server 850) continuously or near-continuously as more ECG data is being recorded, and the RF-based measurements may be transmitted once the measurement process is complete.

[0414] The monitoring and / or recording of physiological data by the sensor 810 may be periodic and, in some embodiments, may be accomplished on a scheduled basis (i.e., periodically) without delay or latency during transmission of already acquired data to the gateway device 830. For example, the sensor 810 may acquire physiological data from the patient (i.e., the wearer of the sensor 810) in a periodic manner, as described above, and transmit the data to the gateway device 830 in a continuous manner, as described above.

[0415] The sensors 810 may transmit acquired data to the server 850 instead of or in addition to transmitting data to the gateway device 830. The sensors 810 may also store some or all of the acquired physiological data. In some embodiments, transmission of data from the sensors 810 to the gateway device 830 may be accomplished via a wireless (e.g., Bluetooth, etc.) and / or wired connection, e.g., 820. Transmission of data from the gateway device 830 to the server 850 may be accomplished via a wireless (e.g., Bluetooth-to-TCP / IP access point communication, Wi-Fi, cellular, etc.) and / or wired connection, e.g., 840.

[0416] As mentioned above, in some embodiments, data and / or signal transmission occurs over two links 820, 840: a link between the sensor 810 and the gateway device 830 (e.g., a Bluetooth link) and a link between the gateway device 830 and the server 850 (e.g., Wi-Fi, cellular). The Bluetooth link can be a connection bus for communication between the sensor 810 and the server 850, used to pass commands, information about the state of the sensor 810's microprocessor, measurement data, etc. In some embodiments, the sensor 810's microprocessor may initiate communication with the server 850 (and / or the gateway device 830), and once a connection is established, the server 850 may be configured to initiate some or all other communications. In some embodiments, the gateway device 830 may conserve power available to the sensor 810, the device 830, and / or the server 850. For example, one or both of the links 820, 840 may enter a power-saving mode (e.g., sleep mode, off state, etc.) when the connection between the respective devices / servers is unavailable. As another example, transmission of data may be at least temporarily suspended if the link quality (e.g., available bandwidth) is insufficient for at least satisfactory transmission of data. In such an embodiment, the gateway device 830 may function as a master device in relation to one or both of the sensors 810 and the server 850.

[0417] In some embodiments, the gateway device 830 may be thought of as a simple pipe, and the sensor-gateway device-server path may be defined as a single link; i.e., link performance may depend on bottlenecks between the sensor-gateway device and the gateway device-server link. In some embodiments, because the gateway device is carried by the patient in close proximity to the device, at least the primary bottleneck may be the gateway device-server link, while the gateway device-server link (e.g., cellular or WiFi coverage) is expected to be variable. In some embodiments, because the transmitted data is processed (e.g., with some latency) and used to display notifications (e.g., rather than being presented online at a monitoring center), a “best-effort delivery” quality of service for the Bluetooth link and / or TCP / IP link may be sufficient. In some embodiments, a single gateway device 830 may serve multiple sensors; i.e., multiple sensors may be connected to a single gateway device 830 via respective links. In some embodiments, there may be multiple gateway devices serving one or more sensors; i.e., each sensor of one or more sensors may be connected to multiple gateway devices via respective links.

[0418] In some embodiments, the transmission links 820, 840 may be configured to withstand coexistence interference from similar devices in the vicinity and from other devices using the same RF band (e.g., Bluetooth, Cellular, WiFi). Interference issues may be addressed by using standard Bluetooth and / or standard TCP / IP protocols and adding a cyclic redundancy check to the transmitted data. Furthermore, to maintain the security of the wireless signals and data, in some embodiments, data transfer between the sensor and the server may be performed using a proprietary protocol. For example, a TCP / IP link may be secured using the SSL protocol, while a Bluetooth link may be encrypted. As another example, UDP / HTTP may be used for secure transmission of data. In some embodiments, only raw binary data may be transmitted without identifying the patient.

[0419] Examples of types of physiological data that the arrhythmia and fluid monitoring sensor 810 monitors and / or acquires from a patient wearing the sensor 810 include one or more of electrocardiogram (ECG) data, chest impedance, heart rate, respiratory rate, physical activity (e.g., movement), and patient posture. In some embodiments, the physiological data may be acquired and / or transmitted by the sensor 810 to the gateway device 830 or the server 850 continuously, periodically, or as directed by a received signal (e.g., as directed by a signal received from the gateway device 830 and / or the server 850). For example, the sensor wearer or another party (e.g., a health professional) may activate the sensor 810, which can automatically begin monitoring and / or recording any one of the aforementioned physiological parameters without further input from the wearer or party. The sensor 810, or the arrhythmia and fluid monitoring system in general, may require further input (e.g., selection of settings identifying the physiological parameters to be measured) before beginning to monitor and / or record physiological data. In any event, once monitoring and / or recording has commenced, the sensor 810 may transmit acquired data to the gateway device 830 and / or the server 850, for example, in at least a continuous manner as described above.

[0420] In some embodiments, one or more of the aforementioned physiological parameters may be measured periodically, and the sensor 810 may transmit the measurements to the gateway device 830 in at least a continuous manner as they are acquired. For example, the periodic measurements may proceed as scheduled, with transmission to the gateway device 830 occurring with little delay or latency after the data is acquired.

[0421] In some embodiments, the sensor 810, or the arrhythmia and fluid monitoring system in general, may be configured to operate some, but not all, of the available features described above. For example, the sensor 810 may monitor and / or acquire one or more of ECG data, thoracic impedance, heart rate, respiratory rate, physical activity (e.g., movement), patient posture, etc., but not others. For example, the sensor may monitor and / or acquire data such as ECG data, but not respiratory rate, physical activity (e.g., movement), patient posture, etc. Such embodiments may be realized, for example, by including controls in the sensor and / or system that separately control the sensor / system components responsible for the features. For example, the arrhythmia and fluid monitoring system may include controls (e.g., power buttons) that separately control the accelerometer and ECG components of the sensor. By switching on the accelerometer power control and switching off the ECG power control, some embodiments may enable monitoring and / or acquisition of data related to respiratory rate, physical activity, and patient posture while ceasing monitoring and / or acquisition of ECG data.

[0422] In some embodiments, an adhesive patch 860 may be used to attach the sensor 810 to a patient's body surface. FIGS. 9A-9E illustrate a sensor 970 as disclosed herein. The patch 910 attaches the sensor 970 to the patient's body, or at least holds the sensor 970 in close proximity to the skin of the body. An illustration of how the sensor 970 may be attached to the patch 910, according to some embodiments, is shown. The patch 910 may include a patch frame 930 (e.g., a plastic frame) that defines the boundary of the area of ​​the patch 910 that houses the sensor 970. The patch 910 may be disposable (e.g., a single-use or multi-use patch) and may be made of a biocompatible nonwoven material. In some embodiments, the sensor 970 may be designed for long-term use. In such embodiments, the connection between the patch 910 and the sensor 970 may be reversible, i.e., the sensor 970 may be removably attached to the patch 910. For example, the sensor 970 may include components such as snap-in clips 940 that secure the sensor 970 to the patch 910 (e.g., patch frame 930) during attachment (and release the sensor 970 from the patch when separation is desired). The sensor 970 may also include positioning tabs 960 that facilitate the attachment process between the sensor 970 and the patch 910. In...

Claims

1. 1. An arrhythmia monitoring system, comprising:

1. An external cardiac monitoring device for a patient, comprising: a plurality of surface electrocardiogram (ECG) electrodes configured to sense surface ECG activity of the patient; an ECG processing circuit configured to process the surface ECG activity of the patient and provide at least one ECG signal for the patient on at least one ECG channel; at least one first processor operatively connected to the at least one ECG channel, the at least one processor comprising: receiving the at least one ECG signal received via the at least one ECG channel; transmitting the at least one ECG signal; at least one first processor configured to an external cardiac monitoring device for a patient, 1. A gateway device, comprising: a non-transitory computer-readable medium including a rhythm change classifier, the rhythm change classifier including at least one neural network trained based on a historical collection of ECG signal portions having known rhythm change information; at least one second processor operably connected to the non-transitory computer-readable medium, the at least one second processor comprising: receiving the at least one ECG signal from the external cardiac monitoring device; using the rhythm change classifier to detect temporal data corresponding to predetermined rhythm changes in the at least one ECG signal, the temporal data including at least one of a start time, a time interval, or any combination thereof; determining, based on the detected time data, at least one ECG signal portion associated with the detected time data that corresponds to the predetermined rhythm change in the at least one ECG signal; transmitting at least one determined portion of the ECG signal to a remote computer system; at least one second processor configured to a gateway device having An arrhythmia monitoring system comprising:

2. the at least one determined ECG signal portion includes a plurality of determined ECG signal portions, the remote computer system being in communication with the external cardiac monitoring device, the remote computer system: receiving the plurality of determined ECG signal portions from the external cardiac monitoring device; 2. The arrhythmia monitoring system of claim 1, configured to analyze each individually determined ECG signal portion of the plurality of determined ECG signal portions to classify an individual class for each individually determined ECG signal portion, the classes for at least two individually determined ECG signal portions including a first class.

3. 3. The arrhythmia monitoring system of claim 2, wherein the remote computer system is further configured to transmit at least one message related to the at least two individually determined ECG signal portions to a computing device associated with a technician.

4. 4. The arrhythmia monitoring system of claim 3, wherein the computing device associated with the technician is configured to display a graphical user interface for collective review of the at least two individually determined ECG signal portions of the first class.

5. 3. The arrhythmia monitoring system of claim 2, wherein analyzing each individual determined ECG signal portion of the plurality of determined ECG signal portions to classify the individual class for each individual determined ECG signal portion includes bucketing the plurality of determined ECG signal portions into a plurality of buckets, and the first class includes a first bucket of the plurality of buckets.

6. Bucketing the plurality of determined ECG signal portions into a plurality of buckets may include: the output of the neural network, a similarity of characteristics of the plurality of determined ECG signal portions; a similarity of vector representations of the plurality of determined ECG signal portions; or at least one of a combination thereof 6. The arrhythmia monitoring system of claim 5, further comprising grouping the plurality of determined ECG signal portions based on:

7. The arrhythmia monitoring system of claim 1 , wherein the external cardiac monitoring device comprises a wearable patch.

8. The arrhythmia monitoring system of claim 1 , wherein the external cardiac monitoring device comprises a wearable defibrillator.

9. The remote computer system is in communication with the gateway device, the remote computer system comprising: receiving the at least one determined ECG signal portion from the external cardiac monitoring device; analyzing the at least one determined ECG signal portion to classify an arrhythmia type of the rhythm change in the at least one ECG signal. The arrhythmia monitoring system of claim 1 , configured to:

10. 2. The arrhythmia monitoring system of claim 1, wherein the at least one ECG channel comprises at least a first ECG channel and a second ECG channel, the at least one ECG signal comprises a first ECG signal associated with the first ECG channel and a second ECG signal associated with the second ECG channel, and the first individual ECG signal is orthogonal to the second individual ECG signal.

11. further comprising at least one sensor and associated sensor circuitry configured to sense non-ECG biometric data of the patient; 2. The arrhythmia monitoring system of claim 1, wherein the at least one second processor is further configured to detect the predetermined rhythm change based on the at least one ECG signal and the non-ECG biometric data of the patient using the rhythm change classifier.

12. 12. The arrhythmia monitoring system of claim 11, wherein the at least one sensor comprises at least one of an accelerometer, a heart sound detector, or a combination thereof, and the non-ECG biometric data comprises at least one of acceleration data, heart sound data, or any combination thereof.

13. Detecting the predetermined rhythm change includes: at least one baseline ECG signal portion of the patient; at least one reference vector of said patient; at least one calibration measurement of the patient, the at least one calibration measurement being based on at least one second ECG signal from second surface ECG activity sensed by a second plurality of ECG electrodes, the second plurality of ECG electrodes being independent from the plurality of ECG electrodes of the external cardiac monitoring device; at least one previous ECG signal portion; The arrhythmia monitoring system of claim 11 further based on at least one of:

14. the at least one ECG channel comprises a plurality of ECG channels, and the at least one ECG signal comprises at least one individual ECG signal associated with each individual ECG channel of the plurality of ECG channels; the at least one neural network has a plurality of Siamese branches, each individual Siamese branch of the plurality of Siamese branches being associated with an individual ECG channel of the plurality of ECG channels; The arrhythmia monitoring system of claim 1 , wherein the at least one neural network further comprises at least one additional layer connected to the plurality of Siamese branches.

15. 15. The arrhythmia monitoring system of claim 14, wherein each Siamese branch of the plurality of Siamese branches has a plurality of convolutional layers, and a dimension of each of the plurality of convolutional layers of each individual Siamese branch is the same as a dimension of each of the plurality of convolutional layers of each other Siamese branch.

16. 15. The arrhythmia monitoring system of claim 14, wherein the plurality of ECG channels comprises a first ECG channel and a second ECG channel, and the at least one ECG signal comprises a first individual ECG signal associated with the first ECG channel and a second individual ECG signal associated with the second ECG channel, the first individual ECG signal being orthogonal to the second individual ECG signal.

17. 1. An arrhythmia monitoring system, comprising:

1. An external cardiac monitoring device for a patient, comprising: a plurality of surface electrocardiogram (ECG) electrodes configured to sense surface ECG activity of the patient; an ECG processing circuit configured to process the surface ECG activity of the patient and provide at least one ECG signal for the patient on at least one ECG channel; a non-transitory computer-readable medium including a rhythm change classifier, the rhythm change classifier including at least one neural network trained based on a historical collection of ECG signal portions having known rhythm change information; at least one processor operatively connected to the at least one ECG channel and the non-transitory computer-readable medium, the at least one processor comprising: receiving the at least one ECG signal received via the at least one ECG channel; using the rhythm change classifier to detect time data corresponding to predetermined rhythm changes in the at least one ECG signal, the time data including at least one of a start time, a time interval, or any combination thereof; determining, based on the detected time data, at least one ECG signal portion associated with the detected time data that corresponds to the predetermined rhythm change in the at least one ECG signal; transmitting at least one determined portion of the ECG signal. the at least one processor; an external cardiac monitoring device having a remote computer system in communication with the external cardiac monitoring device, the remote computer system comprising: receiving the at least one ECG signal portion from the external cardiac monitoring device; configured to analyze each individual determined ECG signal portion of the at least one determined ECG signal portion to classify an individual class for each individual determined ECG signal portion. Remote computer system and An arrhythmia monitoring system comprising:

18. 18. The arrhythmia monitoring system of claim 17, wherein the at least one determined ECG signal portion includes a plurality of determined ECG signal portions, and the class for at least two distinct determined ECG signal portions includes a first class.

19. 20. The arrhythmia monitoring system of claim 18, wherein the remote computer system is further configured to transmit at least one message related to the at least two individually determined ECG signal portions to a computing device associated with a technician, the computing device associated with the technician configured to display a graphical user interface for collective review of the at least two individually determined ECG signal portions of the first class.

20. 20. The arrhythmia monitoring system of claim 18, wherein analyzing each individual determined ECG signal portion of the plurality of determined ECG signal portions to classify the individual class for each individual determined ECG signal portion comprises bucketing the plurality of determined ECG signal portions into a plurality of buckets, the first class comprising a first bucket of the plurality of buckets, and bucketing the plurality of determined ECG signal portions into a plurality of buckets comprises grouping the plurality of determined ECG signal portions based on at least one of an output of a neural network, a similarity of features of the plurality of determined ECG signal portions, a similarity of vector representations of the plurality of determined ECG signal portions, or a combination thereof.

Images