Non-invasive cardiac monitor and method for inferring or predicting a patient's physiological characteristics - Patents.com

Abstract

The present disclosure relates to a non-invasive cardiac monitoring device that records cardiac data to infer the onset of a human physiological characteristic, such as the onset of cardiac arrhythmia, and recommends treatment / intervention. Some embodiments include clustering patient data history to determine efficacy of treatment based on cluster groups corresponding to the patient data. In further embodiments, processing of detected cardiac heart rate signals can be partially processed in the wearable cardiac monitoring device and partially processed in a remote computing system. Some embodiments include a wearable cardiac monitoring device that is attached to a patient for extended periods of time to enable long-term detection of cardiac heart rate signals.

Description

[Background technology]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to U.S. Provisional Patent Application No. 63 / 265,335, filed December 13, 2021, the entire contents of which are incorporated by reference herein.

[0002] For purposes of this disclosure, certain aspects, advantages, and novel attributes of various embodiments are described herein. It is to be understood that not necessarily all such advantages may be achieved in accordance with any particular embodiment. Thus, various embodiments may be implemented to achieve one or more advantages taught herein, but may not necessarily achieve other advantages that may be taught or suggested herein. Summary of the Invention

[0003] The embodiments described herein are directed to physiological monitoring devices that can be worn continuously and comfortably on a human or animal subject for at least one week or more, more typically 2-3 weeks or more. In one embodiment, the device is specifically designed to sense and record cardiac rhythm (e.g., electrocardiogram, ECG) data, although in various alternative embodiments, one or more additional physiological parameters can be sensed and recorded. Such physiological monitoring devices can include many features to facilitate and / or enhance the patient experience and to make the diagnosis of cardiac arrhythmias more accurate and timely.

[0004] Some embodiments described herein may include a computing system for inferring an onset of cardiac arrhythmia in a user, the computing system including: a sensor configured to detect a cardiac signal from the user; and a hardware processor configured to apply the detected cardiac signal to a first machine learning process, the first machine learning process configured to infer an onset of cardiac arrhythmia based on the detected cardiac signal of the user, the first machine learning process being trained by: accessing historical cardiac signal patient data for a plurality of patients; accessing cardiac arrhythmia status for the plurality of patients, the cardiac arrhythmia status for the plurality of patients being determined since a time when the historical cardiac signal patient data was recorded; and training the first machine learning process based on the historical cardiac signal patient data and the cardiac arrhythmia status to infer an onset of cardiac arrhythmia, the user being different from the plurality of patients.

[0005] In some embodiments, the hardware processor is further configured to apply a second machine learning process, the second machine learning process trained to remove or detect noise segments or artifacts from the detected cardiac signal of the user, and the first machine learning process receives the noise segment or artifact removed cardiac signal as an input. In some embodiments, the hardware processor is further configured to apply a second machine learning process, the second machine learning process trained to reduce noise from the detected cardiac signal of the user, and the first machine learning process receives the noise reduced cardiac signal as an input. In some embodiments, the onset of cardiac arrhythmia comprises a probability of confirming the cardiac arrhythmia at a later time point. In some embodiments, at least a portion of the historical cardiac signal patient data is not indicative of cardiac arrhythmia at the time of recording the historical cardiac signal patient data. In some embodiments, the first machine learning process is further configured to process photoplethysmography (PPG) data, heart rate or acceleration data to infer the onset of cardiac arrhythmia. In some embodiments, the cardiac arrhythmia comprises at least one of ventricular tachycardia, supraventricular tachycardia, cardiac ectropion, or ventricular fibrillation. In some embodiments, the computing system comprises a wearable smart watch. In some embodiments, the sensor is an electrode.

[0006] Some embodiments include an electronic device for monitoring a physiological signal of a user, the electronic device including: an adhesive assembly including a housing and flexible wings, the housing containing a circuit board, the flexible wings extending from the housing and configured to conform to a surface of a user; a sensor coupled to the flexible wings, the sensor in electrical communication with the circuit board and configured to be positioned to conform and contact a surface of the user to detect the physiological signal; and one or more hardware processors configured to apply a machine learning process to the detected physiological signal, the machine learning process configured to infer an onset of cardiac arrhythmia based on the detected physiological signal, the machine learning process being trained by: accessing historical cardiac signal patient data for a plurality of patients; accessing cardiac arrhythmia status for the plurality of patients, the cardiac arrhythmia status for the plurality of patients being determined since a time when the historical cardiac signal patient data was recorded; and training the machine learning process based on the historical cardiac signal patient data and the cardiac arrhythmia status to infer an onset of cardiac arrhythmia, the user being different from the plurality of patients.

[0007] In some embodiments, inferring the onset of the cardiac arrhythmia includes generating a cardiac arrhythmia risk score at a time after the physiological signal is detected by the sensor. In some embodiments, the one or more hardware processors are further configured to determine that a cardiac arrhythmia is not currently present in the detected physiological signal prior to inferring the onset of the cardiac arrhythmia. In some embodiments, inferring the onset of the cardiac arrhythmia is responsive to determining that a cardiac arrhythmia is not currently present in the detected physiological signal. In some embodiments, the one or more hardware processors are further configured to adjust a function of the electronic device based on the onset of the inferred cardiac arrhythmia. In some embodiments, adjusting a function of the electronic device includes adjusting a window of the physiological signal that is processed by the electronic device. In some embodiments, in response to the risk score meeting the threshold score, the one or more hardware processors are further configured to send an alert to a user or physician device. In some embodiments, the one or more hardware processors are further configured to recommend a therapy or intervention for the onset of the cardiac arrhythmia. In some embodiments, recommending the therapy or intervention is based on clustering of prior patient data. In some embodiments, the clusters include multiple centroids, and the one or more hardware processors apply a distance process from the centroid of the corresponding cluster to the user's record to recommend a therapy. In some embodiments, the clustering process clusters patients based on cardiac arrhythmia status, intervention type, and outcomes of application of the intervention type.

[0008] Some embodiments include a method of training a machine learning process to infer the onset of a cardiac arrhythmia, the method including: accessing historical cardiac signal patient data for a plurality of patients, where at least a portion of the historical cardiac signal patient data for the cardiac signals does not indicate a cardiac arrhythmia at the time the cardiac signal was recorded; accessing cardiac arrhythmia status for a plurality of patients, where the status of the plurality of patients is determined after the time the cardiac signal was recorded; and training a first machine learning process based on the historical cardiac signal patient data and the cardiac arrhythmia status to infer the onset of a cardiac arrhythmia, where a user is different from the plurality of patients.

[0009] In some embodiments, training the first machine learning process further comprises training the first machine learning process to infer a hospitalization, heart failure onset, stroke onset, or death condition. In some embodiments, the method further comprises training a second machine learning process to infer a hospitalization, heart failure onset, stroke onset, or death condition, and the input to the second machine learning process comprises a cardiac signal of a longer duration than the cardiac signal input to the first machine learning process. In some embodiments, the method further comprises determining an atrial fibrillation burden from the detected cardiac signal, the atrial fibrillation burden comprising an amount of time the user spent in atrial fibrillation during a period of time.

[0010] Some embodiments include a method comprising: detecting a cardiac signal from a user; and applying the detected cardiac signal to a first machine learning process, the first machine learning process being configured to infer the onset of cardiac arrhythmia based on input data, the first machine learning process being trained by: accessing past cardiac signal patient data for a plurality of patients; accessing cardiac arrhythmia status for the plurality of patients, the status of the plurality of patients being determined since a time when the past cardiac signal patient data was recorded; and training the first machine learning process based on the past cardiac signal patient data and the cardiac arrhythmia status to infer the onset of cardiac arrhythmia for a user, the user being different from the plurality of patients.

[0011] These and other aspects and embodiments of the present invention are described in further detail below with reference to the drawings. [Brief description of the drawings]

[0012] [Figure 1A] FIG. 1A is a perspective view of an example of a physiological monitoring device. [Figure 1B] FIG. 1B is an exploded view of an example physiological monitoring device. [Figure 2A] FIG. 2A is a top perspective view of an example physiological monitoring device. [Figure 2B] FIG. 2B is a bottom view of an example physiological monitoring device. [Figure 2C] FIG. 2C is a top perspective view of an example physiological monitoring device including a liner. [Figure 2D] FIG. 2D is a bottom view including a liner of an example physiological monitoring device. [Figure 3A] FIG. 3A is a perspective view of an example of a physiological monitoring device. [Figure 3B] FIG. 3B is a top view of an example physiological monitoring device. [Figure 3C]FIG. 3C is a bottom view of an example physiological monitoring device. [Figure 3D1] FIG. 3D1 is a side view of an example of a physiological monitoring device. [Figure 3D2] FIG. 3D2 illustrates a side view of a ridge for sealing the top and bottom of a housing of an example physiological monitoring device. [Figure 3E] FIG. 3E is a bottom view of an example physiological monitoring device with layers transparent to show a perspective view of the device. [Figure 3F] FIG. 3F is a top view of an example physiological monitoring device with layers transparent to show a perspective view of the device. [Figure 3G] FIG. 3G is an exploded view of various components of an example physiological monitoring device. [Figure 3H] FIG. 3H is an exploded view of various components of an example physiological monitoring device. [Figure 4A] FIG. 4A is a schematic diagram of one embodiment of a system for predicting rhythm annotations using neural network encoding. [Figure 4B] FIG. 4B is a schematic diagram of one embodiment of a first and second subset of layers within a single neural network. [Diagram 5] FIG. 5 is a flow diagram of one embodiment of the system and ambulatory monitor. [Figure 6] FIG. 6 shows a schematic diagram of one embodiment of an autoencoder. [Figure 7A] FIG. 7A is a schematic diagram of an embodiment of applying an LSTM transformer block at the output of a decoder. [Figure 7B] FIG. 7B illustrates an autoencoder architecture of an embodiment that applies an LSTM transformer block at the output of the encoder. [Figure 7C] FIG. 7C shows a self-attention architecture for inferring the occurrence of cardiac arrhythmias. [Figure 7D]FIG. 7D shows a network training architecture for applying an LSTM transformer block at the output of the decoder. [Figure 7E] FIG. 7E shows a network inference architecture for applying an LSTM transformer block at the output of the decoder to generate hazard and / or survival probabilities. [Figure 7F] FIG. 7F shows a network training architecture for applying a self-attention architecture using a loss function. [Figure 7G] FIG. 7G shows a network inference architecture for applying a self-attention architecture to generate hazard and / or survival probabilities. [Figure 8] FIG. 8 illustrates an embodiment of grouping individual signatures (represented as projections on a 2D plane) from individuals who may have different progressions relative to the event time frame or who may respond better to a particular intervention. [Figure 9] FIG. 9 shows a schematic of a risk score alert for patients who have not experienced an event within the current patch wear time, but whose records contain markers indicating a high certainty of atrial fibrillation (AF) onset / detection within a pre-specified follow-up period. [Figure 10] FIG. 10 shows a graph according to an embodiment of risk curves stratified by demographic or clinical characteristics. [Figure 11] FIG. 11 shows an example of a report of a patient with no confirmed atrial fibrillation. [Figure 12A] FIG. 12A shows an example of a report of a patient with atrial fibrillation identified in a cardiac monitor recording. [Figure 12B] FIG. 12B shows an example of a report of a patient with atrial fibrillation identified in a cardiac monitor recording. [Figure 13A] FIG. 13A shows an example of a flow diagram for training an encoder and decoder and applying the encoder to a wearable device to infer the onset of cardiac arrhythmias. [Figure 13B]FIG. 13B shows an example of a flow diagram for training an encoder and a machine learning model, applying the encoder on a wearable device and applying the machine learning model on a remote computing device to infer the onset of a cardiac arrhythmia. [Figure 13C] FIG. 13C shows an example of a flow diagram for training an encoder, decoder, and machine learning model using historical patient cardiac data to infer the onset of cardiac arrhythmia. [Figure 14] FIG. 14 is a schematic diagram of an embodiment of a computer network system. [Figure 15] FIG. 15 is a schematic diagram of one embodiment of a programming and distribution module. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] The following description is directed to a number of different embodiments. However, the described embodiments can be implemented and / or varied in many different ways. For example, the described embodiments can be implemented in any suitable device, apparatus, or system for monitoring any of a number of physiological parameters. For example, the following discussion focuses primarily on long-term patch-based heart rate monitoring devices. In one alternative embodiment, the physiological monitoring device can be used, for example, for pulse oximetry and obstructive sleep apnea diagnosis. The manner in which the physiological monitoring device is used also varies. In some embodiments, the device can be worn for one week or less, while in other embodiments, the device can be worn for at least seven days and / or more than seven days, for example, 14 to 21 days or more.

[0014] Abnormal heart rhythms and arrhythmias often have other, less serious causes, so determining which of these symptoms are due to arrhythmias is a key challenge. In many cases, arrhythmias occur infrequently and / or episodically, making a rapid and definitive diagnosis difficult. Currently, cardiac monitoring is primarily performed using short-term (one day or less) electrodes attached to the chest, such as Holter monitors. The electrodes are connected to a recording device by wires, usually worn on a belt. The electrodes must be replaced daily, and the wires are cumbersome. Devices also have limited memory and recording time. Wearing the device often prevents patients from moving around and prevents them from certain activities while being monitored, such as bathing. Furthermore, Holter monitors are capital equipment and have limited availability, often leading to supply constraints and associated delays in testing. Such limitations severely impede the diagnostic utility of the device, patient compliance with using the device, and the possibility of capturing all important information. Lack of compliance and shortcomings of the device often lead to the need for additional devices, follow-up monitoring, or other testing to make a correct diagnosis.

[0015] Current methods for correlating symptoms with the occurrence of arrhythmias, including the use of heart rate monitoring devices such as Holter monitors and cardiac event recorders, are often insufficient to make an accurate diagnosis. In fact, Holter monitors have been shown to be non-diagnostic 90% of the time (DE Ward et al., "Assessment of the Diagnostic Value of 24-Hour Ambulatory Electrocardiographic Monitoring," Biotelemetry Patient Monitoring, vol. 7, published in 1980).

[0016] Furthermore, the medical process of actually obtaining a heart rate monitoring device and starting monitoring is generally very complicated. There are usually many steps involved in ordering such monitoring devices, tracking, monitoring, retrieving, and analyzing the data. Currently, heart monitoring devices are almost always ordered by a cardiologist or cardiac electrophysiologist (EP) rather than the patient's primary care physician (PCP). This is important because the PCP is often the first doctor to see the patient and determine that the patient's symptoms may be due to arrhythmia. After the patient sees the PCP, the PCP schedules the patient to see the cardiologist or EP. This appointment is usually made several weeks after the PCP's initial consultation, which in itself can lead to a delayed diagnosis and increases the likelihood that an arrhythmic episode will not be diagnosed if it occurs. When the patient finally sees the cardiologist or EP, a heart rate monitoring device is usually ordered. The monitoring period can be 24 to 48 hours (Holter monitor) or up to one month (cardiac event monitor or mobile telemetry device). Once monitoring is finished, the patient must usually return the device to the clinic. The data is processed by the monitoring company or by technicians at a hospital or office, and the report is finally sent to the cardiologist or EP for analysis. As a result of this complex process, the number of patients receiving cardiac rhythm monitoring is lower than the ideal number of patients who would receive it.

[0017] To address some of these problems with cardiac monitoring, the assignee of the present application has developed various embodiments of a small, long-term wearable physiological monitoring device. One embodiment of this device is the Zio® Patch and Zio® Monitoring Platform / System. Various embodiments are also described, for example, in U.S. Patent Nos. 8,150,502, 8,160,682, 8,244,335, 8,560,046, 8,538,503, 9,597,004, and 11,246,524, the entire disclosures of which are incorporated herein by reference. In general, the physiological patch monitors described in the above documents are designed to be worn comfortably on the patient's chest and for at least one week, typically two to three weeks. The monitor detects and records heart rate signal data continuously while the device is worn, and this heart rate data is then available for processing and analysis.

[0018] Physiological Monitoring Devices 1A-1B show multiple views of another example of a physiological monitoring device 100, such as those features described in U.S. Patent Nos. 11,350,864 and 11,337,632, which are incorporated by reference in their entireties. The physiological monitoring device 100 may include one or more of the components described elsewhere herein. The physiological monitoring device 100 may include a housing 115 including an upper housing 140 and a lower housing 145 configured to fit together with a flexible body 110 sandwiched between the upper housing 140 and the lower housing 145. The flexible body 110 may include a trace layer 109 and one or more substrate layers that form wings of the physiological monitoring device 100. The wings may include an adhesive layer 140 and electrodes 150, as described elsewhere herein. The rigid body and / or housing 115 may enclose the PCBA 120 , the flexible top frame 142 , the battery 160 , the battery terminal connector 150 , a portion of the trace layer 109 , the spring contact spacers 132 , and the springs 165 .

[0019] Figure 1A is a perspective view of an example of a physiological monitoring device 100. Figure 1B is an exploded view of a physiological monitoring device 600.

[0020] The upper housing 140 and the lower housing 145 can sandwich the flexible body 110 as described elsewhere herein. In some embodiments, the flexible body 110 can include one or more apertures 132 extending through one or more of the substrate layers to provide breathability and moisture management and / or to facilitate drug delivery to the overlying skin as described elsewhere herein. An upper gasket layer 160 and / or a lower gasket layer 170 (not shown) can be provided on opposing sides of the flexible body 110 (not shown). The gasket layers 160, 170 can be adhesive for adhering to the flexible body 110. A compressive seal can be formed above and / or below the flexible body 110. In some embodiments, a compressive seal can be formed with the flexible upper frame 142. The battery 160 can be disposed below the flexible body 110 including the trace layer 109. The PCBA 120 may be disposed above the flexible body 110, which includes the trace layer 109. The battery terminal connector 150 may be glued or otherwise coupled to the battery 160 such that the first and second battery traces 152, 153 are exposed on an outer surface of the battery terminal connector 150 on the top side of the battery 160. The first and second battery traces 152, 153 may be exposed to the interior volume of the upper housing 140 through a large central opening in the housing area of ​​the trace layer 109.

[0021] Electrical contact between the PCBA 120 and the first and second battery traces 152, 153 and / or between the PCBA 120 and the ECG interface portion 113 of the electrical traces 111, 112 may be established by spring contacts 137. The spring contacts 137 may be coupled to a bottom surface of the PCBA 120. The housing 115 may include a spring contact spacer 132 disposed below the PCBA 120. In some examples, the spring contact spacer 132 may be rigidly affixed (e.g., glued) to the bottom surface of the PCBA 120. In some examples, the spring contact spacer is attached to or integrated with the flexible body 110. In some examples, the spring contact spacer may be integrated into the battery terminal connector. The spring contact spacer 132 may include a flat body and a plurality of legs 133 extending downwardly. The legs 133 may be configured to seat against the top and / or sides of the battery 160 such that the spring contact spacer 132 maintains a minimum separation distance between the battery 160 and the PCBA 120 and provides sufficient space for the spring contacts 137. The spring contact spacer 132 may include one or more holes 134 through which the spring contacts 137 may extend downward from the bottom surface of the PCBA 120. The lower housing 145 may include a spring 165 disposed below the battery 160, as described elsewhere herein. The spring 165 may bias the battery 160 upward and bias the first and second battery traces 152, 153 into physical and electrical contact with the corresponding spring contacts 137. The ECG interface portions 113 of the traces 111, 112 may rest on top of the battery 160 such that biasing the battery 160 upward also brings the ECG interface portions 113 of the traces 111, 112 into physical and electrical contact with the corresponding spring contacts 137. The substantially constant spacing between the traces and the PCBA 120 provided by the springs 165 and the spring contact spacers 132 may reduce, minimize, or eliminate noise in the electrical signal caused by varying degrees of electrical contact between the spring contacts 137 and the traces.The assembly may include at least one spring contact 137 for each of the first battery trace 152, the second battery trace 153, the first electrical trace 111, and the second electrical trace 112. The assembly may include multiple spring contacts 137 for some or all of the traces. The spring contacts 137 may be configured to establish an electrical path between each trace and the PCBA 120 under compression induced by the arrangement of various components including the spring 165. The compressive contact between the spring contacts 137 and the traces may be maintained even under small changes in the separation distance between the traces and the PCBA 120 (e.g., caused by movement) because the spring contacts 137 extend further downward when the separation distance increases and the biasing force correspondingly decreases. In some embodiments, the first and second battery traces 152, 153 may be configured to be located on opposite sides of the housing 115 from the first and second electrical traces 111, 112. In some embodiments, the spring contacts may be configured to transmit electrical signals from a battery or electrocardiogram signals by contacting electrical traces applied to the top housing 140 or bottom housing 145. These electrical traces may be applied to the housing using laser direct structuring, plating onto a plateable substrate applied in a secondary molding process, or printing by aerosol jet, inkjet, or screen printing of conductive materials. In some examples, an RF antenna for wireless communications (such as Bluetooth) may be configured through the use of such electrical traces on the top housing 140 or bottom housing 145.

[0022] 2A-2D depict multiple views of an example of a physiological monitoring device 200 similar to the physiological monitoring device depicted in FIGS. 1A-1B. Here, the physiological monitoring device includes a central housing 202 including an upper housing 202 and a lower housing 206 sandwiched on a flexible substrate 210. One of ordinary skill in the art will appreciate that the housing may be constructed of any suitable material disclosed herein, for example, a hard polymer or a soft flexible polymer. In some examples, the housing may include an indicator 208, which may be any suitable shape, such as an oval, a circle, a square, or a rectangle. The indicator may include an LED light source (not shown) or any suitable light source, and may be overlaid by a transparent or translucent viewing layer disposed against an inner surface of the upper housing. The viewing layer may be constructed of thermoplastic polyurethane or any suitable material. The indicator may be used to indicate a status of the physiological monitoring device, such as the battery life of the physiological monitoring device. In some examples, the indicator may indicate whether the physiological monitoring device is collecting data, transmitting data, pausing, experiencing an error, analyzing data, etc. The indicator may display any suitable color, such as red, amber, green, etc.

[0023] Extending outwardly from the housing are a number of wings 212. Although two wings are depicted here, one skilled in the art will appreciate that some embodiments of the physiological monitoring device 200 may include more than two wings. As described elsewhere herein, the wings may be shaped to improve adhesion to and retention of the physiological monitoring device against the skin. In an example, the wings may be asymmetric, with a majority of one wing (upper lobe) 214 lying above the longitudinal line and a majority of another wing (lower lobe) 216 lying below the longitudinal line, thereby allowing the physiological monitoring device to be positioned at an angle above the heart such that the lower lobe is below the heart when the patient is in a standing position.

[0024] Extending outwardly from the housing and contained on or within the wings are electrode traces 218 similar to those described elsewhere herein, such as in FIG. 1A. As described elsewhere herein, the electrode traces may be printed directly onto a flexible substrate that may be part of a multi-layer flexible assembly 220. Additional printed lines 222 may surround the electrode traces 218 for visual enhancement of the physiological monitoring device, but may be printed on a different layer of the flexible substrate than the electrode traces are printed on. The printed lines may be printed to blend with the shape of the electrode traces. As described elsewhere herein, the electrode traces may surround a series of breathing holes 224 that allow the passage of air to the underlying hydrogel. In examples, there may be one, two, three, four, or more breathing holes. As described elsewhere herein, the openings 226 may extend through one or more layers of the physiological monitoring device to provide breathability and moisture management. In an embodiment, an adhesive boundary layer 228 may extend outward from the wings, thereby improving adhesion. FIG. 2B depicts the underside of the physiological monitoring device 200 depicted in FIG. 2A. Here, the lower housing 206 is clearly visible, as are the electrode traces 218 and printed lines 222 extending outward from the housing. FIGS. 2C and 2D depict the physiological monitoring device 200 of FIGS. 2A-2B, now including an exterior-facing top liner 226 and a skin-facing patient release liner 228 that surround the housing 202 over the wings. Such a release liner serves to protect the physiological monitoring device 200 during storage, and in particular protects the adhesive surface of the physiological monitoring device. In an embodiment, the liner may be shaped such that two sides meet to form an opening for the housing to extend vertically beyond the liner.

[0025] In some implementations, an abrader can be used to abrade the patient's skin prior to adhering the physiological monitoring device 200 (as described elsewhere herein) to the patient. The abrader can be used to remove a top layer of skin from the patient to improve long-term adhesion of the physiological monitoring device and / or signal quality of the physiological monitoring device.

[0026] In various alternative examples, the shape of a particular physiological monitoring device may vary. The shape, footprint, perimeter or boundary of the device may be, for example, circular, elliptical, triangular, compound curve, etc. In some examples, the compound curve may include one or more concave curves and one or more convex curves. There may be concave portions between convex shapes. The concave portions may be between convex portions on the housing and convex portions on the electrodes. In some examples, the concave portions may at least partially correspond to hinges, hinge regions, or areas of reduced thickness between the body and the wings.

[0027] Although described in the context of a cardiac monitor, the device improvements described herein are not so limited. The improvements described herein may be applied to any of a wide variety of physiological data monitoring, recording and / or transmission devices. The improved adhesive design features may also be applied to devices useful for electronic control and / or time-release delivery of pharmacological agents or blood tests, such as glucose monitors and other blood testing devices. As such, the descriptions, characteristics and functions of the components described herein may be modified as necessary to include specific components for a particular application, such as electronics, antennas, power or charging connections, data ports or connections for downloading or offloading information from the device, adding or offloading fluids from the device, monitoring or sensing elements such as electrodes, probes or sensors or other components, or components required for device-specific functionality. Additionally or alternatively, the devices described herein may be used to detect, record or transmit signals or information related to signals generated by the body, including, but not limited to, one or more of ECG, EEG and / or EMG. In certain examples, additional data channels may be included to collect additional data such as, for example, device movement, device flexion or bed, heart rate and / or ambient electrical or acoustic noise.

[0028] The physiological monitors described above and elsewhere herein can be further combined with data processing and transmission methods and systems that improve data collection from the monitor. Additionally, the methods and systems described below can improve the performance of the monitor by allowing timely transmission of clinical information while maintaining high patient compliance and ease of use of the monitor described above. For example, the data processing and transmission methods and systems described in this section or elsewhere herein can help to extend the battery life of the monitor, improve the accuracy of the monitor, and / or provide other improvements and advantages as described in this section or elsewhere herein.

[0029] 3A-3H depict an example of a physiological monitoring device 300 similar to that depicted in U.S. Pat. No. 11,350,864, which is incorporated by reference in its entirety. FIG. 3A depicts a perspective view of the physiological monitoring device. The physiological monitoring device 300 may include wings 330, 331, each of which is asymmetrical about a longitudinal axis extending generally between an electrode interface portion 302 that covers an electrode disposed on the underside of the wing. Electrode traces 304 extend from the housing to the electrodes and provide electrical communication between the electrodes and the central housing. One wing 330 includes a body disproportionately disposed above the longitudinal axis, and the other wing 331 includes a body disproportionately disposed below the longitudinal axis. Thus, the wings 330, 331 can make the flexible body asymmetrical about a lateral axis that is perpendicular to the longitudinal axis and extends through the housing 306, which lateral axis can include the patient trigger 307, as well as other patient triggers disclosed in this section or elsewhere herein. As described elsewhere herein, in certain embodiments, the patient trigger is about 10-30% of the total top area, such as about 20% of the total top area, or about 23%, such as about 22.8% of the total top area. In certain examples, the patient trigger can encompass about 20% or more, about 30% or more, about 40% or more, about 50% or more, or about 75% or more. In some embodiments, the patient trigger can encompass the entire top surface of the housing. The wings 330, 331 can be constructed from the same shape that is inverted or mirrored about both the longitudinal and lateral axes, as shown in FIGS. 3A-3C. In some embodiments, the wings may be asymmetric in size and shape, for example, upper wing 330 may be larger than lower wing 331 or vice versa. The shape of wings 330, 331 may be different such that the relative shape of upper wing 330 is different from the relative shape of lower wing 331.In some embodiments, the upper wing 330 may be under greater tension than the lower wing 331, or vice versa, and thus the different sizes and shapes between the two wings may help address unique force vectors during use of the physiological monitoring device. The wing configuration is particularly suited for placing electrodes diagonally relative to the subject's height, potentially reducing delamination due to gravity. Those skilled in the art will appreciate that the orientation of the wings may be altered so that the wings are mirror-like, rather than being disproportionately positioned above or below the longitudinal axis. Additionally, those skilled in the art will appreciate that the shape of the wings as described herein may differ from the generally rounded shapes depicted in FIGS. 3A-3H. For example, the wings may be angular, such as square, rectangular, triangular, pentagonal, or any suitable polygonal shape. These polygonal shapes may have rounded corners to reduce the possibility of delamination from the corners. A liner 308, as depicted elsewhere herein, can be used to cover and protect any adhesive prior to application of the physiological monitoring device to a patient or user. In an example, the liner is separated into two pieces, one over each wing.

[0030] In certain embodiments, an additional visualization pattern 310 may extend through the wings. The visualization pattern 310 may be of any suitable size or shape to outline the electrode traces and outline the shape of the wings, for example, the visualization pattern 310 may be in the form of lines, such as rounded lines, to reflect the outline of the electrode traces and the shape of the wings. In certain examples, the lines may be one, two, three, four, or more. In some embodiments, the visualization pattern may be formed from a pattern of dots, shapes, or other combinations, such that the visual cleanliness of the device is maintained even if an otherwise clear adhesive layer becomes less visually acceptable to the user over the course of the wear period (e.g., if the adhesive layer picks up foreign matter and / or becomes cloudy due to absorption of moisture). In certain embodiments, the visualization pattern may have another functional purpose of alerting the user how long the device has been worn, for example, by changing color over time or with wear. Such a change in appearance can alert the user to remove the device at the appropriate time. FIG. 3B shows a top view, FIG. 3C shows a bottom view, and FIG. 3D1 shows a side view of an example of a physiological device 300. In FIG. 3C, a flexible electrode 312 is visible. As shown in FIG. 3D1, a top 314 and bottom 316 portion of the housing may be disposed above and below a flexible body 318. FIGS. 3E and 3F show the underside and top side of the physiological monitoring device 300, with each layer being transparent so that all layers are visible. Each layer is described in more detail below in an exploded view of the physiological monitoring device 300. An opening 320 may be disposed in a substrate layer disposed above the adhesive layer. As described in more detail above, such an opening may provide ventilation through one or more layers and facilitate evaporation of moisture from below the adhesive layer through the layer or layers that make up the opening. As shown in FIG. 3D2, in an embodiment, a gasket 319 may be disposed between an upper housing cover 314 and a lower housing 316 that are co-molded to one or more housings.The gasket may compress the adhesive assembly and ridge interface (shown below in FIG. 3D2) or another gasket on the opposing housing to provide water resistance to the internal electronics hardware. As depicted in FIG. 3B2, a ridge 321 may be disposed on the top edge of the lower housing 316, the ridge 321 configured to press against the adhesive layer 319. One skilled in the art will appreciate that the ridge 321 may be any suitable shape, such as, for example, a bordered ridge as depicted in FIG. 3D1. In certain embodiments, the ridge may be rounded, square, and / or polygonal. In certain examples, the height of the ridge may be about 0.15 mm, such as about 0.01 mm to 0.5 mm, about 0.05 mm to 0.4 mm, about 0.1 mm to 0.3 mm, about 0.1 mm to 0.2 mm, or about 0.13 mm.

[0031] FIG. 3G shows an exploded view of one example of a flexible body 301 of a physiological monitoring device 300 described herein and elsewhere herein. The housing 306 is not shown. As will be appreciated by those skilled in the art, the image of FIG. 3G is oriented upside down with respect to positioning on the skin. Following the numbering in FIG. 3G, #7 depicts a release liner, which protects the adhesive layer (340 and the hydrogel electrodes 350). Directly over the adhesive layer is a perforated layer 304 (including openings as described herein) and a flap layer 303. In certain embodiments, the perforated layer and the flap layer may be comprised of any suitable material, such as polyethylene terephthalate (PET) and / or polyurethane. Directly over the perforated layer may be a lower substrate layer #1, which may be comprised of polyurethane. In embodiments, the lower substrate layer may have at least one textured side, which may be positioned such that the textured side faces the flap layer #3. In embodiments, the flap layer #3 may also include at least a textured side. This textured side may be configured to face the bottom substrate layer #1. The conductive electrode traces may be printed on an additional separate substrate (311, 312). Alternatively, in some examples, the conductive electrode traces may be printed directly on the substrate layer #1. Overlying the conductive electrode traces may be a top substrate layer 300. Overlying the top substrate layer may be an additional carrier layer #10, followed by an adhesive layer #11 and a top rigid liner #9. One skilled in the art will appreciate that such layer arrangements are applicable to any example of the physiological monitor described herein, such as the examples of Figures 3A-3F.

[0032] FIG. 3H illustrates an exploded view of an example housing 306 of a physiological monitor device 300 through which the flexible body 301, described in detail above, passes. The upper housing cover 314 can include a patient trigger 307. The upper housing cover can encase a circuit board 322. A spacer 323 disposed on the underside of the circuit board is configured to maintain a constant spacing between the conductive contact springs on the underside of the circuit board and the battery terminals / ECG trace contacts. The spacer can also provide electrical insulation between the circuit board and the battery. The spacer can have holes for the conductive contact springs to pass through, and the contact springs are connected to the circuit board. A battery terminal 325 can be disposed below the flexible body 301 and the circuit board 322, and thereby above a wave spring 326. In an embodiment, the battery terminal 325 can be wrapped around and glued to a coin cell battery 328. The battery terminals 325 can be configured as a flex circuit with conductive vias 327 that allow the positive underside of the coin cell battery 328 to be connected up to the negative top side of the battery, with both the negative and positive terminals presented on the top side of the battery to contact the contact springs on the circuit board. Alternatively, the battery contacts on the bottom housing can bring the positive underside of the coin cell battery up to the negative top side to contact the circuit board. The vent layer 329 can be positioned against the lower housing portion 316 over the vent holes 332 in the lower housing. In an embodiment, the vent layer can be constructed of a material that allows the passage of gases while preventing the passage of liquids, such as ePTFE or any other suitable material. The combination of the vent holes 332 and the vent layer allows for normalization of air pressure between the exterior and interior of the housing. In an embodiment, the vent holes 332 in combination with the ventilation layer prevent the buttons and / or triggers 307 from being blown out or sucked in depending on the outside air pressure, for example, if the patient is at a different altitude, such as in an airplane. The ventilation layer may be thin and round with a ring of adhesive on the bottom. The area of ​​the ventilation layer covered with adhesive may not be gas permeable, and the center portion may be gas permeable and liquid impermeable.A central portion of the vent layer can be positioned over the vent hole, thereby allowing gas to enter and exit the housing while restricting liquid from entering and exiting. In some embodiments, the vent layer can be molded integrally into the bottom housing, or it can be ultrasonically welded to the bottom housing or adhered by any suitable means.

[0033] A system for estimating loads and / or predicting rhythm annotations using neural network encoding Some embodiments disclose a wearable device that can process a detected biosignal through an encoder that includes a first subset of layers of a neural network, such as the features described in U.S. Pat. No. 11,246,524, which is incorporated by reference in its entirety. FIG. 4A is a schematic diagram of an example of a system that predicts rhythm annotations using neural network encoding. In some embodiments, the system can make one or more predictions, such as predicting rhythm annotations or estimating workload for cardiac rhythms including atrial fibrillation and / or atrial flutter. A wearable device, such as a cardiac monitor patch 402, can process an ECG input 404 through a first subset of layers 406 of a neural network, such as an encoder. The wearable device can receive the output of the first subset of the neural network's layers 406 and transmit the output to a computing device 408 (e.g., an external system such as a smartphone or a server) for further processing the data via a decoder including a second subset of the same neural network's layers 410 to derive characteristics of the user 412, such as indications or predictions of past cardiac arrhythmias and / or predictions of future arrhythmia occurrences.

[0034] In some embodiments, the first and second subsets of layers are within one neural network. The neural network may be designed such that the output of the first subset of layers may output data with less dimensionality than the input to the neural network, and the output of the second subset of layers may be designed to provide an indication of a user characteristic, such as a past or future prediction of atrial fibrillation. FIG. 4B is a schematic diagram of an example of the first and second subsets of layers within a single neural network. The neural network may be trained simultaneously on both the first subset of layers 422 and the second subset of layers 424. For example, if a neural network has 10 hidden layers, the first four layers are processed on a wearable device such as a heart monitor patch 426, and the output of the fourth layer, which has a lower dimensionality (e.g., has better data compression capabilities) than the input to the first layer, is sent to an external computing system such as a server 428. The external computing system processes the output of layers 4 through 5-10. The dimensionality of each layer of the neural network can be designed to output a certain data size (e.g., the output dimension of each convolution layer or pooling layer). For example, the patch 426 can include an ECG encoder 430 that receives ECG data 432 at 2400 bits per second (bps). The patch 426 can process the ECG data 432 through a first subset of the layers 422 of the neural network (e.g., in the ECG encoder 430) and output data 434 of a smaller dimensionality, such as data at 128 bps. The output data 434 can be sent to an external server 428. The external server 428 can include a classifier 438 that processes the output data 434 through a second subset of the layers 424 of the neural network and outputs a patient indication or prediction 436 on which it was trained. The entire neural network, including the first subset 422 and the second subset 424 of layers, can be designed and trained as a single neural network.

[0035] In some embodiments, the output of the first subset of layers can be of a smaller dimensionality than the input to the neural network. Thus, instead of transmitting the entire ECG signal (e.g., the input to the neural network) from the wearable device to an external computing device, the wearable device can transmit a smaller amount of data to the external computing device, such as the 128 bps output data 434 of the first subset of layers, instead of the entire 2400 bps ECG signal 432. Advantageously, less network throughput is required to derive indications of past cardiac arrhythmias and / or predict the onset of future arrhythmias.

[0036] Further, instead of processing the ECG signal through all layers of the neural network at the wearable device to derive signs of past cardiac arrhythmias and / or predict future onset of arrhythmias, the wearable device may process the ECG signal through only a first subset of the layers of the neural network via the ECG encoder 430 (e.g., through an encoder) and transmit the output of the first subset to an external device 428 that processes a second subset of the layers (e.g., through a decoder or classifier 438).

[0037] Prognosis and Prediction Embodiments Some embodiments described herein improve upon conventional systems that use long-term continuous monitoring techniques to detect atrial fibrillation. Some conventional monitors typically include an implant device with leads that are embedded in the heart. However, such approaches are highly invasive, lack suitable communication technology, and are very difficult to update the hardware / software.

[0038] Furthermore, conventional monitors do not provide predictive outcomes based on current cardiac signals or current treatments, let alone treatment recommendations such as intervention or hospitalization recommendations.

[0039] Furthermore, conventional implantable technologies do not allow for continuous capture of ECGs. These implantable monitors measure ECG data and capture events detected within the ECG data. These implantable monitors capture event data based on a small data window, such as 10 seconds of ECG data, which may be insufficient to provide outcome predictions regarding the current cardiac signal, to predict outcomes in advance regarding the current treatment provided in the patent, or to determine recommendations regarding treatment recommendations.

[0040] FIG. 5 is a flow diagram 500 of an example of the system and a monitor, such as an ambulatory monitor. The monitor can be an implantable, mountable, and / or wearable recording device. In block 502, the monitor can capture continuous and / or discrete biosignal data. Some embodiments described herein include monitors that can continuously monitor cardiac signals over extended periods of time, such as 7, 14, 21, 30 days, or more, using electrodes and monitoring circuitry. Some embodiments can provide outcome predictions based on the patient's currently measured cardiac signals over periods of time longer than seconds, minutes, or hours. Some embodiments can provide predicted outcomes for treatments currently being provided to the patient. Some embodiments can provide treatment recommendations for the patient based on the currently measured cardiac signals and / or current treatments.

[0041] In some implementations, the monitor may measure one or more physiological signals such as, but not limited to, atrial fibrillation, ECG, PPG, accelerometer data, heart rate, pulse transit time, impedance measurements, acoustic measurements, RR intervals, blood pressure measurements, temperature measurements, and / or glucose levels.

[0042] In some embodiments, the system can continuously measure, record, and / or analyze biosignals via devices implanted, attached, or worn by a person in an ambulatory environment. The signals may include raw measurements or features extracted based on signal processing, machine learning (ML), or deep learning (DL) methods.

[0043] In block 503, the monitor may analyze the received continuous or discrete biosignal data to remove noisy segments and unwanted artifacts. The monitor may apply signal processing machine learning algorithms to classify noise segments from segments below a noise threshold level, such as using a pre-trained model. The monitor may apply signal processing machine learning algorithms to determine whether the received signal is noise, an artifact (such as when the user is sleeping, engaged in high intensity activity, wearing a pacemaker, etc.), and / or the like. This technique may be used to remove or reduce noise and improve the signal-to-noise ratio. The noise may be generated by internal or external sources, such as environmental noise, noise due to imperfect contact with the electrodes, and / or the like.

[0044] In some implementations, the signal processing techniques for removing noise and / or unwanted artifacts may include a machine learning model, such as a neural network, which may be a different machine learning model than a neural network for making inferences on the physiological data (discussed further herein).

[0045] In some embodiments, a noise machine learning algorithm can be trained to determine whether artifacts are present in the data. The noise machine learning algorithm can mitigate noise in the recorded data. In some examples, the same machine-learned algorithm can identify artifacts and / or mitigate noise. In other examples, separate machine learning algorithms are used. In some implementations, prediction of noise and / or artifacts can be performed on a sample-by-sample basis or for groups of samples.

[0046] In some embodiments, artifact segment identification, noise segment identification, and / or noise mitigation may be performed using one or more of the following architectures: a) Fully connected Neural Networks (NNs): For example, fully connected NNs can learn complex relationships between signal morphology and the actual labels of whether the signal is an artifact or noise. Training with a diverse set of signals representing artifacts / noise or useful signals, combined with dropout, can increase the network's ability to learn important relationships between signals and labels (useful signals, noise, artifacts, etc.), while reducing the chance of overfitting. For example, the system can randomly disable certain nodes of a layer or neural network during training to avoid overfitting and / or to more restrictively train a neural network that learns weights that are more robust and predictive for the training objective. b) Encoder / Decoder Architectures and / or Fully Connected NNs Followed by Convolutional Layers: The use of convolutional layers allows for longer signal sequences to be used as inputs that provide better context for identifying onsets and / or offsets of noise / artifact regions (e.g., transitions from useful ECG signals to noise). The encoder / decoder section can advantageously identify latent variables that can effectively learn transitions into and / or out of noise / artifact regions, as well as other signal / noise characteristics, using a smaller set of features. These latent variables can be applied in fully connected layers to generate complex prediction functions. Due to the small size of the network, the implementation of noise prediction can be accomplished on the device itself, or the transmission of the encoder output can be efficiently and continuously transferred over a wireless connection in real time, and the prediction can be performed on a server. c) The convolutional layer is followed by an LSTM layer (e.g., in some embodiments a bidirectional LSTM layer), possibly followed by many fully connected layers: Adding LSTM layers enhances the predictive capabilities of the network by allowing it to recognize parts of the signal that are highly correlated with other parts of the signal. As an example, at the transition between a noiseless and a noisy signal, the LSTM may be able to identify more correlation in the waveform before and after the transition point, and may adjust the sequence of output weights of the system to better identify the transition point between the noisy and noiseless segments. d) Transformer architectures that can exploit similarities / dependencies in the characteristics of successive signals to provide accurate classification that takes into account the context in which noise / artifacts may be present: Transformer networks can advantageously provide a more flexible structure in encoding signal similarities / differences between sections of a signal. As an example, the transformer architecture can adjust its processing based on user selection of the temporal distance at which similarities are seen, the sequence of frequencies of the signal to be considered, etc. For example, the transformer architecture can adjust the length of successive segments at a particular length, such as looking at 1 minute intervals, 1 second intervals, 1 hour intervals, 6 hour intervals, and / or the like. e) Self-supervised learning to extract latent features from a large number of unlabeled signals, including noise and non-noise regions with diverse types of rhythms, and train a model to classify signals in the noise regions: The pre-trained features can be combined as input to a smaller fully connected layer to generate a final prediction of noise or artifacts. Since these features are rich, but the final classification network needs to be smaller, the implementation can be efficiently performed both on the device itself and / or on a server. For example, a first pre-trained model can extract latent features, and an additional layer of a second pre-trained model can receive the latent features as input and be applied after the first pre-trained features to make a classification of noise or artifacts, and / or an additional layer that takes the pre-trained features from the self-supervised model and applies the features to make a prediction. f) Siamese network: In this approach, the system can train a network to identify specific waveforms selected by the user as noise / artifact segments. The system can then make a classification that is more suited to the application at hand. Furthermore, if new versions of the signal are considered as segments that contain artifacts and / or noise, these segments can be added to the process by adding a newly trained network for the new waveform of interest. For example, if there is a model already trained to identify noise, a model can be trained to also identify artifacts without disturbing the part of the model already trained to identify noise.

[0047] In block 504, the monitor can perform an analysis on the captured vital signs to determine whether an arrhythmia is present and / or predict the likelihood of a future onset of an arrhythmia. Some embodiments described herein include a monitor that can determine that a patient does not have a current atrial fibrillation condition. The monitor can predict the onset of a potential atrial fibrillation condition in the future. Conventional monitoring devices are limited to providing information about a current atrial fibrillation condition.

[0048] At block 506, the monitor determines whether an arrhythmia is present in the cardiac signal and / or whether an onset of an arrhythmia is detected. If the determination is yes, the flow diagram proceeds to block 508 where a communication of the presence or onset of the arrhythmia is sent to a physician or a user on a device such as a mobile device and / or the monitor may change functionality, such as analyzing a longer period of data or collecting more data over a longer period of time, after which the flow diagram may proceed to block 509. In some implementations, the monitor may update functionality based on the arrhythmia determined at block 509.

[0049] Advantageously, an alert can be sent to a physician remote from the patient, and the alert can be sent in real time or substantially real time when the arrhythmia or onset is detected. The physician's computing device and / or an application on the computing device can be launched by the system to display the alert. In this manner, even if the physician's computing device does not have the application for the alert open, the system described herein can send an alert that launches the application to alert the physician. In some implementations, if an arrhythmia or onset is not detected, the flow diagram proceeds to block 510.

[0050] Some embodiments described herein include ambulatory monitors that can monitor cardiac signals and determine associations with future potential stroke, death, atrial fibrillation, heart failure, need for hospitalization, and the like. Such ambulatory monitors can make these predictions (such as predicting the onset of atrial fibrillation, or potential stroke or hospitalization) based on cardiac signals in the patient's normal state, such as while the patient is not hospitalized and is going about their daily life. For example, in predicting stroke, the ambulatory monitor can apply multiple weeks of cardiac monitor data, such as two weeks, to make such predictions. Predicting the onset of atrial fibrillation can be made with the same and / or less data volume as predicting the onset of stroke. The volume of data for predicting the onset of atrial fibrillation can depend on the patient's age, as the prevalence of atrial fibrillation may increase faster in patients over 70 years of age.

[0051] Some embodiments described herein include ambulatory monitors that may include an encoder on the patch. For example, for ambulatory monitors that process less data (e.g., less than a week), an encoder may be included on the patch, which may process the data to output data with a smaller number of dimensions for subsequent transmission to an external computing system, such as a server. The server may process the output of the encoder to determine physiological characteristics, such as whether atrial fibrillation can be detected from the cardiac monitoring data, or whether the onset of atrial fibrillation can be predicted. Advantageously, the signal may be partially processed on the patch and partially processed by a server, such as a cloud server, during the wear time of the patch. The cloud server may predict the onset of atrial fibrillation and / or provide treatment recommendations, etc., while the user has the patch placed on his or her body.

[0052] In some implementations, the cloud server can send an alert to a computing device, such as a user's mobile phone and / or a physician's computing device, to notify the patient of the onset condition. The encoder, due to its low dimensionality, can send less data to the server side, thus providing technical advantages such as automatic encryption, extended battery life, reduced processing power required for the patch while the server side further processes the data, and reduced network data throughput used, as further described herein. Furthermore, such immediate processing and reduced use of critical hospital resources is crucial when the patient is already hospitalized, such as the onset of a stroke, where response time largely corresponds to survival rate. The ambulatory monitor can provide predictions of physiological characteristics, such as stroke, and / or provide treatment recommendations, such as sending the patient to an immediate care unit (ICU). The encoder can be trained to output signal features derived from the monitored cardiac signal. For example, the encoder can output cardiac signal features, such as boundaries, edges, and / or the like.

[0053] In block 510, the monitor can upload the recorded continuous and / or discrete biosignals and / or associated derived signal features output from the encoder. The encoder output and / or a portion of the measured biosignals can be continuously uploaded to an external computer such as the cloud. The encoder output can be uploaded by connecting a patch to a device and the device transmitting the data to an external computer, for example to a mobile phone via a USB connector, which transmits the data to a server. Advantageously, if only the encoder output without the measured biosignals or only a portion of the measured biosignals is transmitted to the cloud, the monitor can conserve battery life by storing less data, transmitting less data, requiring less network connections, requiring less transmission of data, and / or requiring less frequent transmission of data.

[0054] In block 512, an external computing device, such as a server, can analyze the recorded and / or associated derived features to identify a set of cardiac features. For example, features extracted from machine learning and / or deep learning techniques may include one or more of the following: atrial fibrillation burden, total time in atrial fibrillation, longest atrial fibrillation episode, quantiles of heart rate, counts of rhythms such as supraventricular tachycardia (SVT), ventricular tachycardia (VT), atrial fibrillation, premature ventricular contractions (PVC), premature atrial contractions (PAC), atrioventricular block (AVB), time and density of events (day, night, week), other arrhythmia burden, number and duration of episodes at activity levels (low, medium, high), transition pattern features / clustering of features, transition time features, rhythm categories, arrhythmia density and density change, ectopic density and density change, activity load, rest load, and / or sleep arrhythmia / load.

[0055] In some embodiments, if an onset of atrial fibrillation is detected, the monitor can adjust settings and / or the external computing system can instruct the monitor to adjust settings, such as observing a longer monitor window. In some embodiments, if the monitor is measuring a longer monitor window, the monitor can change the data sent to the external computing system. For example, a longer monitor window can cause the monitor to only send the encoder output and not send the measured ECG signal, or the monitor can send the encoder output with a smaller or larger window of the measured ECG signal.

[0056] At block 514, a system, such as an external computing device, can calculate an associated risk for each predicted endpoint based on the set of biosignal recorded cardiac features. Some embodiments include a system that can determine a risk score based on measured cardiac signals indicative of the onset of atrial fibrillation. The system can determine the risk score based on a particular intervention and a particular outcome, such as by clustering previous patient data, as further disclosed herein. For example, a clustering algorithm can cluster patients based on their current condition, the type of intervention, and the outcome of the type of intervention. The current patient can then be identified as similar to one of the clusters.

[0057] In some implementations, based on the identified associated clusters, the system can recommend interventions for physiological conditions, such as the onset of atrial fibrillation. The clustering algorithm can be based on years of clinical trials in which patients are monitored for current and / or onset conditions, treated with various intervention types or not treated at all, and the outcomes of these interventions are recorded. Such data is processed by the clustering algorithm to create clusters of patients based on current / onset conditions, intervention types, and outcomes.

[0058] At block 516, the system may determine whether a particular intervention is associated with a predicted endpoint, and if yes, the flow diagram may proceed to block 518, where the system may calculate a predicted impact for each predicted endpoint with an associated intervention based on the vital sign recording cardiac feature set and the predicted endpoint risk assessment. At block 520, the system may populate a report and display the report on the patient and / or physician device with the predicted impact associated with the predicted endpoint and, if applicable, the intervention option. For example, an intervention may be associated with a long-term change in the course of a disease. An intervention may be associated with a short-term treatment, such as when a patient is already hospitalized.

[0059] Thus, implementation of the present system and / or ambulatory monitor advantageously provides significant advantages in cardiac monitoring technology. The present system and / or ambulatory monitor provides prognosis, risk prediction / stratification, and treatment guidance for the outcome of, or the numerous consequences and associated costs associated with, the presence of symptomatic and asymptomatic cardiac arrhythmias, such as atrial fibrillation (AF). As a result, risk stratification systems outperform traditional risk factors. Whereas traditional system diagnostics are limited to providing information regarding potential treatment or the need for continuous monitoring at a population level, the system and / or ambulatory monitor described herein provides individualized guidance for continuous monitoring (timing and intensity) and initiation and titration of treatment (e.g., depending on the current level of severity of the arrhythmia and the predicted future severity of the outcome if left untreated). In particular, the system and / or ambulatory monitor may use partial or complete recording signatures created from wearable devices, such as signal sequences of multiple modalities and / or different time horizons, to predict risk of adverse outcome, need for intervention, estimated intensity, cost of treatment, hospitalization, and / or the like.

[0060] In some implementations, the system and / or ambulatory monitor can use latent multimodal features (encoded markers) that encapsulate the individual's temporal and physiological characteristics that can be used as predictors of multiple future health-related endpoints. The system and / or ambulatory monitor can inform the user or physician of specific combinations of explainable features that can be used to access the role of asymptomatic and symptomatic arrhythmias on the risk of adverse outcomes. The system and / or ambulatory monitor can assess the probability of the presence of undiagnosed (and often asymptomatic) atrial fibrillation, often referred to as silent atrial fibrillation, or predict on-site for atrial fibrillation, even if the diagnostic device does not detect atrial fibrillation during the wear time. Detection of atrial fibrillation is a function of the incidence of atrial fibrillation and the probability that the occurrence of the event coincides with the wear time of the device. Identifying markers indicative of active but undiagnosed atrial fibrillation based on single-lead ECG or PPG sensors worn during daily activities rather than during clinical visits can guide retesting strategies for patients with high confidence scores based on predictive models, leading to earlier diagnosis of the disease. The system and / or ambulatory monitor can provide a time frame for follow-up testing for asymptomatic atrial fibrillation and inform recommended timing of this testing.

[0061] Although the prediction method is described for atrial fibrillation, the same method can be applied to other arrhythmias, e.g., tachycardia (including supraventricular tachycardia, ventricular tachycardia), bradycardia events (including pauses, atrioventricular block). Prediction of ectopic beats (e.g., premature atrial beats and ventricular beats) may also be developed. Indeed, predicting future arrhythmia combinations or categories, rather than specializing in one arrhythmia type, may provide an advantage in predictive accuracy.

[0062] In some embodiments, the use of a single-lead continuous monitoring device can provide the advantage of being able to detect and / or predict the presence of asymptomatic atrial fibrillation from signals associated with the patient's normal activity, an advantage that is unlikely to be provided by a 12-lead ECG in a clinic, which measures only for a short period of time (usually 10 seconds) in a controlled environment. Conventional systems cannot predict future episodes of atrial fibrillation. Examples The systems described herein can be useful for multiple device solutions, for example, informing the patient if they should wear another ambulatory monitoring patch or participating in an extended monitoring paradigm (up to 2 years) with a watch / ring or implantable monitor.

[0063] In some embodiments, the system and / or ambulatory monitor can evaluate cluster patient records within a particular disease and / or risk group, evaluate clinical trajectory / journey based on multiple signatures acquired over time, and evaluate the probability of response to a particular or family of interventions. This advantage is distinct from predicting response to an intervention, as the group informs a combination of disease trajectories and a history of use of a particular family of interventions that improve outcomes. This further personalizes medical care, as it links disease characteristics with response to a particular intervention. The system and / or ambulatory monitor can predict the patient's potential course of disease progression and the effectiveness of the intervention. The system and / or ambulatory monitor can use the risk score to select individuals for inclusion in clinical trials for a given intervention or treatment. The system and / or ambulatory monitor can determine the detectability of future events during the monitoring period, which can be utilized by clinicians to access the need for continued / extended / termination of monitoring.

[0064] In some implementations, after the external computing device determines AF, onset of AF, and / or intervention, the external computing system can send data back to the monitor. Advantageously, the sent data can initiate changes in the monitor, such as how much data is being measured, the algorithm being applied on the monitor, the amount of data (e.g., window length) being processed on the monitor and / or sent to the external computing device, and / or the like. In some implementations, the external computing device notifies a physician, such as a cardiologist or supervising clinician, to indicate that an atrial fibrillation condition has been detected, that onset atrial fibrillation is likely, and / or a recommendation regarding intervention (e.g., ablation or provision of medication based on the clustering algorithms described herein).

[0065] In some embodiments, the system can connect to two wearable devices. The patient can wear one device and upload data to be analyzed in a cloud environment. The system can connect to a second device. In some embodiments, the system connects to the second device at the same time as the first device. The system can obtain similar data to correlate the two and / or receive different data, such as different types of data, to further enhance the processing on the system side. In some implementations, the output of the autoencoder architecture recommends to the patient to wear the second device, such as what type of device and / or when to wear the second device.

[0066] In some implementations, the system can connect to a second device at a later time and correlate the recording with a previous recording of the same patient. Thus, the system can create longitudinal data sets of a single modality (such as ECG) and / or a set of simultaneously measured modalities (such as heart rate, PPG, blood pressure, acceleration, acoustics, etc.). Various modalities can be combined to effectively create a more progressive score. Advantageously, different modalities can provide a much better picture and prediction of atrial fibrillation onset and other physiological characteristics, as well as the progression of interventional treatments. For example, distinguishing between signals under stress and signals while the patient is sleeping can provide a more accurate picture of the onset of atrial fibrillation. Collection of data by wearable devices can include continuous signals from long-term continuous monitoring devices such as heart monitor patches, smart watches, smart fabrics, smart eyewear, and / or the like. The wearable devices can post uploads to the system, and the system can perform analysis of the recorded signals in a mobile computing or cloud environment. Algorithms, data, and results of such analysis can be stored in the cloud for long-term storage.

[0067] In some embodiments, data received from different devices can result in separate risk scores that can indicate progression of a physiological characteristic. For example, based on the measurements of a smartwatch-type wearable device (e.g., a score of 20 at time 1 and a score of 50 at time 2), the system can indicate that atrial fibrillation is progressing quickly and recommend wearing a patch.

[0068] Embodiments of risk prediction In some embodiments, the risk prediction model can require inputs of variables / features and an algorithm that can appropriately weight these inputs to generate a risk score for a given outcome for a particular time frame. The system can select inputs from any number of features as described herein. For machine learning, statistical modeling, and deep learning models, discrete features from the aforementioned features can be combined to generate a risk estimate using the methods described herein. For continuously measured signals, or signals with a time dimension (e.g., rhythm labels derived from an ECG through an annotation algorithm based on the waveform within a specified time interval), the system can use a number of numerical methods to generate inputs of appropriate form for calculating a risk score through a trained algorithm. The system can use these methods for individual signal measurements or can be combined to convey information simultaneously.

[0069] In some embodiments, combinations of these signals can be aligned in time and sampled (upsampled, downsampled, or maintained) at the same sampling rate to generate a simultaneous set of measurements. Each type of one-dimensional signal may be scaled and combined to form a strip of data. The X-axis of the strip is the time axis and the Y-axis represents the stack of one-dimensional signal modalities. Advantageously, structuring the data in this way allows features based on different signal modalities to be created simultaneously. This data structure can be further expanded by exploiting signal transformations and interactions of the individual signals. Such transformations include Fourier and wavelet transforms of the original waveforms. Interactions include simple multiplication of two or more signals, powers of these signals, or time-lagged convolutions of the signals.

[0070] In some embodiments, the system can form the data structure as an image and apply deep learning image analysis techniques to generate a reduced representation of the signal. Specifically, the system can use an autoencoder model that utilizes a two-dimensional convolution of the signal to create a latent representation of the full signal. Such latent features can be of various dimensions and can encapsulate relationships of different signal information in a low dimensionality. In some embodiments, these latent variables are extracted from the bottleneck layer of the autoencoder.

[0071] In some embodiments, the system can apply an autoencoder architecture that represents a dimensionality reduction technique that, when combined with appropriate weights and layer architectures, can extract a set of features that can reconstruct the original signal with high fidelity. This dimensionality reduction allows the autoencoder to create a "record signature" that can uniquely represent a recording, such as a partial or complete recording of a signal, or unique features of signals measured at the same time. Representations corresponding to "similar" recordings can have similar "record signatures" in an appropriate distance metric. In another implementation, instead of an autoencoder, a dimensionality reduction technique such as principal component analysis or random projection can be used. The system can perform feature selection and / or extraction in a supervised or unsupervised manner.

[0072] Example of supervised feature extraction and model development 6 illustrates a schematic of an embodiment of an autoencoder architecture 600. The autoencoder architecture may include a noise machine learning model 603 that removes unwanted artifacts and reduces noise (as described further herein), an encoder 604, and inputs to the encoder such as rhythm data 602A, ECG data 602B, PPG 602C, Q95 heart rate 602D, and acceleration 602E. The autoencoder architecture may also include an output of the encoder 606, a decoder 608 that receives the output of the encoder 606, and output features that represent the same inputs to the encoder such as rhythm data 602A, ECG data 602B, PPG 602C, Q95 heart rate 602D, and acceleration 602E.

[0073] In some embodiments, the encoder 604 can be trained along with the decoder 608 so that the decoder 608 outputs an approximation of the same data input to the encoder 604 (often with a reduced level of noise). The encoder 604 and the decoder 608 can be trained with multiple modalities so that the encoder 604 can be trained to analyze each of the data input to the encoder to extract features of the encoder 606 output. The inputs of the encoder 604, such as rhythm data 602A, ECG data 602B, PPG 602C, Q95 heart rate 602D, and acceleration 602E, can be aligned in time. Because the encoder-decoders (604, 606, 608) have been trained, the encoder 604 can automatically determine, based on the training, that for a particular data set of a patient, the end point of the ECG that aligns with a low point in the PPG where the heart rate is at the 5th percentile and the acceleration of that frame is high corresponds to the onset of AF. Advantageously, the encoder can be trained to evaluate multiple modalities at once to provide predictions of atrial fibrillation signatures and / or onset.

[0074] In some embodiments, the desired outcome can guide the feature extraction process. An autoencoder architecture can be trained to create a compressed representation of a large number of samples. The encoder 604 can be connected to another deep learning network architecture trained for classification or risk prediction using an appropriate loss function. The encoder can be implemented on the device and send the encoded data to a server, where the rest of the deep learning network uses the encoded features to predict the relevant outcome (classification probability, survival probability, outcome label, e.g., AF).

[0075] In some embodiments, the encoder can be trained with the decoder to generate an approximation of the original data, and then the pre-trained encoder can be connected to other deep learning network architectures that are trained simultaneously to predict multiple desired outcomes. Each architecture may predict a single outcome based on the same output 606 of the encoder, or may use a multi-task architecture to predict multiple outcomes simultaneously (such as risk of stroke, heart failure, hospitalization, etc.). During this training, the weights of the encoder can be kept frozen or allowed to continue to adjust. The advantage of using a pre-trained encoder with frozen weights is that the output can be a feature of layer 606. This single result can be transmitted and used from multiple deep learning networks trained with the encoder to predict one outcome per network. Such a solution reduces the amount of data transmitted, since one output can be used from multiple deep learning networks for prediction of multiple desired outcomes or a single multi-task deep learning network. Furthermore, the dimensionality of the bottleneck layer of the encoder, or the output of the encoder 606, can be adjusted to meet conflicting operational requirements and performance. For example, the number of dimensions may be reduced to reduce data transmission, to require less processing of the input data by the server, to require less storage, and / or the like. The number of dimensions of the encoder may be selected based on the desired algorithms on the server side, such as having multiple algorithms for different scenarios and / or desired results.

[0076] In some embodiments, the dimensionality of the encoder can be determined based on the desired accuracy of the output. For example, a smaller dimensionality will result in less accuracy but less network bandwidth usage. In some embodiments, the dimensionality can be frozen for predicting a particular set of outcomes. However, if a different set of outcomes is desired, the system can adjust the dimensionality to be appropriate for predicting a different set of outcomes, such as predicting hospitalizations versus predicting strokes.

[0077] In some embodiments, the training labels include one or more of disease / free classifications for different outcomes (stroke, MI, etc.), a set of time to event values ​​with failure or censoring indices, a set of economic outcomes related to medical costs, a set of length of hospital stays, a set of time to next device prescription, and / or a set of time to arrhythmia (e.g., atrial fibrillation) onset with failure or censoring indices.

[0078] In some implementations, the encoder 604 and decoder 608 are trained together. After training, the encoder 604 is applied to the patient data and the output of the encoder 606 is processed through a machine learning (ML) time-to-event model 612 that predicts the probability of survival / failure (occurrence or non-occurrence of an event) within a specified time window. The ML model is trained to predict the probability using the output 606 in combination with a failure time 614A (time until an event occurs) and a censor 614B (whether a failure was observed at that time) as the outcome.

[0079] Two options for risk-based model development are highlighted in FIG. 5. The autoencoder network receives a 2D tensor of time-aligned measurements or transforms (e.g., heart rate extracted from an ECG or PPG signal) at the encoder 604 and, through a process of reconstructing the signal, can reduce its dimensional representation (the bottleneck layer) as the output of the encoder 606. The 2D tensor has time as its first dimension and any combination of the aforementioned quantities as its second dimension. By using multiple of these inputs from the same, or multiple, devices, the autoencoder architecture can record and reconstruct a reduced dimensional representation of the partial or complete device signal tensor. In a first version, the bottleneck layer can be extracted and introduced as its input to the machine learning model 612 (if the bottleneck layer is more than 2D, it can be flattened into a vector). Standard parametric and non-parametric time-to-event statistical models (e.g., Cox proportional hazards model, parametric (Weibull, lognormal, etc.), or survival tree models) can be used to appropriately weight and integrate the most informative parameters from the bottleneck layer to fit a survival model.

[0080] In some embodiments, the transducer architecture can incorporate demographic data (age, race, place of residence, etc.), clinical data (current measurements), and past historical health data and features (either explainable or latent) to predict a sequence of outcomes, interventions, and associated costs that can be used to predict a patient's progress. These architectures can also incorporate longitudinal recordings to identify recordings or segments of recordings that best predict desired outcomes. The architecture can be used to predict a sequence of events of interest (such as the occurrence of a future atrial fibrillation episode, stroke, hospitalization, treatment, etc.) using longitudinal monitoring. Context can include events within a recording, events between longitudinally obtained recordings, and / or future events related to the measurements at hand (such as stroke, death, etc.). As an example, a cardiac monitor patch may include features indicative of atrial fibrillation that has not yet been detected by the patient.

[0081] In some embodiments, subsequent applications of monitoring patches may diagnose atrial fibrillation and corresponding stress based on information from the first patch. A series of patches may be used to estimate the probability of occurrence of an adverse event, such as a stroke. In this way, a transducer-based model trained with appropriate data may be used to automatically select the most relevant records to predict the next event of interest (e.g., patient unaware of atrial fibrillation status → asymptomatic atrial fibrillation → symptomatic atrial fibrillation → stroke) or incorporate treatment decisions to reduce the probability of such events. Additionally, the system may include predictions of hospitalization procedures and corresponding costs associated with treatment and hospitalization.

[0082] Advantageously, the encoder architecture can apply an encoder to generate features that can be processed through another algorithm, such as a neural network, to determine a desired outcome, such as an atrial fibrillation onset condition. Traditional algorithms have limited ability to capture temporal relationships in data over long time frames. However, architectures such as those described herein (LSTM, transformers, etc.) can naturally incorporate these relationships to make such predictions.

[0083] In some implementations, a machine learning model for cleaning noise (herein referred to as a noise machine learning model) can be trained together with a machine learning model for making inferences (herein referred to as an inference machine learning model). Training data can be input to the noise machine learning model, and the output of the noise machine learning model (or a derived signal thereof) can be input to the inference machine learning model. It is understood that additional computations or signal processing techniques can be performed before and / or after any of the machine learned models.

[0084] In other embodiments, the machine learning model can be the same model as the neural network that performs the inferences. A single machine learning model can be trained to both clean the data from unwanted artifacts and / or noise and to perform inferences on the data.

[0085] In other embodiments, the training of the machine-learned models can be done remotely from the monitor. However, one or more machine-learned models can be applied at the monitor. In some embodiments, one or more of the machine-learned models can be applied at the server side (such as all of the machine-learned models). For example, for short-term prediction of data, the monitor can run one or more machine-learned models. Advantageously, an alert can be displayed immediately to the user, either from the monitor or another computing device such as a mobile phone. Furthermore, when an inference is made, the device can trigger another process, such as applying a long-term prediction model or a different algorithm that makes different inferences (such as one that looks at different characteristics of the user). Such long-term prediction models can be run at the server side.

[0086] In some embodiments, data cleaning can be done at specific time periods, such as every second, every minute, every hour, and / or other. For example, artifacts can be found and / or noise can be reduced based on a 30-minute window every 30 minutes. In other implementations, a sliding window of a period of time can be used and data cleaning can be performed based on the amount the sliding window slides. For example, a 30-minute data window can be used and predictions can be made every 5 minutes.

[0087] In some embodiments, for long-term predictions that require longer periods of data, one or more of the machine-learned models can be run on the server side (such as all of the machine-learned models). For example, a monitor can record data for days, weeks, months, and / or such long periods. The data on the monitor can be downloaded to the server side, and a remote server can run long-term predictions on the data.

[0088] In some embodiments, a subset of the machine-learned models and / or a portion of the machine-learned models may run on the monitor side, with the remaining architecture running on the server side. For example, the noise machine-learned models run on the monitor, and the inference machine-learned models run on a remote server. In some embodiments, a portion of the inference machine-learned models (e.g., a first subset of layers) may run on the monitor, with the remaining portion of the models (e.g., a second subset of layers) running on the server side.

[0089] Advantageously, not all data needs to be sent from the monitor to the server, reducing the amount of data sent. This reduces network bandwidth requirements. Additionally, the monitor requires less memory since it does not need to store the entire dataset before sending it. The data can be processed through one or more machine learning models to output specific data, such as vectors. These vectors are then stored and sent to a remote server.

[0090] FIG. 7A shows an example network architecture 700 for applying one or more LSTM transformer blocks at the output of a decoder. The network architecture 700 can include input features to an encoder 604, an encoder 604, an encoder output 606, a decoder 608, RNN and / or LSTM blocks 702, 704, 706, 708. The architecture is trained using data including failure times 710A, and censored outputs 710B to generate probabilities of outcomes of interest within a pre-specified time window. The autoencoder can be pre-trained separately or can be part of the complete model training. It is included as a feature extraction / compression and noise reduction tool, especially for long (time-wise) records of combined inputs, when a method applicable to image recognition is desired, otherwise it is removed and the inputs are connected to the remaining layers.

[0091] FIG. 7B shows an example deep learning architecture 750 for applying a series of LSTM and / or transformer blocks at the output of the encoder. In some embodiments, the autoencoder can be composed of a combination of LSTM layers, RNNs and / or transformer blocks capturing unidirectional or bidirectional relationships. Advantageously, these architectures can incorporate signal features, their interactions with other signal features over time, and physical distances that inform a person's health status and risk of developing health problems in different time frames. In this way, the encoder architecture can naturally encode information of correlated features and / or timing between features (groups of patterns and timing characteristics), which can then be processed by other layers incorporating temporal relationships to become more powerful in predicting time to events.

[0092] In FIG. 7B , the inputs 602A, 602B, 602C, 602D, 602E, the encoder 604, and the output 606 of the encoder (e.g., a bottleneck) may be attached to a number of other layers (perhaps with repeated such blocks built into the architecture of the model), followed by one or more models of dense layers, which may also incorporate demographic or clinical variables as inputs, connected to an output layer representing right censored data for either AF, stroke, or other clinical outcome of interest.

[0093] In some implementations, the autoencoder architecture 750 can be trained from front to end. The autoencoder architecture 750 can be trained using the patient's past cardiac data (such as PPG, ECG, heart rate, and / or the like) and the resulting condition of the same patient. For example, the training data can include patient data from the past years with the current patient condition and / or subsequent patient condition and / or subsequent conditions from cardiac data recordings. The training data can be used to train the system by adjusting the weights in the encoder and transformer blocks. The pre-learned weights are adjusted in the encoder. The transformer following the encoder can be trained with random weights. During training, the encoder weights are expected to change more slowly than the rest of the network, including the transformer weights and other layers. Through training, the encoder weights are adjusted to emphasize specific features suitable for detecting atrial fibrillation, atrial fibrillation onset, and / or other arrhythmia components.

[0094] FIG. 7C shows a self-attention architecture for inferring the onset of cardiac arrhythmias. The self-attention architecture 770 may include multiple input nodes 772A, 772B, 772C, 772D (collectively referred to as input nodes 772), encoders 774A, 774B, 774C, 774D (collectively referred to as encoders 774), latent features 776A, 776B, 776C, 776D (collectively referred to as latent features 776), self-attention modules 778A, 778B, 778C, 778D (collectively referred to as self-attention modules 778), pooling and / or smoothing layers 780A, 780B, 780C, 780D (collectively referred to as pooling and / or smoothing layers 780), a concatenation module 782 that concatenates the outputs, and output nodes such as a time-to-event loss function 784, a time-to-failure 786, a censoring 788, and / or the like.

[0095] In some embodiments, the self-attention architecture can receive multiple different types of data as input, such as heart rate data 772A, ECG data 772B, PPG data 772C, acceleration data 772D, and / or the like. The architecture can include multiple encoders 774, each of which can take in one or more of the input data. For example, each encoder can be trained to output latent features for a respective input data type (e.g., a first encoder for rhythm data, a second encoder for ECG data).

[0096] In some embodiments, the self-attention module 778 can receive as input the latent features output by the encoder. The self-attention module 778 can find similarities between these latent features based on a distance metric or loss function. The self-attention module 778 can calculate a weighted output vector based on these similarities and the input latent features. For example, the self-attention module 778A can determine the output by weighting the rhythm data of the 2nd, 5th, and 10th minutes higher than the rest, and the self-attention module 778B can determine the output by weighting the ECG data of the 1st, 4th, and 6th minutes higher than the rest.

[0097] In some embodiments, the output of the self-attention module 778 may be pooled and / or flattened into a single value or vector. For example, a maximum or average value may be taken from the output of the self-attention module. The self-attention architecture may concatenate the outputs into a single vector. In some cases, the outputs 782 are aligned in time, so the lengths of the outputs 782 do not need to be the same size.

[0098] In some examples, the self-attention architecture may be trained using patient outcome information including failure time and censoring variables via a time-to-event loss function (e.g., negative log partial likelihood of Cox proportional hazards) 784 that predicts the probability of survival / failure (occurrence or non-occurrence of an event) within a specified time window, failure time 786 (time until an event occurs), and censoring 788 (whether failure was observed at that time). In some embodiments, the output may include a single-valued probability of survival or death, or a probability vector of survival or death (e.g., a probability vector over different time periods).

[0099] FIG. 7D illustrates a network training architecture for applying an LSTM transform block to the output of the decoder. In this example, the autoencoder and LSTM transformer blocks of FIG. 7A are shown. However, it is understood that other neural networks and / or architectures can be applied. During training, input data (such as historical patient data including rhythm, ECG, PPG, and / or the like) is input to the encoder 604. The data is processed through the encoder 604, the decoder 608, and the LSTM transformer blocks 702, 704, 706, 708. The output of the last block 708 of the LSTM transformer is input to a loss function 722. The true values ​​(e.g., from failure times 701A, censoring 710B) are also input to the loss function 722. Based on the output of the loss function 722, the system adjusts the weights of the neural network. For example, if the output of the last block 708 of the LSTM converter is close to the true value, the loss function 722 can output the minimum loss, which can cause the system to strengthen the weights of the neural network (e.g., increase the weights such that the neural network is strengthened to produce outputs closer to the true value).

[0100] FIG. 7E illustrates a network inference architecture for applying an LSTM transform block to the output of the decoder to generate hazard and / or survival probabilities. In this example, the autoencoder and LSTM transformer blocks of FIG. 7A are shown. However, it is understood that other neural networks and / or architectures can be applied. During inference, input data (such as current patient data including rhythm, ECG, PPG, and / or the like) is input to the encoder 604. The data is processed through the encoder 604, the decoder 608, and the LSTM transformer blocks 702, 704, 706, 708. The output of the last block 708 of the LSTM transformer is output via an output node or layer 730.

[0101] In some implementations, the potential outputs can include patient or user probabilities, such as hazard probability or survival probability. The hazard probability can be calculated based on the time t i Such probabilities may include the probability of failure (e.g., death of the patient) at time t i If a patient survives until time t i The probability of dying at time t can therefore be determined by filtering out users who have survived to that time and determining the probability of a user based on a subset of past users who have survived to that time. i It may include the probability of a user surviving beyond

[0102] FIG. 7F illustrates a network training architecture for applying a self-attention architecture using a loss function. In this example, the self-attention architecture of FIG. 7C is shown. However, it is understood that other neural networks and / or architectures can be applied. During training, input data (such as historical patient data including rhythm, ECG, PPG, and / or the like) is input to an encoder 774. The data is processed through the encoder 774, a self-attention architecture 778, a pooling / flattening layer 780, and a concatenation 782. The output of the concatenation 782 is input to a time-to-event loss function 784. The true values ​​(e.g., from the time to failure 786, censoring 788) are also input to the loss function 784. Based on the output of the loss function 784, the system adjusts the weights of the neural network.

[0103] FIG. 7G illustrates a network inference architecture for applying a self-attention architecture to generate hazard and / or survival probabilities. In this example, the self-attention architecture of FIG. 7C is illustrated. However, it is understood that other neural networks and / or architectures can be applied. During inference, input data (such as current patient data including rhythm, ECG, PPG, and / or the like) is input to an encoder 774. The data is processed through the encoder 774, a self-attention architecture 778, a pooling / flattening layer 780, and a concatenation 782. The output of the concatenation 782 is output via an output node or layer 796. Potential outcomes, such as hazard and / or survival probabilities 798, are determined.

[0104] Examples of unsupervised feature extraction, model development and patient clustering In some embodiments, the model is referred to as a "feature engineering" engine. Specifically, the model can provide a representation of the complexity and interaction structure of the features that most distinguish one record from another. As examples, autoencoder, principal component, and / or random projection architectures may be able to extract features of a complete or partial record that represent the overall information content of the record.

[0105] In some embodiments, all the records can be run through a model to obtain a "feature representation summary" (or "signature"), which can then be used with many other modeling techniques to identify the feature summary that best correlates with or predicts a particular outcome of the set described above. A "signature" can include a condensed mathematical representation of the original signal. A signature can consist of a set of numbers that represent the original signal in a lower dimensional space. A signature can be represented as a vector or tensor, where each number in the representation can be thought of as a coordinate of the original signal in a new k-dimensional space. As an example, in FIG. 8 (described further herein), the set of individual signatures run through the model are represented as points that have been projected into a two-dimensional space and clustered based on their similarity. A higher dimensional representation, including but not limited to the original signal, is given in the form of a line plot and color strips that represent the magnitude of the projection of the original signal into the greater than two dimensional coordinate space that the points represent, relative to the centroid of each cluster. The advantage of these models is that they extract information content that is not specific to a given outcome or modeling approach (e.g., classification vs. survival rate).

[0106] In some embodiments, every new recording can be run through the model to create a new "signature" for that recording. These signatures are stored in a database and clustered based on a suitable similarity metric. Such clusters can be created based on traditional machine learning clustering approaches or deep learning representations such as sparse autoencoders. This database, along with demographic and clinical data, can be used to develop supervised models for outcomes of interest.

[0107] Alternatively, based on the creation of a universe of patient signatures, which may also include demographic and clinical data, the system can assess the proximity of new records to existing ones and, if the latter are annotated with specific class labels and / or risk groups, the assignment of clinically useful features to the new records. Furthermore, sequential results can generate signatures for a single patient, providing information about the clinical course of a patient over time and allowing the evaluation of the speed and severity of transitions from healthy to diseased states and to new states after interventions.

[0108] Risk Prediction In some embodiments, many categories of algorithms are available for use with the extracted features as described herein. Examples of such algorithms include: time-to-event approaches (parametric / non-parametric survival analysis) in combination with demographic and clinical factors as predictors (combination of multiple methods) to time-to-event approaches; classification of event / no-event probabilities resulting from ML / DL classification methods in combination with demographic and clinical factors as predictors (combination of multiple methods) to time-to-event approaches; and / or deep learning approaches for event risk prediction, including methods using loss functions to those used in traditional statistical models for survival, LSTM, RNN and transformer models.

[0109] Reporting forecasts In some embodiments, the system can provide a digital representation of the risk prediction. The system can provide, for example, a risk score out of 100, where 0 is no risk and 100 guarantees the occurrence of a future event, and the corresponding risk of a particular event(s) for a given individual within a pre-specified follow-up time (see, e.g., FIG. 9). The system can include corresponding risk curves from validation studies as supporting evidence (see, e.g., FIG. 10). The curves may be stratified based on demographic characteristics, symptoms for monitoring, clinical characteristics such as CHA2DS2-VASc score, HAS-BLED score, treatment regimen, and / or the like. The system can include corresponding treatment benefit curves as a function of event risk score and other demographic / clinical characteristics based on retrospective or randomized prospective studies as supporting evidence of treatment benefit. The system can visually display "record signatures" and a measure of proximity to nearby records. The system can report treatment choices (specific interventions, medications, percentages of such information) and outcomes for patients in the vicinity of the record of interest.

[0110] Example of patient data clustering FIG. 8 shows an embodiment 800 of grouping (represented as a projection on a 2D plane) individual signatures from individuals who may progress differently with respect to an event window or may respond better to a particular intervention (e.g., such that the signature indicates the probability of responding to a given intervention). For example, various group signatures 802A, 802B, 802C can be projected onto the 2D plane as circles 804A, 804B, 804C of various groupings. Stars 806A, 806B, 806C represent representative signatures (centroids) of groups with similar characteristics, and circles represent individual observations that are "closer" in the sense of mathematical distance to a particular centroid. Among other dimensionality reduction methods, the encoder can create a reduced-dimensional representation of the bottleneck information, such as a two-dimensional vector, from a large amount of data. If it is desired to reduce the dimensionality to two dimensions, the system can determine a projection into these two dimensions. For example, the second dimension can contain two different groups of people, the first group benefiting from an intervention and the second group not, and if a patient is close to the centroid of the group of people who benefited from a treatment, the system can predict that the patient could benefit from a certain type of intervention.

[0111] In some embodiments, the system can create clusters from multiple modalities. The system can use one or more neural networks (such as deep learning algorithms) to generate a representation of the signals in a latent space. These neural networks can be trained to incorporate a contrastive loss to assign distances between signals. The contrastive loss can include training examples whose values ​​are "close" (e.g., measurement vectors that have similar characteristics and correspond to individuals with the same or similar clinical outcomes) along with other characteristics that are "distant" from the reference sample.

[0112] In some embodiments, the system can take a sample as a reference and contrast it with other samples that are "close" and many other samples that are "distant." A neural network is trained to generate and apply a function that can compute a distance matrix between samples. This distance metric is based on a latent representation of each sample (e.g., a projection into a coordinate system that places more similar elements closer together and more dissimilar elements farther apart).

[0113] In some embodiments, the system can apply these latent vectors as the basis of a clustering algorithm (e.g., k-means, hierarchical clustering, BIRCH, etc.) to create clusters of patients with similar outcomes and establish their representative samples ("centroids"). Advantageously, the system can make better predictions based on a holistic view of the data (neighborhoods of patient samples with similar outcomes) rather than corresponding to any single patient sample. New samples can then be clustered using a combination of the latent projections obtained from the contrastive loss-based network and a clustering algorithm. In another embodiment, the same latent features can be used to create a classification of outcome classes through the use of a classification neural network. The trained network can then be used to predict any new observations.

[0114] FIG. 9 shows a schematic diagram 900 of a risk score alert for a patient whose record contains markers indicating high confidence of atrial fibrillation onset / detection within a pre-specified follow-up period, but no events have occurred within the current patch wear time. The risk score can be a score between 1-100, such as 75, or other scores within a score range, such as between 1-5, 1-10, or other scores, such as letter grading. The atrial fibrillation portion and the rest (non-atrial fibrillation) portion can represent a population of people. For example, 80% of the population does not have atrial fibrillation, while 20% does. The risk score can be applied to those who do not have atrial fibrillation to help develop a monitoring strategy to manage future risk. The risk score can also be applied to those who have atrial fibrillation to manage their atrial fibrillation condition, such as pharmacological, surgical, or lifestyle interventions. The risk score can be applied to the entire population, whether the patient has atrial fibrillation or not, to optimize the public health of the population.

[0115] FIG. 10 shows a graph 1000 of an embodiment of risk curves stratified by demographic or clinical characteristics. The graph can show a risk score curve based on cumulative hazard over time, such as hours / days / years. The same risk score may have different probabilities of outcomes based on such characteristics. For example, a low risk curve is suggestive for patients aged 30-50. Thus, risk scores may have different meanings for patients of different ages. In some embodiments, the risk curves can be stratified by clinical characteristics, such as the patient's current measurements (e.g., blood pressure, PPG). In some embodiments, the risk curves can be stratified by past health history.

[0116] Example of report implementation FIG. 11 shows an example of a report 1100 for a patient with no identified atrial fibrillation. The report can include a display of an atrial fibrillation risk score 1102 indicating the likelihood that the patient will develop atrial fibrillation in the future. The number 78 can represent a high score (and / or be displayed within a circle). The report can include a display of a relative risk 1104. 6X can represent the relative risk as a fold change for this patient, for example, relative to someone with the same age and sex characteristics but without the abnormal characteristics. The relative risk can be derived from a model and / or can be another representation of risk determined based on morphology, heart rate, demographics, and / or other findings.

[0117] In some embodiments, the report 1100 can include a graph 1106 of the predicted atrial fibrillation risk over the next five years. In some embodiments, the graph can span different time periods, such as days, weeks, months, etc. The graph 1106 can display a risk curve as a function of the score (after five years). The report 1100 can include another graph 1108 that provides a prediction of the person's absolute risk progression as a function of time. The absolute risk can be displayed as a percentage, such as an absolute risk of 18% after five years. The report 1100 can include a table 1110 that provides the risk of onset over time, such as months, and the progression of the atrial fibrillation risk score and / or risk.

[0118] 12A and 12B show an embodiment of a report 1200, 3500 of a patient with atrial fibrillation identified in a cardiac monitor recording. The report 1200, 1250 can indicate the patient's risk of adverse outcome over time. The outcome can correspond to a risk score indicating the likelihood of an event occurring within a time period, such as 36 months. The score and / or risk curve can be generated by a deep learning algorithm, which can take into account, among other things, the patient's ECG morphology, findings, heart rate, and / or demographic conditions. The report 1200, 3500 can include a risk curve graph representing the risk associated with an outcome of interest, such as stroke 1202, heart failure 1204, hospitalization 1206, and / or mortality 1208. Each of the risk curve graphs can include risk curves without intervention (solid line), with atrial fibrillation ablation (first dotted line), and with anticoagulant therapy (second dotted line). For example, a risk curve without intervention can represent risk as a function of time for a cohort with the same characteristics but without intervention. Risk circles 1203, 1205, 1207, 1209 can show levels of risk, such as high risk with a score of 78. Risk curves can show reduction in risk based on specific treatments. As an example, in stroke curve 1202, a patient's risk is reduced by 28% with atrial fibrillation ablation and 23% with anticoagulation therapy.

[0119] In some embodiments, the system can generate risk scores and / or curves for one or more other physiological signals. For example, the system can generate curve graphs of future risk for stroke, heart failure, cardiovascular hospitalization, structural heart disease (left / right ventricular hypertrophy, underinfusion rate, atrial / mitral regurgitation, stenosis, tricuspid regurgitation, etc.), and / or the like.

[0120] Flow Diagram Example 13A illustrates an example of a flow diagram 1300 for training an encoder and a decoder and applying the encoder to a wearable device to infer the onset of cardiac arrhythmia. At block 1302, the system can receive cardiac data and train the encoder and the decoder to output the cardiac data. The encoder can output features of the cardiac data. The encoder and the decoder can be trained to output data that represents the cardiac data with sufficient fidelity, such as an input threshold level of similarity.

[0121] At block 1304, the system may generate a software package including an encoder for the wearable patch for the wearable patch to apply the encoder to the new patient's cardiac data. At block 1306, the wearable patch may record the patient's cardiac data, such as using electrodes on the wearable patch. At block 1308, the wearable device may apply the patient's cardiac data using the encoder to generate features of the patient's cardiac data. The encoder may output features such as boundaries, edges, linear slopes, offsets, signal magnitudes, and / or the like. The output of the encoder may be of a smaller dimensionality than the data input to the encoder.

[0122] At block 1310, the wearable device can transmit the features output from the encoder to a remote computing device. Advantageously, the wearable device may have many technical advantages, such as longer battery life, smaller storage requirements, and / or advantages as further described herein. At block 1312, the remote computing device can apply the features to a machine learning process for inferring the onset of cardiac arrhythmia. The machine learning process can be a separate process from the encoder and decoder. The machine learning process can be trained separately from the encoder and decoder, as further described herein.

[0123] 13B shows an example of a flow diagram 1330 for training an encoder and a machine learning model, applying the encoder on a wearable device, and applying the machine learning model on a remote computing device to infer the onset of cardiac arrhythmia. At block 1332, the system can receive cardiac data and train the encoder and a separate machine learning model to output an inference of the onset of cardiac arrhythmia. The training can apply past cardiac data of the patient and the cardiac arrhythmia status of the same patient at a later time, such as one month later, five months later, one year later, five years later, etc. Advantageously, the encoder can be specifically trained to output the most relevant features of the cardiac data to infer the onset of cardiac arrhythmia.

[0124] At block 1334, the system may generate a software package including an encoder for the wearable patch for the wearable patch to apply the encoder to the new patient's cardiac data, such as at block 1304 and / or as further described herein. At block 1336, the wearable patch may record the patient's cardiac data, such as by using electrodes on the wearable patch, such as at block 1306 and / or as further described herein. At block 1338, the wearable device may apply the patient cardiac data using the encoder to generate features of the patient cardiac data, such as at block 1308 and / or as further described herein.

[0125] At block 1340, the wearable device can transmit the features output from the encoder to a remote computing device, such as at block 1310 and / or as further described herein. At block 1342, the remote computing device can apply a machine learning process to the features to infer the onset of cardiac arrhythmia. The machine learning process is trained with the encoder; however, the encoder is applied on the wearable device and the machine learning process is applied on the remote computing device.

[0126] 13C illustrates an example of a flow diagram 1360 for training an encoder, decoder, and machine learning model using past patient cardiac data to infer the onset of cardiac arrhythmia. At block 1362, the system can access records of past patient cardiac data. For example, the past patient cardiac data can include data of hundreds or thousands of past patients and their past corresponding cardiac data. At block 1364, the system can train the encoder and decoder using the past patient cardiac data, such as at block 1302 and / or as further described herein.

[0127] In block 1366, the system can access the patient status of cardiac arrhythmia at a later time in the cardiac data measurement record. For example, a patient's past cardiac data may be from five years ago, and the data did not indicate a cardiac arrhythmia five years ago. However, the patient's status five years later is that of a patient who later developed a cardiac arrhythmia.

[0128] At block 1368, the system can train a machine learning process to infer the onset of cardiac arrhythmia based on the patient's condition. Data that can be used can include features output by an encoder trained using past patient cardiac data. Thus, the condition and the past patient cardiac data can correspond to the same individual.

[0129] At block 1370, the wearable patch can record patient cardiac data, such as using electrodes on the wearable patch, such as at block 1306 and / or as further described herein. At block 1372, the wearable device can apply the patient cardiac data using an encoder to generate features of the patient cardiac data, such as at block 1308 and / or as further described herein. At block 1374, the remote computing device can apply the features to a machine learning process to infer the onset of a cardiac arrhythmia, such as at block 1312 and / or as further described herein.

[0130] Computing system and method In some examples, the systems, tools, and methods of using same described above enable interactivity and data collection performed by a computing system 1400. Figure 14 is a block diagram illustrating an example in which a computing system 1400 is also in communication with a network 14002 and various external computing systems 1404, such as a wearable system 14005, a gateway device 13006, etc. The computing system 1400 may be used to implement the systems and methods described herein. Although the external systems 1404 are shown grouped, it will be appreciated that each system may be external and / or remotely located from one another.

[0131] Computing system and method In some examples, the systems, tools, and methods of using same described above enable interactivity and data collection performed by a computing system 1400. Figure 15 is a block diagram illustrating an example in which a computing system 1400 is also in communication with a network 14002 and various external computing systems 1404, such as a wearable system 14005, a gateway device 13006, etc. The computing system 1400 may be used to implement the systems and methods described herein. Although the external systems 1404 are shown grouped, it will be appreciated that each system may be external and / or remotely located from one another.

[0132] In some embodiments, the computing system 1400 includes one or more computing devices, such as, for example, a server, a laptop computer, a mobile device (e.g., a smartphone, a smartwatch, a tablet, a personal digital assistant), a kiosk, an automobile console, or a media player. In one implementation, the computing device 1400 includes one or more central processing units (CPUs) 1405, each of which may include a conventional or proprietary microprocessor. The computing device 1400 further includes one or more memories 1430, such as a random access memory (RAM) for temporary storage of information, one or more read-only memories (ROM) for permanent storage of information, and one or more mass storage devices 1420, such as a hard drive, a diskette, a solid-state drive, or an optical media storage device. In certain embodiments, a processing device, a cloud server, a server, or a gateway device may be implemented as the computing system 1300. In one example, the modules of the computing system 1400 are connected to the computer using a standard-based bus system. In different embodiments, the standards-based bus system may be implemented in, for example, Peripheral Component Interconnect (PCI), MicroChannel, Small Computer computing system Interface (SCSI), Industrial Standard Architecture (ISA), and Extended ISA (EISA) architectures. Additionally, the functionality provided in the components and modules of computing device 1400 may be combined into fewer components and modules or further separated into additional components and modules.

[0133] Computing device 1400 may be controlled and coordinated by operating system software, such as iOS, Windows XP, Windows Vista, Windows 7, Windows 8, Windows 10, Windows Server, Embedded Windows, Unix, Linux, Ubuntu Linux, SunOS, Solaris, Blackberry OS, Android, or other operating systems. In a Macintosh system, the operating system may be an available operating system, such as MAC OS X. In other embodiments, computing device 1400 may be controlled by its own operating system. A conventional operating system, among other things, controls and schedules computer processes for execution, performs memory management, provides file system, networking, I / O services, and provides a user interface, such as a graphical user interface (GUI).

[0134] The exemplary computing device 1400 may include one or more I / O interfaces and devices 1410, such as a touchpad or touchscreen, but also a keyboard, a mouse, and a printer. In one implementation, the I / O interfaces and devices 1410 include one or more display devices (such as a touchscreen or monitor) that enable visual presentation of data to a user. More specifically, the display devices may provide for presentation of, for example, GUIs, application software data, and multimedia presentations. The computing system 1400 may also include one or more multimedia devices 1440, such as, for example, a camera, speakers, a video card, a graphics accelerator, and a microphone.

[0135] In one embodiment of the computing system and application tools, the I / O interfaces and devices 1410 may provide a communication interface to various external devices. In one embodiment, the computing device 1400 is electronically coupled to a network 14002, including one or more of a local area network, a wide area network, and / or the Internet, via a wired, wireless, or combination wired and wireless communication link 1415, for example. The network 14002 may communicate with various sensors, computing devices, and / or other electronic devices via the wired or wireless communication link.

[0136] In some embodiments, the filter criteria, signals, and data may be processed by the rhythm inference module application tool and provided to the computing system 1400 from one or more data sources 1410 over the network 14002 in accordance with the methods and systems described herein. The data sources may include one or more internal and / or external databases, data sources, and physical data stores. The data sources 1410, the external computing system 1404, and the rhythm interface module 1490 may include databases for storing data (e.g., feature data, raw signal data, patient data) in accordance with the systems and methods described above, and databases for storing data processed in accordance with the systems and methods described above (e.g., data sent to the sensor, data sent to the clinician). In one embodiment of FIG. 6, the sensor data 1550 may, in some embodiments, store data received from the sensor, data received from the clinician, etc. The rules database 1560 may, in some embodiments, store data (e.g., instructions, preferences, profiles) that establish threshold parameters for analyzing the feature data. In some embodiments, one or more of the databases or data sources may be implemented using relational databases, such as Sybase, Oracle, CodeBase, MySQL, SQLite, and Microsoft® SQL Server, as well as other types of databases, such as flat file databases, entity-relationship databases, and object-oriented databases, NoSQL databases, and / or record-based databases.

[0137] The computing system includes a rhythm interface module 1490, which in one embodiment may be stored in mass storage device 1420 as executable software code executed by CPU 1405. The rhythm interface module 1490 may include a feature module 1510, an alternative data module 1520, an inference module 1530, a feedback module 1540, a sensor data database 1550, and a rules database 1560. These modules may include, by way of example only, software components, object-oriented software components, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, variables, and other components. These modules are configured to perform the processes disclosed herein.

[0138] In some implementations, the various components shown or described herein may be implemented as software and / or firmware on a processor, controller, ASIC, FPGA, and / or dedicated hardware. The software or firmware may include instructions stored in a non-transitory computer-readable memory. The instructions may be executed by the processor, controller, ASIC, FPGA, or dedicated hardware. Hardware components such as controllers, processors, ASICs, FPGAs, etc. may include logic circuitry.

Claims

1. 1. A computing system for inferring an onset of cardiac arrhythmia in a user, comprising: The computing system is a sensor configured to detect a cardiac signal from a user; a hardware processor configured to apply the detected cardiac signal to a first machine learning process; The first machine learning process is configured to infer an onset of a cardiac arrhythmia based on the detected cardiac signal of the user; The first machine learning process is accessing historical cardiac signal patient data for a plurality of patients; accessing cardiac arrhythmia status for a plurality of patients, the cardiac arrhythmia status for the plurality of patients being determined since a time that past cardiac signal patient data was recorded; training a first machine learning process based on past cardiac signal patient data and cardiac arrhythmia status to infer the onset of cardiac arrhythmia, where the user is different from the plurality of patients; Computing system.

2. The hardware processor is further configured to apply a second machine learning process; a second machine learning process trained to remove or detect noise segments or artifacts from the detected cardiac signal of the user; The first machine learning process receives as an input a cardiac signal from which noise segments or artifacts have been removed; The computing system of claim 1 .

3. The hardware processor is further configured to apply a second machine learning process; a second machine learning process trained to reduce noise from the detected cardiac signal of the user; The first machine learning process receives as an input the noise-reduced cardiac signal. A computing system according to claim 1 or 2.

4. The occurrence of cardiac arrhythmia, including the probability of identifying a cardiac arrhythmia at a later time point, A computing system according to any one of claims 1 to 3.

5. at least a portion of the historical cardiac signal patient data is not indicative of a cardiac arrhythmia at the time the historical cardiac signal patient data is recorded; A computing system according to any one of claims 1 to 4.

6. The first machine learning process is further configured to process the photoplethysmogram (PPG) data, the heart rate or the acceleration data to infer the onset of cardiac arrhythmia. A computing system according to any one of claims 1 to 5.

7. The cardiac arrhythmia includes at least one of ventricular tachycardia, supraventricular tachycardia, ectropion, or ventricular fibrillation; A computing system according to any one of claims 1 to 6.

8. The computing system includes a wearable smart watch. A computing system according to any one of claims 1 to 7.

9. The sensor is an electrode. A computing system according to any one of claims 1 to 8.

10. 1. An electronic device for monitoring a physiological signal of a user, comprising: The electronic device is an adhesive assembly including a housing and flexible wings, the housing containing the circuit board, the flexible wings extending from the housing and configured to conform to a surface of a user; a sensor coupled to the flexible wing, the sensor being in electrical communication with the circuit board and configured to be positioned in conformal contact with a surface of a user to detect a physiological signal; one or more hardware processors configured to apply a machine learning process to the detected physiological signals; The machine learning process is configured to infer an onset of a cardiac arrhythmia based on the detected physiological signal; The machine learning process is accessing historical cardiac signal patient data for a plurality of patients; accessing cardiac arrhythmia status for a plurality of patients, the cardiac arrhythmia status for the plurality of patients being determined since a time that past cardiac signal patient data was recorded; training a machine learning process based on past cardiac signal patient data and cardiac arrhythmia status to infer the onset of cardiac arrhythmia, where the user is different from the plurality of patients; Trained by electronic equipment.

11. Inferring the onset of the cardiac arrhythmia includes generating a cardiac arrhythmia risk score at a time point after the physiological signal is detected by the sensor. The electronic device according to any one of claims 1 to 10.

12. the one or more hardware processors are further configured to determine that no cardiac arrhythmia is currently present in the detected physiological signal prior to inferring an onset of the cardiac arrhythmia. An electronic device according to any one of claims 1 to 11.

13. Inferring the onset of a cardiac arrhythmia is responsive to determining that a cardiac arrhythmia is not currently present in the detected physiological signal. An electronic device according to any one of the preceding claims.

14. The one or more hardware processors are further configured to adjust functionality of the electronic device based on the inferred onset of cardiac arrhythmia. An electronic device according to any one of claims 1 to 13.

15. Adjusting the function of the electronic device includes adjusting a window of a physiological signal processed by the electronic device. An electronic device according to any one of the preceding claims.

16. In response to the risk score satisfying the threshold score, the one or more hardware processors are further configured to send an alert to a user or physician device. An electronic device according to any one of the preceding claims.

17. The one or more hardware processors are further configured to recommend a therapy or intervention for an episode of cardiac arrhythmia. An electronic device according to any one of the preceding claims.

18. Recommending treatment or intervention is based on clustering of previous patient data. An electronic device according to any one of the preceding claims.

19. A cluster contains multiple centroids, the one or more hardware processors apply a distance process from the centroid of the corresponding cluster to the user's record to recommend a treatment; An electronic device according to any one of the preceding claims.

20. The clustering clusters patients based on cardiac arrhythmia status, intervention type, and outcome of application of the intervention type; An electronic device according to any one of the preceding claims.

21. 1. A method for training a machine learning process to infer the onset of cardiac arrhythmia, comprising: The method is: accessing historical cardiac signal patient data for a plurality of patients, at least a portion of the historical cardiac signal patient data being not indicative of a cardiac arrhythmia at the time the cardiac signal was recorded; accessing cardiac arrhythmia status for a plurality of patients, the status of the plurality of patients being determined subsequent to the time the cardiac signals were recorded; training a first machine learning process based on past cardiac signal patient data and cardiac arrhythmia status to infer an onset of cardiac arrhythmia, where the user is distinct from the plurality of patients; method.

22. The method is: detecting a cardiac signal from a user; applying the detected cardiac signal to a first machine learning process; The first machine learning process is configured to infer an onset of a cardiac arrhythmia based on the input data; The first machine learning process is accessing historical cardiac signal patient data for a plurality of patients; accessing cardiac arrhythmia status for a plurality of patients, the status of the plurality of patients being determined since a time that past cardiac signal patient data was recorded; training a first machine learning process to infer an occurrence of cardiac arrhythmia of a user based on past cardiac signal patient data and a cardiac arrhythmia status, the user being distinct from the plurality of patients; method.

23. When performed by a processor, accessing historical cardiac signal patient data for a plurality of patients, at least a portion of the historical cardiac signal patient data being not indicative of a cardiac arrhythmia at the time the cardiac signal was recorded; accessing cardiac arrhythmia status for a plurality of patients, the status of the plurality of patients being determined subsequent to the time the cardiac signals were recorded; A non-transitory computer storage medium storing computer-implementable instructions for causing a processor to execute a method including the steps of: training a first machine learning process based on past cardiac signal patient data and cardiac arrhythmia status to infer the onset of cardiac arrhythmia, wherein the user is distinct from a plurality of patients.

24. When performed by a processor, detecting a cardiac signal from a user; applying the detected cardiac signal to a first machine learning process; The first machine learning process is configured to infer an onset of a cardiac arrhythmia based on the input data; The first machine learning process is accessing historical cardiac signal patient data for a plurality of patients; accessing cardiac arrhythmia status for a plurality of patients, the status of the plurality of patients being determined since a time that past cardiac signal patient data was recorded; A non-transitory computer storage medium storing computer-implementable instructions that cause a processor to execute a method comprising: training a first machine learning process to infer an onset of cardiac arrhythmia in a user based on past cardiac signal patient data and cardiac arrhythmia status, the user being distinct from a plurality of patients.

25. Training the first machine learning process further includes training the first machine learning process to infer a hospitalization, heart failure onset, stroke onset, or death status. The method according to any one of claims 1 to 24.

26. The method is: training a second machine learning process to infer a hospitalization, heart failure event, stroke event, or death status; the input to the second machine learning process includes a cardiac signal of a longer duration than the cardiac signal input to the first machine learning process; The method according to any one of claims 1 to 25.

27. The method is: determining an atrial fibrillation load from the detected cardiac signal; Atrial fibrillation burden includes the amount of time a user spends in atrial fibrillation during a period of time; The method according to any one of claims 1 to 26.

Images