阅读说明：本技术 基于机器学习的去极化识别和心律失常定位可视化 (Machine learning based depolarization identification and arrhythmia localization visualization ) 是由 T·D·哈达德 N·查克拉瓦希 D·R·马斯格鲁夫 A·拉德克 E·N·沃曼 R·卡特拉 于 2020-04-13 设计创作，主要内容包括：公开了包括对医疗设备存储的发作数据,包括心电描记图应用机器学习模型的技术。在一些实例中,基于对所述发作数据应用一个或多个机器学习模型,对于多个心律失常类型分类中的每一个,处理电路系统导出分类激活数据,其指示所述分类在与发作相关联的时间段内的变化可能性。所述处理电路系统可显示所述心律失常类型分类在所述时间段内的所述变化可能性的曲线图。在一些实例中,处理电路系统可使用心律失常类型可能性和去极化可能性识别在所述发作期间的去极化,例如QRS波群。(Techniques are disclosed that include applying a machine learning model to episode data, including electrocardiography, stored on a medical device. In some examples, based on applying one or more machine learning models to the episode data, for each of a plurality of arrhythmia type classifications, processing circuitry derives classification activation data that indicates a likelihood of change of the classification over a time period associated with an episode. The processing circuitry may display a graph of the likelihood of change in the arrhythmia type classification over the period of time. In some examples, the processing circuitry may use the arrhythmia type likelihood and the depolarization likelihood to identify depolarizations, such as QRS complexes, during the episode.)

1. A medical device system, comprising:

a medical device configured to:

sensing an electrocardiography of a patient via a plurality of electrodes, an

Storing episode data for an episode, wherein the episode is associated with a time period and the episode data includes the electrocardiography sensed by the medical device during the time period; and

processing circuitry configured to:

the seizure data is received and the data is transmitted,

applying one or more machine learning models to the episode data, the one or more machine learning models configured to output a respective likelihood value for each of a plurality of arrhythmia type classifications, each of the likelihood values representing a likelihood that the respective arrhythmia type classification occurred at any point during the period of time,

deriving, for each of the arrhythmia type classifications, classification activation data indicative of a likelihood of change of the respective arrhythmia type classification over the period of time based on applying the one or more machine learning models to the episode data, and

displaying a graph of the likelihood of the change in the arrhythmia type classification over the time period to a user.

2. The medical device system of claim 1, wherein the processing circuitry is configured to display the graph in conjunction with the electrocardiography.

3. The medical device system of claim 1 or 2, wherein the processing circuitry is configured to indicate on the graph a time of at least one higher likelihood of the at least one arrhythmia type classification relative to other times on the graph of at least one of the arrhythmia type classifications.

4. The medical device system of claim 3, wherein the processing circuitry is configured to:

indicating, based on the output of the one or more machine learning models, that the at least one arrhythmia type classification occurred at any point during the time period; and

indicating on the graph a time of at least one higher likelihood of the at least one arrhythmia type classification in response to indicating that the at least one arrhythmia type classification occurred at any point during the time period.

5. The medical device system of any of claims 1-4, wherein the plurality of arrhythmia type classifications includes a plurality of bradycardias, pauses, ventricular tachycardias, ventricular fibrillation, supraventricular tachycardia, atrial fibrillation, atrial flutter, sinus tachycardia, ventricular premature beats, atrial premature beats, broad group tachycardia, and atrioventricular conduction blocks.

6. The medical device system of any of claims 1-5, wherein each of the one or more machine learning models includes a plurality of layers, and wherein deriving the activation data includes deriving the activation data from outputs of intermediate layers of the plurality of layers.

7. The medical device system of claim 6, wherein the intermediate layer comprises a global average pooling layer.

8. The medical device system of any one of claims 1-7,

wherein the one or more machine learning models include one or more arrhythmia classification machine learning models configured to output a set of respective arrhythmia type likelihood values for each of the plurality of arrhythmia type classifications, each of the arrhythmia type likelihood values of the set representing a likelihood that the respective arrhythmia type classification occurred at a respective time during the time period, and

wherein the processing circuitry is configured to:

applying one or more depolarization detection machine learning models to the episode data, the one or more depolarization detection machine learning models configured to output a set of depolarization likelihood values, each of the depolarization likelihood values of the set representing a likelihood of depolarization occurring at a respective time during the time period; and

identifying one or more depolarizations during the episode based on the arrhythmia type likelihood value and the depolarization likelihood value.

9. The medical device system of claim 8, wherein each of the one or more arrhythmia classification machine learning models includes a plurality of layers, and the processing circuitry is configured to derive the set of arrhythmia type likelihood values from outputs of intermediate layers of the plurality of layers.

10. The medical device system of claim 8 or 9, wherein to identify the one or more depolarizations based on the arrhythmia type likelihood value and the depolarization likelihood value, the processing circuitry is configured to apply the one or more depolarization detection machine learning models to the episode data and the arrhythmia type likelihood value.

11. The medical device system of any of claims 8-10, wherein to identify the one or more depolarizations based on the arrhythmia type likelihood value and the depolarization likelihood value, the processing circuitry is configured to at least one of:

modifying one or more of the likelihood of depolarization values based on one or more of the arrhythmia type likelihood values; or

Modifying a depolarization likelihood threshold based on one or more of the arrhythmia type likelihood values.

12. The medical device system of any of claims 8-11, wherein the processing circuitry is configured to label each of the one or more identified depolarizations as one of a plurality of depolarization types based on the arrhythmia type likelihood value, wherein the plurality of depolarization types include a plurality of normal, ventricular premature beats, atrial premature beats, noise, or artifacts.

13. The medical device system of any of claims 1-12, wherein the processing circuitry includes processing circuitry of a computing device.

14. The medical device system of any one of claims 1-13, wherein the medical device is implantable.

15. A non-transitory computer-readable medium comprising instructions that, when executed by processing circuitry of a computing system, cause the computing system to:

receiving episode data for an episode stored by a medical device of a patient, wherein the episode is associated with a time period and the episode data includes an electrocardiography sensed by the medical device during the time period;

applying one or more machine learning models to the episode data, the one or more machine learning models configured to output a respective likelihood value for each of a plurality of arrhythmia type classifications, each of the likelihood values representing a likelihood that the respective arrhythmia type classification occurred at any point during the period of time;

deriving, for each of the arrhythmia type classifications, classification activation data indicative of a likelihood of change of the respective arrhythmia type classification over the period of time based on applying the one or more machine learning models to the episode data; and

displaying a graph of the likelihood of the change in the arrhythmia type classification over the period of time.

Technical Field

The present disclosure relates generally to medical devices, and more particularly to analysis of signals sensed by medical devices.

Background

Medical devices may be used to monitor physiological signals of a patient. For example, some medical devices are configured to sense an Electrocardiography (EGM) signal, such as an Electrocardiogram (ECG) signal, which indicates electrical activity of the heart via electrodes. Some medical devices are configured to detect the occurrence of an arrhythmia, commonly referred to as a seizure, based on cardiac EGMs and, in some cases, data from additional sensors. Example arrhythmia types include asystole, bradycardia, ventricular tachycardia, supraventricular tachycardia, broad group tachycardia, atrial fibrillation, atrial flutter, ventricular fibrillation, atrioventricular conduction block, ventricular premature beats, and atrial premature beats. The medical device may store cardiac EGMs and other data collected during a time period that includes an episode as episode data. The medical device may also store episode data for a period of time in response to user input, such as input from a patient.

The computing system may obtain episode data from the medical device to allow a clinician or other user to review episodes. A clinician may diagnose a medical condition of the patient based on the identified occurrence of the intra-episode arrhythmia. In some instances, a clinician or other reviewer may review the issue data to annotate the episode, including determining whether an arrhythmia detected by the medical device actually occurred, prioritizing the episode, and generating a report for further review by the clinician prescribing the medical device for the patient or otherwise responsible for the care of the particular patient.

Disclosure of Invention

In general, this disclosure describes techniques for classifying, annotating, reporting episode data, including cardiac EGM data, using one or more machine learning models. In some examples, the processing circuitry applies one or more arrhythmia classification machine learning models to the episode data. The one or more machine learning models output respective likelihood values representing a likelihood that the respective arrhythmia type classification occurred at any point during the episode. The processing circuitry also derives classification activation data for each arrhythmia type classification that indicates a likelihood of a change in the classification over a period of time, and may display a graph of the likelihood of a change over a period of time. The displayed likelihood graph over time may help the user understand the reasoning of any arrhythmia detection during an episode, particularly when displayed in conjunction with the underlying cardiac EGM.

In some examples, the processing circuitry applies one or more depolarization detection machine learning models to the onset data. The one or more depolarization detection machine learning models are configured to output a set of depolarization likelihood values that represent the likelihood of depolarization at different times during the episode. The processing circuitry may use the depolarization and arrhythmia type classification likelihood values to improve its ability to detect depolarizations during an episode.

In one example, the present disclosure describes a computer-implemented method that includes receiving, by processing circuitry of a medical device system, episode data for an episode stored by a medical device of a patient, wherein the episode is associated with a time period, and the episode data includes an electrocardiography sensed by the medical device during the time period. The method also includes applying, by the processing circuitry, one or more machine learning models to the episode data, the one or more machine learning models configured to output a respective likelihood value for each of a plurality of arrhythmia type classifications, each of the likelihood values representing a likelihood that the respective arrhythmia type classification occurred at any point during the time period. The method also includes deriving, by the processing circuitry and for each of the arrhythmia type classifications, classification activation data indicative of a likelihood that the classification varies over the period of time based on applying one or more machine learning models to the episode data. The method also includes displaying, by the processing circuitry and to a user, a graph of a likelihood that the arrhythmia type classification changes over a period of time.

In another example, a computer-implemented method includes receiving, by processing circuitry of a medical device system, episode data for an episode stored by a medical device of a patient, wherein the episode is associated with a time period and the episode data includes an electrocardiography sensed by the medical device during the time period. The method also includes applying, by the processing circuitry, one or more arrhythmia classification machine learning models to the episode data, the one or more arrhythmia classification machine learning models configured to output a set of respective arrhythmia type likelihood values for each of a plurality of arrhythmia type classifications, each of the arrhythmia type likelihood values in the set representing a likelihood that the respective arrhythmia type classification occurred at a respective time during the time period. The method also includes applying, by the processing circuitry, one or more depolarization detection machine learning models to the onset data, the one or more depolarization detection machine learning models configured to output a set of depolarization likelihood values, each of the set depolarization likelihood values representing a likelihood of depolarization occurring at a respective time during the time period. The method also includes identifying, by the processing circuitry, one or more depolarizations during the episode based on the arrhythmia type likelihood value and the depolarization likelihood value.

This summary is intended to provide an overview of the subject matter described in this disclosure. It is not intended to provide an exclusive or exhaustive explanation of the apparatus and methods described in detail in the following figures and description. Additional details of one or more examples are set forth in the accompanying drawings and the description below.

Drawings

Fig. 1 is a conceptual diagram illustrating an example of a medical device system configured to detect cardiac depolarizations and arrhythmias using a machine learning model according to the techniques of this disclosure.

Fig. 2 is a block diagram illustrating an example configuration of the Implantable Medical Device (IMD) of fig. 1.

Fig. 3 is a conceptual side view illustrating an example configuration of the IMD of fig. 1 and 2.

FIG. 4 is a functional block diagram illustrating an example configuration of the computing system of FIG. 1.

Fig. 5 is a flow diagram illustrating example operations for providing a visualization of the likelihood of each of a plurality of arrhythmia classifications over time during an episode.

Fig. 6 is a diagram illustrating an example episode review screen including visualizations of the likelihoods of each of a plurality of arrhythmia classifications over time during an episode.

Fig. 7 is a diagram illustrating another example episode review screen including a visualization of the likelihood of each of a plurality of arrhythmia classifications over time during an episode.

Fig. 8 is a flow diagram illustrating example operations for identifying depolarizations during an episode based on arrhythmia classification and depolarization likelihood.

Fig. 9 is a conceptual diagram illustrating an example technique for identifying depolarizations during an episode based on arrhythmia classification and depolarization likelihood.

Like reference numerals refer to like elements throughout the drawings and the description.

Detailed Description

Various types of implantable and external medical devices detect arrhythmic episodes based on sensed cardiac EGMs and, in some cases, other physiological parameters. External devices that may be used for non-invasive sensing and monitoring of cardiac EGMs include wearable devices, such as patches, watches, or necklaces, with electrodes configured to contact the patient's skin. One example of a wearable physiological monitor configured to sense cardiac EGM is SEEQ available from Medtronic pic, of Dublin, Ireland, Dublin, Ireland, Dublin, Ireland, and Dublin, and the like^TMA mobile cardiac telemetry system. Such external devices can facilitate relatively long-term monitoring of a patient during normal daily activities, and can periodically transmit collected data, e.g., episode data for a detected arrhythmia episode, to a remote patient monitoring system, such as a Medtronic Carelink^TMA network.

Implantable Medical Devices (IMDs) also sense and monitor cardiac EGMs and detect arrhythmic episodes. Monitoring cardiac EGM example IMDs include pacemakers and implantable cardioverter defibrillators, which may be coupled to intravascular or extravascular leads, as well as pacemakers having housings configured for implantation within the heart, which may be leadless. Some IMDs that do not provide therapy, such as implantable patient monitors, sense cardiac EGMs. An example of such an IMD is the regenerative LINQ available from Meindon force Inc^TMA heart monitor (ICM) may be inserted, which may be subcutaneously inserted. Such IMDs may facilitate relatively long patient length during normal daily activitiesPhase monitoring and collected data, e.g., episode data of a detected arrhythmia episode, can be periodically transmitted to a remote patient monitoring system, such as a Medtronic Carelink^TMA network.

Such web services may support centralized or clinical-based review, annotation, and reporting of arrhythmia episodes by uploading episode data from the medical devices and distributing the episode data to various users. The episode data may include an indication of one or more arrhythmias detected by the medical device during the episode. The episode data may also include data collected by the medical device during a time period that includes times before and after the medical device determines that one or more arrhythmias have occurred. Episode data may include cardiac EGMs digitized during a time period, heart rates or other parameters derived from EGMs during the time period, and any other physiological parameter data collected by a medical device during the time period.

Fig. 1 is a conceptual diagram illustrating an example of a medical device system 2 configured to detect cardiac depolarizations and arrhythmias with a machine learning model in accordance with the techniques of this disclosure. Example techniques may be used with IMD 10, which IMD 10 may wirelessly communicate with external device 12. In some examples, IMD 10 is implanted outside of the chest of patient 4 (e.g., subcutaneously in the position of the chest illustrated in fig. 1). IMD 10 may be positioned near the sternum near or just below the cardiac level of patient 4, e.g., at least partially within the cardiac contour. IMD 10 includes a plurality of electrodes (not shown in fig. 1) and is configured to sense cardiac EGMs via the plurality of electrodes. In some examples, IMD 10 employs LINQ^TMForm of ICM. Although described primarily in the context of an example in which the medical device collecting episode data takes the form of an ICM, the techniques of this disclosure may be implemented in a system including any one or more implantable or external medical devices, including monitors, pacemakers, or defibrillators.

External device 12 is a computing device configured for wireless communication with IMD 10. External device 12 may be configured to communicate with computing system 24 via network 25. In some instances, external device 12 may provide a user interface and allow a user to interact with IMD 10. Computing system 24 may include a computing device configured to allow a user to interact with IMD 10 or data collected from the IMD via network 25.

External device 12 may be used to retrieve data from IMD 10 and may transmit the data to computing system 24 via network 25. The retrieved data may include values of physiological parameters measured by IMD 10, indications of arrhythmic episodes or other ailments detected by IMD 10, episode data collected for episodes, and other physiological signals recorded by IMD 10. Episode data may include EGM segments recorded by IMD 10, for example, because IMD 10 determines that an episode of arrhythmia or another disease occurred during the segments, or in response to a request for a recorded segment from patient 4 or another user.

In some instances, computing system 24 includes one or more handheld computing devices, computer workstations, servers, or other networked computing devices. In some examples, computing system 24 may include one or more devices, including processing circuitry and storage devices, that implement monitoring system 450. In some examples, computing system 24, network 25, and monitoring system 450 may be implemented via Medtronic Carelink^TMNetwork or other patient monitoring system.

Network 25 may include one or more computing devices (not shown), such as one or more non-edge switches, routers, hubs, gateways, security devices (e.g., firewalls), intrusion detection and/or prevention devices, servers, computer terminals, laptops, printers, databases, wireless mobile devices (e.g., cellular phones or personal digital assistants), wireless access points, bridges, cable modems, application accelerators, or other network devices. Network 25 may include one or more networks managed by a service provider and, thus, may form part of a large-scale public network infrastructure (e.g., the internet). Network 25 may provide computing devices, such as computing system 24 and IMD 10, with access to the internet and may provide a connectivity framework that allows the computing devices to communicate with one another. In some examples, network 25 may be a private network that provides a communication framework that allows computing system 24, IMD 10, and/or external device 12 to communicate with one another but isolates computing system 24, IMD 10, or external device 12 from devices external to network 25 for security purposes. In some examples, communications between computing system 24, IMD 10, and external device 12 are encrypted.

Monitoring system 450, for example, implemented by processing circuitry of computing system 24, may implement techniques of the present disclosure including applying a machine learning model to episode data to detect cardiac depolarizations and arrhythmias. Monitoring system 450 may receive episode data for an episode from a medical device that includes IMD 10, which may store the episode data in response to detection of its arrhythmia and/or user input. Based on the application of one or more arrhythmia classification machine learning models, monitoring system 450 may determine a likelihood that one or more arrhythmias of one or more types occur during an episode, which in some instances includes an arrhythmia identified by the medical device storing the episode data.

Monitoring system 450 may also derive and plot activation data indicating the likelihood of various arrhythmia type classifications over the time period of the episode, and apply one or more depolarization detection machine learning models to the episode data to identify the occurrence of depolarizations during the episode, such as the R-wave or QRS complex. Monitoring system 450 may, for example, display, in combination, one or more of an arrhythmia type classification, an arrhythmia type classification likelihood map, a flag indicating the time of identified depolarizations, and an episode of cardiac EGM, which may facilitate user review and understanding of classification(s) of episode data by monitoring system 450. Although the techniques are described herein as being performed by monitoring system 450, and thus processing circuitry of computing system 24, the techniques may be performed by processing circuitry of any one or more devices or systems of a medical device system, such as computing system 24, external device 12, or IMD 10. The machine learning model may include, as examples, a neural network, a deep learning model, a convolutional neural network, or other types of predictive analysis systems.

Fig. 2 is a block diagram illustrating an example configuration of IMD 10 of fig. 1. As shown in fig. 2, IMD 10 includes processing circuitry 50, sensing circuitry 52, communication circuitry 54, memory 56, sensors 58, switching circuitry 60, and one or more of electrodes 16A, 16B (hereinafter "electrodes 16") may be disposed on a housing of IMD 10. In some examples, memory 56 includes computer readable instructions that, when executed by processing circuitry 50, cause IMD 10 and processing circuitry 50 to perform various functions attributed to IMD 10 and processing circuitry 50 herein. Memory 56 may include any volatile, nonvolatile, magnetic, optical, or electrical media, such as Random Access Memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), Electrically Erasable Programmable ROM (EEPROM), flash memory, or any other digital media.

The processing circuitry 50 may include fixed function circuitry and/or programmable processing circuitry. The processing circuitry 50 may include any one or more of the following: a microprocessor, controller, Digital Signal Processor (DSP), Application Specific Integrated Circuit (ASIC), Field Programmable Gate Array (FPGA), or equivalent discrete or analog logic circuitry. In some examples, processing circuitry 50 may include multiple components (e.g., any combination of one or more microprocessors, one or more controllers, one or more DSPs, one or more ASICs, or one or more FPGAs), as well as other discrete or integrated logic circuitry. The functionality attributed herein to processing circuitry 50 may be embodied as software, firmware, hardware, or any combination thereof.

Sensing circuitry 52 may be selectively coupled to electrodes 16A, 16B via switching circuitry 60 as controlled by processing circuitry 50. Sensing circuitry 52 may monitor signals from electrodes 16A, 16B to monitor electrical activity of the heart of patient 4 of fig. 1 and generate cardiac EGM data for patient 4. In some examples, processing circuitry 50 may identify sensed characteristics of the cardiac EGM to detect an arrhythmic episode of patient 4. Processing circuitry 50 may store characteristics of the digitized cardiac EGM and the EGM used to detect the arrhythmia episode in memory 56 as episode data for the detected arrhythmia episode. In some examples, processing circuitry 50 stores one or more segments of cardiac EGM data, features derived from cardiac EGM data, and other episodes in response to instructions from external device 12 (e.g., when patient 4 experiences one or more symptoms of arrhythmia and inputs a command to external device 12 instructing IMD 10 to upload data for analysis by a monitoring center or clinician).

In some examples, processing circuitry 50 transmits seizure data for patient 4 to an external device, such as external device 12 of fig. 1, via communication circuitry 54. For example, IMD 10 sends digitized cardiac EGM and other episode data to network 25 for processing by monitoring system 450 of fig. 1.

Sensing circuitry 52 and/or processing circuitry 50 may be configured to detect cardiac depolarizations (e.g., P-waves of atrial depolarizations or R-waves of ventricular depolarizations) when the cardiac EGM amplitude crosses a sensing threshold. In some examples, for cardiac depolarization detection, the sensing circuitry 52 may include a rectifier, filter, amplifier, comparator, and/or analog-to-digital converter. In some examples, sensing circuitry 52 may output an indication to processing circuitry 50 in response to the sensing of cardiac depolarization. In this manner, the processing circuitry 50 may receive detected cardiac depolarization indications corresponding to the occurrence of detected R-waves and P-waves in respective chambers of the heart. The processing circuitry 50 may use indications of the detected R-waves and P-waves to determine characteristics of the central EGM including depolarization intervals, heart rate, and detection of arrhythmias, such as tachyarrhythmias and asystole. Sensing circuitry 52 may also provide one or more digitized cardiac EGM signals to processing circuitry 50 for analysis, e.g., for cardiac rhythm identification and/or to identify and characterize cardiac EGM characteristics, such as QRS amplitude and/or width, or other morphological characteristics.

In some examples, IMD 10 includes one or more sensors 58, such as one or more accelerometers, microphones, optical sensors, and/or pressure sensors. In some examples, the sensing circuitry 52 may include one or more filters and amplifiers for filtering and amplifying signals received from one or more of the electrodes 16A, 16B and/or other sensors 58. In some examples, the sensing circuitry 52 and/or the processing circuitry 50 may include rectifiers, filters, and/or amplifiers, sense amplifiers, comparators, and/or analog-to-digital converters. The processing circuitry 50 may determine a value of a physiological parameter of the patient 4 based on the signal from the sensor 58, which may be used to identify an arrhythmia episode and stored as episode data in the memory 56.

Communication circuitry 54 may include any suitable hardware, firmware, software, or any combination thereof for communicating with another device, such as external device 12. Under the control of processing circuitry 50, communication circuitry 54 may receive downlink telemetry from external device 12 or another device, and send uplink telemetry thereto, by way of an internal or external antenna (e.g., antenna 26). In some examples, processing circuitry 50 may be via an external device (e.g., external device 12) and Medtronic as developed by Medtronic corporation of dublin, irelandA computer network of networks communicates with networked computing devices.

Although described herein in the context of example IMD 10, the techniques for arrhythmia detection disclosed herein may be used with other types of devices. For example, the techniques may use an additional cardiac defibrillator coupled to electrodes external to the cardiovascular system, a transcatheter pacemaker configured for implantation within the heart (e.g., Micra, commercially available from medtronic, dublin, ireland^TMTranscatheter pacing system), an insertable cardiac monitor (e.g., a regenerative LINQ)^TMICM, which is also commercially available from mayonney), neurostimulator, drug delivery device, medical device external to patient 4, wearable device, such as wearable cardioverter defibrillator, fitness tracker, or other wearable device, mobile device, such as a mobile phone, "smart" phone, laptop, tablet, Personal Digital Assistant (PDA), or "smart" garment, such as "smart" glasses, "smart" patch, or "smart" watch.

Fig. 3 is a conceptual side view illustrating an example configuration of IMD 10. In the example shown in fig. 3, IMD 10 may include a leadless, subcutaneously implantable monitoring device having a housing 14 and an insulating cover 74. Electrodes 16A and 16B may be formed or placed on the outer surface of cover 74. The circuitry 50-56 and 60 described above with respect to fig. 2 may be formed or placed on an inner surface of the cover 74 or within the housing 14. In the illustrated example, the antenna 26 is formed or placed on an inner surface of the cover 74, but in some examples, may be formed or placed on an outer surface. In some examples, sensor 58 may also be formed or placed on an inner or outer surface of cover 74. In some examples, insulating cover 74 may be positioned over open housing 14 such that housing 14 and cover 74 enclose antenna 26, sensor 58, and circuitry 50-56 and 60, and protect the antenna and circuitry from fluids such as bodily fluids.

One or more of the antenna 26, the sensor 58, or the circuitry 50-56 may be formed on the insulative cover 74, such as by using flip-chip technology. The insulating cover 74 may be flipped over to the housing 14. When inverted and placed onto housing 14, the components of IMD 10 formed on the inside of insulative cover 74 may be positioned in gap 76 defined by housing 14. The electrodes 16 may be electrically connected to the switching circuitry 60 by one or more through-holes (not shown) formed through the insulating cover 74. The insulating cover 74 may be formed from sapphire (i.e., corundum), glass, parylene, and/or any other suitable insulating material. The housing 14 may be formed of titanium or any other suitable material (e.g., a biocompatible material). The electrode 16 may be formed of any of stainless steel, titanium, platinum, iridium, or alloys thereof. Further, the electrode 16 may be coated with a material such as titanium nitride or fractal titanium nitride, although other suitable materials and coatings for such electrodes may also be used.

Fig. 4 is a block diagram illustrating an example configuration of computing system 24. In the illustrated example, the computing system 24 includes processing circuitry 402 for executing an application 424 that includes a monitoring system 450 or any other application described herein. The computing system 24 may be any component or system that includes processing circuitry or other suitable computing environment for executing software instructions, and for example, need not include one or more of the elements shown in fig. 4 (e.g., the input device 404, the communication circuitry 406, the user interface device 410, or the output device 412; and in some instances, components such as the storage device(s) 408 may be co-located with other components or in the same rack). In some examples, computing system 24 may be a cloud computing system distributed across multiple devices.

In the example of fig. 4, computing system 24 includes processing circuitry 402, one or more input devices 404, communication circuitry 406, one or more storage devices 408, User Interface (UI) device(s) 410, and one or more output devices 412. In some examples, the computing system 24 also includes one or more applications 424 that may be executed by the computing system 24, such as a monitoring system 450, and an operating system 416. Each of the components 402, 404, 406, 408, 410, and 412 may be coupled (physically, communicatively, and/or operatively) for inter-component communication. In some examples, the communication channel 414 may include a system bus, a network connection, an interprocess communication data structure, or any other method for communicating data. As one example, components 402, 404, 406, 408, 410, and 412 may be coupled by one or more communication channels 414.

In one example, the processing circuitry 402 is configured to implement functions and/or processing instructions for execution within the computing system 24. For example, the processing circuitry 402 may be capable of processing instructions stored in the storage 408. Examples of processing circuitry 402 may include any one or more of the following: a microprocessor, a controller, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or equivalent discrete or integrated logic circuitry.

One or more storage devices 408 may be configured to store information within computing device 400 during operation. In some examples, storage device 408 is described as a computer-readable storage medium. In some instances, storage device 408 is a temporary memory, meaning that the primary purpose of storage device 408 is not long-term storage. In some instances, storage 408 is described as volatile memory, meaning that storage 408 does not maintain stored content when the computer is turned off. Examples of volatile memory include Random Access Memory (RAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), and other forms of volatile memory known in the art. In some examples, the storage device 408 is used to store program instructions that are executed by the processing circuitry 402. In one example, storage 408 is used by software or applications 424 running on computing system 24 to temporarily store information during program execution.

In some examples, storage 408 also includes one or more computer-readable storage media. Storage device 408 can be configured to store larger amounts of information than volatile memory. Storage device 408 may additionally be configured for long-term storage of information. In some examples, storage device 408 includes non-volatile storage elements. Examples of such non-volatile storage elements include magnetic hard disks, optical disks, floppy disks, flash memory, or various forms of electrically programmable memory (EPROM) or electrically erasable and programmable memory (EEPROM).

In some examples, computing system 24 also includes communication circuitry 406 to communicate with other devices and systems, such as IMD 10 and external device 12 of fig. 1. The communication circuitry 406 may include a network interface card, such as an ethernet card, an optical transceiver, a radio frequency transceiver, or any other type of device that may send and receive information. Other examples of such network interfaces may include 3G and WiFi radios.

In one example, the computing system 24 also includes one or more user interface devices 410. In some examples, the user interface device 410 is configured to receive input from a user through tactile, audio, or video feedback. Examples of user interface device(s) 410 include a presence-sensitive display, a mouse, a keyboard, a voice response system, a camera, a microphone, or any other type of device for detecting commands from a user. In some examples, the presence-sensitive display includes a touch-sensitive screen.

One or more output devices 412 may also be included in the computing system 24. In some examples, output device 412 is configured to provide output to a user using tactile, audio, or video stimuli. In one example, output device 412 includes a presence-sensitive display, a sound card, a video graphics adapter card, or any other type of device for converting signals into an appropriate form understandable to humans or machines. Additional examples of output device 412 include a speaker, a Cathode Ray Tube (CRT) monitor, a Liquid Crystal Display (LCD), or any other type of device that can generate an understandable output to a user.

The computing system 24 may include an operating system 416. In some examples, operating system 416 controls the operation of components of computing system 24. For example, in one example, the operating system 416 facilitates communication of one or more applications 424 and the monitoring system 450 with the processing circuitry 402, the communication circuitry 406, the storage 408, the input device 404, the user interface device 410, and the output device 412.

Applications 424 may also include program instructions and/or data that may be executed by computing device 400. Example application(s) 424 executable by computing device 400 may include monitoring system 450. Other additional applications not shown may alternatively or additionally be included to provide other functionality described herein and are not depicted for simplicity.

In accordance with the techniques of this disclosure, computing system 24 receives episode data for an episode stored by a medical device, such as IMD 10, via communication circuitry 406. Storage device 408 may store seizure data for the seizure in storage device 408. Episode data may have been collected by the medical device in response to the medical device detecting an arrhythmia and/or directing user input that stores episode data.

The monitoring system 450, as implemented by the processing circuitry 402, may review and annotate the episode and, after annotation, generate a report or other presentation of the episode for review by a clinician or other reviewer. Monitoring system 450 may utilize input device 404, output device 412, and/or communication circuitry 406 to display episode data, arrhythmia type classification, mapping, identified depolarizations, and any other information described herein to the user, and receive any annotations or other input regarding the episode data from the user.

To review and annotate episodes, monitoring system 450 may apply episode data as input to the selected one or more machine learning models. In the example illustrated by fig. 4, monitoring system 450 may apply episode data to one or more arrhythmia classification machine learning models 452 and/or one or more depolarization detection models 454. As an example, the machine learning models 452 and 454 may include a neural network, such as a deep neural network, which may include a convolutional neural network, a multi-layer perceptron, and/or an echo state network, as examples.

The arrhythmia classification machine learning model 452 may be configured to output a value for each of a plurality of arrhythmia type classifications that indicates a likelihood that an arrhythmia of the type occurs at any point during the episode. Monitoring system 450 may apply a configurable threshold (e.g., 50%, 75%, 90%, 95%, 99%) to the likelihood value to annotate the episode as including one or more arrhythmia types, e.g., based on the likelihood of the classification meeting or exceeding the threshold.

In some examples, arrhythmia classification machine learning model 452 is trained with training data that includes cardiac EGMs or other episode data for a plurality of patients labeled with descriptive metadata. For example, during the training phase, monitoring system 450 processes multiple cardiac EGM waveforms. Typically, the multiple cardiac EGM waveforms are from multiple different patients, but may be from a single patient. Each cardiac EGM waveform is labeled with one or more episodes of one or more types of arrhythmia.

For example, a training cardiac EGM waveform may include a plurality of segments, each labeled with a descriptor specifying the absence of an arrhythmia or the presence of a particular classification of arrhythmia (e.g., bradycardia, pause, tachycardia, atrial fibrillation, atrioventricular conduction block, or ventricular fibrillation). In some examples, the clinician manually marks the presence of an arrhythmia in each cardiac EGM waveform. In some examples, the presence of arrhythmias in each cardiac EGM waveform is based on markers classified by a cardiac EGM feature delineation algorithm, e.g., similar to techniques used by IMD 10 to identify arrhythmias based on rate, interval, and state features derived from the cardiac EGM.

Monitoring system 450 is operable to convert training data into vectors and multidimensional arrays to which monitoring system 450 may apply mathematical operations, such as linear algebra, non-linear, or alternative computational operations. The monitoring system 450 uses the training data to teach one or more arrhythmia classification machine learning models 452 to weight different features depicted in the cardiac EGM data. In some examples, monitoring system 450 uses the cardiac EGM data to teach a machine learning model to apply different coefficients that indicate that one or more features in the cardiac EGM are of greater or lesser importance for the occurrence of a particular classification of arrhythmia. By processing a plurality of such waveforms labeled with episodes of arrhythmia, monitoring system 450 may build and train one or more arrhythmia classification machine learning models 452 to receive cardiac EGM data from a patient, such as patient 4 of fig. 1, that was not previously analyzed by monitoring system 450, and process such cardiac EGM data to detect the presence or absence of different classifications of arrhythmia types in the patient with high accuracy. In general, the greater the amount of cardiac EGM data from which one or more arrhythmia classification machine learning models 452 are trained, the greater the accuracy of the machine learning models in detecting or classifying arrhythmias in new cardiac EGM data.

After monitoring system 450 has trained one or more arrhythmia classification machine learning models 452, monitoring system 450 may receive episode data, such as Electrocardiography (EGM) data, for a particular patient, such as patient 4. Monitoring system 450 applies one or more trained arrhythmia classification machine learning models 452 to the episode data to determine whether one or more arrhythmia types occurred at any point during the episode.

In some examples, monitoring system 450 may process one or more characteristics of the cardiac EGM data instead of or in addition to the raw cardiac EGM data itself. One or more features may be obtained via feature delineation performed by IMD 10 and/or monitoring system 450. These features may include, for example, one or more of heart rate, depolarization intervals, other intervals between features of the cardiac EGM, one or more amplitude, width, or morphology features of QRS waves or other features of the cardiac EGM, variability of any of these features, T-wave alternans, or other types of cardiac features not explicitly described herein. In such example embodiments, instead of or in addition to tagging episodes of arrhythmia with a plurality of cardiac EGM waveforms as described above, monitoring system 450 may train one or more arrhythmia classification machine learning models 452 via a plurality of training cardiac features tagged with episodes of arrhythmia.

In further examples, monitoring system 450 may generate an intermediate representation of cardiac EGM data from cardiac EGM data. For example, the monitoring system 450 may apply one or more of signal processing, downsampling, normalization, signal decomposition, wavelet decomposition, filtering, noise reduction, or neural network-based feature representation operations to the electrocardiograph data to generate an intermediate representation of the electrocardiograph data. Monitoring system 450 may process such intermediate representations of the cardiac EGM data to detect and classify various types of arrhythmias for patient 4. Further, instead of a plurality of raw cardiac EGM waveforms labeled with episodes of arrhythmia as described above, monitoring system 450 may train one or more arrhythmia classification machine learning models 452 via a plurality of training intermediate representations labeled with episodes of arrhythmia. Using such intermediate representations of cardiac EGM data may allow for training and development of a low-weight, less computationally complex arrhythmia classification machine learning model 452 by monitoring system 450. In addition, the use of such intermediate representations of electrocardiographic data may require fewer iterations and less training data to build an accurate machine learning model than training the machine learning model using the original cardiac EGM data.

In some examples, based on applying the episode data to one or more arrhythmia classification machine learning models 452, for each of the arrhythmia type classifications, the monitoring system 450 may derive classification activation data that indicates a likelihood of a change in the classification over a period of time of the episode waveform. For a given arrhythmia type, the magnitude of such likelihood values at different times corresponds to the probability that the arrhythmia occurred at that time, with higher values corresponding to higher probabilities.

The classification activation map may identify regions of the input time series, e.g., regions of cardiac EGM data, that constitute the reasons for the time series given a particular classification by one or more arrhythmia classification machine learning models 452. The classification activation map for a given classification may be a univariate time series, where each element (e.g., each timestamp at the sampling frequency of the input time series) may be a weight and or other value derived from the output of an intermediate layer of a neural network or other machine learning model. For each classification, the middle layer may be the last layer before the global average pooling layer and/or the output layer neurons.

Monitoring system 450 may display and/or communicate via communication circuitry 406 a graph of activation data over a period of an episode with another device, e.g., via output device(s) 412. In some instances, monitoring system 450 may display the classification activation data, for example, on the same screen and simultaneously in conjunction with the input cardiac EGM. While one or more arrhythmia classification machine learning models 452 may be configured to provide outputs indicating the likelihood that different arrhythmia type classifications occur during an entire episode, classification activation data may allow monitoring system 450 and/or a user to identify times during an episode and points during cardiac EGMs at which one or more types of one or more arrhythmias are likely to occur.

Post-processing of episode data stored by IMD 10 may include identifying the occurrence of depolarizations within cardiac EGM data, such as R-waves and/or QRS complexes. The identification of depolarization during post-processing may be different than IMD 10 during detection of the episode and storage of the episode data, providing evidence that IMD 10 may misclassify the episode. Annotating the cardiac EGM data with an indication of the occurrence of an identified depolarization may also facilitate review of the episode data by a user and/or monitoring system 450. Feature delineation techniques to detect depolarization may include filtering cardiac EGM data, feature extraction (e.g., using a rectified power signal), peak detection, and refractory analysis or other further processing. Such feature delimitation may require feature engineering and detection rule development.

According to example techniques of this disclosure, monitoring system 450 may apply one or more of the functions of seizure data, such as cardiac EGM dataA plurality of depolarization detection machine learning models 454. The depolarization detection machine learning model 454 may be configured to output a set of depolarization likelihood values, each of the set of depolarization likelihood values representing a likelihood that a depolarization occurred at a respective time during the time period. The set of depolarization possibilities may be expressed as probabilities over time, e.g. p (QRS) in case of detection of the QRS complex_(t))。

Monitoring system 450 may use a training set of cardiac EGM data to train one or more depolarization detection machine learning models 454 in a similar manner as described above with respect to training arrhythmia classification machine learning model 452. The training set of cardiac EGM data is annotated, e.g., by a user and/or a feature-delimiting algorithm, to indicate which portions of cardiac EGM data are depolarized. When presented with new cardiac EGM data, such as episode data stored by IMD 10 from patient 4, monitoring system 450 may identify one or more depolarizations in the cardiac EGM based on a likelihood of depolarization value output by one or more depolarization detection machine learning models 454. For example, monitoring system 450 may compare the likelihood of depolarization value to a configurable and/or adaptive threshold and identify depolarizations when the likelihood of depolarization value meets or exceeds the threshold. In some examples, monitoring system 450 may further post-process the depolarization occurrence times identified based on the likelihood threshold. For example, monitoring system 450 may align the identified time with an R-wave peak identified in the cardiac EGM waveform. As another example, monitoring system 450 may apply a treatment to the identified depolarization time that should not be treated, such as to eliminate a time not greater than a threshold time interval from another candidate time.

In some cases, the monitoring system 450 and one or more depolarization detection machine learning models 454 may have difficulty detecting a particular depolarization in the cardiac EGM data with a high likelihood, for example, due to a smaller training data set with depolarization annotations and/or the occurrence of arrhythmias, such as tachycardia, pause, and atrial fibrillation, which may alter the morphology and rate of depolarization. To increase the ability of monitoring system 450 to identify depolarizations, monitoring system 450 may be based on timeA depolarization likelihood value and an arrhythmia type likelihood value, such as activation data, to identify depolarizations. For example, an arrhythmia type likelihood value indicates that the probability of atrial tachycardia is relatively high at a time in the cardiac EGM data at which the monitoring system 450 may take one or more actions that result in an effective increase in the depolarization likelihood value or a decrease in the depolarization detection threshold compared to the depolarization likelihood value. Such actions may result in an effective modulation of the likelihood of depolarization over the likelihood of an arrhythmia or certain types of arrhythmias. For example, if the arrhythmia classification machine learning model 452 provides a probability p1 of an arrhythmia type classification X at intervals during an episode, such as one second (X)_(t)) And depolarization detection machine learning model 454 provides probability p2 (QRS)_(t)) Then when considering the arrhythmia likelihood, the probability of QRS can be expressed as p3 (QRS)_(t)|p1&p2)。

Fig. 5 is a flow diagram illustrating example operations for providing a visualization of the likelihood of each of a plurality of arrhythmia classifications over time during an episode. For clarity of description, the operations and techniques illustrated in fig. 5-9 are described as being performed by the monitoring system 450, and thus by the processing circuitry 402 of the computing system 24. However, these operations and techniques, as well as any other operations and techniques described herein, may be performed by processing circuitry of any one or more devices of medical device system 2, such as IMD 10, external device 12, and computing device 24.

According to the example illustrated by fig. 5, monitoring system 450 receives episode data for an episode recorded by IMD 10 (500). Episodes are associated with a time period, and episode data includes cardiac EGMs sensed by IMD 10 during the time period. Episode data may include other signals sensed by IMD 10 during a time period, or data derived from cardiac EGM or other signals.

Monitoring system 450 applies one or more arrhythmia classification machine learning models 452(502) to episode data, such as cardiac EGM data. The one or more arrhythmia classification machine learning models 452 output a respective likelihood value for each of a plurality of arrhythmia type classifications, each of the likelihood values representing a likelihood that the respective arrhythmia type classification occurs at any point during the time period. Based on applying one or more arrhythmia classification machine learning models 452 to the episode data, for each arrhythmia type classification, monitoring system 450 also derives classification activation data indicating a likelihood that the classification varies over a period of time (506). As described herein, the monitoring system 450 may derive activation data from the output of an intermediate layer of the machine learning model 452, such as a final layer and/or a global average pooling layer preceding the classification layer.

Monitoring system 450 plots activation data for various arrhythmia type classifications over time (506), and displays a graph of the activation data plot to the user (508). In some examples, monitoring system 450 displays a plot in conjunction with an episode of cardiac EGM, which may allow a user to associate times of relatively high likelihood of a particular arrhythmia type with portions of cardiac EGM that cause relatively high likelihood. In some examples, monitoring system 450 may indicate, for example, a likelihood of annotating or highlighting other times of the plot of at least one of the temporal arrhythmia type classifications relative to at least one arrhythmia type classification. In some examples, monitoring system 450 may further indicate a relatively high likelihood time of arrhythmia type classification on a plot of activation data when monitoring system 450 indicates that a particular type of arrhythmia occurred at some unspecified point during the episode based on the output of one or more arrhythmia detection machine learning models 452, such that a user may understand the reasoning of classifying by one or more arrhythmia detection machine learning models 452, for example, by reference to corresponding portions of cardiac EGMs.

Fig. 6 and 7 are diagrams illustrating example episode review screens 600, 700 that include visualizations of the likelihoods of each of a plurality of arrhythmia classifications over time during an episode. Each of screens 600 and 700 displays a respective cardiac EGM 602, 702 in conjunction with a graph 604, 704 of arrhythmia type classification activation data. Each of the graphs 604 and 704 includes a set of respective plots 606, 706 and respective keywords 608, 708. Each of plots 606 and 706 is a plot of likelihood values over time for a respective arrhythmia type classification. Keywords 608 and 708 identify the arrhythmia type classification for each of the plots.

The graph 604 also includes an annotation 610 identifying a region of the graph 604 in which a plot 606 associated with one of the categories of arrhythmia types, e.g., bradycardia, is associated with a relatively high likelihood value of the associated category of onset as having at least one arrhythmia of the type. Similarly, graph 704 includes annotations 710, 712, and 714, each of which indicates an area of graph 704 in which a respective plot 606 associated with one of the categories of arrhythmia types, e.g., PVC, atrial fibrillation, and sinus tachycardia, has a relatively high likelihood value associated with the associated category of onset as having at least one arrhythmia of the type. By displaying cardiac EGMs 602, 702, graphs 604, 704, and annotations 610, 710, 712, and 714 in conjunction on screens 600, 700, monitoring system 450 may allow a user to more easily review, understand, and confirm (or reject) arrhythmia detection and classification of episodes by monitoring system 450.

By displaying the cardiac EGMs 602, 702, graphs 604, 704, and annotations 610, 710, 712, and 714 in conjunction on the screens 600, 700, the monitoring system 450 may allow a user to quickly identify the portion of the cardiac EGM of interest when reviewing an episode to identify/confirm the presence or absence of an arrhythmia. In some examples, monitoring system 450 may also use the activation data to automatically identify segments of episode data from an episode, such as segments of cardiac EGMs, for inclusion in a report reviewed by the user. In some cases, the duration of the episode is long (e.g., 1 minute) and contains multiple arrhythmias at different times during the episode. While reports generated by such episodes may include a presentation of the entire cardiac EGM stored for the episode, such reports may additionally or alternatively include representative segments, where each segment may include multiple consecutive seconds of cardiac EGM data.

Monitoring system 450 may identify one or more times (in episode duration) at which activation data for a particular arrhythmia type is highest and/or exceeds a threshold. Monitoring system 450 may select a segment of cardiac EGM for presentation to the user as representative of the cause of detecting the type of arrhythmia based on the identified time. For example, referring to fig. 7, the highest likelihood value of the PVC classification activation map occurs at a spike of about 18 seconds after the onset of the episode. For example, if a representative PVC segment of this episode needs to be reported, the monitoring system 450 may present a six second segment of 15-21 seconds of cardiac EGMs.

Fig. 8 is a flow diagram illustrating example operations for identifying depolarizations during an episode based on arrhythmia classification and depolarization likelihood. According to the example illustrated by fig. 8, monitoring system 450 receives episode data for an episode recorded by IMD 10 (800). Episodes are associated with a time period, and episode data includes cardiac EGMs sensed by IMD 10 during the time period. Episode data may include other signals sensed by IMD 10 during a time period, or data derived from cardiac EGM or other signals.

Monitoring system 450 applies one or more arrhythmia classification machine learning models 452(802) to episode data, such as cardiac EGM data. For each arrhythmia type classification, monitoring system 450 may also derive classification activation data that indicates a likelihood of a change in the classification over a period of time. Monitoring system 450 also applies one or more depolarization detection machine learning models 454 to the episode data, the one or more depolarization detection machine learning models 454 configured to output a set of depolarization likelihood values, each of the set of depolarization likelihood values representing a likelihood of depolarization occurring at a respective time during the time period (804). Monitoring system 450 identifies one or more depolarizations during the episode based on the arrhythmia type likelihood value and the depolarization likelihood value (806).

Monitoring system 450 may use the arrhythmia type likelihood value to identify depolarizations in a variety of ways. For example, the monitoring system may apply a time series arrhythmia type likelihood value, e.g., activation data, as illustrated by activation data plots 606 and 706 illustrated by fig. 7, along with episode data, e.g., cardiac EGM data, as input to one or more depolarization detection machine learning models 454. One or more depolarization detection machine learning models 454 may be trained to determine the likelihood of depolarization at a given time of onset based on cardiac EGM inputs and likelihood values classified by one or more arrhythmia types. In some examples, the likelihood values for certain arrhythmia type classifications are selected as inputs to the one or more depolarization detection machine learning models 454, rather than all arrhythmia type classifications for which the one or more arrhythmia detection machine learning models 452 output likelihood values.

In some examples, monitoring system 450 may apply heuristic rules to one or more arrhythmia type classification likelihood values, such as the selected arrhythmia type classification. Based on the rules and the arrhythmia type classification likelihood values at certain times during the episode, the monitoring system 450 may modify, e.g., increase or decrease, the depolarization likelihood values provided by one or more depolarization detection models 454 at certain times during the episode. Additionally or alternatively, monitoring system 450 may similarly modify a depolarization likelihood threshold based on rules and arrhythmia type classification likelihood values at certain times during the episode, to which the monitoring system compares the depolarization likelihood values at certain times during the episode. In either case, modification of the depolarization likelihood value and/or the depolarization likelihood threshold may result in the detection of depolarizations by monitoring system 450 that are otherwise undetected. The heuristic rules may be configured to modify the depolarization likelihood value and/or the depolarization likelihood threshold at a time after or at a particular time of a relatively high likelihood of a certain type of arrhythmia based on an understanding of the timing of depolarizations during the type of arrhythmia. For example, the heuristic rules may be configured to modify a likelihood of depolarization value and/or a likelihood of depolarization threshold at a particular time after a time of relatively high likelihood of a tachyarrhythmia based on a depolarization rate or polarization interval of a typical tachyarrhythmia.

Fig. 9 is a conceptual diagram illustrating an example technique for identifying depolarizations during an episode based on arrhythmia classification and depolarization likelihood. Specifically, FIG. 9 illustrates an electrocardiography 900 that includes depolarizations 902A-902C (collectively "depolarizations 902"). Fig. 9 also illustrates a depolarization likelihood value plot 904. Depolarization likelihood value plot 904 includes regions 906A-906C (collectively, "regions 906") having relatively high depolarization likelihood values and corresponding in time to depolarizations 902. A likelihood of depolarization threshold 908 is also illustrated. The depolarization likelihood value of plot 904 at region 906B does not exceed depolarization likelihood threshold 908. Thus, monitoring system 450 will not detect depolarization 902B.

Fig. 9 also illustrates a plot of arrhythmia classification likelihood values 910 for tachyarrhythmias relative to a baseline 912 and a threshold 914. The threshold 914 may be established by heuristic rules applied by the monitoring system 450. As illustrated in fig. 9, the arrhythmia classification likelihood value 910 exceeds the threshold 914 at time 916.

Figure 9 also illustrates a depolarization likelihood value plot 920 that includes regions 922A-922C (collectively, "regions 922") having relatively high likelihood of depolarization values and corresponding in time to depolarization 902. Regions 922A and 922C are the same as regions 906A and 906C, i.e., have not been modified by monitoring system 450 based on the arrhythmia classification likelihood value 910 heuristic rules. However, based on arrhythmia classification likelihood value 910 and heuristic rules, monitoring system 450 has increased region 922B relative to region 906B based on arrhythmia classification likelihood value 910 having exceeded threshold 908. Thus, monitoring system 450 may detect depolarization 902B.

In addition to using depolarization and arrhythmia likelihood data to identify depolarizations within an episode, such as the techniques described with respect to fig. 8 and 9, using the techniques herein, monitoring system 450 may use the depolarization and arrhythmia likelihood data to determine the type of depolarization that depolarizes within the episode. As an example, the types of depolarization may include normal, ventricular premature beats, atrial premature beats, and artifact/noise. One or more arrhythmia classification machine learning models 452 used by the monitoring system to classify episodes may be configured to classify episodes as ventricular premature beats, atrial premature beats, and artifact/noise, among other classifications. Thus, the monitoring system 450 may be configured to derive categorical activation values over time for ventricular premature beats, atrial premature beats, and artifacts/noise. The monitoring system 450 may also use the classified activation data for those arrhythmia types to determine which depolarizations are of multiple types, such as normal, ventricular premature beats, atrial premature beats, and artifact/noise.

In some examples, monitoring system 450 may label depolarizations that occur when the likelihood value of the noise/artifact classification reaches (e.g., exceeds) a threshold as artifact/noise. Monitoring system 450 may consider depolarization intervals of remaining (e.g., non-artifact/noise) depolarizations during the episode and determine whether any of these remaining depolarizations are "premature" based on their depolarization intervals (e.g., R-R intervals) reaching (e.g., being less than) a threshold. The depolarization interval threshold may be determined based on an average or median of the depolarization intervals during the episode, e.g., the threshold may be a fraction or percentage of the average or median of the depolarization intervals during the episode. Monitoring system 450 may flag premature depolarization as ventricular premature or atrial premature based on a likelihood value corresponding in time to the premature depolarization reaching (e.g., exceeding) a threshold. The monitoring system 450 may mark depolarizations identified during an episode but not determine one of ventricular premature beats, atrial premature beats, or artifact/noise as normal.

In some examples, the techniques of this disclosure include a system comprising an apparatus to perform any of the methods described herein. In some examples, the techniques of this disclosure include a computer-readable medium comprising instructions that cause processing circuitry to perform any of the methods described herein.

It should be understood that the various aspects disclosed herein may be combined in different combinations than those specifically presented in the specification and drawings. It will also be understood that, depending on the example, certain acts or events of any of the processes or methods described herein may be performed in a different order, may be added, merged, or omitted altogether (e.g., all described acts or events may not be necessary for performing the techniques). Additionally, although certain aspects of the disclosure are described as being performed by a single module, unit, or circuit for clarity, it should be understood that the techniques of this disclosure may be performed by a combination of units, modules, or circuits associated with, for example, a medical device.

In one or more examples, the described techniques may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions may be stored on a computer-readable medium in the form of one or more instructions or code and may be executed by a hardware-based processing unit. The computer-readable medium may include a non-transitory computer-readable medium corresponding to a tangible medium such as a data storage medium (e.g., RAM, ROM, EEPROM, flash memory, or any other medium that can be used to store desired program code in the form of instructions or data structures and that can be accessed by a computer).

The instructions may be executed by one or more processors, such as one or more Digital Signal Processors (DSPs), general purpose microprocessors, Application Specific Integrated Circuits (ASICs), field programmable logic arrays (FPGAs), or other equivalent integrated or discrete logic circuitry. Thus, the term "processor" or "processing circuitry" as used herein may refer to any of the foregoing structures or any other physical structure suitable for implementing the described techniques. Furthermore, the techniques may be fully implemented in one or more circuits or logic elements.

The following examples illustrate the techniques described herein.

Example 1: a computer-implemented method, comprising: receiving, by processing circuitry of a medical device system, episode data for an episode stored by a medical device of a patient, wherein the episode is associated with a time period and the episode data contains an electrocardiograph sensed by the medical device during the time period; applying, by processing circuitry, one or more machine learning models to the episode data, the one or more machine learning models configured to output a respective likelihood value for each of a plurality of arrhythmia type classifications, each of the likelihood values representing a likelihood that the respective arrhythmia type classification occurred at any point during the time period; deriving, by the processing circuitry and for each of the arrhythmia type classifications, classification activation data indicative of a likelihood of a change in the classification over the period of time based on applying one or more machine learning models to the episode data; and displaying, by the processing circuitry and to a user, a graph of the likelihood of the change in the arrhythmia type classification over the time period.

Example 2: the method of example 1, wherein displaying a graph comprises displaying a graph in conjunction with an electrocardiograph.

Example 3: the method of example 1 or 2, further comprising indicating, by the processing circuitry, on the graph, times of at least one higher likelihood of at least one arrhythmia type classification relative to other times of the graph of at least one of the arrhythmia type classifications.

Example 4: the method of example 3, further comprising indicating, by the processing circuitry and based on the output of the one or more machine learning models, that the at least one arrhythmia type classification occurred at any point during the time period, wherein indicating on the graph a time of at least one higher likelihood of the at least one arrhythmia type classification comprises indicating on the graph a time of at least one higher likelihood of the at least one arrhythmia type classification in response to indicating that the at least one arrhythmia type classification occurred at any point during the time period.

Example 5: the method of any of examples 1-4, wherein the plurality of arrhythmia type classifications includes a plurality of bradycardias, pauses, ventricular tachycardias, ventricular fibrillation, supraventricular tachycardias, atrial fibrillation, atrial flutter, sinus tachycardias, ventricular premature beats, atrial premature beats, broad group tachycardias, and atrioventricular conduction blocks.

Example 6: the method of any of examples 1-5, wherein each of the one or more machine learning models comprises a plurality of layers, and wherein deriving activation data comprises deriving an activation layer from an output of an intermediate layer of the plurality of layers.

Example 7: the method of example 6, wherein the middle layer comprises a global average pooling layer.

Example 8: the method of any one of examples 1-7, further comprising: selecting segments from the electrocardiography based on the classification activation data; and displaying the selected segments of the electrocardiography.

Example 9: the method of example 8, wherein selecting a fragment comprises: identifying a time during the episode based on the classified activation data for one arrhythmia type classification; and selecting a segment of the electrocardiography map that includes the identified time.

Example 10: a computer-implemented method, comprising: receiving, by processing circuitry of a medical device system, episode data for an episode stored by a medical device of a patient, wherein the episode is associated with a time period and the episode data contains an electrocardiography sensed by the medical device during the time period; applying, by processing circuitry, one or more arrhythmia classification machine learning models to the episode data, the one or more arrhythmia classification machine learning models configured to output a set of respective arrhythmia type likelihood values for each of a plurality of arrhythmia type classifications, each of the set of arrhythmia type likelihood values representing a likelihood that the respective arrhythmia type classification occurred at a respective time during the time period; applying, by processing circuitry, one or more depolarization detection machine learning models to the episode data, the one or more depolarization detection machine learning models configured to output a set of depolarization likelihood values, each of the set of depolarization likelihood values representing a likelihood of depolarization occurring at a respective time during the time period; and identifying, by the processing circuitry, one or more depolarizations during the episode based on the arrhythmia type likelihood value and the depolarization likelihood value.

Example 11: the method of example 10, wherein each of the one or more arrhythmia classification machine learning models comprises a plurality of layers, the method further comprising deriving, by the processing circuitry, a set of arrhythmia type likelihood values from outputs of intermediate layers of the plurality of layers.

Example 12: the method of example 11, wherein the middle layer comprises a global average pooling layer.

Example 13: the method of any of examples 10-12, wherein applying one or more depolarization detection machine learning models to the episode data and identifying one or more depolarizations based on the arrhythmia type likelihood value and the depolarization likelihood value includes applying one or more depolarization detection machine learning models to the episode data and the arrhythmia type likelihood value.

Example 14: the method of any of examples 10-13, wherein identifying one or more depolarizations based on the arrhythmia type likelihood value and the depolarization likelihood value comprises: modifying one or more of the depolarization likelihood values based on one or more of the arrhythmia type likelihood values; and identifying one or more depolarizations based on the one or more modified depolarization likelihood values.

Example 15: the method of any of examples 10-14, wherein identifying one or more depolarizations based on an arrhythmia type likelihood value and a depolarization likelihood value comprises: modifying a depolarization likelihood threshold based on one or more of the arrhythmia type likelihood values; comparing the depolarization likelihood value to a modified depolarization likelihood threshold; and identifying one or more depolarizations based on the comparison.

Example 16: the method of any of examples 10-15, wherein depolarization comprises at least one of an R-wave or a QRS complex.

Example 17: the method of any of examples 10-16, further comprising labeling each of the one or more identified depolarizations as one of a plurality of depolarization types based on an arrhythmia type likelihood value.

Example 18: the method of example 17, wherein the plurality of depolarization types includes a plurality of normal, ventricular premature beats, atrial premature beats, noise, or artifacts.

Example 19: a medical device system, comprising a medical device configured to: sensing an electrocardiography of a patient via a plurality of electrodes; and storing episode data for the episode, wherein the episode is associated with a time period and the episode data comprises an electrocardiography sensed by the medical device during the time period. The medical device system further includes processing circuitry configured to perform the method of any of examples 1-18.

Example 20: the medical device system of example 19, wherein the processing circuitry includes processing circuitry of a computing device.

Example 21: the medical device system of examples 19 or 20, wherein the medical device is implantable.

Example 22: a non-transitory computer-readable medium containing instructions that, when executed by processing circuitry of a computing system, cause the computing system to perform the method of any of examples 1-18.

Various examples have been described. These and other examples are within the scope of the following claims.