Data preparation for AI-based arrhythmia detection

Abstract

Techniques for preparing data for artificial intelligence (AI)-based arrhythmia detection are disclosed. According to the disclosed techniques, a computing system may obtain cardiac electrogram (EGM) strips representing waveforms of a cardiac rhythm of the same patient. Additionally, the computing system may pre-process the cardiac EGM strips. The computing system may then apply a deep learning model to the pre-processed cardiac EGM strips to generate arrhythmia data indicating whether the cardiac EGM strips represent one or more occurrences of one or more arrhythmias.

Description

Technical Field

[0001] The present disclosure relates generally to health monitoring, and more particularly, to monitoring cardiac health. Background Art

[0002] Malignant tachyarrhythmias, such as ventricular fibrillation, are uncoordinated contractions of the myocardium of the ventricles in the heart and are the most commonly identified arrhythmias in cardiac arrest patients. If such an arrhythmia persists for more than a few seconds, it may lead to cardiogenic shock and cessation of effective blood circulation. As a result, sudden cardiac death (SCD) can occur within minutes.

[0003] An implantable or non-implantable medical device can monitor a patient's heart for arrhythmias. A user, such as a physician, can view data generated by the medical device regarding arrhythmias. The user can diagnose the patient's medical condition based on the arrhythmias. Summary of the invention

[0004] In general, the present disclosure describes techniques for preparing data for artificial intelligence (AI)-based arrhythmia detection. As described herein, a computing system can obtain a cardiac electrogram (EGM) strip representing a waveform of a patient's cardiac rhythm. Additionally, the computing system can preprocess the cardiac EGM strip. The computing system can then apply a deep learning model to the preprocessed cardiac EGM strip to generate arrhythmia data indicating whether the cardiac EGM strip represents one or more occurrences of one or more arrhythmias.

[0005] On the one hand, the present disclosure describes a method comprising: obtaining, by a computing system, one or more cardiac electrogram (EGM) strips representing a waveform of a patient's cardiac rhythm; preprocessing, by the computing system, the one or more cardiac EGM strips; and applying, by the computing system, a deep learning model to the one or more preprocessed cardiac EGM strips to generate arrhythmia data indicating whether the one or more cardiac EGM strips represent one or more occurrences of one or more arrhythmias.

[0006] On the other hand, the present disclosure describes a computing system comprising: a storage device configured to store one or more cardiac electrogram (EGM) strips representing a waveform of a patient's cardiac rhythm; one or more processing circuits configured to: pre-process the one or more cardiac EGM strips; and apply a deep learning model to the one or more pre-processed cardiac EGM strips to generate arrhythmia data indicating whether the one or more cardiac EGM strips represent one or more occurrences of one or more arrhythmias.

[0007] On the other hand, the present disclosure describes a computer-readable storage medium having instructions stored thereon that, when executed, cause a computing system to obtain one or more cardiac electrogram (EGM) strips representing a waveform of a patient's cardiac rhythm; preprocess the one or more cardiac EGM strips; and apply a deep learning model to the one or more preprocessed cardiac EGM strips to generate arrhythmia data indicating whether the one or more cardiac EGM strips represent one or more occurrences of one or more arrhythmias.

[0008] 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 devices and methods described in detail in the following figures and description. Additional details of one or more examples are set forth in the following figures and description. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 is a block diagram illustrating a system for analyzing one or more aspects of a patient's cardiac rhythm in accordance with one or more techniques of the present disclosure.

[0010] Figure 2 It shows in more detail Figure 1 Conceptual diagram of the implantable medical device (IMD) and leads of the system.

[0011] Figure 3 is a block diagram of an example implantable medical device in accordance with one or more techniques of this disclosure.

[0012] Figure 4 is a block diagram illustrating an example computing device operating in accordance with one or more techniques of this disclosure.

[0013] Figure 5 is a flow diagram illustrating example operation of an artificial intelligence (AI) system in accordance with one or more techniques of this disclosure.

[0014] Figure 6 is a flow diagram illustrating a first example operation for preprocessing one or more electrocardiac electrogram (EGM) strips in accordance with one or more techniques of the present disclosure.

[0015] Figure 7 is a flow chart illustrating a second example operation for pre-processing a cardiac EGM strip in accordance with one or more techniques of the present disclosure.

[0016] Fig. 8A is a conceptual diagram illustrating example waveforms and envelopes determined according to one or more techniques of this disclosure.

[0017] Figure 8Bis a conceptual diagram illustrating an example normalized waveform generated according to one or more techniques of this disclosure.

[0018] Fig. 9 is a flow diagram illustrating example operation of an AI system in accordance with one or more techniques of this disclosure.

[0019] Fig. 10A is a conceptual diagram showing a graph of example cardiac waveforms and QRS markers detected by a device.

[0020] Fig. 10B is a conceptual diagram showing an example of a plot of heart rate over time based on a marker channel.

[0021] Fig.11 is a flow chart illustrating a third example operation for pre-processing a cardiac EGM strip in accordance with one or more techniques of the present disclosure.

[0022] Fig.12 is a conceptual diagram illustrating an example original cardiac waveform and waveforms of channels corresponding to different frequency bands, in accordance with one or more techniques of the present disclosure.

[0023] Fig.13 is a flow chart illustrating a fourth example operation for pre-processing a cardiac EGM strip in accordance with one or more techniques of the present disclosure.

[0024] Fig.14 is a flow diagram illustrating example operation of an AI system in accordance with one or more techniques of this disclosure.

[0025] Fig.15 is a flow chart illustrating a fifth example operation for pre-processing a cardiac EGM strip in accordance with one or more techniques of the present disclosure.

[0026] Fig.16 is a conceptual diagram illustrating example operations of an AI system including an autoencoder in accordance with one or more techniques of this disclosure.

[0027] Fig.17 is a flow diagram illustrating example operations in which an autoencoder is used to confirm device classification in accordance with one or more techniques of this disclosure.

[0028] Like reference numerals refer to like elements throughout the drawings and the description. DETAILED DESCRIPTION

[0029] Figure 1 is a block diagram illustrating a system 10 for analyzing one or more aspects of a cardiac rhythm of a patient 14 in accordance with the techniques of the present disclosure. The system 10 includes a medical device 16. An example of such a medical device is an implantable medical device (IMD), such as Figure 1As shown. Figure 1 As shown in the example system 10 in FIG. 1 , in some instances, the medical device 16 can be, for example, an implantable cardiac monitor, an implantable cardiac pacemaker, an implantable cardioverter / defibrillator (ICD), or a pacemaker / cardioverter / defibrillator. In some instances, the medical device 16 is a non-implantable medical device, such as a non-implantable cardiac monitor (e.g., a Holter monitor).

[0030] exist Figure 1 In an example of , the medical device 16 is connected to the leads 18, 20 and 22 and is communicatively coupled to an external device 27, which in turn is communicatively coupled to a computing system 24 via a communication network 25. The medical device 16 senses electrical signals associated with depolarization and repolarization of the heart 12, such as a cardiac electrogram (EGM), via electrodes on one or more leads 18, 20 and 22 or a housing of the medical device 16. The medical device 16 can also deliver therapy to the heart 12 in the form of electrical signals via electrodes positioned on one or more leads 18, 20 and 22 or a housing of the medical device 16. The therapy can be pacing, cardioversion and / or defibrillation pulses. The medical device 16 can monitor the cardiac EGM signals collected by the electrodes on the leads 18, 20 or 22, and based on the cardiac EGM signals, diagnose and treat arrhythmias.

[0031] In some examples, medical device 16 includes communication circuitry 17 including a device for communicating with another device such as Figure 1 The medical device 16 may include any suitable circuitry, firmware, software, or any combination thereof, for communicating with an external device 27. For example, the communication circuitry 17 may include one or more processors, memories, radios, antennas, transceivers, receivers, modulation and demodulation circuitry, filters, amplifiers, etc., for radio frequency communication with other devices such as the computing system 24. The medical device 16 may use the communication circuitry 17 to receive downlink data to control one or more operations of the medical device 16 and / or to send uplink data to the external device 27.

[0032] Leads 18, 20, 22 extend into heart 12 of patient 14 to sense electrical activity of heart 12 and / or deliver electrical stimulation to heart 12. Figure 1 In the example shown, the right ventricular (RV) lead 18 extends through one or more veins (not shown), the superior vena cava (not shown), and the right atrium 26, and into the right ventricle 28. The left ventricular (LV) lead 20 extends through one or more veins, the vena cava, the right atrium 26, and into the coronary sinus 30 to a region adjacent to the free wall of the left ventricle 32 of the heart 12. The right atrial (RA) lead 22 extends through one or more veins and the vena cava, and into the right atrium 26 of the heart 12.

[0033] although Figure 1 The example system 10 of illustrative embodiments depicts a medical device 16, but in other examples, the techniques of the present disclosure may be applied to other types of medical devices that are not necessarily implantable. For example, a medical device according to the techniques of the present disclosure may include a wearable medical device or "smart" clothing worn by a patient 14. For example, such a medical device may take the form of a watch worn by the patient 14 or a circuit system attached to the patient 14 in an adhesive manner. In another example, a medical device as described herein may include an external medical device having implantable electrodes.

[0034] In some examples, external device 27 takes the form of an external programmer or mobile device, such as a mobile phone, a "smart" phone, a laptop computer, a tablet computer, a personal digital assistant (PDA), etc. In some examples, external device 27 is the CareLink 3000 available from Medtronic, Inc. TM Monitor. A user, such as a physician, technician, surgeon, electrophysiologist, or other clinician, may interact with external device 27 to retrieve physiological or diagnostic information from medical device 16. A user, such as patient 14 or clinician as described above, may also interact with external device 27 to program medical device 16, for example, to select or adjust values ​​for operating parameters of medical device 16. External device 27 may include processing circuitry, memory, a user interface, and communication circuitry capable of transmitting and receiving information to and from each of medical device 16 and computing system 24.

[0035] In some instances, the computing system 24 takes the form of a handheld computing device, a computer workstation, a server or other networked computing device, a smart phone, a tablet computer or an external programmer, which includes a user interface for presenting information to a user and receiving input from the user. In some instances, the computing system 24 may include one or more devices implementing a machine learning system, such as a neural network, a deep learning system or other types of machine learning systems. A user (such as a physician, technician, surgeon, electrophysiologist or other clinician) can interact with the computing system 24 to retrieve physiological or diagnostic information from the medical device 16. The user can also interact with the computing system 24 to program the medical device 16, for example, to select values ​​for the operating parameters of the IMD. The computing system 24 may include a processor configured to evaluate the cardiac EGM (or its segment) and / or other sensing signals transmitted from the medical device 16 to the computing system 24.

[0036] The network 25 may include one or more computing devices (not shown), such as one or more non-edge switches, routers, hubs, gateways, security devices such as firewalls, intrusion detection and / or intrusion prevention devices, servers, computer terminals, laptops, printers, databases, wireless mobile devices such as cellular phones or personal digital assistants, wireless access points, bridges, cable modems, application accelerators or other network devices. The network 25 may include one or more networks managed by a service provider and may therefore form part of a large-scale public network infrastructure such as the Internet. The network 25 may provide computing devices such as computing systems 24 and medical devices 16 that access the Internet, and may provide a communication framework that allows computing devices to communicate with each other. In some instances, the network 25 may be a dedicated network that provides a communication framework that allows computing systems 24, medical devices 16, and EMR databases 66 to communicate with each other, but isolates computing systems 24, medical devices 16, and EMR databases 66 from external devices for security purposes. In some instances, the communication between computing systems 24, medical devices 16, and EMR databases 66 is encrypted.

[0037] External device 27 and computing system 24 may communicate via wireless or non-wireless communications over network 25 using any technology known in the art. In some examples, computing system 24 is a remote device that communicates with external device 27 via an intermediate device located in network 25, such as a local access point, wireless router, or gateway. Figure 1 In some examples, the external device 27 and the computing system 24 communicate via the network 25, but in some examples, the external device 27 and the computing system 24 communicate directly with each other. Examples of communication technology may include, for example, Other communication technologies are also contemplated. The computing system 24 may also communicate with one or more other external devices using a variety of known wired and wireless communication technologies.

[0038] The EMR database 66 stores EMR data for the patient 14. The EMR database 66 may include a processing circuit system and one or more storage media (e.g., random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electronically erasable programmable read-only memory (EEPROM), or flash memory. In some instances, the EMR database 66 is a cloud computing system. In some instances, the functionality of the EMR database 66 is distributed across multiple computing systems.

[0039] In one example, computing system 24 receives patient data collected by medical device 16 of patient 14. In some examples, the patient data includes physiological data of patient 14, such as one or more of the following: activity level of patient 14, heart rate of patient 14, posture of patient 14, electrocardiogram of patient 14, blood pressure of patient 14, pulse transit time of patient 14, respiratory rate of patient 14, hypopnea index or apnea of ​​patient 14, accelerometer data of patient 14, features derived from accelerometer data of patient 14, such as activity counts, posture, statistical control process variables, etc., raw electromyogram or cardiac EGM of patient 14, one or more features derived from raw electromyogram of patient 14, such as heart rate variability, t-wave alternans, QRS morphology, etc., interval data and features derived from interval data, heart sounds, potassium level, blood glucose index, temperature of patient 14, or any data that can be derived from the above parameterized data or any other type of patient parameter data. In some examples, medical device 16 or another device can automatically generate patient parameter data by processing information from one or more sensors. For example, medical device 16 may determine via one or more sensors that patient 14 has fallen, that patient 14 is frail or ill, or that patient 14 is suffering from a condition of sleep apnea.

[0040] In some instances, the patient data includes environmental data, such as air quality measurements, ozone levels, particle counts, or pollution levels, ambient temperature, or daylight hours in proximity to the patient 14. In some instances, one of the medical device or external device 27 can sense the environmental data via one or more sensors. In another instance, the environmental data is received by the external device 27 via an application (such as a weather application) executed on the external device 27 and uploaded to the computing system 24 over the network 25. In another instance, the computing system 24 collects the environmental data directly from a cloud service that has location-based data of the patient 14.

[0041] In some examples, the patient data includes patient symptom data uploaded by the patient 14 via an external device, such as the external device 27. For example, the patient 14 may upload the patient symptom data via an application executed on a smart phone. In some examples, the patient 14 may upload the patient symptom data via a user interface ( Figure 1 (not depicted) such as uploading patient symptom data via a touch screen, keyboard, graphical user interface, voice command, etc.

[0042] In some instances, the patient data includes device-related data, such as one or more of the impedance of one or more electrodes of the medical device, the selection of electrodes, a drug delivery schedule for the medical device, a history of electrical pacing therapy delivered to the patient, or diagnostic data of the medical device. In some instances, the medical device collecting the patient data is an IMD. In other instances, the medical device collecting the patient data is another type of patient device, such as a wearable medical device or a mobile device (e.g., a smart phone) of the patient 14. In some instances, the computing system 24 receives the patient data periodically, such as daily.

[0043] In some instances, computing system 24 further receives EMR data for patient 14 from EMR database 66. EMR data may be considered another form of patient data. In some instances, the EMR data stored by EMR database 66 may contain many different types of historical medical information about patient 14. For example, as examples, EMR database 66 may store a patient's medication history, a patient's surgical history, a patient's hospitalization history, a patient's potassium level over time, one or more laboratory test results for patient 14, a cardiovascular history for patient 14, or a comorbidity of patient 14, such as atrial fibrillation, heart failure, or diabetes.

[0044] The computing system 24 may implement a cardiac EGM monitoring system that can help manage chronic heart disease. According to the techniques of the present disclosure, to implement the cardiac EGM monitoring system, the computing system 24 may apply artificial intelligence (AI) techniques to analyze patient data, such as cardiac EGM data. Example AI techniques may include deep learning or other machine learning techniques. A neural network algorithm is an example of a deep learning algorithm.

[0045] An AI system is a computing system that includes memory and one or more processing circuits configured to perform AI techniques. Figure 1 In the context of , the AI ​​system may be the medical device 16, the computing system 24, the external device 27, or another device or system of devices. Thus, in this disclosure, unless otherwise indicated, discussions of actions performed by the AI ​​system may apply to actions performed by any of these devices.

[0046] The AI ​​system can generate data about one or more aspects of the cardiac rhythm of patient 14. For example, the AI ​​system can generate arrhythmia data based at least in part on a cardiac EGM strip obtained from one or more medical devices such as medical device 16, the data indicating whether the cardiac EGM strip represents one or more occurrences of one or more arrhythmias. The cardiac EGM strip includes data representing the cardiac rhythm of the patient in a continuous time period (e.g., 30 seconds, 45 seconds, etc.). The cardiac EGM strip may include a series of samples of waveforms representing cardiac rhythms. A user (e.g., a technician, a physician, a patient, a healthcare professional, or other type of user) can view the occurrence of one or more detected arrhythmias for diagnostic purposes or as part of performing ongoing care for patient 14. In addition to the cardiac EGM strip, the AI ​​system can also use one or more other types of patient data to detect the occurrence of arrhythmias, such as information from the electronic medical record of patient 14.

[0047] The AI ​​system may be trained to identify one or more aspects of the cardiac rhythm of patient 14 that are of interest for a given cardiac EGM strip by applying one or more deep learning models that have been trained to identify such aspects of the cardiac rhythm of patient 14. The various aspects of the cardiac rhythm of patient 14 may include various arrhythmias, the location of such arrhythmias within one or more cardiac EGM strips (which reflects the time at which the arrhythmias occur), morphological aspects of the occurrence of arrhythmias, etc. The cardiac rhythm classification model may be trained on cardiac EGM strips extracted from a population of subjects, individual patients, groups of patients, and in some instances other data.

[0048] The performance of the deep learning model can be improved by preprocessing the input data provided to the deep learning model. For example, as described in the present disclosure, preprocessing the data enables the deep learning model to be used with cardiac EGM strips generated by various types of devices. Therefore, according to the technology of the present disclosure, the AI ​​system can obtain a cardiac EGM strip representing a waveform of the patient's cardiac rhythm. Additionally, the AI ​​system can preprocess the cardiac EGM strip. The computing system can then apply the deep learning model to the preprocessed cardiac EGM strip to generate arrhythmia data indicating whether the cardiac EGM strip represents one or more occurrences of one or more arrhythmias. For example, a deep learning system can be trained to generate arrhythmia data including an element vector corresponding to different arrhythmias. The value of the element in the vector indicates whether the occurrence of the corresponding arrhythmia appears in the cardiac EGM strip.

[0049] Figure 2 It shows in more detail Figure 120, 22 of system 10. In the illustrated example, bipolar electrodes 40 and 42 are positioned adjacent to the distal end of lead 18, and bipolar electrodes 48 and 50 are positioned adjacent to the distal end of lead 22. In addition, four electrodes 44, 45, 46, and 47 are positioned adjacent to the distal end of lead 20. Lead 20 may be referred to as a quadripolar LV lead. In other examples, lead 20 may include more or fewer electrodes. In some examples, LV lead 20 includes segmented electrodes, for example, wherein each of the plurality of longitudinal electrode locations of the lead, such as the locations of electrodes 44, 45, 46, and 47, includes a plurality of discrete electrodes arranged at corresponding circumferential locations around the circumference of the lead.

[0050] In the illustrated example, electrodes 40 and 44-48 take the form of ring electrodes, and electrodes 42 and 50 may take the form of extendable spiral tip electrodes telescopically mounted within insulating electrode heads 52 and 56, respectively. Leads 18 and 22 also include elongated electrodes 62 and 64, respectively, which may take the form of coils. In some examples, each of electrodes 40, 42, 44-48, 50, 62, and 64 is electrically coupled to a corresponding conductor within the lead body of its associated lead 18, 20, 22, and thereby coupled to the circuit system within medical device 16.

[0051] In some examples, the medical device 16 includes one or more housing electrodes, such as Figure 2 The housing electrode 4 shown may be integrally formed with an outer surface of the hermetically sealed housing 8 of the medical device 16 or otherwise coupled to the housing 8. In some examples, the housing electrode 4 is defined by an uninsulated portion of the exterior-facing portion of the housing 8 of the medical device 16. Other divisions between insulated and non-insulated portions of the housing 8 may be used to define two or more housing electrodes. In some examples, the housing electrode includes substantially all of the housing 8.

[0052] The housing 8 encloses signal generation circuitry for generating therapeutic stimulation (e.g., cardiac pacing, cardioversion, and defibrillation pulses), and sensing circuitry for sensing electrical signals associated with depolarization and repolarization of the heart 12. The housing 8 may also enclose a memory for storing the sensed electrical signals. The housing 8 may also enclose communication circuitry 17 for communication between the medical device 16 and the computing system 24.

[0053] Medical device 16 senses electrical signals accompanying depolarization and repolarization of heart 12 via electrodes 4, 40, 42, 44-48, 50, 62, and 64. Medical device 16 may sense such electrical signals via any bipolar combination of electrodes 40, 42, 44-48, 50, 62, and 64. In addition, any of electrodes 40, 42, 44-48, 50, 62, and 64 may be combined with housing electrode 4 for unipolar sensing.

[0054] The shown number and configuration of leads 18, 20, and 22 and electrodes are examples only. Other configurations, i.e., the number and positioning of leads and electrodes, are also possible. In some examples, system 10 may include additional leads or lead segments having one or more electrodes positioned at different locations in the cardiovascular system for sensing and / or delivering therapy to patient 14. For example, instead of or in addition to intercardiac leads 18, 20, and 22, system 10 may include one or more epicardial or extravascular (e.g., subcutaneous or substernal) leads that are not positioned within heart 12.

[0055] The medical device 16 may send patient data to the computing system 24 (e.g., via the external device 27). The patient data may include data based on electrical signals detected by the electrodes 4, 40, 42, 44-48, 50, 62, and / or 64. For example, the medical device 16 may collect cardiac EGM data and send it to the computing system 24. According to the techniques of the present disclosure, an AI system that may be implemented by the computing system 24, the medical device 16, or another device may pre-process the patient data and use the pre-processed patient data to determine to generate arrhythmia data or other data about the cardiac rhythm of the patient 14. In some instances, the medical device 16 may pre-process the cardiac EGM strip and the computing system 24 or the external device 27 may apply a deep learning model to the pre-processed cardiac EGM strip to generate arrhythmia data.

[0056] Although described herein in the context of an example medical device 16 that provides therapeutic electrical stimulation, the techniques disclosed herein can be used with other types of devices. For example, the techniques can be implemented with an external cardiac defibrillator coupled to electrodes external to the cardiovascular system, a transcatheter pacemaker configured to be implanted within the heart, such as the Micra 1000, commercially available from Medtronic, Inc. of Dublin, Ireland. TM Transcatheter pacing systems, insertable cardiac monitors such as the Reveal LINQ also commercially available from Medtronic TMICM, neurostimulator, drug delivery device, wearable device such as a wearable cardioverter defibrillator, fitness tracker or other wearable device, mobile device such as a mobile phone, "smart" phone, laptop, tablet computer, personal digital assistant (PDA) or "smart" clothing such as "smart" glasses or "smart" watch.

[0057] Figure 3 is a block diagram of an example medical device 16 according to the techniques of the present disclosure. In the example shown, the medical device 16 includes a processing circuit system 58, a memory 59, a communication circuit system 17, a sensing circuit system 50, a therapy delivery circuit system 52, a sensor 57, and a power supply 54. The memory 59 includes computer-readable instructions that, when executed by the processing circuit system 58, cause the medical device 16 and the processing circuit system 58 to perform various functions attributed to the medical device 16 and the processing circuit system 58 herein (e.g., performing short-term prediction of arrhythmias, delivering therapies such as anti-tachycardia pacing, bradycardia pacing, and post-shock pacing therapy, etc.). The memory 59 may include any volatile media, non-volatile media, magnetic media, optical media, or electronic 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 or analog media.

[0058] Processing circuit system 58 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 analog logic circuit system. In some examples, processing circuit system 58 may include multiple components (such as any combination of one or more microprocessors, one or more controllers, one or more DSPs, one or more ASICs, or one or more FPGAs) and other discrete or integrated logic circuit systems. The functions attributed to processing circuit system 58 herein may be embodied in software, firmware, hardware, or any combination thereof.

[0059] The processing circuit system 58 controls the therapy delivery circuit system 52 to deliver stimulation therapy to the heart 5 according to therapy parameters that can be stored in the memory 59. For example, the processing circuit system 58 can control the therapy delivery circuit system 52 to deliver electrical pulses having an amplitude, pulse width, frequency, or electrode polarity specified by the therapy parameters. In this way, the therapy delivery circuit system 52 can deliver pacing pulses (e.g., ATP pulses, bradycardia pacing pulses, or post-shock pacing therapy) to the heart 5 via the electrodes 34 and 40. In some instances, the therapy delivery circuit system 52 can deliver pacing stimulation in the form of voltage or current electrical pulses, such as ATP therapy, bradycardia therapy, or post-shock pacing therapy. In other instances, the therapy delivery circuit system 52 can deliver one or more of these types of stimulation in the form of other signals (such as sine waves, square waves, or other substantially continuous time signals).

[0060] The therapy delivery circuit system 52 is electrically coupled to the electrodes 34 and 40 carried on the housing of the medical device 16. Although the medical device 16 may include only two electrodes, such as electrodes 34 and 40, in other instances, the medical device 16 may utilize three or more electrodes. The medical device 16 may use any combination of electrodes to deliver therapy and / or detect electrical signals from the patient 12. In some instances, the therapy delivery circuit system 52 includes a charging circuit, one or more pulse generators, capacitors, converters, switching modules, and / or other components capable of generating and / or storing energy for delivery as pacing therapy, cardiac resynchronization therapy, other therapies, or a combination of therapies. In some instances, the therapy delivery circuit system 52 delivers therapy as one or more electrical pulses according to one or more sets of therapy parameters that define the amplitude, frequency, voltage, or current of the therapy or other parameters of the therapy.

[0061] The sensing circuit system 50 monitors the signals from the electrodes 4, 40, 42, 44-48, 50, 62 ( Figure 2 ) and 64( Figure 2 ) to monitor the electrical activity, impedance or other electrical phenomena of the heart 12. In some instances, the sensing circuit system 50 includes one or more analog components, digital components or a combination thereof. In some instances, the sensing circuit system 50 includes one or more sense amplifiers, comparators, filters, rectifiers, threshold detectors, analog-to-digital converters (ADCs), etc. In some instances, the sensing circuit system 50 can convert the sensed signals into digital form and provide the digital signals to the processing circuit system 58 for processing or analysis. In one instance, the sensing circuit system 50 amplifies the signals from the electrodes 4, 40, 42, 44-48, 50, 62 and 64 and converts the amplified signals into multi-bit digital signals through the ADC.

[0062] In some examples, the sensing circuit system 50 performs sensing of the cardiac electrogram to determine the heart rate or heart rate variability, or detects arrhythmias (e.g., tachyarrhythmias or bradycardias) or senses other parameters or events from the cardiac electrogram. The sensing circuit system 50 may also include a switching circuit system for selecting which electrode (and electrode polarity) of the available electrodes to use to sense cardiac activity based on which electrode combination or electrode vector is used in the current sensing configuration. The processing circuit system 58 may control the switching circuit system to select the electrode used as the sensing electrode and its polarity. The sensing circuit system 50 may include one or more detection channels, each of which may be coupled to a selected electrode configuration to detect cardiac signals via the electrode configuration. In some examples, the sensing circuit system 50 may compare the processed signal with a threshold value to detect the presence of atrial or ventricular depolarization and indicate the presence of atrial depolarization (e.g., P wave) or ventricular depolarization (e.g., R wave) to the processing circuit system 58. Sensing circuitry 50 may include one or more amplifiers or other circuitry for comparing the cardiac electrogram amplitude to a threshold value, which may be adjustable.

[0063] The processing circuit system 58 may include a timing and control module, which may be embodied as hardware, firmware, software, or any combination thereof. The timing and control module may include a dedicated hardware circuit (such as an ASIC) separate from other processing circuit system 58 components (such as a microprocessor), or a software module executed by a component of the processing circuit system 58, which may be a microprocessor or an ASIC. The timing and control module may implement a programmable counter. If the medical device 16 is configured to generate bradycardia pacing pulses and deliver them to the heart 12, such a counter may control the basic time intervals associated with DDD, VVI, DVI, VDD, AAI, DDI, DDDR, VVIR, DVIR, VDDR, AAIR, DDIR, or other pacing modes.

[0064] In some examples, the processing circuit system 58 of the medical device 16 implements the AI ​​system 300. For example, the processing circuit system 58 can apply a deep learning model to the cardiac EGM strip, as described elsewhere in this disclosure. The processing circuit system 58 can implement the AI ​​system 300 using dedicated circuitry or by executing software instructions stored on a computer-readable medium such as a memory 59. The sensing circuit system 50 can generate a cardiac EGM strip based on data received from the electrodes 4, 40, 42, 44-48, 50, 62, and 64. The communication circuit system 17 can transmit the cardiac EGM strip and / or other data to the external device 27.

[0065] Memory 59 may be configured to store various operating parameters, therapy parameters, sensed and detected data, and any other information relevant to therapy and treatment of patient 12. Figure 3 In examples of memory 59, memory 59 may store, for example, sensed cardiac EGMs associated with detected or predicted arrhythmias, and therapy parameters defining therapy delivery provided by therapy delivery circuitry 52. ​​In other examples, memory 59 may act as a temporary buffer for storing data until the data can be uploaded to computing system 24.

[0066] The communication circuit system 17 includes any suitable circuit system, firmware, software or any combination thereof for communicating via Figure 1 The communication circuit system 17 may be configured to communicate with another device, such as the computing system 24, via the network 25 of the present invention. For example, the communication circuit system 17 may include one or more antennas, modulation and demodulation circuit systems, filters, amplifiers, etc., for radio frequency communication with other devices, such as the computing system 24, via the network 25. Under the control of the processing circuit system 58, the communication circuit system 17 may receive downlink telemetry from the computing system 24 and send uplink telemetry to the computing system via the antenna, which may be internal and / or external. The processing circuit system 58 may provide data to be transmitted uplink to the computing system 24 and control signals for telemetry circuits within the communication circuit system 17, for example, via an address / data bus. In some examples, the communication circuit system 17 may provide the received data to the processing circuit system 58 via a multiplexer.

[0067] The power source 54 may be any type of device configured to hold an electrical charge to operate the circuitry of the medical device 16. The power source 54 may be provided as a rechargeable or non-rechargeable battery. In other examples, the power source 54 may be incorporated into an energy scavenging system that stores electrical energy from movement of the medical device 16 within the patient 12.

[0068] According to the technology of the present disclosure, the medical device 16 collects patient data of the patient 14 via the sensing circuit system 50 and / or the sensor 57. The sensor 57 may include one or more sensors, such as one or more accelerometers, pressure sensors, optical sensors for O2 saturation, etc. In some instances, the patient data includes one or more of the following: the activity level of the patient 14, the heart rate of the patient 14, the posture of the patient 14, the electrocardiogram of the patient 14 (e.g., a cardiac EGM strip of the patient 14), the blood pressure of the patient 14, the accelerometer data of the patient 14, or other types of patient parameter data. The medical device 16 uploads the patient parameter data to the computing system 24 via the communication circuit system 17 through the network 25. In some instances, the medical device 16 uploads the patient parameter data to the computing system 24 every day. In some instances, the patient parameter data includes one or more values ​​representing average measurements of the patient 14 over a long period of time (e.g., about 24 hours to about 48 hours). For example, one or more other devices, such as a wearable medical device or a mobile device (eg, a smart phone) of patient 14 , may collect patient parameter data and upload the patient parameter data to external device 27 and / or computing system 24 .

[0069] Although described herein in the context of an example medical device 16 that provides therapeutic electrical stimulation, the techniques disclosed herein for short-term prediction of cardiac arrhythmias can be used with other types of devices. For example, the techniques can be implemented with a transcatheter pacemaker configured to be implanted within the heart, such as the Micra 1000, commercially available from Medtronic plc of Dublin, Ireland. TM Transcatheter pacing systems, insertable cardiac monitors such as the Reveal LINQ also commercially available from Medtronic TM ICM, neurostimulator, drug delivery device, wearable device such as a wearable cardioverter defibrillator, fitness tracker or other wearable device, mobile device such as a mobile phone, "smart" phone, laptop, tablet computer, personal digital assistant (PDA) or "smart" clothing such as "smart" glasses or "smart" watch.

[0070] Figure 4 4 is a block diagram illustrating an example computing system 24 operating in accordance with one or more techniques of the present disclosure. In one example, computing system 24 includes processing circuitry 402 for executing application programs 424 including monitoring system 450 or any other application programs described herein. Figure 4 24, but the computing system 24 may be any component or system including a processing circuit system or other suitable computing environment for executing software instructions, and need not include, for example, Figure 4 406; and in some instances, components such as storage device 408 may not be co-located or located in the same rack as other components. In some instances, computing system 24 may be a cloud computing system distributed across multiple devices.

[0071] like Figure 4 As shown in the example of , the computing system 24 includes a processing circuit system 402, one or more input devices 404, a communication circuit system 406, one or more output devices 412, one or more storage devices 408, and a user interface (UI) device 410. In one example, the computing system 24 further includes one or more applications 424 (such as monitoring system 450) and an operating system 416 that can be executed by the computing system 24. Each of the components 402, 404, 406, 408, 410 and 412 is coupled (physically, communicatively and / or operatively) for inter-component communication. In some examples, the communication channel 414 can include a system bus, a network connection, an inter-process communication data structure, or any other method for transmitting data. As an example, the components 402, 404, 406, 408, 410 and 412 can be coupled via one or more communication channels 414.

[0072] In one example, processing circuitry 402 is configured to implement functions and / or process instructions for execution within computing system 24. For example, processing circuitry 402 may be capable of processing instructions stored in storage device 408. Examples of processing circuitry 402 may include any one or more of 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.

[0073] One or more storage devices 408 may be configured to store information within the computing system 24 during operation. In some instances, the storage device 408 is described as a computer-readable storage medium. In some instances, the storage device 408 is a temporary memory, which means that the primary purpose of the storage device 408 is not long-term storage. In some instances, the storage device 408 is described as a volatile memory, which means that when the computer is turned off, the storage device 408 does not maintain the stored content. 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 instances, the storage device 408 is used to store program instructions executed by the processing circuit system 402. In one example, the storage device 408 is used by software or application 424 running on the computing system 24 to temporarily store information during program execution.

[0074] In some instances, storage device 408 also includes one or more computer-readable storage media. Storage device 408 can be configured to store an amount of information larger than a volatile memory. Storage device 408 can be further configured to store information for a long time. In some instances, storage device 408 includes a non-volatile storage element. Examples of such non-volatile storage elements include magnetic hard disks, optical disks, floppy disks, flash memories, or various forms of electrically programmable memory (EPROM) or electrically erasable and programmable (EEPROM) memory. In some instances, storage device 408 can store patient data, such as cardiac EGM strips.

[0075] In some examples, computing system 24 also includes communication circuit system 406. In one example, computing system 24 uses communication circuit system 406 to communicate with external devices (such as Figure 1 The communication circuit system 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 can send and receive information. Other examples of such network interfaces may include 3G, 4G, 5G, and WI-FI. TM radio.

[0076] In one example, computing system 24 also includes one or more user interface devices 410. In some examples, user interface device 410 is configured to receive input from a user through tactile, audio, or video feedback. Examples of user interface device 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, a presence-sensitive display includes a touch-sensitive screen.

[0077] One or more output devices 412 may also be included in the computing system 24. In some instances, the output device 412 is configured to provide output to the user using tactile, audio, or video stimulation. In one instance, the output device 412 includes a display, a sound card, a video graphics adapter card, or any other type of device for converting signals into a suitable form understandable to humans or machines. In some instances, the output device 412 includes a display device. Other instances of the 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 understandable output to the user.

[0078] Computing system 24 may include operating system 416. In some examples, operating system 416 controls the operation of components of computing system 24. For example, in one example, operating system 416 facilitates communication of one or more applications 424 and monitoring system 450 with processing circuitry 402, communication circuitry 406, storage devices 408, input devices 404, user interface devices 410, and output devices 412.

[0079] Applications 422 may also include program instructions and / or data that may be executed by computing system 24. An example application 422 that may be executed by computing system 24 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.

[0080] According to the technology of the present disclosure, application 424 includes monitoring system 450. Monitoring system 450 can be configured to receive patient data, evaluate patient data, and generate output data. For example, in one example, when monitoring system 450 determines that patient 14 ( Figure 1 ) is likely to have experienced one or more arrhythmic events belonging to one or more types of arrhythmias, the monitoring system 450 can generate a notification. In another example, the monitoring system 450 can generate one or more graphs showing changes in one or more aspects of the patient's 14 heart rhythm.

[0081] like Figure 4As shown in the example of , in some examples, the monitoring system 450 can implement an AI system 451 that includes a preprocessing unit 452 and a deep learning model 454. Although the remainder of the present disclosure refers to the preprocessing unit 452 and the deep learning model 454, such references can be equally applicable to preprocessing units and deep learning models implemented in devices or systems other than the computing system 24 (such as in the medical device 16 or the external device 27, or a combination thereof). In addition, the AI ​​system 451, the preprocessing unit 452, and the deep learning model 454 can be implemented outside the context of the monitoring system. The preprocessing unit 452 can preprocess the input data that the AI ​​system 451 provides as input to the deep learning model 454, including cardiac EGM strips.

[0082] In some instances, the deep learning model 454 is implemented using one or more neural network systems, deep learning systems, or other types of supervised or unsupervised machine learning systems. For example, the deep learning model 454 can be implemented by a feedforward neural network such as a convolutional neural network, a radial basis function neural network, a recursive neural network, a modular or associative neural network. In some instances, the AI ​​system 451 trains the deep learning model 454 with patient data of one or more patient groups including cardiac EGM strips to generate data about one or more aspects of the cardiac rhythm of patients in the group. In some instances, after the AI ​​system 451 has pre-trained the deep learning model 454, the AI ​​system 451 can further train the deep learning model 454 with patient data specific to patient 14 or a smaller group of patients.

[0083] In some instances, the AI ​​system 451 trains the deep learning model 454 with patient data of a patient population, determines an error rate of the deep learning model 454, and then feeds the error rate back to the deep learning model 454 to allow the deep learning model 454 to update its prediction based on the error rate. In some instances, the error rate may correspond to the difference between the output data determined by the deep learning model 454 based on the input data and the pre-labeled output data of the same input data. In some instances, the AI ​​system 451 may use an error function to determine the error rate. The error function may be implemented using signal processing techniques and heuristics in a manner conventionally used to detect the occurrence of arrhythmias. In some instances, the monitoring system 450 may receive feedback from a user (e.g., patient 14, clinician, or another type of person) indicating whether the detected arrhythmia occurred in patient 14 within a specific time period. In some instances, the monitoring system 450 may receive a message from a medical device 16 indicating that the medical device 16 has detected (or has not detected) the occurrence of an arrhythmia in patient 14. In some instances, the monitoring system 450 may obtain feedback in other ways, such as by periodically checking the EMR data to determine whether an arrhythmia has occurred. The monitoring system 450 may update the deep learning model 454 with the feedback. Thus, the training process may occur iteratively to gradually improve the data generated by the deep learning model 454 by "learning" from correct and incorrect data generated by the deep learning model 454 in the past. Further, the training process may be used to further fine-tune the deep learning model 454 trained using population-based data to generate more accurate data for a particular individual. In some instances, a person at the monitoring service may provide feedback.

[0084] In some instances, the deep learning model 454 is implemented using a neural network. The neural network may include an input layer and an output layer, as well as one or more hidden layers between the input layer and the output layer. Each layer of the neural network may include one or more artificial neurons, which are referred to as neurons in the present disclosure. The input layer of the neural network includes multiple input neurons. The input layer may include a separate input neuron for each sample value of a segment of the cardiac EGM strip. In some instances, the segment may be connected to the cardiac EGM strip. In other instances, the segment may be a sub-segment of the cardiac EGM strip. For example, in an instance where the cardiac EGM strip includes samples representing 45 seconds of the cardiac rhythm of patient 14, the segment may include samples representing the first 10 seconds of the cardiac EGM strip.

[0085] The AI ​​system 451 can provide overlapping segments of the cardiac EGM strip to the deep learning model 454. For example, the AI ​​system 451 can provide a segment including samples representing the 0th to 10th seconds of the cardiac EGM strip, and then provide a segment including samples representing the 5th to 15th seconds of the cardiac EGM strip, and then provide a segment including samples representing the 10th to 20th seconds of the cardiac EGM strip, and so on. In some instances, the computing system 24 can provide segments that span two or more cardiac EGM strips. For ease of explanation, applying a deep learning model to a cardiac EGM strip can actually refer to applying a deep learning model to a segment of the cardiac EGM strip.

[0086] In some instances, the deep learning model 454 includes a convolutional neural network (CNN). For example, in one instance, a convolutional layer may be followed by an input layer of the type described above. A first convolutional layer neuron may receive input from a first group of input layer neurons consisting of a given number of consecutive input layer neurons; a second convolutional layer neuron may receive input from a second group of input layer neurons, the second group of input layer neurons consisting of the same given number of consecutive input layer neurons, but offset by a stride length from the first input layer neuron of the first group of input layer neurons; a third convolutional layer neuron may receive input from a third group of input layer neurons, the third group of input layer neurons consisting of the same given number of consecutive input layer neurons, but offset by a stride length from the first input layer neuron of the second group of input layer neurons; and so on. The given number of consecutive input neurons and the stride length are different hyperparameters of the CNN. One or more fully connected hidden layers may follow the convolutional layer.

[0087] In some examples of the present disclosure, for each corresponding arrhythmia in a set of one or more arrhythmias, the deep learning model 454 can generate data indicating whether one or more occurrences of the corresponding arrhythmia are represented in a segment of the cardiac EGM strip. For example, in one example, the hidden layer of the deep learning model 454 provides input data to the output layer of the deep learning model 454. For each corresponding arrhythmia in a set of arrhythmias, the output layer of the deep learning model 454 includes a separate output neuron corresponding to the corresponding arrhythmia. The output neuron corresponding to the corresponding arrhythmia receives input data from a single neuron in the hidden layer of the deep learning model 454, and the input data also corresponds to the corresponding arrhythmia type. The data corresponding to the corresponding arrhythmia generated by the hidden layer neurons includes a probability value indicating the probability that the occurrence of the arrhythmia has appeared in the segment of the cardiac EGM strip. The activation function of the output neuron can apply a threshold function to the probability value generated by the hidden layer neuron. For each output neuron, the threshold function may cause the output neuron to generate a first value (e.g., 1) if the probability value provided to the output neuron is greater than a threshold, and to generate a second value (e.g., 0) if the probability value provided to the output neuron is less than the same threshold.

[0088] Furthermore, in the example of the previous paragraph, the AI ​​system 451 can use the probability values ​​generated by the hidden layer to track the location within the cardiac EGM strip where an occurrence of an arrhythmia occurs. For example, as described above, the cardiac EGM strip can be subdivided into segments and the AI ​​system 451 provides the segments as input to the deep learning model 454. Thus, by determining which segment of the cardiac EGM strip results in the highest probability value corresponding to an arrhythmia, the AI ​​system 451 can determine which segment is most likely to represent the occurrence of an arrhythmia.

[0089] As noted elsewhere in this disclosure, the input provided to the deep learning model 454 may include patient data in addition to the segments of the cardiac EGM strip. For example, in some instances, the patient data may additionally include data about the patient's physiological state (e.g., the patient's physiological state, such as activity, posture, breathing, etc.), which may also be captured by the medical device 16. Patient data corresponding to different physiological conditions (e.g., rest, nighttime rest, nighttime high-profile angle rest, etc.) can be used as additional parameters for model training or input data for the deep learning model 454. The use of such data can enable the AI ​​system 450 to detect the occurrence of arrhythmias during other disease conditions (e.g., a sensitive model of tachycardia during rest can be used to monitor heart failure (HF) patients; a model of bradycardia during activity can be used to monitor chronotropic insufficiency in patients). In some instances, the monitoring system 450 receives EMR data of the patient 14 from the EMR database 66 via the communication circuit system 406. In some instances, the EMR data stored by the EMR database 66 may contain many different types of historical medical information about the patient 14. For example, the EMR database 66 may store a patient's medication history, a patient's surgical history, a patient's hospitalization history, a patient's potassium level over time, or one or more laboratory test results for a patient, etc. The EMR data may form part of the patient data used as input to the deep learning model 454.

[0090] In some examples, the deep learning model 454 converts the patient data into one or more vectors and tensors (e.g., multidimensional arrays) representing the patient data. The deep learning model 454 can apply mathematical operations to the one or more vectors and tensors to generate a mathematical representation of the patient data. The deep learning model 454 can determine different weights corresponding to the identified relationship between the patient data and the occurrence of arrhythmia. The deep learning model 454 can apply different weights to the patient data to generate a probability value.

[0091] Figure 5 is a flow chart illustrating an example operation according to the technology of the present disclosure. Figure 1 and Figure 4 describe Figure 5 The flowcharts of the present disclosure are presented as examples. Other examples of the techniques according to the present disclosure may include more, fewer, or different actions, or may perform actions in a different order or in parallel. Figure 5 The operations may be performed by an AI system implemented on one or more of the medical device 16, the computing system 24, the external device 27, and / or other devices.

[0092] exist Figure 5In an example of , the AI ​​system can obtain one or more cardiac EGM strips (500) of patient 14 (i.e., the current patient). The AI ​​system can obtain one or more cardiac EGM strips of patient 14 in one or more of various ways. For example, in an example in which the AI ​​system is implemented by computing system 24, computing system 24 can obtain one or more cardiac EGM strips of patient 14 from medical device 16 (e.g., via external device 27 and network 25). In some examples in which the AI ​​system is implemented in medical device 16, the AI ​​system can obtain one or more cardiac EGM strips by generating one or more cardiac EGM strips based on data from electrodes. In some examples, the AI ​​system can obtain one or more cardiac EGM strips of the current patient from a database (e.g., EMR database 66) that stores one or more cardiac EGM strips of the current patient. Other examples of obtaining cardiac EGM strips are described elsewhere in this disclosure.

[0093] In addition, Figure 5 In the example of , the preprocessing unit 452 can preprocess one or more cardiac EGM strips (502). The preprocessing unit 452 can preprocess one or more cardiac EGM strips in one or more of a variety of ways. For example, Figure 6 , 7 , 9, 11 and 15 show example ways to preprocess one or more cardiac EGM strips. In some instances, the deep learning model 454 can be trained using cardiac EGM strips having certain characteristics such as signal polarity, signal amplitude levels and variations, hardware characteristics, etc. In such instances, after training, the preprocessing unit 452 can preprocess one or more cardiac EGM strips so that the one or more cardiac EGM strips have the characteristics of the training cardiac EGM strips. This can allow the deep learning model 454 to be used with cardiac EGM strips generated by more types of medical devices. For example, as described in detail below, Figure 7 and Fig.15 is an example in which the preprocessing unit 452 preprocesses one or more cardiac EGM strips to conform to one or more characteristics of the cardiac EGM strips on which the deep learning model 454 is trained. In addition, in some examples, the deep learning model 452 can be trained using the preprocessed cardiac EGM strips generated by the preprocessing unit 452. For example, in Figure 6 , Figure 7 , Fig. 9 and Fig.15 In the example, the AI ​​system can use preprocessed cardiac EGM strips to train the deep learning model 454.

[0094] After the preprocessing unit 452 preprocesses one or more cardiac EGM strips, the AI ​​system can apply the deep learning model 454 to the one or more preprocessed cardiac EGM strips to generate arrhythmia data (504) indicating whether the one or more cardiac EGM strips represent one or more occurrences of one or more arrhythmias. That is, the input to the deep learning model 454 can include one or more preprocessed cardiac EGM strips or segments thereof. The deep learning model 454 can be trained according to any of the examples described elsewhere in this disclosure. When the AI ​​system applies the deep learning model 454 to one or more preprocessed cardiac EGM strips, the AI ​​system can perform a feed-forward pass through the neurons of the neural network of the deep learning model 454. The arrhythmia data can be the output of the neural network.

[0095] In addition, in some instances, the AI ​​system can generate output data based on the arrhythmia data. For example, the AI ​​system can generate a table indicating the time and type of occurrence of arrhythmias detected. The time can be determined based on a timestamp provided by a medical device 16 with a cardiac EGM strip. In another example, the AI ​​system can generate a graph showing the rate at which the patient 14 experiences arrhythmia occurrences over time. The AI ​​system can present the output data to one or more types of users. For example, the AI ​​system can present the output data to the patient 14, the patient's 14 healthcare provider, a user of a healthcare monitoring organization, or other types of people.

[0096] Figure 6 is a flow chart illustrating a first example operation for preprocessing one or more cardiac EGM strips in accordance with one or more techniques of the present disclosure. Figure 6 In an instance of the deep learning model, the pre-processing unit 452 may generate one or more time-modified cardiac EGM strips (602) by modifying the temporal resolution of one or more cardiac EGM strips to match the expected temporal resolution of the deep learning model. For example, the pre-processing unit 452 may modify the temporal resolution of one or more cardiac EGM strips from 128 samples / second to 200 samples / second. The pre-processing unit 452 may modify the one or more cardiac EGM strips in one of a variety of ways. For example, in an instance where the expected temporal resolution of the deep learning model is greater than the temporal resolution of the one or more cardiac EGM strips, the pre-processing unit 452 may interpolate samples between samples of the one or more cardiac EGM strips. In an instance where the expected temporal resolution of the deep learning model is less than the temporal resolution of the one or more cardiac EGM strips, the pre-processing unit 452 may extract samples from the one or more cardiac EGM strips. Other techniques for modifying the temporal resolution of one or more cardiac EGM strips are possible.

[0097] Additionally, the preprocessing unit 452 may generate one or more preprocessed cardiac EGM strips by subtracting an average of samples of one or more time-modified cardiac EGM strips from samples of the one or more time-modified cardiac EGM strips (604). For example, the preprocessing unit 452 may determine an average of samples of one or more time-modified cardiac EGM strips and then subtract the average from each of the samples of the one or more time-modified cardiac EGM strips. In another example, the preprocessing unit 452 may subtract an average of samples of one or more raw cardiac EGM strips from one or more raw cardiac EGM strips and then modify the temporal resolution of the one or more resulting cardiac EGM strips, thereby reversing the order of actions 600 and 602.

[0098] Figure 7 is a flowchart showing a second example operation for preprocessing a cardiac EGM strip according to one or more techniques of the present disclosure. The medical device can be used for patients with different physiological conditions (e.g., fat / muscle mass) or underlying heart disease. Therefore, the signal levels in the cardiac EGM strips can be different. There may also be amplitude variations in the signal levels in the cardiac EGM strips of the same patient, for example, due to posture changes or premature ventricular contractions (PVCs). Because there may not be enough training data to train the deep learning model 454 from scratch or for transfer learning (e.g., a deep learning model developed using an external / resting Holter monitor is applied to a cardiac EGM strip from an implanted monitor of an active patient) under such conditions, applying additional normalization to the waveform to remove the mean value can improve the arrhythmia and QRS detection performance of the deep learning model.

[0099] Therefore, in Figure 7 In an example of the present invention, the pre-processing unit 452 may generate one or more time-modified cardiac EGM strips (700) by modifying the temporal resolution of the one or more cardiac EGM strips to match the expected temporal resolution of the deep learning model. Additionally, the pre-processing unit 452 may generate one or more mean-subtracted cardiac EGM strips (702) by subtracting the mean of the samples of the one or more time-modified cardiac EGM strips from the samples of the one or more time-modified cardiac EGM strips. The pre-processing unit 452 may generate one or more mean-subtracted cardiac EGM strips (702) in a manner similar to that described with respect to Figure 6 Actions 700 and 702 are performed in the same manner as described for actions 600 and 602.

[0100] In addition, Figure 7In an example of the present invention, to normalize the varying amplitude levels, the pre-processing unit 452 can calculate a moving signal envelope over a sliding window of one or more mean-subtracted cardiac EGM strips (704). The sliding window can be 0.5 seconds in duration, 1 second in duration, 1.5 seconds in duration, or have another duration. In some examples, the envelope can be the standard deviation of the waveform samples in the moving window or the 99th and 1st percentiles of the waveform samples in the moving window, or the maximum and minimum values ​​of the waveform samples in the moving window.

[0101] Fig. 8A is a conceptual diagram showing example waveforms and envelopes determined according to one or more techniques of the present disclosure. Specifically, Fig. 8A Shown are examples of waveforms (solid lines) and envelopes calculated over a 1 second moving window with the 99th (upper dashed line) and 1st (lower dashed line) percentile samples in the moving window.

[0102] In addition, Figure 7 In an example of , the pre-processing unit 452 can normalize the mean-subtracted cardiac EGM strip based on the moving signal envelope (706). For example, in one example, let d indicate the waveform, dU indicate the upper envelope, and dL indicate the lower envelope (e.g., as Fig. 8A In this example, the normalized waveform can be sampled and calculated as d / (dU-dL). Figure 8B is a conceptual diagram showing an example normalized waveform generated according to one or more techniques of the present disclosure. Figure 8B In the waveform, the QRS amplitude variation is minimized. Figure 7 In an example, the AI ​​system may provide normalized mean-subtracted cardiac EGM strips as input to the deep learning model 454.

[0103] In other examples, the pre-processing unit 452 can perform additional pre-processing actions. For example, the pre-processing unit 452 can normalize the entire waveform by its standard deviation, normalize the waveform by a constant gain factor, and / or perform other actions.

[0104] Fig. 9is a flowchart showing an example operation of an AI system according to one or more techniques of the present disclosure. The above examples relate to deep learning models applied only to cardiac EGM waveforms. Because some devices (e.g., medical device 16) have limited memory and battery power, the device stores and transmits limited waveform segments. However, such devices can also generate and store additional marker channel information. The marker channel information can correspond to the time period during and before the time period corresponding to the recorded cardiac EGM waveform. The marker channel information can provide additional arrhythmia diagnosis. For example, in one instance, a marker channel can indicate a detected QRS complex wave. Therefore, in this example, the marker channel can include a series of samples, each of which indicates whether a QRS complex wave is detected during the time period corresponding to the sample. Based on this data, the RR rate of the cardiac rhythm of patient 14 can be determined. Fig. 9 The example involves using marker channels to confirm and expand the deep learning model 454 for arrhythmia detection.

[0105] like Fig. 9 As shown in the example of , the AI ​​system uses a marker channel to confirm and extend the arrhythmia detection of the deep learning model 454. The waveform channel (e.g., cardiac EGM strip) and the marker channel can be processed by two separate models: A.) a waveform-based deep learning model 454 and B.) a marker channel-based model. If the arrhythmia detected by the two models is consistent during the duration of the waveform, the AI ​​system will increase the confidence of the arrhythmia detection and extend the time course of the arrhythmia detection to just beyond the recorded waveform duration. In other words, the AI ​​system can cause the device that generates the waveform channel and the marker channel to continue to generate the waveform channel. However, if the two models are inconsistent, the AI ​​system can adjust the confidence of the arrhythmia detection (e.g., if the marker-based model has high reliability), but will not extend the time course of the arrhythmia detection. In other words, the AI ​​system can cause the device that generates the waveform channel and the marker channel to stop generating the waveform channel. Not generating the waveform channel can reduce the power consumption of the device.

[0106] exist Fig. 9 In an example of , the AI ​​system can obtain one or more cardiac EGM strips and a marker channel (900). In some examples, the marker channel indicates a QRS complex wave detected in the cardiac rhythm of the patient 14. The pre-processing unit 452 can resample and normalize the one or more cardiac EGM strips, for example, as described with respect to Figure 7 Additionally, the AI ​​system may apply the deep learning model 454 (i.e., a waveform-based deep learning model) to one or more pre-processed cardiac EGM strips to generate a first set of arrhythmia data (904). The AI ​​system may use the deep learning model 454 (i.e., a waveform-based deep learning model) to generate a first set of arrhythmia data (904). Figure 5 The deep learning model 454 is applied in the same manner as described.

[0107] In addition, Fig. 9 In an example of, the preprocessing unit 452 can preprocess the marker channel (906). For example, the preprocessing unit 452 can determine the RR interval, the average heart rate, one or more moving average heart rate trends, outlier removal, the marker-based Lorenz plot, heart rate variability, or other types of data based on the marker channel. Additionally, the preprocessing unit 452 can apply a deep learning model based on the marker channel to the preprocessed marker channel to generate second arrhythmia data indicating a second set of occurrences of one or more arrhythmias (908). In this example, the deep learning model based on the marker channel can be implemented as a neural network that is trained to identify the occurrence of arrhythmias in the preprocessed marker channel.

[0108] exist Fig. 9 In an example, the AI ​​system can determine whether the first set of arrhythmia data and the second set of arrhythmia data are consistent (910). For example, the AI ​​system can check whether there are any detected arrhythmia events that are not in the combination of the first set and the second set of arrhythmia data. In addition, the AI ​​system can adjust the confidence level of the occurrence of one of the arrhythmias based on whether the occurrence of the arrhythmia is in both the first set of occurrences and the second set of occurrences or only in one of the first set of occurrences and the second set of occurrences. For example, if the first set of arrhythmia data and the second set of arrhythmia data are consistent with the detected arrhythmia occurrence (the "yes" branch of 910), the AI ​​system can increase the confidence level of the detected arrhythmia occurrence and expand the arrhythmia detection (912). For example, the AI ​​system can cause the medical device 16 to continue to generate a cardiac EGM strip. On the other hand, if the first set of arrhythmia data and the second set of arrhythmia data are inconsistent with the detected arrhythmia occurrence (the "NO" branch of 910), the AI ​​system can adjust the confidence level of the detected arrhythmia occurrence and do not expand the arrhythmia detection (914). For example, the AI ​​system may cause the medical device 16 to stop generating cardiac EGM strips. Thus, the AI ​​system may perform one or more of the following: adjusting a confidence level of the occurrence of one of the arrhythmias based on whether the occurrence of the arrhythmia is in both the first group of occurrences and the second group of occurrences or only in one of the first group of occurrences and the second group of occurrences, a duration of a monitoring session of the medical device generating one or more cardiac EGM strips.

[0109] The confidence level can be used in various ways. For example, the monitoring system 450 can output the confidence level for display. In this example, displaying the confidence level can help a monitoring professional or a healthcare professional determine how to act on the information about the arrhythmia. In some examples, if the confidence level is below a predetermined threshold, the monitoring system 450 does not present the information about the arrhythmia to the user.

[0110] Additionally, in some instances, the medical device 16 generates a cardiac EGM strip of limited duration (e.g., 45 seconds) upon detection of an arrhythmia or patient-triggered occurrence. The diagnostic device has a patient trigger (e.g., a button on the device) that the patient can press when they feel symptoms. These are considered symptomatic ECG episodes. Around the time of patient activation / triggering, the device captures an extended waveform signal that is transmitted for review. Additionally, in such instances, the medical device 16 generates marker channels for a more extended duration (e.g., 5 minutes before and 5 minutes after the detection of an arrhythmia or patient-triggered occurrence). Fig. 9 In such instances, the AI ​​system may pre-process the cardiac EGM strip (902) and apply the deep learning model 454 to the pre-processed cardiac EGM strip to generate arrhythmia data (904). Further, in such instances, the AI ​​system may apply a marker channel-based model to a portion of the marker channel marker corresponding to a time before and during a time corresponding to the cardiac EGM strip (908). In such instances, the AI ​​system may determine whether the outputs of the waveform-based deep learning model 454 and the marker channel-based deep learning model are consistent. For example, if the deep learning model 454 determines that the waveform of the cardiac EGM strip contains an occurrence of ventricular tachycardia and the marker-based deep learning model detects a high rate episode with low RR variability, the AI ​​system may determine that the output of the waveform-based deep learning model 454 is consistent with the output of the marker channel-based deep learning model.

[0111] Based on whether the outputs of the two models are consistent, the AI ​​system can adjust the likelihood of an overall arrhythmia occurring in the waveform. For example, if there is consistency between the marker channel-based model and the waveform-based deep learning model 454 (e.g., the marker channel detected a tachyarrhythmia greater than or equal to 120 beats per minute (BPM) for the duration of the recorded waveform, and the waveform-based deep learning model 45 detected ventricular tachycardia (VT)), the monitoring system 450 can use the trend of the marker channel to present information about what happened before the onset of the arrhythmia detected in the cardiac EGM strip. For example, the monitoring system 450 can output a graph showing how the patient's heart performed before the onset of an arrhythmia, such as Fig. 10BThis can help physicians diagnose the patient's 14 heart condition.

[0112] In one example, the medical device 16 can generate one or more cardiac EGM strips representing a waveform of limited duration in response to an onboard detection algorithm that determines that an arrhythmia is occurring. The medical device 16 itself can implement the onboard detection algorithm. In one example, the onboard detection algorithm can determine that a 45-second tachycardia episode occurred from 10:30:00 am to 10:30:45 am. In this example, the medical device 16 can provide one or more cardiac EGM strips representing a 45-second waveform to the AI ​​system to report the arrhythmia. In addition, in this example, the AI ​​system can apply a waveform-based deep learning model 454 to determine whether one or more cardiac EGM strips contain any arrhythmias during the 45-second time period (i.e., from 10:30:00 am to 10:30:45 am). The monitoring system 450 can report any arrhythmias detected by the deep learning model 454 based on one or more cardiac EGM strips. Additionally, in some instances, the medical device 16 may generate a marker channel covering a longer duration (e.g., from 10:20:00 a.m. to 10:35:00 a.m.). The medical device 16 may also provide this marker channel to the AI ​​system for arrhythmia reporting. The AI ​​system's marker channel-based model may use the marker channel to detect arrhythmias during and adjacent to the time period corresponding to one or more cardiac EGM strips (e.g., 10:29:00 a.m. to 10:31:00 a.m.). If the arrhythmia detected by the marker channel-based model during and adjacent to the time period corresponding to one or more cardiac EGM strips is consistent with the arrhythmia detected by the waveform-based deep learning model 454, the monitoring system 450 may output an indication of the arrhythmia detected by the waveform-based deep learning model 454 based on one or more cardiac EGM strips in the time period from 10:30:00 a.m. to 10:30:45 a.m. Additionally, the monitoring system 450 can output additional information from the marker channel (from 10:20:00 AM to 10:35:00 AM), such as the average HR and HR variability trend before, during, and after the recorded waveform duration. In this way, the monitoring system 450 can output information that provides additional arrhythmia-related information to the physician. For example, the physician may be able to see if there is a sudden onset of tachycardia.

[0113] In some examples, if the medical device 16 implements both the waveform-based deep learning model 454 and the marker channel-based model, the medical device 16 can determine whether there is consistency between the arrhythmias identified by the waveform-based deep learning model 454 and the marker channel-based model over a period of time, and the medical device 16 can keep recording the waveform for a longer duration. If the medical device 16 determines that there is no consistency, the medical device 16 can stop recording / delete this waveform segment.

[0114] Fig. 10A is a conceptual diagram showing a graph of example cardiac waveforms and QRS markers detected by a device. Fig. 10B is a conceptual diagram showing an example of a plot of heart rate over time based on a marker channel. Fig. 10A and Fig. 10B For the same data and aligned in time on the x-axis. Fig. 10A In the example of , the marker channel information is shown as a series of points arranged on a horizontal line. The dark jagged line starting around the 95th second is a waveform represented by one or more cardiac EGM strips. Fig. 10A In the example, the markers are stored on the medical device 16 starting from the time period before the finite waveform segment. Generating and storing the marker data may require fewer computing resources than a cardiac EGM strip. Fig. 10B As shown in the example, the waveform-based deep learning model and the marker channel-based deep learning model agree in detecting the occurrence of ventricular tachycardia (VT) at approximately the 130-second mark, and the time course of arrhythmia detection extends from only a 120-130-second segment (within the waveform duration) to approximately 2 minutes before the onset of the arrhythmia; this indicates a rapid increase in HR and VT events.

[0115] Although other examples described in the present disclosure use the deep learning model 452 directly on the preprocessed cardiac EGM strips, in other examples, the AI ​​system can perform deep learning QRS and arrhythmia detection on the transformed signal (e.g., in the frequency domain). For example, in one example, the neural network can have several initial convolutional layers to extract low-level features of the input signal. In some examples, the AI ​​system can learn the convolution kernel / filter parameters directly from the data. In some examples, in order to reduce computational complexity while maintaining performance, the deep learning data can be preprocessed using existing expert knowledge. For example, as part of preprocessing one or more cardiac EGM strips, the AI ​​system can decompose the waveform identified by one or more cardiac EGM strips into decomposed waveforms corresponding to different frequency bands. Different frequency bands can correspond to different aspects of the ECG signal, such as high-frequency QRS signals, low-frequency P wave signals, etc. In this example, the AI ​​system can generate arrhythmia data by applying the deep learning model 452 to the decomposed waveform.

[0116] Fig.11 is a flow chart illustrating a third example operation for pre-processing a cardiac EGM strip in accordance with one or more techniques of the present disclosure. Fig.11 In the example of FIG. 1 , as part of preprocessing one or more cardiac EGM strips, the preprocessing unit 452 can scale the waveform represented by the one or more cardiac EGM strips ( 1100 ). Scaling in this example and other parts of the present disclosure can be used to limit the range of EGM amplitude scaling on which the model needs to be trained. Additionally, in Fig.11 In an example of , the preprocessing unit 452 may decompose a waveform represented by one or more cardiac EGM strips into a plurality of channels corresponding to different frequency bands (1002). Subsequently, as part of applying the deep learning model 454 to the segments of the preprocessed cardiac EGM strips, the AI ​​system may apply the deep learning model to the channels to generate arrhythmia data.

[0117] Therefore, in Fig.11 In the example, after waveform scaling is performed, the signal is decomposed into multiple channels, where each channel corresponds to a different frequency band. Fig.12 is a conceptual diagram illustrating an example original cardiac waveform and waveforms of channels corresponding to different frequency bands, in accordance with one or more techniques of the present disclosure. Fig.12 An example is shown in which, instead of the original waveform 1200 for deep learning, 3 channels (1202, 1204, and 1206) derived from the original waveform 1200 using (i) stationary wavelet decomposition and (ii) band-specific time delays for QRS alignment are used. It should be noted that channel 1202 is primarily composed of high-frequency features (typically corresponding to the QRS segment) and band 1206 is composed of low-frequency features (e.g., p waves). Therefore, instead of using a 1x N vector as input to the deep learning model 454, a 3x N matrix (with stationary wavelet decomposition and aligned signals) is used as input. Having a preprocessed signal can help accelerate learning and / or reduce the complexity of the deep learning model 454.

[0118] Fig.13 is a flow chart illustrating a fourth example operation for pre-processing a cardiac EGM strip in accordance with one or more techniques of the present disclosure. Fig.13 As an example, as part of preprocessing one or more cardiac EGM strips, the preprocessing unit 452 can scale the waveform represented by the one or more cardiac EGM strips (1300). The preprocessing unit 452 can scale the waveform represented by the one or more cardiac EGM strips (1300). Fig.11 Use any of the provided instances to scale the waveform.

[0119] In addition, Fig.13In an example, the pre-processing unit 452 may generate one or more transformed cardiac EGM strips (1302) by transforming the scaled waveform into a two-dimensional time-frequency domain. For example, the pre-processing unit 452 may transform the scaled waveform into a spectrogram. Subsequently, the AI ​​system may apply the deep learning model 454 to one or more pre-processed cardiac EGM strips by applying the deep learning model to the one or more transformed cardiac EGM strips. The monitoring system 450 may output an image of the spectrogram for display. In addition, in some examples, the deep learning model 454 may be implemented in a manner similar to an image recognition deep learning model. For example, the deep learning model 454 may include a convolutional layer that applies individual filters to the spectrogram. In addition, in some examples, by presenting the entire spectrogram to the deep learning model 454, the AI ​​system may enable the deep learning model 454 to determine a set of optimal frequencies and features for this group of signals.

[0120] Fig.14 is a conceptual diagram illustrating an example operation of an AI system according to one or more techniques of this disclosure. Fig.14 In the example of , the preprocessing unit 452 can preprocess the cardiac EGM strip (1400). For example, in Fig.14 In an example of, the pre-processing unit 452 can apply the learned scaling factor to the waveform represented by the cardiac EGM strip. In some examples, the AI ​​system can learn the scaling factor by repeatedly testing different scaling factors with training data and determining which scaling factors result in the best performance in correctly detecting arrhythmias.

[0121] The AI ​​system may provide pre-processed cardiac EGM strips as input to the deep neural network 1402. The deep neural network 1402 may be part of the deep learning model 454.

[0122] exist Fig.14 In an example, as part of applying the deep learning model 454, the AI ​​system can perform QRS detection (1402). As part of performing QRS detection, the deep neural network 1402 can determine QRS probability values ​​corresponding to multiple time points. Each of the QRS probability values ​​indicates a respective probability that a peak of a QRS complex wave occurs during a time point corresponding to the QRS probability value. For example, a set of initial layers of the deep neural network can determine the QRS probability values ​​based on preprocessed EGM strips.

[0123] In addition, the deep neural network 1402 can perform arrhythmia classification (1406) based on the preprocessed cardiac EGM strips and the QRS probability values. That is, the deep neural network 1402 can detect a set of occurrences of one or more arrhythmias by providing the QRS probability values ​​and the preprocessed cardiac EGM strips as inputs to the neural network that generates arrhythmia data. Therefore, in this step, the QRS probability values ​​are combined with deeper layers of the deep learning model 454 to detect arrhythmia events.

[0124] like Fig.14 As shown in the example, the AI ​​system can output QRS probability values ​​and occurrence times, respectively, and arrhythmia data indicating whether and where the cardiac EGM strip represents one or more occurrences of one or more arrhythmias. This can provide the advantage of providing two separate paths for the QRS, and arrhythmia detection is interpretable because ECG readers / technicians trained in the field typically use the QRS as the first step in arrhythmia determination. In addition, the 2-way interaction between the QRS and the arrhythmia detection module can help provide arrhythmia-specific QRS detection and enhanced QRS detection based on the occurrence of detected arrhythmias (for example, if the QRS model marks PVC / bigeminy but is not sure which, and if the arrhythmia model claims its PVC, then the rhythm is bigeminy, otherwise it is T wave oversensing).

[0125] Fig.15 is a flow chart illustrating a fifth example operation for pre-processing a cardiac EGM strip in accordance with one or more techniques of the present disclosure. Fig.15 In an example, the operation includes signal polarity and signal characteristic preprocessing according to one or more techniques of the present disclosure. In some examples, the deep learning model 454 is trained with data from a certain type of hardware and cardiac EGM signal characteristics. For example, a user may want to use a large data set to train the deep learning model 454. The data set may contain "upright" waveform morphology, but the deep learning model 454 may also need to be used with devices where the waveform morphology is not always upright. For example, some devices generate cardiac EGM signals in which the R wave is initially deflected in the negative direction, and other devices may generate cardiac EGM signals in which the R wave is initially deflected in the positive direction. To address such situations, the deep learning model 454 can be trained on a set of original waveforms and their polarity-reversed versions, both of which have the same arrhythmia content. In cases where the AI ​​system needs to use a pre-existing deep learning model (i.e., where retraining is not possible), the waveform signal can be transformed to meet the input characteristics of the deep learning.

[0126] Therefore, in Fig.15In an example of , the AI ​​system may determine the polarity of the cardiac EGM strip (1500). The AI ​​system may then determine whether the polarity of the cardiac EGM strip is the same as the expected polarity of the deep learning model 454 (1502). In response to determining that the polarity of the cardiac EGM strip is not the expected polarity of the deep learning model 454 ("No" of 1502), the AI ​​system may invert the polarity of the cardiac EGM strip (1504). Otherwise, the AI ​​system does not invert the polarity of the cardiac EGM strip (1506).

[0127] The AI ​​system may determine the polarity of the cardiac EGM strip in one of a variety of ways. For example, in one example, when the medical device 16 is implanted in the patient 14, the implanting physician may program a setting to indicate whether the polarity is reversed. In this example, the medical device 16 may include data in the cardiac EGM strip generated by the medical device 16 indicating the polarity of the cardiac EGM strip generated by the medical device 16.

[0128] In another example, when viewing and analyzing data from a medical device 16, such as a cardiac EGM strip, at a monitoring center, if the waveform morphology of the patient 14 is reversed, the monitoring center technician may flag the data. For short- to medium-term monitoring, such flagging may only need to be performed at the beginning of monitoring. For long-term applications, such morphology flagging may be performed periodically (e.g., every month) to account for any device drift.

[0129] In another example of determining the polarity of a cardiac EGM strip, the AI ​​system can use the P-wave and T-wave morphology to estimate whether the waveform morphology is flipped. That is, the P-wave and T-wave always deflect in the same direction at the beginning and end of the cardiac cycle. Therefore, based on the initial detection of the P-wave and T-wave, the AI ​​system can determine the polarity of the cardiac EGM strip.

[0130] In another example of determining the polarity of a cardiac EGM strip, the AI ​​system can use a deep learning model to detect waveform polarity as a precursor to applying a deep learning model (e.g., deep learning model 454) for arrhythmia detection. Thus, in this example, deep learning model 454 can be considered a first deep learning model and the AI ​​system can apply additional deep learning models to the cardiac EGM strip to determine the polarity of the cardiac EGM strip. Additional deep learning models can include an artificial neural network that is trained to classify the polarity of the cardiac EGM strip.

[0131] In a related example, the AI ​​system may use a deep learning similarity model to check whether the input signal morphology is similar to the signal morphology required for the arrhythmia detection model and performing appropriate signal transformations. The deep learning similarity model may take multiple waveforms as input and may generate output data indicating whether the waveforms are similar. The waveform in this example may be a waveform represented by a cardiac EGM strip. In this example, "similarity" may refer to the same morphology / polarity. In other words, the deep learning similarity model may compare waveforms of cardiac EGM strips from multiple devices to determine whether the waveforms have similar morphologies. The deep learning similarity model may be implemented as a neural network.

[0132] In addition to or as an alternative to the polarity of the cardiac EGM strip, the AI ​​system may modify the properties of the cardiac EGM strip. For example, the AI ​​system may generate device classification data indicating the category of the device that generates the cardiac EGM strip. Devices of different categories may have different hardware characteristics (e.g., the bandwidth of the input signal). Therefore, it may be advantageous to refilter and / or transform the signal of the cardiac EGM strip to match the input characteristics of the deep learning model 454 before applying the deep learning model to the cardiac EGM strip. For example, the AI ​​system may filter the signal of the cardiac EGM strip to change the bandwidth of the signal. For example, in this example, the AI ​​system may scale the signal samples of the cardiac EGM strip so that the sample distribution is within the expected sample value range of the deep learning model 454. In some examples, as part of filtering the signal of the cardiac EGM strip to change the bandwidth of the signal, the AI ​​system may increase or decrease the sampling rate of the signal of the cardiac EGM strip to match the expected sampling rate of the deep learning model 454. The AI ​​system may use interpolation to adjust the sampling rate. Due to the different characteristics of different categories of devices, the deep learning model 454 may not correctly identify the occurrence of arrhythmias in cardiac EGM strips generated by multiple types of devices.

[0133] Fig.16 is a conceptual diagram illustrating an example operation of an AI system including an autoencoder according to one or more techniques of the present disclosure. Fig.16 In an example of , the AI ​​system uses auto-encoding to determine the device classification. In other words, the AI ​​system can use auto-encoding to check whether the classification data assigned to the cardiac EGM strip is correct. The classification data assigned to the cardiac EGM strip may include data generated by another device (e.g., medical device 16) indicating an arrhythmia detected in the cardiac EGM strip.

[0134] like Fig.16As shown in the example of , the input to the AI ​​system can include one or more preprocessed cardiac EGM strips 1600 and classification data 1602. The preprocessed cardiac EGM strips 1600 represent the waveform of the cardiac rhythm of the patient 14. The preprocessed cardiac EGM strips 1600 can be preprocessed in one or more of a variety of ways, including according to any of the examples provided elsewhere in this disclosure. The classification data 1602 can include data indicating the class of arrhythmia detected in the cardiac EGM strip, and the preprocessed cardiac EGM strips 1600 are based on the data. For example, the classification data 1602 can indicate that the medical device 16 has determined that the cardiac EGM strip contains an occurrence of atrial fibrillation.

[0135] The autoencoder 1604 classifies the probability that the classified data 1602 is correct. The autoencoder 1604 can be implemented as a deep neural network. The deep neural network of the autoencoder 1604 includes an input layer 1604, a set of hidden layers, and an output layer. The autoencoder 1604 is trained so that when a preprocessed cardiac EGM strip and a set of classified data are provided as input to the autoencoder 1604, the output layer outputs a reconstructed version of the preprocessed cardiac EGM strip. The autoencoder 1604 can be trained according to one of the various techniques known in the art for training autoencoders. For example, the AI ​​system may compare the reconstructed version of the preprocessed cardiac EGM strip with the original preprocessed cardiac EGM strip 1600 to determine an error value that can be used in a back-propagation algorithm to update the parameters of the deep neural network of the autoencoder 1604. Fig.16 As shown in the example, the AI ​​system can provide one or more of the preprocessed cardiac EGM strip 1600 or segments thereof and device classification data 1602 to the input layer 1606 of the deep neural network of the autoencoder 1604.

[0136] The output of one of the hidden layers of the deep neural network of the autoencoder 1604 can generate probability values ​​1606. Each of the probability values ​​corresponds to a different class of arrhythmias and indicates a confidence level that the classified data 1602 correctly identifies the class of arrhythmias in the cardiac EGM strip. In some examples, to speed up the training of the deep neural network of the autoencoder 1604, the layers up to the middle layers are pre-trained separately from the subsequent layers of the deep neural network of the autoencoder 1604.

[0137] In addition, Fig.16In an example of the present invention, the AI ​​system may use probability values ​​1606 to determine whether classification data 1602 correctly identifies an arrhythmia in the cardiac EGM strip. For example, the AI ​​system may identify the highest probability value in probability values ​​1606 and compare the arrhythmia corresponding to the highest probability value in probability values ​​1606 with the arrhythmia indicated by classification data 1602 to determine whether classification data 1602 indicates the arrhythmia corresponding to the highest probability value in probability values ​​1606.

[0138] Fig.17 is a flow chart illustrating example operations in which an autoencoder is used to confirm device classification in accordance with one or more techniques of the present disclosure. Fig.17 As shown in the example and with Fig.16 Consistent with the example of , the AI ​​system can obtain training input vectors (1700). Each of the training input vectors includes a segment of a training cardiac EGM strip and device classification data indicating one or more arrhythmias detected in the training cardiac EGM strip. The AI ​​system can train an autoencoder 1604 based on the training input vectors to reconstruct the training cardiac EGM strips (1702) of the training input vectors.

[0139] In addition, Fig.17 In an example, the AI ​​system may obtain additional classification data (1704). For example, in one example, the AI ​​system may receive additional classification data from a medical device such as medical device 16.

[0140] The AI ​​system may provide the segments of the preprocessed cardiac EGM strip and the classification data (1706) to the input layer 1606 of the autoencoder 1604. In addition, the AI ​​system may determine whether the classification data correctly identifies an arrhythmia, if any, in the preprocessed cardiac EGM strip based on probability values ​​generated by the intermediate layers of the autoencoder 1604 (1708). Each of the probability values ​​corresponds to a different arrhythmia and indicates a confidence level that the preprocessed cardiac EGM strip contains the arrhythmia.

[0141] The following is a set of non-limiting examples according to one or more techniques of the present disclosure.

[0142] Example 1. A method comprising: obtaining, by a computing system, one or more cardiac electrogram (EGM) strips representing a waveform of a patient's cardiac rhythm; preprocessing, by the computing system, the one or more cardiac EGM strips; and applying, by the computing system, a deep learning model to the one or more preprocessed cardiac EGM strips to generate arrhythmia data indicating whether the one or more cardiac EGM strips represent one or more occurrences of one or more arrhythmias.

[0143] Example 2. A method according to Example 1, wherein preprocessing the one or more cardiac EGM strips includes: generating, by the computing system, one or more time-modified cardiac EGM strips by modifying the time resolution of the one or more cardiac EGM strips to match the expected time resolution of the deep learning model; and generating, by the computing system, one or more preprocessed cardiac EGM strips by subtracting the average value of samples of the one or more time-modified cardiac EGM strips from the samples of the one or more time-modified cardiac EGM strips.

[0144] Example 3. The method of example 1, wherein preprocessing the one or more cardiac EGM strips comprises:

[0145] The computing system generates one or more time-modified cardiac EGM strips by modifying the time resolution of the one or more cardiac EGM strips to match the expected time resolution of the deep learning model; the computing system generates one or more mean-subtracted cardiac EGM strips by subtracting the mean of samples of the one or more time-modified cardiac EGM strips from the samples of the one or more time-modified cardiac EGM strips; the computing system calculates a moving signal envelope on a sliding window of the one or more mean-subtracted cardiac EGM strips; and the computing system normalizes the one or more mean-subtracted cardiac EGM strips based on the moving signal envelope.

[0146] Example 4. A method according to any one of Examples 1 to 3, wherein: the arrhythmia data is first arrhythmia data; the method further includes: obtaining a marker channel by the computing system; preprocessing the marker channel by the computing system; applying a deep learning model based on the marker channel to the preprocessed marker channel by the computing system to generate second arrhythmia data indicating a second group of occurrences of the one or more arrhythmias; determining by the computing system whether the first arrhythmia data is consistent with the second arrhythmia data; and at least one of the following: adjusting by the computing system a confidence level of the occurrence of one of the arrhythmias based on whether the occurrence of the arrhythmia is in both the first group of occurrences and the second group of occurrences or only in one of the first group of occurrences and the second group of occurrences; or extending by the computing system the duration of a monitoring session of a medical device that generates the one or more cardiac EGM strips based on the occurrence of the arrhythmia being in both the first group of occurrences and the second group of occurrences.

[0147] Example 5. The method of Example 4, wherein the marker channel indicates a detected QRS complex.

[0148] Example 6. A method according to any one of Examples 1 to 5, wherein: preprocessing the one or more cardiac EGM strips includes: scaling the waveform represented by the one or more cardiac EGM strips by the computing system; and decomposing the waveform represented by the one or more cardiac EGM strips into multiple channels corresponding to different frequency bands by the computing system, and applying the deep learning model to the one or more preprocessed cardiac EGM strips includes applying the deep learning model to the channels by the computing system to generate the arrhythmia data.

[0149] Example 7. A method according to any one of Examples 1 to 6, wherein: preprocessing the one or more cardiac EGM strips includes: scaling the waveform represented by the one or more cardiac EGM strips by the computing system; generating a transformed waveform by the computing system by transforming the scaled waveform into a two-dimensional time-frequency domain, and applying the deep learning model to the one or more preprocessed cardiac EGM strips includes applying the deep learning model to the transformed waveform.

[0150] Example 8. A method according to any one of Examples 1 to 7, wherein: preprocessing the one or more cardiac EGM strips includes applying, by the computing system, a learned scaling factor to the waveform represented by the one or more cardiac EGM strips, and applying the deep learning model includes: determining, by the computing system, QRS probability values ​​corresponding to the multiple time points, each of the QRS probability values ​​indicating a corresponding probability that a peak of a QRS complex wave occurs during the time point corresponding to the QRS probability value; and detecting, by the computing system, the set of occurrences of the one or more arrhythmias by providing the QRS probability values ​​and the one or more preprocessed cardiac EGM strips as input to a neural network that generates the arrhythmia data.

[0151] Example 9. A method according to any one of Examples 1 to 8, wherein preprocessing the one or more cardiac EGM strips includes: determining, by the computing system, the polarity of the one or more cardiac EGM strips; and reversing, by the computing system, the polarity of the one or more cardiac EGM strips based on the polarity of the one or more cardiac EGM strips being not the expected polarity of the deep learning model.

[0152] Example 10. A method according to Example 9, wherein: the deep learning model is a first deep learning model, and determining the polarity of the one or more cardiac EGM bands includes applying a second deep learning model by the computing system to the one or more cardiac EGM bands to determine the polarity of the one or more cardiac EGM bands.

[0153] Example 11. A method according to any one of Examples 1 to 10, wherein the method further includes: obtaining training input vectors by the computing system, wherein each of the training input vectors includes a segment of a training cardiac EGM strip and device classification data indicating one or more arrhythmias detected in the training cardiac EGM strip; training an autoencoder by the computing system based on the training input vectors to reconstruct the training cardiac EGM strips of the training input vectors; obtaining additional device classification data by the computing system; providing the one or more preprocessed cardiac EGM strips and the additional device classification data to an input layer of the autoencoder by the computing system; and determining by the computing system whether the classification data correctly identifies the arrhythmia in the preprocessed cardiac EGM strip, if any, based on probability values ​​generated by an intermediate layer of the autoencoder, wherein each of the probability values ​​corresponds to a different arrhythmia and indicates a confidence level that the one or more preprocessed cardiac EGM strips contain the arrhythmia.

[0154] Example 12. A method according to any one of Examples 1 to 11, wherein preprocessing the one or more cardiac EGM strips includes one or more of the following: scaling samples of the signals of the one or more cardiac EGM strips so that the samples of the signals of the one or more cardiac EGM strips are distributed within the expected sample value range of the deep learning model, or increasing or decreasing the sampling rate of the signals of the one or more cardiac EGM strips to match the expected sampling rate of the deep learning model.

[0155] Example 13. A computing system comprising: a storage device configured to store one or more cardiac electrogram (EGM) strips representing a waveform of a patient's cardiac rhythm; one or more processing circuits configured to: pre-process the one or more cardiac EGM strips; and apply a deep learning model to the one or more pre-processed cardiac EGM strips to generate arrhythmia data indicating whether the one or more cardiac EGM strips represent one or more occurrences of one or more arrhythmias.

[0156] Example 14. The computing system of example 13, further configured to perform the method of any one of examples 2 to 12.

[0157] Example 15. A computer-readable storage medium having instructions stored thereon, which, when executed, cause a computing system to perform the method according to any one of Examples 1 to 12.

[0158] Example 16. A method as described in the specification.

[0159] In some examples, the technology of the present disclosure includes a system including means for performing any of the methods described herein. In some examples, the technology of the present disclosure includes a computer-readable medium including instructions for causing a processing circuit system to perform any of the methods described herein.

[0160] It should be understood that the various aspects and examples disclosed herein may be combined in different combinations other than those specifically presented in the specification and drawings. It should also be understood that, depending on the example, the actions or events of any process or method described herein may be performed in a different order, may be added, merged, or excluded entirely (e.g., all described actions and events may not be necessary for performing the technology). In addition, although certain aspects of the present disclosure are described as being performed by a single module, unit, or circuit for the sake of clarity, it should be understood that the technology of the present disclosure may be performed by a combination of units, modules, or circuit systems associated with, for example, a medical device.

[0161] In one or more instances, the described techniques may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions may be stored in the form of one or more instructions or codes on a computer-readable medium and may be executed by a hardware-based processing unit. A computer-readable medium may include a non-transitory computer-readable medium, which corresponds 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).

[0162] 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 terms "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. Moreover, the techniques may be fully implemented in one or more circuits or logic elements.

[0163] Various examples have been described. These and other examples are within the scope of the following claims.

Claims

1. A method for arrhythmia detection, comprising: obtaining, by a computing system, training input vectors, wherein each of the training input vectors comprises a segment of a training cardiac EGM strip and device classification data indicative of one or more arrhythmias detected in the training cardiac EGM strip; training, by the computing system, an autoencoder based on the training input vector to reconstruct a training cardiac EGM strip of the training input vector; obtaining, by the computing system, one or more electrocardiogram (EGM) strips representing a waveform of a cardiac rhythm of the patient and additional device classification data; preprocessing, by the computing system, the one or more cardiac EGM strips; applying, by the computing system, the deep learning model to one or more preprocessed cardiac EGM strips to generate arrhythmia data indicating whether the one or more cardiac EGM strips represent one or more occurrences of one or more cardiac arrhythmias; providing, by the computing system, the one or more pre-processed cardiac EGM strips and the additional device classification data to an input layer of the autoencoder; as well as The computing system determines whether the classification data correctly identifies an arrhythmia in the preprocessed cardiac EGM strips, if any, based on probability values ​​generated by an intermediate layer of the autoencoder, wherein each of the probability values ​​corresponds to a different arrhythmia and indicates a confidence level that the one or more preprocessed cardiac EGM strips contain the arrhythmia.

2. The method of claim 1, wherein pre-processing the one or more cardiac EGM strips comprises: generating, by the computing system, one or more temporally modified cardiac EGM strips by modifying a temporal resolution of the one or more cardiac EGM strips to match an expected temporal resolution of a deep learning model; as well as One or more pre-processed cardiac EGM strips are generated by the computing system by subtracting an average of the samples of the one or more time-modified cardiac EGM strips from the samples of the one or more time-modified cardiac EGM strips.

3. The method of claim 1 , wherein pre-processing the one or more cardiac EGM strips comprises: generating, by the computing system, one or more time-modified cardiac EGM strips by modifying a temporal resolution of the one or more cardiac EGM strips to match an expected temporal resolution of the deep learning model; generating, by the computing system, one or more mean-subtracted cardiac EGM strips by subtracting a mean of the samples of the one or more time-modified cardiac EGM strips from the samples of the one or more time-modified cardiac EGM strips; calculating, by the computing system, a moving signal envelope over a sliding window of the one or more mean-subtracted cardiac EGM strips; as well as The one or more mean-subtracted cardiac EGM strips are normalized by the computing system based on the movement signal envelope.

4. The method according to any one of claims 1 to 3, wherein: The arrhythmia data is first arrhythmia data, the one or more occurrences of the one or more arrhythmias being a first set of occurrences of the one or more arrhythmias; The method further comprises: Obtaining a marker channel by the computing system; Preprocessing the marker channel by the computing system; applying, by the computing system, the marker channel-based deep learning model to the preprocessed marker channels to generate second arrhythmia data indicative of a second set of occurrences of the one or more arrhythmias; determining, by the computing system, whether the first arrhythmia data is consistent with the second arrhythmia data; and At least one of the following: adjusting, by the computing system, a confidence level for the occurrence of one of the arrhythmias based on whether the occurrence of the arrhythmia is in both the first group of occurrences and the second group of occurrences or in only one of the first group of occurrences and the second group of occurrences; or A duration of a monitoring session of a medical device generating the one or more cardiac EGM strips is extended by the computing system based on the occurrence of the arrhythmia being in both the first group of occurrences and the second group of occurrences.

5. The method of claim 4, wherein the marker channel indicates a detected QRS complex.

6. The method according to any one of claims 1 to 3, wherein: Preprocessing the one or more cardiac EGM strips comprises: scaling, by the computing system, a waveform represented by the one or more cardiac EGM strips; and decomposing, by the computing system, a waveform represented by the one or more cardiac EGM strips into a plurality of channels corresponding to different frequency bands, and Applying the deep learning model to the one or more preprocessed cardiac EGM strips includes applying, by the computing system, the deep learning model to the channels to generate the arrhythmia data.

7. The method according to any one of claims 1 to 3, wherein: Preprocessing the one or more cardiac EGM strips comprises: scaling, by the computing system, a waveform represented by the one or more cardiac EGM strips; and generating, by the computing system, a transformed waveform by transforming the scaled waveform into a two-dimensional time-frequency domain, and Applying the deep learning model to the one or more preprocessed cardiac EGM strips includes applying the deep learning model to the transformed waveform.

8. The method according to any one of claims 1 to 3, wherein: pre-processing the one or more cardiac EGM strips includes applying, by the computing system, a learned scaling factor to waveforms represented by the one or more cardiac EGM strips; and Applying the deep learning model includes: determining, by the computing system, QRS probability values ​​corresponding to a plurality of time points, each of the QRS probability values ​​indicating a respective probability that a peak of a QRS complex occurs during the time point corresponding to the QRS probability value; as well as The one or more occurrences of the one or more arrhythmias are detected by the computing system by providing the QRS probability value and the one or more pre-processed cardiac EGM strips as inputs to a neural network that generates the arrhythmia data.

9. The method of any one of claims 1 to 3, wherein pre-processing the one or more cardiac EGM strips comprises: determining, by the computing system, a polarity of the one or more cardiac EGM strips; as well as Based on the polarity of the one or more cardiac EGM strips not being an expected polarity by the deep learning model, the polarity of the one or more cardiac EGM strips is reversed by the computing system.

10. The method according to claim 9, wherein: The deep learning model is a first deep learning model, and Determining the polarity of the one or more cardiac EGM strips includes applying, by the computing system, a second deep learning model to the one or more cardiac EGM strips to determine the polarity of the one or more cardiac EGM strips.

11. The method of any one of claims 1 to 3, wherein pre-processing the one or more cardiac EGM strips comprises one or more of: scaling the samples of the signal of the one or more cardiac EGM strips so that the samples of the signal of the one or more cardiac EGM strips are distributed within the expected sample value range of the deep learning model, or Increase or decrease the sampling rate of the signal of the one or more cardiac EGM strips to match the expected sampling rate of the deep learning model.

12. A computing system comprising: a storage device configured to store one or more electrocardiogram (EGM) strips representing a waveform of a patient's heart rhythm and additional device classification data; One or more processing circuits configured to: obtaining training input vectors, wherein each of the training input vectors comprises a segment of a training cardiac EGM strip and device classification data indicative of one or more arrhythmias detected in the training cardiac EGM strip; training an autoencoder based on the training input vector to reconstruct a training cardiac EGM strip of the training input vector; pre-processing the one or more cardiac EGM strips; applying the deep learning model to the one or more preprocessed cardiac EGM strips to generate arrhythmia data indicating whether the one or more cardiac EGM strips represent one or more occurrences of one or more cardiac arrhythmias; providing the one or more preprocessed cardiac EGM strips and the additional device classification data to an input layer of the autoencoder; as well as Determining whether the classification data correctly identifies an arrhythmia in the preprocessed cardiac EGM strips, if any, based on probability values ​​generated by an intermediate layer of the autoencoder, wherein each of the probability values ​​corresponds to a different arrhythmia and indicates a confidence level that the one or more preprocessed cardiac EGM strips contain the arrhythmia.

13. The computing system of claim 12, wherein the one or more processing circuits are configured to perform the method of any one of claims 2 to 11.

14. A computer-readable storage medium having stored thereon instructions which, when executed, cause a computing system to: obtaining one or more electrocardiogram (EGM) strips representing a waveform of a cardiac rhythm of the patient and additional device classification data; obtaining training input vectors, wherein each of the training input vectors comprises a segment of a training cardiac EGM strip and device classification data indicative of one or more arrhythmias detected in the training cardiac EGM strip; training an autoencoder based on the training input vector to reconstruct a training cardiac EGM strip of the training input vector; pre-processing the one or more cardiac EGM strips; applying the deep learning model to the one or more preprocessed cardiac EGM strips to generate arrhythmia data indicating whether the one or more cardiac EGM strips represent one or more occurrences of one or more cardiac arrhythmias; providing the one or more preprocessed cardiac EGM strips and the additional device classification data to an input layer of the autoencoder; as well as Determining whether the classification data correctly identifies an arrhythmia in the preprocessed cardiac EGM strips, if any, based on probability values ​​generated by an intermediate layer of the autoencoder, wherein each of the probability values ​​corresponds to a different arrhythmia and indicates a confidence level that the one or more preprocessed cardiac EGM strips contain the arrhythmia.

15. The computer-readable storage medium of claim 14, wherein the instructions, when executed, cause the computing system to perform the method of any one of claims 2 to 11.

Images