Low-power machine learning system for arrhythmia detection

Abstract

The present invention discloses a technique for reducing the impact of machine learning arrhythmia detection on power consumption of a medical device using feature delineation. In one example, a medical device performs feature-based delineation of cardiac electrogram data sensed from a patient to obtain cardiac features indicative of an arrhythmia episode in the patient. The medical device determines whether the cardiac features meet a threshold criterion for applying a machine learning model to verify the feature-based delineation of the cardiac electrogram data. In response to determining that the cardiac features meet the threshold criterion, the medical device applies the machine learning model to the sensed cardiac electrogram data to verify that the arrhythmia episode has occurred or to determine a classification of the arrhythmia episode.

Description

Technical Field

[0001] The present disclosure relates generally to medical devices, and more particularly to medical devices configured to detect cardiac arrhythmias. 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. Therefore, sudden cardiac death (SCD) may occur within minutes.

[0003] In patients with a high risk of ventricular fibrillation, the use of implantable medical devices (IMDs), such as implantable cardioverter defibrillators (ICDs), has been shown to be beneficial in preventing SCD. An ICD is a battery-powered shock device that may include electrical housing electrodes (sometimes referred to as can electrodes), which are typically coupled to one or more electrical leads placed in the heart. If an arrhythmia is sensed, the ICD may send pulses through the electrical leads to shock the heart and restore its normal rhythm. Some ICDs have been configured to attempt to terminate detected tachyarrhythmias by delivering anti-tachycardia pacing (ATP) before delivering the shock. In addition, the ICD has been configured to deliver a relatively high magnitude post-shock pacing after successfully terminating a tachyarrhythmia with a shock, so as to support the heart in recovering from the shock. Some ICDs also deliver bradycardia pacing, cardiac resynchronization therapy (CRT) or other forms of pacing.

[0004] Other types of medical devices may be used for diagnostic purposes. For example, an implanted or non-implanted medical device may monitor a patient's heart. A user, such as a physician, may review data generated by a medical device for the occurrence of arrhythmias, such as atrial or ventricular tachyarrhythmias or cardiac arrest. The user may diagnose the patient's medical condition based on the identified occurrence of arrhythmias. Summary of the invention

[0005] According to the technology of the present disclosure, a medical device system is described herein, which uses feature delineation and machine learning to perform arrhythmia detection and classification. Specifically, the medical device system as described herein can use feature delineation to perform preliminary detection of arrhythmias in a patient, and only use a machine learning model to verify the arrhythmia episode detected by the feature delineation or to classify such episode detected by the feature delineation as a specific type of arrhythmia.

[0006] As described herein, feature description refers to the use of features obtained by signal processing for detecting or classifying arrhythmias. Typically, feature description involves the use of engineering rules to identify or extract features in electrocardiogram data, measure the characteristics of such features, and use measurements to detect or classify arrhythmias. For example, feature description can be used to identify features such as R waves, QRS complexes, P waves, T waves, the rate of such features, the interval between such features, feature morphology, the width or amplitude of such features, or other or other types of cardiac features or characteristics of such features that are not clearly described in this article. Feature description can include feature extraction, signal filtering, peak detection, refractory analysis, or other types of signal processing, feature engineering, or detection rule development. Feature description algorithms can be optimized for real-time, embedded, and low-power applications, such as for use in implantable medical devices. However, feature description algorithms may require expert design and feature engineering to accurately detect arrhythmias in patients.

[0007] In contrast to feature delineation techniques for arrhythmia detection and classification, machine learning techniques may be used for arrhythmia detection and classification. As described herein, machine learning refers to the use of a machine learning model, such as a neural network or a deep learning model, trained on a training data set to detect arrhythmias from electrocardiogram data. Machine learning techniques may be contrasted with feature delineation because feature delineation relies on signal processing, and the machine learning system may "learn" potential features present in electrocardiogram data that indicate the onset of an arrhythmia without requiring knowledge or understanding on behalf of the system designer of the relationship between the features and the onset of an arrhythmia.

[0008] In one example, a medical device, such as an IMD, senses electrocardiogram data of a patient. The medical device performs a feature-based delineation of the electrocardiogram data to obtain cardiac features indicative of an arrhythmia episode in the patient. The medical device determines whether the cardiac features meet threshold criteria for applying a machine learning model to validate the feature-based delineation of the electrocardiogram data. In response to determining that the cardiac features meet the threshold criteria for applying the machine learning model, the medical device applies the machine learning model to the sensed electrocardiogram data, for example, to validate that an arrhythmia episode has occurred in the patient or to detect one or more other types of arrhythmias among the arrhythmias that have occurred in the patient.

[0009] In another example, the medical device compares a first cardiac feature of the cardiac electrogram data to a cardiac feature defined by an entry of an arrhythmia dictionary. In response to determining that the first cardiac feature of the cardiac electrogram data is not similar to the cardiac feature defined by the entry of the arrhythmia dictionary, the medical device applies a machine learning model to determine a classification of an arrhythmia episode evidenced by the first cardiac feature. The medical device may store the determined arrhythmia classification and cardiac feature as a new entry in the arrhythmia dictionary in order to construct the arrhythmia dictionary. When a second cardiac feature similar to the first cardiac feature is subsequently detected through feature delineation, the medical device determines that the second cardiac feature indicates an arrhythmia episode having the same classification as the arrhythmia episode evidenced by the first cardiac feature.

[0010] The techniques of the present disclosure may provide specific improvements in the field of arrhythmia detection and classification by medical devices. For example, the techniques of the present disclosure may use machine learning models only to analyze cardiac features that have been identified through feature delineation as likely to present a patient with an arrhythmia attack. By using machine learning models to verify arrhythmia detection in patients, the techniques of the present disclosure may improve the accuracy of arrhythmia detection. In addition, by using low-power feature delineation to limit the use of computationally complex, power-intensive machine learning models to only the most relevant patient data, the techniques of the present disclosure may effectively implement machine learning models to detect arrhythmia detection without adversely increasing the power usage of such medical devices and reducing their battery life.

[0011] In one example, the present disclosure describes a method comprising: sensing electrocardiogram data of a patient by a medical device including a processing circuit system and a storage medium; performing, by the medical device, a feature-based delineation of the sensed electrocardiogram data to obtain a cardiac feature present in the electrocardiogram data and indicative of an arrhythmia episode in the patient; determining, by the medical device and based on the feature-based delineation, that the cardiac feature satisfies a threshold criterion for applying a machine learning model to verify that the patient has experienced the arrhythmia episode; in response to determining that the cardiac feature satisfies the threshold criterion, applying, by the medical device, electrocardiogram data of a plurality of patients to the medical device to verify that the arrhythmia episode has occurred. The trained machine learning model is applied to the sensed cardiac electrogram data to verify that the patient has experienced the arrhythmia attack based on the machine learning model; and in response to verifying that the patient has experienced the arrhythmia attack by the machine learning model: generating a report by the medical device including an indication that the patient has experienced the arrhythmia attack and one or more of the cardiac characteristics consistent with the arrhythmia attack; and output by the medical device and used for displaying the report including the indication that the patient has experienced the arrhythmia attack and the one or more of the cardiac characteristics consistent with the arrhythmia attack.

[0012] In another example, the present disclosure describes a method comprising: sensing, by a medical device comprising a processing circuit system and a storage medium, cardiac electrogram data of a patient; performing, by the medical device, a feature-based delineation of the sensed cardiac electrogram data to obtain cardiac features present in the cardiac electrogram data; determining, by the medical device, a similarity of the obtained cardiac features to cardiac features of each of a plurality of entries in an arrhythmia dictionary of the medical device, wherein each of the plurality of entries in the arrhythmia dictionary comprises a cardiac feature of a plurality of arrhythmia classifications of the patient; an arrhythmia classification and a cardiac feature proving the arrhythmia classification; in response to determining that the obtained cardiac feature is not similar to the cardiac feature of each of the multiple entries in the arrhythmia dictionary, applying, by the medical device, a machine learning model trained using cardiac electrogram data of multiple patients to the sensed cardiac electrogram data to determine, based on the machine learning model, that the patient has experienced an arrhythmia episode of a first classification; and storing, by the medical device, a first entry in the arrhythmia dictionary including the first classification of the arrhythmia episode and the obtained cardiac feature.

[0013] The present summary is intended to provide an overview of the subject matter described in this disclosure. The present summary is not intended to provide an exclusive or exhaustive explanation of the apparatus and methods described in detail in the following figures and description. Additional details of one or more examples are set forth in the following figures and description. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a conceptual diagram illustrating an example of a medical device system including a leadless implantable medical device and an external device in conjunction with a patient for predicting cardiac arrhythmia according to the presently disclosed techniques.

[0015] Figure 2 It is shown Figure 1 A block diagram of an example of an implantable medical device.

[0016] Figure 3 It is shown Figure 1 A block diagram of an example configuration of an implantable medical device.

[0017] Figure 4 is a flow chart illustrating example operations in accordance with the techniques of this disclosure.

[0018] Figure 5 is a flow chart illustrating example operations in accordance with the techniques of this disclosure.

[0019] Figure 6 is a flow chart illustrating example operations in accordance with the techniques of this disclosure.

[0020] Figure 7 is a flow chart illustrating example operations in accordance with the techniques of this disclosure.

[0021] Figure 8 is a flow chart illustrating example operations in accordance with the techniques of this disclosure.

[0022] Like reference numerals refer to like elements throughout the drawings and the description. DETAILED DESCRIPTION

[0023] Techniques for effectively using machine learning methods for arrhythmia detection in medical devices are disclosed. Characterization algorithms can use cardiac electrogram data sensed from a patient to perform, for example, QRS detection and / or arrhythmia detection. Such characterization algorithms can be optimized for real-time, embedded, and low-power applications, such as for use with implantable medical devices. However, the characterization algorithms may require expert design and feature engineering to accurately detect arrhythmias in patients.

[0024] Machine learning methods such as deep learning and artificial intelligence (AI) for arrhythmia detection provide a flexible platform for developing algorithms with different goals. For example, with a high degree of accuracy, machine learning systems can, for example, detect atrial fibrillation (AF), exclude episodes that do not exhibit arrhythmias, etc., without the need for expert design and feature engineering required for arrhythmia algorithms such as feature delineation. However, machine learning systems may be computationally prohibitive for implementation in medical devices such as IMDs or medical devices that operate on battery power. Frequent use of computationally expensive machine learning models on medical devices may affect battery life.

[0025] As described in detail herein, techniques, methods, systems, and devices are disclosed for physiological, device-based, and algorithm-based approaches to regulating the use of on-device machine learning systems to ensure efficient power usage. As set forth herein, a medical device system is described that allows for the use of on-device machine learning arrhythmia detection, such as deep learning or AI, in a power efficient manner to enable the use of machine learning arrhythmia detection by medical devices performing short-term or long-term diagnostic analysis or monitoring.

[0026] Figure 1 An example medical device system 2 is shown in the context of a patient 4 and a heart 6 in accordance with certain examples of apparatus and methods described herein. The example techniques may be used with an IMD 10, which may be leadless and communicate wirelessly with an external device 12, such as Figure 1 In some instances, IMD 10 can be coupled to one or more leads. In some instances, IMD 10 can be implanted outside of the chest of patient 4 (e.g., subcutaneously in the Figure 1 The IMD 10 may be positioned adjacent to the sternum near and / or just below the level of the heart 6.

[0027] In some examples, IMD 10 may take the form of Reveal LINQ, both available from Medtronic plc, of Dublin, Ireland. TM In the form of an Insertable Cardiac Monitor (ICM) or a Holter Heart Monitor. As discussed herein, the techniques of the present disclosure may be performed by an implantable device such as IMD 10.

[0028] In other examples, the techniques described herein may be performed by an external medical device, such as external device 12, in addition to or in lieu of IMD 10. Such an external medical device may be positioned external to patient 4 (e.g., positioned on the skin of patient 4) and may perform any or all of the functions described herein with respect to IMD 10. External device 12 may be a computing device configured for use in an environment such as a home, clinic, or hospital, and may be additionally configured to communicate with IMD 10 via wireless telemetry. For example, external device 12 may be coupled to a remote patient monitoring system, such as the AT&T® Patient Monitoring System, available from Medtronic plc of Dublin, Ireland. In some examples, the external device 12 may include a programmer, an external monitor, or a 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, the external device 12 is a wearable electronic device such as the SEEQ available from Medtronic, Inc. of Dublin, Ireland. TM A mobile cardiac telemetry (MCT) system or another type of wearable "smart" electronic garment such as a "smart" watch, "smart" patch or "smart" glasses.

[0029] In some examples, a user, such as a physician, technician, surgeon, electrophysiologist, or other clinician, may interact with external device 12 to retrieve physiological or diagnostic information from IMD 10. In some examples, a user, such as patient 4 or a clinician as described above, may also interact with external device 12 to program IMD 10, e.g., to select or adjust values ​​for operating parameters of IMD 10. In some examples, external device 12 acts as an access point to facilitate communications with IMD 10.

[0030] A user, such as a physician, technician, surgeon, electrophysiologist, or other clinician, may interact with external device 12 to retrieve physiological or diagnostic information from IMD 10. The user may also interact with external device 12 to program IMD 10, e.g., to select values ​​for operating parameters of the IMD. External device 12 may include a processor configured to evaluate the EGM and / or other sensed signals transmitted from IMD 10 to external device 12.

[0031] In any such instances, the processing circuitry of the medical device system 2 may send a signal to a remote computer (e.g., Figure 1 The processing circuitry of medical device system 2 may transmit patient data including electrocardiogram data of patient 4 to external device 12 in the medical device system or another device not depicted. In some examples, the processing circuitry of medical device system 2 may transmit a determination that patient 4 is experiencing an episode of arrhythmia, such as an episode of bradycardia, tachycardia, atrial fibrillation, or ventricular fibrillation.

[0032] External device 12 may be a computing device that communicates with IMD 10 via wireless telemetry (e.g., for use in a home, outpatient, clinic, or hospital setting). External device 12 may include or be coupled to a computer such as that available from Medtronic plc of Dublin, Ireland. In some examples, external device 12 may receive data, alarms, patient physiological information, or other information from IMD 10.

[0033] The external device 12 can be used to program commands or operating parameters into the IMD 10 to control its operation (e.g., when configured as a programmer of the IMD 10). In some instances, the external device 12 can be used to query the IMD 10 to retrieve data, including device operating data and physiological data accumulated in the IMD memory. This query can occur automatically according to a schedule and / or can occur in response to a remote or local user command. Programmers, external monitors, and consumer devices are examples of external devices 12 that can be used to query the IMD 10. Examples of communication technologies used by the IMD 10 and the external device 12 include radio frequency (RF) telemetry, which can be an RF link established by Bluetooth, WiFi, or a medical implant communication service (MICS). In some instances, the external device 12 can include a user interface configured to allow a patient 4, a clinician, or another user to interact remotely with the IMD 10. In some such instances, the external device 12 and / or any other device of the medical device system 2 can be a wearable device (e.g., in the form of a watch, necklace, or other wearable item).

[0034] Medical device system 2 is an example of a medical device system configured to perform arrhythmia detection, verification, and reporting. According to the technology of the present disclosure, medical device system 2 implements machine learning arrhythmia detection and characterization to detect and classify arrhythmias in patient 4. Additional examples of one or more other implanted devices or external devices may include an implanted multi-channel pacemaker, ICD, IPG, leadless (e.g., intracardiac) pacemaker, extravascular pacemaker and / or ICD or other IMD or combination of such IMDs configured to deliver CRT to heart 6, external monitor, and external therapy delivery device or drug pump such as external pacing or electrical stimulation device. In some examples, IMD 10 implements a machine learning system, such as a neural network, a deep learning system, or other type of predictive analysis system.

[0035] The communication circuitry of each of the devices of medical device system 2 (e.g., IMD 10 and external device 12) may enable the devices to communicate with each other. In addition, although one or more sensors (e.g., electrodes) are described herein as being positioned on the housing of IMD 10, in other instances, such sensors may be positioned on the housing of another device implanted within or external to patient 4. In such instances, one or more of the other devices may include processing circuitry configured to receive signals from electrodes or other sensors on the respective devices and / or communication circuitry configured to transmit signals from electrodes or other sensors to another device (e.g., external device 12) or a server.

[0036] According to the technology of the present disclosure, the medical device system 2 uses feature delineation and machine learning to perform arrhythmia detection and classification. Specifically, a medical device such as IMD 10 or external device 12 uses feature delineation to perform preliminary detection of arrhythmias in patient 4. In some instances, the medical device applies a machine learning model to the electrocardiogram data of patient 2 to verify that the feature delineation of the electrocardiogram data has correctly detected an arrhythmia attack. In some instances, the medical device applies a machine learning model to the electrocardiogram data of patient 2 to verify that the feature delineation of the electrocardiogram data has correctly classified an arrhythmia attack as a specific type of arrhythmia. For ease of explanation, the following section describes the technology of the present disclosure as being performed by IMD 10. However, the technology of the present disclosure may be performed by other types of medical devices, such as external device 12 or a combination of medical devices (e.g., IMD 10 and external device 12) that operate in conjunction with each other.

[0037] In one example of the technology disclosed herein, IMD 10 senses cardiac electrogram data of patient 4. IMD 10 performs a feature-based delineation of the cardiac electrogram data to obtain cardiac features indicative of an arrhythmia episode in patient 4. IMD 10 determines whether the cardiac features meet threshold criteria for applying a machine learning model to validate the feature-based delineation of the cardiac electrogram data. In some examples, IMD 10 further determines that a noise of at least one of the cardiac features is less than a predetermined threshold. In some examples, IMD 10 further determines that the patient is in a first posture state among a plurality of posture states or a first activity state among a plurality of activity states. In response to determining that the cardiac features meet the threshold criteria, IMD 10 applies the machine learning model to the sensed cardiac electrogram data to, for example, validate that an arrhythmia episode has occurred in patient 4 or to detect one or more additional types of arrhythmias that have occurred in patient 4.

[0038] In one example of the disclosed technology, IMD 10 may classify an arrhythmia episode by comparing cardiac features consistent with an arrhythmia episode to cardiac features of an arrhythmia dictionary maintained by IMD 10. IMD 10 compares a first cardiac feature of the electrocardiogram data to cardiac features defined by entries of the arrhythmia dictionary. For example, in response to determining that the first cardiac feature of the electrocardiogram data is similar to a cardiac feature defined by an entry of the arrhythmia dictionary, IMD 10 determines that the first cardiac feature indicates that an arrhythmia episode has occurred in patient 4, which is a classification defined by a matching entry within the arrhythmia dictionary.

[0039] As another example, in response to determining that a first cardiac feature of the cardiac electrogram data is not similar to a cardiac feature defined by any entry of an arrhythmia dictionary, the IMD 10 applies a machine learning model to determine a classification of an arrhythmia episode evidenced by the first cardiac feature. The IMD 10 stores the determined arrhythmia classification and cardiac feature as a new entry in the arrhythmia dictionary to construct the arrhythmia dictionary. When a second cardiac feature is subsequently detected by feature delineation that is similar to a cardiac feature of an entry of the arrhythmia dictionary, the IMD 10 determines that the second cardiac feature indicates an arrhythmia episode having the same classification as an arrhythmia episode defined in an entry of the arrhythmia dictionary and containing cardiac features that match the second cardiac feature.

[0040] The techniques of the present disclosure may provide specific improvements in the field of arrhythmia detection and classification by a medical device such as IMD 10. For example, the techniques of the present disclosure may use a machine learning model to analyze only those electrocardiogram signals that have been identified through feature delineation as being likely to present to the patient with an arrhythmia episode. By using a machine learning model to validate arrhythmia detection of patient 4 performed through feature delineation, the techniques of the present disclosure may utilize machine learning to improve the accuracy and flexibility of arrhythmia detection. Additionally, by using low-power feature delineation to limit the use of computationally complex, power-intensive machine learning models to only the most relevant patient data, the techniques of the present disclosure may efficiently implement a machine learning model to detect arrhythmia detection without adversely increasing the power usage of such a medical device and reducing its battery life.

[0041] Figure 2 It is shown Figure 1 A block diagram of an example of an implantable medical device. Figure 2As shown, IMD 10 includes processing circuitry 50, sensing circuitry 52, communication circuitry 54, memory 56, sensor 58, switching circuitry 60, characterization circuitry 260, and electrodes 16A, 16B (hereinafter referred to as "electrodes 16"), one or more of which may be disposed within a housing of IMD 10. In some examples, memory 56 includes computer-readable instructions that, when executed by processing circuitry 50, cause IMD 10 and processing circuitry 50 to perform various functions attributed herein to IMD 10 and processing circuitry 50. Memory 56 may include any volatile, nonvolatile, magnetic, optical, or electrical media, such as random access memory (RAM), read-only memory (ROM), nonvolatile RAM (NVRAM), electrically erasable programmable ROM (EEPROM), flash memory, or any other digital media.

[0042] Processing circuit system 50 may include fixed function circuit system and / or programmable processing circuit system. Processing circuit system 50 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 an equivalent discrete or analog logic circuit system. In some examples, processing circuit system 50 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, as well as other discrete or integrated logic circuit systems. The functions attributed to processing circuit system 50 herein may be embodied as software, firmware, hardware, or any combination thereof.

[0043] Sensing circuitry 52 and communication circuitry 54 may be selectively coupled to electrodes 16A, 16B via switching circuitry 60 as controlled by processing circuitry 50. Sensing circuitry 52 may monitor signals from electrodes 16A, 16B to monitor Figure 1 The processing circuit system 50 senses the electrical activity of the heart of the patient 4 and generates electrocardiogram data of the patient 4. In some examples, the processing circuit system 50 performs characterization of the sensed electrocardiogram data through the characterization circuit system 260 to detect an arrhythmia episode in the patient 4. In some examples, the processing circuit system 50 sends a signal to the processing circuit system 54 through the communication circuit system 54. Figure 1 For example, the IMD 10 transmits the electrocardiogram data of the patient 4 to an external device such as an external device 12. Figure 1The IMD 10 transmits the digitized electrocardiogram data to the external device 12 for data processing or viewing by a clinician. In some instances, the IMD 10 transmits one or more segments of the electrocardiogram data in response to detecting an arrhythmia episode by the characterization circuit system 260. In another instance, the IMD 10 transmits one or more segments of the electrocardiogram data in response to an instruction from the external device 12 (e.g., when the patient 4 experiences one or more symptoms of an arrhythmia and inputs a command to the external device 12, the command instructing the IMD 10 to upload the electrocardiogram data for analysis by a monitoring center or clinician).

[0044] In some examples, IMD 10 includes one or more sensors 58, such as one or more accelerometers, microphones, and / or pressure sensors. Sensing circuitry 52 can monitor signals from sensors 58 and transmit patient data obtained from sensors 58 to, for example, Figure 1 In some examples, the sensing circuit system 52 may include one or more filters and amplifiers for filtering and amplifying signals received from one or more electrodes 16A, 16B and / or other sensors 58. In some examples, the sensing circuit system 52 and / or the processing circuit system 50 may include rectifiers, filters and / or amplifiers, sense amplifiers, comparators, and / or analog-to-digital converters.

[0045] The communication circuit system 54 may include any suitable hardware, firmware, software, or any combination thereof, for communicating with another device, such as the external device 12, or another medical device or sensor, such as a pressure sensing device. Under the control of the processing circuit system 50, the communication circuit system 54 may receive downlink telemetry from the external device 12 or another device and send uplink telemetry thereto via an internal or external antenna, such as antenna 26. In some examples, the communication circuit system 54 may communicate with the external device 12. In addition, the processing circuit system 50 may communicate with the external device (e.g., the external device 12) and a Medtronic® medical imaging system, such as that developed by Medtronic plc of Dublin, Ireland. Computer networks such as the Internet communicate with networked computing devices.

[0046] A clinician or other user may retrieve data from IMD 10 using external device 12 or by using another local or networked computing device configured to communicate with processing circuitry 50 via communications circuitry 54. The clinician may also program parameters of IMD 10 using external device 12 or another local or networked computing device. In some examples, the clinician may select one or more parameters that define how IMD 10 senses electrocardiogram data of patient 4.

[0047] One or more components of IMD 10 may be coupled to a power source ( Figure 2 In the embodiment of the present invention, the power source may include a rechargeable or non-rechargeable battery positioned within the housing of IMD 10. The non-rechargeable battery may be selected to last for several years, while the rechargeable battery may be inductively charged from an external device, for example, on a daily or weekly basis.

[0048] According to the technology of the present disclosure, the processing circuit system 50 utilizes the sensing circuit system 52 and senses the electrocardiogram data of the patient 4 through the electrode 16. In some instances, the electrocardiogram data is the ECG of the patient 4. The processing circuit system 50 performs a feature description of the electrocardiogram data through the feature description circuit system 260 to obtain one or more cardiac features present in the electrocardiogram data. The feature description circuit system 260 can further perform preliminary detection of arrhythmia attacks. In some instances, the feature description circuit system 260 includes a circuit system configured to perform one or more of QRS detection, inappropriate processing, noise processing, or description of electrocardiogram data. For example, the feature description circuit system 260 receives the original signal from the sensing circuit system 50 and / or the sensor 58, and extracts one or more cardiac features from the original signal. In some instances, the characterization circuit system 260 identifies one or more cardiac characteristics, such as one or more of the RR intervals present in the electrocardiogram data, an average heart rate present in the electrocardiogram data, a minimum heart rate present in the electrocardiogram data, a maximum heart rate present in the electrocardiogram data, a PR interval present in the electrocardiogram data, variability of the heart rate present in the electrocardiogram data, one or more amplitudes of one or more features of the ECG, intervals between said or more features of the ECG, T wave alternans, QRS morphology measures, or other types of cardiac characteristics not explicitly described herein.

[0049] As an example, the feature delineation circuit system 260 identifies one or more features of the T wave of the electrocardiogram of patient 4 to detect an arrhythmic episode of patient 4. In some instances, the one or more identified features are one or more amplitudes of the T wave. In some instances, the one or more identified features are the frequency of the T wave. In some instances, the one or more identified features include at least the amplitude of the T wave and the frequency of the T wave. In some instances, the feature delineation circuit system 260 identifies one or more relative changes in the one or more identified features that indicate an arrhythmic episode of patient 4. In some instances, the feature delineation circuit system 260 identifies one or more interactions between multiple identified features that indicate an arrhythmic episode of patient 4. In some instances, the feature delineation circuit system 260 analyzes patient data representing one or more values ​​averaged over a short period of time (e.g., approximately 3 minutes). For example, the electrocardiogram data may include one or more of the average frequency or average amplitude of the T wave or QRS wave of the electrocardiogram of patient 4 to detect an arrhythmic episode.

[0050] Processing circuitry 50 may apply characterization via characterization circuitry 260 to determine that one or more cardiac characteristics are indicative of an arrhythmia episode. In some instances, processing circuitry 50 applies characterization via characterization circuitry 260 to classify a detected arrhythmia episode as a particular type of arrhythmia episode (e.g., bradycardia, tachycardia, atrial fibrillation, or ventricular fibrillation). In some instances, processing circuitry 50 performs characterization of sensed cardiac electrogram data via characterization circuitry 260, as described in more detail below. In some instances, the characterization performed by IMD 10 is of reduced complexity in order to conserve power to IMD 10. This may enable characterization circuitry 260 to perform initial or preliminary detection of an arrhythmia.

[0051] In addition, as described in detail below, processing circuit system 50 applies machine learning system 250 to the cardiac electrogram data to verify or classify the detection of an arrhythmia episode by characterization circuit system 260. Although machine learning system 250 can perform a more comprehensive and detailed analysis of the cardiac electrogram data to more accurately detect an arrhythmia by characterization circuit system 260, machine learning system 250 may require more computing resources and power than characterization circuit system 260. By using machine learning system 250 to verify or classify the detection of an arrhythmia episode by characterization circuit system 260, IMD 10 can take advantage of the high accuracy provided by machine learning system 250 while minimizing power consumption or battery life of IMD 10. In some examples, processing circuit system 50 transmits one or more of the cardiac electrogram data, one or more cardiac features present in the cardiac electrogram data, an indication of an arrhythmia episode verified by machine learning system 250, or an indication of a classification of a detected arrhythmia episode as determined by machine learning system 250 to external device 12 via communication circuit system 54.

[0052] In some instances, the machine learning model implemented by the machine learning system 250 is trained with training data, which includes electrocardiogram data of multiple patients labeled with descriptive metadata. For example, during the training phase, the machine learning system 250 processes multiple ECG waveforms. Typically, the multiple ECG waveforms are from multiple different patients. Each ECG waveform is labeled with one or more arrhythmia attacks of one or more types. For example, a training ECG waveform may include multiple segments, each segment is labeled with a descriptor, and the descriptor specifies the absence or presence of an arrhythmia of a specific classification (e.g., bradycardia, tachycardia, atrial fibrillation, or ventricular fibrillation). In some instances, a clinician manually marks the presence of an arrhythmia in each ECG waveform. In some instances, the presence of an arrhythmia in each ECG waveform is marked according to the classification of the feature description algorithm. The machine learning system 250 can be operated to convert the training data into vectors and tensors (e.g., multi-dimensional arrays) on which the machine learning system 250 can apply mathematical operations, such as linear algebra, nonlinear, or alternative computational operations. The machine learning system 250 uses the training data 104 to teach the machine learning model to weigh different features depicted in the electrocardiogram data. In some instances, the machine learning system 250 uses the electrocardiogram data to teach the machine learning model to apply different coefficients that represent one or more features in the electrocardiogram as having more or less importance relative to the occurrence of a particular classification of arrhythmia. By processing many such ECG waveforms marked with arrhythmia episodes, the machine learning system 250 can build and train a machine learning model to obtain a prediction of arrhythmia from the following examples: Figure 1 Patient 4 and the like receive electrocardiogram data that has not been previously analyzed by the machine learning system 250, and process the electrocardiogram data to detect the presence or absence of different classifications of arrhythmias in the patient with high accuracy. Generally, the greater the amount of electrocardiogram data that the machine learning system 250 is trained on, the higher the accuracy of the machine learning model in detecting or classifying arrhythmias in new electrocardiogram data.

[0053] After the machine learning system 250 has trained the machine learning model, the machine learning system 250 may receive patient data, such as electrocardiogram data, for a particular patient, such as patient 4. The machine learning system 250 applies the trained machine learning model to the patient data to detect an arrhythmic episode for patient 4. Additionally, the machine learning system 250 applies the trained machine learning model to the patient data to classify the patient's arrhythmic episode as indicating a particular type of arrhythmia. In some instances, the machine learning system 250 may output a preliminary determination that the arrhythmic episode is indicative of a particular type of arrhythmia, and an estimate of certainty in the determination. In response to determining that the estimate of certainty in the determination is greater than a predetermined threshold (e.g., 50%, 75%, 90%, 95%, 99%), the processing circuit system 50 may classify the arrhythmic episode as a particular type of arrhythmia. As described herein, the processing circuit system 50 uses the machine learning system 250 to verify that the characterization circuit system 260 has correctly detected the arrhythmic episode or that the characterization circuit system 260 has correctly classified the arrhythmic episode as a particular type.

[0054] In some instances, the machine learning system may process one or more cardiac features of the electrocardiogram data rather than the raw electrocardiogram data itself. The one or more cardiac features may be obtained by feature description performed by the IMD 10, as described above. The cardiac features may include, for example, one or more RR intervals present in the electrocardiogram data, an average heart rate present in the electrocardiogram data, a minimum heart rate present in the electrocardiogram data, a maximum heart rate present in the electrocardiogram data, a PR interval present in the electrocardiogram data, a variability of the heart rate present in the electrocardiogram data, one or more amplitudes of one or more features of the ECG, T wave alternations, QRS morphology measurements, or other types of cardiac features not explicitly described herein. In such example embodiments, the machine learning system may train the machine learning model by using a plurality of training cardiac features marked with arrhythmia episodes rather than a plurality of ECG waveforms marked with arrhythmia episodes as described above.

[0055] In some examples, machine learning system 250 may apply the machine learning model to other types of data to determine that an arrhythmic episode has occurred in patient 4. For example, machine learning system 250 may apply the machine learning model to one or more characteristics of the electrocardiogram data related to the patient's arrhythmia, the activity level of IMD 10, the input impedance of IMD 10, or the battery level of IMD 10.

[0056] In another example, the processing circuit system 50 may generate an intermediate representation of the electrocardiogram data from the electrocardiogram data. For example, the processing circuit system 50 may apply one or more signal processing, signal decomposition, wavelet decomposition, filtering or noise reduction operations to the electrocardiogram data to generate an intermediate representation of the electrocardiogram data. In this example, the machine learning system 250 processes this intermediate representation of the electrocardiogram data to detect and classify arrhythmia episodes of the patient 4. In addition, the machine learning system 250 may train the machine learning model by using multiple training intermediate representations marked with arrhythmia episodes instead of multiple original ECG waveforms marked with arrhythmia episodes as described above. The use of this intermediate representation of the electrocardiogram data may allow the machine learning system 250 to train and develop a lighter, less computationally complex machine learning model. In addition, as opposed to using the original electrocardiogram data to train the machine learning model, using this intermediate representation of the electrocardiogram data may require fewer iterations and less training data to build an accurate machine learning model.

[0057] In some instances, the memory 56 includes an arrhythmia dictionary 270. In some instances, the arrhythmia dictionary 270 includes a plurality of entries. Each of the plurality of entries includes a classification of one or more specific types of arrhythmias (e.g., bradycardia, tachycardia, atrial fibrillation, or ventricular fibrillation). In addition, the entry includes one or more cardiac features indicating the classification of the arrhythmia. As described in more detail below, the processing circuit system 50 uses the arrhythmia dictionary 270 to classify arrhythmia episodes detected by the feature description circuit system 260 as specific types of arrhythmias. In addition, the processing circuit system 50 applies the machine learning system 250 to classify detected arrhythmia episodes for which the arrhythmia dictionary 270 does not contain a corresponding entry, so as to establish robust entries of the arrhythmia dictionary 270.

[0058] Although described herein in the context of an example IMD 10 sensing electrocardiogram data of a patient 4, the techniques for arrhythmia detection disclosed herein may be used with other types of devices. For example, the techniques may be used with an external cardiac defibrillator coupled to electrodes external to the cardiovascular system, a device configured to be implanted within the heart such as the Micra 300 commercially available from Medtronic plc of Dublin, Ireland, or a device configured to be implanted within the heart. TM A transcatheter pacemaker such as a transcatheter pacing system, such as the Reveal LINQ also commercially available from Medtronic TM ICM and other insertable cardiac monitors, SEEQ available from Medtronic in Dublin, Ireland TMIt can be implemented by a mobile cardiac telemetry (MCT) device, a neurostimulator, a drug delivery device, a medical device external to the patient 4, a wearable device such as a wearable cardioverter-defibrillator, a fitness tracker or other wearable device, a mobile device such as a mobile phone, a "smart" phone, a laptop computer, a tablet computer, a personal digital assistant (PDA), or a "smart" garment such as "smart" glasses, a "smart" patch or a "smart" watch.

[0059] Figure 3 It is shown Figure 1 A block diagram of an example configuration of an implantable medical device. Figure 3 The components may not necessarily be drawn to scale but may be exaggerated to show details. Figure 3 yes Figure 1 A block diagram of a top view of an example configuration of an IMD 10.

[0060] Figure 3 is to show that it can contain substantially similar Figure 1 Components of IMD 10 are shown in a conceptual diagram of IMD 10. Figure 1 and 2 In addition to the components shown, Figure 3 The illustrated example of IMD 10 may also include a wafer-level insulating cover 74 that can help insulate electrical signals transmitted between electrodes 16A, 16B located on housing 14 and processing circuit system 50. In some examples, insulating cover 74 can be positioned over open housing 14 to form a housing for components of IMD 10B. One or more components of IMD 10B (e.g., antenna 26, processing circuit system 50, sensing circuit system 52, communication circuit system 54, and / or switching circuit system 60) can be formed on the bottom side of insulating cover 74, such as by using flip-chip technology. Insulating cover 74 can be flipped onto housing 14. When flipped and placed onto housing 14, the components of IMD 10 formed on the bottom side of insulating cover 74 can be positioned in a gap 78 defined by housing 14. Housing 14 can be formed of titanium or any other suitable material (e.g., a biocompatible material) and can have a thickness of about 200 microns to about 500 microns. These materials and dimensions are merely examples, and other materials and other thicknesses are possible for the devices of the present disclosure.

[0061] In some instances, the IMD 10 collects patient data including electrocardiogram data of the patient 4 through the sensing circuit system 50 and / or the sensor 58. The sensor 58 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 patient's activity level, the patient's heart rate, the patient's posture, the patient's electrocardiogram, the patient's blood pressure, the patient's accelerometer data, or other types of patient parameter data. In some instances, the IMD 10 uploads the patient data to the external device 12 through the communication circuit system 54, and the external device can then upload this data to a remote monitoring center or a patient monitoring network. In some instances, the IMD 10 uploads patient data every day. In some instances, the patient data includes one or more values ​​representing the average measurement value of the patient 4 over a long period of time (e.g., about 24 hours to about 48 hours). In this instance, the IMD 10 uploads the patient data and performs arrhythmia detection and classification (described below) on the patient 4. However, in other instances, the medical device that processes the patient data to detect and / or classify arrhythmias of patient 4 is different from the medical device that performs short-term monitoring of patient 4. For example, IMD 10 may perform short-term monitoring of patient 4, and external device 12 processes the patient data to detect and / or classify arrhythmias of patient 4.

[0062] Figure 4 is a flow chart illustrating an example operation according to the technology of the present disclosure. Figure 1 describe Figure 4 In some instances, Figure 4 The operation is an operation for detecting and classifying arrhythmias in patient 4. Figure 4 In operation, system 2 combines the accuracy of the machine learning model of machine learning system 250 in detecting arrhythmias with the low power consumption of the feature delineation algorithm. Figure 1 The IMD 10 describes Figure 4 operation, but in other instances, Figure 4 The operations may be performed by, for example, external device 12 or a combination of IMD 10 and external device 12.

[0063] like Figure 4 As depicted, IMD 10 senses electrocardiogram data (402) of patient 4. The electrocardiogram data may be, for example, a paroxysmal ECG of patient 4 or a full disclosure ECG of patient 4. Additionally, the electrocardiogram data of patient 4 may be from a single channel or a multi-channel system. For simplicity, in FIG. Figure 4 In the example of , the electrocardiogram data of patient 4 is described as single-channel paroxysmal ECG data.

[0064] The IMD 10 applies the feature description to the electrocardiogram data to detect one or more cardiac features (404). The IMD 10 further applies the feature description to the electrocardiogram data to detect one or more arrhythmia attacks. For example, the IMD 10 can apply QRS detection description and noise markers to the electrocardiogram data to provide arrhythmia characteristics and / or cardiac characteristics (e.g., heart rate variability during atrial fibrillation attacks, duration of pauses) of the detected arrhythmia attacks. In some instances, the feature description is a one-size-fits-all type algorithm. In other instances, the feature description can be an algorithm with parameters programmed by a clinician or specific to the patient 4. In some instances, the IMD 10 applies the feature description to the electrocardiogram data to classify the detected arrhythmia attacks as specific types of arrhythmias.

[0065] The IMD 10 further determines the triggering of the detected arrhythmia episode. For example, the arrhythmia episode can be detected in response to an automatic trigger, in response to a baseline trigger, or in response to an input received from the patient 4. For example, the baseline trigger may occur not in response to a detected arrhythmia, but in response to a periodically occurring cardiac electrogram segment capture operation (e.g., a segment captured when the IMD 10 is implanted or attached or a segment captured once every 24 hours, etc.). In some instances, the IMD 10 may apply the machine learning system 250 to verify only the arrhythmia episode detected in response to the automatic trigger, while the arrhythmia episode detected in response to the baseline trigger or the patient input may be reported without verification by the machine learning system 250. For example, in response to determining that the automatic triggering did not detect the arrhythmia episode (e.g., the "No" box of 406), the IMD 10 generates a report of the arrhythmia (410) and outputs the report to a clinician or a monitoring center for review (412), as described below.

[0066] In some instances, the IMD 10 may use the machine learning system 250 to perform a number of tasks. For example, the IMD 10 may use the machine learning system 250 to verify that the detection of an arrhythmia made by feature delineation is appropriate. Using the machine learning system 250 in this manner may limit the computational complexity of the machine learning model and may provide diagnosis-related information to clinicians. For example, a machine learning model may be used with a high arrhythmia detection sensitivity to verify that an automatic triggering of atrial fibrillation is associated with an atrial fibrillation episode in patient 4. As another example, a machine learning model may be used to ensure that only portions of the electrocardiogram that are consistent with the duration of a verified atrial fibrillation episode in patient 4 are used to calculate the atrial fibrillation burden in patient 4. As described herein, the burden of an arrhythmia episode is the ratio of the length of time of the arrhythmia episode to the length of time during which the IMD 10 monitors the patient.

[0067] As another example, IMD 10 may use machine learning system 250 to detect the presence of other arrhythmic episodes that are not detected by feature delineation but are consistent with arrhythmic episodes detected by feature delineation. For example, IMD 10 may use machine learning system 250 to detect the presence of "high importance" arrhythmias, such as pauses, atrial fibrillation, or ventricular tachycardia. The use of a machine learning model designed to detect the presence or absence of an arrhythmic episode among a set of types of arrhythmias may be more limited in computational complexity than a machine learning model designed to detect exact types and occurrence instances of an arrhythmic episode.

[0068] For example, about Figure 4 In response to determining that an arrhythmia episode was detected in response to the automatic trigger (e.g., “Yes” block of 406), IMD 10 applies the machine learning model of machine learning system 250 to the electrocardiogram data to verify the detection of the arrhythmia episode by the feature delineation (408). In some examples, IMD 10 applies the machine learning model of machine learning system 250 to the electrocardiogram data to verify the classification of the arrhythmia episode by the feature delineation.

[0069] In some instances, arrhythmia episodes detected in response to automatic triggering have the same limited duration (e.g., 30 seconds before and after arrhythmia detection) and are typically shorter in duration than patient-triggered episodes. In some instances, the machine learning model receives as input at least a portion of the electrocardiogram waveform that caused the automatic triggering and the triggering cause (e.g., detection by characteristic delineation of atrial fibrillation, bradycardia, or pause). The electrocardiogram waveform can be, for example, an ECG waveform.

[0070] In some instances, the machine learning model receives a preprocessed version of a cardiac electrogram waveform or a signal decomposition of the cardiac electrogram waveform as input. For example, the IMD 10 can apply preprocessing to the cardiac electrogram waveform by applying downsampling or signal normalization operations to the waveform to generate an intermediate representation of the cardiac electrogram. Signal decomposition is to decompose the frequency band or frequency domain representation (e.g., spectrogram) using wavelet. In some instances, the IMD 10 can apply preprocessing to the cardiac electrogram waveform to reduce the computational complexity of the machine learning model. For example, by downsampling a record that was originally sampled at 200 samples per second to 100 samples per second, the IMD 10 may require half the computing resources to process the downsampled record into the original record. Similarly, by using the same signal decomposition as the signal decomposition for arrhythmia detection, the IMD 10 can use pre-calculated data to reduce the computational complexity of the machine learning system 250 because the machine learning model does not have to learn the optimal set of signal convolution layers.

[0071] In response to verifying that the detection of an arrhythmia episode by characterization is correct, the IMD 10 stores the electrocardiogram data of the episode and / or transmits the electrocardiogram data of the episode to a monitoring center or clinician for review. For example, the IMD 10 generates a report of arrhythmias (410) and outputs the report to the clinician or monitoring center (412). Typically, only those arrhythmia episodes detected by characterization that are verified by the machine learning system 250 as appropriate or important are stored and transmitted for review by the monitoring center and / or clinician. For example, if the machine learning system 250 detects a bradycardia episode and the characterization performed on the electrocardiogram data indicates that 4 of 4 non-noise heartbeats are less than 30 beats per minute (BPM), the IMD 10 generates a report notifying the physician of the occurrence of an arrhythmia episode.

[0072] In one instance, the report includes an indication that the patient has had an arrhythmia attack and one or more cardiac features consistent with the arrhythmia attack. In some instances, the report further includes classifying the arrhythmia attack as a specific type of arrhythmia. In some instances, the report includes a sub-segment of electrocardiogram data consistent with the arrhythmia attack obtained from patient 4. For example, IMD 10 can identify a sub-segment of electrocardiogram data of patient 4, wherein the sub-segment includes electrocardiogram data of a first time period before the arrhythmia attack (e.g., typically less than 10 minutes before the onset of the arrhythmia attack), a second time period during the occurrence of the arrhythmia attack, and a third time period after the arrhythmia attack (e.g., typically less than 10 minutes after the arrhythmia attack stops). In some instances, the duration of the attack varies depending on the device type and may further depend on the use case of the medical device, one or more settings of the medical device, or the specific type of arrhythmia sensed. For example, some types of arrhythmias self-terminate quickly (resulting in short duration episodes), while other types of arrhythmias are continuous and have a certain length, so that the recorded duration of the episode can depend on the designated memory space on the medical device. Typically, the temporal length of the patient's electrocardiogram data is greater than the first time period, the second time period, and the third time period. In addition, the IMD 10 identifies one or more of the cardiac characteristics consistent with the first time period, the second time period, and the third time period. The IMD 10 includes in the report a sub-segment of the electrocardiogram data and one or more of the cardiac characteristics consistent with the first time period, the second time period, and the third time period.

[0073] In some examples, IMD 10 receives one or more adjustments to one or more parameters used by IMD 10 to sense electrocardiogram data of patient 4 in response to the report. IMD 10 performs such adjustments to subsequently sense electrocardiogram data of patient 4.

[0074] Figure 5 is a flow chart illustrating an example operation according to the technology of the present disclosure. Figure 1 describe Figure 5 In some instances, Figure 5 The operation is an operation for detecting and classifying arrhythmias in patient 4. Figure 5 In operation, system 2 combines the accuracy of the machine learning model of machine learning system 250 in detecting arrhythmias with the low power consumption of the feature delineation algorithm. Figure 1 The IMD 10 describes Figure 5 operation, but in other instances, Figure 5 The operations may be performed by, for example, external device 12 or a combination of IMD 10 and external device 12.

[0075] like Figure 5 As depicted, IMD 10 senses electrocardiogram data of patient 4 (502). IMD 10 applies characterization to the electrocardiogram data to detect one or more cardiac characteristics (504). IMD 10 further determines whether an arrhythmia episode is detected in response to automatic triggering (506). For example, in response to determining that an arrhythmia episode is not detected by automatic triggering (e.g., "No" block of 506), as may be the case for an arrhythmia episode detected in response to a baseline trigger or in response to input received from patient 4, IMD 10 generates a report of the arrhythmia (510) and outputs the report to a clinician or monitoring center for review (512). The operations of steps 502, 504, 506, 510, and 512 may be performed in a manner similar to that described in the accompanying drawings. Figure 4 Steps 402, 404, 406, 410, and 412 occur in a substantially similar manner.

[0076] In response to determining that an arrhythmia episode was detected in response to an automatic trigger (e.g., "yes" block of 506), the IMD processes the electrocardiogram data to produce filtered electrocardiogram data (507). For example, the IMD 10 may adjust the electrocardiogram data, such as one or more ECG episodes, based on the device and physiological parameters such as input impedance, activity level, and posture changes to produce filtered electrocardiogram data. For example, during time periods in which the patient 4 exhibits high activity levels or during time periods in which the IMD 10 experiences highly varying levels of input impedance, the feature description may erroneously detect an arrhythmia episode due to automatic triggering in response to noise in the electrocardiogram data. Using filtered electrocardiogram data may allow the IMD 10 to discard these noisy data periods that may be subject to signal artifacts such as amplitude level variations. In addition, using filtered electrocardiogram data may reduce the computational complexity of the machine learning model because the machine learning model may be designed to analyze non-noisy data. Additionally, use of filtered electrocardiogram data may allow IMD 10 to avoid using a machine learning model to verify that an arrhythmia episode was correctly detected by characterization, where the arrhythmia episode may have been falsely triggered due to noise in the electrocardiogram data, thereby further reducing power consumption of IMD 10. In some instances, automatically triggered arrhythmia episodes associated with pause-related syncope or bradycardia may not be suppressed by the machine learning model.

[0077] IMD 10 applies the machine learning model of machine learning system 250 to the filtered electrocardiogram data to verify the detection of the arrhythmia episode by the feature delineation (508). In some instances, IMD 10 applies the machine learning model of machine learning system 250 to the filtered electrocardiogram data to verify the classification of the arrhythmia episode by the feature delineation. In response to verifying that the detection of the arrhythmia episode by the feature delineation is correct, IMD 10 stores the electrocardiogram data of the episode and / or transmits the electrocardiogram data of the episode to a monitoring center or clinician for review. For example, as described above, IMD 10 generates a report of the arrhythmia (510) and outputs the report to the clinician or monitoring center (512).

[0078] Figure 6 is a flow chart illustrating an example operation according to the technology of the present disclosure. Figure 1 and 2 describe Figure 6 In some instances, Figure 6 The operation is an operation for detecting and classifying arrhythmias of patient 4 using the arrhythmia dictionary constructed by the machine learning system 250. Figure 6 In operation, system 2 combines the accuracy of the machine learning model of machine learning system 250 in detecting arrhythmias with the low power consumption of the feature delineation algorithm. Figure 1 The IMD10 describes Figure 6 operation, but in other instances, Figure 6 The operations may be performed by, for example, external device 12 or a combination of IMD 10 and external device 12.

[0079] like Figure 6 As depicted, IMD 10 senses electrocardiogram data of patient 4 (602). IMD 10 applies characterization to the electrocardiogram data to detect one or more cardiac features (604). The operations of steps 602 and 604 may be performed in the same manner as described above. Figure 4 Steps 402 and 404 occur in a substantially similar manner.

[0080] The IMD 10 determines whether the arrhythmia dictionary 270 contains at least one entry (606). Each entry of the arrhythmia dictionary 270 contains a classification that defines an arrhythmia episode as an episode of a particular type of arrhythmia or a group of arrhythmias, such as an episode of bradycardia, tachycardia, atrial fibrillation, ventricular fibrillation, or AV block. Each entry further contains one or more cardiac features associated with the corresponding classified arrhythmia episode. In response to determining that the arrhythmia dictionary 270 contains at least one entry (e.g., the "yes" box of 606), the IMD 10 determines whether the cardiac features obtained by feature delineation are similar to the cardiac features of the entry in the arrhythmia dictionary 270 (614). For example, the IMD 10 may compare one or more parameters of the cardiac features obtained by feature delineation with one or more parameters of the cardiac features of the entry in the arrhythmia dictionary. In some examples, IMD 10 may determine the similarity between cardiac features obtained through characterization and cardiac features of entries in arrhythmia dictionary 270 by applying computationally efficient methods, such as L1 distance, similarity percentage, or if-then-else rules.

[0081] For example, IMD 10 determines the similarity of the L1 distance of the cardiac feature obtained by the feature delineation and the L1 distance of the cardiac feature of each of the multiple entries of the arrhythmia dictionary. In another example, IMD 10 determines whether the difference between at least one parameter of the cardiac feature obtained by the feature delineation and at least one parameter of the cardiac feature of each of the multiple entries of the arrhythmia dictionary 270 is greater than a predetermined threshold.

[0082] In response to determining that the cardiac characteristics obtained by the feature delineation are similar to the cardiac characteristics of the entry in the arrhythmia dictionary (e.g., the "yes" box of 614), IMD 10 determines whether the entry in the arrhythmia dictionary defines the cardiac characteristics as indicating the arrhythmia episode of concern (616). In response to determining that the entry in the arrhythmia dictionary defines the cardiac characteristics as indicating the arrhythmia episode of concern (e.g., the "yes" box of 616), IMD 10 generates a report of the arrhythmia (610) and outputs the report to a clinician or monitoring center for review (612). The operations of steps 610 and 612 may be performed in the same manner as described above. Figure 4 Steps 410 and 412 occur in a substantially similar manner.

[0083] In response to determining that an entry in the arrhythmia dictionary defines the cardiac characteristic as not indicative of an arrhythmia episode of interest (e.g., "No" block of 616), the IMD 10 may archive the sensed electrocardiogram data for review by a monitoring center or clinician at a later time (618). For example, if the arrhythmia episode is an episode of normal sinus rhythm (NSR), premature ventricular contraction (PVC), premature atrial contraction (PAC), or one or more signal artifacts due to noise, the IMD 10 stores the characteristics of the detected episode (e.g., mean, median RR, RR variation, mean QRS morphology, QRS morphology deviation) for later reference. In some instances, the electrocardiogram data of the episode is not used for further processing or reporting.

[0084] In response to determining that arrhythmia dictionary 270 does not contain at least one entry (e.g., "No" block of 606), or in response to determining that the cardiac features obtained by the feature delineation are not similar to any entry in arrhythmia dictionary 270 (e.g., "No" block of 614), IMD 10 applies the machine learning model of machine learning system 250 to the cardiac electrogram data to verify the detection of the arrhythmia episode made by the feature delineation (608) and / or detect any other arrhythmias of interest. The operation of step 608 may be performed in the same manner as described above. Figure 4 408 occurs in a substantially similar manner. In response to verifying that the detection of an arrhythmia episode made by the feature delineation is correct or that other arrhythmias of interest are present, the IMD 10 stores the cardiac features in the arrhythmia dictionary 270 (620). For example, the IMD 10 creates a new entry in the arrhythmia dictionary 270 that defines a classification of an arrhythmia episode and includes one or more cardiac features obtained by the feature delineation indicating the classification of the arrhythmia episode. Thus, the IMD 10 may use the machine learning system 250 to build, maintain, and update the arrhythmia dictionary 270 for detecting and classifying arrhythmia episodes. In addition, using only the machine learning system 250 to verify and classify the first instance of each type of arrhythmia may limit the use of the machine learning system 250, thereby saving power to the IMD 10. In some instances, the machine learning system 250 is used only to verify and classify arrhythmia episodes detected in response to automatic triggering. In some examples, after storing the cardiac characteristics as a new entry in arrhythmia dictionary 270 , IMD 10 generates a report of the arrhythmia ( 610 ) and outputs the report to a clinician or monitoring center for review ( 612 ).

[0085] Figure 7 is a flow chart illustrating an example operation according to the technology of the present disclosure. Figure 1 describe Figure 7 In some instances, Figure 7 The operation is an operation for detecting and classifying arrhythmia of patient 4. Figure 7 The operation may provide a trade-off between battery life and device life for arrhythmia detection accuracy in patients, where for example Figure 4 The procedure for detecting and classifying arrhythmias did not appropriately trigger or detect an arrhythmic episode in Patient 4. Figure 1 The IMD 10 describes Figure 7 operation, but in other instances, Figure 7 The operations may be performed by, for example, external device 12 or a combination of IMD 10 and external device 12.

[0086] In most patients (eg, between about 80%-90%), the Figure 4 and 5 The described arrhythmia detection operation provides increased accuracy of arrhythmia detection while avoiding disadvantageous power consumption. However, in the remaining subgroup, e.g. Figure 4 and 5 The described operation may not provide high accuracy in arrhythmia detection due to various physiological reasons, such as the placement of the IMD 10 (e.g., which may cause the morphology of the cardiac electrogram to change due to displacement or movement of the IMD 10) or patient characteristics. For arrhythmia diagnosis, patients in such a subpopulation may benefit from a more sensitive and more complex arrhythmia detection scheme, even if such a scheme may consume more power (and result in a commensurate reduction in the maximum battery life of the IMD 10), rather than removing the IMD 10 or replacing the IMD 10 with another device. Figure 7 The operation of the present invention describes a method for transferring arrhythmia detection methods from, for example, Figure 4 or 5. Switch to a higher complexity machine learning arrhythmia detection solution.

[0087] like Figure 7 As depicted, IMD 10 senses electrocardiogram data of patient 4 (702). The operation of step 702 may be performed in the same manner as described above. Figure 4 Step 402 of FIG. 10 occurs in a substantially similar manner. IMD 10 determines whether IMD 10 is operating in a default arrhythmia detection mode (e.g., Figure 4or 5) (703). Typically, the IMD 10 begins operating in this default mode. In response to determining that the IMD 10 is operating in the default mode (e.g., the "Yes" box of 703), the IMD 10 applies the feature delineation to the cardiac electrogram data to detect one or more cardiac features indicative of an arrhythmia attack (704). In some instances, the IMD 10 obtains additional device and physiological parameters (706). The IMD 10 applies the machine learning model of the machine learning system 250 to the cardiac electrogram data to verify the detection of the arrhythmia attack made by the feature delineation (708). In some instances, the IMD 10 applies the machine learning model to the cardiac electrogram data to detect one or more arrhythmia attacks of other classifications that were not detected by the feature delineation. In some instances, the IMD 10 applies the machine learning model to the cardiac electrogram data as well as the device and physiological parameters to perform the verification or detection. The operations of steps 704 and 708 can be performed in the same manner as described above. Figure 4 Steps 404 and 408 occur in a substantially similar manner.

[0088] IMD 10 verifies whether the feature delineation correctly detects the arrhythmia episode based on the machine learning model (709). In response to verifying that the detection of the arrhythmia episode made by the feature delineation is correct (e.g., "yes" box of 709), IMD 10 generates a report of the arrhythmia (710) and outputs the report to the clinician or monitoring center (712). The operations of steps 704 and 708 can be performed in the same manner as Figure 4 Steps 704 and 708 occur in a substantially similar manner. In some examples, IMD 10 may continue to operate according to a default arrhythmia detection protocol.

[0089] In response to verifying that the detection of an arrhythmia episode by the characterization is incorrect (e.g., "No" block of 709), IMD 10 increments a counter of inappropriate instances of detected arrhythmias (716). In some examples, IMD 10 increments the counter when the characterization incorrectly determines that an arrhythmia episode has occurred in patient 4. In some examples, IMD 10 increments the counter when the characterization incorrectly classifies an arrhythmia episode in patient 4 as a particular type of arrhythmia episode.

[0090] If, according to the machine learning model, the characteristic depiction continues to inappropriately detect arrhythmia episodes or episodes that do not contain arrhythmias, the IMD 10 switches from the default arrhythmia detection mode to a second mode using a machine learning model for arrhythmia detection. For example, the IMD determines whether a counter of inappropriate instances of detected arrhythmias is greater than a predetermined threshold (718). If the counter is not greater than the predetermined threshold (e.g., a "No" box of 718), the IMD 10 continues to operate according to the default arrhythmia detection scheme. If the counter is greater than the predetermined threshold (e.g., a "Yes" box of 718), the IMD 10 switches to an alternative arrhythmia detection mode (720). In some instances, the IMD 10 does not switch detection modes until physiological characteristics and device characteristics are met. For example, the IMD 10 may postpone switching detection modes until suitable device adhesion (patient 4 is at a low or medium activity level) and no patient's arrhythmia trigger is active. In some examples, IMD 10 turns off machine learning arrhythmia detection for the duration of a patient-activated trigger that captures electrocardiogram data.

[0091] During subsequent operation, IMD 10 senses electrocardiogram data of patient 4 (702). In response to determining that IMD 10 is not operating in the default mode (e.g., "No" block of 703), the IMD processes the electrocardiogram data to produce filtered electrocardiogram data (707). The operation of step 707 may be performed in the same manner as described above. Figure 5 507 of . IMD 10 applies a second machine learning model of machine learning system 250 to the filtered electrocardiogram data to detect arrhythmia episodes present in the filtered electrocardiogram data (724). In some examples, the second machine learning model receives as input at least a portion of the electrocardiogram waveform, one or more device parameters, or one or more physiological parameters. The electrocardiogram waveform can be, for example, an ECG waveform.

[0092] Because the second machine learning model has a higher complexity than the feature delineation detection of the arrhythmia, the second machine learning model may adversely affect the power consumption and battery life of the IMD 10. Therefore, the IMD 10 updates the estimate of battery life (726). In some instances, the IMD 10 can adjust the frequency or period of transmission to the external device 12. For example, to save battery life, the IMD 10 may switch from uploading patient data such as sensed cardiac electrogram data daily to uploading such data weekly to increase device life. The IMD 10 generates a report of the arrhythmia (710) and outputs the report to a clinician or monitoring center (712).

[0093] Figure 8is a flow chart illustrating an example operation according to the technology of the present disclosure. Figure 1 and 2 describe Figure 8 In some instances, Figure 8 The operation is an operation for detecting and classifying arrhythmias of patient 4 using the arrhythmia dictionary constructed by the machine learning system 250. Figure 8 In operation, system 2 combines the accuracy of the machine learning model of machine learning system 250 in detecting arrhythmias with the low power consumption of the feature delineation algorithm. Figure 1 The IMD10 describes Figure 8 operation, but in other instances, Figure 8 The operations may be performed by, for example, external device 12 or a combination of IMD 10 and external device 12.

[0094] In some applications such as syncope and atrial fibrillation monitoring, the arrhythmias of interest (e.g., pauses, sinus bradycardia, atrial fibrillation) and the corresponding cardiac characteristics of such arrhythmias are known to the clinician. Therefore, the techniques of the present disclosure can be implemented on the IMD 10 using appropriate algorithms for automatic triggering of arrhythmia detection. However, for other types of applications, such as post-myocardial infarction monitoring (e.g., characteristics of interest may be such as premature ventricular contractions (PVC) burden, ST segment, etc.) or post-transcatheter aortic valve replacement (TAVR) monitoring, the exact arrhythmia or condition of interest may not be known. Therefore, Figure 8 The implementation of a generic machine learning model that can be used across multiple applications without the need to develop specific arrhythmia detection algorithms.

[0095] Figure 8 The operation can be basically similar to Figure 6 However, as discussed in more detail below, Figure 8 The operation applies a machine learning system to determine whether the cardiac features obtained by feature delineation are similar to the cardiac features of an entry in the arrhythmia dictionary 270, thereby determining whether the cardiac features obtained by feature delineation indicate an arrhythmia episode of the same classification as the classification of the entry in the arrhythmia dictionary 270.

[0096] like Figure 8 As depicted, IMD 10 senses electrocardiogram data of patient 4 (802). IMD 10 applies characterization to the electrocardiogram data to detect one or more cardiac features (804). The operations of steps 802 and 84 may be performed in the same manner as described above. Figure 4Steps 402 and 404 occur in a substantially similar manner. In some instances, the characterization performed by IMD 10 is adjusted for high sensitivity. For example, IMD 10 may implement a high sensitivity auto-trigger algorithm to identify an episode of concern. In some instances, the characterization is configured to classify a segment of the electrocardiogram of patient 4 as an arrhythmia episode of concern, wherein the segment exhibits an RR rate greater than or equal to 90 beats per minute or less than 60 beats per minute, an RR variability greater than 50 milliseconds, or any morphological change in the absence of noise. In some instances, the characterization is configured to classify a segment of the electrocardiogram of patient 4 as an arrhythmia episode of concern, wherein the segment is of non-noise, non-NSR duration.

[0097] IMD 10 determines whether arrhythmia dictionary 270 contains at least one entry (806). Each entry of arrhythmia dictionary 270 contains a classification that defines an arrhythmia episode as an episode of a particular type of arrhythmia, such as bradycardia, tachycardia, atrial fibrillation, ventricular fibrillation, or AV block. Each entry further contains one or more cardiac features associated with the corresponding classified arrhythmia episode.

[0098] In response to determining that arrhythmia dictionary 270 contains at least one entry (e.g., "yes" box of 806), IMD 10 applies machine learning system 250 to determine whether the cardiac characteristics obtained by the feature delineation are similar to the cardiac characteristics of the entries of arrhythmia dictionary 270 (816). In some examples, machine learning system 250 determines whether the cardiac characteristics obtained by the feature delineation are similar, dissimilar, to the cardiac characteristics of the entries of arrhythmia dictionary 270. In some examples, machine learning system 250 may not determine a specific type of arrhythmia, but rather detect any changes in the cardiac electrogram other than noise or NSR or an entry in arrhythmia dictionary 270.

[0099] In response to determining that the cardiac characteristics obtained by the feature delineation are similar to the cardiac characteristics of the entry in the arrhythmia dictionary 270 (e.g., the "Yes" box of 814), the IMD 10 determines that the cardiac characteristics obtained by the feature delineation indicate an arrhythmia episode of the same classification as the classification of the entry in the arrhythmia dictionary 270. In some instances, the IMD 10 updates a counter of the arrhythmia episode of the classification of the entry in the arrhythmia dictionary 270 to track the total number of arrhythmia episodes of the classification that have occurred in the patient 4 (818). The IMD 10 stores the value of the counter together with the entry in the arrhythmia dictionary 270 (820). After storing the value of the counter in the arrhythmia dictionary 270, the IMD 10 generates a report of the arrhythmia (810) and outputs the report to a clinician or a monitoring center for review (812). The operations of steps 810 and 812 can be respectively performed in the same manner as described above. Figure 6 Steps 610 and 612 occur in a substantially similar manner.

[0100] In response to determining that the arrhythmia dictionary 270 does not contain at least one entry (e.g., a "No" box at 806), or in response to determining that the cardiac features obtained by the feature delineation are not similar to any entry in the arrhythmia dictionary 270 (e.g., a "No" box at 814), the IMD 10 applies the machine learning model of the machine learning system 250 to the cardiac electrogram data to characterize the arrhythmia episode (808). For example, the IMD 10 applies the machine learning model to verify that the arrhythmia episode has occurred, and / or to determine a classification of the arrhythmia episode. In some instances, the machine learning model is a deep learning model that characterizes the episode using "deep features" (e.g., a final deep learning arrhythmia detection model state before softmax classification).

[0101] The IMD 10 stores the cardiac features in the arrhythmia dictionary 270 together with the value of a counter for the number of occurrences of the arrhythmia type (e.g., a "1" corresponding to the first occurrence of the new entry) as a new entry in the arrhythmia dictionary 270 (820). In some instances, the IMD 10 stores the depth features of the newly detected arrhythmia attack in the entry. After creating the new entry and storing the value of the counter in the arrhythmia dictionary 270, the IMD 10 generates a report of the arrhythmia (810) and outputs the report to a clinician or monitoring center for review (812). For example, the IMD 10 transmits cardiac electrogram data corresponding to the detected attack for expert review and confirmation by a clinician. The review is performed by an expert who can look for signal changes related to other patient characteristics such as medication changes in addition to typical arrhythmias.

[0102] The following examples may illustrate one or more aspects of the present disclosure.

[0103] Example 1. A method comprising: sensing electrocardiogram data of a patient by a medical device including a processing circuit system and a storage medium; performing, by the medical device, a feature-based delineation of the sensed electrocardiogram data to obtain a cardiac feature present in the electrocardiogram data and indicating an arrhythmia episode in the patient; determining, by the medical device and based on the feature-based delineation, that the cardiac feature satisfies a threshold criterion for applying a machine learning model to verify that the patient has experienced the arrhythmia episode; in response to determining that the cardiac feature satisfies the threshold criterion for applying the machine learning model, applying, by the medical device, electrocardiograms of a plurality of patients to the patient's cardiac electrogram data; The machine learning model trained with data is applied to the sensed cardiac electrogram data to verify that the patient has experienced the arrhythmia attack based on the machine learning model; and in response to verifying that the patient has experienced the arrhythmia attack by the machine learning model: a report is generated by the medical device including an indication that the patient has experienced the arrhythmia attack and one or more of the cardiac characteristics consistent with the arrhythmia attack; and the report is output by the medical device and used for display, including the indication that the patient has experienced the arrhythmia attack and the one or more of the cardiac characteristics consistent with the arrhythmia attack.

[0104] Example 2. A method according to Example 1, wherein the arrhythmia episode of the patient is an arrhythmia episode of a first classification for the patient, wherein applying the machine learning model to the sensed electrocardiogram data to verify that the patient has experienced the arrhythmia episode includes applying the machine learning model to the sensed electrocardiogram data to verify that the patient has experienced the arrhythmia episode of the first classification, wherein the method further includes, in response to determining that the patient has experienced the arrhythmia episode of the first classification based on the feature-based depiction, applying the machine learning model to the sensed electrocardiogram data to determine that the patient has experienced an arrhythmia episode of a second classification, and wherein generating the report including the indication that the patient has experienced the arrhythmia episode and one or more of the cardiac features consistent with the arrhythmia episode includes generating a report including the indication that the patient has experienced the arrhythmia episode of the first classification, the indication that the patient has experienced the arrhythmia episode of the second classification, and the one or more of the cardiac features consistent with the arrhythmia episode of the first classification.

[0105] Example 3. A method according to any one of Examples 1 or 2, wherein the electrocardiogram data includes an electrocardiogram (ECG) of the patient.

[0106] Example 4. A method according to any one of Examples 1 to 3, wherein performing a feature-based depiction of the electrocardiogram data to obtain the cardiac features present in the electrocardiogram data includes performing at least one of QRS detection, refractory processing, noise processing or depiction of the electrocardiogram data to obtain the cardiac features present in the electrocardiogram data.

[0107] Example 5. A method according to any one of Examples 1 to 4, wherein applying the machine learning model to verify that the patient has experienced the arrhythmia episode includes applying the machine learning model to verify that the patient has experienced an episode of at least one of bradycardia, tachycardia, atrial fibrillation, ventricular fibrillation, or AV block.

[0108] Example 6. A method according to any one of Examples 1 to 5, wherein the cardiac characteristics present in the electrocardiogram data are one or more RR intervals among the RR intervals present in the electrocardiogram data, an average heart rate present in the electrocardiogram data, a minimum heart rate present in the electrocardiogram data, a maximum heart rate present in the electrocardiogram data, a PR interval present in the electrocardiogram data, variability of the heart rate present in the electrocardiogram data, one or more amplitudes of one or more characteristics of an electrocardiogram (ECG), or intervals between one or more characteristics of the ECG.

[0109] Example 7. A method according to any one of Examples 1 to 6, wherein the machine learning model trained using the electrocardiogram data of the multiple patients includes a machine learning model trained using multiple electrocardiogram (ECG) waveforms, each ECG waveform being labeled with one or more classified arrhythmia episodes of one of the multiple patients.

[0110] Example 8. A method according to any one of Examples 1 to 7, wherein determining that the cardiac characteristic satisfies the threshold criterion includes determining that at least one of a physiological parameter of the patient or a parameter of the medical device satisfies the threshold criterion.

[0111] Example 9. A method according to any one of Examples 1 to 8, wherein applying the machine learning model to the sensed electrocardiogram data further includes applying the machine learning model to at least one of: one or more characteristics of the sensed electrocardiogram data associated with the patient's arrhythmia; an activity level of the medical device; an input impedance of the medical device; or a battery level of the medical device.

[0112] Example 10. A method according to any one of Examples 1 to 9, wherein applying the machine learning model to the sensed cardiac electrogram data includes applying the machine learning model to the sensed cardiac electrogram data in response to determining that the cardiac features meet the threshold criteria and determining that the noise of at least one of the cardiac features is less than a predetermined threshold.

[0113] Example 11. A method according to any one of Examples 1 to 10, wherein applying the machine learning model to the sensed electrocardiogram data includes applying the machine learning model to the sensed electrocardiogram data in response to determining that the cardiac characteristic satisfies the threshold criteria and determining that the patient is in a first posture state among multiple posture states.

[0114] Example 12. A method according to any one of Examples 1 to 11, wherein applying the machine learning model to the sensed electrocardiogram data includes applying the machine learning model to the sensed electrocardiogram data in response to determining that the cardiac characteristic satisfies the threshold criteria and determining that the patient is in a first activity state among multiple activity states.

[0115] Example 13. A method according to any one of Examples 1 to 12, wherein in response to outputting the report including an indication that the patient has experienced the arrhythmia episode and the one or more of the cardiac features consistent with the arrhythmia episode, the method further includes: receiving, by the medical device from a user, an adjustment to the feature-based depiction of the cardiac electrogram data; and performing a feature-based depiction of the cardiac electrogram data based on the adjustment to obtain a second cardiac feature present in the cardiac electrogram data.

[0116] Example 14. A method according to any one of Examples 1 to 13, wherein the electrocardiogram data of the patient includes an electrocardiogram (ECG) of the patient, and wherein generating the report including the indication that the patient has experienced the arrhythmia attack and one or more of the cardiac features consistent with the arrhythmia attack includes: identifying a sub-segment of the ECG of the patient, wherein the sub-segment includes ECG data of a first time period before the arrhythmia attack, a second time period during the arrhythmia attack, and a third time period after the arrhythmia attack, and wherein the time length of the ECG of the patient is greater than the first time period, the second time period, and the third time period; identifying one or more of the cardiac features consistent with the first time period, the second time period, and the third time period; and including the sub-segment of the ECG and the one or more of the cardiac features consistent with the first time period, the second time period, and the third time period in the report.

[0117] Example 15. A method according to any one of Examples 1 to 14, wherein determining based on the feature-based depiction that the cardiac feature satisfies threshold criteria for applying the machine learning model to verify that the patient has experienced the arrhythmia episode includes determining based on the feature-based depiction that the cardiac feature indicates that the patient has experienced a first classification of arrhythmia episode, and wherein applying the machine learning model to the sensed cardiac electrogram data to verify that the patient has experienced the arrhythmia episode includes applying the machine learning model to the sensed cardiac electrogram data to verify the determination based on the feature-based depiction that the cardiac feature indicates the first classification of the arrhythmia episode.

[0118] Example 16. A method according to any one of Examples 1 to 15, further comprising determining a burden of the arrhythmia episode for the patient in response to verifying that the patient has experienced the arrhythmia episode, wherein the burden of the arrhythmia episode for the patient comprises a ratio of a length of time of the arrhythmia episode to a length of time during which the medical device monitors the patient.

[0119] Example 17. A method according to any one of Examples 1 to 16, wherein the method further comprises: processing the sensed electrocardiogram data by the medical device to generate filtered electrocardiogram data, wherein applying the machine learning model to the sensed electrocardiogram data to verify that the patient has experienced the arrhythmia episode comprises applying the machine learning model to the filtered electrocardiogram data to verify that the patient has experienced the arrhythmia episode.

[0120] Example 18. A method according to Example 17, wherein processing the sensed electrocardiogram data to generate the filtered electrocardiogram data includes: determining at least one of a time period of high activity level of the patient or a time period of highly varying input impedance of the medical device; in response to determining at least one of the time period of high activity level of the patient or the time period of highly varying input impedance of the medical device, discarding at least a portion of the sensed electrocardiogram data that coincides with at least one of the time period of high activity level of the patient or the time period of highly varying input impedance of the medical device to generate the filtered electrocardiogram data.

[0121] Example 19. A method according to Example 18, wherein discarding at least a portion of the sensed electrocardiogram data that coincides with at least one of the time periods of high activity level of the patient or the time periods of highly varying input impedance of the medical device includes discarding one or more of the sensed input impedance of the medical device, the sensed activity level of the patient, or the sensed posture changes of the patient that coincide with at least one of the time periods of high activity level of the patient or the time periods of highly varying input impedance of the medical device.

[0122] Example 20. A method according to any one of Examples 1 to 19, further comprising: wherein applying the machine learning model to the sensed electrocardiogram data to verify that the patient has had the arrhythmia episode includes at least one of a first determination based on the machine learning model that the patient has not had the arrhythmia episode or a second determination based on the machine learning model that the patient has had a different type of arrhythmia episode; in response to at least one of the first determination and the second determination, updating a counter of arrhythmia episodes of the patient that were not correctly detected by the medical device; and in response to determining that the value of the counter is greater than a predetermined threshold, switching from performing a feature-based delineation of the sensed electrocardiogram data to obtain cardiac features present in the electrocardiogram data and indicating an arrhythmia episode of the patient to applying a second machine learning model trained using electrocardiogram data of multiple patients to the sensed electrocardiogram data to obtain cardiac features present in the electrocardiogram data and indicating an arrhythmia episode of the patient based on the machine learning model.

[0123] Example 21. A method according to Example 20, wherein sensing the electrocardiogram data of the patient includes sensing first electrocardiogram data of the patient; and wherein the method further includes: sensing second electrocardiogram data of the patient by the medical device; applying the second machine learning model to the sensed second electrocardiogram data to obtain a second cardiac feature present in the second electrocardiogram data and indicating a second arrhythmia attack in the patient based on the machine learning model; in response to obtaining the second cardiac feature indicating the second arrhythmia attack in the patient by the second machine learning model: generating a second report by the medical device including an indication that the patient has had the second arrhythmia attack and one or more of the second cardiac features consistent with the second arrhythmia attack; and outputting by the medical device and used to display the second report including the indication that the patient has had the second arrhythmia attack and the one or more second cardiac features consistent with the second arrhythmia attack.

[0124] Example 22. A method according to any one of Examples 21 or 22, further comprising: in response to determining that the value of the counter is greater than the predetermined threshold: updating, by the medical device, an estimate of the power consumption of the medical device; and outputting by the medical device and used for displaying the estimate of the power consumption of the medical device.

[0125] Example 23. A method according to any one of Examples 1 to 22, wherein the method further includes processing the sensed electrocardiogram data by the medical device to generate an intermediate representation of the sensed electrocardiogram data, wherein applying the machine learning model trained using the electrocardiogram data of the multiple patients to the sensed electrocardiogram data to verify that the patient has experienced the arrhythmia episode includes applying the machine learning model trained using the intermediate representation of the electrocardiogram data of the multiple patients to the intermediate representation of the sensed electrocardiogram data and the cardiac features present in the electrocardiogram data to verify that the patient has experienced the arrhythmia episode.

[0126] Example 24. A method according to Example 23, wherein processing the sensed electrocardiogram data to generate the intermediate representation of the sensed electrocardiogram data includes at least one of: applying a filter to the sensed electrocardiogram data; performing signal decomposition on the sensed electrocardiogram data.

[0127] Example 25. The method of Example 24, wherein performing signal decomposition on the sensed electrocardiogram data comprises performing wavelet decomposition on the sensed electrocardiogram data.

[0128] Example 26. A method comprising: sensing cardiac electrogram data of a patient by a medical device including a processing circuit system and a storage medium; performing, by the medical device, a feature-based delineation of the sensed cardiac electrogram data to obtain cardiac features present in the cardiac electrogram data; determining, by the medical device, a similarity between the obtained cardiac features and cardiac features of each of a plurality of entries in an arrhythmia dictionary of the medical device, wherein each of the plurality of entries in the arrhythmia dictionary includes one arrhythmia classification among a plurality of arrhythmia classifications of the patient and a cardiac feature that justifies the arrhythmia classification; in response to determining that the obtained cardiac features are not similar to the cardiac features of each of the plurality of entries in the arrhythmia dictionary, applying, by the medical device, a machine learning model trained using cardiac electrogram data of a plurality of patients to the sensed cardiac electrogram data to determine, based on the machine learning model, that the patient has experienced an arrhythmia episode of a first classification; and storing, by the medical device, a first entry in the arrhythmia dictionary including the first classification of the arrhythmia episode and the obtained cardiac features.

[0129] Example 27. A method according to Example 26, further comprising: after storing an entry including the first classification of the arrhythmia attack and the obtained cardiac characteristics: generating by the medical device a report including an indication that the patient has had the arrhythmia attack of the first classification and one or more of the obtained cardiac characteristics consistent with the arrhythmia attack of the first classification; and outputting by the medical device and used for displaying the report including the indication that the patient has had the arrhythmia attack of the first classification and one or more of the obtained cardiac characteristics consistent with the arrhythmia attack of the first classification.

[0130] Example 28. A method according to any one of Examples 26 or 27, further comprising: sensing second electrocardiogram data of a patient by the medical device comprising a processing circuit system and a storage medium; performing a feature-based depiction of the sensed second electrocardiogram data by the medical device to obtain a second cardiac feature present in the second electrocardiogram data; determining by the medical device that the second cardiac feature is similar to the cardiac feature of the first entry among the multiple entries in the arrhythmia dictionary of the medical device; in response to determining that the second cardiac feature is similar to the cardiac feature of the first entry among the multiple entries in the arrhythmia dictionary: generating by the medical device a report including an indication that the patient has had an arrhythmia episode of the first classification and one or more of the second cardiac features consistent with the arrhythmia episode of the first classification; and outputting by the medical device and used for displaying the report including an indication that the patient has had an arrhythmia episode of the first classification and one or more of the second cardiac features consistent with the arrhythmia episode of the first classification.

[0131] Example 29. A method according to any one of Examples 26 to 28, wherein determining the similarity of the obtained cardiac feature to the cardiac feature of each of the multiple entries in the arrhythmia dictionary of the medical device includes determining that the L1 distance of the obtained cardiac feature is not similar to the L1 distance of the cardiac feature of each of the multiple entries in the arrhythmia dictionary of the medical device.

[0132] Example 30. A method according to any one of Examples 26 to 28, wherein determining the similarity of the obtained cardiac characteristic and the cardiac characteristic of each of the multiple entries in the arrhythmia dictionary of the medical device includes determining that a difference between at least one parameter of the obtained cardiac characteristic and at least one parameter of the cardiac characteristic of each of the multiple entries in the arrhythmia dictionary of the medical device is greater than a predetermined threshold.

[0133] Example 31. A method according to any one of Examples 26 to 28, wherein determining the similarity of the obtained cardiac feature to the cardiac feature of each of the multiple entries in the arrhythmia dictionary of the medical device includes: applying a second machine learning model trained using cardiac electrogram data of multiple patients to the obtained cardiac feature and the cardiac feature of each of the multiple entries in the arrhythmia dictionary of the medical device to determine that the obtained cardiac feature is not similar to the cardiac feature of each of the multiple entries in the arrhythmia dictionary of the medical device.

[0134] Example 32. A method according to any one of Examples 26 to 31, further comprising: in response to determining that the obtained cardiac characteristics are similar to cardiac characteristics of a first entry among the multiple entries in the arrhythmia dictionary of the medical device: updating, by the medical device, a counter of arrhythmia episodes of a first classification defined by the first entry among the multiple entries in the arrhythmia dictionary of the medical device; and storing, by the medical device, the counter in the first entry among the multiple entries in the arrhythmia dictionary of the medical device.

[0135] Example 33. A method according to any one of Examples 26 to 32, further comprising determining, by the medical device, that the obtained cardiac characteristic is not similar to the patient's normal sinus rhythm (NSR) or noise, wherein in response to determining that the obtained cardiac characteristic is not similar to the cardiac characteristic of each of the multiple entries in the arrhythmia dictionary, applying the machine learning model to the sensed cardiac electrogram data to determine that the patient has experienced the arrhythmia episode of the first classification includes applying the machine learning model to the sensed cardiac electrogram data to determine that the patient has experienced the arrhythmia episode of the first classification in response to determining that the obtained cardiac characteristic is not similar to the cardiac characteristic of each of the multiple entries in the arrhythmia dictionary and in response to determining that the obtained cardiac characteristic is not similar to the patient's NSR.

[0136] In some examples, the technology of this disclosure includes a system including means for performing any of the methods described herein. In some examples, the technology of this disclosure includes a computer-readable medium including instructions for causing a processing circuit system to perform any of the methods described herein.

[0137] It should be understood that the various aspects disclosed herein may be combined in combinations different from those specifically presented in the specification and drawings. It should also be understood that, depending on the example, certain actions or events of any of the processes or methods described herein may be performed in different orders, may be added, merged, or omitted entirely (e.g., all described actions or events may not be necessary for performing the technique). In addition, although for clarity, certain aspects of the present disclosure are described as being performed by a single module, unit, or circuit, it should be understood that the techniques of the present disclosure may be performed by a combination of units, modules, or circuit systems associated with, for example, a medical device.

[0138] 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 can be accessed by a computer).

[0139] 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. Furthermore, the techniques may be fully implemented in one or more circuits or logic elements.

Claims

1. A medical device comprising: Storage media; as well as processing circuitry operably coupled to the storage medium and configured to: sensing electrocardiogram data of the patient; determining that the cardiac electrogram is indicative of an arrhythmia, wherein, to determine that the cardiac electrogram is indicative of an arrhythmia, the processing circuitry is configured to perform a feature-based delineation of the sensed cardiac electrogram data to obtain cardiac features present in the cardiac electrogram data and indicative of an arrhythmic episode in the patient; In response to determining that the cardiac electrogram indicates an arrhythmia, determining based on the feature-based delineation whether the cardiac feature satisfies a threshold criterion for applying a machine learning model to verify that the patient has experienced an episode of the arrhythmia; In response to determining that the cardiac feature satisfies the threshold criteria for applying the machine learning model, applying the machine learning model trained using electrocardiogram data of a plurality of patients to the sensed electrocardiogram data to verify that the patient has experienced the arrhythmia episode based on the machine learning model, wherein the machine learning model applied for verification is not used to perform the feature-based delineation of the sensed electrocardiogram data; In response to verifying, by the machine learning model, that the patient has experienced the arrhythmic episode, generating a report for transmission to an external device, the report including an indication that the patient has experienced an arrhythmic episode and one or more of the cardiac characteristics consistent with the arrhythmic episode; and In response to determining that the cardiac characteristic does not meet the threshold criteria for applying the machine learning model, transmitting an indication that the arrhythmic episode has occurred to the external device without using the machine learning model to verify that the patient has experienced the arrhythmic episode.

2. The medical device according to claim 1, wherein the arrhythmic episode of the patient is an arrhythmic episode of a first classification of the patient, wherein in order to apply the machine learning model to the sensed electrocardiogram data to verify that the patient has experienced the arrhythmia episode, the processing circuit system is configured to apply the machine learning model to the sensed electrocardiogram data to verify that the patient has experienced the arrhythmia episode of the first classification, wherein the processing circuitry is further configured to, in response to determining that the patient has experienced the first classified arrhythmia episode based on the feature-based delineation, apply the machine learning model to the sensed electrocardiogram data to determine that the patient has experienced a second classified arrhythmia episode, and Wherein, in order to generate the report including the indication that the patient has experienced the arrhythmia episode and one or more of the cardiac characteristics consistent with the arrhythmia episode, the processing circuit system is configured to generate a report including the indication that the patient has experienced the arrhythmia episode of the first classification, the indication that the patient has experienced the arrhythmia episode of the second classification, and the one or more of the cardiac characteristics consistent with the arrhythmia episode of the first classification.

3. A medical device according to any one of claims 1 to 2, wherein in order to perform a feature-based depiction of the electrocardiogram data to obtain the cardiac features present in the electrocardiogram data, the processing circuit system is configured to perform at least one of QRS detection, refractory processing, noise processing or depiction of the electrocardiogram data to obtain the cardiac features present in the electrocardiogram data.

4. The medical device of any one of claims 1 to 2, wherein in order to apply the machine learning model to verify that the patient has experienced the arrhythmia episode, the processing circuit system is configured to apply the machine learning model to verify that the patient has experienced an episode of at least one of bradycardia, tachycardia, atrial fibrillation, ventricular fibrillation, or AV block.

5. A medical device according to any one of claims 1 to 2, wherein the machine learning model trained using the cardiac electrogram data of the multiple patients includes a machine learning model trained using multiple electrocardiogram (ECG) waveforms, each ECG waveform is labeled with one or more classified one or more arrhythmia episodes of one of the multiple patients.

6. The medical device according to any one of claims 1 to 2, wherein in order to determine that the cardiac characteristic meets the threshold criterion, the processing circuit system is configured to determine that at least one of the patient's physiological parameters or the parameters of the medical device meets the threshold criterion.

7. A medical device according to any one of claims 1 to 2, wherein, in order to apply the machine learning model to the sensed cardiac electrogram data, the processing circuit system is configured to apply the machine learning model to the sensed cardiac electrogram data in response to determining that the cardiac characteristics meet the threshold criteria and determining that the noise of at least one of the cardiac characteristics is less than a predetermined threshold.

8. A medical device according to any one of claims 1 to 2, wherein, in order to apply the machine learning model to the sensed electrocardiogram data, the processing circuit system is configured to apply the machine learning model to the sensed electrocardiogram data in response to determining that the cardiac characteristic satisfies the threshold criterion and determining that the patient is in a first posture state among multiple posture states.

9. The medical device according to any one of claims 1 to 2, wherein the electrocardiogram data of the patient comprises an electrocardiogram (ECG) of the patient, and wherein to generate the report including the indication that the patient has experienced the arrhythmic episode and the one or more of the cardiac characteristics consistent with the arrhythmic episode, the processing circuitry is configured to: identifying a sub-segment of the ECG of the patient, wherein the sub-segment includes ECG data for a first time period before the onset of the arrhythmia, a second time period during the onset of the arrhythmia, and a third time period after the onset of the arrhythmia, and wherein the temporal length of the ECG of the patient is greater than the first time period, the second time period, and the third time period; identifying one or more of the cardiac characteristics consistent with the first time period, the second time period, and the third time period; and The sub-segment of the ECG and the one or more of the cardiac characteristics consistent with the first time period, the second time period, and the third time period are included in the report.

10. The medical device according to any one of claims 1 to 2, wherein to determine, based on the feature-based delineation, that the cardiac feature satisfies a threshold criterion for applying the machine learning model to verify that the patient has experienced an arrhythmic episode, the processing circuitry is configured to determine, based on the feature-based delineation, that the cardiac feature indicates that the patient has experienced an arrhythmic episode of a first classification, and Wherein, in order to apply the machine learning model to the sensed electrocardiogram data to verify that the patient has experienced the arrhythmia episode, the processing circuit system is configured to apply the machine learning model to the sensed electrocardiogram data to verify that the cardiac characteristics determined based on the depiction of the features indicate the arrhythmia episode of the first classification.

11. The medical device according to any one of claims 1 to 2, wherein the processing circuitry is further configured to process the sensed electrocardiogram data to generate filtered electrocardiogram data, Wherein, in order to apply the machine learning model to the sensed electrocardiogram data to verify that the patient has experienced the arrhythmia episode, the processing circuit system is further configured to apply the machine learning model to the filtered electrocardiogram data to verify that the patient has experienced the arrhythmia episode.

12. The medical device of any one of claims 1 to 2, wherein applying the machine learning model to the sensed electrocardiogram data to verify that the patient has had the arrhythmic episode comprises at least one of a first determination based on the machine learning model that the patient has not had the arrhythmic episode or a second determination based on the machine learning model that the patient has had a different type of arrhythmic episode, The medical device is further configured to: In response to at least one of the first determination and the second determination, updating, by the medical device, a counter of incorrectly detected arrhythmia episodes of the patient; and In response to determining that the value of the counter is greater than a predetermined threshold, switching from performing a feature-based characterization of the sensed electrocardiogram data to obtain cardiac features present in the electrocardiogram data and indicative of an arrhythmia episode in the patient to applying a second machine learning model trained using electrocardiogram data of multiple patients to the sensed electrocardiogram data to obtain cardiac features present in the electrocardiogram data and indicative of an arrhythmia episode in the patient based on the machine learning model.

13. A computing device comprising: means for sensing electrocardiographic data of a patient; means for determining that the cardiac electrogram is indicative of an arrhythmia, wherein, to determine that the cardiac electrogram is indicative of an arrhythmia, the means performs a feature-based delineation of the sensed cardiac electrogram data to obtain cardiac features present in the cardiac electrogram data and indicative of an arrhythmic episode in the patient; means for determining, in response to determining that the cardiac electrogram indicates an arrhythmia, whether the cardiac feature satisfies threshold criteria for applying a machine learning model to verify that the patient has experienced an episode of the arrhythmia based on the feature-based depiction; means for applying the machine learning model, in response to determining that the cardiac feature satisfies the threshold criteria for applying the machine learning model, to the sensed electrocardiogram data to verify that the patient has experienced the arrhythmia episode based on the machine learning model, wherein the machine learning model applied for verification is not used to perform the feature-based delineation of the sensed electrocardiogram data; and means for generating a report for transmission to an external device in response to verifying by the machine learning model that the patient has experienced the arrhythmic episode, the report comprising an indication that the patient has experienced an arrhythmic episode and one or more of the cardiac characteristics consistent with the arrhythmic episode; and Means for transmitting to the external device an indication that the arrhythmic episode has occurred, in response to determining that the cardiac characteristic does not satisfy the threshold criteria for applying the machine learning model, without using the machine learning model to verify that the arrhythmic episode has occurred in the patient.

14. A medical system comprising the medical device according to any one of claims 1 to 12.

15. A medical system comprising the computing device according to claim 13.

Images