Ventricular premature beat (PVC) detection using artificial intelligence model

Abstract

An example medical system includes a plurality of electrodes configured to sense an electrocardiogram of a patient; and processing circuitry configured to: perform feature extraction of the sensed electrocardiogram to extract a plurality of features from the sensed electrocardiogram; converting one or more of the plurality of extracted features into corresponding feature images; applying a machine learning model trained using electrocardiogram data for the plurality of patients to the images of the sensed electrocardiogram and at least the corresponding feature images to determine whether one or more particular heartbeats in the sensed electrocardiogram are indicative of ventricular premature beat (PVC) bounces; and in response to the determination that the one or more particular heartbeats indicate PVC beating, outputting a classification that the one or more particular heartbeats are PVC beating.

Description

[0001] This application is an international application claiming priority to U.S. Patent Application No. 18 / 050,814, filed Oct. 28, 2022, the entire content of which is incorporated herein by reference. Technical Field

[0002] The present disclosure generally relates to medical device systems, and more particularly to medical device systems configured to detect premature ventricular contractions (PVCs). Background Art

[0003] Medical devices can be used to monitor a patient's physiological signals. For example, some medical devices are configured to sense electrocardiogram (EGM) signals, which indicate the electrical activity of the heart via electrodes. Some medical devices can be configured to deliver therapy in combination with or independent of the monitoring of physiological signals.

[0004] PVCs are premature beats originating from the ventricles. PVCs are premature because they occur before the regular heartbeat originating from the sinoatrial node. During a PVC event, the ventricles discharge and contract prematurely before the normal impulse arrives from the sinoatrial node. PVCs can occur in healthy individuals. As examples, PVCs can be caused by caffeine, smoking, alcohol, stress, fatigue, pharmacological toxicity, electrolyte imbalance, hypoxia, and heart attacks. Common symptoms associated with PVCs include palpitations, dizziness, fatigue, dyspnea, chest pain, and lightheadedness. PVCs are generally considered benign, but can lead to cardiomyopathy, ventricular arrhythmias, and heart failure.

[0005] Management strategies for PVC-induced cardiomyopathy include medical therapy and catheter ablation, with the role of catheter ablation increasing given its potential for permanent PVC suppression. Ablation of PVCs can lead to improvement in left ventricular systolic dysfunction (LVSD) and normalization of left ventricular ejection fraction (LVEF). PVC burden, which is a quantification of the amount of PVCs over a period of time, can be an independent predictor of PVC-induced cardiomyopathy. Currently, 24-hour Holter monitoring is the most commonly used method for determining PVC burden. Summary of the Invention

[0006] Generally speaking, the present disclosure relates to techniques for detecting premature ventricular contractions (PVCs) using electrocardiograms (EGMs) sensed by a medical device to, for example, facilitate determination of the PVC burden. More specifically, the present disclosure relates to techniques for evaluating cardiac EGMs to determine whether a particular heartbeat is a PVC. A processing circuit may determine whether a particular heartbeat is a PVC based on extracting particular features from the cardiac EGM, converting the extracted features into corresponding feature images, and applying an image of the cardiac EGM and one or more feature images to an artificial intelligence model, such as a machine learning model or other suitable model. In some examples, the processing circuit may weight and / or normalize the cardiac EGM signal image and the feature images, and apply the weighted and / or normalized cardiac EGM signal image and feature images to a machine learning model to determine whether a particular heartbeat is a PVC.

[0007] For some medical devices, for example, those that utilize external, subcutaneous, or other extravascular electrodes, the position and orientation of the electrodes used to sense cardiac EGMs may vary between patients and over time within a given patient relative to the heart and other tissues. According to the techniques of the present disclosure, the criteria used by the processing circuit to determine whether a particular heartbeat is a PVC may improve the sensitivity and / or specificity of PVC detection. In some examples, improving the sensitivity and / or specificity of PVC detection may facilitate a more accurate determination of the PVC burden, cardiac health, and risk of sudden cardiac death, and may lead to clinical interventions to suppress PVCs, such as medications and PVC ablation.

[0008] Unlike conventional PVC detection systems, the techniques and systems of the present disclosure can use a machine learning model to more accurately determine whether one or more specific heartbeats in a sensed electrocardiogram indicate premature ventricular contractions (PVCs). In some examples, a machine learning model is trained using a set of training instances, where one or more of these training instances include data indicating the relationship between various EGM features (including features for a specific heartbeat) and a classification of whether a PVC occurs for such features. Since the machine learning model is trained with potentially thousands or millions of training instances, the machine learning model can reduce the amount of classification errors in classifying one or more heartbeats as PVCs when compared to conventional PVC detection systems. Additionally, the techniques and systems of the present disclosure can be implemented in an implantable medical device (IMD) that can continuously and / or periodically sense EGMs without human intervention when subcutaneously implanted in a patient for months or years, and perform millions of operations per second on the patient's EGM data to identify PVCs using the machine learning model. Using the techniques of the present disclosure (using a machine learning model) in an IMD can be advantageous when a physician cannot be present with a patient for weeks or months to evaluate the EGM, and / or when performing millions of operations on weeks or months of EGM data cannot practically be performed in a physician's mind.

[0009] Utilizing a machine learning model implementing the techniques of the present disclosure to reduce classification errors for PVCs may provide one or more technical and clinical advantages. In some examples, a medical system that converts extracted features of EGM data into corresponding feature images and applies a trained machine learning model to the sensed images of the EGM and the converted feature images may help determine with higher specificity and sensitivity that a particular heartbeat indicates a PVC beat. For example, when a system or computing device that determines that a particular heartbeat indicates a PVC beat has higher specificity and sensitivity, the number of false positives may be reduced. For example, using a machine learning model as described in the present disclosure may require less power because reducing the number of false positives may require less PVC communications from the IMD to other computing devices and improve the life of the IMD when it is subcutaneously implanted in a patient's body for many years of use. In some examples, using a machine learning model as described in the present disclosure may result in higher specificity and sensitivity for determining whether a particular heartbeat indicates a PVC beat. This higher specificity and sensitivity may increase the reliability of another device, user, and / or clinician in determining the accuracy of whether a particular heartbeat indicates a PVC beat. In some examples, such improved reliability regarding the accuracy of determining whether a particular heartbeat indicates a PVC can result in improved usefulness of the system or computing device because a clinician, user, or other computing device may not use and / or may not rely on determinations that are not at or above specificity and sensitivity thresholds. The systems and techniques of the present disclosure using machine learning models can also more flexibly classify or predict PVCs from specific portions of the EGM by eliminating the need to configure explicit rule sets in the IMD, which may otherwise be too large in size to be practically implemented and processed for each new portion of the EGM sensed for the patient. In addition, reducing the number of true and false positives of PVCs when using the techniques of the present disclosure can reduce the clinical burden on physicians and other caregivers to review and identify true positive PVCs. Physicians and caregivers can also provide better customized care, treatment, and interventions for patients experiencing PVCs through more accurate PVC classifications provided by machine learning models used with the techniques of the present disclosure.

[0010] In one example, a medical system includes: a plurality of electrodes configured to sense an electrocardiogram of a patient; and a processing circuit configured to: perform feature extraction of the sensed electrocardiogram to extract a plurality of features from the sensed electrocardiogram; convert one or more of the extracted plurality of features into a corresponding feature image; apply a machine learning model trained using electrocardiogram data for a plurality of patients to the image of the sensed electrocardiogram and at least the corresponding feature image to determine whether one or more specific heartbeats in the sensed electrocardiogram indicate a premature ventricular contraction (PVC); and in response to the determination that the one or more specific heartbeats indicate a PVC beat, output a classification that the one or more specific heartbeats are PVC beats.

[0011] As another example, a computing device includes: a memory; and a processing circuit coupled to the memory, the processing circuit being configured to: receive a sensed electrocardiogram of a patient; perform feature extraction of the sensed electrocardiogram to extract multiple features from the sensed electrocardiogram; convert one or more of the extracted multiple features into a corresponding feature image; apply a machine learning model trained using electrocardiogram data for multiple patients to the image of the sensed electrocardiogram and at least the corresponding feature image to determine whether one or more specific heartbeats in the sensed electrocardiogram indicate a premature ventricular contraction (PVC); and in response to the determination that the one or more specific heartbeats indicate a PVC beat, output a classification that the one or more specific heartbeats are PVC beats.

[0012] As another example, a method includes: receiving a sensed electrocardiogram of a patient; performing feature extraction of the sensed electrocardiogram to extract multiple features from the sensed electrocardiogram; converting one or more of the extracted multiple features into a corresponding feature image; applying a machine learning model trained using electrocardiogram data for multiple patients to the image of the sensed electrocardiogram and at least the corresponding feature image to determine whether one or more specific heartbeats in the sensed electrocardiogram indicate a premature ventricular contraction (PVC); and in response to the determination that the one or more specific heartbeats indicate a PVC beat, outputting a classification that the one or more specific heartbeats are PVC beats.

[0013] The present disclosure is intended to provide an overview of the subject matter described in the present disclosure. The present disclosure is not intended to provide an exclusive or exhaustive explanation of the systems, devices, and methods described in detail in the following figures and description. Further details of one or more examples of the present disclosure are set forth in the following figures and description. Other features, objectives, and advantages will be apparent from the description and drawings, as well as from the claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The environment of an example medical system is illustrated in conjunction with a patient.

[0015] Figure 2 is an example Figure 1 A functional block diagram of an example configuration of an implantable medical device (IMD) for a medical system.

[0016] Figure 3 is an example Figure 1 and Figure 2 Conceptual side view of an example configuration of an IMD.

[0017] Figure 4A is an example Figures 1 to 3 Conceptual side view of an example configuration of an IMD.

[0018] Figure 4B is an example Figure 4A Schematic diagram of an example configuration of an IMD.

[0019] Figure 4C is an example Figures 1 to 3 Conceptual side view of an example configuration of an IMD.

[0020] Figure 5 is an example Figure 1 A functional block diagram of an example configuration of an external device to FIG. 4 .

[0021] Figure 6 is a block diagram illustrating an example system including an access point, a network, an external computing device such as a server, and one or more other computing devices that may be coupled to Figures 1 to 5 of IMDs and external devices.

[0022] Figure 7A is a graph illustrating a cardiac EGM according to some examples of the present disclosure.

[0023] Figure 7B is a graph illustrating multiple correlation coefficients determined from a cardiac EGM according to some examples of the present disclosure.

[0024] Figure 7C is a graph illustrating feature images of multiple correlation coefficients according to some examples disclosed herein.

[0025] Figure 8A is a graph illustrating multiple RR intervals determined from a cardiac EGM according to some examples of the present disclosure.

[0026] Figure 8B is a graph illustrating characteristic images of multiple RR intervals according to some examples of the present disclosure.

[0027] Figure 9is a graph illustrating weighting of cardiac EGM based images according to some examples of the present disclosure.

[0028] Figure 10 is a graph illustrating a weighted image of extracted features from a cardiac EGM of a normal heartbeat according to some examples of the present disclosure.

[0029] Figure 11A and Figure 11B is a graph illustrating a non-axis-normalized weighted image of extracted features from a cardiac EGM according to some examples of the present disclosure.

[0030] Figure 12A and Figure 12B is a graph illustrating normalized and weighted images of extracted features from a cardiac EGM according to some examples of the present disclosure.

[0031] Figure 13 is a graph illustrating the results of normalized and weighted images of extracted features from cardiac EGM applied to a machine learning model according to some examples currently disclosed.

[0032] Figure 14 is an example of some examples disclosed in the present disclosure. Figure 13 Plot of the confusion matrix for the validation dataset.

[0033] Figure 15 is a conceptual diagram illustrating an example machine learning model configured to determine whether a particular heartbeat in a sensed electrocardiogram is a PVC beat and / or a score indicating whether a particular heartbeat may be a PVC beat.

[0034] Figure 16 is a conceptual diagram illustrating an example training process for an artificial intelligence model according to an example of the present disclosure.

[0035] Figure 17 is a flow chart illustrating an example technique for operating a system to determine whether a particular beat in a sensed electrocardiogram is a PVC beat.

[0036] Like reference characters represent like elements throughout the specification and drawings. DETAILED DESCRIPTION

[0037] Various types of medical devices sense cardiac EGM. In some examples, the EGM may also include an electrocardiogram (ECG or EKG). Some medical devices that sense cardiac EGM are non-invasive, for example using multiple electrodes placed in contact with an external portion of the patient, such as at various locations on the patient's skin. As an example, the electrodes used to monitor cardiac EGM in these non-invasive procedures can be attached to the patient using an adhesive, a cloth tape, a belt, or a vest, and electrically coupled to a monitoring device, such as an electrocardiograph, a Holter monitor, or other electronic device. The electrodes are configured to sense electrical signals associated with the electrical activity of the patient's heart or other cardiac tissue, and provide these sensed electrical signals to an electronic device for further processing and / or display of the electrical signals. Non-invasive devices and methods can be utilized on a temporary basis, for example to monitor patients during clinical visits, such as during a doctor's appointment, or for example within a predetermined time period, such as a day (twenty-four hours), or a period of several days.

[0038] External devices that can be used to non-invasively sense and monitor cardiac EGMs include wearable devices such as patches, watches, or necklaces having electrodes configured to contact the patient's skin. One example of a wearable physiological monitor configured to sense cardiac EGMs is the SEEQ, commercially available from Medtronic plc of Dublin, Ireland. TM Mobile cardiac telemetry systems. Such external devices can facilitate relatively long-term monitoring of patients during normal daily activities and can periodically transmit the collected data to a web service such as Medtronic's Carelink TM network.

[0039] Some implantable medical devices (IMDs) also sense and monitor cardiac EGMs. The electrodes used by the IMD to sense the cardiac EGMs are typically integrated with the housing of the IMD and / or coupled to the IMD via one or more elongated leads. Example IMDs that monitor cardiac EGMs include: pacemakers and implantable cardioverter-defibrillators, which may be coupled to intravascular or extravascular leads; and pacemakers having a housing configured for implantation within the heart, which may be leadless. An example of a pacemaker configured for intracardiac implantation is the Micra 1000, commercially available from Medtronic. TM Transcatheter pacing systems. Some IMDs that do not provide therapy (e.g., implantable patient monitors) sense cardiac EGMs. An example of such an IMD is the Reveal LINQ II commercially available from Medtronic plc. TM and LINQ II TMInsertable cardiac monitors (ICMs), which can be inserted subcutaneously. Such IMDs can facilitate relatively long-term monitoring of patients during normal daily activities and can periodically transmit collected data to a web service, such as Medtronic's Carelink TM network.

[0040] Any medical device configured to sense the cardiac EGM via implanted electrodes or external electrodes, including the examples identified herein, can implement the techniques of the present disclosure for evaluating the cardiac EGM to determine whether a particular heartbeat in the cardiac EGM is a PVC. For example, features can be extracted from the cardiac EGM, the extracted features can be modified into corresponding feature images, and the image of the EGM and one or more feature images can be applied to a machine learning model to determine whether a particular heartbeat is a PVC, which results in an accurate determination of the PVC with high specificity and sensitivity. The techniques of the present disclosure for determining whether a particular heartbeat is a PVC can facilitate the determination of PVC burden, cardiac health, and risk of sudden cardiac death, and may lead to clinical interventions to inhibit PVCs, such as medications and PVC ablation.

[0041] Figure 1 The example medical system 2 environment in conjunction with a patient 4 according to one or more techniques of the present disclosure is illustrated. The example techniques may be used with an IMD 10 that may be used with an external device 12 and Figure 1 In some examples, IMD 10 is implanted outside the chest of patient 4 (e.g., implanted subcutaneously). Figure 1 The IMD 10 may be positioned near the sternum at or just below the level of the patient's 4 heart, for example at least partially within the outline of the heart. The IMD 10 includes a plurality of electrodes ( Figure 1 10 is not shown in the figure) and is configured to sense cardiac EGM via a plurality of electrodes. In some examples, the IMD 10 employs Reveal LINQ TM or LINQ IIICM TM of the form, or similar to, for example, LINQ TM A version of an ICM or a modified version of another ICM.

[0042] External device 12 may be a computing device having a display viewable by a user and an interface (i.e., a user input mechanism) for providing input to external device 12. In some examples, external device 12 may be a laptop computer, a tablet computer, a workstation, one or more servers, a cellular phone, a smart phone, a personal digital assistant, or another computing device that can run an application that enables the computing device to interact with IMD 10. External device 12 is configured to communicate with IMD 10 and, optionally, with another computing device (e.g., a computer) via wireless communication. Figure 1 For example, the external device 12 may communicate via near field communication technology (e.g., inductive coupling, NFC, or other communication technology that can operate at a range of less than 10 cm to 20 cm) and far field communication technology (e.g., according to 802.11 or RF telemetry or other communication technologies that can operate at a range greater than near field communication technologies) based on a set of specifications.

[0043] In some examples, external device 12 may be a wearable computing device 12B or additionally include a wearable computing device 12B. Wearable computing device 12B may include electrodes and other sensors to sense physiological signals of patient 4, and may collect and store physiological data and detect seizures based on such signals. Wearable computing device 12B may be incorporated into the apparel of patient 4, such as in clothing, shoes, glasses, a watch or wristband, a hat, etc. In some examples, computing device 12B is a smartwatch or other accessory or peripheral device for external device 12, such as when external device 12 is a smartphone or tablet.

[0044] The external device 12 may be used to configure the operating parameters of the IMD 10. The external device 12 may be used to retrieve data from the IMD 10. The retrieved data may include values ​​of physiological parameters measured by the IMD 10, indications of arrhythmias or other disease episodes detected by the IMD 10, and physiological signals recorded by the IMD 10. For example, the external device 12 may retrieve information related to the detection of PVCs by the IMD 10, such as a count or other quantification of the PVCs, e.g., over a period of time since the external device last retrieved the information. The external device 12 may also retrieve cardiac EGM segments recorded by the IMD 10, e.g., because the IMD 10 determined that an arrhythmia or other disease episode occurred during the segment, or in response to a request to record a segment from the patient 4 or another user. As described below with respect to Figure 6 As discussed in greater detail, one or more remote computing devices may interact with IMD 10 via a network in a manner similar to external device 12 , such as to program IMD 10 and / or retrieve data from IMD 10 .

[0045] Processing circuitry of medical system 2, such as processing circuitry of IMD 10, external device 12, and / or one or more other computing devices, may be configured to perform example techniques of the present disclosure for determining whether a particular heartbeat is a PVC. In some examples, processing circuitry of medical system 2 may be configured to perform example techniques of the present disclosure for determining whether a cardiac EGM includes a PVC doublet, triplet, and / or non-sustained ventricular tachycardia (VT). In some examples, processing circuitry of medical system 2 analyzes cardiac EGM sensed by IMD 10 to determine whether a particular heartbeat is a PVC based on whether a particular heartbeat and adjacent heartbeats (e.g., comparisons between these heartbeats) in the cardiac EGM meet multiple criteria. In some examples, processing circuitry of medical system 2 analyzes cardiac EGM sensed by IMD 10 to determine whether a particular heartbeat is a PVC based on whether a particular heartbeat pair or heartbeat triplet in the cardiac EGM and adjacent heartbeats (e.g., comparisons between these heartbeats) of the heartbeat pair or heartbeat triplet meet multiple criteria to determine whether a particular heartbeat pair or heartbeat triplet is a corresponding PVC doublet or triplet. The criteria may include noise criteria, inter-depolarization interval (e.g., RR interval) criteria, and / or morphology criteria, as described in more detail below. Although described in the context of an example in which the IMD 10 that senses the cardiac EGM includes an insertable cardiac monitor, example systems including one or more implantable or external devices of any type configured to sense the cardiac EGM may be configured to implement the techniques of the present disclosure.

[0046] Figure 2 is an example of one or more techniques described herein. Figure 1 1 is a functional block diagram of an example configuration of an IMD 10. In the illustrated example, IMD 10 includes electrodes 16A and 16B (collectively, "electrodes 16"), antenna 26, processing circuitry 50, sensing circuitry 52, communication circuitry 54, storage device 56, switching circuitry 58, and sensor 62. Although the illustrated example includes two electrodes 16, in some examples, an IMD that includes or is coupled to more than two electrodes 16 may implement the techniques of this disclosure.

[0047] Processing circuit 50 may include fixed function circuits and / or programmable processing circuits. Processing circuit 50 may include any one or more of a microprocessor, a controller, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or equivalent discrete or analog logic circuits. In some examples, processing circuit 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) and other discrete or integrated logic circuits. The functions attributed to processing circuit 50 herein may be embodied in software, firmware, hardware, or any combination thereof.

[0048] The sensing circuit 52 can be selectively coupled to the electrode 16 via the switching circuit 58, for example, to select the electrode 16 and polarity for sensing the cardiac EGM, referred to as a sensing vector, as controlled by the processing circuit 50. The sensing circuit 52 can sense signals from the electrode 16, for example to generate a cardiac EGM, in order to monitor the electrical activity of the heart. As an example, the sensing circuit 52 can also monitor signals from the sensor 62, which can include one or more accelerometers, pressure sensors, and / or optical sensors. In some examples, in addition to the sensed cardiac EGM, the signal received from the sensor 62 can also be used to detect whether a particular heartbeat is a PVC. In some examples, the sensing circuit 52 may include one or more filters and amplifiers for filtering and amplifying the signals received from the electrode 16 and / or the sensor 62.

[0049] The sensing circuit 52 and / or the processing circuit 50 may be configured to detect cardiac depolarization (e.g., a P wave for atrial depolarization or an R wave for ventricular depolarization) when the cardiac EGM amplitude exceeds a sensing threshold. In some examples, for cardiac depolarization detection, the sensing circuit 52 may include a rectifier, a filter, an amplifier, a comparator, and / or an analog-to-digital converter. In some examples, the sensing circuit 52 may output an indication to the processing circuit 50 in response to the sensing of cardiac depolarization. In this manner, the processing circuit 50 may receive an indication of detected cardiac depolarization corresponding to the presence of detected R and P waves in the corresponding chambers of the heart. The processing circuit 50 may use the indication of the detected R and P waves to determine the depolarization interval, the heart rate, and detect arrhythmias, such as tachyarrhythmias and asystole.

[0050] In accordance with the techniques of the present disclosure, sensing circuitry 52 may also provide one or more digitized cardiac EGM signals to processing circuitry 50 for analysis, e.g., for cardiac rhythm differentiation, and / or for analysis to determine whether one or more PVC detection criteria are met. In some examples, processing circuitry 50 may store the digitized cardiac EGM in storage device 56. In accordance with the techniques of the present disclosure, processing circuitry 50 of IMD 10 and / or processing circuitry of another device that retrieves data from IMD 10 may analyze the cardiac EGM to determine whether one or more PVC detection criteria are met.

[0051] Communication circuitry 54 may include any suitable hardware, firmware, software, or any combination thereof, for communicating with another device, such as external device 12, another networked computing device, or another IMD or sensor. Under the control of processing circuitry 50, communication circuitry 54 may receive downlink telemetry from external device 12 or another device and send uplink telemetry to it, via an internal or external antenna, such as antenna 26. In addition, processing circuitry 50 may communicate with external devices (e.g., external device 12) and computer networks (such as Medtronic The antenna 26 and the communication circuit 54 may be configured to transmit and / or receive signals via inductive coupling, electromagnetic coupling, near field communication (NFC), radio frequency (RF) communication, Bluetooth, WiFi, or other proprietary or non-proprietary wireless communication schemes.

[0052] In some examples, storage device 56 includes computer-readable instructions that, when executed by processing circuit 50, cause IMD 10 and processing circuit 50 to perform various functions attributed to IMD 10 and processing circuit 50 herein. Storage device 56 may include any volatile, non-volatile, magnetic, optical, or electrical media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), electrically erasable programmable ROM (EEPROM), flash memory, or any other digital media. As an example, storage device 56 may store programmed values ​​for one or more operating parameters of IMD 10 and / or data collected by IMD 10 for transmission to another device using communication circuit 54. As an example, data stored by storage device 56 and transmitted by communication circuit 54 to one or more other devices may include PVC detection quantification and / or digitized cardiac EGM.

[0053] Figure 3 is an example Figure 1 and Figure 2 1 is a conceptual side view of an example configuration of an IMD 10. Figure 3 In the example shown, IMD 10 may include a leadless subcutaneously implantable monitoring device having housing 15 and insulating cover 76. Electrodes 16A and 16B may be formed or disposed on an outer surface of cover 76. Figure 2The described circuits 50 to 62 may be formed or placed on an inner surface of the cover 76 or within the housing 15. In the illustrated example, the antenna 26 is formed or placed on the inner surface of the cover 76, but in some examples, may be formed or placed on the outer surface. In some examples, one or more sensors 62 may be formed or placed on the outer surface of the cover 76. In some examples, the insulating cover 76 may be positioned over the open housing 15 so that the housing 15 and the cover 76 surround the antenna 26 and the circuits 50 to 62 and protect the antenna and the circuits from fluids (such as body fluids).

[0054] One or more of the antenna 26 or the circuits 50 to 62 may be formed on the inner side of the insulating cover 76, such as by using flip chip technology. The insulating cover 76 may be flipped onto the housing 15. When flipped and placed onto the housing 15, the components of the IMD 10 formed on the inner side of the insulating cover 76 may be positioned in the gap 78 defined by the housing 15. The electrode 16 may be electrically connected to the switching circuit 58 by one or more through holes (not shown) formed through the insulating cover 76. The insulating cover 76 may be formed of sapphire (i.e., corundum), glass, polyparaxylene, and / or any other suitable insulating material. The housing 15 may be formed of titanium or any other suitable material (e.g., a biocompatible material). The electrode 16 may be formed of any one of stainless steel, titanium, platinum, iridium, or their alloys. In addition, the electrode 16 may be coated with a material such as titanium nitride or fractal titanium nitride, but other suitable materials and coatings for such electrodes may be used.

[0055] Figure 4A is a conceptual diagram illustrating an example IMD 10A, which may be an ICM Figures 1 to 3 Example configuration of an IMD 10. Figure 4A In the example shown, IMD 10A can be embodied as a monitoring device having housing 15, proximal electrode 16A, and distal electrode 16B. Housing 15 can also include first major surface 14, second major surface 18, proximal end 20, and distal end 22. Housing 15 encloses electronic circuitry located within IMD 10A and protects the circuitry contained therein from bodily fluids. Electrical feedthroughs provide electrical connections for electrodes 16A and 16B.

[0056] exist Figure 4A In the example shown, IMD 10A is defined by a length L, a width W, and a thickness or depth D, and is in the form of an elongated rectangular prism, where the length L is much greater than the width W, which in turn is greater than the depth D. In one example, the geometry of IMD 10A—particularly the width W being greater than the depth D—is selected to allow IMD 10A to be inserted beneath the patient's skin using a minimally invasive procedure and maintained in a desired orientation during insertion. For example, Figure 4AThe device shown includes radial asymmetry (particularly rectangular shape) along the longitudinal axis, which maintains the device in the correct orientation after insertion. For example, the spacing between the proximal electrode 64 and the distal electrode 66 can be in the range of 30 millimeters (mm) to 55mm, 35mm to 55mm and 40mm to 55mm, and can be any range or separate spacing of 25mm to 60mm. In addition, the IMD 10A can have a length L in the range of 30mm to about 70mm. In other examples, the length L can be in the range of 5mm to 60mm, 15mm to 50mm, 40mm to 60mm, 45mm to 60mm, and can be any length or length range between about 5mm and about 80mm. In addition, the width W of the main surface 14 can be in the range of 5mm to 15mm, 3mm to 10mm, and can be any single width or width range between 3mm and 15mm. The thickness of the depth D of the IMD 10A can be in the range of 2mm to 9mm. In other examples, the depth D of the IMD 10A can be in the range of 2 mm to 5 mm, can be in the range of 5 mm to 15 mm, and can be any single depth or depth range of 2 mm to 15 mm. In addition, the IMD 10A according to examples of the present disclosure has a geometry and size designed for ease of implantation and patient comfort. The examples of the IMD 10A described in the present disclosure can have a volume of three cubic centimeters (cm) or less, 1.5 cubic centimeters or less, or any volume between three cubic centimeters and 1.5 cubic centimeters.

[0057] exist Figure 4A In the example shown, once inserted into the patient, the first major surface 14 faces outward, toward the patient's skin, and the second major surface 18 is located opposite the first major surface 14. Figure 4A In the example shown, proximal end 20 and distal end 22 are rounded to reduce discomfort and irritation to surrounding tissue once inserted beneath the patient's skin. IMD 10A, including instruments and methods for inserting IMD 10, is described, for example, in U.S. Patent Publication No. 2014 / 0276928, which is incorporated herein by reference in its entirety.

[0058] Proximal electrode 16A and distal electrode 16B are used to sense cardiac signals, such as intrathoracic or extrathoracic EGM signals, which may be submuscular or subcutaneous. EGM signals may be stored in a memory of IMD 10A, and the data may be transmitted via integrated antenna 30A to another medical device, which may be another implanted device or an external device such as external device 12. In some examples, electrodes 16A and 16B may additionally or alternatively be used to sense any biopotential signal of interest from any implanted location, which may be, for example, an EGM, EEG, EMG, or neural signal.

[0059] exist Figure 4A In the example shown, the proximal electrode 16A is near the proximal end 20, and the distal electrode 16B is near the distal end 22. In this example, the distal electrode 16B is not limited to a flat, outwardly facing surface, but can extend from the first major surface 14 around the rounded edge 24 and / or the end surface 25 to the second major surface 18, so that the electrode 16B has a three-dimensional curved configuration. In some examples, the electrode 16B is a non-insulated portion of the metal (e.g., titanium) portion of the housing 15.

[0060] exist Figure 4A In the example shown, the proximal electrode 16A is located on the first major surface 14 and is substantially flat and outward facing. However, in other examples, the proximal electrode 16A can utilize a three-dimensional curved configuration of the distal electrode 16B, thereby providing a three-dimensional proximal electrode (not shown in this example). Similarly, in other examples, the distal electrode 16B can utilize a substantially flat outward facing electrode located on the first major surface 14, which is similar to the electrode shown with respect to the proximal electrode 16A.

[0061] Various electrode configurations allow for configurations in which the proximal electrode 16A and the distal electrode 16B are located on both the first major surface 14 and the second major surface 18. In other configurations, such as Figure 4A In the configuration shown, only one of proximal electrode 16A and distal electrode 16B is located on both major surfaces 14 and 18, while in other configurations, both proximal electrode 16A and distal electrode 16B are located on one of first major surface 14 or second major surface 18 (i.e., proximal electrode 16A is located on first major surface 14 and distal electrode 16B is located on second major surface 18). In another example, IMD 10A may include electrodes on both major surfaces 14 and 18 at or near the proximal and distal ends of the device, such that a total of four electrodes are included on IMD 10A. Electrodes 16A and 16B may be formed of a variety of different types of biocompatible conductive materials (e.g., stainless steel, titanium, platinum, iridium, or alloys thereof), and may utilize one or more coatings, such as titanium nitride or fractal titanium nitride.

[0062] existFigure 4A In the example shown, the proximal end 20 includes a head assembly 28 that includes one or more of the proximal electrode 16A, an integrated antenna 30A, an anti-migration protrusion 32, and / or a suture hole 34. The integrated antenna 30A is located on the same major surface (e.g., the first major surface 14) as the proximal electrode 16A and is also included as part of the head assembly 28. The integrated antenna 30A allows the IMD 10A to transmit and / or receive data. In other examples, the integrated antenna 30A may be formed on a major surface opposite the proximal electrode 16A, or may be incorporated into the housing 15 of the IMD 10A. Figure 4A In the example shown, the anti-migration protrusion 32 is located adjacent to the integrated antenna 30A and protrudes away from the first major surface 14 to prevent longitudinal movement of the device. Figure 4A In the example shown, the anti-migration protrusions 32 include a plurality (e.g., nine) of small bumps or protrusions extending away from the first major surface 14. As discussed above, in other examples, the anti-migration protrusions 32 can be located on the major surface opposite the proximal electrode 16A and / or the integrated antenna 30A. Figure 4A In the example shown, head assembly 28 includes suture holes 34 that provide another means of securing IMD 10A to the patient to prevent movement after insertion. In the example shown, suture holes 34 are located adjacent to proximal electrode 16A. In one example, head assembly 28 is a molded head assembly made of a polymer or plastic material that may be integrated with or separate from the main portion of IMD 10A.

[0063] Figure 4B According to the embodiments of the present disclosure Figure 4A 10A is a functional schematic diagram of the IMD 10A shown. The IMD 10A may include a housing 15, a proximal electrode 16B located at the proximal end 24, a distal electrode 16A located at the distal end 22, an integrated antenna 30A, a circuit 400, and a power supply 402. In particular, the circuit 400 is coupled to the proximal electrode 16B and the distal electrode 16A to sense cardiac signals and monitor events. The circuit 400 may also be connected to transmit and receive communications via the integrated antenna 30A. The power supply 402 provides power to the circuit 400 and any other components that require power. The power supply 402 may include one or more energy storage devices, such as one or more rechargeable or non-rechargeable batteries. In some examples, the circuit 400 includes a processing circuit 50 and a storage device 56, such as a memory, such as Figure 2 As shown, the memory is operably coupled to the processing circuit 56 and is configured to store the machine learning model.

[0064] exist Figure 4BIn the example shown, the circuit 400 can receive the raw EGM signal monitored by the proximal electrode 16B and the distal electrode 16A. The circuit 400 may include a component / module for converting the raw EGM signal into a processed EGM signal, which can be analyzed to detect a sensing event. Although not shown, the circuit 400 may include any discrete and / or integrated electronic circuit components that implement analog and / or digital circuits that can generate the described functions for analyzing EGM signals to detect / verify PVC. For example, the circuit 400 may include analog circuits, such as pre-amplification circuits, filtering circuits, and / or other analog signal conditioning circuits. These modules may also include digital circuits, such as digital filters, combinational or sequential logic circuits, state machines, integrated circuits, processors (shared, dedicated, or groups) that execute one or more software or firmware programs, memory devices, or any other suitable components or combinations thereof that provide the described functionality.

[0065] In one example, the circuit 400 includes a sensing unit for monitoring EGM signals detected by the corresponding proximal electrode 16A and distal electrode 16B, respectively. In one example, the circuit 400 includes a processing circuit 50 for receiving information about the sensed event and implementing one or more algorithms for determining whether a PVC beat has occurred. In addition, the analog voltage signals received from the electrodes 16A and 16B can be passed to an analog-to-digital (A / D) converter included in the circuit 400 and stored in a memory unit (not shown) included as part of the circuit 400 for subsequent analysis using firmware executed by a processor included as part of the circuit 400.

[0066] In some examples, housing 15 may be a hermetically sealed housing configured for subcutaneous implantation within a patient, with at least power source 402, memory, and processing circuitry 50 within the hermetically sealed housing.

[0067] Figure 4C is a perspective view illustrating another IMD 10B, which may be from Figures 1 to 3 Another example configuration of IMD 10 is shown. Figure 4C The IMD 10B can be basically similar to Figure 4A The IMD 10A is configured and the differences between them are discussed in this article.

[0068] IMD 10B may include a leadless subcutaneously implantable monitoring device, such as an ICM. IMD 10B includes a housing having a base 40 and an insulating cover 42. Proximal electrode 16C and distal electrode 16D may be formed or placed on an outer surface of cover 42. For example, as described below with respect to Figure 3The various circuits and components of the described IMD 10B may be formed or placed on an inner surface of cover 42 or within base 40. In some examples, a battery or other power source for IMD 10B may be included within base 40. In the illustrated example, antenna 30B is formed or placed on an outer surface of cover 42, but in some examples may be formed or placed on an inner surface. In some examples, insulating cover 42 may be positioned over open base 40 such that base 40 and cover 42 enclose the circuits and other components and protect them from fluids such as bodily fluids.

[0069] Circuits and components can be formed on the inner side of insulating cover 42, such as by using flip chip technology. Insulating cover 42 can be flipped onto base 40. When flipped and placed onto base 40, the components formed on the inner side of insulating cover 42 of IMD 10B can be positioned in gap 44 defined by base 40. Electrodes 16C and 16D and antenna 30B can be electrically connected to the circuit formed on the inner side of insulating cover 42 through one or more through holes (not shown) formed through insulating cover 42. Insulating cover 42 can be formed of sapphire (i.e., corundum), glass, poly-p-xylene and / or any other suitable insulating material. Base 40 can be formed of titanium or any other suitable material (e.g., biocompatible material). Electrodes 16C and 16D can be formed of any one of stainless steel, titanium, platinum, iridium or their alloys. In addition, electrodes 16C and 16D can be coated with materials such as titanium nitride or fractal titanium nitride, but other suitable materials and coatings for such electrodes can also be used.

[0070] exist Figure 4C In the example shown, the housing of the IMD 10B defines a length L, a width W, and a thickness or depth D, and is in the form of an elongated rectangular prism, wherein the length L is much greater than the width W, which in turn is greater than the depth D, similar to Figure 4C The IMD 10A of the present invention can be configured to have a depth D of 2 mm to 15 mm, 3 mm to 5 mm, or about 4 mm. For example, the spacing between the proximal electrode 64 and the distal electrode 66 can be in the range of 30 millimeters (mm) to 50 mm, 35 mm to 45 mm, or about 40 mm. In addition, the IMD 10B can have a length L in the range of 30 mm to about 70 mm. In other examples, the length L can be in the range of 5 mm to 60 mm, 40 mm to 60 mm, 45 mm to 55 mm, or about 45 mm. In addition, the width W can be in the range of 3 mm to 15 mm, such as about 8 mm. The thickness of the depth D of the IMD 10B can be in the range of 2 mm to 15 mm, 3 mm to 5 mm, or about 4 mm. The IMD 10B can have a volume of three cubic centimeters (cm) or less, or 1.5 cubic centimeters or less (such as about 1.4 cubic centimeters).

[0071] exist Figure 4C In the example shown, once subcutaneously inserted into the patient's body, the outer surface of the cover 42 faces outward, toward the patient's skin. Figure 4C As shown, the proximal end 46 and the distal end 48 are rounded to reduce discomfort and irritation to surrounding tissue once inserted.

[0072] Figure 5 1 is a block diagram illustrating an example configuration of components of the external device 12. Figure 5 In the example of FIG. 8 , the external device 12 includes a processing circuit 80 , a communication circuit 82 , a storage device 84 , and a user interface 86 .

[0073] The processing circuit 80 may include one or more processors configured to implement functions and / or processing instructions for execution within the external device 12. For example, the processing circuit 80 may be capable of processing instructions stored in the storage device 84. The processing circuit 80 may include, for example, a microprocessor, a DSP, an ASIC, an FPGA, or an equivalent discrete or integrated logic circuit or a combination of any of the foregoing devices or circuits. Thus, the processing circuit 80 may include any suitable structure, whether hardware, software, firmware, or any combination thereof, to perform the functions attributed to the processing circuit 80 herein.

[0074] The communication circuit 82 may include any suitable hardware, firmware, software, or any combination thereof for communicating with another device such as the IMD 10. Under the control of the processing circuit 80, the communication circuit 82 may receive downlink telemetry from the IMD 10 or another device, and send uplink telemetry to the IMD or another device. The communication circuit 82 may be configured to transmit or receive signals via inductive coupling, electromagnetic coupling, near field communication (NFC), radio frequency (RF) communication, Bluetooth, WiFi, or other proprietary or non-proprietary wireless communication schemes. The communication circuit 82 may also be configured to communicate with a device other than the IMD 10 via any of various forms of wired and / or wireless communications and / or network protocols.

[0075] Storage device 84 may be configured to store information within external device 12 during operation. Storage device 84 may include a computer-readable storage medium or a computer-readable storage device. In some examples, storage device 84 includes one or more of short-term memory or long-term memory. Storage device 84 may include, for example, RAM, DRAM, SRAM, a magnetic disk, an optical disk, flash memory, or various forms of EPROM or EEPROM. In some examples, storage device 84 is used to store data indicating instructions for execution by processing circuit 80. Storage device 84 may be used by software or applications running on external device 12 to temporarily store information during program execution.

[0076] The data exchanged between the external device 12 and the IMD 10 may include operating parameters. The external device 12 may transmit data including computer-readable instructions that, when implemented by the IMD 10, may control the IMD 10 to change one or more operating parameters and / or export collected data. For example, the processing circuit 80 may transmit instructions to the IMD 10 requesting the IMD 10 to export collected data (e.g., PVC detection data and / or digitized cardiac EGM) to the external device 12. In turn, the external device 12 may receive the data collected from the IMD 10 and store the collected data in the storage device 84. The processing circuit 80 may implement any of the techniques described herein to analyze the cardiac EGM received from the IMD 10, for example, to determine whether a particular heartbeat is a PVC.

[0077] A user, such as a clinician or patient 4, can interact with the external device 12 via the user interface 86. The user interface 86 includes a display (not shown), such as a liquid crystal display (LCD) or a light emitting diode (LED) display or other type of screen, wherein the processing circuit 80 can present information related to the IMD 10, such as a cardiac EGM, an indication of PVC detection, and a quantification of detected PVCs, such as a quantification of PVC burden. In addition, the user interface 86 may include an input mechanism for receiving input from a user. The input mechanism may include, for example, any one or more of a button, a keypad (e.g., an alphanumeric keypad), a peripheral pointing device, a touch screen, or another input mechanism that allows a user to navigate through a user interface presented by the processing circuit 80 of the external device 12 and provide input. In other examples, the user interface 86 also includes an audio circuit for providing auditory notifications, instructions or other sounds to the user, receiving voice commands from the user, or both.

[0078] Figure 6 is a block diagram illustrating an example system including an access point 90, a network 92, an external computing device (such as a server 94), and one or more other computing devices 100A-100N (collectively, “computing devices 100”) that may be coupled to IMD 10 and external device 12 via network 92, in accordance with one or more techniques described herein. In this example, IMD 10 may use communication circuitry 54 to communicate with external device 12 via a first wireless connection and to communicate with access point 90 via a second wireless connection. Figure 5 In the example of FIG. 1 , access point 90 , external device 12 , server 94 , and computing device 100 are interconnected and can communicate with each other via network 92 .

[0079] The access point 90 may include a device that is connected to the network 92 via any of a variety of connections, such as a telephone dial-up, a digital subscriber line (DSL), or a cable modem connection. In other examples, the access point 90 may be coupled to the network 92 via different forms of connection, including a wired connection or a wireless connection. In some examples, the access point 90 may be a user device that can be co-located with the patient, such as a tablet or a smart phone. The IMD 10 may be configured to transmit data, such as PVC detection information, PVC quantification, and / or cardiac EGM, to the access point 90. The access point 90 may then transmit the retrieved data to the server 94 via the network 92.

[0080] In some cases, server 94 may be configured to provide a secure storage site for data that has been collected from IMD 10 and / or external devices 12. In some cases, server 94 may compile the data in a web page or other document via computing device 100 for viewing by a trained professional, such as a clinician. Figure 6 One or more aspects of the illustrated system may be used with a Medtronic The server 94 may be implemented using general network technologies and functionalities similar to those provided by the network. In some examples, the server 94 may communicate with the computing device 100 via the network 92. For example, the server 94 may transmit data analysis such as PVC detection information to the computing device 100, the external device 12, or any other computing device via the network 92. For example, the server 94 may transmit a specific heartbeat as a PVC beat to the computing device 100, the external device 12, or any other computing device via the network 92.

[0081] In some examples, one or more of the computing devices 100 may be a tablet or other smart device located with a clinician, through which the clinician can program the IMD 10, receive warnings from the IMD, and / or query the IMD. For example, the clinician can access data collected by the IMD 10 through the computing device 100, such as when the patient 4 is between clinician visits, to check the status of a medical condition. In some examples, the clinician may enter instructions for medical intervention for the patient 4 into an application executed by the computing device 100, such as based on the status of the patient's condition determined by the IMD 10, the external device 12, the server 94, or any combination thereof, or based on other patient data known to the clinician. Subsequently, the device 100 may transmit instructions for medical intervention to another computing device in the computing device 100 located with the patient 4 or the caregiver of the patient 4. For example, such instructions for medical intervention may include instructions to change the dosage, timing, or selection of a drug, instructions to schedule a clinician visit, or instructions to seek medical attention. In another example, computing device 100 may generate a warning to patient 4 based on the status of patient 4's medical condition, which may enable patient 4 to proactively seek medical attention before receiving instructions for medical intervention. In this way, patient 4 may autonomously take action to address his or her medical condition as needed, which may help improve patient 4's clinical outcomes.

[0082] In by Figure 6 In the example shown, server 94 includes, for example, a storage device 96 and processing circuitry 98 for storing data retrieved from IMD 10. Figure 5Not shown, but computing device 100 may similarly include storage devices and processing circuits. Processing circuit 98 may include one or more processors configured to implement functions and / or process instructions for execution within server 94. For example, processing circuit 98 may be capable of processing instructions stored in memory 96. Processing circuit 98 may include or be coupled to a communication circuit, which may include any suitable hardware, firmware, software, or any combination thereof for communicating with another device. In some examples, a description of a processing circuit 98 output signal (such as a classification) may include processing circuit 98 causing a communication circuit of server 94 to output a signal. Processing circuit 98 may include, for example, a microprocessor, a DSP, an ASIC, an FPGA, or an equivalent discrete or integrated logic circuit or a combination of any of the foregoing devices or circuits. Therefore, processing circuit 98 may include any suitable structure, whether hardware, software, firmware, or any combination thereof, to perform the functions attributed to processing circuit 98 herein. Processing circuitry 98 of server 94 and / or processing circuitry of computing device 100 may implement any of the techniques described herein to analyze cardiac EGMs received from IMD 10, eg, to determine whether a particular heartbeat is a PVC.

[0083] Storage device 96 may include a computer-readable storage medium or a computer-readable storage device. In some examples, memory 96 includes one or more of short-term memory or long-term memory. Storage device 96 may include, for example, RAM, DRAM, SRAM, a magnetic disk, an optical disk, flash memory, or various forms of EPROM or EEPROM. In some examples, storage device 96 is used to store data indicating instructions for execution by processing circuit 98.

[0084] Although techniques for determining whether a particular heartbeat is a PVC beat are primarily described herein (e.g., with respect to Figures 6 to 14 ) is performed by processing circuitry 98 of server 94, but such techniques may be performed in whole or in part by processing circuitry of any one or more devices of system 2 (such as processing circuitry 80 of external device 12, processing circuitry 50 of IMD 10, or processing circuitry of one or more computing devices 100).

[0085] Figure 7Ais a graph illustrating an example of a sensed cardiac EGM signal 120 image that includes one or more specific heartbeats 124, a previous adjacent heartbeat 122 immediately before the one or more specific heartbeats 124, and a subsequent adjacent heartbeat 126 immediately after the one or more specific heartbeats 124. The sensed cardiac EGM signal 120 may also include data about the heartbeat immediately before the previous adjacent heartbeat 122, and may include data about the heartbeat immediately after the subsequent adjacent heartbeat 126. For example, this data can help determine some RR intervals. In some examples, the sensed cardiac EGM signal 120 may be a window or segment that starts at 75 samples before the R peak of the heartbeat 122 and ends at 95 samples after the R peak of the heartbeat 126. However, other sampling amounts may also be selected. Although Figure 7A An example of a sensed cardiac EGM signal having three consecutive heartbeats is illustrated, but in some examples, the sensed cardiac EGM signal may include more than three consecutive heartbeats, such as but not limited to four consecutive heartbeats or five consecutive heartbeats. In some examples, one or more specific heartbeats 124 may include two consecutive heartbeats, such as for detecting a PVC doublet, or may include three consecutive heartbeats, such as for detecting a PVC triplet. In these examples, the previous adjacent heartbeat 122 and the subsequent adjacent heartbeat 126 may be about two consecutive heartbeats of one or more specific heartbeats 124 in a PVC doublet or three consecutive heartbeats of one or more specific heartbeats 124 in a PVC triplet. Although the following discussion as a whole relates to one or more specific heartbeats 124 being a specific heartbeat 124, it is not necessarily limited to the specific heartbeat 124 being one heartbeat, and may be, for example, two consecutive heartbeats or three consecutive heartbeats, as discussed above.

[0086] For each beat, different features of the sensed cardiac EGM signal 120 may be extracted to help distinguish PVC beats from non-PVC beats, which are then fed to a machine learning model for classification. In some examples, initially, a heartbeat (such as a particular heartbeat 124) may be processed to determine whether the heartbeat is noisy. If the heartbeat is determined to be noisy, the machine learning model may be bypassed for the heartbeat, and the heartbeat may be automatically classified as noise or a non-PVC beat.

[0087] Processing circuitry 98 can detect R-wave peaks of corresponding heartbeats in an EGM signal, such as sensed cardiac EGM signal 120. To detect noise, processing circuitry 98 can determine whether the R-wave of a heartbeat, such as a particular heartbeat 124, has noise. In some examples, to detect whether the R-wave of a heartbeat has noise, processing circuitry 98 can be configured to: 1) consider the number of samples before (e.g., 25 samples) and after (e.g., 25 samples) an R-wave peak detected by the IMD 10 during an initial collection of the EGM signal; 2) subtract an average value from the section of the EGM signal and calculate the absolute value of the section; 3) determine the sample with the largest absolute value as the R-wave peak, which may be different from the R-wave peak identified by the IMD 10 when the EGM signal was initially collected; 4) after determining the R-wave peak, consider a 32-sample window around the R-wave peak (15 samples before and 16 samples after); 5) determine a first difference and a second difference of the window of the signal; 6) for each sample, determine whether there is a sign change between the current sample and the next sample for the first difference; 7) determine the number of such samples where a detected sign change exists and the corresponding second difference is greater than or less than a threshold (e.g., greater than 40 or less than -40); 8) if the number of such samples is greater than a threshold number of samples (e.g., 8), then the R-wave is considered to have noise.

[0088] If the R-wave is determined to have noise, processing circuitry 98 can mark the current beat as a non-PVC beat and bypass the machine learning model for that beat. If the R-wave is determined to have no noise, processing circuitry 98 can extract features from the sensed EGM signal 120 and feed the sensed cardiac EGM signal 120 and the different extracted features to the machine learning model for classification, as discussed below.

[0089] There are several characteristics of the cardiac EGM signal that can help distinguish PVCs from normal heartbeats. For example, the coupling interval for a PVC heartbeat can be shorter than the coupling interval for a normal heartbeat. Thus, the RR interval between a PVC heartbeat and the previous QRS complex can be shorter than the RR interval for a normal heartbeat. In some examples, a compensatory pause can occur after a PVC beat, which results in the RR interval between the PVC heartbeat and the subsequent heartbeat being longer than the RR interval between a normal heartbeat and the subsequent heartbeat. In some examples, the QRS and T-wave morphologies can be different for PVC heartbeats as well as for normal heartbeats. In some examples, there may be no P-wave before the QRS complex for a PVC heartbeat.

[0090] In some examples, one or more sensed cardiac EGM signals 120 may be received periodically. For example, the sensed cardiac EGM signals may be received daily, hourly, every two hours, every thirty minutes, or at any other time interval. In some examples, a clinician and / or user may request when to obtain the cardiac EGM signals.

[0091] Figure 7B is an example graph illustrating a plurality of correlation coefficients determined for Figure 7A the image of the sensed cardiac EGM signal 120 in. As Figure 7B shown in the example of, C1 represents the correlation coefficient between heartbeat 122 and heartbeat 124, C2 represents the correlation coefficient between heartbeat 122 and heartbeat 126, and C3 represents the correlation coefficient between heartbeat 124 and heartbeat 126. The processing circuit 98 may perform feature extraction and extract correlation coefficients from the sensed cardiac EGM signal 120, such as C1, C2, and C3. In some examples, the processing circuit 98 may consider 25 samples before and 75 samples after each corresponding R-wave peak of heartbeats 122, 124, 126 in the sensed cardiac EGM signal when determining the correlation coefficients C1, C2, C3.

[0092] As Figure 7C shown in the example of, the processing circuit 98 may convert the extracted correlation coefficients (such as C1, C2, and C3) into a feature image 130 for application to a machine learning model. In some examples, the feature image 130 may be a digital array, such as a two-dimensional digital array or a one-dimensional digital array. In some examples, the feature image 130 may be a digital image. For example, the processing circuit 98 may use a programming script such as a script to convert the extracted correlation coefficients into a corresponding digital image (e.g., an example of the feature image 130) such as a JPEG image for application to a machine learning model. In some examples, when a particular heartbeat 124 is a PVC beat, the values of the correlation coefficients C1 and C3 may be small because the morphology of the particular heartbeat 124 will be different from the adjacent normal heartbeats 122, 126 of the particular heartbeat 124. Additionally, when a particular heartbeat 124 is a PVC beat, C2 may be large because the morphologies of the two normal heartbeats 122, 126 will be similar.

[0093] Figure 8A is an example graph illustrating a plurality of RR intervals determined for Figure 7A the sensed cardiac EGM signal 120 in. As Figure 8AAs shown in the example in , it is a graph exemplifying four RR intervals RR1, RR2, RR3, and RR4 of the sensed cardiac EGM 120. The RR interval is determined as the time interval between the maximum R-wave amplitude in a beat and the maximum R-wave amplitude of an adjacent beat. For example, RR1 is the time interval between the maximum R-wave amplitude of heartbeat 122 and the maximum R-wave amplitude of the heartbeat before heartbeat 122. RR2 is the time interval between the maximum R-wave amplitude of heartbeat 122 and the maximum R-wave amplitude of heartbeat 124. RR3 is the time interval between the maximum R-wave amplitude of heartbeat 124 and the maximum R-wave amplitude of heartbeat 126. RR4 is the time interval between the maximum R-wave amplitude of heartbeat 126 and the maximum R-wave amplitude of the heartbeat immediately following heartbeat 126.

[0094] The processing circuit 98 can perform feature extraction and extract RR intervals, such as RR1, RR2, RR3, and RR4, from the sensed cardiac EGM signal 120. In some examples, the processing circuit 98 can consider 25 samples before and 75 samples after each corresponding R-wave peak of the heartbeats in the sensed cardiac EGM signal when determining the RR intervals RR1, RR2, RR3, and RR4.

[0095] As Figure 8B shown in the example in , the processing circuit 98 can convert the extracted RR intervals (such as RR1, RR2, RR3, and RR4) into a feature image 132 for application to a machine learning model. In some examples, when a particular heartbeat 124 is a PVC beat, RR2 can be the shortest RR interval and RR3 can be the longest RR interval because a compensatory pause typically exists after a PVC beat.

[0096] The processing circuit 98 can perform feature extraction and extract one or more of the maximum QRS amplitude 141, minimum QRS amplitude 142, difference 143 between the maximum QRS amplitude and the minimum QRS amplitude, number of samples 144 between the maximum QRS amplitude and the minimum QRS amplitude, maximum slope 145 of the QRS wave, minimum slope 146 of the QRS wave, slope value difference 147 between the maximum slope and the minimum slope of the QRS wave, and number of samples 148 between the maximum slope and the minimum slope of the QRS wave from the sensed cardiac EGM signal 120.

[0097] As Figure 9 shown in the example in , the processing circuit 98 can convert one or more of the extracted values 141, 142, 143, 144, 145, 146, 147, 148 discussed above into corresponding feature images 151, 152, 153, 154, 155, 156, 157, 158 for application to a machine learning model.

[0098] As shown in the example of Figure 9 the processing circuit 98 may weight one or more of the following: the sensed cardiac EGM signal 120 image, the feature correlation coefficient image 130, the feature RR interval image 132, the feature maximum QRS amplitude image 151, the feature minimum QRS amplitude image 152, the feature difference image 153 between the maximum QRS amplitude and the minimum QRS amplitude, the feature sample number image 154 between the maximum QRS amplitude and the minimum QRS amplitude, the feature maximum slope image 155 of the QRS wave, the feature minimum slope image 156 of the QRS wave, the slope value feature difference image 157 between the maximum slope and the minimum slope of the QRS wave, or the feature sample number image 158 between the maximum slope and the minimum slope of the QRS wave. In some examples, weighting an image includes adjusting the corresponding size of the image for application to a machine learning model. For example, if the sensed cardiac EGM signal 120 image has a weight of four, the feature correlation coefficient image 130 has a weight of two, and the feature maximum QRS amplitude image 151 has a weight of one, then the size of the sensed cardiac EGM signal 120 image will be four times as large as the feature maximum QRS amplitude image 151 and twice as large as the feature correlation coefficient image 130.

[0099] In some examples, the machine learning model may assign a greater weight to a larger image than to a smaller image. The processing circuit 98 may provide a greater weight to the extracted feature, which results in a larger feature image of the corresponding extracted feature, which has a greater impact on determining whether a particular heartbeat 124 is a PVC.

[0100] In some examples, the processing circuit 98 may assign the highest weight to the sensed cardiac EGM signal 120 image. For example, in Figure 9 the sensed cardiac EGM signal 120 image is assigned a weight of four. In other examples, the processing circuit 98 may assign different weight values. The processing circuit 98 may assign one or more of the feature correlation coefficient image 130 or the feature RR interval image 132 the next highest weight after the sensed cardiac EGM signal 120 image. For example, in Figure 9 the feature correlation coefficient image 130 or the feature RR interval image 132 is each assigned a weight of 2. In other examples, the processing circuit 98 may assign different weight values.

[0101] For example, the sensed cardiac EGM signal 120 image, the feature correlation coefficient image 130, and the feature RR interval image 132 may have the greatest effect of the feature images of the extracted features in determining whether a particular heartbeat 124 is a PVC. Accordingly, the processing circuit 98 may apply the maximum weight to the sensed cardiac EGM signal 120 image, the feature correlation coefficient image 130, and the feature RR interval image 132

[0102] The processing circuit 98 may assign the lowest weight to one or more of the following: the feature maximum QRS amplitude image 151, the feature minimum QRS amplitude image 152, the feature difference image 153 between the maximum QRS amplitude and the minimum QRS amplitude, the feature number of samples image 154 between the maximum QRS amplitude and the minimum QRS amplitude, the feature maximum slope image 155 of the QRS wave, the feature minimum slope image 156 of the QRS wave, the feature difference image 157 of the slope value between the maximum slope of the QRS wave and the minimum slope of the QRS wave, or the feature number of samples image 158 between the maximum slope of the QRS wave and the minimum slope of the QRS wave. For example, in Figure 9 each of the feature images 151, 152, 153, 154, 155, 156, 157, 158 is assigned a weight of one. In other examples, the processing circuit 98 may assign different weight values.

[0103] For example, in Figure 9 due to the weights applied by the processing circuit 98, the image of the sensed cardiac EGM signal 120 is twice as large as each of the feature correlation coefficient image 130 and the feature RR interval image 132. Additionally, due to the weights applied by the processing circuit 98, the image of the sensed cardiac EGM signal 120 is four times as large as each of the feature images 151, 152, 153, 154, 155, 156, 157, 158.

[0104] The processing circuit 98 may apply one or more of the weighted images 120, 130, 132, 151, 152, 153, 154, 155, 156, 157, 158 to a machine learning model to determine whether a particular heartbeat in the sensed electrocardiogram is a PVC beat. In response to determining that the particular heartbeat is a PVC beat, the processing circuit 98 may cause a classification that the particular heartbeat is a PVC beat to be output. For example, the classification may be output to a clinician computing device.

[0105] Figure 10An example of a weighted image of a cardiac EGM signal 120 image for sensing of normal heartbeats, a feature correlation coefficient image 130, a feature RR interval image 132, a feature maximum QRS amplitude image 151, a feature minimum QRS amplitude image 152, a feature difference image 153 between the maximum QRS amplitude and the minimum QRS amplitude, a feature sample number image 154 between the maximum QRS amplitude and the minimum QRS amplitude, a feature maximum slope image 155 of the QRS complex, a feature minimum slope image 156 of the QRS complex, a feature difference image 157 of the slope values between the maximum slope and the minimum slope of the QRS complex, or a feature sample number image 158 between the maximum slope and the minimum slope of the QRS complex is shown.

[0106] In some examples, the processing circuit 98 can normalize the axes of the sensed cardiac EGM signal 120 image and can apply one or more of the feature images 130, 132, 151, 152, 153, 154, 155, 156, 157, 158 to a machine learning model. The processing circuit 98 can normalize the axes of the weighted sensed cardiac EGM signal 120 image and can apply one or more of the weighted feature images 130, 132, 151, 152, 153, 154, 155, 156, 157, 158 to a machine learning model. In some examples where the processing circuit 98 weights the axes of the image, the processing circuit 98 can weight the cardiac EGM signal 120 image and one or more of the images 130, 132, 151, 152, 153, 154, 155, 156, 157, 158 before or after the processing circuit 98 weights the axes.

[0107] Figure 11A and Figure 11B An example of a weighted image of the sensed cardiac EGM signal 120 image, the feature correlation coefficient image 130, the feature RR interval image 132, the feature maximum QRS amplitude image 151, the feature minimum QRS amplitude image 152, the feature difference image 153 between the maximum QRS amplitude and the minimum QRS amplitude, the feature sample number image 154 between the maximum QRS amplitude and the minimum QRS amplitude, the feature maximum slope image 155 of the QRS complex, the feature minimum slope image 156 of the QRS complex, the feature difference image 157 of the slope values between the maximum slope and the minimum slope of the QRS complex, or the feature sample number image 158 between the maximum slope and the minimum slope of the QRS complex without axis normalization is shown.

[0108] Figure 12A and Figure 12BAn example of processing circuitry 98 is shown that normalizes axes of a weighted image of a sensed cardiac EGM signal 120N image, a feature correlation coefficient image 130N, a feature RR interval image 132N, a feature maximum QRS amplitude image 151N, a feature minimum QRS amplitude image 152N, a feature difference image 153N between the maximum QRS amplitude and the minimum QRS amplitude, a feature sample count image 154N between the maximum QRS amplitude and the minimum QRS amplitude, a feature maximum slope image 155N of the QRS complex, a feature minimum slope image 156N of the QRS complex, a feature difference image 157N of slope values between the maximum slope of the QRS complex and the minimum slope of the QRS complex, or a feature sample count image 158N between the maximum slope of the QRS complex and the minimum slope of the QRS complex for application to a machine learning model.

[0109] In some examples, processing circuitry 98 may normalize one or more axes, such as the y-axis, of the transformed image and / or the weighted image based on minimum and maximum values of each respective feature determined from a development data set. In some examples, processing circuitry 98 may normalize one or more axes of the transformed image and / or the weighted image to between 0 and 1 based on minimum and maximum values in the data set of the image or by dividing each data point in the data set of the respective image by a constant, such as the standard deviation of the data set of the respective image, the maximum value of the data set of the respective image, the minimum value of the data set of the respective image, the average value of the data set of the respective image, the median value of the data set of the respective image, etc.).

[0110] In some examples, by extracting specific features corresponding to PVCs from cardiac EGMs that have been determined in accordance with the inventive techniques of the present disclosure, as discussed above, for application to a machine learning model, the machine learning model can be more focused and spend most of its nodes on determining whether a particular heartbeat is a PVC, which can result in more accurate results.

[0111] In some examples, by extracting one or more specific features from the sensed cardiac EGM signal 120, converting the extracted features into an image, and applying the EGM signal 120 image and the feature image to a machine learning model to determine whether a particular heartbeat 124 is a PVC, the processing circuitry 98 may be able to determine whether a particular heartbeat 124 is a PVC with high accuracy, specificity, and sensitivity. Additionally, in some examples, by weighting and / or normalizing the EGM signal 120 image and the feature image and applying the weighted and / or normalized EGM signal 120 image and feature image to a machine learning model to determine whether a particular heartbeat 124 is a PVC, the processing circuitry 98 may be able to determine whether a particular heartbeat 124 is a PVC with high accuracy, specificity, and sensitivity. This may facilitate determination of PVC burden, cardiac health, and risk of sudden cardiac death and may lead to clinical intervention to suppress PVCs, such as medications and PVC ablation.

[0112] The processing circuitry 98 may be configured to execute an artificial intelligence (AI) engine that operates according to one or more models, such as a machine learning model. The machine learning model may include any number of different types of machine learning models, such as neural networks, deep neural networks, convolutional neural networks, recurrent neural networks, such as long short-term memory networks, dense neural networks, and the like. In some examples, the various feature inputs to the AI engine may be fed as direct inputs into different layers of the network and not necessarily before a convolutional layer. Although described with respect to machine learning models, the techniques described in this disclosure may also be applicable to other types of AI models, including rule-based models, finite state machines, and the like.

[0113] Machine learning generally enables a computing device to analyze input data and identify actions to perform in response to that input data. Each machine learning model may be trained using training data that reflects possible input data. The training data may be labeled or unlabeled (meaning that the correct action to take based on a sample of the training data is either explicitly stated or not explicitly stated, respectively).

[0114] The training of a machine learning model may be supervised (because a designer, such as a computer programmer, may guide the training to direct the machine learning model to identify the correct action given the input data) or unsupervised (because the designer does not guide the machine learning model to identify the correct action given the input data). In some instances, a machine learning model is trained with a combination of labeled and unlabeled training data, a combination of supervised and unsupervised training, or a combination of possible ones of them. Examples of machine learning include nearest neighbor, naive Bayes, decision tree, linear regression, support vector machine, neural network, k-means clustering, Q-learning, temporal difference, deep adversarial network, evolutionary algorithms, or other supervised, unsupervised, semi-supervised, or reinforcement learning algorithms for training one or more models.

[0115] Processing circuit 98 can use machine learning such as deep learning algorithms or models (e.g., neural networks or deep belief networks) to generate a score indicating whether a particular heartbeat 124 may be a PVC to determine whether a particular heartbeat 124 is a PVC. The processing circuit 98 can train a deep learning model to represent the relationship between the images discussed above and whether a particular heartbeat is a PVC. For example, the processing circuit 98 can use images of EGM signals from other patients to train the deep learning model. In some examples, the processing circuit 98 can train the deep learning model by adjusting the weights of the hidden layers of the neural network model according to whether the heartbeat is a PVC to balance the contribution of each input (e.g., the characteristics of the image input).

[0116] Once the deep learning model is trained, the processing circuit 98 can obtain data (such as images of the EGM signal 120 and feature images of features extracted from the EGM signal 120) and apply it to the trained deep learning model. For example, the input image may include a normalized and weighted image of one or more of a sensed cardiac EGM signal 120N image, a feature correlation coefficient image 130N, a feature RR interval image 132N, a feature maximum QRS amplitude image 151N, a feature minimum QRS amplitude image 152N, a feature difference image 153N between the maximum QRS amplitude and the minimum QRS amplitude, a feature sample number image 154N between the maximum QRS amplitude and the minimum QRS amplitude, a feature maximum slope image 155N of the QRS wave, a feature minimum slope image 156N of the QRS wave, a feature difference image 157N of the slope value between the maximum slope and the minimum slope of the QRS wave, or a feature sample number image 158N between the maximum slope and the minimum slope of the QRS wave.

[0117] The output of the deep learning model may include a score indicating whether a particular heartbeat is a PVC. For example, the score can be the probability that a particular heartbeat is a PVC. The processing circuit 98 can cause the score to be displayed to the clinician to assist in determining whether the patient should receive treatment based on having PVCs. In some examples, the output of the deep learning model may include a classification of the type of PVC beat, such as monomorphic, polymorphic, bigeminy, trigeminy, and interpolated PVCs. In some examples, the processing circuit 98 can determine a classification of the type of PVC beat, such as monomorphic, polymorphic, bigeminy, trigeminy, and interpolated PVCs, based on the determination of whether a particular heartbeat is a PVC and / or a score indicating whether a particular heartbeat is a PVC for a sequence of heartbeats. For example, if every third beat in a sequence of heartbeats is determined to be a PVC, the processing circuit 98 can determine that the patient 4 has trigeminy. For example, the sequence of heartbeats can include 5, 10, 15, 20, or 50 heartbeats. However, other quantities of heartbeats, such as greater than 50 heartbeats or less than 50 heartbeats, can also be used for the sequence of heartbeats. In some examples, the deep learning model can also be trained to determine the location of the origin of the PVC, such as whether the detected PVC beat originated from the septum, left ventricle, right ventricle, right ventricular outflow tract, etc. The processing circuit 98 can obtain data (such as EGM signal 120 images and feature images of features extracted from the EGM signal 120) and apply it to the trained deep learning model to determine whether a particular heartbeat is a PVC beat and / or to determine the location of the origin of the detected PVC beat, such as whether the detected PVC beat originated from the septum, left ventricle, right ventricle, right ventricular outflow tract, etc.

[0118] In some examples, when the processing circuit 98 determines that a particular heartbeat is a PVC beat, the score and / or correlation coefficient of the PVC beat can be stored and compared with the scores and correlation coefficients of previously detected PVC beats to determine whether they are monomorphic or polymorphic PVC beats. In some examples, when the processing circuit 98 determines that the scores and / or correlation coefficients of all PVC beats are similar, the processing circuit 98 can determine that the PVC beats can be monomorphic PVC beats exhibiting the same morphology. In some examples, when the processing circuit 98 determines that the scores and correlation coefficients have a difference greater than a difference threshold, the processing circuit 98 determines that the PVC beats can be polymorphic PVCs. In some examples, the processing circuit 98 can determine how many different morphologies of PVC beats are present in the patient based on the scores and / or correlation coefficients.

[0119] For example, a ventricular bigeminy pattern is when each normal beat is followed by a PVC beat. When the processing circuit 98 detects the following PVC pattern: normal beat - PVC beat - normal beat - PVC beat - normal beat - PVC beat, the processing circuit 98 determines that the PVC pattern is a bigeminy pattern and can provide a corresponding alert. A ventricular trigeminy pattern is when a PVC beat occurs every three beats. When the processing circuit 98 detects the following PVC pattern: normal beat - normal beat - PVC beat - normal beat - normal beat - PVC beat - normal beat - normal beat - PVC beat, the processing circuit 98 determines that the PVC pattern is a trigeminy pattern and can provide a corresponding alert. When the processing circuit 98 detects that a PVC beat occurs every four beats, the processing circuit 98 can determine the PVC pattern as a quadrigeminy pattern. For example, the quadrigeminy pattern will look like: normal beat - normal beat - normal beat - PVC beat - normal beat - normal beat - normal beat - PVC beat - normal beat - normal beat - normal beat - PVC beat.

[0120] The detected PVCs can be classified as interpolated PVCs based on the RR interval pattern. If the detected PVCs do not have a compensatory pause and the P-P interval for normal beats remains approximately constant, then these PVCs can be classified as interpolated PVCs by AI.

[0121] Figure 13 An example of the processing circuit 98 is shown, which will apply the normalized weighted images of the sensed cardiac EGM signal 120N image, feature correlation coefficient image 130N, feature RR interval image 132N, feature maximum QRS amplitude image 151N, feature minimum QRS amplitude image 152N, feature difference image 153N between the maximum QRS amplitude and the minimum QRS amplitude, feature sample number image 154N between the maximum QRS amplitude and the minimum QRS amplitude, feature maximum slope image 155N of the QRS wave, feature minimum slope image 156N of the QRS wave, feature difference image 157N of the slope values between the maximum slope and the minimum slope of the QRS wave, or feature sample number image 158N between the maximum slope and the minimum slope of the QRS wave as shown in Figure 12A and Figure 12B to a machine learning model. As shown in the example of Figure 13 , the machine learning model produces an accuracy of 98.93%.

[0122] Figure 14 An example of the confusion matrix on the validation dataset of Figure 13 is shown. As shown in the example of Figure 14 , the sensitivity of PVC detection is 97.3%, and the specificity is 99.2%.

[0123] Figure 15 is a conceptual diagram illustrating an example machine learning model 1500 that is configured to determine whether a particular heartbeat in a sensed electrocardiogram is a PVC beat. Machine learning model 1500 is an example of the machine learning models discussed above. Machine learning model 1500 is an example of a deep learning model or deep learning algorithm that is trained to determine whether a particular heartbeat in a sensed electrocardiogram is a PVC beat and / or a score indicating whether a particular heartbeat may be a PVC beat. One or more of the IMD 10, external device 12, server 94, and / or computing system 100 may train, store, and / or utilize machine learning model 1500, but in other examples, other devices may apply inputs associated with a particular patient to machine learning model 1500. As discussed above, other types of machine learning and deep learning models or algorithms may be utilized in other examples. For example, a convolutional neural network model of ResNet-18 may be used. Some non-limiting examples of models that may be used for transfer learning include AlexNet, VGGNet, GoogleNet, ResNet50, or DenseNet, among others. Some non-limiting examples of machine learning techniques include support vector machines, K-nearest neighbor algorithms, and multi-layer perceptrons.

[0124] As Figure 15As shown in the example of, the machine learning model 1500 can include three layers. These three layers include an input layer 1502, a hidden layer 1504, and an output layer 1506. The output layer 1506 includes the output of the transfer function 1505 from the output layer 1506. The input layer 1502 represents each of the input values X1 to X4 provided to the machine learning model 1500. In some examples, as described above, the input values can include any of the values input into the machine learning model. For example, the input values can be one or more of the weighted images 120, 130, 132, 151, 152, 153, 154, 155, 156, 157, 158 as described above. In some examples, the input values can be the sensed cardiac EGM signal 120N image, the feature correlation coefficient image 130N, the feature RR interval image 132N, the feature maximum QRS amplitude image 151N, the feature minimum QRS amplitude image 152N, the feature difference image 153N between the maximum QRS amplitude and the minimum QRS amplitude, the feature sample number image 154N between the maximum QRS amplitude and the minimum QRS amplitude, the feature maximum slope image 155N of the QRS wave, the feature minimum slope image 156N of the QRS wave, the feature difference image 157N of the slope values between the maximum slope and the minimum slope of the QRS wave, or the feature sample number image 158N between the maximum slope and the minimum slope of the QRS wave, which are normalized and weighted images. Additionally, in some examples, the input values of the machine learning model 1500 can include additional data, such as other data associated with one or more additional parameters of patient 4.

[0125] Each input value among the input values for each node in the input layer 1502 is provided to each node of the hidden layer 1504. In Figure 15 the example of, the hidden layer 1504 includes two layers, one layer having four nodes and the other layer having three nodes, but fewer or more nodes can be used in other examples. Each input from the input layer 1502 is multiplied by a weight and then summed at each node of the hidden layer 1504. During the training of the machine learning model 1500, the weights for each input are adjusted to establish relationships between electrocardiograms to determine whether a particular heartbeat in the sensed electrocardiogram is a PVC beat and / or to determine a score indicating whether a particular heartbeat could be a PVC. In some examples, one hidden layer can be incorporated into the machine learning model 1500, or three or more hidden layers can be incorporated into the machine learning model 1500, where each layer includes the same or different numbers of nodes.

[0126] Apply the result of each node within the hidden layer 1504 to the transfer function of the output layer 1506. The transfer function can be linear or non-linear, depending on the number of layers within the machine learning model 1500. Example non-linear transfer functions can be sigmoid functions or rectifier functions. The output 1507 of the transfer function can be a classification of whether a particular heartbeat is a PVC beat and / or an indication of a score of whether a particular heartbeat can be a PVC beat generated by a computing device or computing system (such as by processing circuitry 98) in response to applying one or more of the weighted images 120, 130, 132, 151, 152, 153, 154, 155, 156, 157, 158 to the machine learning model 1500.

[0127] As shown in the above example, by extracting one or more specific features from the sensed cardiac EGM signal 120, and converting the extracted features into images, and applying the EGM signal 120 images and the feature images to a machine learning model (such as the machine learning model 1500) to determine whether a particular heartbeat 124 is a PVC, the processing circuitry 98 is able to determine whether a particular heartbeat 124 is a PVC with high accuracy, specificity, and sensitivity. Additionally, in some examples, by weighting and / or normalizing the EGM signal 120 images and the feature images and applying the weighted and / or normalized EGM signal 120 images and feature images to the machine learning model to determine whether a particular heartbeat 124 is a PVC, the processing circuitry 98 is able to determine whether a particular heartbeat 124 is a PVC with high accuracy, specificity, and sensitivity. This can facilitate the determination of PVC burden, cardiac health, and the risk of sudden cardiac death and may lead to clinical intervention to suppress PVCs, such as medications and PVC ablation.

[0128] Figure 16is an example of using supervised and / or reinforcement learning techniques to train a machine learning model 1602. The machine learning model 1602 can be implemented using any number of models for supervised learning and / or reinforcement learning, such as but not limited to artificial neural networks, decision trees, naive Bayesian networks, support vector machines, or k-nearest neighbor models, to name just a few examples. In some examples, one or more of the IMD 10, external device 12, server 94, and / or computing device 100 initially trains the machine learning model 1602 based on a metric-based training set and corresponding to PVC beats. The training set 1600 can include a set of feature vectors, where each feature in the feature vector represents a value for a specific metric. One or more of the IMD 10, external device 12, server 94, and / or computing device 100 can select a training set that includes a set of training instances, each training instance including an association between one or more corresponding electrocardiogram features of a corresponding feature image and a corresponding PVC beat. The prediction or classification of the machine learning model 1602 can be compared 1604 with a target output 1603, and an error signal and / or machine learning model weight modification can be sent / applied to the machine learning model 1602 based on this comparison to modify / update the machine learning model 1602. For example, for each training instance in the training set, one or more of the IMD 10, external device 12, server 94, and / or computing device 100 can modify the machine learning model 1602 based on the corresponding electrocardiogram features and corresponding PVC beats of the training instance to change the score generated by the machine learning model 1602 in response to subsequent PVC beats applied to the machine learning model 1602.

[0129] Figure 17 is a flowchart illustrating example techniques for a medical system 2. As Figure 17 shown, the processing circuit 98 can receive a sensed electrocardiogram (1610) of a patient. The processing circuit 98 can extract a plurality of features (1620) from the sensed electrocardiogram. The processing circuit 98 can convert one or more of the extracted plurality of features into corresponding feature images (1630). The processing circuit 98 can apply a machine learning model, such as machine learning model 1500, to the image of the sensed electrocardiogram and the one or more feature images (1640). The machine learning model can be trained using electrocardiogram data for a plurality of patients. The processing circuit 98 can determine (1650) whether a particular heartbeat in the sensed electrocardiogram is a PVC beat based at least in part on the applied machine learning model. In response to the processing circuit 98 determining that the particular heartbeat is a PVC beat, the processing circuit 98 can output a classification that the particular heartbeat is a PVC beat (1660).

[0130] The techniques described in this disclosure may be implemented, at least in part, in the form of hardware, software, firmware, or any combination thereof. For example, aspects of these techniques may be implemented in one or more microprocessors, DSPs, ASICs, FPGAs, or any other equivalent integrated or discrete logic QRS circuitry, as well as any combination of such components, embodied in an external device such as a physician or patient programmer, simulator, or other device. The terms "processor" and "processing circuitry" generally may refer to any of the foregoing logic circuitry alone or in combination with other logic circuitry, or any other equivalent circuitry alone or in combination with other digital or analog circuitry.

[0131] For aspects implemented in software, at least some of the functionality attributable to the systems and devices described in this disclosure may be embodied as instructions on a computer-readable storage medium, such as RAM, DRAM, SRAM, disks, optical disks, flash memory, or various forms of EPROM or EEPROM. The instructions may be executed to support one or more aspects of the functionality described in this disclosure.

[0132] In addition, in some aspects, the functionality described herein may be provided within dedicated hardware and / or software modules. Describing different features as modules or units is intended to highlight different functional aspects and does not necessarily imply that such modules or units must be implemented by separate hardware or software components. Instead, the functionality associated with one or more modules or units may be performed by separate hardware or software components, or integrated within common or separate hardware or software components. Additionally, these techniques may be implemented entirely in one or more circuits or logic elements. The techniques of this disclosure may be implemented in a variety of devices or apparatuses, including an IMD, an external programmer, a combination of an IMD and an external programmer, an integrated circuit (IC) or a set of ICs, and / or discrete circuitry residing within an IMD and / or an external programmer.

[0133] Various aspects of the techniques may implement the following embodiments.

[0134] Embodiment 1: A medical system, the medical system comprising: a plurality of electrodes configured to sense an electrocardiogram of a patient; and processing circuitry configured to: perform feature extraction of the sensed electrocardiogram to extract a plurality of features from the sensed electrocardiogram; convert one or more of the extracted plurality of features into corresponding feature images; apply a machine learning model trained using electrocardiogram data of a plurality of patients to an image of the sensed electrocardiogram and at least the corresponding feature images to determine whether one or more specific heartbeats in the sensed electrocardiogram indicate premature ventricular contractions (PVCs); and output a classification that the one or more specific heartbeats are PVCs in response to the determination that the one or more specific heartbeats indicate PVCs.

[0135] Example 2: The medical system according to Example 1, wherein the sensed electrocardiogram includes signals of consecutive heartbeats, and the signals of the consecutive heartbeats include signals of the one or more specific heartbeats, signals of the heartbeats immediately preceding the one or more specific heartbeats, and signals of the heartbeats immediately following the one or more specific heartbeats.

[0136] Example 3: The medical system according to Example 2, wherein the plurality of extracted features includes a plurality of correlation coefficients and a plurality of R-R intervals.

[0137] Example 4: The medical system according to Example 3, wherein the plurality of correlation coefficients includes three correlation coefficients, and the plurality of R-R intervals includes four R-R intervals.

[0138] Example 5: The medical system according to Example 4, wherein the plurality of extracted features further includes one or more of the following: maximum QRS amplitude, minimum QRS amplitude, amplitude difference between the maximum QRS amplitude and the minimum QRS amplitude, number of samples between the maximum QRS amplitude and the minimum QRS amplitude, maximum slope of the QRS wave, minimum slope of the QRS wave, slope value difference between the maximum slope and the minimum slope of the QRS wave, or number of samples between the maximum slope and the minimum slope of the QRS wave.

[0139] Example 6: The medical system according to Example 5, wherein the processing circuit is further configured to: adjust the size of the image of the sensed electrocardiogram corresponding to the respective electrocardiogram weights; adjust the size of the feature image corresponding to the respective image weights; and apply the machine learning model to the size-adjusted image of the sensed electrocardiogram and the size-adjusted feature image to determine whether one or more specific heartbeats in the sensed electrocardiogram are PVC beats.

[0140] Example 7: The medical system according to Example 6, wherein the processing circuit is further configured to: normalize the axis of the image of the sensed electrocardiogram; normalize the feature image; and apply the machine learning model to the normalized image of the sensed electrocardiogram and the normalized feature image to determine whether one or more specific heartbeats in the sensed electrocardiogram are PVC beats.

[0141] Example 8: The medical system according to any one of Examples 6 and 7, wherein the image of the sensed electrocardiogram is weighted more than the feature image.

[0142] Example 9: The medical system according to Example 8, wherein the feature image corresponding to the plurality of extracted correlation coefficients is weighted more than one or more of the following feature images: the maximum QRS amplitude, the minimum QRS amplitude, the amplitude difference between the maximum QRS amplitude and the minimum QRS amplitude, the number of samples between the maximum QRS amplitude and the minimum QRS amplitude, the maximum slope of the QRS wave, the minimum slope of the QRS wave, the slope value difference between the maximum slope of the QRS wave and the minimum slope of the QRS wave, and the number of samples between the maximum slope of the QRS wave and the minimum slope of the QRS wave, and the feature image corresponding to the plurality of extracted R-R intervals is weighted more than each of the following feature images: the maximum QRS amplitude, the minimum QRS amplitude, the amplitude difference between the maximum QRS amplitude and the minimum QRS amplitude, the number of samples between the maximum QRS amplitude and the minimum QRS amplitude, the maximum slope of the QRS wave, the minimum slope of the QRS wave, the slope value difference between the maximum slope of the QRS wave and the minimum slope of the QRS wave, and the number of samples between the maximum slope of the QRS wave and the minimum slope of the QRS wave.

[0143] Example 10: The medical system according to any one of Examples 1 to 9, wherein the processing circuit is further configured to apply the machine learning model to the image of the sensed electrocardiogram and the one or more feature images to determine the location of the origin of the PVC beats.

[0144] Example 11: The medical system according to any one of Examples 1 to 10, wherein the medical system is an implantable cardiac monitor, the implantable cardiac monitor comprising: a power source operatively coupled to the processing circuit; a memory operatively coupled to the processing circuit and configured to store the machine learning model; a distal electrode operatively coupled to the processing circuit; a proximal electrode operatively coupled to the processing circuit; and a hermetically sealed housing configured for subcutaneous implantation in a patient, wherein at least the power source, the memory, and the processing circuit are within the hermetically sealed housing, and wherein the housing has a length, a width, and a depth, wherein the length is greater than the width and the width is greater than the depth, wherein the length is in the range of 5 millimeters (mm) to 60 mm, wherein the width is in the range of 5 mm to 15 mm, and wherein the depth is in the range of 5 mm to 15 mm.

[0145] Example 12: The medical system according to any one of Examples 1 to 11, wherein before the machine learning model is applied to the corresponding feature image, the machine learning model is trained by: selecting a training set including a set of training instances, each training instance including an association between one or more corresponding electrocardiogram features of the corresponding feature image and a corresponding PVC beat; and for each training instance in the training set, modifying the machine learning model based on the corresponding electrocardiogram features and the corresponding PVC beat of the training instance to change a score generated by the machine learning model in response to a subsequent PVC beat applied to the machine learning model.

[0146] Example 13: A medical device, the medical device comprising: a memory; and a processing circuit coupled to the memory, the processing circuit being configured to: receive a sensed electrocardiogram of a patient; perform feature extraction of the sensed electrocardiogram to extract a plurality of features from the sensed electrocardiogram; convert one or more of the extracted plurality of features into a corresponding feature image; apply a machine learning model trained using electrocardiogram data of a plurality of patients to the image of the sensed electrocardiogram and at least the corresponding feature image to determine whether one or more specific heartbeats in the sensed electrocardiogram indicate premature ventricular contractions (PVC) beats; and output a classification that the one or more specific heartbeats are PVC beats in response to the determination that the one or more specific heartbeats indicate PVC beats.

[0147] Example 14: The computing device according to Example 13, wherein the sensed electrocardiogram includes signals of consecutive heartbeats, the signals of the consecutive heartbeats including signals of the one or more specific heartbeats, signals of heartbeats immediately preceding the one or more specific heartbeats, and signals of heartbeats immediately following the one or more specific heartbeats.

[0148] Example 15: The computing device according to Example 14, wherein the extracted plurality of features includes a plurality of correlation coefficients and a plurality of R-R intervals.

[0149] Example 16: The computing device according to Example 15, wherein the plurality of correlation coefficients includes three correlation coefficients, and the plurality of R-R intervals includes four R-R intervals.

[0150] Example 17: The computing device according to Example 16, wherein the plurality of extracted features further includes one or more of the following: the maximum QRS amplitude, the minimum QRS amplitude, the amplitude difference between the maximum QRS amplitude and the minimum QRS amplitude, the number of samples between the maximum QRS amplitude and the minimum QRS amplitude, the maximum slope of the QRS wave, the minimum slope of the QRS wave, the slope value difference between the maximum slope and the minimum slope of the QRS wave, or the number of samples between the maximum slope and the minimum slope of the QRS wave.

[0151] Example 18: The computing device according to Example 17, wherein the processing circuit is further configured to: adjust the size of the image of the sensed electrocardiogram corresponding to the respective electrocardiogram weights; adjust the size of the feature image corresponding to the respective image weights; apply the machine learning model to the size-adjusted image of the sensed electrocardiogram and the size-adjusted feature image to determine whether one or more specific heartbeats in the sensed electrocardiogram are PVC beats.

[0152] Example 19: The computing device according to Example 18, wherein the processing circuit is further configured to: normalize the axis of the image of the sensed electrocardiogram; normalize the feature image; and apply the machine learning model to the normalized image of the sensed electrocardiogram and the normalized feature image to determine whether one or more specific heartbeats in the sensed electrocardiogram are PVC beats.

[0153] Example 20: The computing device according to any one of Examples 18 and 19, wherein the image of the sensed electrocardiogram is weighted more than the feature image.

[0154] Example 21: The computing device according to Example 20, wherein the feature image corresponding to the plurality of extracted correlation coefficients is weighted more than one or more of the following feature images: the maximum QRS amplitude, the minimum QRS amplitude, the amplitude difference between the maximum QRS amplitude and the minimum QRS amplitude, the number of samples between the maximum QRS amplitude and the minimum QRS amplitude, the maximum slope of the QRS wave, the minimum slope of the QRS wave, the slope value difference between the maximum slope and the minimum slope of the QRS wave, and the number of samples between the maximum slope and the minimum slope of the QRS wave, and the number of samples between the maximum slope and the minimum slope of the QRS wave.

[0155] Example 22: The computing device according to any one of Examples 13 to 21, wherein the processing circuit is further configured to apply the machine learning model to the image of the sensed electrocardiogram and the one or more feature images to determine the location of the origin of the PVC beat.

[0156] Example 23: The computing device according to any one of Examples 13 to 22, wherein the machine learning model is trained by: selecting a training set including a set of training instances, each training instance including an association between one or more corresponding electrocardiogram features of a corresponding feature image and a corresponding PVC beat; and for each training instance in the training set, modifying the machine learning model based on the corresponding electrocardiogram features and the corresponding PVC beat of the training instance to change the score generated by the machine learning model in response to a subsequent PVC beat applied to the machine learning model.

[0157] Example 24: A method, the method comprising: receiving a sensed electrocardiogram of a patient; performing feature extraction of the sensed electrocardiogram to extract a plurality of features from the sensed electrocardiogram; converting one or more of the extracted plurality of features into corresponding feature images; applying a machine learning model trained using electrocardiogram data of a plurality of patients to an image of the sensed electrocardiogram and at least the corresponding feature images to determine whether one or more specific heartbeats in the sensed electrocardiogram indicate premature ventricular contractions (PVC) beats; and in response to the determination that the one or more specific heartbeats indicate PVC beats, outputting a classification that the one or more specific heartbeats are PVC beats.

[0158] Example 25: The method according to Example 24, wherein the sensed electrocardiogram includes signals of consecutive heartbeats, and the signals of the consecutive heartbeats include signals of the one or more specific heartbeats, signals of heartbeats immediately preceding the one or more specific heartbeats, and signals of heartbeats immediately following the one or more specific heartbeats.

[0159] Example 26: The method according to Example 25, wherein the extracted plurality of features include a plurality of correlation coefficients and a plurality of R-R intervals.

[0160] Example 27: The method according to Example 26, wherein the plurality of correlation coefficients include three correlation coefficients, and the plurality of R-R intervals include four R-R intervals.

[0161] Example 28: The method according to Example 27, wherein the plurality of extracted features further includes one or more of the following: maximum QRS amplitude, minimum QRS amplitude, amplitude difference between the maximum QRS amplitude and the minimum QRS amplitude, number of samples between the maximum QRS amplitude and the minimum QRS amplitude, maximum slope of the QRS wave, minimum slope of the QRS wave, slope value difference between the maximum slope and the minimum slope of the QRS wave, or number of samples between the maximum slope and the minimum slope of the QRS wave.

[0162] Example 29: The method according to Example 28, wherein the method further includes: adjusting the size of the image of the sensed electrocardiogram corresponding to the respective electrocardiogram weights; adjusting the size of the feature image corresponding to the respective image weights; and applying the machine learning model to the size-adjusted image of the sensed electrocardiogram and the size-adjusted feature image to determine whether one or more specific heartbeats in the sensed electrocardiogram are PVC beats.

[0163] Example 30: The method according to Example 29, wherein the method further includes: normalizing the axis of the image of the sensed electrocardiogram; normalizing the feature image; and applying the machine learning model to the normalized image of the sensed electrocardiogram and the normalized feature image to determine whether one or more specific heartbeats in the sensed electrocardiogram are PVC beats.

[0164] Example 31: The method according to any one of Examples 29 and 30, wherein the image of the sensed electrocardiogram is weighted more than the feature image.

[0165] Example 32: The method according to Example 31, wherein the feature image corresponding to the plurality of extracted correlation coefficients is weighted more than one or more of the following feature images: the maximum QRS amplitude, the minimum QRS amplitude, the amplitude difference between the maximum QRS amplitude and the minimum QRS amplitude, the number of samples between the maximum QRS amplitude and the minimum QRS amplitude, the maximum slope of the QRS wave, the minimum slope of the QRS wave, the slope value difference between the maximum slope and the minimum slope of the QRS wave, and the number of samples between the maximum slope and the minimum slope of the QRS wave, and the number of samples between the maximum slope and the minimum slope of the QRS wave.

[0166] Example 33: The method according to any one of Examples 24 to 32, wherein the method further comprises: applying the machine learning model to the image of the sensed electrocardiogram and the one or more feature images to determine the location of the origin of the PVC beats.

[0167] Various embodiments have been described. These and other examples are within the scope of the appended claims.

Claims

1. A medical system, the medical system comprising: a plurality of electrodes configured to sense an electrocardiogram of a patient; and a processing circuit configured to: perform feature extraction of the sensed electrocardiogram to extract a plurality of features from the sensed electrocardiogram; convert one or more of the extracted plurality of features into corresponding feature images; apply a machine learning model trained using electrocardiogram data of a plurality of patients to the image of the sensed electrocardiogram and at least the corresponding feature images to determine whether one or more specific heartbeats in the sensed electrocardiogram indicate premature ventricular contractions (PVC) beats; and in response to the determination that the one or more specific heartbeats indicate PVC beats, output a classification that the one or more specific heartbeats are PVC beats.

2. The medical system according to claim 1, wherein the sensed electrocardiogram includes signals of consecutive heartbeats, and the signals of the consecutive heartbeats include signals of the one or more specific heartbeats, signals of the heartbeats immediately preceding the one or more specific heartbeats, and signals of the heartbeats immediately following the one or more specific heartbeats.

3. The medical system according to any one of claims 1 to 2, wherein the extracted plurality of features includes a plurality of correlation coefficients and a plurality of R-R intervals.

4. The medical system according to claim 3, wherein the plurality of correlation coefficients includes three correlation coefficients, and the plurality of R-R intervals includes four R-R intervals.

5. The medical system according to any one of claims 1 to 4, wherein the extracted plurality of features further includes one or more of the following: maximum QRS amplitude, minimum QRS amplitude, amplitude difference between the maximum QRS amplitude and the minimum QRS amplitude, number of samples between the maximum QRS amplitude and the minimum QRS amplitude, maximum slope of the QRS wave, minimum slope of the QRS wave, slope value difference between the maximum slope and the minimum slope of the QRS wave, or number of samples between the maximum slope and the minimum slope of the QRS wave.

6. The medical system according to any one of claims 1 to 5, wherein the processing circuit is further configured to: adjust the size of the image of the sensed electrocardiogram corresponding to the respective electrocardiogram weights; adjust the size of the feature image corresponding to the respective image weights; and apply the machine learning model to the size-adjusted image of the sensed electrocardiogram and the size-adjusted feature image to determine whether one or more specific heartbeats in the sensed electrocardiogram are PVC beats.

7. The medical system according to any one of claims 1 to 6, wherein the processing circuit is further configured to: normalize the axis of the image of the sensed electrocardiogram; normalize the feature image; and apply the machine learning model to the normalized image of the sensed electrocardiogram and the normalized feature image to determine whether one or more specific heartbeats in the sensed electrocardiogram are PVC beats.

8. The medical system according to any one of claims 3 to 7, wherein the feature image corresponding to the plurality of extracted correlation coefficients is weighted more than one or more of the feature images for: the maximum QRS amplitude, the minimum QRS amplitude, the amplitude difference between the maximum QRS amplitude and the minimum QRS amplitude, the number of samples between the maximum QRS amplitude and the minimum QRS amplitude, the maximum slope of the QRS wave, the minimum slope of the QRS wave, the slope value difference between the maximum slope of the QRS wave and the minimum slope of the QRS wave, and the number of samples between the maximum slope of the QRS wave and the minimum slope of the QRS wave, and the feature image corresponding to the plurality of extracted R-R intervals is weighted more than each of the feature images for: the maximum QRS amplitude, the minimum QRS amplitude, the amplitude difference between the maximum QRS amplitude and the minimum QRS amplitude, the number of samples between the maximum QRS amplitude and the minimum QRS amplitude, the maximum slope of the QRS wave, the minimum slope of the QRS wave, the slope value difference between the maximum slope of the QRS wave and the minimum slope of the QRS wave, and the number of samples between the maximum slope of the QRS wave and the minimum slope of the QRS wave.

9. The medical system according to any one of claims 1 to 8, wherein the processing circuit is further configured to apply the machine learning model to the image of the sensed electrocardiogram and the one or more feature images to determine the location of the origin of the PVC beats.

10. The medical system according to any one of claims 1 to 9, wherein the medical system is an implantable cardiac monitor, the implantable cardiac monitor comprises: a power supply operatively coupled to the processing circuit; a memory operatively coupled to the processing circuit and configured to store the machine learning model; a distal electrode operatively coupled to the processing circuit; a proximal electrode operatively coupled to the processing circuit; and an airtight housing configured for subcutaneous implantation in the patient, wherein at least the power supply, the memory, and the processing circuit are within the airtight housing, and wherein the housing has a length, a width, and a depth, wherein the length is greater than the width, and the width is greater than the depth, wherein the length is in the range of 5 millimeters (mm) to 60 mm, wherein the width is in the range of 5 mm to 15 mm, and wherein the depth is in the range of 5 mm to 15 mm.

11. The medical system according to any one of claims 1 to 10, wherein the machine learning model is trained by the following method before being applied to the corresponding feature image: selecting a training set including a set of training instances, each training instance including the association between one or more corresponding electrocardiogram features of the corresponding feature image and the corresponding PVC beat; and for each training instance in the training set, modifying the machine learning model based on the corresponding electrocardiogram features and the corresponding PVC beat of the training instance, so as to change the score generated by the machine learning model in response to subsequent PVC beats applied to the machine learning model.

12. A computing device, the computing device comprising: a memory; and a processing circuit coupled to the memory, the processing circuit being configured to: receive a sensed electrocardiogram of a patient; perform feature extraction on the sensed electrocardiogram to extract a plurality of features from the sensed electrocardiogram; convert one or more of the extracted plurality of features into corresponding feature images; apply a machine learning model trained using electrocardiogram data of multiple patients to the image of the sensed electrocardiogram and at least the corresponding feature image to determine whether one or more specific heartbeats in the sensed electrocardiogram indicate premature ventricular contractions (PVC) beats; and in response to the determination that the one or more specific heartbeats indicate PVC beats, output a classification that the one or more specific heartbeats are PVC beats.

13. The computing device according to claim 12, wherein the extracted plurality of features include a plurality of correlation coefficients and a plurality of R-R intervals.

14. The computing device according to any one of claims 12 to 13, wherein the processing circuit is further configured to: adjust the size of the image of the sensed electrocardiogram corresponding to the corresponding electrocardiogram weights; adjust the size of the feature image corresponding to the corresponding image weights; apply the machine learning model to the size-adjusted image and size-adjusted feature image of the sensed electrocardiogram to determine whether one or more specific heartbeats in the sensed electrocardiogram are PVC beats.

15. The computing device according to any one of claims 12 to 14, wherein the processing circuit is further configured to: normalize the axis of the image of the sensed electrocardiogram; normalize the feature image; and apply the machine learning model to the normalized image and normalized feature image of the sensed electrocardiogram to determine whether one or more specific heartbeats in the sensed electrocardiogram are PVC beats.

Images