AI-based detection of physiological events

Abstract

Systems and methods for detecting physiological events or estimating physiological parameters using a movable electrogram of a subject are discussed. An exemplary system includes a computing device that may receive a movable electrogram collected by a movable medical device (AMD) associated with a subject and apply the movable electrogram to a trained machine learning model to estimate a physiological parameter in the subject or detect a physiological event thereof. The same or different machine learning models may be trained using the movable electrograms to detect the operational state of the AMD. The system includes an output device for outputting the estimated physiological parameter, the detected physiological event, or the detected operational status of the device to a user or process, such as initiating or titrating therapy.

Description

[0001] Related Applications

[0002] This application claims the benefit of U.S. patent application serial number 63 / 334,239, filed on April 25, 2022, which is incorporated herein by reference in its entirety. Technical Field

[0003] This document relates generally to medical devices, and more particularly to systems, devices, and methods for artificial intelligence (AI)-based detection of physiological events using active electrograms. Background Art

[0004] Ambulatory medical devices (AMDs), such as implantable medical devices (IMDs), have been used to monitor a patient's health or disease state and provide treatment. For example, an implantable cardioverter-defibrillator (ICD) can be used to monitor certain abnormal heart rhythms and deliver electrical energy to the heart to correct the abnormal heart rhythm. Some IMDs can be used to monitor chronic deterioration of cardiac hemodynamic performance (such as due to congestive heart failure (CHF)) and provide cardiac stimulation therapy, including cardiac resynchronization therapy (CRT), to correct cardiac asynchrony within or between ventricles.

[0005] Some AMDs have sensors that can sense physiological information from the patient, such as an electrocardiogram (ECG) or electrogram (EGM) that represents the electrical activity of the heart. For example, a Holter monitor is a wearable device with skin electrodes to monitor a movable ECG. ECGs have been used to diagnose and monitor a variety of heart conditions or diseases, such as arrhythmias, coronary heart disease, heart attacks, cardiomyopathy, left ventricular systolic dysfunction, etc. ECGs have also been used to investigate symptoms such as chest pain, palpitations, dizziness, and shortness of breath, as well as other heart-related problems. Summary of the invention

[0006] Artificial intelligence (AI) can be used to interpret ECGs and assist in detecting heart conditions such as arrhythmias. Machine learning (ML) models that mimic the function of brain neurons using a series of interconnected computational statistical algorithms or nodes can be trained on large data sets of ECGs to identify specific heart conditions or diagnoses. AI-based ECG analysis can also provide valuable insights into some non-cardiac conditions and seemingly unrelated or indirect biological parameters such as cardiac ejection fraction or blood potassium.

[0007] Electrocardiograms are typically measured in clinical settings using a 12-lead system. This may be inconvenient for some patients, especially those who require long-term continuous or frequent ECG monitoring. Mobile monitoring via devices such as implantable cardiac devices, insertable cardiac monitors (ICMs), subcutaneous ICDs (S-ICDs), wearable or portable devices, and other mobile monitors can also allow ECG or EGM (e.g., endocardial or epicardial EGM) data collection, analysis, and detection of cardiac events. Compared to clinical 12-lead ECGs, wearable or implantable devices can acquire mobile ECGs or EGMs at a higher frequency or even continuously without changing the electrode configuration commonly used in the clinic. Mobile ECGs or EGMs can be analyzed to produce repetitive measurements of physiological parameters. Measurements can be trended over time and used to provide more accurate early detection or prediction of cardiac or other physiological events.

[0008] The present disclosure describes an AI-based system, device and method for estimating the physiological parameters of a subject or detecting its physiological events using a movable EGM. According to one embodiment, the system includes a computing device configured to receive a movable EGM collected by an AMD worn or implanted in a subject, and the received movable EGM is applied to a trained machine learning (ML) model to estimate the physiological parameters in the subject or detect the physiological events in the subject. The trained ML model is trained using a training data set, which includes an EGM collected from a patient population and an assessment of the physiological parameters or physiological events in the patient population. In some examples, the same or different ML models can be trained to detect the operating state of AMD from the movable EGM of the subject. The estimated physiological parameters, detected physiological events or detected AMD operating states can be provided to a user, or provided to a process in order to initiate or titrate treatment. In some examples, when conditions are met, the estimated physiological parameters can be used to trigger a sensor to directly measure the physiological parameters. Direct measurement can be used to adjust the trained ML model to improve its parameter estimation performance.

[0009] Example 1 is a system for detecting physiological events in a subject, the system comprising: a computing device configured to: receive an ambulatory electrogram of the subject collected by an ambulatory medical device (AMD); and apply the received ambulatory electrogram to a trained machine learning model to estimate a physiological parameter in the subject or detect a physiological event thereof; and an output unit configured to output the estimated physiological parameter or detected physiological event to a user or process.

[0010] In Example 2, the subject matter of Example 1 optionally includes a computing device, which may also be configured to: determine or confirm a physical characteristic of at least one sensor using a movable electrogram; and detect a physiological event in a subject or estimate a physiological parameter thereof using the movable electrogram and the determined or confirmed physical characteristic of at least one sensor, wherein in order to detect the physiological event, the computing device is configured to: apply the received movable electrogram to a trained machine learning model to estimate a physiological parameter in the subject or detect a physiological event thereof using the movable electrogram sensed by the at least one sensor, and wherein the physical characteristic is a sensor type or form factor of the at least one sensor.

[0011] In Example 3, the subject matter of Example 2 optionally includes a computing device, which may include a training module configured to generate a trained machine learning model, the training module including: constructing a training data set, the data set including movable electrograms collected from a patient population and assessments of physiological parameters or physiological events in the patient population; and training the machine learning model using the constructed training data set until convergence or training stopping criteria are met, the trained machine learning model representing the correspondence between the movable electrograms of the patient population and the physiological parameters or physiological events in the patient population.

[0012] In Example 4, the subject matter of Example 3 optionally includes at least one sensor that can be configured to sense physiological information other than a movable electrogram from the subject, and wherein the training module is configured to train the machine learning model using a deep learning algorithm including a deep neural network.

[0013] In Example 5, the subject matter of any one or more of Examples 2-4 optionally includes a computing device that can be configured to apply the received electrogram to a trained machine learning model to estimate a physiological parameter including at least one of: a cardiac parameter; a respiratory parameter; a circulating biomarker; or a systemic or local fluid state.

[0014] In Example 6, the subject matter of any one or more of Examples 2-5 optionally includes a computing device that can be configured to apply the received electrogram to a trained machine learning model to detect physiological events including at least one of: an arrhythmia; a heart failure exacerbation event; a comorbidity of heart failure; a neurological disease; or a response to a medication.

[0015] In Example 7, the subject matter of any one or more of Examples 2-6 optionally includes a computing device that can be further configured to apply the received electrogram to the trained machine learning model to detect an operational state of AMD.

[0016] In Example 8, the subject matter of Example 7 optionally includes an operational state of the AMD, the operational state indicating at least one of: a change in position, posture, or orientation of the AMD; or a change in a device-tissue interface of the AMD.

[0017] In Example 9, the subject matter of any one or more of Examples 1-8 may optionally include an AMD which may be configured to continuously or periodically collect ambulatory electrograms of the subject via one or more attachable or implantable electrodes.

[0018] In Example 10, the subject matter of any one or more of Examples 1-9 optionally includes an AMD which may include at least one of: an insertable cardiac monitor; a subcutaneously implantable cardioverter-defibrillator; or a wearable or portable cardiac monitor.

[0019] In Example 11, the subject matter of any one or more of Examples 1-10 may optionally include an AMD, which may include a therapy circuit configured to initiate or adjust therapy of the subject based on the estimated physiological parameter or detected physiological event.

[0020] In Example 12, the subject matter of any one or more of Examples 2-11 optionally includes a computing device that can be configured to: apply the received electrogram to a trained machine learning model to estimate a physiological parameter; and in response to the estimated physiological parameter satisfying a condition, trigger at least one of the at least one sensors to directly measure the physiological parameter.

[0021] In Example 13, the subject matter of Example 12 optionally includes a computing device, which may include a calibration circuit configured to adjust the trained machine learning model based at least on the directly measured physiological parameter.

[0022] In Example 14, the subject matter of any one or more of Examples 1-13 may optionally include a computing device that may be configured to perform parallel computations to estimate multiple physiological parameters or detect multiple physiological events substantially simultaneously.

[0023] In Example 15, the subject matter of Example 14 optionally includes a computing device, which may include: multiple processors, or a multi-core processor including multiple computing units, each of the multiple processors or the multiple computing units is configured to: apply a portion of the subject's movable electrogram to a corresponding trained machine learning model to estimate a corresponding physiological parameter in the subject or detect a corresponding physiological event in the subject.

[0024] Example 16 is a method for detecting physiological events in a subject, the method comprising: receiving an ambulatory electrogram of the subject collected by an ambulatory medical device (AMD); determining, via a computing device, a form factor of at least one sensor using the ambulatory electrogram, the at least one sensor being configured to sense physiological information from the subject that is different from the ambulatory electrogram; using the ambulatory electrogram and the physiological information sensed by the at least one sensor to continuously detect physiological events in the subject or estimate physiological parameters thereof; and providing the estimated physiological parameters or detected physiological events to a user or process.

[0025] In Example 17, the subject matter of Example 16 optionally includes continuously detecting physiological events, which may include applying the received electrogram to a trained machine learning model to estimate a physiological parameter in the subject or detect a physiological event thereof.

[0026] In Example 18, the subject matter of Example 17 optionally includes: constructing a training dataset comprising movable electrograms collected from a patient population and assessments of physiological parameters or physiological events in the patient population; and using the constructed training dataset to train a machine learning model until convergence or training stopping criteria are met, the trained machine learning model representing the correspondence between the movable electrograms of the patient population and the physiological parameters or physiological events in the patient population.

[0027] In Example 19, the subject matter of any one or more of Examples 17-18 optionally includes applying the received electrogram to the trained machine learning model and detecting an operational state of the AMD.

[0028] In Example 20, the subject matter of any one or more of Examples 16-19 optionally includes initiating or adjusting treatment of the subject via AMD based on the estimated physiological parameter or detected physiological event.

[0029] In Example 21, the subject matter of any one or more of Examples 16-20, in response to an estimated physiological parameter satisfying a condition, optionally includes: triggering direct measurement of the physiological parameter using at least one of the at least one sensors; and adjusting the trained machine learning model based at least on the direct measurement of the physiological parameter.

[0030] This summary is an overview of some of the teachings of the present application, and it is not intended to be an exclusive or exhaustive treatment of the subject matter. Further details about this subject matter can be found in the detailed description and the appended claims. Other aspects of the present disclosure will be apparent to those skilled in the art upon reading and understanding the following detailed description and viewing the accompanying drawings, each of which should not be construed as limiting. The scope of the present disclosure is defined by the appended claims and their legal equivalents. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Various embodiments are shown by way of example in the figures of the accompanying drawings. Such embodiments are exemplary and are not intended to be exhaustive or exclusive embodiments of the subject matter.

[0032] Figure 1 An example of a patient management system and portions of an environment in which the system may operate are generally shown.

[0033] Figure 2 An example system is shown that includes an movable medical device configured to sense or detect physiological information from a patient via one or more sensors.

[0034] Figure 3 is a block diagram illustrating a portion of an exemplary AI-based physiological event detection system configured to detect physiological events or estimate physiological parameters using an active EGM.

[0035] Figure 4A-4B is a diagram showing an example of training a machine learning (ML) model and using the trained ML model to estimate a physiological parameter, detect a physiological event, or detect an operating status of a device.

[0036] Figure 5 is a flow chart illustrating an example of a method for detecting a physiological event or estimating a physiological parameter using an active EGM.

[0037] Figure 6 A block diagram of an example machine is generally shown upon which any one or more of the techniques (eg, methodologies) discussed herein may be performed. DETAILED DESCRIPTION

[0038] Disclosed herein are AI-based systems, devices, and methods for detecting physiological events or estimating physiological parameters using a movable EGM. An exemplary system includes a computing device configured to receive a movable EGM collected by a movable medical device (AMD) associated with a subject, and apply the movable EGM to a trained machine learning (ML) model to estimate physiological parameters in the subject or detect its physiological events. The same or different machine learning models can be trained using movable electrograms to detect the operating state of AMD. The system includes an output device to output estimated physiological parameters, detected physiological events, or detected device operating states to a user or process in order to initiate or titrate treatment.

[0039] The systems, devices, and methods discussed in this document can improve the prediction or early detection of various physiological events, such as cardiovascular or respiratory diseases, systemic or local fluid status, circulating biomarkers, or neurological diseases. Although alternative solutions, such as direct measurements using dedicated sensors, can provide the desired detection performance, the operation of dedicated sensors is generally associated with higher system complexity and implementation costs, and some sensors may not be sensitive enough to provide early indications of events. The present disclosure describes early physiological event detection using a movable ECG or EGM, which can be quickly acquired using lower cost, simpler or less obtrusive systems, devices, and methods. The availability of large amounts of ECG or EGM data makes it suitable for AI-based applications including training and validating ML models for estimating physiological parameters and detecting physiological events. Early detection of physiological events can be used to pre-screen patients for further evaluation using, for example, more complex and expensive tests, or for medical intervention or treatment optimization (e.g., drug titration or timing, etc.).

[0040] In some examples, the systems, devices, and methods discussed herein can provide additional uses for existing sensors, reduce sensor costs, enable early intervention, improve patient prognosis, and reduce overall medical system costs. In some examples, the systems and methods described herein represent improved forms of physiological event detection and patient intervention that are superior to the prior art. In some examples, patients can be monitored and alerts of changes in patient conditions can be issued to patients, caregivers, clinicians, or one or more other systems or users. In other examples, the systems and methods described herein can provide intervention or treatment optimization recommendations, or can directly provide or change treatment to patients.

[0041] Figure 1 An example patient management system 100 and a portion of an environment in which the patient management system 100 may operate are shown. The patient management system 100 may perform a range of activities including remote patient monitoring and diagnosis of disease conditions. Such activities may be performed in the vicinity of the patient 101, such as at the patient's home or office, through a centralized server, such as at a hospital, clinic, or doctor's office, or through a remote workstation, such as a secure wireless mobile computing device.

[0042] The patient management system 100 may include one or more removable medical devices, an external system 105, and a communication link 111 that provides communication between the one or more removable medical devices and the external system 105. The one or more removable medical devices may include an implantable medical device (IMD) 102, a wearable medical device 103, or one or more other implantable, leadless, subcutaneous, external, wearable, or removable medical devices configured to monitor, sense, or detect information from the patient 101, determine physiological information about the patient 101, or provide one or more therapies to treat various conditions of the patient 101, such as one or more cardiac or non-cardiac conditions (e.g., dehydration, sleep apnea, etc.).

[0043] In an example, the implantable medical device 102 may include one or more conventional cardiac rhythm management devices implanted in the patient's chest, such as a pacemaker, a pacemaker / defibrillator, a cardiac resynchronization therapy (CRT) device, a cardiac remodeling control therapy (RCT) device, etc. The implantable medical device 102 may have a lead system including one or more transvenous, subcutaneous or non-invasive leads or catheters to position one or more electrodes or other sensors (e.g., heart sound sensors) in, on, or around the heart of the patient 101 or one or more other locations in the chest, abdomen, or neck. By way of example and not limitation, the lead system may be inserted or positioned transvenously on the surface of a portion of the heart (such as the right atrium (RA), right ventricle (RV), left atrium (LA), or left ventricle (LV), or any tissue between or near the heart parts). In another example, the implantable medical device 102 may include, for example, a monitor implanted subcutaneously in the chest of the patient 101. Examples of such subcutaneous devices may include an insertable cardiac monitor (ICM), a subcutaneous ICD (S-ICD), etc. Implantable medical device 102 includes a housing containing circuitry and, in some examples, one or more sensors, such as a temperature sensor, an accelerometer for sensing heart sounds, and the like.

[0044] The implantable medical device 102 may include a sensing circuit configured to sense cardiac electrical activity using electrodes on a lead system. The sensed cardiac electrical activity is often referred to as an electrogram (EGM). In an example, the sensing circuit may sense the EGM continuously or periodically at a particular frequency. The implantable medical device 102 may also include an evaluation circuit configured to detect or determine specific physiological information of the patient 101, or to determine one or more physiological conditions, or to provide information or an alert to a user, such as the patient 101 (e.g., a patient), a clinician, or one or more other caregivers or processes.

[0045] The implantable medical device 102 may alternatively or additionally be configured as a therapeutic device configured to treat one or more medical conditions of the patient 101. The therapy may be delivered to the patient 101 via a lead system and associated electrodes or using one or more other delivery mechanisms. The therapy may include delivering one or more drugs to the patient 101, such as using the implantable medical device 102 or one or more other movable medical devices. In some examples, the therapy may include cardiac resynchronization therapy for correcting asynchrony and improving cardiac function in patients with heart failure. In other examples, the implantable medical device 102 may include a drug delivery system, such as a drug infusion pump, to deliver drugs to the patient for managing arrhythmias or complications caused by arrhythmias, hypertension, or one or more other physiological conditions. In other examples, the implantable medical device 102 may include one or more electrodes configured to stimulate the patient's nervous system or provide stimulation to the muscles of the patient's airway, etc.

[0046] The wearable medical device 103 may include one or more wearable or external medical sensors or devices (e.g., an automatic external defibrillator (AED), a Holter monitor, a patch-based device, a smart watch, a smart accessory, a wrist or finger-worn medical device such as a finger-based photoplethysmography sensor, etc.). The wearable medical device 103 may include a sensing circuit coupled to one or more electrodes or sensors. In an example, the sensing circuit of the wearable medical device 103 may be configured to sense cardiac electrical activity, such as an ambulatory ECG, via electrodes electrically coupled to the sensing circuit.

[0047] The external system 105 may include a dedicated hardware / software system, such as a programmer, a remote server-based patient management system, or alternatively a system defined primarily by software running on a standard personal computer. The external system 105 may manage the patient 101 through the implantable medical device 102 or one or more other movable medical devices connected to the external system 105 via the communication link 111. In other examples, the implantable medical device 102 may be connected to the wearable medical device 103, or the wearable medical device 103 may be connected to the external system 105 via the communication link 111. This may include, for example, programming the implantable medical device 102 to perform one or more of acquiring physiological data, performing at least one self-diagnostic test (such as for device operating status), analyzing physiological data, or optionally delivering or adjusting treatment for the patient 101. In addition, the external system 105 may send information to the implantable medical device 102 or the wearable medical device 103 via the communication link 111, or receive information from the implantable medical device 102 or the wearable medical device 103. Examples of information may include: real-time or stored physiological data from patient 101; diagnostic data, such as detection of patient hydration status, hospitalization, response to therapy delivered to patient 101, or device operating status of implantable medical device 102 or wearable medical device 103 (e.g., battery status, lead impedance, etc.). Communication link 111 may be an inductive telemetry link, a capacitive telemetry link, or a radio frequency (RF) telemetry link, or wireless telemetry based on, for example, "strong" Bluetooth or IEEE 802.11 Wireless Fidelity "Wi-Fi" interface standards. Other configurations and combinations of patient data source interfaces are also possible.

[0048] The external system 105 may include an external device 106 located near one or more movable medical devices, and a remote device 108 located relatively far from one or more movable medical devices to communicate with the external device 106 via a communication network 107. Examples of the external device 106 may include a medical device programmer. The remote device 108 may be configured to evaluate the collected patient or patient information and provide alarm notifications, as well as other possible functions. In an example, the remote device 108 may include a centralized server that acts as a central hub for the storage and analysis of collected data. The server may be configured as a single, multiple or distributed computing and processing system. The remote device 108 may receive data from multiple patients. The data may be collected by one or more movable medical devices and other data acquisition sensors or devices associated with the patient 101. The server may include a memory device to store the data in a patient database. The server may include an alarm analyzer circuit to evaluate the collected data to determine whether a specific alarm condition is met. The satisfaction of the alarm condition may trigger the generation of an alarm notification, such as provided by one or more human-perceivable user interfaces. In some examples, the alarm condition may alternatively or additionally be evaluated by one or more movable medical devices (such as implantable medical devices). By way of example, the alarm notification may include a web page update, a telephone or pager call, an email, SMS, a text or "instant" message, and a message to the patient and a direct notification to both emergency services and clinicians. Other alarm notifications are also possible. The server may include an alarm prioritizer circuit configured to prioritize the alarm notifications. For example, a similarity metric between physiological data associated with the detected medical event and physiological data associated with historical alarms may be used to prioritize alarms for detected medical events.

[0049] The remote device 108 may additionally include one or more locally configured clients or remote clients securely connected to the server via the communication network 107. Examples of clients may include personal desktops, laptops, mobile devices, or other computing devices. System users, such as clinicians or other qualified medical professionals, can use the client to securely access stored patient data assembled in a database in the server, and select and prioritize patients and alarms for healthcare provision. In addition to generating alarm notifications, the remote device 108, including the server and interconnected clients, can also execute a follow-up plan by sending a follow-up request to one or more movable medical devices, or by sending a message or other communication to the patient 101 (e.g., the patient), a clinician, or an authorized third party as a compliance notification.

[0050] The communication network 107 can provide wired or wireless interconnection. In an example, the communication network 107 can be based on the Transmission Control Protocol / Internet Protocol (TCP / IP) network communication specification, although other types or combinations of networking implementations are also possible. Similarly, other network topologies and arrangements are also possible.

[0051] One or more of the external device 106 or remote device 108 can use information collected from the IMD 102 or the wearable medical device 103 (such as an ambulatory EGM or ambulatory ECG collected from the patient 101) to estimate physiological parameters, detect physiological events, or detect the operating state of the IMD 102 or the wearable medical device 103. In various examples, artificial intelligence (AI) or machine learning (ML) can be used to assist in estimating physiological parameters, or detecting physiological events or device operating states. For example, one or more of the external device 106 or remote device 108 can include an ML engine that uses a trained ML model to evaluate and identify different physiological events or physiological parameters. In some examples, one or more of the external device 106 or remote device 108 can include a computing platform that utilizes parallel processing with interconnected processing nodes and queues that form a workflow for substantially simultaneously estimating multiple physiological parameters or detecting multiple physiological events.

[0052] In some examples, the IMD 102 or wearable medical device 103 may collect other sensor signals other than the ambulatory ECG or EGM, as described below with reference to Figure 2Such sensors may have different physical properties, such as different form factors or sensor types. For example, different sensors may have form factors suitable for wearing on different body parts (e.g., helmets, glasses, headphones, necklaces, chest straps, belts, watches, bracelets, gloves, rings, ankle straps, shoes, socks, removable patches or clothing) or implanted in different body positions. In addition, specific sensors designed for specific locations or general purposes may have different profiles, surface areas or shapes, which may affect the received signals in different ways. Such sensor signals may also be used to estimate physiological parameters, detect physiological events, or detect the operating state of a device. In an example, a movable ECG or EMG and one or more other sensors of the corresponding type and form factor may be used to continuously detect physiological events. Continuous detection may involve using a movable ECG or EMG to estimate physiological parameters or generate initial detection of physiological events, and when the estimated physiological parameters or initial detection of physiological events meet specific conditions, one or more other dedicated sensors are triggered to directly measure physiological parameters or detect physiological events. ECG or EGM-based parameter estimation event detection can be confirmed, rejected, or modified by direct measurement or detection by one or more dedicated sensors. In some examples, direct measurement or detection by dedicated sensors can be used to calibrate ECG or EGM-based physiological parameter estimation or physiological event detection. Figure 3 Examples of AI-based physiological parameter estimation and physiological event detection using movable EGM are discussed.

[0053] One or more of the external device 106 or remote device 108 may output the detected medical event to a system user, such as a patient or clinician, or to a process including, for example, an instance of a computer program executable in a microprocessor. In an example, the process may include automatically generating a recommendation for antiarrhythmic therapy, or a recommendation for further diagnostic testing or treatment. In an example, the external device 106 or remote device 108 may include a corresponding display unit for displaying physiological or functional signals, or an alarm, an alert, an emergency call, or other form of warning to signal the detection of an arrhythmia. In some examples, the external system 105 may include an external data processor configured to analyze physiological or functional signals received by one or more movable medical devices and confirm or reject the detection of an arrhythmia. Computationally intensive algorithms, such as machine learning algorithms, may be implemented in the external data processor to retrospectively process data to detect arrhythmias.

[0054] Portions of one or more movable medical devices or external systems 105 may be implemented using hardware, software, firmware, or a combination thereof. Portions of one or more movable medical devices or external systems 105 may be implemented using dedicated circuits that may be constructed or configured to perform one or more functions, or may be implemented using general-purpose circuits that may be programmed or otherwise configured to perform one or more functions. Such general-purpose circuits may include a microprocessor or a portion thereof, a microcontroller or a portion thereof, or a programmable logic circuit, a memory circuit, a network interface, and various components for interconnecting these components. For example, a "comparator" may include, among other things, an electronic circuit comparator that may be constructed to perform a specific function of a comparison between two signals, or the comparator may be implemented as a portion of a general-purpose circuit that may be driven by a code that instructs a portion of a general-purpose circuit to perform a comparison between two signals. A "sensor" may include an electronic circuit configured to receive information and provide an electronic output representing such received information.

[0055] The therapeutic device 110 can be configured to send information to or receive information from one or more of the removable medical device or the external system 105 using the communication link 111. In an example, one or more removable medical devices, the external device 106, or the remote device 108 can be configured to control one or more parameters of the therapeutic device 110. The external system 105 can allow programming of one or more removable medical devices, and can receive information about one or more signals acquired by one or more removable medical devices, such as can be received via the communication link 111. The external system 105 can include a local external implantable medical device programmer. The external system 105 can include a remote patient management system, which can monitor the patient's status or adjust one or more treatments, such as from a remote location.

[0056] Figure 2 An example system 200 is shown, which includes an active medical device (AMD) 202 configured to sense or detect physiological information from a patient 101 via one or more sensors. The AMD 202 can be a single device or multiple medical devices or monitors. The AMD 202 can be an example of one or more of an implantable medical device 102 or a wearable medical device 103, and is configured to use the patient's active physiological information to estimate physiological parameters, detect physiological events, or detect device operating status.

[0057] AMD 202 may include one or more physiological sensors configured to sense corresponding physiological information of patient 101. Examples of physiological information may include cardiac or respiratory parameters, circulating biomarkers, systemic or local fluid states, neurological conditions, and patient response to medication or other treatment regimens, among others. AMD 202 may include circuits or microprocessors that may use at least some of the sensor data (such as a mobile ECG or EGM collected by the cardiac sensor 210) to estimate physiological parameters, detect physiological events, or detect device operating status. In examples of AI-based physiological parameter estimation and physiological event detection systems, sensor data may be used to establish training data sets for training and validating ML models. In some examples, ML models trained for estimating physiological parameters using ECG or EGM may be updated or refined using direct measurements of physiological parameters by dedicated sensors, as described below with reference to Figure 3 Further described.

[0058] By way of example and not limitation, the sensor may include one or more of the following: a respiration sensor 204, a heart sound sensor 206, an impedance sensor 208, a heart sensor 210, an activity sensor 212, a posture sensor 214, a pressure sensor 216, a temperature sensor 218, a blood oxygen sensor 220, or a chemical sensor 222. The respiration sensor 204 may be configured to receive respiration information including, but not limited to, indications of respiration rate (RR), respiration volume (tidal volume), RSBI, dyspnea, shortness of breath, etc. Such respiration information may be used to detect respiratory events such as pulmonary edema, dyspnea, and pneumonia, etc.

[0059] The heart sound sensor 206 can be in the form of an accelerometer or a microphone sensor, and can be configured to receive heart sound information, including but not limited to the intensity (e.g., amplitude) of one or more of the S1, S2, S3, or S4 heart sound components, and timing parameters such as the systolic timing interval measured using the heart sound components. Such heart sound information can be used to detect deterioration of cardiac function (e.g., decreased contractility, decreased ejection fraction, decreased systolic blood pressure, increased end-diastolic volume, and decreased cardiac output), diastolic abnormalities, myocardial depression, ischemia and microvascular dysfunction, arrhythmias, and WHF, etc. For example, a decrease in the S1 amplitude can indicate decreased myocardial contractility, a decrease in the S2 amplitude can indicate decreased arterial blood pressure and cardiac output, and an increase in S3, S4, or the S4 / S3 ratio can indicate deterioration of diastolic function.

[0060] The microphone or accelerometer sensor can sense physiological information other than heart sounds. In an example, a microphone can be used to sense information about the patient's voice, which can also be analyzed to detect patient signs or symptoms associated with specific physiological events. The microphone sensor can sense respiration (such as breathing rate), which can be used to detect shortness of breath or other breathing disorders. The accelerometer sensor can be used to sense physical activity that can indicate fatigue or neurological disorders (such as tremors, gait changes, or seizures), or sense body tremors that can indicate stiffness.

[0061] The impedance sensor 208 can be configured to receive impedance information (e.g., intracardiac, intrathoracic, or transthoracic impedance). A decrease in body impedance can indicate fluid accumulation, such as in the lungs known as pulmonary edema. In an example, the impedance sensor 208 can detect pulmonary edema based on a decrease in thoracic impedance. In some examples, the impedance sensor 208 can detect an increase in body impedance indicating dehydration.

[0062] The cardiac sensor 210 may be configured to receive cardiac electrical information, such as that described above with reference to Figure 1 The ECG or EGM. ECG or EGM can be used to detect cardiac parameters or cardiac events, such as heart rate, heart rate variability, cardiac synchrony or various arrhythmias.

[0063] Activity sensor 212 can be configured to receive information about body movement (e.g., activity, steps, etc.), and posture sensor 214 can be configured to receive posture or position information. Changes in body activity or changes in posture can indicate the development or worsening of heart failure. The body activity or movement information and posture or position information can be used additionally or alternatively to trigger one or more other physiological sensors, such as heart sounds, impedance or pressure data obtained at a specific body activity level or a specific posture.

[0064] The pressure sensor 216 can be configured to receive pressure information. In an example, the pressure sensor 216 is a blood pressure sensor configured to sense blood pressure, which can also be used to detect hypotension. In another example, the pressure sensor 216 is configured to sense abdominal pressure.

[0065] The temperature sensor 218 may be configured to receive body temperature information. Examples of body temperature sensors may include thermocouples, thermistors, infrared sensors, or temperature sensing integrated circuits. Since fever is one of the adverse systemic inflammatory responses to various medical diseases, high body temperature (e.g., above 38° C.) and other sensor information may be used to detect physiological events, or to calibrate physiological event detectors based on other physiological information (such as an ambulatory ECG or EGM).

[0066] Blood oxygen sensor 220 can be configured to receive information about blood oxygen saturation. In an example, blood oxygen sensor 220 is a pulse oximeter. Patients with certain medical conditions may experience hypoxia, which is a condition where the body or body part lacks adequate oxygen supply at the tissue level. Low arterial oxygen saturation (e.g., below 92%) and other sensor information can be used to detect physiological events, or to calibrate a physiological event detector based on other physiological information (such as an ambulatory ECG or EGM).

[0067] Chemical sensor 222 can be configured to receive information of one or more blood chemicals or circulating biomarkers. Circulating biomarkers are typically nucleic acids, extracellular vesicles, proteins, and metabolites from all metastatic sites, as well as normal organ physiological turnover or the effects of systemic drug therapy. Examples of biomarkers can include plasma glucose; biomarkers of renal injury, such as blood urea nitrogen (BUN) and creatinine; cardiac biomarkers such as natriuretic peptides, myoglobin, troponin, and creatine kinase, etc. Circulating biomarkers can additionally include blood electrolyte levels, such as potassium, sodium, or calcium levels. Data from chemical sensor 222 can be used to detect various types of physiological events, such as arrhythmias, heart failure, renal dysfunction, to control conditions such as electrolyte imbalances, such as hyperkalemia or hypokalemia (abnormally high or low blood potassium levels), or hyperglycemia or hypoglycemia (abnormally high or low blood glucose levels), whether as an intrinsic physiological response or a response to drug therapy (e.g., side effects, inappropriate treatment, or insufficient dosage).

[0068] Figure 3 is a block diagram illustrating a portion of an exemplary AI-based physiological event detection system 300 that can be configured to detect physiological events or estimate physiological parameters using patient data (such as an ambulatory ECG or EGM). At least a portion of the system 300 can be implemented in one or more of the implantable medical device 102, the wearable medical device 103, or the AMD 202. Additionally or alternatively, at least a portion of the system 300 can be implemented in a device in a remote patient management system (such as the external device 106 or the remote device 108).

[0069] System 300 may include a patient data receiver 310, a processor 320, a user interface 330, and a device controller 340. In some examples, system 300 may include or be communicatively coupled to a database 350. Patient data receiver 310 may be configured to receive data from one or more physiological sensors (such as those described above with reference to Figure 2The sensors 204-222 discussed above receive patient information, such as physiological information of a patient (or a group of patients). In an example, a patient data receiver 310 may be included in a wearable or implantable device and collects ambulatory patient data, such as an ambulatory ECG or EGM, continuously or periodically or at a specific frequency (such as daily, weekly, biweekly, or monthly).

[0070] The processor 320 may include one or more of a physiological parameter estimator 322, a physiological event detector 324, or a device operating state detector 326. The physiological parameter estimator 322 may estimate the physiological parameters using at least a portion of the received patient information (such as an ambulatory ECG or EGM). In an example, the estimated physiological parameters may include one or more cardiac parameters, such as ECG parameters (e.g., QT interval prolongation, ST segment elevation indicating acute myocardial infarction, T wave inversion, heart rate, heart rate variability, premature atrial contraction (PAC), premature ventricular contraction (PVC), cardiac timing interval, etc.). Cardiac parameters may also include systolic parameters (e.g., left ventricular ejection fraction, right ventricular ejection fraction, stroke volume, heart wall thickness indicating hypertrophy, heart wall motion or wall motion changes, ejection time, pre-ejection period, etc.), diastolic parameters, such as those obtained from transthoracic Doppler echocardiographic analysis (e.g., E and A peak velocities and their ratio E / A, E wave deceleration time, isovolumetric relaxation time), and valve function parameters (e.g., stenosis or regurgitation associated with one or more of the aortic, mitral, pulmonary, or tricuspid valve values). In another example, the estimated physiological parameters may include one or more respiratory parameters, such as indications of chronic obstructive pulmonary disease (COPD), central sleep apnea or obstructive sleep apnea, and other respiratory diseases. In another example, the estimated physiological parameters may include systemic or local fluid status, such as fluid accumulation in the lungs or other organs or tissues. Fluid overload is associated with increased mortality and can also lead to several complications such as pulmonary edema. In yet another example, the estimated physiological parameters may include one or more circulating biomarkers, such as blood glucose levels, various renal injury biomarkers, arrhythmia and heart failure biomarkers, or blood electrolyte levels.

[0071] The physiological event detector 324 may detect physiological events using at least a portion of the received patient information, such as an ambulatory ECG or EGM. In some examples, the physiological event detector 324 may detect physiological events using estimated physiological parameters provided by the physiological parameter estimator 322. The detected physiological events may include, for example, arrhythmias, such as atrial fibrillation (AF), atrial flutter, supraventricular tachycardia, ventricular tachycardia, ventricular fibrillation, Wolff-Parkinson-White (WPW) syndrome, reentrant arrhythmias, and the like. In an example, the detected physiological events may include an arrhythmia burden indicating the frequency or time spent on a particular type of arrhythmia, such as an AF burden represented by the proportion (e.g., percentage) of time that an individual is in AF during monitoring. In an example, the detected physiological events may include a worsening heart failure (WHF) event, or heart failure comorbidities, such as renal insufficiency, diabetes, chronic obstructive pulmonary disease (COPD), sleep disorders such as obstructive and central sleep apnea, anemia, liver disease such as cirrhosis, or other volume regulation diseases. In another example, the detected physiological events may include neurological diseases, such as sympathovagal imbalance syndrome, orthostatic hypotension, pain, migraine, and mood disorders, etc. In another example, the detected physiological events may include the patient's response to a drug or other treatment regimen, such as the side effects experienced by the patient. In yet another example, the detected physiological events may include cardiotoxicity or cardiopulmonary toxicity of immunotherapy, which may be represented by worsening cardiac function (e.g., decreased contractility, decreased ejection fraction, decreased systolic pressure, increased end-diastolic volume, and decreased cardiac output), diastolic abnormalities, myocardial depression, ischemia and microvascular dysfunction, arrhythmias, and WHF, etc.

[0072] The device operating state detector 326 can detect the operating state of a movable device, such as one or more of the implantable medical device 102, the wearable medical device 103, or the AMD 202. In an example, the device operating state may include a change in the position, posture, or orientation of a movable medical device on a subject, such as a flipping of an implantable device (e.g., the implantable medical device 102) in a skin pocket, or rotation syndrome associated with inadvertent or intentional manipulation of the implantable device within its skin pocket by the patient. Flipping or rotation of the implantable device may cause the lead to curl or shift, resulting in the inability to provide treatment or providing inappropriate treatment. In another example, the device operating state may include a change in the device-tissue interface, such as the presence of air in a skin pocket for the implantable device. In yet another example, the device operating state may include a state of its operating environment, such as the presence of noise or interference or the co-implantation of another device.

[0073] In some examples, the patient data receiver 310 can continuously or periodically collect mobile patient data, such as a mobile ECG or EGM. Therefore, one or more of the physiological parameter estimator 322, the physiological event detector 324, or the device operating state detector 326 can perform its corresponding tasks continuously or periodically at a specific frequency (such as daily, weekly, biweekly, or monthly). In some examples, the patient data receiver 310 can collect mobile patient data at a variable frequency (such as changing from daily to weekly to biweekly to monthly). The continuously estimated parameters and detected events can be trended over time to provide a more accurate assessment of disease progression and patient health.

[0074] One or more of the physiological parameter estimator 322, the physiological event detector 324, or the device operating state detector 326 can perform its corresponding parameter estimation or event detection task based on the analysis of the mobile ECG or EGM. Although the mobile ECG or EGM does not provide direct measurement of certain physiological parameters, which may be directly measured using dedicated sensors (e.g., respiratory parameters, circulatory biomarkers, and certain cardiac parameters), the operation of dedicated sensors is generally associated with higher system complexity and implementation costs, and some sensors are not sensitive to providing early indications of events. In contrast, systems, devices, and methods that are less expensive, simpler, or less obtrusive can quickly acquire a mobile ECG or EGM and can provide estimates of physiological parameters and early detection of physiological events or device operating states.

[0075] Artificial intelligence (AI) or machine learning (ML) can be used to improve ECG- or EGM-based physiological parameter estimation or physiological event detection. Figure 3As shown, one or more of the physiological parameter estimator 322, the physiological event detector 324, or the device operating state detector 326 can use one or more trained machine learning (ML) models 321 to perform their corresponding parameter estimation or event detection tasks. One or more trained ML models 321 may have a neural network structure including an input layer, one or more hidden layers, and an output layer. Patient information received by the patient data receiver 310 (such as a mobile ECG or EGM, or a signal metric trend of the ECG or EGM) may be provided to the input layer of one or more ML models 321. One or more ML models 321 may propagate input data or data trends to the output layer through one or more hidden layers. One or more ML models 321 may provide the system 300 with the ability to perform tasks (without explicit programming) by inferring based on patterns found in data analysis. One or more ML models 321 explore the research and construction of algorithms (e.g., ML algorithms) that can learn from existing data and predict new data. Such algorithms operate by constructing one or more ML models 321 from training data to make data-driven predictions or decisions that are represented as outputs or evaluations.

[0076] One or more ML models 321 may be trained using supervised learning or unsupervised learning. Supervised learning uses prior knowledge (e.g., examples relating inputs to outputs or outcomes) to learn the relationship between inputs and outputs. The goal of supervised learning is to learn a function that, given some training data, best approximates the relationship between the training inputs and outputs so that the ML model can enforce the same relationship to generate the corresponding output when given an input. Unsupervised learning is the use of information that is neither classified nor labeled to train an ML algorithm and allow the algorithm to act on that information without guidance. Unsupervised learning is useful in exploratory analysis because it can automatically identify structure in the data.

[0077] Some common tasks for unsupervised learning include clustering, representation learning, and density estimation. Some examples of commonly used unsupervised learning algorithms are K-means clustering, principal component analysis, and autoencoders. Some common tasks for supervised learning are classification problems and regression problems. Classification problems (also known as grouping problems) aim to classify items into one of several category values. Regression algorithms aim to quantify some items (e.g., by providing scores for some input values). Some examples of commonly used supervised ML algorithms are logistic regression (LR), naive Bayes, random forest (RF), neural networks (NN), deep neural networks (DNN), matrix factorization, and support vector machines (SVM).

[0078] In an example, deep learning can be used to train an ML model. The ML model has an architecture of a deep neural network example of a DNN, including a convolutional neural network (CNN), a recurrent neural network (RNN), a deep belief network (DBN), a long and short-term memory (LSTM) network, a transfer learning network, or a hybrid neural network including two or more neural network models of different types or different model configurations.

[0079] Another type of ML is collaborative learning, which trains algorithms across multiple decentralized devices holding local data without exchanging data. This approach contrasts with traditional centralized machine learning techniques (where all local data sets are uploaded to one server) and more classic decentralized approaches (which typically assume that local data samples are distributed the same). Collaborative learning enables multiple participants to build common, robust machine learning models without sharing data, allowing key issues such as data privacy, data security, data access rights, and heterogeneous data access to be addressed.

[0080] like Figure 3 As shown, one or more trained ML models 321 may be generated by a training module 328. The training module 328 may be included in Figure 3 In the processor 320 shown, or alternatively implemented in a separate unit. In order to train the ML model, a training data set can be constructed using the patient information collected from the patient population, and each patient population has the corresponding AMD (e.g., implantable device or wearable device) of the same or similar type. The information of the patient population can be stored in the database 350. The training data can include the patient's physiological data collected by its corresponding AMD, and the evaluation of the specific physiological parameters, physiological events or running state of the corresponding AMD. The evaluation can be provided by a user (e.g., a clinician), and includes the features of physiological parameters or detected events (e.g., for the start, termination, duration, rate, rhythm and heart time and synchronization of arrhythmia attacks). In some examples, it can be evaluated based on the diagnostic test using a dedicated sensor or instrument (e.g., echocardiogram or other diagnostic imaging tests, electrical mapping for diagnosing arrhythmia, invasive glucose tests via finger puncture, glucose strips or continuous glucose monitors). The training data can additionally include the demographic data, medical history and health status, treatment and results of the patient.

[0081] In the example of training an EGM-based ML model, the training data may include, in addition to other patient information, data trends of EGMs or EGM-generated signal metrics collected from a patient population, and assessments of physiological parameters or identification of physiological events, which may be provided as output labels by a human expert. Training of the ML model may be performed continuously or periodically, or in near real time as additional patient information is received by the patient data receiver 310. Training involves algorithmically adjusting one or more ML model parameters until the trained ML model satisfies specific training convergence criteria (such as the model output is as close as possible (within a specific margin) to the output label).

[0082] In some examples, multiple ML models can be trained separately and applied in different applications. For example, a first ML model (or a first set of ML models) can be trained to establish a correspondence between EGMs collected from a population (optionally with other information) and an evaluation of a physiological parameter. The physiological parameter estimator 322 can apply the EGMs collected from the patient to one or more trained first ML models to estimate one or more physiological parameters. In another example, a second ML model (or a second set of ML models) can be trained to establish a correspondence between EGMs collected from a population (optionally with other information) and an evaluation of a physiological event. The physiological event detector 324 can apply the EGMs sensed from the patient to one or more trained second ML models to detect one or more physiological events. In another example, a third ML model (or a third set of ML models) can be trained to establish a correspondence between EGMs collected from a population (optionally with other information) and an evaluation of a device operating state. The device operating state detector 326 can apply the EGMs sensed from the patient to one or more trained third ML models to detect the device operating state. In some examples, one or more trained third ML models can detect device operating states (e.g., airbag / tissue contact, or device flipping or rotation) based on signal features extracted from the input EGM (e.g., EGM noise or low amplitude EGM). Figure 4A-4B Examples of training ML models and using the trained ML models to estimate physiological parameters or detect physiological events using a patient's EGM are discussed.

[0083] In some examples, processor 320 can be configured to perform parallel computations, where multiple computing resources are used to perform corresponding computing tasks, such as estimating multiple physiological parameters or detecting multiple physiological events substantially or simultaneously. In an example, processor 320 can be a multi-core processor including multiple computing units, each of which is configured to apply a corresponding portion of the subject's movable EGM to a corresponding trained ML model to estimate a corresponding physiological parameter in the subject or detect a corresponding physiological event thereof. In some examples, multiple processors (each similar to processor 320 or a portion thereof) can perform substantially concurrent computations, such as estimating multiple physiological parameters or detecting multiple physiological events.

[0084] The user interface 330 may include an output unit configured to output one or more of the estimated physiological parameters, detected physiological events, or detected device operating states to a user (e.g., a patient, a caregiver, or a clinician), such as to a display or one or more other user interfaces. The output may include a score, trend, or other indication. The output may include recommendations for conducting further diagnostic tests, consulting a caregiver, taking or adjusting medications, lifestyle management, comorbidity assessment, management or prevention of major adverse cardiovascular events and / or cardiovascular mortality, and other medical standards depending on the type of physiological event. In other examples, the evaluation circuit 304 may be configured to provide an output to another circuit, machine, or process.

[0085] The device controller 340 can control, adjust or stop the treatment of the medical device, the drug delivery system, etc. based on the estimated physiological parameters, the detected physiological events or the detected device operating state. In some examples, when the estimated physiological parameters meet specific conditions, the device controller 340 can trigger a dedicated sensor to directly measure the physiological parameters. Based on the physiological parameter estimation or physiological event detection based on ECG or EGM, a dedicated sensor of a specific type and / or a specific form factor can be determined, or information about a specific sensor can be confirmed (for example, the sensor type or shape received can be confirmed using the sensed information). This determination or confirmation can improve the system performance of different sensors, or enable the device controller to be used with different sensors or information from different sensors or sources. For example, if the trained ML model estimates the abnormal level of a biomarker (for example, plasma glucose level) based on the input EGM, the glucose sensor can be triggered to perform a direct measurement of the glucose level. The movable EGM and dedicated sensor signals can be collected continuously or periodically at a corresponding frequency (such as daily, weekly, biweekly or monthly). In some examples, the movable EGM and sensor signals can be collected at a variable frequency (such as varying from daily to weekly, biweekly to monthly). ECG or EGM-based physiological parameter estimates or physiological event detection can be confirmed, rejected, or modified by direct measurement or detection by dedicated sensors. This continuous detection involving initial parameter estimates based on ECG or EGM and event detection followed by confirmation detection using dedicated sensors can improve the efficiency and accuracy of physiological event detection, reduce system complexity, and save system and implementation costs.

[0086] In some examples, the device controller 340 can use direct measurements of physiological parameters by dedicated sensors to calibrate one or more of the physiological parameter estimator 322, the physiological event detector 324, or the device operating state detector 326. For example, if the trained ML model detects an abnormal level of a renal injury biomarker based on the input EGM, the direct measurement of the level of the renal injury biomarker can be taken and used to adjust the EGM acquisition (e.g., time, sampling rate, duration) and analysis. In some examples, the direct measurement by the dedicated sensor can be used to update one or more trained ML models 321 (e.g., by adjusting the weights assigned to one or more nodes of the neural network), or retrain one or more ML models 321.

[0087] In some examples, the processor 320 can detect information about device interference from a co-implant or an external noise source from received patient data (e.g., an ambulatory ECG or EGM). Examples of device interference can include pacing pulses from a co-implant, electromagnetic noise, optical sensing noise, or accelerometer interference, among others. Based on the detected interference information, the device controller 340 can calibrate one or more of the physiological parameter estimator 322, the physiological event detector 324, or the device operating state detector 326, or adjust one or more trained ML models 321. In some examples, the device controller 340 can adjust the EGM sensing configuration or other ambulatory sensor settings to avoid or reduce interference and improve signal quality.

[0088] Portions of the system 300, such as the processor 320 (or a portion thereof, such as one or more of the physiological parameter estimator 322, the physiological event detector 324, the device operating state detector 326, or the training module 328) or one or more of the device controllers 340, may include a corresponding circuit set including one or more other circuits or subcircuits. The circuits or subcircuits may perform the functions, methods, or techniques described herein alone or in combination. In an example, the hardware of the circuit set may be invariably designed to perform a specific operation (e.g., hardwired). In an example, the hardware of the circuit set may include variably connected physical components (e.g., execution units, transistors, simple circuits, etc.) that include physically modified computer-readable media (e.g., magnetic, electrical, movable placement of invariant aggregated particles, etc.) to encode instructions for specific operations. When the physical components are connected, the basic electrical characteristics of the hardware components change, for example, from an insulator to a conductor, and vice versa. The instructions enable embedded hardware (e.g., an execution unit or a loading mechanism) to create components of the circuit set in hardware via variable connections to perform parts of a specific operation when operated. Thus, when the device is operating, the computer-readable medium is communicatively coupled to other components of the circuit set component. In examples, any physical component can be used in more than one component of more than one circuit set. For example, under operation, an execution unit can be used in a first circuit of a first circuit set at one point in time and reused by a second circuit in the first circuit set, or reused by a third circuit in the second circuit set at a different time.

[0089] Portions of the system 300, such as the processor 320 (or a portion thereof) or one or more of the device controller 340, may be implemented as part of a microprocessor circuit. The microprocessor circuit may be a special-purpose processor, such as a digital signal processor, an application-specific integrated circuit (ASIC), a microprocessor, or other type of processor for processing information including physiological signals received from the patient data receiver 310. Alternatively, the microprocessor circuit may be a general-purpose processor that can receive and execute instruction sets that perform the functions, methods, or techniques described herein.

[0090] In some examples, portions of system 300 may be implemented in a cloud-based computing platform ("cloud") that provides cloud-based services including, for example, configuration of data storage, computing services, and customer service. For example, database 350 may reside in a cloud storage device. Additionally or alternatively, the cloud may provide computing services for one or more of physiological parameter estimator 322, physiological event detector 324, device operating state detector 326, or training module 328 to achieve their respective goals. An authorized user may remotely access the cloud storage device to retrieve physiological information from database 350, or use the computing services therein to perform data analysis, such as estimating physiological parameters in a patient or detecting their physiological events, or detecting the operating state of a patient's AMD.

[0091] Figure 4A-4B is a diagram showing an example of training an EGM-based ML model and using the trained model to estimate physiological parameters, detect physiological events, or detect device operating status based on input EGM sensed from a patient. Figure 4A The ML model training (or learning) phase is shown, during which the ML model 420 can be trained using training data including EGM 410 collected from the group. For each EGM, the training data can also include corresponding annotation data 415, which includes an assessment of the physiological parameters, physiological events or device operating status of the corresponding EGM. The assessment can be provided by the user, such as based on direct measurement of physiological parameters by dedicated sensors, or diagnostic tests using dedicated instruments (e.g., echocardiograms or other diagnostic imaging tests, electrical mapping for diagnosing arrhythmias, invasive glucose tests via finger punctures, glucose strips or continuous glucose monitors). The ML model 740 can be trained using supervised learning, unsupervised learning or reinforcement learning. Examples of ML model architectures and algorithms can include, for example, decision trees, neural networks, support vector machines, or deep neural networks, etc. Examples of deep neural networks can include convolutional neural networks (CNNs), recursive neural networks (RNNs), deep belief networks (DBNs), long and short-term memory (LSTM) networks, transfer learning networks, or hybrid neural networks including two or more neural network models of different types or different model configurations. The training of the ML model can be performed continuously or periodically, or in near real time as additional patient data is obtained. Training involves adjusting one or more ML model parameters through an algorithm until the trained ML model meets specific training convergence criteria. The trained ML model 420 can establish a correspondence between the EGM and one or more of a physiological parameter, a physiological event, or a device operating state.

[0092] Figure 4BAn inference phase is shown, during which the EGM 430 sensed from the patient can be applied to the trained ML model 420 to automatically estimate one or more physiological parameters, detect physiological events, or detect device operating states. The estimated parameters and / or detected events 450 can be provided to a user (e.g., a patient, a caregiver, or a clinician), such as to a display or one or more other user interfaces. Additionally or alternatively, the estimated parameters and / or detected events 450 can be output to another circuit, machine, or process, such as to control, adjust, or stop a treatment of a medical device, a drug delivery system, or the like. In some examples, the estimated parameters and / or detected events 450 can be used to trigger direct measurement of physiological parameters using a dedicated sensor, or to calibrate one or more of the physiological parameter estimator 322, the physiological event detector 324, or the device operating state detector 326, or to update the trained ML model 420, such as by adjusting the weights assigned to one or more nodes in the neural network.

[0093] Figure 5 is a flow chart illustrating an example of a method 500 for detecting physiological events or estimating physiological parameters using patient information such as an electrocardiogram (EGM). The method 500 may be implemented in and performed by a computing device, such as one or more of an implantable medical device 102, a wearable medical device 103, or an AMD 202. Alternatively, the method 500 may be implemented in and performed by a device in a remote patient management system, such as an external device 106, a remote device 108, or a cloud-based computing platform.

[0094] Method 500 begins at 510, where physiological information may be received from a patient using an active medical device (AMD) such as implantable medical device 102, wearable medical device 103, or AMD 202. Physiological information may be received using corresponding sensors or electrodes such as described above with reference to Figure 2 The sensors 204-222 discussed above are used to receive various physiological information. In the example, the received physiological information includes an active EGM. The active EGM, optionally together with other physiological information, can be collected continuously or periodically at a specific frequency (such as daily, weekly, biweekly or monthly).

[0095] At 520, the received movable EGM may be analyzed using a computing device or a portion thereof, such as the processor 320 of the AI-based physiological event detection system 300. The analysis may include applying the received movable EGM to one or more trained machine learning (ML) models, such as one or more trained ML models 321. As described above with reference to Figure 3 and 4AAs discussed, an ML model can be trained using a training data set that includes movable EGMs collected from a patient population and assessments of physiological parameters or physiological events in the patient population. The ML model can be trained using supervised learning or unsupervised learning. In an example, deep learning can be used to train the ML model. The ML model has the architecture of a deep neural network (DNN). Training of the ML model can be performed continuously or periodically, or in near real time when additional patient information is received. Training involves adjusting one or more ML model parameters by an algorithm until the trained ML model meets specific training convergence criteria. The trained ML model represents the correspondence between the movable EGM of the patient population and the physiological parameters or physiological events in the patient population. Such a trained ML model is also referred to as an EGM-based ML model.

[0096] At 530, based on the analysis of the EGM using one or more trained ML models, physiological parameters can be estimated and / or physiological events can be detected. Physiological parameter estimation and physiological event detection can be respectively performed using the methods described above with reference to Figure 3 The present invention relates to a physiological parameter estimator 322 and a physiological event detector 324 as discussed above. Examples of estimated physiological parameters may include cardiac parameters, respiratory parameters, circulating biomarkers, or systemic or local fluid states, etc. Examples of detected physiological events include arrhythmias, heart failure exacerbation events, heart failure complications, neurological diseases, or responses to drugs, etc.

[0097] In some examples, at step 520, an ML model (or a separate ML model in addition to an ML model trained for estimating physiological parameters or detecting physiological events) can be trained to establish a correspondence between EGMs collected from the population and assessments of device operating states, such as changes in position, posture, or orientation of the AMD, or changes in the device-tissue interface of the AMD. An animated EGM received from the subject can be applied to the trained model, and the operating state of the AMD can be detected at 530, such as using a device operating state detector 326.

[0098] At 540, the estimated physiological parameter or detected physiological event may be provided to a user or process, such as to a display or one or more other user interfaces. The output may include a recommendation or alert to any adverse event to a clinician. The estimated physiological parameter, detected physiological event, or detected device operating state may alternatively or additionally be used to control, adjust, or stop a treatment of a medical device, a drug delivery system, or the like. In some examples, additionally or alternatively, when the estimated physiological parameter meets a specific condition, a dedicated sensor may be triggered to directly measure the physiological parameter. Direct measurement of the physiological parameter may be used to calibrate the process of estimating the physiological parameter or detecting the physiological event or device operating state. In some examples, direct measurement of the physiological parameter may be used to adjust the trained ML model or used in the process of retraining the ML model.

[0099] Figure 6 Generally, a block diagram of an example machine 600 is shown on which any one or more techniques (eg, methods) discussed herein may be performed. Portions of this description may be applied to the computing framework of various portions of the AI-based physiological event detection system 300.

[0100] In an alternative embodiment, the machine 600 can operate as a standalone device, or can be connected (e.g., networked) to other machines. In a networked deployment, the machine 600 can operate in a server-client network environment with the capabilities of a server machine, a client machine, or both. In an example, the machine 600 can act as a peer machine in a peer-to-peer (P2P) (or other distributed) network environment. The machine 600 can be a personal computer (PC), a tablet PC, a set-top box (STB), a personal digital assistant (PDA), a mobile phone, a network device, a network router, a switch or a bridge, or any machine capable of executing instructions (sequentially or otherwise) specifying the actions to be taken by the machine. In addition, although only a single machine is shown, the term "machine" should also be considered to include any collection of the following machines, which independently or jointly execute one (or more) instruction sets to implement any one or more methods discussed herein, such as cloud computing, software as a service (SaaS), and other computer cluster configurations.

[0101] As described herein, examples may include logic or multiple components or mechanisms, or may be operated by them. A circuit set is a collection of circuits implemented in a tangible entity, which includes hardware (e.g., simple circuits, gates, logic, etc.). Circuit set components may be flexible over time and potential hardware variability. A circuit set includes components that can perform specific operations individually or in combination when operating. In an example, the hardware of a circuit set may be immutably designed to perform specific operations (e.g., hardwired). In an example, the hardware of a circuit set may include variably connected physical components (e.g., execution units, transistors, simple circuits, etc.), which include physically modified computer-readable media (e.g., magnetic, electrical, movable placement of invariant aggregated particles, etc.) to encode instructions for specific operations. When connecting physical components, the basic electrical properties of the hardware components change, for example, from an insulator to a conductor, and vice versa. Instructions enable embedded hardware (e.g., an execution unit or a loading mechanism) to create components of a circuit set in hardware via variable connections to perform parts of specific operations when operating. Thus, when the device is in operation, the computer-readable medium is communicatively coupled to other components of the circuit group component. In an example, any physical component can be used in more than one component of more than one circuit set. For example, in operation, an execution unit can be used in a first circuit of a first circuit set at one point in time and reused by a second circuit in the first circuit set, or reused by a third circuit in the second circuit set at a different time.

[0102] The machine (e.g., computer system) 600 may include a hardware processor 602 (e.g., a central processing unit (CPU), a graphics processing unit (GPU), a hardware processor core, or any combination thereof), a main memory 604, and a static memory 606, some or all of which may communicate with each other via an interconnect (e.g., a bus) 608. The machine 600 may also include a display unit 610 (e.g., a raster display, a vector display, a holographic display, etc.), an alphanumeric input device 612 (e.g., a keyboard), and a user interface (UI) navigation device 614 (e.g., a mouse). In an example, the display unit 610, the input device 612, and the UI navigation device 614 may be a touch screen display. The machine 600 may additionally include a storage device (e.g., a drive unit) 616, a signal generating device 618 (e.g., a speaker), a network interface device 620, and one or more sensors 621, such as a global positioning system (GPS) sensor, a compass, an accelerometer, or other sensors. The machine 600 may include an output controller 628, such as a serial (e.g., universal serial bus (USB), parallel or other wired or wireless (e.g., infrared (IR), near field communication (NFC), etc.) connection to communicate with or control one or more peripheral devices (e.g., printer, card reader, etc.).

[0103] The storage device 616 may include a machine-readable medium 622 having stored thereon one or more sets of data structures or instructions 624 (e.g., software) that embody or are utilized by any one or more of the techniques or functionality described herein. The instructions 624 may also reside, completely or at least partially, within the main memory 604, the static memory 606, or the hardware processor 602 during execution of the instructions 624 by the machine 600. In an example, one or any combination of the hardware processor 602, the main memory 604, the static memory 606, or the storage device 616 may constitute a machine-readable medium.

[0104] Although the machine-readable medium 622 is illustrated as a single medium, the term “machine-readable medium” may include a single medium or multiple media (eg, a centralized or distributed database, and / or associated caches and servers) configured to store one or more instructions 624 .

[0105] The term "machine-readable medium" may include any medium capable of storing, encoding or carrying instructions executed by the machine 600 and causing the machine 600 to perform any one or more of the techniques of the present disclosure, or any medium capable of storing, encoding or carrying data structures used by or associated with these instructions. Non-limiting machine-readable medium examples may include solid-state memory, optical media, and magnetic media. In an example, the aggregated machine-readable medium includes a machine-readable medium having a plurality of particles with a constant (e.g., stationary) mass. Therefore, the aggregated computer-readable medium is not a transient propagation signal. Specific examples of aggregated machine-readable media may include: non-volatile memory, such as semiconductor memory devices (e.g., electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM)) and flash memory devices; magnetic disks, such as internal hard disks and removable magnetic disks; magneto-optical disks; and CD-ROM and DVD-ROM disks.

[0106] The term "machine-readable medium" may include any medium capable of storing, encoding or carrying instructions executed by the machine 600 and causing the machine 600 to perform any one or more of the techniques of the present disclosure, or any medium capable of storing, encrypting or carrying data structures used by or associated with these instructions. Non-limiting examples of machine-readable media may include solid-state memory and optical and magnetic media. In one example, the aggregated machine-readable medium includes a machine-readable medium having a plurality of particles with a constant (e.g., stationary) mass. Therefore, the aggregated computer-readable medium is not a transient propagating signal. Specific examples of mass machine-readable media may include: non-volatile memory, such as semiconductor storage devices (e.g., electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM)) and flash memory devices; disks, such as internal hard disks and removable disks; magneto-optical disks; and CD-ROM and DVD-ROM optical disks. The instructions 624 may be further sent or received over a communication network 626 using a transmission medium via the network interface device 620 using any of a variety of transmission protocols (e.g., frame relay, Internet Protocol (IP), transmission control protocol (TCP), user datagram protocol (UDP), hypertext transfer protocol (HTTP), etc.). Example communication networks may include a local area network (LAN), a wide area network (WAN), a packet data network (e.g., the Internet), a mobile telephone network (e.g., a cellular network), a plain old telephone (POTS) network, and a wireless data network (e.g., a wireless network known as a wireless network). The Institute of Electrical and Electronics Engineers (IEEE) 802.11 family of standards, known as The IEEE 802.16 family of standards), the IEEE 802.15.4 family of standards, peer-to-peer (P2P) networks, and the like. In an example, the network interface device 620 may include one or more physical jacks (e.g., Ethernet, coaxial, or telephone jacks) or one or more antennas to connect to the communication network 626. In an example, the network interface device 620 may include multiple antennas to wirelessly communicate using at least one of single-input multiple-output (SIMO), multiple-input multiple-output (MIMO), or multiple-input single-output (MISO) technology. The term "transmission medium" should be understood to include any intangible medium that is capable of storing, encoding, or carrying instructions executed by the machine 800, and includes digital or analog communication signals or other intangible media to facilitate the communication of such software.

[0107] Various embodiments are shown in the above figures. One or more features from one or more of these embodiments may be combined to form other embodiments.

[0108] The method examples described herein may be at least partially machine or computer implemented. Some examples may include a computer-readable medium or machine-readable medium encoded with instructions, which are operable to configure an electronic device or system to perform the method described in the above example. The implementation of this method may include code, such as microcode, assembly language code, high-level language code, or the like. Such code may include computer-readable instructions for performing various methods. The code may form part of a computer program product. In addition, the code may be tangibly stored on one or more volatile or non-volatile computer-readable media during execution or at other times.

[0109] The foregoing detailed description is intended to be illustrative rather than limiting.The scope of the disclosure should, therefore, be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled.

Claims

1. A system for detecting a physiological event in a subject, the system comprising: A computing device configured to: receiving an ambulatory electrogram of the subject collected by an ambulatory medical device (AMD); and applying the received electrogram to a trained machine learning model to estimate a physiological parameter in the subject or detect a physiological event in the subject; as well as An output unit is configured to output the estimated physiological parameters or detected physiological events to a user or a process.

2. The system of claim 1, wherein the computing device is further configured to: determining or confirming a physical characteristic of at least one sensor using the active electrogram; and detecting a physiological event in the subject or estimating a physiological parameter in the subject using the active electrogram and the determined or confirmed physical characteristic of the at least one sensor, in, To detect the physiological event, the computing device is configured to: apply the received electrogram to a trained machine learning model to estimate the physiological parameter in the subject or detect the physiological event in the subject using the electrogram sensed by the at least one sensor, and Therein, the physical characteristic is a sensor type or a form factor of the at least one sensor.

3. The system of claim 2, wherein the computing device comprises a training module configured to generate the trained machine learning model, comprising: constructing a training data set comprising ambulatory electrograms collected from a patient population and an assessment of a physiological parameter or physiological event in the patient population; as well as The machine learning model is trained using the constructed training data set until convergence or training stop criteria are met, and the trained machine learning model represents the correspondence between the movable electrograms in the patient group and the physiological parameters or the physiological events in the patient group.

4. The system of claim 3, wherein the at least one sensor is configured to sense physiological information from the subject other than the electrogram, and The training module is configured to train the machine learning model using a deep learning algorithm including a deep neural network.

5. The system of any one of claims 2-4, wherein the computing device is configured to apply the received electrogram to a trained machine learning model to estimate the physiological parameter comprising at least one of: Cardiac parameters; Respiratory parameters; Circulating biomarkers; or Systemic or local fluid status.

6. The system of any one of claims 2-4, wherein the computing device is configured to apply the received electrogram to a trained machine learning model to detect the physiological event comprising at least one of: Heart rhythm disorders heart failure worsening events; Comorbidities of heart failure; neurological disease; or Response to medication.

7. The system of any one of claims 2-5, wherein the computing device is further configured to apply the received electrogram to a trained machine learning model to detect an operational state of the AMD.

8. The system of claim 7, wherein the operating status of the AMD indicates at least one of the following: a change in the position, posture or orientation of the AMD; or Changes in the device-tissue interface of AMD.

9. The system of any one of claims 1-8, comprising the AMD configured to continuously or periodically collect the ambulatory electrogram of the subject via one or more attachable or implantable electrodes.

10. The system of any one of claims 1-9, wherein the AMD comprises at least one of: Insertable heart monitor; a subcutaneous implantable cardioverter-defibrillator; or Wearable or portable heart monitor.

11. The system of claims 1-10, wherein the AMD comprises a therapy circuit configured to initiate or adjust therapy for the subject based on the estimated physiological parameter or detected physiological event.

12. The system of claim 2, wherein the computing device is configured to: applying the received electrogram to the trained machine learning model to estimate the physiological parameter; and In response to the estimated physiological parameter satisfying a condition, at least one of the at least one sensor is triggered to directly measure the physiological parameter.

13. A system according to claim 12, wherein the computing device includes a calibration circuit configured to adjust the trained machine learning model based at least on the directly measured physiological parameters.

14. The system of any one of claims 1-13, wherein the computing device is configured to perform parallel computations to estimate multiple physiological parameters or detect multiple physiological events substantially simultaneously.

15. The system of claim 14, wherein the computing device comprises: A plurality of processors, or a multi-core processor comprising a plurality of computing units, wherein each of the plurality of processors or the plurality of computing units is configured to apply a portion of the subject's movable electrogram to a corresponding trained machine learning model to estimate a corresponding physiological parameter in the subject or detect a corresponding physiological event in the subject.

Images