Probability threshold selection for generating arrhythmia notifications

Abstract

A technique for monitoring the occurrence of cardiac arrhythmias in patients is disclosed. A computational system generates sample probability values ​​by applying a machine learning model to sample patient data. The machine learning model determines a corresponding probability value indicating the probability of the arrhythmia occurring during each corresponding time window. The computational system outputs a user interface including graphical data based on the sample probability values, and receives instructions via the user interface for selecting a probability threshold for a patient. The computational system receives patient data of the patient and applies the machine learning model to the patient data to determine a current probability value. In response to determining that the current probability exceeds the patient's probability threshold, the computational system generates an alert indicating that the patient may have experienced the occurrence of the arrhythmia.

Description

Technical Field

[0001] This disclosure generally relates to health monitoring, and more specifically, to monitoring heart health. Background Technology

[0002] Malignant tachyarrhythmias, such as ventricular fibrillation, are uncoordinated contractions of the myocardium in the ventricles of the heart and are the most common arrhythmias in patients with cardiac arrest. If such arrhythmia lasts longer than a few seconds, it can lead to cardiogenic shock and cessation of effective blood circulation. Therefore, sudden cardiac death (SCD) can occur within minutes.

[0003] In patients at high risk of ventricular fibrillation, the use of implantable medical devices (IMDs), such as implantable cardioverter defibrillators (ICDs), has been shown to be beneficial in preventing sudden cardiac death (SCD). An ICD is a battery-powered shock device that typically contains an electrical housing electrode (sometimes referred to as a can electrode) coupled to one or more electrical leads placed within the heart. If an arrhythmia is sensed, the ICD can deliver a pulse via the leads to shock the heart and restore its normal rhythm. Some ICDs have been configured to attempt to terminate a detected tachyarrhythmia by delivering antitachycardia pacing (ATP) before delivering a shock. Additionally, ICDs have been configured to deliver relatively high-amplitude post-shock pacing after the shock has successfully terminated the tachyarrhythmia to support the heart's recovery from the shock. Some ICDs also deliver bradycardia pacing, cardiac resynchronization therapy (CRT), or other forms of pacing.

[0004] Other types of medical devices can be used for diagnostic purposes. For example, implantable or non-implantable medical devices can monitor a patient's heart. Users, such as physicians, can view data generated by the medical device regarding the occurrence of arrhythmias such as atrial or ventricular tachyarrhythmias or cardiac arrest. Users can diagnose a patient's medical condition based on the identified arrhythmias. Summary of the Invention

[0005] In general, this disclosure describes a technique for monitoring the occurrence of cardiac arrhythmias in patients. A computing system generates sample probability values ​​by applying a machine learning model to sample patient data. In some instances, the computing system is a cloud computing system. The machine learning model determines a corresponding probability value indicating the probability of an arrhythmia occurring during each corresponding time window. The computing system outputs a user interface including graphical data based on the sample probability values, and receives instructions via the user interface for selecting a probability threshold for a patient. The computing system receives patient data for the patient and applies the machine learning model to the patient data to determine a current probability value. In response to determining that the current probability exceeds the patient's probability threshold, the computing system generates a notification indicating that the patient may have experienced the arrhythmia. In this way, a user can effectively configure the computing system regarding its sensitivity and specificity using the user interface.

[0006] In one aspect, this disclosure describes a method comprising: generating a set of sample probability values ​​by a computing system including a processing circuitry system and a storage medium by applying a machine learning model to a sample set of patient data, wherein: the machine learning model is trained using patient data from a plurality of patients, the sample set including a plurality of time windows; and for each respective time window of the plurality of time windows, the machine learning model is configured to determine a corresponding probability value in the sample probability value set that indicates the probability of an arrhythmia occurring during the respective time window; generating graphical data by the computing system based on the sample probability values; and outputting a user interface for display on a display device by the computing system, the user interface including... The graphical data; the computing system receiving, via the user interface, an instruction for selecting a probability threshold for a patient; the computing system receiving patient data of the patient, wherein the patient data is collected by one or more medical devices; the computing system applying the machine learning model to the patient data to determine a current probability value indicating the probability that the patient has experienced an arrhythmia; the computing system determining that the current probability value exceeds the patient's probability threshold; and in response to determining that the current probability value is greater than or equal to the patient's probability threshold, the computing system generating a notification indicating that the patient may have experienced the occurrence of the arrhythmia.

[0007] On the other hand, this disclosure describes a computing system comprising one or more processing circuits; and a storage medium storing instructions that, when executed, configure the one or more processing circuits to: generate a set of sample probability values ​​by applying a machine learning model to a sample set of patient data, wherein: the machine learning model is trained using patient data from multiple patients, the sample set comprising multiple time windows; and for each of the multiple time windows, the machine learning model is configured to determine a corresponding probability value in the sample probability value set that indicates the probability of an arrhythmia occurring during the corresponding time window; and generate a graph number based on the sample probability values. The system outputs a user interface for display on a display device, the user interface including the graphical data; receives, via the user interface, an instruction for selecting a probability threshold for a patient; receives patient data of the patient, wherein the patient data is collected by one or more medical devices; applies the machine learning model to the patient data to determine a current probability value indicating the probability that the patient has experienced an arrhythmia; determines that the current probability value exceeds the patient's probability threshold; and generates a notification indicating that the patient may have experienced the occurrence of the arrhythmia in response to determining that the current probability value is greater than or equal to the patient's probability threshold.

[0008] On the other hand, this disclosure describes a computer-readable medium having instructions stored thereon that, when executed, cause one or more processing circuits of a computing system to: generate a set of sample probability values ​​by applying a machine learning model to a sample set of patient data, wherein: the machine learning model is trained using patient data from a plurality of patients, the sample set comprising a plurality of time windows; and for each of the plurality of time windows, the machine learning model is configured to determine a corresponding probability value in the sample probability value set that indicates the probability of an arrhythmia occurring during the corresponding time window; generate graphical data based on the sample probability values; output a user interface for display on a display device, the user interface including the graphical data; receive, through the user interface, an instruction for selecting a probability threshold for a patient; receive patient data of the patient, wherein the patient data is collected by one or more medical devices; apply the machine learning model to the patient data to determine a current probability value indicating the probability that the patient has experienced an arrhythmia; determine that the current probability value exceeds the probability threshold for the patient; and generate a notification indicating that the patient may have experienced the occurrence of the arrhythmia in response to determining that the current probability value is greater than or equal to the probability threshold for the patient.

[0009] The present invention is intended to provide an overview of the subject matter described herein. It is not intended to provide an exclusive or exhaustive interpretation of the apparatuses and methods described in detail in the following drawings and description. Further details of one or more examples are set forth in the following drawings and description. Attached Figure Description

[0010] Figure 1 This is a block diagram illustrating a system for monitoring cardiac arrhythmias in patients according to the technology disclosed herein.

[0011] Figure 2 It shows in more detail Figure 1 A conceptual diagram of the implantable medical device (IMD) and leads for the system.

[0012] Figure 3 This is a block diagram of an example implantable medical device based on the technology disclosed herein.

[0013] Figure 4 This is a block diagram illustrating an example computing device operating according to one or more techniques disclosed herein.

[0014] Figure 5 This is a flowchart illustrating example operations based on the technology disclosed herein.

[0015] Figure 6 This is a flowchart illustrating a first example operation of generating graphical data and receiving instructions for selecting a probability threshold according to the technology disclosed herein.

[0016] Figure 7 This is a flowchart illustrating an example process for selecting models and operating points based on monitoring reasons, according to the technology disclosed herein.

[0017] Figure 8 This is a flowchart illustrating an example process for selecting a model and operating point based on the monitored arrhythmia, according to the technology disclosed herein.

[0018] Figure 9 This is a flowchart illustrating a second example operation of generating graphical data and receiving instructions for selecting a probability threshold according to the technology disclosed herein.

[0019] Figure 10 It is a conceptual diagram of the technology disclosed herein, which includes an example diagram of the original cardiac electrical waveform during a certain time period and an example diagram of the probability of the occurrence of arrhythmia during the same time period.

[0020] Throughout the accompanying drawings and description, similar reference numerals refer to similar elements. Detailed Implementation

[0021] Figure 1 This is a block diagram illustrating a system 10 for monitoring cardiac arrhythmias in a patient according to the technology of this disclosure. System 10 includes a medical device 16. An example of such a medical device is as follows: Figure 1 The implantable medical device (IMD) shown is illustrated. (As shown via...) Figure 1 As illustrated in Example System 10, in some instances, the medical device 16 may be, for example, an implantable cardiac monitor, an implantable pacemaker, an implantable cardioverter-defibrillator (ICD), or a pacemaker / cardioverter-defibrillator. In some instances, the medical device 16 is a non-implantable medical device, such as a non-implantable cardiac monitor (e.g., a Holt monitor).

[0022] exist Figure 1 In this example, medical device 16 is connected to leads 18, 20, and 22 and communicatively coupled to an external device 27, which in turn is communicatively coupled to a computing system 24 via a communication network 25. Medical device 16 senses electrical signals, such as electrocardiograms (EGMs), accompanying depolarization and repolarization of the heart 12 via electrodes on one or more leads 18, 20, and 22 or via the housing of medical device 16. Medical device 16 can also deliver therapy to the heart 12 in the form of electrical signals via electrodes positioned on one or more leads 18, 20, and 22 or via the housing of medical device 16. The therapy may be a pacing pulse, a cardioversion pulse, and / or a defibrillation pulse. Medical device 16 can monitor the EGM signals collected via electrodes on leads 18, 20, and 22 and diagnose and treat arrhythmias based on the EGM signals.

[0023] In some instances, the medical device 16 includes a communication circuitry 17, which includes components for communication with, for example, Figure 1 The communication circuit system 17 may include any suitable circuitry, firmware, software, or any combination thereof for communicating with another device, such as the external device 27. For example, the communication circuitry 17 may include one or more processors, memories, radios, antennas, transmitters, receivers, modulation and demodulation circuitry, filters, amplifiers, etc., for radio frequency communication with other devices such as the computing system 24. The medical device 16 may use the communication circuitry 17 to receive downlink data to control one or more operations of the medical device 16 and / or to transmit uplink data to the external device 27.

[0024] Leads 18, 20, and 22 extend into the heart 12 of the patient 14 to sense the electrical activity of the heart 12 and / or deliver electrical stimulation to the heart 12. Figure 1In the example shown, the right ventricle (RV) lead 18 extends through one or more veins (not shown), the superior vena cava (not shown), and the right atrium 26, and enters the right ventricle 28. The left ventricle (LV) lead 20 extends through one or more veins, the vena cava, and the right atrium 26, and enters the coronary sinus 30, reaching the region adjacent to the free wall of the left ventricle 32 of the heart 12. The right atrium (RA) lead 22 extends through one or more veins and the vena cava, and enters the right atrium 26 of the heart 12.

[0025] Although Figure 1 Example system 10 depicts medical device 16, but in other instances, the technology of this disclosure can be applied to other types of medical devices that are not necessarily implantable. For example, a medical device according to the technology of this disclosure may comprise a wearable medical device or “smart” clothing worn by patient 14. For example, such a medical device may take the form of a watch worn by patient 14 or a circuit system adhesively attached to patient 14, such as Seeq, which is commercially available from Medtronic plc of Dublin, Ireland. TM Mobile cardiac telemetry system. In another example, a medical device as described herein may comprise an external medical device with implantable electrodes.

[0026] In some instances, external device 27 takes the form of an external programmer or mobile device such as a mobile phone, smartphone, laptop computer, tablet computer, or personal digital assistant (PDA). In some instances, external device 27 is a CareLink device commercially available from Medtronic. TM A monitor. Users such as physicians, technicians, surgeons, electrophysiologists, or other clinicians can interact with external device 27 to retrieve physiological or diagnostic information from medical device 16. Users such as patients 14 or clinicians, as described above, can also interact with external device 27 to program medical device 16, for example, by selecting or adjusting the values ​​of operating parameters of medical device 16. External device 27 may include processing circuitry, memory, user interface, and communication circuitry capable of transmitting and receiving information to and from medical device 16 and computing system 24.

[0027] In some instances, the computing system 24 takes the form of a handheld computing device, computer workstation, server or other networked computing device, smartphone, tablet computer, or external programmer, and includes a user interface for presenting information to the user and receiving input from the user. In some instances, the computing system 24 may include one or more devices implementing a machine learning system, such as a neural network, deep learning system, or another type of machine learning system. Users, such as physicians, technicians, surgeons, electrophysiologists, or other clinicians, can interact with the computing system 24 to retrieve physiological or diagnostic information from the medical device 16. Users can also interact with the computing system 24 to program the medical device 16, for example, by selecting values ​​for operating parameters of the IMD. The computing system 24 may include a processor configured to evaluate the EGM and / or other sensed signals transmitted from the medical device 16 to the computing system 24.

[0028] Network 25 may include one or more computing devices (not shown), such as one or more non-edge switches, routers, hubs, gateways, security devices such as firewalls, intrusion detection and / or intrusion prevention devices, servers, computer terminals, laptops, printers, databases, wireless mobile devices such as cellular phones or personal digital assistants, wireless access points, bridges, cable modems, application accelerators, or other network devices. Network 25 may include one or more networks managed by a service provider and may thus form part of a large public network infrastructure such as the Internet. Network 25 may provide Internet-accessible computing devices such as computing system 24 and medical device 16, and may provide a communication framework that allows the computing devices to communicate with each other. In some instances, network 25 may be a private network providing a communication framework that allows computing system 24, medical device 16, and EMR database 66 to communicate with each other, but isolates computing system 24, medical device 16, and EMR database 66 from external devices for security purposes. In some instances, communication between computing system 24, medical device 16, and EMR database 66 is encrypted.

[0029] External device 27 and computing system 24 can communicate via network 25 using any technology known in the art, via wired and / or wireless communication. In some instances, computing system 24 is a remote device that communicates with external device 27 through an intermediary device located in network 25, such as a local access point, wireless router, or gateway. Although in Figure 1 In some instances, external device 27 and computing system 24 communicate via network 25; however, in others, external device 27 and computing system 24 communicate directly with each other. Examples of communication technologies may include, for example, those based on… Communication may be via the BLE protocol. Other communication technologies are also considered. The computing system 24 can also communicate with one or more other external devices using a variety of known wired and wireless communication technologies. In some instances, the computing system 24 and network 25 may be implemented and provide access to Carelink, managed by Medtronic. TM System access.

[0030] EMR database 66 stores the EMR data of patient 14. EMR database 66 may include processing circuitry and one or more storage media, such as random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electronically erasable programmable read-only memory (EEPROM), or flash memory. In some instances, EMR database 66 is a cloud computing system. In some instances, the functionality of EMR database 66 is distributed across multiple computing systems.

[0031] In one instance, computing system 24 receives patient data from patient 14 collected by medical device 16. In some instances, the patient data includes physiological data of patient 14, such as one or more of the following: patient 14's activity level, patient 14's heart rate, patient 14's posture, patient 14's electrocardiogram, patient 14's blood pressure, patient 14's pulse transit time, patient 14's respiratory rate, patient 14's hypoventilation index or apnea, patient 14's accelerometer data, features derived from patient 14's accelerometer data such as activity counts, posture, statistical control process variables, patient 14's raw electromyography, one or more features derived from patient 14's raw electromyography such as heart rate variability, T-wave alternation, QRS morphology, patient 14's interval data and features derived from the interval data, heart sounds, potassium levels, blood glucose levels, temperature, or any data that can be derived from the parameter data mentioned above or any other type of patient parameter data. In some instances, medical device 16 and / or other devices can automatically generate patient parameter data by processing information from one or more sensors. For example, sensors in medical device 16 and / or other devices can determine that patient 14 has fallen, that patient 14 is weak or ill, or that patient 14 is experiencing sleep apnea.

[0032] In some instances, the patient data includes environmental data such as air quality measurements, ozone levels, particulate counts or pollution levels near patient 14, ambient temperature, or daytime. In some instances, one of the medical device or external device 27 may sense the environmental data via one or more sensors. In another instance, the environmental data is received by external device 27 via an application such as a weather app running on external device 27 and uploaded to computing system 24 via network 25. In yet another instance, computing system 24 collects the environmental data directly from a cloud service containing location-based data about patient 14.

[0033] In some instances, the patient data includes patient symptom data uploaded by patient 14 via an external device such as external device 27. For example, patient 14 can upload patient symptom data via an application running on a smartphone. In some instances, patient 14 can upload patient symptom data via a user interface (...). Figure 1 (Not described in the text) Such as uploading patient symptom data via touch screen, keyboard, graphical user interface, voice command, etc.

[0034] In some instances, the patient data includes device-related data such as one or more of the following: impedance of one or more electrodes of the medical device, electrode selection, medication delivery schedule of the medical device, history of electrical pacing therapy delivered to the patient, or diagnostic data of the medical device. In some instances, the medical device that collects the patient data is an IMD (Integrated Device Management) of the medical device. In other instances, the medical device that collects the patient data is another type of patient device, such as a wearable medical device or mobile device (e.g., a smartphone) for patient 14. In some instances, the computing system 24 receives patient data periodically, for example daily.

[0035] In some instances, computing system 24 further receives EMR data of patient 14 from EMR database 66. This EMR data can be considered another form of patient data. In some instances, the EMR data stored by EMR database 66 may contain many different types of historical medical information about patient 14. For example, EMR database 66 may store the patient's medication history, surgical procedure history, hospitalization history, potassium levels over time, one or more laboratory test results of patient 14, cardiovascular history of patient 14, or comorbidities of patient 14, such as atrial fibrillation, heart failure, or diabetes.

[0036] The computing system 24 applies one or more machine learning models trained using patient data from multiple patients to the patient data of patient 14 to monitor for the occurrence of arrhythmias in patient 14. For example, the computing system 24 may receive patient data from patient 14. The patient data is collected by one or more medical devices, such as medical device 16. Furthermore, in this example, the computing system 24 may apply a machine learning model to the patient data to determine a current probability value indicating the likelihood that patient 14 has experienced an arrhythmia. In this example, the computing system 24 may further determine whether the current probability value exceeds a probability threshold for patient 14. In response to determining that the current probability value is greater than or equal to the probability threshold for patient 14, the computing system 24 may generate a notification indicating that patient 14 may have experienced an arrhythmia. The probability value may be a value between 0 and 1; or a value within another range, where the value indicates probability or likelihood. The probability threshold may also be within such ranges.

[0037] Setting the probability threshold can determine the review burden and diagnostic positivity rate associated with notifications generated by the computing system 24. Users reviewing notifications generated by the computing system 24 may experience a review burden associated with the notifications, as they may need to spend time reviewing them and determining whether patient 14 actually experienced the arrhythmia indicated by the notification. The diagnostic positivity rate associated with the notification can correspond to the percentage of notifications generated by the computing system 24 that actually produce valuable diagnostic information, or other metrics. If the review burden is too high, it may be impractical for users to use the computing system 24 to monitor patient 14. Similarly, if the diagnostic positivity rate is too low, users may not find it useful to use the computing system 24 to monitor patient 14.

[0038] Therefore, according to the technology of this disclosure, the computing system 24 allows users to adjust probability thresholds in a manner that can alter the review burden and diagnostic positivity rate associated with notifications generated by the computing system 24. Thus, in some instances of this disclosure, the computing system 24 can generate a set of sample probability values ​​by applying a machine learning model to a sample set of patient data. The machine learning model may be trained using patient data from multiple patients. Furthermore, the sample set may include multiple time windows. For each of the multiple time windows, the machine learning model is configured to determine a corresponding probability value in the sample probability value set that indicates the probability of an arrhythmia occurring during the corresponding time window. In some instances, each of the time windows in the sample set includes a corresponding cardiac EGM strip.

[0039] The computing system 24 can generate graphical data based on the sample probability values ​​and can output a user interface for display on a display device. The user interface may include the graphical data. As described in this disclosure, the graphical data may include one or more receiver-operator curves (ROCs), graphs, and / or other types of graphical data. The computing system 24 can receive user input instructions for selecting a probability threshold for patient 14 through the user interface. For example, the computing system 24 can receive instructions for clicking, typing, tapping, swiping, dragging, pinching, or other types of user input for one or more features of the user interface. The computing system 24 can receive patient data for patient 14 and apply the machine learning model to the patient data to determine a current probability value indicating the probability that patient 14 has experienced an arrhythmia. The computing system 24 can then determine whether the current probability value exceeds the probability threshold set for patient 14.

[0040] In this example, in response to determining that the current probability value is greater than or equal to the probability threshold set for patient 14, computing system 24 can generate a notification indicating that patient 14 may have experienced the occurrence of the arrhythmia. Computing system 24 can store a copy of the generated notification. By allowing the user to appropriately set the probability threshold, computing device 24 can avoid generating and storing unnecessary notifications. This can save storage space on the computer-readable medium of computing system 24. Furthermore, in some instances, computing system 24 can transmit notifications to devices used by one or more users monitoring patient 14. By avoiding the generation and transmission of multiple notifications from users that are unsuitable for diagnostic purposes, network bandwidth can be saved, and battery life of the receiving device can be conserved.

[0041] Figure 2 It shows in more detail Figure 1 A conceptual diagram of system 10, medical device 16, and leads 18, 20, and 22. In the illustrated example, bipolar electrodes 40 and 42 are located adjacent to the distal end of lead 18, and bipolar electrodes 48 and 50 are located adjacent to the distal end of lead 22. Additionally, four electrodes 44, 45, 46, and 47 are located adjacent to the distal end of lead 20. Lead 20 may be referred to as a quadrupole LV lead. In other examples, lead 20 may contain more or fewer electrodes. In some examples, LV lead 20 includes segmented electrodes, for example, each of the multiple longitudinal electrode locations of the lead, such as the locations of electrodes 44, 45, 46, and 47, comprises multiple discrete electrodes arranged circumferentially around the circumference of the lead at the respective circumferential location.

[0042] In the illustrated examples, electrodes 40 and 44-48 are in the form of ring electrodes, and electrodes 42 and 50 may be in the form of extendable spiral-tipped electrodes retractably mounted within insulated electrode heads 52 and 56, respectively. Leads 18 and 22 also include elongated electrodes 62 and 64, which may be in the form of coils. In some examples, electrodes 40, 42, 44-48, 50, 62, and 64 are electrically coupled to corresponding conductors within the lead bodies of their associated leads 18, 20, and 22, and thereby coupled to the circuitry within the medical device 16.

[0043] In some instances, the medical device 16 includes one or more housing electrodes, such as Figure 2 The housing electrode 4 shown may be integrally formed with the outer surface of the hermetically sealed housing 8 of the medical device 16, or may be otherwise coupled to the housing 8. In some instances, the housing electrode 4 is defined by the uninsulated portion of the externally facing part of the housing 8 of the medical device 16. Other divisions between the insulating and uninsulated portions of the housing 8 may be used to define two or more housing electrodes. In some instances, the housing electrode comprises substantially all of the housing 8.

[0044] The housing 8 encloses a signal generation circuit system for generating therapeutic stimuli such as cardiac pacing pulses, cardioversion pulses, and defibrillation pulses, as well as a sensing circuit system for sensing electrical signals accompanying the depolarization and repolarization of the heart 12. The housing 8 may also enclose a memory for storing the sensed electrical signals. The housing 8 may also enclose a communication circuit system 17 for communication between the medical device 16 and the computing system 24.

[0045] The medical device 16 senses electrical signals associated with the depolarization and repolarization of the heart 12 via electrodes 4, 40, 42, 44-48, 50, 62, and 64. The medical device 16 can sense such electrical signals using any bipolar combination of electrodes 40, 42, 44-48, 50, 62, and 64. Furthermore, any of electrodes 40, 42, 44-48, 50, 62, and 64 can be combined with housing electrode 4 for unipolar sensing.

[0046] The number and configuration of leads 18, 20, and 22 and electrodes shown are merely examples. Other configurations, i.e., the number and positioning of leads and electrodes, are also possible. In some instances, system 10 may include additional leads or lead segments having one or more electrodes positioned at different locations in the cardiovascular system for sensing and / or delivering therapy to patient 14. For example, as an alternative to or supplement to intracardiac leads 18, 20, and 22, system 10 may include one or more epicardial or extravascular (e.g., subcutaneous or substernal) leads not positioned within the heart 12.

[0047] Medical device 16 transmits patient data to computing system 24 (e.g., via external device 27). The patient data may include data based on electrical signals detected by electrodes 4, 40, 42, 44-48, 50, 62, and / or 64. For example, medical device 16 may collect cardiac EGM and other data and transmit them to computing system 24. According to the techniques of this disclosure, computing system 24 can use the patient data to determine probability values ​​indicating the likelihood that patient 14 has experienced one or more arrhythmias.

[0048] Although described herein in the context of an example medical device 16 providing therapeutic electrical stimulation, the techniques disclosed herein can be used with other types of devices. For example, the techniques can be implemented with devices such as: an external cardiac defibrillator with electrodes coupled to the outside of the cardiovascular system; or a transcatheter pacemaker configured for implantation in the heart, such as the Micra, commercially available from Medtronic in Ireland. TM Transcatheter pacing systems; insertable cardiac monitors, such as the Reveal LINQ, which is also commercially available from Medtronic. TM ICM; neurostimulators; drug delivery devices; wearable devices such as wearable cardioverter defibrillators, fitness trackers or other wearable devices; mobile devices such as mobile phones, “smart” phones, laptops, tablets, personal digital assistants (PDAs) or “smart” clothing such as “smart” glasses or “smart” watches.

[0049] Figure 3 This is a block diagram of an example medical device 16 according to the technology of this disclosure. In the illustrated example, the medical device 16 includes a processing circuitry system 58, a memory 59, a communication circuitry system 17, a sensing circuitry system 50, a therapy delivery circuitry system 52, a sensor 57, and a power supply 54. The memory 59 contains computer-readable instructions that, when executed by the processing circuitry system 58, cause the medical device 16 and the processing circuitry system 58 to perform various functions attributed herein to the medical device 16 and the processing circuitry system 58 (e.g., performing short-term prediction of arrhythmias, delivering therapies such as antitachycardia pacing therapy, bradycardia pacing therapy, and post-shock pacing therapy). The memory 59 may contain any volatile, non-volatile, magnetic, optical, or electrical medium, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), electrically erasable programmable ROM (EEPROM), flash memory, or any other digital or analog medium.

[0050] The processing circuit system 58 may include any one or more of the following: a microprocessor, a controller, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or an equivalent discrete or analog logic circuit system. In some instances, the processing circuit system 58 may include multiple components such as any combination of one or more microprocessors, one or more controllers, one or more DSPs, one or more ASICs, or one or more FPGAs, as well as other discrete or integrated logic circuit systems. The functionality attributed to the processing circuit system 58 herein may be embodied in software, firmware, hardware, or any combination thereof.

[0051] The processing circuit system 58 can control the therapy delivery circuit system 52 to deliver stimulation therapy to the heart 5 according to therapy parameters that can be stored in the memory 59. For example, the processing circuit system 58 can control the therapy delivery circuit system 52 to deliver electrical pulses with amplitude, pulse width, frequency, or electrode polarity specified by the therapy parameters. In this way, the therapy delivery circuit system 52 can deliver pacing pulses (e.g., ATP pulses, bradycardia pacing pulses, or post-shock pacing therapy) to the heart 5 via electrodes 34 and 40. In some instances, the therapy delivery circuit system 52 can deliver pacing stimulation in the form of voltage or current pulses, such as ATP therapy, bradycardia treatment, or post-shock pacing therapy. In other instances, the therapy delivery circuit system 52 can deliver one or more of these types of stimulation in the form of other signals such as sine waves, square waves, and / or other substantially continuous time signals.

[0052] Therapy delivery circuitry 52 is electrically coupled to electrodes 34 and 40 carried on the housing of medical device 16. While medical device 16 may contain only two electrodes, such as electrodes 34 and 40, in other instances, medical device 16 may use three or more electrodes. Medical device 16 may use any combination of electrodes to deliver therapy and / or detect electrical signals from patient 12. In some instances, therapy delivery circuitry 52 includes charging circuitry, one or more pulse generators, capacitors, transformers, switching modules, and / or other components capable of generating and / or storing energy for delivery as pacing therapy, cardiac resynchronization therapy, other therapies, or combinations thereof. In some instances, therapy delivery circuitry 52 delivers therapy in the form of one or more electrical pulses based on one or more sets of therapy parameters defining amplitude, frequency, voltage or current of the therapy, or other parameters of the therapy.

[0053] The sensing circuit system 50 monitors signals from electrodes 4, 40, 42, 44-48, 50, and 62. Figure 2 ) and 64 ( Figure 2The sensing circuitry 50 uses signals from one or more combinations of two or more electrodes to monitor the electrical activity, impedance, or other electrical phenomena of the heart 12. In some instances, the sensing circuitry 50 includes one or more analog components, digital components, or combinations thereof. In some instances, the sensing circuitry 50 includes one or more sensing amplifiers, comparators, filters, rectifiers, threshold detectors, analog-to-digital converters (ADCs), etc. In some instances, the sensing circuitry 50 converts the sensed signals into digital form and provides the digital signals to the processing circuitry 58 for processing or analysis. In one instance, the sensing circuitry 50 amplifies signals from electrodes 4, 40, 42, 44-48, 50, 62, and 64, and converts the amplified signals into multi-bit digital signals using an ADC.

[0054] In some instances, the sensing circuitry system 50 performs sensing of the cardiac electrogram to determine heart rate or heart rate variability, or to detect arrhythmias (e.g., tachyarrhythmias or bradycardia), or to sense other parameters or events from the cardiac electrogram. The sensing circuitry system 50 may also include a switching circuitry system for selecting which of the available electrodes (and electrode polarities) are used to sense cardiac activity, depending on which electrode combination or electrode vector is used in the current sensing configuration. The processing circuitry system 58 can control the switching circuitry system to select the electrodes that act as sensing electrodes and their polarities. The sensing circuitry system 50 may include one or more detection channels, each of which can be coupled to a selected electrode configuration to detect cardiac signals through that configuration. In some instances, the sensing circuitry system 50 compares the processed signal to a threshold to detect the presence of atrial or ventricular depolarization and indicates to the processing circuitry system 58 the presence of atrial depolarization (e.g., P wave) or ventricular depolarization (e.g., R wave). The sensing circuit system 50 may include one or more amplifiers or other circuit systems for comparing the amplitude of the cardiac electrogram with a threshold that may be adjustable.

[0055] The processing circuitry 58 may include a timing and control module, which may be embodied in hardware, firmware, software, or any combination thereof. The timing and control module may include dedicated hardware circuitry, such as an ASIC, separate from other components of the processing circuitry 58, such as a microprocessor, or a software module executed by components of the processing circuitry 58 that may be a microprocessor or an ASIC. The timing and control module may implement a programmable counter. If the medical device 16 is configured to generate and deliver bradycardia pacing pulses to the heart 12, such a counter may control the basic time interval associated with DDD, VVI, DVI, VDD, AAI, DDI, DDDR, VVIR, DVIR, VDDR, AAIR, DDIR, and other pacing modes.

[0056] The memory 59 can be configured to store various operating parameters, therapeutic parameters, sensed and detected data, and any other information related to the therapy and treatment of the patient 12. Figure 3 In one example, memory 58 may store, for example, sensed cardiac EGM associated with a detected or predicted arrhythmia, and therapy parameters defining the delivery of the therapy provided by therapy delivery circuitry system 52. In other examples, memory 58 may be used as a temporary buffer to store data until the data can be uploaded to computing system 24.

[0057] Communication circuit system 17 includes a means for transmitting signals via... Figure 1 The network 25 communicates with another device, such as the computing system 24, using any suitable circuitry, firmware, software, or any combination thereof. For example, the communication circuitry 17 may include one or more antennas, modulation and demodulation circuitry, filters, amplifiers, etc., for radio frequency communication with other devices, such as the computing system 24, via the network 25. Under the control of the processing circuitry 58, the communication circuitry 17 can receive downlink telemetry from the computing system 24 and send uplink telemetry to the computing system via an antenna (which may be internal and / or external). The processing circuitry 58 may, for example, provide data to be uplinked to the computing system 24 and control signals for the telemetry circuitry within the communication circuitry 17 via an address / data bus. In some instances, the communication circuitry 17 may provide the received data to the processing circuitry 58 via a multiplexer.

[0058] The power source 54 can be any type of device configured to maintain charge to operate the circuitry of the medical device 16. The power source 54 can be provided as a rechargeable or non-rechargeable battery. In other instances, the power source 54 can be incorporated into an energy extraction system that stores electrical energy generated by the movement of the medical device 16 within the patient 12.

[0059] According to the technology disclosed herein, medical device 16 collects patient data of patient 14 via sensing circuitry system 50 and / or sensors 57. Sensors 57 may include one or more sensors, such as one or more accelerometers, pressure sensors, optical sensors for O2 saturation, etc. In some instances, the patient data includes one or more of the following: patient activity level, patient heart rate, patient posture, patient electrocardiogram, patient blood pressure, patient accelerometer data, or other types of patient parameter data. Medical device 16 uploads the patient parameter data to computing system 24 via network 25 through communication circuitry system 17. In some instances, medical device 16 uploads patient parameter data to computing system 24 daily. In some instances, patient parameter data includes one or more values ​​representing average measurements of patient 14 over a long period of time (e.g., approximately 24 hours to approximately 48 hours). For example, one or more other devices, such as patient 14's wearable medical device or mobile device (e.g., smartphone), may collect patient parameter data and upload it to computing system 24.

[0060] Although described herein in the context of an example medical device 16 providing therapeutic electrical stimulation, the techniques disclosed herein for short-term prediction of arrhythmias can be used with other types of devices. For example, the techniques can be implemented with devices configured for implantation in the heart, such as the Micra device, commercially available from Medtronic in Ireland. TM Transcatheter pacing systems; insertable cardiac monitors, such as the Reveal LINQ, which is also commercially available from Medtronic. TM ICM; neurostimulators; drug delivery devices; wearable devices such as wearable cardioverter defibrillators, fitness trackers or other wearable devices; mobile devices such as mobile phones, “smart” phones, laptops, tablets, personal digital assistants (PDAs) or “smart” clothing such as “smart” glasses or “smart” watches.

[0061] Figure 4 This is a block diagram illustrating an example computing system 24 operating according to one or more techniques of this disclosure. In one example, the computing system 24 includes a processing circuitry 402 for executing an application 424 including a monitoring system 450 or any other application described herein. Although for illustrative purposes... Figure 4 The diagram shows a standalone computing system 24, but computing system 24 can be any component or system that includes a processing circuitry system or other suitable computing environment for executing software instructions, and does not necessarily need to include, for example, a processing circuitry system or other suitable computing environment for executing software instructions. Figure 4One or more components are shown (e.g., communication circuitry 406; and in some instances, components such as storage device 408 may be located in a different location or in the same rack than other components). In some instances, computing system 24 may be a cloud computing system distributed across multiple devices.

[0062] like Figure 4 As illustrated in the example, computing system 24 includes processing circuitry 402, one or more input devices 404, communication circuitry 406, one or more output devices 412, one or more storage devices 408, and user interface (UI) devices 410. In one example, computing system 24 further includes one or more application programs 424 executable by computing system 24, such as machine learning model 450 and operating system 416. Each of components 402, 404, 406, 408, 410, and 412 is coupled (physically, communicatively, and / or operatively) for inter-component communication. In some examples, communication channel 414 may include a system bus, network connection, inter-process communication data structure, or any other method for communicating data. As an example, components 402, 404, 406, 408, 410, and 412 may be coupled via one or more communication channels 414.

[0063] In one instance, the processing circuitry 402 is configured to perform functions and / or process instructions for execution within the computing system 24. For example, the processing circuitry 402 may be able to process instructions stored in the storage device 408. Instances of the processing circuitry 402 may include any one or more of the following: a microprocessor, a controller, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or an equivalent discrete or integrated logic circuitry system.

[0064] One or more storage devices 408 may be configured to store information within the computing system 24 during operation. In some instances, storage device 408 is described as a computer-readable storage medium. In some instances, storage device 408 is temporary memory, meaning that the primary purpose of storage device 408 is not long-term storage. In some instances, storage device 408 is described as volatile memory, meaning that storage device 408 does not maintain its stored contents when the computer is turned off. Examples of volatile memory include random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), and other forms of volatile memory known in the art. In some instances, storage device 408 is used to store program instructions executed by the processing circuitry system 402. In one instance, storage device 408 is used by software or application 424 running on the computing system 24 to temporarily store information during program execution.

[0065] In some instances, storage device 408 also includes one or more computer-readable storage media. Storage device 408 can be configured to store a larger amount of information than volatile memory. Storage device 408 can also be configured for long-term storage of information. In some instances, storage device 408 includes non-volatile storage elements. Examples of such non-volatile storage elements include magnetic hard disks, optical disks, floppy disks, flash memory, or various forms of electrically programmable memory (EPROM) or electrically erasable and programmable (EEPROM) memory.

[0066] In some instances, the computing system 24 also includes a communication circuit system 406. In one instance, the computing system 24 utilizes the communication circuit system 406 to communicate with, for example, Figure 1 The communication circuit system 406 communicates with external devices such as the IMD 17 and EMR database 66. The communication circuit system 406 may include a network interface card, such as an Ethernet card, optical transceiver, radio frequency transceiver, or any other type of device capable of sending and receiving information. Other examples of such network interfaces may include 3G, 4G, 5G, and Wi-Fi™ radios.

[0067] In one instance, the computing system 24 further includes one or more user interface devices 410. In some instances, the user interface device 410 is configured to receive input from a user via haptic, audio, or video feedback. Examples of user interface devices 410 include presence-sensitive displays, mice, keyboards, voice response systems, cameras, microphones, or any other type of device for detecting commands from the user. In some instances, presence-sensitive displays include touchscreens.

[0068] One or more output devices 412 may also be included in the computing system 24. In some instances, the output device 412 is configured to provide output to a user using tactile, audio, or video stimuli. In one instance, the output device 412 includes an presence-sensitive display, sound card, video graphics adapter card, or any other type of device for converting signals into a suitable form that is understandable to humans or machines. In some instances, the output device 412 includes a display device. Additional examples of the output device 412 include a speaker, a cathode ray tube (CRT) monitor, a liquid crystal display (LCD), or any other type of device capable of generating understandable output to a user.

[0069] The computing system 24 may include an operating system 416. In some instances, the operating system 416 controls the operation of the components of the computing system 24. For example, in one instance, the operating system 416 facilitates communication between one or more applications 424 and the monitoring system 450 and the processing circuitry 402, the communication circuitry 406, the storage device 408, the input device 404, the user interface device 410, and the output device 412.

[0070] Application 424 may also contain program instructions and / or data executable by computing system 24. Example application 424 executable by computing system 24 may include monitoring system 450. Other additional applications, not shown, may be included to provide other functionalities described herein, and are not depicted for simplicity.

[0071] According to the technology disclosed herein, application 424 includes monitoring system 450. Monitoring system 450 is configured to receive patient data, evaluate the patient data, and identify patient 14 (…) within monitoring system 450. Figure 1 A notification may be generated when one or more arrhythmias have occurred.

[0072] like Figure 4 As shown in the examples, the monitoring system 450 may contain a set of machine learning models 452. In summary, the machine learning models 452 may be referred to as a machine learning model library or a machine learning model suite. Each machine learning model in the machine learning models 452 may be configured to generate a probability value indicating the probability that patient 14 has experienced one or more arrhythmias based on patient data provided to the machine learning model. In some instances, each machine learning model in the machine learning models 452 is implemented using one or more neural network systems, deep learning systems, or other types of supervised or unsupervised machine learning systems. For example, the machine learning models may be implemented using feedforward neural networks such as convolutional neural networks, radial basis function neural networks, recurrent neural networks, modular or associative neural networks. In some instances, the monitoring system 450 trains the machine learning models 452 with patient data from multiple patients to generate probability values ​​for arrhythmias. In some instances, after the machine learning models have been pre-trained with patient data from multiple patients (and in some instances, EMR data), the monitoring system 450 may further train the machine learning models with patient data specific to patient 14.

[0073] In some instances, monitoring system 450 trains one or more machine learning models in machine learning model 452 with patient data from multiple patients, determines the error rate of the machine learning model, and then feeds the error rate back to the machine learning model so that the machine learning model updates its predictions based on the error rate. Monitoring system 450 may use a backpropagation algorithm, such as gradient descent, to feed the error rate back to the machine learning model. The error rate may correspond to the difference between a probability value determined by the machine learning model based on the input data and a pre-labeled probability value for the same input data. In some instances, monitoring system 450 may use an error function to determine the error rate. The error function may be implemented using signal processing techniques and trial and error in a manner conventionally used for detecting the occurrence of arrhythmias. In some instances, the error function may return an element vector, each element indicating whether the machine learning model correctly identified the occurrence of the corresponding arrhythmia.

[0074] In some instances, monitoring system 450 may receive feedback from patient 14 or a clinician indicating whether a predicted arrhythmia occurred in patient 14 within a specific time period. In some instances, monitoring system 450 may receive a message from medical device 16 indicating that medical device 16 has detected (or has not yet detected) the occurrence of an arrhythmia in patient 14. In some instances, monitoring system 450 may obtain feedback in other ways, such as by periodically checking EMR data to determine if an arrhythmia has occurred. Monitoring system 450 may use the feedback to update the machine learning model. Thus, the training process can occur iteratively to incrementally improve the data generated by the machine learning model by “learning” from both correct and incorrect data generated by the machine learning model in the past. Additionally, the training process can be used to further fine-tune the machine learning model trained with population-based data to provide more accurate predictions for specific individuals. In some instances, feedback may be provided by personnel monitoring the service.

[0075] According to the techniques disclosed herein, different machine learning models in machine learning model 452 can correspond to different monitoring causes. In some instances, different machine learning models in machine learning model 452 correspond to the 10th Revision of the International Classification of Diseases and Related Health Problems (ICD-10). thDifferent ICD-10 codes correspond to different codes defined in the revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10). Different ICD-10 codes may correspond to different reasons for prescribing a medical device 16 to a patient 14. For example, different ICD-10 codes may exist for atrial fibrillation (AF) management, stroke, syncope, and other medical conditions. In some instances, the monitoring system 450 may use training data generated for different ICD-10 codes to train one or more machine learning models in the machine learning model 452.

[0076] Monitoring system 450 can be used for different use cases where the same model may not provide optimal arrhythmia detection performance. Therefore, a suite of machine learning models can be generated and applied to identify the occurrence of arrhythmias of interest. For example, in one instance, machine learning model 452 may include a machine learning model to detect any occurrence of arrhythmias such as AF or sinus bradycardia or cardiac arrest (e.g., pause). In this instance, the user may not be concerned about other arrhythmias (e.g., normal sinus rhythm (NSR), atrial premature beats (PAC), ventricular premature beats (PVC), sinus tachycardia) in patient 14. In this instance, a general practitioner could use this machine learning model to look for “significant” arrhythmias. In another instance, monitoring system 450 can use a machine learning model configured to determine probability values ​​indicating the probability that patient 14 has experienced any arrhythmia belonging to the sinus bradycardia or AV block type. In this instance, the user may not be concerned about other arrhythmias in patient 14. In this instance, such a machine learning model could be used for monitoring after transcatheter aortic valve replacement (TAVR).

[0077] The input data used by machine learning model 452 may include patient data. The patient data may include data representing one or more electrical signals, such as an EGM signal. In some instances, the patient data may include data about the patient's physiological state (e.g., activity, posture, respiration, etc.), which may also be captured by medical device 16. Training data corresponding to different physiological conditions (e.g., rest, nighttime rest, nighttime high-posture angle rest, etc.) may be used as additional parameters for model training or as input data for machine learning model 452. Using such data can enable monitoring system 450 to detect arrhythmias during other disease conditions (e.g., a sensitive model for resting tachycardia can be used to monitor patients with heart failure (HF); a model for active bradycardia can be used to monitor patients with chronotropic dysfunction).

[0078] Once one or more of the machine learning models in machine learning model 452 have been trained, monitoring system 450 can use machine learning model 452 to detect the occurrence of arrhythmias experienced by patient 14. For example, monitoring system 450 (which may be executed by processing circuitry system 402) may receive patient data via communication circuitry system 406. The patient data may be collected, in whole or in part, by medical device 16 of patient 14. In some instances, the patient data includes one or more of the following: EGM data, patient activity level, patient heart rate, patient posture, patient electrocardiogram, patient blood pressure, patient accelerometer data, EMR data from EMR database 66, and / or other types of patient data. In some instances, medical device 16 is an IMD. In other instances, medical device 16 is another type of patient device, such as a wearable medical device or mobile device (e.g., a smartphone) of patient 14. In some instances, monitoring system 450 receives patient data from medical device 16 daily.

[0079] In some instances, monitoring system 450 receives EMR data of patient 14 from EMR database 66 via communication circuitry system 406. In some instances, the EMR data stored by EMR database 66 may contain many different types of historical medical information about patient 14. For example, EMR database 66 may store the patient's medication history, surgical procedure history, hospitalization history, potassium levels over time, or one or more laboratory test results. The EMR data may form part of the patient data used as input to one or more machine learning models in machine learning model 452.

[0080] In some instances, each machine learning model in machine learning model 452 transforms patient data into one or more vectors and tensors (e.g., multidimensional arrays) representing the patient data. The machine learning model may apply mathematical operations to the one or more vectors and tensors to generate a mathematical representation of the patient data. The machine learning model may determine different weights corresponding to the identifying relationship between the patient data and the occurrence of arrhythmias. The machine learning model may apply different weights to the patient data to generate probability values.

[0081] Figure 5 This is a flowchart illustrating example operations according to the technology disclosed herein. For convenience, regarding... Figure 1 Described Figure 5 The flowcharts in this disclosure are presented as examples. Other examples of the technology according to this disclosure may include more, fewer, or different actions, or the actions may be performed in different orders or in parallel.

[0082] exist Figure 5 In one instance, the computing system 24 can generate a set of sample probability values ​​(500) by applying a machine learning model to a sample set of patient data. The machine learning model can be machine learning model 452. Figure 4 One of them. As discussed above, the machine learning model can be trained using patient data from a sample of patients. The sample of patients can include a single patient or multiple patients. The sample of patients may or may not include the current patient who will be monitored for one or more arrhythmias.

[0083] The sample set comprises multiple time windows. In some instances, the time windows overlap. In other instances, the time windows do not overlap. Each time window of the sample set may include one or more sample series of at least one cardiac electrical waveform from patients in the set of sample patients. In instances where the time windows overlap, the same sample may fall within two or more time windows.

[0084] For each of the plurality of time windows, the machine learning model is configured to determine a corresponding probability value in the set of sample probability values ​​that indicates the probability of an arrhythmia occurring during the corresponding time window. As discussed elsewhere in this disclosure, the machine learning model may include a neural network trained to generate probability values ​​indicating the probability that a patient has experienced one or more arrhythmias. The input to the neural network may contain data corresponding to the corresponding time window.

[0085] In addition, Figure 5 In one example, the computing system 24 can generate graphical data (502) based on the sample probability values. As discussed elsewhere in this disclosure, the computing system 24 can generate one or more types of graphical data from various types based on the sample probability values. For example, in some instances, the computing system 24 generates a graphical representation of ROC. This is described in detail below. Figure 6 This is a sample flowchart containing the actions used to generate the ROC. It is described in detail below. Figure 9 It is an example flowchart containing actions for generating a probability map of the occurrence of arrhythmias plotted over time.

[0086] The computing system 24 can output a user interface (504) for display on a display device. Figure 5In some instances, the user interface includes graphical data. For example, in instances where the graphical data includes ROCs, the user interface may include a graph showing the ROCs. In instances where the graphical data includes a probability graph of the occurrence of arrhythmias plotted over time, the user interface may include the graph. The display device may be one of the output devices 412, one of the user interface devices 410, or another device for displaying data.

[0087] The computing system 24 can output the user interface in one or more of various forms. For example, the computing system 24 can render and output a webpage containing the graphical data. In another example, the computing system 24 can output a graphical user interface for a local application.

[0088] In addition, Figure 5 In one instance, the computing system 24 can receive information for selecting patient 14 through the user interface. Figure 1 The computer system 24 may receive user input instructions (506) for a probability threshold of 14. For example, in an instance where the user interface includes a graph showing the ROC, the computer system 24 may receive user input instructions for selecting a point on the ROC. In an instance where the user interface includes a probability graph of the occurrence of arrhythmias plotted over time, the user interface may further include a threshold indicator, and the computing system 24 may receive user input instructions for positioning the threshold indicator in the graph at a location corresponding to the probability threshold of patient 14. In some instances, the computing system 24 may not require user input instructions for selecting the probability threshold of patient 14. Instead, in the absence of user input instructions for selecting the probability threshold of patient 14, the computing system 24 may set the probability threshold to a default probability threshold or make the probability threshold equal to the default probability threshold. In some such instances, the default probability threshold may be a probability threshold that maximizes the diagnostic positivity rate without considering the expected review burden.

[0089] The computing system 24 can receive patient data (508) from patient 14. The patient data is generated by, for example, medical device 16 ( Figure 1 The patient data may be collected by one or more medical devices, such as 14 and / or other types of devices. As discussed elsewhere in this disclosure, the patient data may include one or more sample sequences of at least one cardiac electrical waveform of patient 14.

[0090] In addition, Figure 5In one instance, the computing system 24 may apply the machine learning model to the patient data to determine a current probability value (510) indicating the probability that patient 14 has experienced an arrhythmia. For example, in an instance where the arrhythmia is atrial fibrillation, the computing system 24 may apply the machine learning model to the patient data and determine that the probability that patient 14 has experienced atrial fibrillation is 0.98.

[0091] Additionally, the computing system 24 can determine whether the current probability value exceeds the probability threshold of patient 14 (512). In response to determining that the current probability value is greater than or equal to the probability threshold of patient 14 (the "yes" branch at 512), the computing system 24 can generate a notification (514) indicating that the patient may have experienced the occurrence of the arrhythmia. The computing system 24 can generate the notification in one or more ways. For example, in one instance, the computing system 24 can send a message to a monitoring user (e.g., a text message, SMS message, instant message, email message, in-app message, voice message, video message, etc.). In this instance, the message notifies the monitoring user that patient 14 may have experienced the occurrence of the arrhythmia. In some instances, the computing system 24 may not generate a notification for each occurrence of the arrhythmia, but rather for groups of arrhythmias. In some instances, the user interface may present a list of generated notifications.

[0092] Otherwise, in Figure 5 In one instance, in response to determining that the current probability value is not greater than or equal to the probability threshold for patient 14 (the "No" branch of 512), the computing system 24 does not generate the notification and can continue to receive the patient's patient data (508). In other instances, in response to determining that the current probability value is not greater than or equal to the probability threshold for patient 14, the computing system 24 may perform other actions. In some instances, for each corresponding arrhythmia among multiple arrhythmias, the computing system 24 may perform... Figure 5 The operation.

[0093] As described above, in some instances, the computing system 24 does not require user input instructions for selecting a probability threshold for a patient, such as patient 14. Therefore, in such instances, the computing system 24 can receive patient data collected by one or more medical devices. In this instance, the computing system 24 can apply the machine learning model to the patient data to determine a current probability value indicating the probability that the patient has experienced an arrhythmia. Furthermore, in this example, the computing system 24 can determine that the current probability value exceeds a default probability threshold, which is set to maximize the diagnostic positivity rate. In response to determining that the current probability value is greater than or equal to the default probability threshold, the computing system 24 can generate a notification indicating that the patient may have experienced the occurrence of the arrhythmia.

[0094] Figure 6 This is a flowchart illustrating a first example operation of generating graphical data and receiving instructions for selecting a probability threshold according to the technology disclosed herein. Figure 6 The examples provide information on how to calculate system 24. Figure 5 How to generate graphical data in action (502) and how to calculate system 24 can be Figure 5 Example details of receiving user input for selecting a probability threshold for patient 14 in action (506).

[0095] exist Figure 6 In this example, as part of generating graphical data based on the sample probability values, the computational system 24 can generate an ROC (600). The ROC is a curve plotting the sensitivity values ​​against the specificity values. Typically, when the sensitivity value is high, the probability that the computational system 24 will not generate a notification indicating that the patient 14 has experienced an arrhythmia is low when the patient 14 has actually experienced one. Therefore, when the sensitivity value is high, there may be more false positives, but the computational system 24 is less likely to miss the occurrence of arrhythmias. When the specificity value is high, the probability that the computational system 24 will generate a notification indicating that the patient 14 has experienced an arrhythmia is low when the patient 14 has not actually experienced one. Therefore, when the specificity value is high, there may be few false positives, but the computational system 24 may miss more occurrences of arrhythmias. Therefore, the sensitivity value decreases as the specificity value increases, and vice versa.

[0096] As part of generating the ROC, the computation system 24 may perform actions (602) to (608) for each corresponding probability threshold in the set of evaluation probability thresholds. The set of evaluation probability thresholds may contain two or more evaluation probability thresholds. Generally, the larger the number of evaluation probability thresholds used, the more data the computation system 24 can use to generate the ROC.

[0097] exist Figure 6 In this example, the computational system 24 can determine the sensitivity value (602) of the corresponding evaluation probability threshold. As mentioned above regarding... Figure 5 As discussed, the computing system 24 can generate a set of sample probability values ​​by applying machine learning patterns to a patient data sample set. Each sample probability value in the sample probability values ​​indicates the probability of the occurrence of the arrhythmia during the corresponding time window. The computing system 24 can determine the sensitivity value of the corresponding evaluation probability threshold as a ratio of (i) the total number of sample probability values ​​in the sample probability value set that are greater than or equal to the corresponding evaluation probability threshold and (ii) the total number of time windows in the sample set that actually contain the occurrence of the arrhythmia.

[0098] In addition, Figure 6 In an example, the computing system 24 can determine the specificity value (604) of the corresponding evaluation probability threshold. The computing system 24 can determine the specificity value of the corresponding evaluation probability threshold as a ratio of (i) the total number of sample probability values ​​that are not greater than or equal to the corresponding evaluation probability threshold and (ii) the total number of time windows in the sample set that actually do not contain the occurrence of the arrhythmia.

[0099] The calculation system 24 can then determine a point (606) on the ROC corresponding to the corresponding probability value. The point on the ROC corresponding to the corresponding probability value is based on the sensitivity value and the specificity value of the corresponding probability threshold. For example, the point can be defined by a pair of coordinates, one of which is the sensitivity value of the corresponding probability threshold and the other is the specificity value of the corresponding evaluation probability threshold. In some instances, the calculation system 24 can apply one or more functions to transform the sensitivity value and the specificity value of the corresponding evaluation probability threshold to determine the coordinates of the point.

[0100] The calculation system 24 can determine whether there are any other evaluation probability thresholds to be evaluated (608). If there are any other evaluation probability thresholds to be evaluated (the "yes" branch of 608), the calculation system 24 can repeat actions (602) to (606) with respect to another evaluation probability threshold among the evaluation probability thresholds. Otherwise (the "no" branch of 608), the calculation system 24 can continue. Figure 5 The operation. In some instances, after determining that there are no other evaluation probability thresholds to evaluate, the computation system 24 can generate curves of smooth or discrete values ​​based on the determined points (e.g., using interpolation, extrapolation, and / or regression).

[0101] Subsequently, the calculation system 24 can receive instructions (506) from user input for selecting a probability threshold for patient 14. Figure 6 In one instance, as part of receiving an instruction for user input to select the probability threshold for patient 14, the computing system 24 may receive an instruction (610) for user input to select a point on the ROC corresponding to the probability threshold for patient 14. The computing system 24 may receive the instruction for user input to select the point on the ROC in one of a variety of ways. For example, in one instance, the user interface may include an indicator element that can slide along the ROC. In this instance, the computing system 24 may receive tap, swipe, or drag input to reposition the indicator element to a location along the ROC. In some instances, the computing system 24 may receive an instruction for user input specifying a probability threshold for patient 14, in which case the computing system 24 may update the positioning of the indicator element such that the indicator element is positioned on the ROC at a location corresponding to the specified probability threshold for patient 14. In some instances, the computing system 24 can receive a user input indicating a specified sensitivity value or specificity value. In this case, the computing system 24 can update the positioning of the indicator element so that the indicator element is positioned on the ROC at a location corresponding to the specified sensitivity value or specificity value.

[0102] Figure 7 This is a flowchart illustrating an example process for selecting a model and operating point based on the cause of monitoring, according to the technology of this disclosure. A user can determine the ICD-10 code (700) corresponding to the cause of monitoring patient 14. Furthermore, the computing system 24 can be configured to use a set of machine learning models. This set of machine learning models may be referred to as a machine learning model library. Each machine learning model may contain one or more neural networks configured to determine the probability value of one or more arrhythmias. Each machine learning model may be associated with a different ICD-10 code, or otherwise associated with different causes of monitoring the patient. The computing system 24 and / or the user can select a machine learning model from the library based on the ICD-10 code associated with the machine learning model (702).

[0103] In addition, Figure 7 In one example, the computing system 24 can present an ROC 704 to a user so that the user can select one or more operation points (706). The operation points may correspond to probability thresholds for the patient 14. The computing system 24 can generate the ROC based on thresholding the probability of arrhythmia (e.g., as shown in the example). Figure 6(As described in the examples). For example, if the monitoring reason is to detect atrial fibrillation (AF) in a stroke patient, the computing system 24 can present a ROC for AF detection from which the user (e.g., a prescribing physician) can select a threshold corresponding to very high sensitivity and low specificity for AF detection. In another example, if the monitoring reason is AF management, the user (e.g., a prescribing physician) can select an operation point from the ROC that has high specificity and low sensitivity for AF. In yet another example, if the monitoring reason is syncope, the physician can select from the ROC a balance of high sensitivity for cardiac arrest and sinus bradycardia, and a balance of sensitivity and specificity for AF and other arrhythmias.

[0104] exist Figure 7 In some instances, a user (e.g., a prescribing physician) can iterate on the diagnostic positivity rate against the review burden of a selected machine learning model and point of action. As discussed elsewhere in this disclosure, review burden can refer to the burden of reviewing a notification, and diagnostic positivity rate can refer to the amount of diagnostically valuable information derived from such notifications. For example, if a user is not using the monitoring service (the "No" branch of 708), the user can assess the review burden on the diagnostic positivity rate once or multiple times before selecting an acceptable balance between review burden and diagnostic positivity rate.

[0105] In some instances, the computing system 24 may present (e.g., output for display on a display device) data indicating the expected review burden versus the expected diagnostic positivity rate for one or more operation points. The operation points for arrhythmias may correspond to probability thresholds. In some instances, the computing system 24 may store historical data for a patient population (e.g., from a monitoring center or database). For each patient in the patient population, the historical data may indicate the patient's operation point, the patient's review burden, and the patient's diagnostic positivity rate. In this example, when an operation point is set for a particular patient, the computing system 24 may identify similar patients in the patient population using said operation point and determine the expected review burden and expected diagnostic positivity rate for said particular patient based on the review burden and diagnostic positivity rate of the identified patients. For example, the computing system 24 may calculate the average of the review burden and diagnostic positivity rate for the identified patients.

[0106] In some instances, the computational system 24 can numerically estimate the expected review burden and expected diagnostic positivity rate at the operation point based on prevalence and algorithm performance. For example, in one instance, it is assumed that the prevalence of AF in the monitored patient population is 20% (i.e., 20% of AF-triggered ECGs from patients in this population actually have AF). This prevalence can be obtained from historical values ​​in the literature or from monitoring centers. Furthermore, in this instance, it is assumed that the user wants to obtain the diagnostic positivity rate and review burden for two AF detection operation points: (i) 95% sensitivity and 70% specificity and (ii) 70% sensitivity and 95% specificity, and reports the corresponding AF episodes detected by the algorithm. AF episodes can correspond to the time window in which the device (e.g., medical device 16) reports the possible presence of AF. In this instance, the "baseline" review burden is the point at which all episodes are reviewed. In this instance, for the first operation point, the review burden is 43% of the baseline. That is, it is assumed that 1000 episodes are presented to the algorithm. Therefore, the expected number of true positives (TP) out of 1000 seizures is TP = 0.95 × 200 = 190; the expected number of false positives (FP) out of 1000 seizures is FP = (1 - 0.7) × 800 = 240; the total number of tests = 190 + 240 = 430; therefore, 430 / 1000 = 43%. 95% (190 out of 200) of the actual seizures are captured, and approximately 44% (190 out of 430) of the reviewed seizures are true AFs. For the second operating point, the review burden is 18% of the baseline. That is, assuming 1000 seizures are presented to the algorithm; TP = 0.7 × 200 = 140; FP = (1 - 0.95) × 800 = 40; total tests = 140 + 40 = 180; therefore, 180 / 1000 = 18%. Here, only 70% (140 out of 200) of the actual seizures were captured, and approximately 78% (140 out of 180) of the reviewed seizures were true atrial fibrillation (AF). Therefore, in this instance, the review burden and diagnostic positivity rate at the first operation point were higher than those at the second operation point.

[0107] In some instances, a user can adjust the review burden versus diagnostic positivity rate of a selected arrhythmia by changing the probability threshold of one or more arrhythmias among the selected arrhythmias (e.g., as described elsewhere in this disclosure). This step helps the user (e.g., a physician) coordinate their operating model to best select one that provides the optimal balance between review burden and diagnostic positivity rate (e.g., if the sensitivity (e.g., the probability threshold) is set to 99%, the review burden may be very high, but if the sensitivity is set to 98%, the review burden may be manageable). The computational system 24 can then use the selected model, and the operating point is used for arrhythmia detection and notification.

[0108] If the user is using the monitoring service (the "Yes" branch of 708) or after assessing the review burden on the diagnostic positivity rate, the computational system 24 can initiate the monitoring process using the selected machine learning model and action point. The monitoring process can receive patient information, apply the selected machine learning model, and generate notifications (e.g., regarding...). Figure 5 (As described in actions (508) to (514)). The monitoring service may be a service that reviews notifications on behalf of a user (e.g., a physician).

[0109] Figure 8 This is a flowchart illustrating an example process for model and operand selection based on a monitored arrhythmia, according to the technology disclosed herein. Figure 8 In this example, a user (e.g., a prescribing physician) can identify a set of arrhythmias to monitor (800). The set of arrhythmias to monitor can be those arrhythmias the user wants to receive notifications about. In this way, the user can select one or more arrhythmias from the set. The selected arrhythmias can be arrhythmias that the user is interested in monitoring.

[0110] In addition, such as Figure 8 As shown in the examples, machine learning models 802A-802N (collectively referred to as "machine learning models 802") can be developed for different groups of arrhythmias. For example, machine learning model 802A can generate probability values ​​indicating the probability that patient 14 has experienced atrial fibrillation (AF), pause, and sinus bradycardia; machine learning model 802B can generate probability values ​​indicating the probability that patient 14 has experienced sinus bradycardia, pause, and atrial-ventricular (AV) block; machine learning model 802C can generate probability values ​​indicating the probability that patient 14 has experienced atrial fibrillation, atrial flutter, and supraventricular tachycardia (SVT); machine learning model 802N can generate probability values ​​indicating the probability that patient 14 has experienced sinus tachycardia, sinus bradycardia, atrial fibrillation (AF), atrial flutter, supraventricular tachycardia (SVT), atrial-ventricular (AV) block, and intraventricular conduction delay (IVCD). Figure 8 In this instance, each machine learning model in machine learning model 802 corresponds to a different category of arrhythmia. In other instances, different machine learning models may exist for different arrhythmias rather than different categories of arrhythmias. In some instances, each machine learning model in machine learning model 802 may be machine learning model 452. Figure 4The different machine learning models in ) . Each machine learning model in machine learning model 802 can be implemented using one or more neural networks. The developed machine learning models 802 can be combined to form a machine learning model "library".

[0111] The user can select one or more machine learning models from the machine learning model library 802 (804). For example, if the user chooses to monitor sinus bradycardia, AV block, and atrial flutter, the user can select machine learning models 802B and 802C. Note that in some cases, the selected machine learning model can be configured to generate probability values ​​for one or more arrhythmias that the user may not necessarily want to monitor. In some instances, the computing system 24 can receive user input instructions to select one or more machine learning models from the machine learning model library 802.

[0112] In addition, Figure 8 In one example, the computing system 24 can present one or more ROCs 808 to a user, allowing the user to select one or more operation points (806) for patient 14. Each operation point in patient 14 can correspond to a probability threshold used to identify the occurrence of different arrhythmias. Therefore, based on the selected arrhythmia to be monitored, a corresponding machine learning model can be selected and a corresponding ROC can be presented to allow the user to select operation points. The user can select different operation points on each of the ROCs. Figure 8 In one instance, a user can select an operation point on the ROC by moving an indicator element 809 on the ROC with user input, thereby providing the computing system 24 with an indication of user input for selecting a point on the ROC corresponding to the probability threshold of the patient 14's arrhythmia.

[0113] exist Figure 8In some instances, a user (e.g., a prescribing physician) can iterate the diagnostic positivity rate against the review burden of a selected machine learning model and action point. As discussed elsewhere in this disclosure, review burden can refer to the burden of reviewing a notification, and diagnostic positivity rate can refer to the amount of diagnostically valuable information derived from such notifications. For example, if the user is not using the monitoring service (the "No" branch of 810), the user can assess the review burden against the diagnostic positivity rate once or multiple times before reaching an acceptable balance between the selected review burden and the diagnostic positivity rate. In some instances, the user can adjust the review burden against the diagnostic positivity rate by changing the probability threshold of one or more arrhythmias among the selected arrhythmias (e.g., as described elsewhere in this disclosure). If the user is using the monitoring service (the "Yes" branch of 810) or after assessing the review burden against the diagnostic positivity rate, the computing system 24 can initiate a monitoring process using the selected machine learning model and action point. The monitoring process can receive patient information, apply the selected machine learning model, and generate notifications (e.g., regarding...). Figure 5 (As described in actions (508) to (514)). In some instances, the computing system 24 may receive subsequent instructions from user input for updating the probability threshold of patient 14.

[0114] Figure 9 This is a flowchart illustrating a second example operation of generating graphical data and receiving instructions for selecting a probability threshold according to the technology disclosed herein. Figure 9 The examples provide information on how to calculate system 24. Figure 5 How to generate graphical data in action (502) and how to calculate system 24 can be Figure 5 Example details of receiving user input for selecting a probability threshold for patient 14 in action (506).

[0115] like Figure 9 As shown in the example, as part of generating graphical data based on the sample probability values, the computing system 24 can plot a graph (900) of the sample probability values ​​over time. (See above regarding...) Figure 5 As discussed, the computational system 24 can generate a set of sample probability values ​​by applying a machine learning model to a sample set of patient data. Each sample probability value indicates the probability of an arrhythmia occurring during the corresponding time window. In some instances, the computational system 24 can generate smooth or discrete curves based on the sample probability values.

[0116] In addition, Figure 9 In this example, the computing system 24 can generate a threshold indicator (902). Additionally, in Figure 9In one instance, as part of an indicator (506) for receiving user input for selecting a probability threshold for patient 14, the computing system 24 may receive an indication (904) for positioning the threshold indicator at a position in the figure corresponding to the probability threshold of patient 14.

[0117] The calculation system 24 can generate the threshold indicator in one or more ways. For example, in one instance, the threshold indicator includes a threshold bar superimposed on the graph and oriented parallel to the time axis of the graph. In this instance, the threshold bar can be superimposed on the graph such that, based on the set of sample probability values, the threshold bar appears above or below the curve or data point. Furthermore, in some instances where the threshold indicator includes a threshold bar, the calculation system 24 can receive a user input instruction for sliding the threshold bar in a direction perpendicular to the time axis to position the threshold indicator in the graph at a location corresponding to the expected probability threshold for patient 14. In some instances, the calculation system 24 can receive a user input instruction (e.g., in text form) for specifying the expected probability threshold for patient 14, and the calculation system 24 can update the position of the threshold indicator to the position in the graph corresponding to the expected probability threshold for patient 14. In some instances, the threshold indicator can include an arrow, pointer, or other type of graphical element in or adjacent to the graph.

[0118] Figure 10 This is a conceptual diagram of example 1000, which includes a raw cardiac electrical waveform over a certain time period, and example 1002, which includes the probability of arrhythmic events during the same time period, according to the technology of this disclosure. Figure 10 In the example, 1000 corresponds to an electrocardiogram (EGM) indicating the overall amplitude of the cardiac electrical potential recorded during the time period described. Figure 10 In one example, the duration of the time period is approximately 45 seconds.

[0119] 1002 contains waveforms corresponding to different arrhythmias within a set of arrhythmias. For example, 1002 may contain waveforms corresponding to different ICD-10 types. In some instances, a set of arrhythmias may be selected based on physician interest or patient condition.

[0120] exist Figure 10 In the example, the set of arrhythmias includes grade 1 atrioventricular block (AVB), atrial fibrillation, premature ventricular contractions (PVC), sinus rhythm, supraventricular tachycardia, and noise / nonphysiological signal segments. Figure 10The arrhythmias are categorized as "artifacts," atrial flutter, sinus bradycardia, and sinus tachycardia. For each arrhythmia, the waveform corresponding to that arrhythmia is based on a sample probability value, which indicates the probability of the arrhythmia occurring within a time window that ends at a time value corresponding to that sample probability value. Figure 10 The waveform shown in the example is determined based on the 1000 waveform. In Figure 10 In this example, for ease of interpretation, probability values ​​are mapped (e.g., linearly expanded) to index values. Figure 10 In this instance, the index value is labeled as a "waveform heatmap" value.

[0121] exist Figure 10 In this example, threshold bar 1004 is superimposed on 1002. Threshold bar 1004 is positioned on 1002 at a location corresponding to the expected probability threshold for patient 14. Figure 10 In this example, threshold bar 1004 is positioned on 1002 at the location corresponding to index value 2. If the probability value of one of the arrhythmias rises above the probability threshold corresponding to the location indicated by threshold bar 1004, the calculation system 24 can generate a notification indicating that patient 14 may have experienced an arrhythmia. For example, in Figure 10 In the example, given the location of threshold bar 1004, the calculation system 24 can determine that the patient may have experienced six occurrences of PVC, two or more occurrences of sinus arrhythmia, and an occurrence of atrial fibrillation.

[0122] like Figure 10 As illustrated in the example, the computing system 24 can receive instructions from user input to update the position of the threshold bar 1004 to a higher position in 1002. Therefore, given the updated position of the threshold bar 1004, the computing system 24 can still determine that the patient 14 may have experienced six occurrences of PVC, but is uncertain whether the patient 14 has experienced an occurrence of sinus arrhythmia or atrial fibrillation.

[0123] In some instances, different threshold bars may exist for different arrhythmias. For example, a first threshold bar may exist for grade 1 AVC; a second threshold bar may exist for atrial fibrillation; a third threshold bar may exist for PVC; and so on. Therefore, users can be able to set different probability thresholds for patient 14 for different arrhythmias. For example, if patient 14 is known to frequently experience atrial flutter without serious consequences, the user may need a high probability threshold for atrial flutter, but may need a lower probability threshold for sinus bradycardia.

[0124] In some instances, the technology of this disclosure includes a system comprising means for performing any of the methods described herein. In some instances, the technology of this disclosure includes a computer-readable medium comprising instructions for causing a processing circuitry system to perform any of the methods described herein.

[0125] The following is a non-limiting list of examples of one or more technologies according to this disclosure.

[0126] Example 1. A method comprising: generating a set of sample probability values ​​by a computing system including a processing circuitry system and a storage medium by applying a machine learning model to a sample set of patient data, wherein: the machine learning model is trained using patient data from a plurality of patients, the sample set including a plurality of time windows; and for each respective time window of the plurality of time windows, the machine learning model is configured to determine a corresponding probability value in the set of sample probability values ​​that indicates the probability of an arrhythmia occurring during the respective time window; generating graphical data by the computing system based on the sample probability values; and outputting a user interface for display on a display device by the computing system, the user interface including the graphical data. The computing system receives, via the user interface, an instruction for selecting a probability threshold for a patient; the computing system receives patient data of the patient, wherein the patient data is collected by one or more medical devices; the computing system applies the machine learning model to the patient data to determine a current probability value indicating the probability that the patient has experienced an arrhythmia; the computing system determines that the current probability value exceeds the patient's probability threshold; and in response to determining that the current probability value is greater than or equal to the patient's probability threshold, the computing system generates a notification indicating that the patient may have experienced the occurrence of the arrhythmia.

[0127] Example 2. According to the method of Example 1, wherein: generating the graphical data includes generating a receiver operating curve (ROC) by the computing system, wherein generating the ROC includes: for each of a plurality of evaluation probability thresholds: the computing system determines the sensitivity value of the corresponding evaluation probability threshold as a ratio of the following two: (i) the total number of sample probability values ​​in the sample probability value set that are greater than or equal to the corresponding evaluation probability threshold and (ii) the total number of time windows in the sample set that actually contain the occurrence of the arrhythmia that actually occurred in the sample set; the computing system determines the specificity value of the corresponding probability value as a ratio of the following two: (i) not The total number of sample probability values ​​greater than or equal to the corresponding evaluation probability threshold and (ii) the total number of time windows in the sample set that actually do not contain the occurrence of the arrhythmia; and the calculation system determines a point on the ROC corresponding to the corresponding probability value, wherein the point on the ROC corresponding to the corresponding probability value is based on the sensitivity value and the specificity value of the corresponding evaluation probability threshold; and receiving an instruction for selecting the probability threshold of the patient includes: the calculation system receiving an instruction for selecting a point on the ROC corresponding to the probability threshold of the patient.

[0128] Example 3. The method according to any one of Examples 1 or 2, wherein: generating the graphical data includes: generating a graph by the computing system that plots the relationship between the sample probability values ​​and time; generating a threshold indicator by the computing system; and receiving the user input instruction includes receiving by the computing system an instruction from the user input for positioning the threshold indicator at a position in the graph corresponding to the probability threshold of the patient.

[0129] Example 4. According to the method of Example 3, wherein the threshold indicator includes a threshold bar superimposed on the graph and oriented parallel to the time axis of the graph.

[0130] Example 5. The method according to any one of Examples 1 to 4, wherein receiving the patient data of the patient includes receiving the patient's cardiac electrical waveform data by the computing system.

[0131] Example 6. The method according to any one of Examples 1 to 5, further comprising receiving, by the computing system, an instruction for updating the probability threshold of the patient.

[0132] Example 7. The method according to any one of Examples 1 to 6, wherein the medical device includes a wearable medical device or an implantable medical device (IMD).

[0133] Example 8. A method according to any one of Examples 1 to 7, wherein: the arrhythmia is a first arrhythmia among a plurality of arrhythmias, and for each corresponding arrhythmia among the plurality of arrhythmias, the method comprises: generating a corresponding sample probability value set by the computing system by applying a corresponding machine learning model to a corresponding patient data sample set, wherein: the corresponding machine learning model is trained using patient data from the plurality of patients, the corresponding sample set comprising a plurality of corresponding time windows; and for each of the plurality of corresponding time windows, the corresponding machine learning model is configured to: determine a corresponding probability value in the corresponding sample probability value set indicating the probability of the corresponding arrhythmia occurring during the corresponding time window; and the computing system... The computing system generates corresponding graphical data from the corresponding sample probability value set; outputs a user interface for display on the display device, such that the user interface includes the corresponding graphical data; receives, through the user interface, an instruction from a user for selecting a corresponding probability threshold for the patient; applies the machine learning model to the patient data to determine a corresponding probability value indicating the probability that the patient has experienced the occurrence of the corresponding arrhythmia; determines that the corresponding probability value exceeds the corresponding probability threshold; and in response to determining that the corresponding probability value is greater than or equal to the corresponding probability threshold, generates a notification indicating that the patient may have experienced the occurrence of the corresponding arrhythmia.

[0134] Example 9. The method according to any one of Examples 1 to 8, further comprising: the calculation system presenting data indicating the expected review burden versus the expected diagnostic positivity rate of the patient's probability threshold.

[0135] Example 10. A computing system comprising a processing circuitry and a storage medium, the computing device being configured to perform the method according to any one of Examples 1 to 9.

[0136] Example 11. A method as described in the specification.

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

[0138] In one or more instances, the described techniques may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, functionality may be stored on a computer-readable medium in the form of one or more instructions or code and may be executed by a hardware-based processing unit. The computer-readable medium may include non-transitory computer-readable media, which corresponds to tangible media such as data storage media (e.g., RAM, ROM, EEPROM, flash memory, or any other medium that can be used to store desired program code in the form of instructions or data structures and that can be accessed by a computer).

[0139] The instructions can be executed by one or more processors, such as one or more digital signal processors (DSPs), general-purpose microprocessors, application-specific integrated circuits (ASICs), field-programmable arrays (FPGAs), or other equivalent integrated or discrete logic circuit systems. Therefore, the terms "processor" or "processing circuit system" as used herein can refer to any of the foregoing structures or any other physical structure suitable for implementing the described techniques. Furthermore, the techniques can be implemented entirely with one or more circuit or logic elements.

[0140] Various examples have been described. These and other examples are within the scope of the following claims.

Claims

1. A computing system comprising: One or more processing circuits; as well as A storage medium storing instructions that, when executed, configure the one or more processing circuits to: A set of sample probability values ​​is generated by applying a machine learning model to a patient data sample set, where: The machine learning model was trained using patient data from multiple patients. The sample set includes multiple time windows; and For each of the plurality of time windows, the machine learning model is configured to determine a corresponding probability value in the sample probability value set that indicates the probability of an arrhythmia occurring during the corresponding time window. Graphical data is generated based on the sample probability values, wherein, as part of generating the graphical data, the one or more processing circuits generate a graph showing the relationship between the sample probability values ​​and time, and generate a threshold indicator. Output a user interface for display on a display device, the user interface including the graphical data; The user interface receives instructions for selecting a probability threshold for a patient. The system receives patient data of the patient, wherein the patient data is collected by one or more medical devices, and as part of an instruction to receive the user input, the one or more processing circuits receive an instruction from the user input for positioning the threshold indicator at a position in the figure corresponding to the probability threshold of the patient; The machine learning model is applied to the patient data to determine a current probability value indicating the probability that the patient has experienced an arrhythmia. It is determined that the current probability value exceeds the patient's probability threshold; and In response to determining that the current probability value is greater than or equal to the patient's probability threshold, a notification indicating that the patient may have experienced the occurrence of the arrhythmia is generated.

2. The computing system according to claim 1, wherein: The one or more processing circuits are configured such that, as part of generating the graphical data, the one or more processing circuits generate a receiver operating curve (ROC), the ROC plotting sensitivity values ​​against specificity values, wherein the one or more processing circuits are configured such that, as part of generating the ROC, the one or more processing circuits: For each of the multiple evaluation probability thresholds: The sensitivity value of the corresponding evaluation probability threshold is determined as the ratio of the following two: (i) the total number of sample probability values ​​in the sample probability value set that are greater than or equal to the corresponding evaluation probability threshold and (ii) the total number of time windows in the sample set that actually contain the occurrence of the arrhythmia that actually occurred in the sample set. The specificity value of the corresponding evaluation probability threshold is determined as the ratio of the following two: (i) the total number of sample probability values ​​that are not greater than or equal to the corresponding evaluation probability threshold and (ii) the total number of time windows in the sample set that do not actually contain the occurrence of the arrhythmia. and Determine the point on the ROC corresponding to the corresponding probability value, wherein the point on the ROC corresponding to the corresponding probability value is based on the sensitivity value of the corresponding evaluation probability threshold and the specificity value of the corresponding evaluation probability threshold; and The one or more processing circuits are configured such that, as part of receiving an instruction for selecting a point on the ROC corresponding to the probability threshold of the patient, the one or more processing circuits receive an instruction for selecting a point on the ROC corresponding to the probability threshold of the patient.

3. The computing system of claim 1, wherein the threshold indicator comprises a threshold bar superimposed on the graph and oriented parallel to the time axis of the graph.

4. The computing system of claim 1 or 2, wherein the one or more processing circuits are configured such that, as part of receiving the patient data, the one or more processing circuits are configured to receive the patient's cardiac electrical waveform data.

5. The computing system of claim 1 or 2, wherein the one or more processing circuits are further configured to receive instructions from user input for updating the probability threshold of the patient.

6. The computing system according to claim 1 or 2, wherein the medical device includes a wearable medical device or an implantable medical device (IMD).

7. The computing system according to claim 1 or 2, wherein: The aforementioned arrhythmia is the first of several arrhythmias, and For each of the plurality of arrhythmias, the computing system is configured to perform the following operation: the computing system generates a set of corresponding sample probability values ​​by applying a corresponding machine learning model to a corresponding patient data sample set, wherein: The corresponding machine learning model was trained using patient data from the multiple patients. The corresponding patient data sample set includes multiple corresponding time windows; and For each of the plurality of corresponding time windows, the corresponding machine learning model is configured to: determine a corresponding probability value in the corresponding sample probability value set that indicates the probability of the corresponding arrhythmia occurring during the corresponding time window; The computing system generates corresponding graphical data based on the corresponding sample probability value set; The computing system outputs a user interface for display on the display device, such that the user interface includes the corresponding graphical data; The computing system receives, through the user interface, an instruction from the user to select the corresponding probability threshold for the patient; The computing system applies the machine learning model to the patient data to determine a probability value indicating the probability that the patient has experienced the corresponding arrhythmia. The calculation system determines that the corresponding probability value exceeds the corresponding probability threshold; and In response to determining that the corresponding probability value is greater than or equal to the corresponding probability threshold, the calculation system generates a notification indicating that the patient may have experienced the occurrence of the corresponding arrhythmia.

8. The computing system of claim 1 or 2, wherein the one or more processing circuits are further configured to present data indicating the expected review burden versus the expected diagnostic positivity rate of the patient's probability threshold.

9. A computer-readable data storage medium having instructions stored thereon, which, when executed, cause one or more processing circuits of a computing system to: A set of sample probability values ​​is generated by applying a machine learning model to a patient data sample set, where: The machine learning model was trained using patient data from multiple patients. The sample set includes multiple time windows; and For each of the plurality of time windows, the machine learning model is configured to determine a corresponding probability value in the sample probability value set that indicates the probability of an arrhythmia occurring during the corresponding time window. Graphical data is generated based on the sample probability values, wherein, as part of generating the graphical data, the one or more processing circuits generate a graph showing the relationship between the sample probability values ​​and time, and generate a threshold indicator. Output a user interface for display on a display device, the user interface including the graphical data; The user interface receives an instruction for selecting a probability threshold for a patient. As part of the instruction to receive the user input, the one or more processing circuits receive an instruction for positioning the threshold indicator in the figure at a position corresponding to the probability threshold of the patient. Receive patient data from the patient, wherein the patient data is collected by one or more medical devices; The machine learning model is applied to the patient data to determine a current probability value indicating the probability that the patient has experienced an arrhythmia. It is determined that the current probability value exceeds the patient's probability threshold; and In response to determining that the current probability value is greater than or equal to the patient's probability threshold, a notification indicating that the patient may have experienced the occurrence of the arrhythmia is generated.

Images