Pace-making ventricular tachycardia detection

Abstract

The invention relates to pacing ventricular tachycardia detection. A physiological monitoring system collects electrocardiogram ("ECG") data from a through-monitoring sensor via a sensor interface. The machine learning model identifies characteristics of ECG data attributable to a cardiovascular device used by a patient. The machine learning model creates and persistently updates at least one normal ECG template for an individual patient. The abnormal ECG dataset is analyzed using a knowledge-based heartbeat classifier to identify a particular arrhythmia. The data structure stores a response to the particular arrhythmia based on the possible impact of the particular arrhythmia on the patient. For example, the response to dangerous arrhythmias may be an emergency alert, while the response to transient arrhythmias quickly corrected by the patient's pacemaker may be a non-urgent message or annotation to be archived.

Description

Technical Field

[0001] The present disclosure relates generally to the field of medical monitoring of physiological parameters of a patient. More specifically, the present disclosure relates to detecting, analyzing, and responding to electrocardiogram (ECG) abnormalities in a patient using a cardiovascular device. Background Art

[0002] This section introduces technical information that may be relevant to the described and / or claimed subject matter. It is intended to provide context to aid in understanding the present disclosure. Under relevant patent law, such potentially relevant "background art" may or may not qualify as "prior art." Therefore, its discussion in this section is not an admission of prior art and should not be construed as such.

[0003] Many advances in medical diagnosis and treatment have stemmed from the ability of clinicians to collect and interpret measurements of observable phenomena related to various aspects of a patient's health (collectively, "physiological parameters"). Collecting and comparing measurements taken at different times or under different conditions ("monitoring" physiological parameters) can reveal meaningful trends, sensitivities, or possible prognoses that can better inform ongoing patient care.

[0004] Automating the measurement and monitoring of physiological parameters offers several valuable benefits. For example, because measurement equipment can be calibrated to uniform standards, variations associated with subjectivity are reduced. Artificial sensors may be able to measure physiological parameters over a wider range with greater sensitivity than human vision, hearing, or touch. Delegating physiological parameter monitoring to equipment frees up clinicians to observe other patient parameters or monitor several patients simultaneously. Beyond the benefits mentioned here, automation of medical monitoring can yield other benefits. Consequently, the development of medical monitoring devices, systems, and processes continues to evolve and develop in line with patient needs and the availability of new technologies.

[0005] The beating of the human heart is one of the most fundamental vital signs, and heart disease is one of the most common causes of death worldwide. A patient whose heart stops beating may die unless the heart can be restarted within a short enough time. However, many types of heart disease can be detected through known early warning signs before the condition progresses enough to cause the heart to stop. Various steps can then be taken to mitigate the risk. Continuous monitoring allows for the effectiveness of mitigation measures to be assessed.

[0006] One type of early warning sign is an arrhythmia, in which the heart beats at an abnormal rate (such as tachycardia (which is too fast) or bradycardia (which is too slow)) or in an abnormal rhythm (missed or extra beats in repeated cycles). There are many different types and degrees of arrhythmias, ranging from minor abnormalities that occur occasionally in an otherwise healthy heart to serious conditions that require immediate medical attention. They can be caused by other heart problems, other diseases, medications, aging, or other factors. The effects felt by the patient can range from mild discomfort to incapacity. Therefore, correctly identifying the type of arrhythmia a patient is experiencing can be crucial to diagnosing potential problems and implementing effective countermeasures.

[0007] An electrocardiogram ("ECG") is one of the most common measurements used to detect and analyze cardiac arrhythmias (although it has many other applications). As the heart beats, the heart muscle surrounding each of the four heart chambers (the right and left atria and the right and left ventricles) can contract and relax in sequence. During the contraction that forces blood out of each chamber, the cells of the muscle tissue involved can electrically depolarize, or discharge. The resulting electric field pulses propagate outward from the heart to produce tiny but detectable voltage changes in the microvolt range at the surface of the skin. The ECG measures these voltage changes through one or more monitoring sensors in contact with the skin. The voltage can be measured over a sufficient time to collect a data set spanning several heart beat cycles (e.g., 2.5 to 10 seconds). The collected data set of voltage relative to time can be called an "ECG" or "trace" or "waveform." The term "ECG" can also refer to the instrument that collects the data, or the process of collecting the data.

[0008] Historically, ECGs were drawn on graph paper using a pen plotter and read by eye. Plotters were typically calibrated to a standard speed and scale, making it easy to compare ECGs collected under different conditions. Reading and interpreting an ECG is a skill that clinicians are required to learn. Beyond the basics of identifying an "average normal" ECG and a few common abnormalities, the amount and level of information that a human reader can extract from an ECG, and the accuracy of the inferences they can extrapolate about a patient's health, vary greatly. In addition to the extremely wide variety of possible abnormalities, some individual patients also have atypical ECGs that are their "personal normal" without underlying disease. Not all of the best ECG readers (whether through experience or a "good eye") can effectively teach others to replicate their success.

[0009] The more complex the trace, the more difficult it is to interpret. Additional complexity can be added to the ECG trace if the action of a cardiovascular device affects the measured voltage changes. For example, a pacemaker can provide timed electrical pulses to maintain a stable heart rhythm that might otherwise develop an arrhythmia. Some pacemakers operate continuously, while others allow the patient's heart to self-regulate until its internal sensors detect an abnormality, such as tachycardia (beating too fast) or bradycardia (beating too slow). Like "intrinsic" electrical signals generated directly by the heart muscle, pacing pulses can be detected by the ECG at the surface of the skin, and the resulting trace reflects a mixed or combined signal from both the heart and the pacemaker. This mixed dataset can be more complex than a dataset generated by the patient's heart alone. Other types of cardiovascular devices add their own complexity. Pacing pulses and other contributions from cardiovascular devices are often referred to as "stimulation artifacts" ("SA"). Unless SA can be detected and analyzed in the ECG of a patient using the device, an ECG trace of a rhythm successfully corrected by the device may be mistaken for an abnormal trace, leading to the recommendation or implementation of unnecessary treatment.

[0010] Although cardiovascular devices are intended to regulate heart rhythm, some devices may allow (or even cause) arrhythmias (e.g., pacemaker-mediated tachycardia or "PMT"). Some of these situations are due to device malfunction or a malfunction that needs to be corrected. However, some newer devices may be designed to temporarily allow one type of arrhythmia (such as tachycardia) in order to prevent a more dangerous arrhythmia (such as fibrillation). As newer cardiovascular devices operate with increasingly sophisticated algorithms, accurate analysis of the resulting complex ECG traces will allow clinicians to discover whether a patient's device is functioning as intended.

[0011] Computers that analyze digitized ECGs have the advantage of being able to mathematically separate mixed signals, such as those from a patient's heart and a cardiovascular device. They can also process the data to enhance detail, convert it from the time domain to the frequency domain, perform interpolation, extrapolation, curve fitting, compare different data sets, and use many other techniques to extract information from the collected signals. In addition, computers can perform many analyses at a sufficient speed to deliver results to clinicians in real time, enabling them to intervene promptly. However, the variety of natural and device-modulated ECG traces that can be generated by the general patient population remains too numerous to be satisfactorily handled by conventional manually written computer instructions based on declarative or procedural knowledge. An alternative would be a valuable addition to patient monitoring technology that reliably warns clinicians when a patient's life is in immediate danger while minimizing the occurrence of false alarms that could waste resources and cause unnecessary pain to the patient. Summary of the Invention

[0012] The following brief summary is intended to provide a basic understanding of some aspects of the present disclosure that may be included in the appended claims. It is not intended to be an exhaustive overview of the subject matter, a complete listing of key or critical elements, or a statement of limitations on what may be claimed. Its sole purpose is to introduce some concepts and terms that will be discussed in greater depth in the detailed description.

[0013] The physiological monitoring system collects electrocardiogram (ECG) data from monitoring sensors via a sensor interface. Analysis software identifies ECG data characteristics that can be attributed to the cardiovascular device used by the patient. Through machine learning, a personalized template of normal ECG ranges for the patient and device is created and continuously updated.

[0014] Abnormal ECG data sets (those that do not match the patient's normal template, and optionally those collected before the patient's normal template is learned) are analyzed by one or more stored heartbeat classifiers representative of specific arrhythmias. The comparison determines what impact the arrhythmia may have on the patient and prompts the system to respond accordingly. For example, arrhythmias that cause severe pain or risk can prompt an emergency alert to the clinician; arrhythmias that cause moderate discomfort can prompt a real-time non-urgent warning to the clinician; and arrhythmias that can be corrected quickly and painlessly by a cardiovascular device can prompt a non-urgent message.

[0015] The disclosed method for detecting arrhythmias in electrocardiogram ("ECG") data includes at least: (1) collecting at least one of (a) an intrinsic ECG signal from a patient and (b) stimulation artifacts from a cardiovascular device used by the patient; (2) creating a normal template from a continuous ECG data set through unsupervised machine learning; (3) detecting the presence of a specific arrhythmia in the ECG data by applying at least one knowledge-based heartbeat classifier; and (4) executing a response corresponding to the expected effect of the specific arrhythmia on the patient.

[0016] The disclosed physiological monitoring system includes at least: (1) at least one electrocardiogram ("ECG") port; (2) a processor communicatively coupled to the ECG port; and (3) a memory communicatively coupled to the ECG port and encoded with (a) a data structure comprising (i) a heartbeat classifier to detect specific arrhythmias in ECG data, (ii) corresponding responses linked to predefined arrhythmias, (iii) a normal template learned from the ECG data, and (iv) a set of instructions that, when executed by the processor, cause the processor to: (A) collect an ECG data set through the ECG port, the data set comprising at least one of (I) an intrinsic ECG signal from a patient and (II) a stimulation artifact from a cardiovascular device of the patient; (B) compare the ECG data set to a normal template; and (C) update the normal template using the ECG data set if the ECG data set matches the normal template.

[0017] The disclosed non-transitory computer-readable storage medium includes instructions that, when executed, cause a processor to at least: (1) collect electrocardiogram ("ECG") data comprising at least one of an intrinsic ECG signal from a patient and stimulation artifacts from a cardiovascular device used by the patient; (2) create a normal template from a continuous ECG data set through unsupervised machine learning; (3) detect the presence of a specific arrhythmia in the ECG data by applying at least one knowledge-based heartbeat classifier; and (4) execute a response corresponding to an expected effect of the specific arrhythmia on the patient.

[0018] Additional methods, systems, and non-transitory storage media are described in this document and accompanying figures.

[0019] The above content is intended to introduce some possible implementations of the subject matter of the present invention, and does not limit the scope of protection granted by the claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings illustrate the described embodiments in detail by way of example. The accompanying drawings and the corresponding description are not intended to limit the scope of the protected subject matter. Instead, it is intended to protect all modifications, equivalents, and alternatives that fall within the spirit and scope of the appended claims.

[0021] Figure 1 A physiological monitoring system according to some embodiments is schematically illustrated.

[0022] Figure 2 A patient undergoing ECG monitoring while using a cardiovascular device is schematically illustrated.

[0023] Figure 3 Examples of selected types of ECG traces and their structural features that can be analyzed by some published methods are presented.

[0024] Figure 4 is a general overview of an analysis model according to some embodiments.

[0025] Figure 5 is a conceptual diagram of a memory configured to implement some embodiments of the disclosed method.

[0026] Figure 6 is a flow chart of instructions for a normal template learning session according to some disclosed methods.

[0027] Figure 7 It is a decision tree for ECG diagnostic analysis using a knowledge-based heartbeat classifier according to some published methods.

[0028] Figure 8 is an example ECG trace of an intrinsic tachycardia that was successfully corrected by a cardiovascular device.

[0029] Figure 9 is an example ECG trace of an intrinsic tachycardia that a cardiovascular device has attempted to correct but failed.

[0030] Figure 10 is an example ECG trace of endless-loop tachycardia ("ELT"), a type of pacemaker-mediated tachycardia.

[0031] Figure 11 is an example ECG trace of a pacemaker-mediated conduction block ventricular tachycardia.

[0032] In the drawings, like reference numbers generally indicate elements of similar function, similar structure, or both. DETAILED DESCRIPTION

[0033] Unless otherwise defined, all terms used herein (including technical and / or scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art. Definitions of some field-specific terms as used in this document are presented below. Definitions of more general or field-independent patent terms are presented near the end of the document.

[0034] "Abnormal" may describe the allowable margin by which a characteristic of the ECG deviates from the patient's stored personalized normal ECG template, or from one or more other stored references or standards.

[0035] An "allowable margin" can refer to the maximum error or deviation that can be measured between one dataset and another, or between one dataset and the average of multiple other datasets, while still considering the datasets a match. This maximum can be preset, derived from a machine learning process, adjusted by a clinician, or derived from other sources.

[0036] As used herein, "collecting ECG data" refers to, but is not limited to, manipulating analog signals to meet the requirements of the next stage for further processing. ECG signal processing may include converting between analog and digital domains (e.g., via analog-to-digital or digital-to-analog converters), amplifying, filtering, converting, biasing, range matching, isolation, and any other process required to make the sensor output suitable for processing.

[0037] As used herein, a "computer" may be any known computing device or apparatus, integrated or distributed, physical or virtual, digital or analog, using binary, ternary, decimal, hexadecimal, or any suitable notation. It may include a processor, a means for entering input, a means for accessing output, and optionally a means for storing and / or retrieving data.

[0038] "Device modulation" refers to a dataset of physiological measurements that are attributable to characteristics of a medical device used by a patient. For example, the ECG of a patient using a pacemaker may exhibit regularly spaced spikes that are identifiable as "pacing pulses" generated by the pacemaker.

[0039] "ECG," when used as a noun, can refer to the machine that performs ECG measurements or to its output. When used as an adjective, it can refer to anything related to the process of ECG measurement.

[0040] In its broadest sense, a "fused beat" refers to an ECG feature that occurs when the myocardium is triggered by more than one source. However, for the purposes of this disclosure, a "fused beat" refers to a stimulation artifact (such as a pacemaker pulse) that coincides so closely with the QRS complex that it is difficult to distinguish between the intrinsic and stimulated contributions. For example, the QRS complex may appear simply to be wider, rather than having a discernible additional peak.

[0041] In the context of programming, a "link" can mean any association or relationship between two or more pieces of information that can be perceived and acted upon by a running program. A link might reside in a non-transient data structure or a set of processing instructions. A link need not be distinct code items but rather the proximity or relative position of two pieces of information, such as in a table or register. In contrast, in the context of communications hardware, a "link" might be any suitable channel over which a communications component can send or receive data.

[0042] A "machine" as used herein is any device, physical or virtual, that transmits or modifies energy of any type to perform or assist in the performance of a human task.

[0043] “Machine learning” can refer to algorithms that look for patterns in data, often large amounts of data. Once a pattern is found, the algorithm can fit new input data into the pattern or predict how various changes might affect the pattern. Rather than being explicitly programmed to fit a function in a predefined way, a machine learning algorithm can use statistical methods to adapt its approach based on “experience” (previous results). Machine learning models can solve problems (such as facial recognition, voice recognition, and customer preference prediction) that would be too expensive, difficult, or time-consuming for traditional algorithms designed by human programmers. (Although a subset of machine learning, called “deep learning,” can use artificial neural networks (ANNs), a wide range of machine learning models can be successful without ANNs.)

[0044] As used herein, a "match" between data sets does not necessarily mean an exact match, but rather allows for a tolerable error margin, which may be predefined and stored in a data structure or derived by a processor from a stored instruction being executed.

[0045] A "model" in machine learning is a program that learns a task, such as recognizing patterns, by training it on one or more training sets.

[0046] A "module" may include hardware, software, or a combination of both, which are assembled to implement a desired functionality. The degree of interchangeability between software and hardware is well known in the art and has applicability in technical contexts such as those disclosed herein. Thus, a module may be implemented, for example, in an electronic logic circuit composed of electronic components such as transistors, resistors, inductors, capacitors, etc. Alternatively, a module may be implemented in software in the form of executable instructions stored by a processor-based resource. Alternatively, a module may be implemented in an integrated circuit such as an ASIC, EPROM, or EEPROM. Alternatively, a module may combine some or all of these technologies.

[0047] “Pacing dependence” may describe a patient whose pacemaker or other cardiovascular device is programmed to run all the time because their heart is unlikely to regulate itself without such help.

[0048] A "sensor" may be a transducer that converts a physical quantity to be measured into an electrical signal (e.g., a current signal or a voltage signal). The physical quantity may include, for example, but is not limited to, electromagnetic radiation (e.g., photons of infrared or visible light), a magnetic field, an electric field, pressure, force, temperature, current, or voltage.

[0049] A "training set" is a dataset that is fed into a machine learning program to enable it to learn a new task. The data can be labeled (inputs linked to predefined outcomes), unlabeled (input values ​​only), or a mixture of both. Once the program has adapted its rules and data structures to successfully perform a task (evolving into a "model"), it is fed a "test set," distinct from the training set, for similar analysis. As the model continues to build experience, part or all of the test set may become the training set for subsequent test sets.

[0050] Illustrative examples of the subject matter claimed below are disclosed. For the sake of clarity, not all features of an actual implementation are described in this specification for each example. It will be appreciated that in the development of any such actual implementation, numerous implementation-specific decisions may be made to achieve the developer's specific goals, such as complying with system-related and business-related constraints, which will vary from implementation to implementation. Moreover, it will be appreciated that such development work, while complex and time-consuming, is a routine undertaking for those of ordinary skill in the art having the benefit of this disclosure.

[0051] Figure 1 A physiological monitoring system according to some embodiments is schematically illustrated. First an overview of the system will be presented, which will then present each component and its options in more detail.

[0052] The physiological monitoring system 7 receives monitoring signals from the sensor 16 at the sensor interface 2 via the lead 17. The sensor interface 2 may include one or more ECG ports for receiving ECG signals. The sensor 16 is strategically positioned relative to the patient P to monitor physiological parameters of the patient P. The one or more sensors 16 may be ECG sensors. The patient P uses a cardiovascular device 14, which, when active, may affect the ECG signal sensed by the sensor 16. As non-limiting examples, the cardiovascular device 14 may include a pacemaker or an implantable cardioverter-defibrillator ("ICD").

[0053] The physiological monitoring system 7 may include modules 2-6 and 8-10 interconnected by an internal bus 5 to perform various functions. The modules may include hardware, software, or a combination thereof. The internal bus 5 may carry data signals, control signals, power, or a combination thereof between the various hardware modules. The sensor interface 2 receives physiological monitoring signals via leads 17, which may include ECG monitoring signals. One or more processors 3 may organize and analyze the physiological monitoring data received at the sensor interface 2, and may also control the general operation of the physiological monitoring system 7. A graphical user interface ("GUI") display 4 may be configured to display various patient data, sensor data, and hospital or patient care information. In addition, a user may query, control, and otherwise interact with the physiological monitoring system 7 through the GUI functionality of the GUI display 4. A power supply 9 provides power to at least some of the power consuming components within or connected to the physiological monitoring system 7.

[0054] The communication interface 6 may permit the physiological monitoring system 7 to communicate directly or indirectly with one or more external devices or networks (not shown). For example, the communication interface 6 may allow information from the GUI display 4 to be viewed on a remote monitor (not shown). Additionally or alternatively, the communication interface 6 may transmit warning messages or other signals from the warning system 10 to a device or destination of interest to the clinician.

[0055] The memory 8 may receive data from the processor(s) 3 or other sources, store it in one or more structures on a short-term or long-term basis, and allow the processor(s) 3 or other destinations to access the data. As a non-limiting example, a data structure 12 in the memory 8 may include a set of stored heartbeat classifiers, each heartbeat classifier being associated with a corresponding response via a link. Some responses may include actions to be performed by the warning system 10. As another non-limiting example, the memory 8 may store instructions 11 for execution by the processor(s) 3, such as organizing ECG data into an ECG data set, identifying the contribution of the cardiovascular device 14, accessing the data structure 12 to check for indicators of arrhythmia, activating the warning system 10 if the selected response 13 includes the warning system 10, and applying a machine learning model to update a personal normal ECG template for the patient P.

[0056] Each of the elements shown may be embodied in various suitable ways, all within the scope of the appended claims, as follows:

[0057] The conductive leads 17 coupling the sensors 16 to the sensor interface 2 may include any type of suitable signal carrying cable (e.g., single wire, twisted pair, coaxial cable, fiber optic). Additionally or alternatively, one or more sensors 16 may be wirelessly connected to the sensor interface 2, in which case the sensor interface 2 may include a wireless communication protocol using any suitable wireless protocol (such as connection, cellular data or near field communication protocols such as ))Circuitry that receives and / or transmits data.

[0058] The data signal received from the sensor 16 may be an analog signal. For example, the ECG data signal may be input to the sensor interface 2, which may include an ECG data acquisition circuit ( Figure 1 (not separately shown in the figure). The ECG data acquisition circuit may include amplification and filtering circuits and analog-to-digital ("A / D") circuits that convert analog signals into digital signals using amplification, filtering, and A / D conversion methods. If the ECG sensor 16 is a wireless sensor, the sensor interface 2 may receive data from the wireless communication module ( Figure 1 Thus, the sensor interface 2 may be configured to interact with one or more sensors 16 and receive sensor data therefrom.

[0059] As described above, a continuous ECG data set is acquired by receiving an ECG signal from one of the ECG sensors 16 via the lead 17 thereof and then processing the acquired ECG signal using the processor(s) 3 of the physiological monitoring device 7. The acquired ECG signal can be buffered as data in the memory 8 or processed in real time. Similarly, once the ECG data set is processed, it can be buffered as data in the memory 8 as needed or desired.

[0060] Thus, ECG signals and ECG data sets can be processed, displayed, or otherwise processed or manipulated in real time or near real time. The term "near real time" means as close to real time as possible within the computing resources permitted by a given embodiment. Some embodiments may also store ECG signals, impedance respiration signals, or other data in memory 8 for a longer period of time. For example, some embodiments may manage processing resources by storing data before pushing it out. Similarly, some embodiments may store data extracted from other locations.

[0061] The processor(s) 3 may be any suitable processor-based resource. They may be, but are not limited to, a central processing unit ("CPU"), a hardware microprocessor, a multi-core processor, a single-core processor, a field programmable gate array ("FPGA"), a controller, a microcontroller, an application specific integrated circuit ("ASIC"), a digital signal processor ("DSP"), or other similar processing device capable of executing any type of instructions, algorithms, or software for controlling the operation and performing the functions of the physiological monitoring system 7. In some embodiments, the processor(s) 3 may include a processor chipset, including, for example, but not limited to, one or more coprocessors. The processors may be centralized or distributed across two or more chips, boards, or other locations.

[0062] The GUI display 4 can interact with one or more input devices (not shown) such as a keyboard, a pointing or tracking device, a microphone, or a camera. The display can include a liquid crystal display ("LCD"), a cathode ray tube ("CRT") display, a thin film transistor ("TFT") display, a light emitting diode ("LED") display, a high definition ("HD") display, or other similar display devices that may include touch screen capabilities. The GUI display 4 can provide a means for inputting instructions or information directly into the physiological monitoring system 7. The displayed patient information may, for example, relate to a measured physiological parameter of the patient P (e.g., an ECG reading).

[0063] The communication interface 6 may allow the physiological monitoring system 7 to communicate directly or indirectly with one or more computing networks and devices, workstations, consoles, computers, monitoring equipment, warning systems, and / or mobile devices (e.g., mobile phones, tablets, or other handheld display devices). The communication interface 6 may include various network cards, interfaces, communication channels, clouds, antennas, and / or circuits to allow wired and wireless communication with such computing networks and devices. The communication interface 6 may be used to implement communication with such computing networks and devices. connection, cellular network connection and / or Example wireless communication connections implemented using the communication interface 6 include those according to, but not limited to, the IEEE 802.11 protocol, the Radio Frequency for Consumer Electronics ("RF4CE") protocol, and / or the IEEE 802.15.4 protocol (e.g., Essentially, any wireless communication protocol can be used.

[0064] Additionally, the communication interface 6 can use, for example, a universal serial bus ("USB") connection or other communication protocol interface to permit direct (i.e., device-to-device) communication (e.g., messaging, handshaking, etc.) to and from the physiological monitoring system 7. The communication interface 6 can also permit direct device-to-device connection with other devices (such as a tablet, computer, or similar electronic device) or external storage or memory.

[0065] The memory 8 can be a single memory device or a group of co-located or distributed memory devices. Locations can include, but are not limited to, being on-chip with the processor or any other functional module of the physiological monitoring system 7, or being off-chip in one or more dedicated memory modules. The memory 8 can be removable or installed, or a combination thereof, and can be volatile or non-volatile, or a combination thereof. The memory 8 can be, for example, RAM, a memory buffer, a hard drive, a database, EPROM, EEPROM, ROM, flash memory, a hard disk, or any other non-transitory computer-readable medium capable of storing data and instructions.

[0066] The power supply 9 can include a self-contained power source (such as a battery pack) and / or include an interface for powering the physiological monitoring system 7 through a power outlet, directly or through a monitor stand or other power connection system. The power supply 9 can also be a removable rechargeable battery to allow replacement. A backup power source (such as an uninterruptible power supply) or a built-in backup power source (such as a backup battery or supercapacitor) can be connected to provide continuous power to the physiological monitoring system 7 during power failures, battery replacement, or other power outages.

[0067] The physiological monitoring system 7 can optionally be deployed as part of a larger medical facility management system, whose functionality may include, but is not necessarily limited to, physiological monitoring. The medical facility management system may include a set of physical and virtual servers and client devices connected in a network or cloud. The physiological monitoring system 7 may exchange information with the medical facility management system via any suitable wired or wireless communication link.

[0068] Figure 2 A patient undergoing ECG monitoring while using a cardiovascular device is schematically illustrated. When the muscles surrounding the chambers of the heart H of the patient P electrically depolarize, the heart H can generate intrinsic ECG signals 231, which trigger mechanical contractions that push blood in and out of the chambers. The resulting electrical signals 231 can propagate outward from the heart H until they reach the skin surface S, where they can be detected by sensors (such as monitoring sensors 16). The monitoring sensors 16 can contact the skin surface S through a layer of conductive gel (not shown) to improve signal reception.

[0069] Meanwhile, the cardiovascular device 14 of the patient P may include a pulse generator module 244, pacing leads and (one or more) electrodes 234, and some type of sensing capability 224. The pulse generator module 244 houses the pulse generation hardware, a storage medium with control software and firmware, and a power source (typically a battery). When the sensing capability 224 senses an arrhythmia, or in other circumstances, such as when the device's programming prompts it to pace or otherwise stimulate the myocardium of the heart H, the pulse generator 244 sends a corrective electrical signal 232 to the heart H via the pacing leads and electrodes 234.

[0070] Although conceptual and not drawn to scale, the cardiovascular device 14 shown here schematically represents an implantable transvenous pacemaker, ICD, etc. The cardiovascular device 14 can be fully implanted in the body, with leads and electrodes 234 passed through a central vein into the interior of the heart H. There, the electrodes 234 directly stimulate electrical activity of the inner walls of one or more heart chambers, and the sensing mechanism 234 directly senses electrical activity of the inner walls of one or more heart chambers.

[0071] Cardiovascular devices can be configured in a variety of different ways to address the needs of different patients. Some devices are intended to be installed temporarily in an emergency, while others are long-term or permanent. In some cardiovascular devices 14, a single pair of electrical contacts performs both sensing and pacing functions. In other devices, these functions may be performed by different subassemblies. Pacing can be unipolar or bipolar. Pacing pulses can be delivered to a single chamber of the heart H (usually the right ventricle) or to two chambers (the right atrium and right ventricle in a dual-chamber device, or the right ventricle and left ventricle in a biventricular device) . Dual-chamber devices can operate in various modes, such as AS-VP (atrial sensing, ventricular pacing), AP-VS (atrial pacing, ventricular sensing), AP-VP (atrial pacing, ventricular pacing), and fusion beats (ventricular pacing, which is timed differently than in other VP situations).

[0072] Biventricular cardiovascular devices can be used for cardiac resynchronization therapy ("CRT"). In one example of a CRT device, a left ventricular ("LV") pacing lead can be added to a standard pacemaker or ICD that already has a right ventricular ("RV") pacing lead and an optional right atrial lead. Unlike some other cardiovascular devices that activate only when a need is detected, CRT is generally most effective in "always on" mode: that is, when the ventricles are paced as close to 100% of the time as possible.

[0073] Leadless cardiovascular devices are integrated assemblies that are inserted completely into the chambers of the heart H, usually through a vein. Subcutaneous cardiovascular devices can be implanted subcutaneously (e.g., on the side of the chest, under the armpit), with electrodes running along the sternum rather than impacting the heart itself.

[0074] For either of these configurations, as the correction signal propagates from the electrode 234 through the muscle of the heart H, it continues to propagate as the device signal 232 to the external skin surface S. Thus, when the cardiovascular device 14 is active, the ECG monitoring sensor 16 receives both the stimulation artifact 232 and the intrinsic ECG signal 231 simultaneously.

[0075] The ECG signals (one or more) 231 and 232 received by the monitoring sensor 16 are transmitted to the physiological monitoring system 7 (specifically, the ECG port of its sensor interface, not shown in this figure) via the sensor leads 17 (as described above, which can be any suitable cable or wireless connection). Their combination (device modulated signal 233) can then be displayed on the GUI display 4 and analyzed by one or more processors (not shown in this figure) of the physiological monitoring system.

[0076] Figure 3This paper presents examples of selected types of ECG traces and their structural features that can be analyzed using published methods. It introduces some intrinsic ECG features to facilitate a later discussion of how to identify arrhythmias in normal ECG datasets and their patient-to-patient variability. This series of plots represents a dataset from a 12-lead ECG channel, a diagnostic standard in cardiology. Each plot in the series plots voltage on the vertical axis and time on the horizontal axis. The ECG is an indirect indicator of myocardial activity because the heart muscle contracts in response to electrical depolarization of its cells.

[0077] Dataset I is the upper lateral bipolar ECG measured between sensors on the patient's wrist (or shoulder). Dataset II is the lower left bipolar ECG measured between sensors on the patient's left wrist (or shoulder) and left ankle. Dataset III is the lower right bipolar ECG measured between sensors on the patient's right wrist (or shoulder) and left ankle (or lower abdomen).

[0078] The datasets aVR, aVL, and aVF may be derived from leads I, II, and III rather than being measured individually.

[0079] Data set V1 is a chest lead ECG measured by a sensor on the 4th intercostal space on the right side of the patient (between the 4th and 5th ribs on the right side). Data set V2 is a chest lead ECG measured by a sensor on the 4th intercostal space on the left side of the patient. Data set V3 is a chest lead ECG measured by a sensor on the 5th rib on the left side of the patient. Data set V4 is a chest lead ECG measured by a sensor at the end of the 5th intercostal space on the left side of the patient closest to the sternum (breastbone). Data set V5 is a chest lead ECG measured by a sensor at the end of the 5th intercostal space on the left side of the patient closest to the axilla (armpit). Data set V6 is a chest lead ECG measured by a sensor on the mid-axillary line (lateral center line of the axilla) on the left side of the patient.

[0080] All datasets I-V6 were collected from the same patient at the same time. They differ from each other for various reasons. Different lead placements create additional differences. Therefore, the analysis of each dataset can be informed by the context provided by the original lead placement.

[0081] A magnified view of the circled area of ​​Dataset II shows a classic view of the characteristics, features, or "waves" (deviations from baseline B) visible in the ECG of an example single heartbeat. Clinicians can extract information about a patient's heart health based on the presence, amplitude, width, shape, and spacing of the waves.

[0082] The small P wave is produced by electrical depolarization of the atria and originates in the sinoatrial ("SA") node. If one or both atria are paced in the "AP" mode of a cardiovascular device, the pacing pulse (a narrow spike, not shown in this figure) may appear in or next to the P wave on the ECG.

[0083] During the baseline pause following the P wave, atrial depolarization (ideally) spreads to the ventricles through the atrioventricular node (or AV node). The resulting electrical depolarization of the ventricles produces a "QRS complex," a series of waves that can include a small, downward Q wave, a tall, upward R wave, and a small, downward S wave. The beginning of the Q wave is called the "QRS onset." The QRS complex is one of the most informative parts of an ECG, containing clues that can help diagnose a variety of heart problems. Conventionally, even though neither a Q wave or an S wave may be present in the ECG, whichever wave is present is referred to as a "QRS complex." If one or both ventricles are paced by a cardiovascular device in "VP" mode, the pacing pulse may appear at or near the QRS onset.

[0084] The small T wave following the QRS complex is generated by ventricular repolarization. The U wave following the T wave is typically much smaller than the T wave and often absent altogether. Its presence sometimes indicates underlying pathology and sometimes not. Further complicating matters, in some patients, the T wave may split into two peaks, with the second peak being mistaken for a U wave. The behavior of the U wave during changes in heart rate is associated with both tachycardia and bradycardia (fast and slow arrhythmias, respectively).

[0085] In a heart without arrhythmias, the QRS complex may be approximately 80-100 milliseconds long, and the frequency of the P wave and the frequency of the QRS complex are equal throughout the length of the ECG data set. Other commonly assessed ECG dimensions include the PQ segment S PQ 、PQ interval I PQ , ST interval I ST and ST segment S ST Preferably, an algorithm analyzing an ECG dataset can detect and evaluate many of the features examined by expert human readers and potentially extract some information that is inaccessible to the human eye.

[0086] Figure 4 This is a general overview of the analysis model according to some embodiments. The introduction of the entire process 400 can provide readers with a sense of orientation so that readers can traverse the more detailed disclosure content that will be provided later.

[0087] While monitoring a patient in process 401, an ECG dataset can be collected and submitted as test data to an analysis model in process 402. The model determines whether the patient has a currently active cardiovascular device at branch point 403 by identifying SA and any other ECG characteristics attributable to such a device. In some embodiments, this is done early in the analysis, or even as a first step before further classification, clustering, or other manipulation of the ECG dataset.

[0088] If cardiovascular device activity is detected at branch point 403, the ECG dataset is treated as device modulated in process 404. In some embodiments, the device modulated dataset can be compared primarily or exclusively to a device modulation template based on a device-modulated heartbeat classifier. If a device modulated rhythm within an acceptable margin is detected at branch point 405, monitoring is continued and returns to process 401. However, if a device modulated tachycardia or other arrhythmia is detected at branch point 405, the system executes a stored response in process 406. The response is linked to the detected arrhythmia via a data structure and corresponds to its predicted impact on the patient. For example, if the patient may be in severe distress, an emergency alert can be sent to a clinician, or if the patient may be uncomfortable but not in immediate danger, a non-emergency warning can be sent.

[0089] If no cardiovascular device activity is detected at branch point 403, the ECG data set is considered intrinsic (affected only by the patient's heart action) in process 407. In some embodiments, the intrinsic data set can be compared primarily or exclusively to an intrinsic template based on an intrinsic heartbeat classifier. If an intrinsic rhythm within an acceptable margin is detected at branch point 408, monitoring is continued and process 401 is returned to. However, if an intrinsic tachycardia or other arrhythmia is detected at branch point 408, the system executes a stored response in process 406. The response is linked to the detected arrhythmia via a data structure and corresponds to its predicted impact on the patient. For example, if the patient may be in severe distress, an emergency alert can be sent to a clinician, or if the patient may be feeling uncomfortable but is not in immediate danger, a non-emergency warning can be sent.

[0090] In some embodiments, each collected ECG dataset can be converted into training data to help analyze subsequently collected test datasets and continue to build the expertise of the model. These options will be described in more detail later in this disclosure.

[0091] Figure 5 is a conceptual diagram of a memory configured to implement some embodiments of the disclosed method. Figure 1The memory 8 described in the embodiment of the present invention may be a monolithic or distributed, integrated or removable, part of a system on a chip (SoC), implemented as part of a virtual machine (VM), or any other suitable form of non-transitory machine-readable data storage device. The memory 8 may communicate with one or more processors, controllers, transceivers, or other electronic components and devices via an internal bus 5, a wireless link 501, or any suitable type of permanent, removable, or switchable communication link.

[0092] The memory 8 includes instructions 11 that are accessed and executed by one or more processors (not shown). The instructions 11 may include one or more machine learning models and one or more explicit programs. The machine learning model may use unsupervised learning, supervised learning, semi-supervised learning, or a combination thereof.

[0093] The memory 8 also includes a data structure 12 that provides data accessed by one or more processors (not shown) during execution of the instructions 11. A knowledge-based heartbeat classifier 502 enables a model to identify specific arrhythmias 503 in an ECG dataset via an explicit algorithm or supervised machine learning. Each specific arrhythmia 503 is associated with a response 505 via an information link 504. Each response 505 is based on the expected impact of the specific arrhythmia 503 on the patient (e.g., an urgent alert in case of pain or danger, a non-urgent warning of discomfort, a log entry or non-urgent message if the condition corrects itself in a short period of time). Any suitable type of link can be used, such as a lookup table or proximity.

[0094] The data structure 12 also includes templates 506, some of which can be created using unsupervised or semi-supervised machine learning. For example, ECGs collected for the same patient that exhibit an acceptable heart rhythm (e.g., sinus beat or effective pacemaker pacing) can be used to create and repeatedly update a normal template for the patient. A flag 508 identifying salient features of the ECG (such as SA and heart rhythm) and an identifier for the patient can be associated with the template 506 using any suitable type of link 507.

[0095] In some embodiments, data structure 12 has two separate substructures that separate intrinsic data from device-modulated data. That is, incoming extrinsic ECG data can be compared to an extrinsic template and analyzed using an extrinsic heartbeat classifier, while incoming device-modulated ECG data can be compared to a device-modulated template and analyzed using a device-modulated heartbeat classifier.

[0096] Figure 6This is a flow chart of instructions for a normal template learning session according to some published methods. Given the differences between patients and the varying activity of cardiovascular devices, a patient's "normal" ECG may not be constant. For example, the ECG may change with the patient's physical or emotional state, respond to medications, or adapt to the cardiovascular device. To address this diversity, complexity, and variability, machine learning can be used to learn a personalized normal template, which increases in scope and flexibility as more ECG data is collected.

[0097] In some embodiments, the machine learning that creates and updates personalized normal templates can be unsupervised (trained from unlabeled data given only input values ​​to discover hidden patterns and groupings) or semi-supervised (trained from a mixture of labeled and unlabeled data). The expected position of SA can be used for feature selection to reduce generalization error and random noise.

[0098] At starting point 601 of process 600, a processor may collect an ECG data set from a patient in process 602. The test data set includes intrinsic ECG signals generated by the patient's heart, and it may or may not include one or more SAs contributed by a cardiovascular device used by the patient.

[0099] In processes 603a and 603b, the model extracts heartbeat features and waveforms from the ECG dataset. While the medical literature varies on the parameters it refers to as "features" compared to the "waveform" components, the model can extract parameters including, but not limited to, beat area, beat width, beat amplitude, QRS polarity, P wave presence, temporal coupling of the P wave to the QRS complex, RR interval, ST segment elevation and depression, T wave abnormalities, pathological Q waves, RR interval, ST interval, PR interval, and QT interval.

[0100] If the signature and waveform analysis reveals no SA attributable to the cardiovascular device at branch point 604, the results are compared to stored intrinsic templates in process 624. If, instead, one or more SAs reveal activity of the cardiovascular device, a corresponding flag (e.g., "pacing") can be attached to the dataset in process 614. From there, the model analyzes the alignment of SA with intrinsic heartbeat features (e.g., whether SA precedes or follows the onset of the QRS) in process 634. In process 654, the results are compared to any stored templates that are also marked as device modulation and have similar SA alignment.

[0101] Regardless of whether comparison process 624 or 654 is performed, if a matching stored template is found in the data structure at branch point 605, the new data is added to update the matching stored template in process 615. In other words, the most recently analyzed test data becomes the training data used to analyze subsequently collected data. If no matching template is found, a new template is allocated and stored in process 625.

[0102] After updating the matching template in process 615 or assigning and storing a new template in process 625, the model determines whether the current learning cycle is complete at branch point 606 by applying criteria such as beat-to-beat correlation. For example, the algorithm may determine that the learning cycle is complete after 15 beats with correlation above a predetermined threshold. If the learning cycle is not complete at branch point 606, the model returns to process 602 to collect more data. If it is complete, the model verifies and stores the new or updated template(s) at process 616, determines the heart rhythm at process 636, and then ends the learning session at endpoint 607.

[0103] Each newly monitored patient can start with a blank template onto which its characteristics are consolidated as successive ECG data sets are collected and analyzed. For example, a patient's normal ECG may exhibit one of the following common QRS patterns. If a cardiovascular device is sensing one or more cardiac chambers but is not currently pacing any of them, a "sinus beat" pattern may be primarily intrinsic. An "AP-VS" pattern may occur when one or both atria are paced and one or both ventricles are sensed. An "AS-VP" pattern may occur when one or both atria are sensed and one or both ventricles are paced. An "AP-VP" pattern may occur when at least one atrium and at least one ventricle are paced. The model can identify any of these patterns by its characteristics and / or waveform, and in some cases can identify the brand or model of the cardiovascular device by its "signature" SA.

[0104] Other variations are possible; there is considerable variability between patients’ intrinsic heart beats, and the customization of cardiovascular devices has become increasingly complex as technology advances. Furthermore, a patient whose device activates only when needed might have both an intrinsic normal template (learned when the device is inactive) and a device-modulated normal template (learned when the device is active). Within any of these categories, the data might form separate clusters based on one variable or the other, which could ultimately become distinct templates.

[0105] Figure 7 is a decision tree for ECG diagnostic analysis using a knowledge-based heartbeat classifier according to some disclosed methods. Method 700 identifies conditions revealed by an ECG, such as arrhythmias.

[0106] Process 701 collects ECG test data in the process of monitoring patient physiological parameters. Optionally, the ECG data set can be a 20-second trace or other suitable time length. The data set can be collected by any suitable number of leads (such as 1, 2, 5, 6 or 12).

[0107] In some embodiments, the diagnostic method 700 is paused by default during a normal template learning cycle, so the branch point 702 determines whether a normal template learning cycle is currently ongoing. If so, it brings the new test data back to Figure 6 6. If the model has completed the learning cycle, it continues the diagnostic process by comparing the new data to the template based heartbeat classifier in process 704. The model determines whether the new ECG data set is intrinsic or device modulated, and whether it is normal (sinus rhythm or successfully paced) or arrhythmic (potentially harmful).

[0108] If the ECG is an intrinsic sinus beat at an acceptable rate, it is added to the matching intrinsic template or, if no match is found in the data structure, it is assigned to a new template in process 705. If the ECG is "sinusoidal" (where one or more SAs indicate an acceptable rate for successful device operation), it is labeled as, for example, "stimulated" or "paced" in process 706. (This label should not be confused with the "flag" in the template learning process 600. The label represents the output value from the model, while the flag is the input value that describes the characteristics of such a label in the dataset.) If no fusion beat is detected at branch point 707, the dataset can be added to the existing matching stimulation template in process 708, or assigned to a new template if no match is found.

[0109] If the new ECG data set is stimulated but arrhythmic, it is marked as "stimulated" (or "paced," or any other useful descriptor) in process 709. Thereafter, other distinctions are made to distinguish specific types of arrhythmias. At branch point 710, the timing of the SA relative to the QRS complex is checked. In many cases, the SA preferably precedes the onset of the QRS. These ECG data sets are marked as "correctly timed" (or other useful descriptors) in process 711 and then checked at branch point 712 to see if the arrhythmia corrects itself within a short period of time (e.g., less than 20 seconds). Figure 8 shown.

[0110] If yes, then the arrhythmia has been corrected by the cardiovascular device and the patient does not need attention (unless the recurrence frequency is higher than expected, which can also be tracked by the model). In the example, this result triggers a low priority response (such as a non-urgent message) in process 713. If the arrhythmia persists, it may be as follows Figure 10 The illustrated infinitely cycling tachycardia, which is typically uncomfortable but not painful, triggers a real-time, non-emergency alert in process 714 so that a clinician can treat the patient's discomfort.

[0111] Returning to branch point 710, if SA occurs after the onset of QRS, it may indicate a more serious condition, such as Figure 11 Pacemaker-mediated conduction block tachycardia is shown. This result triggers an emergency alarm in process 715 for immediate clinician intervention.

[0112] Another possible result of the comparison process 704 is an intrinsic arrhythmia with no SA in the trace. This could be an intrinsic tachycardia that is not corrected by the cardiovascular device, such as Figure 9 As shown. The patient does not have cardiovascular equipment, or the equipment they have is not functioning properly. Since this is also a very serious condition, this result will also trigger an emergency alarm in process 715 for immediate clinician intervention.

[0113] The order of these processes shown is intended to be a non-limiting example. Without exceeding the scope of the claims, the order of some processes may be exchanged and / or some processes may be performed in parallel.

[0114] Figure 8 Figure 800 is an example ECG trace of a cardiovascular device successfully correcting an intrinsic tachycardia. In ECG 800, intrinsic ventricular tachycardia can be identified by the shape, rate, and amplitude of pulse train 801a. The cardiovascular device detects the tachycardia and responds with a series of pacing pulses 802a, which appear to be superimposed on the intrinsic pulse train. Later, pulse train 801b still shows signs of tachycardia. The cardiovascular device may detect the persistence of the tachycardia, or its programmed routine may be to follow its first pacing pulse train 802a with a second pulse train 802b. Pacing pulse 802b effectively stops the tachycardia. The subsequent pulse train 803 is closer to normal in rate, amplitude, and shape. The time required for a cardiovascular device to correct an intrinsic tachycardia may vary depending on the manufacturer's algorithm, but typically it will meet the American Association for the Advancement of Medical Instrumentation ("AAMI") standard of 10 seconds or less.

[0115] The pacing pulses 802a, 802b correct the tachycardia while causing little or no pain to the patient. If the physiological monitoring system alerts the clinician immediately at the onset of the tachycardia, the clinician can intervene with forceful CPR and / or a strong shock from a paddle-type defibrillator. In some cases, such measures can provide benefits that outweigh the risk of collateral damage. Overtreatment of self-terminating ventricular tachycardia is more common when the cardiovascular device begins applying pacing pulses immediately rather than, for example, waiting 6-15 seconds and applying a stronger shock. However, if the cardiovascular device is effectively implemented, such forceful interventions may reduce the patient's quality of life and cause additional wear and tear on the cardiovascular device. In addition, emergency responses triggered by false alarms are costly to the clinic, the patient, and the healthcare system.

[0116] In some embodiments, a heartbeat classifier based in part on data similar to ECG 800 can be included in the data structure of the analysis model. Since the pacemaker has already corrected the tachycardia, clinician intervention is generally unnecessary, and the linked response can include a log entry, a file annotation, or a low-priority message sent to the clinician. Optionally, the model can record the number and frequency of these events: while each event alone may not be of concern to the clinician, a number or frequency of events exceeding a predetermined threshold may require closer monitoring or additional testing of the patient.

[0117] Figure 9 is an example ECG trace of an intrinsic tachycardia that a cardiovascular device attempted to correct but failed. ECG 900 shows a pulse train 901A that can be identified as an intrinsic ventricular tachycardia. Spike 902 could be a single pacing pulse from the cardiovascular device attempting to correct the arrhythmia, or it could be some other pulse. In any case, it is neither a rhythmic pacing pulse train from a well-functioning pacemaker nor a high-amplitude pulse from an ICD. Therefore, the cardiovascular device either did not detect tachycardia 901a, or it attempted to respond but failed for some reason. After the single pacing pulse 902 (or other spike of unknown origin), tachycardia 901a continues unabated, as shown in 901B.

[0118] In some embodiments, the present invention provides a method for the present invention to provide ...

[0119] Figure 10 Figure 1 is an example ECG trace of an endless loop tachycardia ("ELT"), a type of pacemaker-mediated tachycardia. The term "pacemaker-mediated tachycardia" can be used to describe a variety of situations in which a cardiovascular device paces the ventricles at an inappropriately high rate. ELT affects users of dual-chamber pacemakers and similar cardiovascular devices that provide inhibition of ventricular pacing. ELT is generally associated with retrograde ventricular ("AV") conduction (current flowing in the reverse direction through or near the AV node).

[0120] After observing that some patients were sensitive to right ventricular pacing to the point of increasing risk for atrial fibrillation and / or heart failure, cardiovascular device manufacturers introduced algorithms to inhibit RV pacing under certain circumstances, such as when the heart rate approaches the device's programmed upper rate limit. Some of these algorithms had the unintended consequence of causing retrograde VA conduction, leading to ELT. Symptoms of ELT may include palpitations, mild headache, syncope, or chest pain—uncomfortable, but not as severe or dangerous as the atrial fibrillation that the inhibition algorithms were developed to prevent.

[0121] ELT can be classified as a reentrant tachycardia, meaning that an extra conduction circuit is formed in the heart that may discharge between normal heartbeats, causing premature or extra heartbeats that become a sustained tachycardia. The cardiovascular device may form the antegrade limb of the reentrant circuit, and the patient's native AV node (or alternatively, an accessory pathway) forms the retrograde limb. Atrial sensing, ventricular pacing ("AS-VP") cardiovascular devices are most commonly associated with EVL, although similar behavior has also been reported in atrial pacing, ventricular pacing ("AP-VP") cardiovascular devices.

[0122] ECG data set 1000 illustrates the onset of ELT. Signal 1003 includes the atrial sensing and ventricular pacing cardiovascular device actions. At this point, the heart rate is within acceptable limits. The atrial sensing coincides with the onset of the P wave (a very small bump after a long pause), and the ventricular pacing pulse coincides with the onset of the QRS complex (although there is no Q wave in this ECG, which is an unusual phenomenon). Signal 1003 has a normal PQ interval I PQ (approximately 180 milliseconds). At 1004, there is a PQ interval I PQ * Abnormally long (approximately 280 milliseconds) heartbeat. Afterwards, ELT 1001 begins.

[0123] The heartbeat classifier in the tachycardia detection data structure can be based in part on traces similar to ECG dataset 1000 to enable the model to identify cases of ELT. The linked response may include an immediate non-urgent warning to the clinician to address the patient's potential discomfort. Alternatively, if the model has identified the patient as pacing dependent (because their normal template is entirely or almost entirely modulated by the device), the response can be upgraded from a warning to an alarm.

[0124] Figure 11 This is an example ECG trace of a pacemaker-mediated conduction block ventricular tachycardia. This arrhythmia is unusual; as of this writing, only a few clinicians have encountered it. On the other hand, it can pose serious risks to pacemaker-dependent patients. This arrhythmia can be managed by optimizing cardiovascular device programming and, in some cases, ablating scar tissue that has formed on the myocardium. However, a correct diagnosis is essential.

[0125] In some patients, scarring or other electrically abnormal tissue (sometimes called "refractory myocardium" because it conducts electricity differently than normal myocardium) may develop in the heart. In mild cases, depolarization and repolarization currents may be able to bypass the refractory area, but this straining approach may be less effective if additional refractory tissue or ectopic circuits (abnormal current pathways that conduct electricity asynchronously with the sinus node and / or pacemaker) develop.

[0126] When a cardiovascular device such as a pacemaker or ICD is implanted, some patients may develop scar tissue around the point where the device leads enter the heart muscle. Electrolyte abnormalities or stereotactic radiation may also create or enlarge a refractory zone at this critical interface. If the refractory zone prevents excessive current from flowing out of the device, ventricular tachycardia (ventricular tachycardia) may develop (called "pacemaker exit block-mediated ventricular tachycardia"). Patients may experience multiple episodes of tachycardia per day, with each episode elevating the heart rate to 150 beats per minute.

[0127] Particularly for pacemaker-dependent patients, conduction block tachycardias can be both distressing and dangerous, necessitating an emergency response. An effective response, which may include ablation of scar tissue, depends on a timely and accurate diagnosis. The disclosed machine learning model can identify any pacing-dependent patient by the absence of a fully intrinsic ECG in the normal template and optionally flag them for close monitoring and escalated response. Additionally, one or more conduction block heartbeat classifiers can be based in part on an ECG trace (such as trace 1100), with the linked response potentially including issuing an immediate emergency alert to one or more clinicians.

[0128] As improvements in healthcare extend patient lifespans, more patients can use cardiovascular devices to extend their lives, and some conditions now considered rare may become more common. Some disclosed embodiments may be particularly beneficial for diagnosing rare or emergency conditions (such as pacemaker block tachycardia). Even if few clinicians have encountered a certain condition in their reading or experience, software developers can upload the relevant heartbeat classifier to the network once it has been fully tested and approved. Any patient monitoring system connected to the network can then add the new heartbeat classifier as an update to their machine learning model. Optionally, the response to the detection of an abnormal condition can include a link to a relevant peer-reviewed research paper so that local clinicians can quickly understand how to proceed once the patient's condition has stabilized.

[0129] Embodiments of the disclosed method include collecting electrocardiogram ("ECG") data comprising at least one of an intrinsic ECG signal from a patient and stimulation artifacts from a cardiovascular device used by the patient; creating a normal template from a continuous ECG data set through unsupervised machine learning; detecting the presence of a specific arrhythmia in the ECG data by applying at least one knowledge-based heartbeat classifier; and performing a response on the patient that corresponds to an expected effect of the specific arrhythmia.

[0130] In some embodiments, the method further comprises storing the normal template and the heartbeat classifier in a data structure, wherein the data structure links a specific arrhythmia to a corresponding response. In some embodiments, the data structure stores the intrinsic normal template separately from the normal template with stimulation artifacts.

[0131] In some embodiments, the specific arrhythmia includes tachycardia.

[0132] In some embodiments, unsupervised machine learning continues until a threshold correlation is detected between consecutive ECG data sets.

[0133] In some embodiments, the particular arrhythmia comprises a transient intrinsic tachycardia that is corrected by the cardiovascular device within a predetermined time limit, and the corresponding response does not include an emergency alarm or urgent warning requesting a clinician's attention.

[0134] In some embodiments, the particular arrhythmia comprises a pacemaker-mediated endless-cycling tachycardia, and the corresponding response comprises an urgent alert requesting a clinician's prompt attention.

[0135] In some embodiments, the particular arrhythmia comprises a pacemaker-mediated conduction block tachycardia, and the corresponding response comprises an emergency alert requesting immediate intervention by a clinician.

[0136] In some embodiments, the particular arrhythmia comprises a persistent uncorrected intrinsic tachycardia, and the corresponding response comprises an emergency alert requesting immediate intervention by a clinician.

[0137] In some embodiments, the stimulation artifact includes a pacing pulse, and the method includes determining whether the cardiovascular device performs one of the following: (1) atrial pacing, ventricular sensing ("AP-VS"); (2) atrial sensing, ventricular pacing before QRS onset ("AS-VP"); (3) atrial pacing, ventricular pacing before QRS onset ("AP-VP"); and (4) ventricular pacing during the QRS ("fusion beat").

[0138] In some embodiments, the method further comprises inferring information about the structure, programming, or origin of the cardiovascular device based on properties of the stimulation artifact(s) in the ECG data.

[0139] In some embodiments, the method further includes updating the normal template with newly collected ECG test data from the same patient by converting the analyzed test data into training data.

[0140] An embodiment of the disclosed physiological monitoring system includes at least one electrocardiogram ("ECG") port, a processor communicatively coupled to the ECG port, and a memory communicatively coupled to the ECG port and encoded with a data structure. The data structure includes a heartbeat classifier that detects a specific arrhythmia in the ECG data, a corresponding response linked to a predefined arrhythmia, and a normal template learned from the ECG data. The memory is also encoded with a set of instructions that, when executed by the processor, causes the processor to: collect an ECG data set through the ECG port, the data set including at least one of an intrinsic ECG signal from a patient and a stimulation artifact from a cardiovascular device of the patient; compare the ECG data set to a normal template; and, if the ECG data set matches the normal template, update the normal template using the ECG data set.

[0141] In some embodiments, the physiological monitoring system further includes an ECG monitoring sensor communicatively coupled to the ECG port.

[0142] In some embodiments, the cardiovascular device comprises a pacemaker or an implantable cardioverter-defibrillator ("ICD").

[0143] In some embodiments, the physiological monitoring system further comprises a communication interface for coupling to an external network.

[0144] Embodiments of the disclosed non-transitory computer-readable storage medium include instructions that, when executed, cause a processor to: collect electrocardiogram (“ECG”) data comprising at least one of an intrinsic ECG signal from a patient and stimulation artifacts from a cardiovascular device used by the patient; create a normal template from a continuous set of ECG data through unsupervised machine learning; detect the presence of a specific arrhythmia in the ECG data by applying at least one knowledge-based heartbeat classifier; and execute a response corresponding to an expected effect of the specific arrhythmia on the patient.

[0145] The foregoing description and accompanying drawings generally describe a method for analyzing an electrocardiogram (ECG) of a user of a cardiovascular device that affects an ECG. The foregoing description and accompanying drawings generally describe a physiological monitoring system for analyzing an ECG of a user of a cardiovascular device that affects an ECG. The foregoing description and accompanying drawings generally describe a non-transitory computer-readable storage medium containing instructions that, when executed, cause a processor to detect a cardiac arrhythmia in the ECG of a user of a cardiovascular device that affects an ECG.

[0146] Unless otherwise defined, all terms used herein (including technical and / or scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. In addition, unless otherwise defined, all terms defined in general dictionaries shall not be over-interpreted.

[0147] As used herein, the articles "a" or "an" are intended to have their ordinary meaning in the patent arts, ie, "one or more."

[0148] Unless expressly specified otherwise, when the term "about" is applied to a value, the term "about" generally means within the tolerance of the equipment used to produce the value, or in some examples, means plus or minus 10%, or plus or minus 5%, or plus or minus 1%.

[0149] In one or more embodiments, use of the phrases "can," "capable of," "operable to," or "configured to" refers to some device, logic, hardware, and / or element being designed in such a manner to enable the device, logic, hardware, and / or element to be used in the specified manner.

[0150] When an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element, or intervening elements may be present. Conversely, when an element is referred to as being "directly connected" or "directly coupled" to another element, there are no intervening elements. Other words used to describe the relationship between elements should be interpreted in a similar manner (e.g., "between" versus "directly between," "adjacent" versus "directly adjacent," etc.).

[0151] Expressions such as "include" and "may include" that may be used in this disclosure indicate the presence of the disclosed functions, operations, and constituent elements, and do not limit the presence of one or more additional functions, operations, and constituent elements. In this disclosure, terms such as "include" and / or "have" may be interpreted as indicating a certain characteristic, number, operation, constituent element, component, or combination thereof, but should not be interpreted as excluding the possibility of the presence or addition of one or more other characteristics, numbers, operations, constituent elements, components, or combinations.

[0152] The subject matter of the present disclosure is provided as an example of an apparatus, system, method, circuit, and program for implementing the features described in the present disclosure. However, in addition to the features described above, additional features or variations are also contemplated. It is contemplated that any emerging technology may be used to implement the components and functions of the present disclosure, which may replace any of the technologies implemented above.

[0153] The present invention is described in detail with reference to the accompanying drawings, and detailed descriptions are provided to assist in a comprehensive understanding of the various exemplary embodiments of the present invention. The functions and arrangements of the elements discussed may be changed without departing from the spirit and scope of the present invention. Various embodiments may omit, replace, or add various programs or components as appropriate. For example, features described with respect to certain embodiments may be combined into other embodiments. In addition, descriptions of well-known functions and configurations may be omitted for clarity and brevity. Therefore, those skilled in the art will recognize that various changes and modifications may be made to the examples described herein without departing from the spirit and scope of the present invention.

[0154] Therefore, various modifications of the present disclosure will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of the present disclosure. Throughout this disclosure, the terms "example," "examples," or "exemplary" indicate examples or instances and do not imply or require any preference for the examples mentioned. Therefore, the present disclosure is not limited to the examples and designs described herein, but should be accorded the widest scope consistent with the principles and novel features disclosed.

Claims

1. A method comprising: collecting electrocardiogram ("ECG") data comprising at least one of an intrinsic ECG signal from a patient and stimulation artifacts from a cardiovascular device used by the patient; Create normal templates from continuous ECG datasets through unsupervised machine learning; detecting the presence of a specific arrhythmia in the ECG data by applying at least one knowledge-based heartbeat classifier; as well as A response is executed corresponding to the expected effect of the particular arrhythmia on the patient.

2. The method according to claim 1, further comprising storing the normal template and the heartbeat classifier in a data structure, wherein The data structure links the particular arrhythmia to a corresponding response.

3. The method according to claim 2, wherein: The data structure stores the inherent normal template and the normal template with stimulation artifacts separately.

4. The method according to claim 1, wherein The specific arrhythmias include tachycardia.

5. The method according to claim 1, wherein The unsupervised machine learning continues until a threshold correlation is detected between the consecutive ECG data sets.

6. The method according to claim 1, wherein: The specific arrhythmia comprises a transient intrinsic tachycardia that is corrected by the cardiovascular device within a predetermined time limit; and A corresponding response does not include an emergency alert or urgent warning requesting a clinician's attention.

7. The method according to claim 1, wherein: The specific arrhythmias include pacemaker-mediated indefinite-cycling tachycardia; and A responsiveness response includes an urgent warning requesting the clinician's prompt attention.

8. The method according to claim 1, wherein: The specific arrhythmias include pacemaker-mediated conduction block tachycardia; and Corresponding responses include emergency alerts that require immediate intervention by clinicians.

9. The method according to claim 1, wherein: The specific arrhythmia includes a persistent uncorrected intrinsic tachycardia; and Corresponding responses include emergency alerts that require immediate intervention by clinicians.

10. The method according to claim 1, wherein The stimulation artifact includes a pacing pulse, and further comprising determining whether the cardiovascular device performs one of the following: atrial pacing, ventricular sensing ("AP-VS"); Atrial sensing before QRS onset, ventricular pacing ("AS-VP"): Atrial pacing, ventricular pacing before QRS onset ("AP-VP"); and Ventricular pacing during the QRS ("fusion beat").

11. The method of claim 1 , further comprising inferring information about the structure, programming, or origin of the cardiovascular device from properties of stimulation artifact(s) in the ECG data.

12. The method of claim 1, further comprising updating the normal template with newly collected ECG test data from the same patient by converting the analyzed test data into training data.

13. A physiological monitoring system comprising: at least one electrocardiogram ("ECG") port; a processor communicatively coupled to the ECG port; as well as a memory communicatively coupled to the ECG port, the memory being encoded with: A data structure comprising: A heartbeat classifier for detecting specific arrhythmias in ECG data; Linking to corresponding responses for predefined arrhythmias; and Normal templates learned from ECG data; and A set of instructions that, when executed by the processor, cause the processor to: collecting an ECG data set through the ECG port, the data set comprising at least one of an intrinsic ECG signal from a patient and a stimulation artifact from a cardiovascular device of the patient; comparing the ECG dataset with the normal template; and If the ECG dataset matches the normal template, the normal template is updated with the ECG dataset.

14. The system of claim 13, further comprising an ECG monitoring sensor communicatively coupled to the ECG port.

15. The system according to claim 13, wherein: The cardiovascular device includes a pacemaker or an implantable cardioverter-defibrillator ("ICD").

16. The system of claim 13, further comprising a communication interface for coupling to an external network.

17. A non-transitory computer-readable storage medium comprising instructions that, when executed, cause a processor to: collecting electrocardiogram ("ECG") data comprising at least one of an intrinsic ECG signal from a patient and stimulation artifacts from a cardiovascular device used by the patient; Create normal templates from continuous ECG datasets through unsupervised machine learning; detecting the presence of a specific arrhythmia in the ECG data by applying at least one knowledge-based heartbeat classifier; as well as A response is executed corresponding to the expected effect of the particular arrhythmia on the patient.

18. A method of analysing an electrocardiogram (ECG) of a user of cardiovascular equipment affecting said ECG, substantially as hereinbefore described and illustrated in the accompanying drawings.

19. A physiological monitoring system for analysing an electrocardiogram (ECG) of a user of cardiovascular equipment affecting said ECG, substantially as hereinbefore described and illustrated in the accompanying drawings.

20. A non-transitory computer readable storage medium comprising instructions which, when executed, cause a processor to detect an arrhythmia in an electrocardiogram (ECG) of a user of a cardiovascular device that affects the ECG, substantially as hereinbefore described and illustrated.

Images