Systems and methods for detecting arrhythmic heartbeats and detecting cardiac mapping annotations

Abstract

The invention is entitled "System and method for detecting arrhythmic heartbeats and detecting cardiac mapping annotations." Systems and methods for detecting mapping annotations of an electrophysiology (EP) mapping system are disclosed. The system includes a processor including a machine learning algorithm configured to: receive a first heartbeat at an identified cardiac spatial location, the first heartbeat including a first set of attribute information corresponding to the first heartbeat; receive a second heartbeat at the identified cardiac spatial location, the second heartbeat including a second set of attribute information corresponding to the second heartbeat; compare the first set of attribute information to the second set of attribute information; and determine which of the first heartbeat and the second heartbeat has the best characteristics based on the compared attribute information.

Description

[0001] Cross-references to related applications This application claims the benefit of U.S. Provisional Application 63 / 037,259, filed June 10, 2020, which is incorporated herein by reference as fully illustrated. Technical Field

[0002] This disclosure relates to artificial intelligence and machine learning, which are associated with selecting the optimal heartbeat at each spatial location in a stable arrhythmia and calculating and detecting the optimal cardiac mapping annotation. Background Technology

[0003] Treatment of cardiac conditions such as arrhythmias often requires detailed mapping of cardiac tissue, chambers, veins, arteries, and / or electrical pathways. For example, a prerequisite for successful catheter ablation is that the cause of the arrhythmia is accurately located within a cardiac chamber. Such localization can be accomplished via an electrophysiological study, during which potentials are spatially resolved using a mapping catheter introduced into the cardiac chamber. This electrophysiological study (often referred to as electrophysiological (EP) cardiac mapping or cardiac electroanatomical (EA) mapping) provides 3D mapping data that can be displayed on a monitor. In many cases, mapping and therapeutic functions (e.g., ablation) are provided by a single catheter or a group of catheters, allowing the mapping catheter to simultaneously function as a therapeutic (e.g., ablation) catheter.

[0004] Mapping of cardiac regions, such as the heart itself, tissues, veins, arteries, and / or electrical pathways, can lead to the identification of problem areas such as scar tissue, sources of arterial vasculature (e.g., electrocolic vasculature), healthy areas, etc. The cardiac region can be mapped to allow for the provision of a visual rendering of the mapped cardiac region using a display, as further disclosed herein. Additionally, cardiac mapping can include mapping based on one or more modalities, such as, but not limited to, local activation time (LAT), electrical activity, topology, bipolar mapping, dominant frequency, or impedance. Catheters inserted into the patient can be used to capture data corresponding to multiple modalities, and this data can be provided simultaneously or at different times for rendering based on the corresponding settings and / or preferences of a healthcare professional.

[0005] Electrocardiograms (ECGs) and electrograms (EGMs) are examples of cardiac mapping. An ECG is generated from electrical signals from the heart that describe cardiac activity. ECGs are used during cardiac surgery to identify potential sites of origin for heart conditions. ECG signals can also be used to map parts of the heart. An electrogram can be recorded by each of the electrodes in contact with the surface of the heart, relative to a time reference, such as starting with a P wave in the sinus rhythm from a surface ECG. ECG and EGM signals can be used with rule-based algorithms to determine cardiac mapping annotations, such as those described in U.S. Patent Publication US2018 / 0042504. However, during cardiac procedures, physicians may need to manually correct incorrect mapping annotations.

[0006] EP cardiac mapping systems typically acquire and record the first heartbeat at each spatial location, even if the heartbeat has poor characteristics. An improved method and system are needed to select and detect the optimal heartbeat as the mapping annotation at each spatial location. Furthermore, an improved method and system are needed for determining cardiac mapping annotations. Summary of the Invention

[0007] This paper describes methods, apparatus, systems, and models for selecting the optimal heartbeat at each spatial location in stable arrhythmias and for calculating and detecting the optimal cardiac mapping annotation.

[0008] According to one aspect, the subject matter disclosed herein relates to a system for detecting heartbeats having optimal characteristics for electrophysiological (EP) mapping systems. The system includes a processor comprising a machine learning algorithm configured to receive a first heartbeat at an identified cardiac spatial location, the first heartbeat including a first set of attribute information corresponding to the first heartbeat; receive a second heartbeat at the identified cardiac spatial location, the second heartbeat including a second set of attribute information corresponding to the second heartbeat; compare the first set of attribute information with the second set of attribute information; and determine, based on the compared attribute information, which of the first and second heartbeats possesses optimal characteristics.

[0009] According to another aspect, the subject matter disclosed herein relates to a method for detecting heartbeats with optimal characteristics of an EP mapping system using a machine learning algorithm. The method includes receiving first data including a first heartbeat at an identified cardiac spatial location, the first heartbeat including first attribute information corresponding to the first heartbeat; receiving second data including a second heartbeat at the identified cardiac spatial location, the second heartbeat including second attribute information corresponding to the second heartbeat; comparing the first data with the second data; and outputting a binary determination based on the comparison of which of the first and second heartbeats has optimal characteristics.

[0010] According to another aspect, the subject matter disclosed herein relates to a system for detecting annotations in an EP (Electronic Point of Interest) mapping system. The system includes a processor comprising a machine learning algorithm configured to: receive input data comprising attribute data of each of a plurality of heartbeats obtained at the same spatial location; compare the attribute data of each of the plurality of heartbeats with a predefined threshold; and determine which heartbeat to use as the annotation based on the heartbeat with the best attribute data.

[0011] According to another aspect, the subject matter disclosed herein relates to a method for detecting annotations in an EP (Extended Prescription) mapping system using a machine learning algorithm. The method includes receiving input data comprising attribute data for each of a plurality of heartbeats obtained at the same spatial location; comparing the attribute data for each of the plurality of heartbeats with a predefined threshold; determining which heartbeat should be used as the annotation based on the heartbeat with the best attribute data; and outputting the determination to the EP mapping system.

[0012] According to another aspect, the subject matter disclosed herein relates to a system for determining mapping annotations for an EP mapping system. The system includes a processor comprising a machine learning algorithm configured to receive first EP signal data obtained from a cardiac location; determine initial mapping annotations based on the first EP signal data; receive data associated with manually corrected mapping annotations; receive second EP signal data obtained from a cardiac location; and determine new mapping annotations based on the second EP signal data and the manually corrected mapping annotations.

[0013] According to another aspect, the subject matter disclosed herein relates to a method for determining mapping annotations of an EP mapping system using a machine learning algorithm. The method includes receiving first EP signal data obtained from a cardiac location; determining initial mapping annotations based on the first EP signal data; receiving data related to manually corrected mapping annotations at the cardiac location; receiving second EP signal data obtained from the cardiac location; and determining new mapping annotations based on the second EP signal data and the manually corrected mapping annotations. Attached Figure Description

[0014] The above and other features and advantages of the invention will become apparent from the following more detailed description of preferred embodiments of the invention as illustrated in the accompanying drawings.

[0015] Figure 1 This is a block diagram of an exemplary system for remotely monitoring and transmitting patient biometrics according to the subject matter of this application.

[0016] Figure 2 This is a system diagram of an example computing environment for network communication based on the subject matter of this application.

[0017] Figure 3 This is a block diagram of an exemplary apparatus that can implement one or more features of this disclosure according to the subject matter of this application.

[0018] Figure 4 The combination shown is based on the subject matter of this application. Figure 3 A graphical depiction of an exemplary device's artificial intelligence system.

[0019] Figure 5 The subject matter of this application is shown in Figure 4 The methods executed in artificial intelligence systems.

[0020] Figure 6 An example of a probability calculated using Naive Bayes according to the subject matter of this application is shown.

[0021] Figure 7 An exemplary decision tree based on the subject matter of this application is shown.

[0022] Figure 8 An exemplary random forest classifier based on the subject matter of this application is shown.

[0023] Figure 9 An exemplary logistic regression based on the subject matter of this application is shown.

[0024] Figure 10 An exemplary support vector machine according to the subject matter of this application is shown.

[0025] Figure 11 An exemplary linear regression model based on the subject matter of this application is shown.

[0026] Figure 12 An exemplary K-means clustering method based on the subject matter of this application is shown.

[0027] Figure 13 An exemplary ensemble learning algorithm based on the subject matter of this application is shown.

[0028] Figure 14 An exemplary neural network according to the subject matter of this application is shown.

[0029] Figure 15 A hardware-based neural network according to the subject matter of this application is shown.

[0030] Figure 16 An electrocardiogram (ECG) signal generated by the contraction (depolarization) and relaxation (repolarization) of the atrial and ventricular myocardium of the heart, according to the subject matter of this application, is shown.

[0031] Figure 17 An exemplary cardiac ablation system according to the subject matter of this application is shown, wherein one or more features of the disclosed subject matter may be implemented.

[0032] Figure 18 An exemplary embodiment of a convolutional neural network according to this application is shown, which can receive the attributes of two heartbeats and learn to identify which heartbeat is better; Figure 19 This is an exemplary flowchart illustrating a method for training a neural network to obtain the properties of two heartbeats and learning to identify which heartbeat is better, according to the subject matter of this application. Figure 20 This is an exemplary illustration of multiple electrodes sampling multiple contact points in a heart cavity according to an exemplary embodiment of this application; Figure 21 An exemplary embodiment of a recurrent neural network according to the subject matter of this application is shown, which can receive the properties of multiple heartbeats at the same spatial location and learn to identify the best labeling annotation; Figure 22 This is an exemplary illustration of determining the calibration annotation based on the highest peak voltage of the ECG signal surrounding the reference annotation in the subject matter of this application; Figure 23 This is an exemplary flowchart illustrating a system and method for training a machine learning algorithm to determine label annotations according to the subject matter of this application; and Figure 24 These are exemplary illustrations showing EGM signals for mapping annotations determined by a rule-based system and EGM signals for mapping annotations determined manually by a physician. Detailed Implementation

[0033] This invention provides a system and method for selecting and detecting the optimal heartbeat as a mapping annotation at each spatial location, and for improving the determination of cardiac mapping annotations based on machine learning.

[0034] Figure 1 This is a block diagram of an exemplary system 100 for remotely monitoring and transmitting patient biometrics (i.e., patient data). Figure 1 In the example shown, system 100 includes a patient biometric monitoring and processing device 102 associated with patient 104, a local computing device 106, a remote computing system 108, a first network 110, and a second network 120.

[0035] According to an exemplary embodiment, the monitoring and processing device 102 may be a device located inside the patient's body (e.g., subcutaneously implantable). The monitoring and processing device 102 may be inserted into the patient's body via any applicable means, including oral injection, surgical insertion via vein or artery, endoscopic procedure, or laparoscopic procedure.

[0036] According to an exemplary embodiment, the monitoring and processing device 102 may be an external device to the patient. For example, as described in more detail below, the monitoring and processing device 102 may include an attachable patch (e.g., which is attached to the patient's skin). The monitoring and processing device 102 may also include a catheter, probe, blood pressure cuff, scale, bracelet or smartwatch biometric tracker, glucose monitor, continuous positive airway pressure (CPAP) machine, or virtually any device that can provide input related to the patient's health or biometrics, and has one or more electrodes.

[0037] According to an exemplary embodiment, the monitoring and processing device 102 may include components inside the patient and components outside the patient.

[0038] exist Figure 1 A single monitoring and processing device 102 is shown. However, an exemplary system may include multiple patient biometric monitoring and processing devices. These devices may communicate with one or more other patient biometric monitoring and processing devices. Additionally or alternatively, these devices may communicate with a network 110.

[0039] One or more monitoring and processing devices 102 may acquire patient biometric data (e.g., electrical signals, blood pressure, temperature, blood glucose levels, or other biometric data) and receive from one or more other monitoring and processing devices 102 at least a portion of the acquired patient biometric data representing the acquired patient biometrics and additional information associated with the acquired patient biometrics. The additional information may be, for example, diagnostic information and / or additional information obtained from an additional device such as a wearable device. Each monitoring and processing device 102 may process data, including patient biometrics acquired by itself and data received from one or more other monitoring and processing devices 102.

[0040] exist Figure 1 In this context, network 110 is an example of a near-field network (e.g., a local area network (LAN) or a personal area network (PAN)). Information can be transmitted between monitoring and processing device 102 and local computing device 106 via near-field network 110 using any of a variety of near-field wireless communication protocols such as Bluetooth, Wi-Fi, Zigbee, Z-Wave, Near Field Communication (NFC), Ultraband, Zigbee, or Infrared (IR)).

[0041] In an exemplary embodiment, the second network 120 may be a wired network, a wireless network, or include one or more wired and wireless networks. For example, the second network 120 may be a remote network (e.g., a wide area network (WAN), the Internet, or a cellular network). Information may be transmitted via the second network 120 using any of a variety of remote wireless communication protocols (e.g., TCP / IP, HTTP, 3G, 4G / LTE, or 5G / New Radio).

[0042] In an exemplary embodiment, patient monitoring and processing device 102 may include a patient biometric sensor 112, a processor 114, a user input (UI) sensor 116, a memory 118, and a transmitter-receiver (i.e., transceiver) 122. Patient monitoring and processing device 102 may continuously or periodically monitor, store, process, and transmit any number of various patient biometrics via network 110. Examples of patient biometrics include electrical signals (e.g., ECG signals and brain biometrics), blood pressure data, blood glucose data, and temperature data. Patient biometrics may be monitored and transmitted for the treatment of any number of various diseases, such as cardiovascular diseases (e.g., arrhythmias, cardiomyopathy, and coronary artery disease) and autoimmune diseases (e.g., type I and type II diabetes).

[0043] In one embodiment, the patient biometric sensor 112 may include, for example, one or more sensors configured to sense a type of biometric patient biometric. For example, the patient biometric sensor 112 may include electrodes, temperature sensors, blood pressure sensors, blood glucose sensors, blood oxygen sensors, pH sensors, accelerometers, and microphones configured to acquire electrical signals (e.g., cardiac signals, brain signals, or other bioelectrical signals).

[0044] In an exemplary embodiment, as described in more detail below, the patient biometric monitoring and processing device 102 may be an ECG monitor for monitoring ECG signals of the heart. The patient biometric sensor 112 of the ECG monitor may include one or more electrodes for acquiring ECG signals. ECG signals can be used to treat various cardiovascular diseases.

[0045] In an exemplary embodiment, transceiver 122 may include a separate transmitter and receiver. Alternatively, transceiver 122 may include a transmitter and receiver integrated into a single device.

[0046] In an exemplary embodiment, processor 114 may be configured to store patient data, such as patient biometric data acquired by patient biometric sensor 112, in memory 118, and transmit patient data across network 110 via transmitter of transceiver 122. Data from one or more other monitoring and processing devices 102 may also be received by receiver of transceiver 122, as described in more detail below.

[0047] According to an exemplary embodiment, the monitoring and processing device 102 includes a UI sensor 116, which may be a piezoelectric or capacitive sensor configured to receive user input, such as a tap or touch. For example, in response to a patient 104 tapping or touching a surface of the monitoring and processing device 102, the UI sensor 116 can be controlled to achieve capacitive coupling. Gesture recognition can be achieved via any of a variety of capacitance types, such as resistive capacitance, surface capacitance, projected capacitance, surface acoustic waves, piezoelectric, and infrared touch. The capacitive sensor may be positioned over a small area or along the length of a surface, such that a tap or touch on the surface activates the monitoring device.

[0048] As described in more detail below, processor 114 can be configured to selectively respond to different tap patterns (e.g., single or double taps) of a capacitive sensor (which may be UI sensor 116), enabling different tasks of the patch (e.g., data acquisition, storage, or transmission) to be activated based on the detected pattern. In some embodiments, audible feedback can be given to the user from processing device 102 when a gesture is detected.

[0049] In an exemplary embodiment, a local computing device 106 of system 100 communicates with patient biometric monitoring and processing device 102 and may be configured to act as a gateway to remote computing system 108 via a second network 120. For example, local computing device 106 may be a smartphone, smartwatch, tablet, or other portable smart device configured to communicate with other devices via the second network 120. Alternatively, local computing device 106 may be a fixed or standalone device, such as a fixed base station including, for example, modem and / or router capabilities, a desktop or laptop computer using an executable program to transmit information between processing device 102 and remote computing system 108 via a PC's radio module, or a USB dongle. Patient biometrics may be transmitted between local computing device 106 and patient biometric monitoring and processing device 102 via a near-field wireless network 110, such as a local area network (LAN) (e.g., a personal area network (PAN)), using near-field wireless technology standards (e.g., Bluetooth, Wi-Fi, ZigBee, Z-wave, and other near-field wireless standards). In some implementations, the local computing device 106 may also be configured to display the acquired patient electrical signals and information associated with the acquired patient electrical signals, as described in more detail below.

[0050] In some exemplary embodiments, the remote computing system 108 may be configured to receive at least one of monitored patient biometrics and patient-associated information via a second network 120, which is a remote network. For example, if the local computing device 106 is a mobile phone, the second network 120 may be a wireless cellular network and may transmit information between the local computing device 106 and the remote computing system 108 via wireless technology standards such as any of the wireless technologies described above. As described in more detail below, the remote computing system 108 may be configured to provide (e.g., visually displayed and / or audibly provided) at least one of patient biometrics and related information to healthcare professionals (e.g., physicians).

[0051] Figure 2 This is a system diagram of an example computing environment 200 communicating with a second network 120. In some cases, the computing environment 200 is integrated into a public cloud computing platform (such as Amazon Web Services or Microsoft Azure), a hybrid cloud computing platform (such as HP Enterprise OneSphere), or a private cloud computing platform.

[0052] like Figure 2 As shown, computing environment 200 includes remote computing system 108, which is an example of a computing system on which the embodiments described herein can be implemented.

[0053] The telecomputing system 108 can perform various functions via a processor 220, which may include one or more processors. For example, functions may include analyzing monitored patient biometrics and related information, and providing (e.g., via display 266) alerts, additional information, or instructions based on physician-determined or algorithm-driven thresholds and parameters. As described in more detail below, the telecomputing system 108 can be used to provide a patient information dashboard (e.g., via display 266) to healthcare professionals (e.g., physicians), enabling them to identify and prioritize patients with more critical needs than others.

[0054] like Figure 2 As shown, the remote computing system 108 may include communication mechanisms such as bus 221, or other communication mechanisms for transmitting information within the computer system 108. The computer system 108 also includes one or more processors 220 coupled to bus 221 for processing information. Processor 220 may include one or more CPUs, GPUs, or any other processor known in the art.

[0055] Computer system 108 may also include system memory 230 coupled to bus 221 for storing information and instructions to be executed by processor 220. System memory 230 may include computer-readable storage media in the form of volatile and / or non-volatile memory, such as read-only system memory (ROM) 231 and / or random access memory (RAM) 232. System memory RAM 232 may include other dynamic storage devices (e.g., dynamic RAM, static RAM, and synchronous DRAM). System memory ROM 231 may include other static storage devices (e.g., programmable ROM, erasable PROM, and electrically erasable PROM). Furthermore, system memory 230 may be used to store temporary variables or other intermediate information during processor 220 instruction execution. Basic input / output system 233 (BIOS) may be included among elements within computer system 108, such as routines for transferring information during startup, which may be stored in system memory ROM 231. RAM 232 may contain data and / or program modules that are immediately accessible and / or currently operated by processor 220. The system memory 230 may additionally include, for example, an operating system 234, application programs 235, other program modules 236, and program data 237.

[0056] In an exemplary embodiment, computer system 108 also includes a disk controller 240 coupled to bus 221 to control one or more storage devices, such as hard disks 241 and removable media drives 242 (e.g., floppy disk drives, optical disk drives, tape drives, and / or solid-state drives), for storing information and instructions. Storage devices can be added to computer system 108 using appropriate device interfaces, such as Small Computer System Interface (SCSI), Integrated Device Electronics (IDE), Universal Serial Bus (USB), or FireWire.

[0057] Computer system 108 may also include a display controller 265 coupled to bus 221 to control a monitor or display 266, such as a cathode ray tube (CRT) or liquid crystal display (LCD), to display information to a computer user. The illustrated computer system 108 includes a user input interface 260 and one or more input devices, such as a keyboard 262 and a pointing device 261, for interacting with the computer user and providing information to the processor 220. The pointing device 261 may be, for example, a mouse, trackball, or pointer, for transmitting directional information and command selections to the processor 220 and for controlling cursor movement on the display 266. The display 266 may provide a touchscreen interface that allows input to supplement or replace the communication of directional information and command selections by the pointing device 261 and / or keyboard 262.

[0058] In response to processor 220 executing one or more sequences of instructions contained in memory such as system memory 230, computer system 108 may perform some or every one of the functions and methods described herein. Such instructions may be read into system memory 230 from another computer-readable medium such as hard disk 241 or removable media drive 242. Hard disk 241 may contain one or more data repositories and data files used by the embodiments described herein. The contents of the data repositories and data files may be encrypted to enhance security. Processor 220 may also be employed in a multiprocessing arrangement to execute one or more sequences of instructions contained in system memory 230. In alternative embodiments, hardwired circuitry may be used in place of or in combination with software instructions. Therefore, the embodiments are not limited to any particular combination of hardware circuitry and software.

[0059] As described above, computer system 108 may include at least one computer-readable medium or memory for storing instructions programmed according to the embodiments described herein and for containing the data structures, tables, records, or other data described herein. As used herein, the term computer-readable medium refers to any non-transitory tangible medium that participates in providing instructions to processor 220 for execution. Computer-readable media can take many forms, including but not limited to non-volatile media, volatile media, and transmission media. Non-limiting examples of non-volatile media include optical discs, solid-state drives, magnetic disks, and magneto-optical discs, such as hard disk 241 or removable media drive 242. Non-limiting examples of volatile media include dynamic memory, such as system memory 230. Non-limiting examples of transmission media include coaxial cables, copper wires, and optical fibers, including conductors constituting bus 221. Transmission media may also take the form of acoustic or optical waves, such as acoustic or optical waves generated during radio wave and infrared data communication.

[0060] The computing environment 200 may also include a computer system 108 that operates in a networked environment using logical connections to a local computing device 106 and one or more other devices, such as personal computers (laptop or desktop computers), mobile devices (e.g., patient mobile devices), servers, routers, network PCs, peer-to-peer devices, or other public network nodes, and typically includes many or all of the elements described above with respect to computer system 108. When used in a networked environment, computer system 108 may include a modem 272 for establishing communication on a second network 120, such as the Internet. Modem 272 may be connected to system bus 221 via network interface 270 or via another suitable mechanism.

[0061] like Figure 1 and Figure 2As shown, the second network 120 can be any network or system known in the art, including the Internet, intranet, local area network (LAN), wide area network (WAN), metropolitan area network (MAN), direct connection or a series of connections, cellular telephone network, or any other network or medium that can facilitate communication between computer system 610 and other computers (e.g., local computing device 106).

[0062] Figure 3 This is a block diagram of an exemplary device 300 that may implement one or more features of this disclosure. For example, device 300 may be a local computing device 106. Device 300 may include, for example, a computer, gaming device, handheld device, set-top box, television, mobile phone, or tablet computer. Device 300 includes a processor 302, a memory 304, a storage device 306, one or more input devices 308, and one or more output devices 310. Device 300 may also optionally include an input driver 312 and an output driver 314. It should be understood that device 300 may include... Figure 3 Additional components not shown include an artificial intelligence accelerator.

[0063] In various alternatives, processor 302 includes a central processing unit (CPU), a graphics processing unit (GPU), a CPU and GPU located on the same die, or one or more processor cores, wherein each processor core can be a CPU or a GPU. In various alternatives, memory 304 is located on the same die as processor 302 or is located separately from processor 302. Memory 304 includes volatile or non-volatile memory, such as random access memory (RAM), dynamic RAM, or cache.

[0064] Storage device 306 includes fixed or removable storage devices, such as hard disk drives, solid-state drives, optical disks, or flash drives. Input device 308 includes, but is not limited to, keyboards, keypads, touchscreens, touchpads, detectors, microphones, accelerometers, gyroscopes, biometric scanners, or network connectors (e.g., wireless LAN cards for transmitting and / or receiving wireless IEEE 802 signals). Output device 310 includes, but is not limited to, displays, speakers, printers, haptic feedback devices, one or more lights, antennas, or network connections (e.g., wireless LAN cards for transmitting and / or receiving wireless IEEE 802 signals).

[0065] Input driver 312 communicates with processor 302 and input device 308, and allows processor 302 to receive input from input device 308. Output driver 314 communicates with processor 302 and output device 310, and allows processor 302 to send output to output device 310. It should be noted that input driver 312 and output driver 314 are optional components, and device 300 will operate in the same manner if input driver 312 and output driver 314 are not present. Output driver 314 may include an accelerated processing unit (“APD”) 316 coupled to display device 318. APD receives computation commands and graphics rendering commands from processor 302, processes those computation and graphics rendering commands, and provides pixel output to display device 318 for display. As described further in detail below, APD 316 includes one or more parallel processing units to perform computations according to the Single Instruction Multiple Data (“SIMD”) paradigm. Therefore, although various functions are described herein as being performed by or in conjunction with APD 316, in various alternatives, the functions described as being performed by APD 316 may additionally or alternatively be performed by other computing devices with similar capabilities, which are not driven by a host processor (e.g., processor 302) and provide graphics output to display device 318. For example, any processing system that performs processing tasks according to the SIMD paradigm is expected to perform the functions described herein. Alternatively, computing systems that do not perform processing tasks according to the SIMD paradigm are expected to perform the functions described herein.

[0066] Figure 4 The combination is shown Figure 3 A functional graphical depiction of an exemplary device for an artificial intelligence system 400 is provided. System 400 includes data 410, a machine 420, a model 430, multiple predictions 440, and underlying hardware 450. System 400 operates in such a way that it uses data 410 to train machine 420 while simultaneously building model 430 to enable prediction of multiple outcomes 440. System 400 may operate relative to hardware 450. In such a configuration, data 410 may be associated with hardware 450 and may originate from, for example, monitoring and processing device 102. For example, data 410 may be data being generated or output data associated with hardware 450. Machine 420 may function as or be associated with controller or data collection associated with hardware 450. Model 430 may be configured to model the operation of hardware 450 and the data 410 collected from hardware 450 in order to predict the outcomes achieved by hardware 450. Using the predicted outcomes 440, hardware 450 may be configured to provide a desired outcome 440 from hardware 450.

[0067] Figure 5 It shows in Figure 4A general method 500 is executed in an artificial intelligence system. Method 500 includes collecting data from hardware at step 510. This data may include data currently collected from the hardware, historical data from the hardware, or other data, or various combinations thereof. For example, the data may include measurements taken during a surgical procedure and may be correlated with the outcome of the procedure. For example, the temperature of the heart may be collected and correlated with the outcome of a cardiac procedure.

[0068] At step 520, method 500 includes training the machine on hardware. This training may include analysis and correlation of the data collected in step 510. For example, in the case of the heart, the data on temperature and outcomes may be trained to determine whether there is a correlation or relationship between the heart's temperature and the outcome during the procedure.

[0069] At step 530, method 500 includes building a model on data associated with the hardware. Building the model may include physical hardware or software modeling, algorithmic modeling, etc., as will be described below. This modeling may attempt to represent the collected and trained data.

[0070] At step 540, method 500 includes predicting the outcome of a model associated with the hardware. Such a prediction of the outcome may be based on a trained model. For example, in the case of the heart, if a temperature between 97.7 and 100.2 during the procedure produces a positive outcome, the outcome can be predicted based on the heart's temperature during the procedure in a given procedure. While the model is basic, it is provided for illustrative purposes and to enhance understanding of this disclosure.

[0071] The systems and methods of this invention are used to train machines, build models, and predict outcomes using algorithms. These algorithms can be used to solve trained models and predict hardware-related results. These algorithms can generally be categorized into classification, regression, and clustering algorithms.

[0072] For example, classification algorithms are used to categorize a dependent variable (which is the variable being predicted) into multiple classes and predict the class (dependent variable) given an input. Therefore, classification algorithms are used to predict outcomes from a fixed set of predefined results. Classification algorithms can include Naive Bayes, decision trees, random forest classifiers, logistic regression, support vector machines, and k nearest neighbors.

[0073] Generally speaking, the Naive Bayes algorithm follows Bayes' theorem and employs a probabilistic approach. It should be understood that other probability-based algorithms can also be used, and these algorithms typically operate using similar probabilistic principles to those described below for the exemplary Naive Bayes algorithm.

[0074] Figure 6An example of probabilities computed by Naive Bayes is shown. The probabilistic approach of Bayes' theorem essentially means that the algorithm has a set of prior probabilities for each class of the target, rather than jumping directly into the data. After inputting the data, the Naive Bayes algorithm updates the prior probabilities to form the posterior probabilities. This is given by the following formula: Naive Bayes and Bayesian algorithms are often useful when it's necessary to predict whether an input belongs to one of n categories in a given list. Probabilistic methods can be used because the probabilities for all n categories will be quite low.

[0075] For example, such as Figure 6 As shown, a person plays golf, depending on factors including, but not limited to, the external weather shown in the first dataset 610. The first dataset 610 shows the weather in the first column and the golf playing results associated with that weather in the second column. In frequency table 620, the frequency of certain events occurring is generated. In frequency table 620, the frequency with which a person plays or does not play golf under each weather condition is determined. Thus, a likelihood table is compiled to generate initial probabilities. For example, the probability of cloudy weather is 0.29, while the probability of playing golf in general is 0.64.

[0076] Posterior probabilities can be generated from likelihood table 630. These posterior probabilities can be configured to answer questions about weather conditions and whether to play golf under those conditions. For example, the probability of playing golf in sunny weather can be expressed using Bayes' theorem: P(Yes|Sunny) = P(Sunny|Yes) * P(Yes) / P(Sunny) According to likelihood table 630: P(sunny|is) = 3 / 9 = 0.33 P (sunny) = 5 / 14 = 0.36 P(is) = 9 / 14 = 0.64. Therefore, P(is|sunny) = 0.33 * 0.64 / 0.36 or approximately 0.60 (60%).

[0077] Generally, a decision tree is a tree structure similar to a flowchart, where each outer node represents a test on an attribute and each branch represents the result of that test. Leaf nodes contain the actual predicted labels. A decision tree starts at the root, where attribute values ​​are compared until a leaf node is reached. Decision trees can be used as classifiers when dealing with high-dimensional data and when little time has been spent on data preparation. Decision trees can take the form of simple decision trees, linear decision trees, algebraic decision trees, deterministic decision trees, stochastic decision trees, nondeterministic decision trees, and quantum decision trees. The following sections... Figure 7 An example decision tree is provided in the document.

[0078] Figure 7 A decision tree for deciding whether to play golf is shown, following the same structure as the Bayesian example above. In the decision tree, the first node 710 checks the weather, thus considering sunny 712, cloudy 714, and rainy 716 as choices to proceed down the decision tree. If the weather is sunny, the tree's leg follows to a second node 720 that checks the temperature. In this example, the temperature at node 720 can be high 722 or normal 724. If the temperature at node 720 is high 722, the prediction is "No" (don't play) 723 for the golf ball. If the temperature at node 720 is normal 724, the prediction is "Yes" (play) 725 for the golf ball.

[0079] Furthermore, starting from the first node 710, if the weather is cloudy 714, then the prediction result of "yes" 715 for a golf ball occurs.

[0080] Starting with the first node (weather 710), the result of rainy day 716 leads to a (re)check of the temperature at the third node 730. If the temperature at the third node 730 is normal 732, then "yes" (play) 733 golf ball. If the temperature at the third node 730 is low 734, then "no" (don't play) 735 golf ball.

[0081] According to the decision tree, a golfer will play golf if the weather is cloudy (715), sunny at normal temperature (725), or rainy at normal temperature (733). However, a golfer will not play golf if the weather is sunny and hot (723) or rainy and cold (735).

[0082] A random forest classifier is a committee of decision trees, where each decision tree is fed a subset of attributes from the data and makes predictions based on that subset. The mode of the actual predictions of the decision trees is considered to provide the final random forest answer. Random forest classifiers typically mitigate the overfitting present in independent decision trees, resulting in a more robust and accurate classifier.

[0083] Figure 8 An exemplary random forest classifier for classifying the colors of clothing is shown. Figure 8 As shown, the random forest classifier comprises five decision trees 8101, 8102, 8103, 8104, and 8105 (collectively or generally referred to as decision tree 810). Each tree is designed to classify the color of clothing. No discussion of each tree and the decisions made is provided, as each individual tree is typically used as... Figure 7The decision trees are used for operation. In this example, three of the five trees (8101, 8102, 8104) determine that the clothing is blue, one tree determines that the clothing is green (8103), and the remaining tree determines that the clothing is red (8105). The random forest takes these actual predictions from the five trees and calculates the mode of these actual predictions to provide the random forest answer that the clothing is blue.

[0084] Logistic regression is another algorithm used for binary classification tasks. Logistic regression is based on the logistic function (also known as the sigmoid function). This sigmoid curve can take any real-valued number and maps it between 0 and 1, asymptotically approaching those limits. Logistic models can be used to model the probability of a given category or event, such as pass / fail, win / lose, live / die, or healthy / sick. This can be extended to modeling several classes of events, such as determining whether an image contains a cat, dog, lion, etc. Each object detected in the image is assigned a probability between 0 and 1, with the sum of probabilities being 1.

[0085] In a logistic model, the logarithmic odds (logarithm of the odds) of a value labeled "1" is a linear combination of one or more independent variables ("predictors"); these independent variables can each be binary variables (two categories, encoded by indicator variables) or continuous variables (any real value). The corresponding probability of a value labeled "1" can vary between 0 (definitely the value "0") and 1 (definitely the value "1"), hence the label; the function that converts the logarithmic odds to probabilities is a logistic function, hence this name. The unit of measurement for the logarithmic odds scale is called the sublogarithm, derived from the logistic unit, hence this alternative name. Similar models, such as probabilistic models, can also be used with different sigmoid functions instead of logistic functions; the defining characteristic of a logistic model is that the odds of a given outcome are multiplicatively scaled at a constant rate by adding one of the independent variables, where each independent variable has its own parameter; for a binary dependent variable, this summarizes the odds ratio.

[0086] In a binary logistic regression model, the dependent variable has two levels (categorically). Multinomial logistic regression models outputs with more than two values, and if these multiple categories are ordered, it models them using ordinal logistic regression (e.g., proportional dominance ordinal logistic models). The logistic regression model itself simply models the probability of the output against the input and does not perform statistical classification (it is not a classifier), but it can be used as a classifier, for example, by selecting a cutoff value and classifying inputs with probabilities greater than the cutoff value into one category and inputs with probabilities less than the cutoff value into another; this is a common way to create binary classifiers.

[0087] Figure 9An exemplary logistic regression 900 is illustrated. This exemplary logistic regression enables the prediction of outcomes based on a set of variables. For example, based on an individual's GPA, acceptance by a school can be predicted. The past history of GPA and its relationship to acceptance make the prediction possible. Figure 9 Logistic regression (900) allows analysis of the GPA variable (920) to predict outcomes (910) limited to 0 to 1. At the lower end of the S-curve (930), the GPA (920) predicts an unacceptable outcome (910). At the upper end of the S-curve (940), the GPA (920) predicts an acceptable outcome (910). Logistic regression can be used to predict home values, customer life expectancy in the insurance industry, and more.

[0088] Support Vector Machines (SVMs) can be used to classify data with the margin between two classes that are as far apart as possible. This is called the maximum margin. SVMs take support vectors into account when drawing the hyperplane, unlike linear regression, which uses the entire dataset for that purpose.

[0089] Figure 10 An exemplary device actuator is illustrated. In the exemplary SVM 1000, data can be classified into two distinct categories, represented as squares 1010 and triangles 1020. The SVM 1000 operates by drawing a hyperplane 1030. The hyperplane 1030 is monitored by comparing the distances (shown as lines 1040) between the hyperplane 1030 and the nearest data points 1050 from each category. The nearest data points 1050 to the hyperplane 1030 are called support vectors. The hyperplane 1030 is drawn based on these support vectors 1050, and the optimal hyperplane has the maximum distance to each of these support vectors 1050. The distance between the hyperplane 1030 and the support vectors 1050 is called the margin.

[0090] SVM 1000 can be used to classify data by using a hyperplane 1030 such that the distance between the hyperplane 1030 and the support vectors 1050 is maximized. For example, this type of SVM 1000 can be used to predict heart disease.

[0091] K Nearest Neighbors (KNN) is a set of algorithms that typically make no assumptions about the distribution of the underlying data and perform a relatively short training phase. Generally, KNN uses a large number of data points divided into several classes to predict the classification of a newly sampled point. Operationally, KNN specifies an integer N with a new sample. It selects the N entries in the system's model that are closest to the new sample. It then determines the most common class of these entries and assigns that class to the new sample. KNN typically requires more storage space as the training set grows. This also means that the estimation time increases proportionally with the number of training points.

[0092] In regression algorithms, the output is a continuous quantity, so regression algorithms can be used when the target variable is a continuous variable. Linear regression is a general example of a regression algorithm. Linear regression can be used to estimate true quality (housing costs, number of bids, all buyout transactions, etc.) based on one or more consistent variables. A connection between the variables and the outcome is created by fitting a best-fit line (i.e., fitting a linear regression). This best-fit line is called the regression line and is expressed by the direct condition Y = a*X + b. Linear regression works best in methods involving low dimensionality.

[0093] Figure 11 An exemplary linear regression model is illustrated. In this model, the predictor variable 1110 is modeled relative to the measured variable 1120. Clusters of instances of the predictor variable 1110 and the measured variable 1120 are plotted as data points 1130. The data points 1130 are then fitted with a best-fit line 1140. Then, given the measured variable 1120, the best-fit line 1140 is used for subsequent predictions, which predict the predictor variable 1110 for that instance. Linear regression can be used to model and predict outcomes of surgical procedures, performance of financial portfolios, income forecasts, real estate, and traffic at estimated arrival times.

[0094] Clustering algorithms can also be used to model and train datasets. In clustering, inputs are assigned to two or more clusters based on feature similarity. Clustering algorithms typically learn patterns and useful insights from data without any guidance. For example, unsupervised learning algorithms such as K-means clustering can be used to cluster viewers into similar groups based on their interests, age, geography, etc.

[0095] K-means clustering is generally considered a simple unsupervised learning method. In K-means clustering, similar data points are grouped together and bound as clusters. One way to bind data points together is by calculating the centroid of the group. In determining valid clusters, in K-means clustering, the distance between each point and the centroid of the cluster is evaluated. Data points are assigned to the nearest cluster based on their distance to the centroid. The goal of clustering is to determine intrinsic groupings within a set of unlabeled data. The "K" in K-means represents the number of clusters formed. The number of clusters (basically the number of categories that can classify new data instances) can be determined by the user. For example, this determination can be performed during training using feedback and by looking at the size of the clusters.

[0096] K-means is used when the dataset contains points with different and well-spaced intervals; otherwise, if the clusters are not spaced, modeling may lead to inaccurate clustering. Additionally, K-means can be avoided when the dataset contains a large number of outliers or is non-linear.

[0097] Figure 12 This illustrates K-means clustering. In K-means clustering, data points are plotted and assigned K values. For example, for... Figure 12 With K=2, data points are plotted as shown in Figure 1210. These points are then assigned to similar centers in step 1220. Cluster centroids are identified as shown in Figure 1230. Once centroids are identified, these points are reassigned to clusters to provide the minimum distance between data points and their respective cluster centroids, as shown in Figure 1240. New centroids for the clusters can then be determined, as shown in Figure 1250. During the reassignment of data points to clusters, new cluster centroids may be formed, iterated, or a series of iterations may occur to minimize the cluster size and determine the centroids of the optimal centroids. Then, when new data points are measured, they can be compared with the centroids and clusters to identify them together with that cluster.

[0098] Ensemble learning algorithms can be used. These algorithms use multiple learning algorithms to achieve better predictive performance compared to the performance that can be obtained from any of the constituent learning algorithms alone. Ensemble learning algorithms perform the task of searching through a hypothesis space to find suitable hypotheses that will make good predictions for a particular problem. Even if the hypothesis space contains hypotheses that are very well-suited to a particular problem, finding good hypotheses can be very difficult. Ensemble algorithms combine multiple hypotheses to form better hypotheses. The term ensemble is generally reserved for methods that use the same base learner to generate multiple hypotheses. The broader terminology of multiple classifier systems also encompasses the mixing of hypotheses that are not induced by the same base learner.

[0099] Evaluating the predictions of an ensemble typically requires more computation than evaluating the predictions of a single model; therefore, ensembles can be seen as a way to compensate for poor learning algorithms by performing a significant amount of additional computation. Fast algorithms such as decision trees are often used in ensemble methods, such as random forests, although slower algorithms can also benefit from ensemble techniques.

[0100] Ensembles are supervised learning algorithms because they can be trained and then used for prediction. Therefore, a trained ensemble represents a single hypothesis. However, this hypothesis is not necessarily contained within the hypothesis space of the model that constructs it. Thus, ensembles can be shown to have greater flexibility in the functions they can represent. Theoretically, this flexibility allows these ensembles to be better fitted to the training data than a single model, but in practice, some ensemble techniques (especially bagging) tend to reduce the problems associated with overfitting to training data.

[0101] Empirically, ensemble algorithms tend to produce better results when there is significant diversity among the models. Therefore, many ensemble methods attempt to promote diversity among the models they combine. While not intuitive, more stochastic algorithms (such as stochastic decision trees) can produce stronger ensembles than highly deliberate algorithms (such as entropy-reducing decision trees). However, using multiple brute-force learning algorithms has shown to be more effective than techniques that attempt to discard models to promote diversity.

[0102] The number of component classifiers in an ensemble has a significant impact on prediction accuracy. Predetermining the ensemble size, volume, and speed of a large data stream makes this even more critical for online ensemble classifiers. Theoretical frameworks suggest that there exists an ideal number of component classifiers for ensembles, such that having more or fewer classifiers will decrease accuracy. Theoretical frameworks also indicate that using the same number of independent component classifiers as the class labels yields the highest accuracy.

[0103] Some common types of ensembles include Bayesian optimal classifier, bootstrapping aggregation (bagging), boosting, Bayesian model averaging, Bayesian model combination, model storage, and stacking. Figure 13 An exemplary ensemble learning algorithm is shown, in which bagging is performed in parallel 1310 and boosting is performed sequentially 1320.

[0104] During bagging, multiple subsets are created from the original dataset and replaced. A base model is created on each of these subsets. The base models can run in parallel and independently of each other. The final prediction can be determined by combining the predictions from all the base models. Bagging can be an effective method for reducing model variability.

[0105] On the other hand, boosting is an iterative process where each subsequent model attempts to correct the errors of the previous models. Subsequent models depend on the preceding models. First, a subset is created from the original dataset. Initially, all data points are given equal weight. Next, a base model is created on this subset. This model is then used to make predictions on the entire dataset. Errors are calculated using the actual and predicted values. Observations that are not correctly predicted are given higher weights. A second model is then created and makes predictions on the dataset. The second model attempts to correct the errors from the first model. Subsequent models are then created, each correcting the errors of the previous models. The final model (i.e., the strong learner) is a weighted average of all the previous models (i.e., the weak learners). Therefore, boosting can be used to combine multiple weak learners to form a strong learner.

[0106] A neural network is a network or circuit of neurons, or, in the modern sense, an artificial neural network composed of artificial neurons or nodes. The connections of biological neurons are modeled as weights. Positive weights reflect excitatory connections, while negative values ​​represent inhibitory connections. The input is modified by the weights and summed using a linear combination. Activation functions control the amplitude of the output. For example, an acceptable output range is typically between 0 and 1, or the range can be between -1 and 1.

[0107] These artificial networks can be used for predictive modeling, adaptive control, and applications, and can be trained on datasets. Experience-based self-learning can occur within the network, drawing conclusions from complex and seemingly unrelated groups of information.

[0108] For completeness, a biological neural network consists of one or more groups of chemically connected or functionally related neurons. A single neuron can connect to many other neurons, and the total number of neurons and connections in a network can be extensive. Connections (called synapses) typically form from an axon to a dendrite, but dendritic synapses and other connections are also possible. In addition to electrical signals, other forms of signaling, caused by the diffusion of neurotransmitters, exist.

[0109] Artificial intelligence, cognitive modeling, and neural networks are information processing paradigms inspired by how biological nervous systems process data. Artificial intelligence and cognitive modeling attempt to simulate some characteristics of biological neural networks. In the field of artificial intelligence, artificial neural networks have been successfully applied to speech recognition, image analysis, and adaptive control to build software agents or autonomous robots (in computers and video games).

[0110] In terms of artificial neurons, often referred to as artificial neural networks (ANNs) or simulated neural networks (SNNs), a neural network (NN) is a set of interconnected natural or artificial neurons that process information using mathematical or computational models based on computational connection methods. In most cases, an ANN is an adaptive system that changes its structure based on the flow of external or internal information through the network. More practically, neural networks are tools for modeling nonlinear statistical data or making decisions. These terms can be used to model complex relationships between inputs and outputs or to find patterns in data.

[0111] Artificial neural networks involve networks of simple processing elements (artificial neurons) that can exhibit complex global behavior determined by the connections between the processing elements and their parameters.

[0112] A classic type of artificial neural network is the recurrent Hopfield network. The practicality of artificial neural network models lies in their ability to infer functions from observations and also to utilize those functions. Unsupervised neural networks can also be used to learn representations of inputs that capture salient features of the input distribution, and recent deep learning algorithms can implicitly learn the distribution function of observed data. Learning in neural networks is particularly useful in applications where the complexity of the data or task makes manually designing such functions impractical.

[0113] Neural networks can be used in various fields. The tasks applied to artificial neural networks often fall into the following broad categories: function approximation or regression analysis, including time series forecasting and modeling; classification, including pattern and sequence recognition, novelty detection, and order decision-making; and data processing, including filtering, clustering, blind signal separation, and compression.

[0114] ANN applications encompass areas such as nonlinear system recognition and control (vehicle control, process control), game playing and decision making (backgammon, chat, competitions), pattern recognition (radar systems, facial recognition, object recognition), sequence recognition (gesture, speech, handwritten text recognition), medical diagnosis, financial applications, data mining (or knowledge discovery in databases, "KDD"), visualization, and email spam filtering. For example, semantic feature maps of user interests can be created from images trained for object recognition.

[0115] Figure 14 An exemplary neural network 1400 is illustrated. In this neural network, there is an input layer represented by multiple inputs such as 14101 and 14102. Inputs 1410a and 1410b are provided to a hidden layer depicted as including nodes 1420a, 1420b, 1420c, and 1420d. These nodes 1420a, 1420b, 1420c, and 1420d are combined to produce an output 1430 in the output layer. The neural network performs simple processing via hidden layers (nodes 14201, 14202, 14203, and 14204) of simple processing elements, and can exhibit complex global behavior determined by the connections between processing elements and element parameters.

[0116] Figure 14 The neural network 1400 in the example can be implemented in hardware. For example... Figure 15The diagram illustrates a hardware-based neural network. In an exemplary embodiment, and without limitation, the hardware-based neural network 1500 includes RAM 1510, which preferably includes weights, biases, directions, etc., a multiplexer 1520, a finite state machine (FSM) control unit 1530, an operand selector 1540, an arithmetic logic unit 1560, and a multiplier unit 1570. Input 1580 is provided to the neural network 1500, and the neural network 1500 produces an output 1550. For example, in... Figure 15 In the illustrated implementation, input 1580 is provided to multiplexer 1520, which forwards the selected input to multiplier unit 1570. Weights, biases, directions, etc., from RAM 1510 are also forwarded to multiplier unit 1570. Multiplier unit 1570 applies the weights, biases, directions, etc., to the selected input and forwards the result to arithmetic logic unit 1560. Operand selector 1540 is configured to track the number of operands being multiplied. Information from FSM control unit 1530 is also forwarded to arithmetic logic unit 1560. The FSM control unit 1530 controls the overall operation of the neuron block. Arithmetic logic unit 1560 performs one or more arithmetic operations and produces output 1550.

[0117] Treatment of cardiac conditions such as arrhythmias often requires detailed mapping of cardiac tissue, chambers, veins, arteries, and / or electrical pathways. This localization can be accomplished via an electrophysiological study, during which potentials are spatially resolved using a mapping catheter introduced into the cardiac chambers. This electrophysiological study (so-called electroanatomical mapping) thus provides 3D mapping data that can be displayed on a monitor. In many cases, mapping and therapeutic functions (e.g., ablation) are provided by a single catheter or a group of catheters, allowing the mapping catheter to simultaneously function as a therapeutic (e.g., ablation) catheter.

[0118] Cardiac mapping can be achieved using one or more techniques. As an example of a first technique, cardiac mapping can be achieved by sensing the electrical properties of cardiac tissue (e.g., local activation time) based on precise locations within the heart. The corresponding data can be acquired via one or more catheters advanced into the heart using catheters with electrical and position sensors at their distal ends. For example, initially, position and electrical activity can be measured at approximately 10 to approximately 20 points on the inner surface of the heart. These data points are often sufficient to generate a preliminary reconstruction or mapping map of the cardiac surface of satisfactory quality. The preliminary map can be combined with data taken from additional points to produce a more comprehensive map of cardiac electrical activity. In clinical settings, it is not uncommon to accumulate data at 100 or more sites to generate a detailed and comprehensive mapping map of cardiac chamber electrical activity. The resulting detailed map can then serve as a basis for determining therapeutic actions, such as tissue ablation, to alter the propagation of cardiac electrical activity and restore normal heart rhythm.

[0119] Catheters containing position sensors can be used to determine the trajectories of points on the surface of the heart. These trajectories can be used to infer kinematic properties, such as tissue contractility. When trajectory information is sampled at a sufficient number of points in the heart, a mapping depicting these kinematic properties can be constructed.

[0120] Typically, a catheter containing an electrical sensor at or near its distal tip is advanced to a point in the heart, where the sensor contacts the tissue and acquires data at that point, thereby measuring the electrical activity at that point in the heart. Multi-electrode catheters can be implemented using any suitable shape, such as linear catheters with multiple electrodes, balloon catheters including electrodes distributed on multiple ridges that shape the balloon, lasso or ring catheters with multiple electrodes, or any other suitable shape.

[0121] According to one example, a multi-electrode catheter can be advanced into the chambers of the heart. Anterior and posterior fluorescein (AP) and lateral fluorescein maps are obtained to establish the position and orientation of each electrode. An electrogram relative to a time reference (e.g., starting from the P wave in the sinus rhythm from a surface ECG) can be recorded by each of the electrodes in contact with the cardiac surface. As further disclosed herein, the system can distinguish which electrodes record electrical activity from those that do not, due to their less close proximity to the endocardial wall. After recording the initial electrogram, the catheter can be repositioned, and fluorescein and electrograms can be recorded again. An electropatch can then be constructed iteratively based on the above process.

[0122] According to one example, cardiac mapping can be generated based on the detection of the intracardiac electrical potential field. Non-contact techniques for simultaneously acquiring large amounts of cardiac electrical information can be implemented. For example, a catheter with a distal end portion can be provided with a series of sensor electrodes distributed on its surface and connected to an insulating electrical conductor for connection to a signal sensing and processing device. The size and shape of the end portion allow the electrodes to be substantially spaced apart from the walls of the heart chambers. The intracardiac electrical potential field can be detected during a single heartbeat. According to one example, the sensor electrodes can be distributed on a series of circumferences located in planes spaced apart from each other. These planes can be perpendicular to the long axis of the end portion of the catheter. At least two additional electrodes can be provided adjacent to each other at the ends of the long axis of the end portion. As a more specific example, the catheter can include four circumferences, with eight electrodes spaced equally angularly apart on each circumference. Thus, in this specific embodiment, the catheter can include at least 34 electrodes (32 circumferential electrodes and 2 end electrodes).

[0123] According to another example, an electrophysiological cardiac mapping system and technique based on a non-contact, non-expanding multi-electrode catheter can be realized. EGM can be obtained using a catheter with multiple electrodes (e.g., between 42 and 122 electrodes). According to this specific implementation, knowledge of the relative geometry of the probe and the endocardium can be obtained, for example, through independent imaging modalities such as transesophageal echocardiography. After independent imaging, non-contact electrodes can be used to measure the cardiac surface potential and construct a mapping map from it. The technique may include the following steps (after the independent imaging step): (a) measuring the potential using multiple electrodes positioned on a probe located in the heart; (b) determining the geometric relationship between the probe surface and the endocardial surface; (c) generating a coefficient matrix representing the geometric relationship between the probe surface and the endocardial surface; and (d) determining the endocardial potential based on the electrode potentials and the coefficient matrix.

[0124] According to another example, techniques and devices for mapping the potential distribution of cardiac chambers are available. An intracardiac multi-electrode mapping catheter assembly can be inserted into a patient's heart. The mapping catheter assembly may include a multi-electrode array with an integral reference electrode, or preferably, a mating reference catheter. The electrodes may be deployed in the form of a substantially spherical array. The electrode array may be spatially referenced to points on the endocardial surface via the reference electrode or via a reference catheter in contact with the endocardial surface. A preferred electrode array catheter may carry multiple individual electrode sites (e.g., at least 24). Furthermore, this exemplary technique is achieved by understanding the location of each electrode site in the array and by understanding the cardiac geometry. These locations are preferably determined using techniques of impedance plethysmography.

[0125] According to another example, a process for measuring electrophysiological data in the chambers of the heart can be implemented. The method may partially include positioning a set of active and passive electrodes in the heart, supplying current to the active electrodes to generate an electric field in the heart chambers, and measuring the electric field at the sites of the passive electrodes. The passive electrodes are contained in an array positioned on an inflatable balloon of a balloon catheter. In a preferred embodiment, the array reportedly has 60 to 64 electrodes.

[0126] According to another example, cardiac mapping can be achieved using one or more ultrasound transducers. The ultrasound transducer can be inserted into the patient's heart and can collect multiple ultrasound slices (e.g., two-dimensional or three-dimensional slices) at various locations and orientations within the heart. The location and orientation of a given ultrasound transducer can be known, and the collected ultrasound slices can be stored so that they can be displayed at a later time. One or more ultrasound slices can be displayed corresponding to the location of a probe (e.g., a treatment catheter) at a later time, and the probe can be overlaid on one or more ultrasound slices.

[0127] According to other examples, body patches and / or surface electrodes may be positioned on or near the patient's body. A catheter having one or more electrodes may be positioned within the patient's body (e.g., within the patient's heart), and the location of this catheter may be determined by the system based on signals transmitted and received between the one or more electrodes of the catheter and the body patch and / or surface electrodes. Additionally, the catheter electrodes may sense biometric data (e.g., LAT values) from within the patient's body (e.g., within the heart). This biometric data may be correlated with the determined location of the catheter, enabling the display of a rendering of the patient's body part (e.g., the heart), and the display of biometric data overlaid on the shape of the body part, as determined by the location of the catheter.

[0128] Electrical signals, such as electrocardiogram (ECG) signals, are typically detected before and / or during cardiac procedures. For example, ECG signals can be used to identify the potential location within the heart from which arrhythmias causing the signal originate. Generally, an ECG is a signal describing the electrical activity of the heart. ECG signals can also be used to map parts of the heart. ECG signals are generated by the contraction (depolarization) and relaxation (repolarization) of the atrial and ventricular myocardium. Figure 16As shown in signal 1600, an ECG signal includes a P wave (due to atrial depolarization), a QRS complex (due to atrial repolarization and ventricular depolarization), and a T wave (due to ventricular repolarization). To record an ECG signal, electrodes can be placed at specific locations on the body or positioned within the body via a catheter. Artifacts (e.g., noise) are unwanted signals that are combined with electronic signals such as ECG signals and can sometimes hinder the diagnosis and / or treatment of heart conditions. Artifacts in electrical signals can include baseline drift, electric field interference, electromyography (EMG) noise, electric field noise, etc.

[0129] Additionally, biometric (e.g., biopotential) patient monitors can use surface electrodes to measure biopotentials, such as ECG or EEG. The fidelity of these measurements is limited by the effectiveness of the electrode connection to the patient. The resistance of the electrode system to the flow of current (called impedance) characterizes the effectiveness of the connection. Generally, the higher the impedance, the lower the fidelity of the measurement. Several mechanisms can contribute to lower fidelity.

[0130] Figure 17 This is an illustration of an exemplary system 1720 that can implement one or more features of the subject matter of this disclosure. All or part of system 1720 can be used to collect information for a training dataset, and / or all or part of system 1720 can be used to implement a trained model. System 1720 may include components, such as catheter 1740, configured to damage tissue regions of organs within the body. Catheter 1740 may also be further configured to obtain biometric data including electronic signals. While catheter 1740 is shown as a pointed catheter, it should be understood that catheters of any shape, including one or more elements (e.g., electrodes), can be used to implement embodiments disclosed herein. System 1720 includes a probe 1721 having an axis that can be navigated by a physician 1730 to a body part of a patient 1728 lying on a table 1729, such as the heart 1726. Multiple probes may be provided according to embodiments; however, for brevity, a single probe 1721 is described in this example, but it should be understood that probe 1721 may represent multiple probes. Figure 17 As shown, physician 1730 can insert shaft 1722 through sheath 1723 while manipulating the distal end of shaft 1722 using a manipulator near the proximal end of catheter 1740 and / or deflecting it from sheath 1723. As shown in partial enlarged illustration 1725, catheter 1740 can be fitted to the distal end of shaft 1722. Catheter 1740 can be inserted through sheath 1723 in a collapsed state and can then be deployed within heart 1726. As further described herein, catheter 1740 may include at least one ablation electrode 1747 and catheter needle.

[0131] According to one embodiment, catheter 1740 can be configured to ablate tissue regions of the heart chambers of heart 1726. A partial enlarged illustration 1745 shows catheter 1740 within the heart chambers of heart 1726 in a magnified view. As shown, catheter 1740 may include at least one ablation electrode 1747 coupled to the body of the catheter. According to other embodiments, multiple elements may be connected via a strip forming the shape of catheter 1740. One or more other elements (not shown) may be provided, which may be any element configured to ablate or obtain biometric data, and may be an electrode, a transducer, or one or more other elements.

[0132] According to the embodiments disclosed herein, an ablation electrode, such as electrode 1747, can be configured to deliver energy to a tissue region of an organ in the body, such as heart 1726. The energy can be thermal and can cause damage to the tissue region by starting from the surface of the tissue region and extending into the thickness of the tissue region.

[0133] According to the embodiments disclosed herein, biometric data may include one or more of the following: LAT, electrical activity, topology, bipolar mapping, dominant frequency, impedance, etc. LAT can be a time point corresponding to a threshold activity of local activation, calculated based on a normalized initial starting point. Electrical activity can be any applicable electrical signal that can be measured based on one or more thresholds and sensed and / or amplified based on signal-to-noise ratio and / or other filters. Topology can correspond to the physical structure of a body part or a portion of a body part, and can correspond to variations in the physical structure relative to different parts of the body part or relative to different body parts. Dominant frequency can be a frequency or frequency range that is prevalent in a part of a body part and can differ in different parts of the same body part. For example, the dominant frequency of the pulmonary veins of the heart can differ from the dominant frequency of the right atrium of the same heart. Impedance can be a resistance measurement at a given region of a body part.

[0134] like Figure 17 As shown, probe 1721 and catheter 1740 can be connected to console 1724. Console 1724 may include processor 1741 (such as a general-purpose computer) having suitable front-end and interface circuitry 1738 for transmitting and receiving signals to and from the catheter, as well as other components for controlling system 1720. In some embodiments, processor 1741 may be further configured to receive biometric data, such as electrical activity, and determine whether a given tissue region is conductive. According to one embodiment, the processor may be located external to console 1724 and may be located, for example, in a catheter, an external device, a mobile device, a cloud-based device, or may be a stand-alone processor.

[0135] As noted above, processor 1741 may include a general-purpose computer that can be software-programmed to perform the functions described herein. The software may be downloaded to the general-purpose computer electronically, for example, via a network, or alternatively or additionally set and / or stored on a non-transitory tangible medium, such as magnetic storage, optical storage, or electronic storage. Figure 17 The exemplary configuration shown can be modified to implement the embodiments disclosed herein. The embodiments disclosed herein can be applied similarly using other system components and setups. Additionally, system 1720 may include additional components such as elements for sensing electrical activity, wired or wireless connectors, processing and display devices, etc.

[0136] According to one embodiment, a display 1727 connected to a processor (e.g., processor 1741) may be located in a remote location such as a separate hospital or within a separate healthcare provider network. Additionally, system 1720 may be part of a surgical system configured to acquire anatomical and electrical measurements of a patient's organ (such as the heart) and perform cardiac ablation procedures. An example of such a surgical system is the Carto, sold by Biosense Webster. ® system.

[0137] System 1720 can also, and optionally, use ultrasound, computed tomography (CT), magnetic resonance imaging (MRI), or other medical imaging techniques known in the art to obtain biometric data, such as anatomical measurements of a patient's heart. System 1720 can use catheters, electrocardiograms (EKG), or other sensors that measure the electrical properties of the heart to obtain electrical measurements. Figure 17 As shown, biometric data, including anatomical and electrophysiological measurements, can then be stored in memory 1742 of the mapping system 1720. The biometric data can be transferred from memory 1742 to processor 1741. Alternatively or otherwise, biometric data can be transferred to a server 1760, which may be local or remote, using network 1762.

[0138] Network 1762 can be any network or system known in the art, such as an intranet, local area network (LAN), wide area network (WAN), metropolitan area network (MAN), direct connection or a series of connections, cellular telephone network, or any other network or medium capable of facilitating communication between system 1720 and server 1760. Network 1762 can be wired, wireless, or a combination thereof. Wired connections can be implemented using Ethernet, Universal Serial Bus (USB), RJ-11, or any other wired connection known in the art. Wireless connections can be implemented using Wi-Fi, WiMAX and Bluetooth, infrared, cellular networks, satellite, or any other wireless connection method known in the art. Additionally, several networks can operate independently or communicate with each other to facilitate communication within network 1762.

[0139] In some cases, Server 1760 can be implemented as a physical server. In other cases, Server 1760 can be implemented as a public cloud computing provider (e.g., Amazon Web Services (AWS)). ® () virtual server.

[0140] According to an exemplary embodiment, server 1760 may be implemented as a processor storing machine learning algorithms (such as neural network 1790) or communicating with such a processor. In another embodiment, neural network 1790 may be implemented in console 1724. For example, but not limited to, neural network 1790 may be on one or more CPU processors, one or more GPU processors, one or more FPGA chips, or on an ASIC dedicated to performing deep learning computations (such as Intel). ® Nervana ™ The neural network 1790 is implemented on a neural network processor. According to an exemplary embodiment, the neural network 1790 may be located on a server or processor in a medical operating room, a hospital or medical facility, a remote server or processor, or in the cloud.

[0141] The console 1724 can be connected to a surface electrode 1743 via a cable 1739, which may include an adhesive skin patch attached to the patient 1728. A processor, in conjunction with a current tracking module, determines the orientation coordinates of the catheter 1740 within a body part of the patient (e.g., the heart 1726). The orientation coordinates may be based on impedance or electromagnetic field measured between the surface electrode 1743 and other electromagnetic components of the electrode 1747 or catheter 1740. Additionally or alternatively, a positioning pad may be located on the surface of the stage 1729 and may be detachable from the stage 1729.

[0142] Processor 1741 may include a real-time noise reduction circuitry system typically configured as a field-programmable gate array (FPGA), followed by an analog-to-digital (A / D) ECG (electrocardiogram) or EMG (electromyography) signal conversion integrated circuit. Processor 1741 may pass signals from the A / D ECG or EMG circuitry to another processor and / or may be programmed to perform one or more functions disclosed herein.

[0143] The console 1724 may also include an input / output (I / O) communication interface that enables the console to transmit signals from and / or to the electrode 1747.

[0144] During the procedure, processor 1741 may facilitate the presentation of a body part rendering 1735 to physician 1730 on display 1727 and store data representing the body part rendering 1735 in memory 1742. Memory 1742 may include any suitable volatile and / or non-volatile memory, such as random access memory or hard disk drive. In some embodiments, physician 1730 may be able to manipulate the body part rendering 1735 using one or more input devices, such as a touchpad, mouse, keyboard, gesture recognition device, etc. For example, the input device may be used to change the position of catheter 1740, causing the rendering 1735 to be updated. In an alternative embodiment, display 1727 may include a touchscreen that can be configured to accept input from physician 1730 in addition to presenting the body part rendering 1735.

[0145] Processor 1741 can acquire multiple electrical signals from catheter 1740, ECG, or other sensors that measure the electrical properties of the heart, as described above. Processor 1741 can apply algorithms for optimal heartbeat selection. In some embodiments, the algorithm may include dynamic filtering of the electrical signals to determine which electrical signals are integrated into "points" in a "point cloud" representing desired anatomical structures, such as the chambers of the heart.

[0146] The algorithm may include multiple filters. Filters may include, but are not limited to: position stability, internal distance, catheter filter, cycle length, cycle length stability, ventricular activity, position density, respiration, pattern matching, LAT, LAT stability, unipolar slope, bipolar voltage, tissue proximity indicator (TPI), graded stability, and noise level. However, this list is not exhaustive, and dynamic filters may utilize other parameters.

[0147] In some implementations, the algorithm may include two filtering phases: Phase 1 and Phase 2. In Phase 1, one or more first filters may be applied to all collected electrical signals. First filters may include, but are not limited to: position stability, internal distance, catheter filtering, cycle length, ventricular activity, position density, respiratory cycle indication, and pattern matching. If the electrical signals meet certain criteria in Phase 1, one or more second filters may be applied in Phase 2. Second filters may include, but are not limited to: relative LAT stability, slope of unipolar signals, bipolar voltage, TPI, and graded stability.

[0148] In some implementations, one or more parameters can be calculated in phase two based on the characteristics of the signal. For example, in phase two, signals can be analyzed to determine whether they have a LAT (Low Amplitude Attack). Furthermore, bipolar voltages can be calculated and analyzed. Signals can also be analyzed to determine if the electrodes are close to tissue. Force values ​​(in grams) can also be analyzed. The relative position of points within a respiratory cycle can also be analyzed. In some implementations, the algorithm determines whether the bipolar voltage is high or low. For example, a signal can be considered to have a “high” bipolar voltage if the bipolar voltage is greater than or equal to 0.1 mV, and a signal can be considered to have a “low” bipolar voltage if the bipolar voltage is less than 0.1 mV. However, as those skilled in the art will understand, the “high” and “low” thresholds for bipolar voltage can vary based on many factors, which may be patient-specific or non-patient-specific. Thresholds can be input into the system by a physician during the case (e.g., via a graphical user interface). If a signal is determined to have a “high” bipolar voltage with a LAT, the unipolar slope and relative LAT stability of the signal can be calculated and analyzed. It can calculate and analyze the unipolar slope, relative LAT stability, and TPI of a signal.

[0149] This algorithm can determine whether a signal is fragmented. It can also calculate the signal's TPI (Total Productivity). Furthermore, it can calculate the signal's fragmentation stability. Each of these properties can be an independent variable in machine learning.

[0150] Multiple second filters can be applied to signals passing through the first filter. In some implementations, the second filter applied to each signal is based on the characteristics of each signal. For example, the following filter can be applied to a signal determined to have LAT and high voltage with one or more of the following characteristics: a unipolar slope greater than or equal to 0.03 mV / ms and a relative LAT stability of 3 ms. The provided thresholds are given by way of example only, and various other thresholds may be utilized in the algorithm. In addition to having a unipolar slope greater than or equal to 0.03 mV / ms and a relative LAT stability of 3 ms, determining signals with LAT and low voltage may also require a TPI, which indicates the signal collected when catheter 14 contacts tissue to pass through the second filter.

[0151] Machine learning can learn, for example, that signals identified as lacking LAT and not to be graded may require TPI (Transient Pressure Index), which indicates the point at which the signal is collected when the catheter contacts the tissue and is considered a “good” point. In addition to having a TPI indicating contact, signals identified as lacking LAT and to be graded may require satisfactory grading stability to be considered a “good” point.

[0152] Signals that do not pass through the second filter can be discarded. Signals that pass through the second filter can be selected as points in the point cloud to create EP mapping maps of target mapping sites (such as the ventricles).

[0153] According to one embodiment, the neural network 1790 can be provided for selecting and detecting the optimal heartbeat as a cardiac mapping annotation at each spatial location, and for improving the determination of cardiac mapping annotations. According to an exemplary embodiment, the neural network 1790 receives input data to train the neural network. The input data is preferably a set of attributes related to the heartbeat. Non-limiting examples of input data or attribute data may include: - ECG signals received from the mapping electrodes surrounding the reference annotations in the EP mapping diagram; - ECG signals received by one or more body surface electrodes surrounding the reference annotation; - Intracardiac EGM signals received by electrodes or bipolar electrode pairs of a catheter (such as a mapping catheter), wherein the EGM signal may be a bipolar or unipolar EGM of the electrodes of the mapping catheter surrounding a reference annotation in the EP mapping. - Spatial location of the mapping electrode at a reference annotation, which may or may not have respiratory compensation; - Whether the heartbeat is incorporated into the EP mapping (e.g., the heartbeat passes through the filter of the heartbeat selector algorithm); - Local annotation time of the acquired heartbeats, including any physician repairs; - The TPI of the calibration electrode as per the reference notes; - The force value detected by the force sensor of the calibration catheter when referring to the annotation (usually measured in grams); - Location of reference notes within the respiratory cycle during lung movement monitoring; - Breathing gating state is output as a Boolean value; - The difference between the current respiratory cycle length and the previous cycle length; - The ratio of the current respiratory cycle length to the previous cycle length; - The difference between the current respiratory cycle length and the average or median cycle length; - The ratio of the current respiratory cycle length to the average or median cycle length; - The spatial distance between the mapping electrode at the current reference annotation and the same electrode at the previous reference annotation, which can be measured by the actual location or the breathing compensation location; - Is there any surface activation of the body during this heartbeat (i.e., V (Indication of interference); - A discrete Boolean value (i.e., 0 or 1), which serves as an input indicating whether the physician has removed the heartbeat from the mapping; and - Any other EP data measured by the electrodes of the catheter.

[0154] In one implementation, one or more input data 1810 are fed into the neural network 1790. The input data 1810 may be stored in various locations, including but not limited to hospitals or medical facilities, remote server locations, or the cloud, but not limited to these. When the system 1720 is idle, training data may be manually or automatically transferred to the storage device associated with the neural network in real time at predetermined intervals upon request. Therefore, the neural network 1790 can learn in real time or at predetermined times or intervals.

[0155] In one implementation, the output of the neural network 1790 may include, but is not limited to: - Determine, such as a discrete Boolean value, whether the determination indicates whether the current heartbeat has an improved property compared to the previous heartbeat; and - Determine accurate cardiac mapping annotations.

[0156] In the prior art, when the EA point in the cardiac EP mapping system is not good enough (e.g., because the ECG at that EA point is noisy), the physician deletes the EA point and acquires a new EA point at approximately the same spatial location.

[0157] According to an exemplary embodiment of this application, a machine learning algorithm is trained using a pair of deleted and reacquired Explanation Points (EAs) obtained at approximately the same spatial location (i.e., the same location, up to a certain tolerance). Given two EAs, the machine learning algorithm learns that the reacquired EAs have better features than the deleted EAs. Once the machine learning algorithm is trained with sufficient data (given any pair of EAs), it can predict which EA has better features. The EP (Extended Precipitation) mapping system can use this information to automatically replace EAs with inferior features during EP testing. The point is automatically used in the mapping process, ignoring any other points acquired at that location, before the mapping system acquires the point at the spatial location and uses it automatically. In this system, multiple points can be acquired, and the system can determine which points are applied to the mapping system. The input to the machine learning algorithm preferably includes one or more of the aforementioned input data.

[0158] According to an exemplary embodiment, neural network 1790 may include a convolutional neural network (CNN) or a recurrent neural network (RNN), such as a long short-term memory (LSTM) neural network. A convolutional neural network (CNN) is a deep learning algorithm preferably used in the fields of computer vision and / or image recognition. A CNN assigns importance (learnable weights) to various aspects or features in an input image so that they can be distinguished from each other. An LSTM neural network is a recurrent neural network with feedback connections for deep learning.

[0159] In one implementation, the EP mapping system supplier may ship a system with a pre-trained network. Hospitals may preferably continue training the system. In one implementation, a single model may be maintained for all hospitals or a group of hospitals, or each hospital may maintain its own model.

[0160] In an exemplary implementation, after each training iteration, the training model (including its output) can be executed against a standard database (such as a gold standard database) to verify its accuracy. In an exemplary implementation, if the accuracy of a newly trained model is below a threshold, or alternatively, if the accuracy of a newly trained model is lower than that of a previous model, the model can be discarded. Similarly, if the accuracy of a newly trained model is equal to or higher than a threshold, or alternatively, if the accuracy of a newly trained model is higher than that of a previous model, the model can be released to a field labeling system. In an exemplary implementation, the release of a new model can be performed manually, for example, by an operator downloading a file from a URL and uploading it to system 1720. Alternatively, the new model can be pushed to system 1720 in the field via the Internet.

[0161] In endocardial mapping (EP), an in vivo probe is typically inserted into the heart cavity, using a catheter with multiple mapping electrodes positioned near the distal end of the catheter body. Time-varying ECG signals are recorded at multiple contact points between the mapping electrodes and the heart tissue. The multiple ECG electrodes are then moved to different contact locations within the heart tissue, and this process is repeated. A measure of cardiac function is then calculated from the local ECG signals, which are spatially mapped across the entire surface of the heart cavity. This mapping helps medical professionals identify areas of cardiac dysfunction.

[0162] Power sources in the heart, such as the sinoatrial (SA) and atrial-ventricular (AV) nodes, initiate electrical activity waves that propagate across the heart, triggering the contraction of muscular tissue in the atria and ventricles into a characteristic sinus rhythm. Characteristic ECG waveforms are detected at multiple mapping electrodes as the active wavefront arrives at them during each cardiac cycle. These waveforms are time-shifted due to the different arrival times of the same wavefront at different electrodes that contact the tissue at different spatial locations along the surface of the heart chambers.

[0163] The arrival times of ECG waveforms detected at multiple mapping electrodes can be used to map the propagation time and / or velocity of active waves on the heart. Mapping of active waves is performed relative to a single time reference indicating the cardiac cycle (referred to as the reference annotation time).

[0164] Reference annotation time can be calculated by processing ECG signals obtained from a surface electrode or an IC reference electrode on an additional catheter and placed in contact with the surface of the heart chambers. Typically, the physician specifies whether the reference annotation time is calculated from the BS channel or the IC channel based on the suspected pathology. However, if the ECG reference channel assigned during the mapping procedure fails (e.g., due to poor reference electrode contact, system noise, or other damage), remapping is required. Cardiac remapping is time-consuming and uncomfortable for the patient.

[0165] Figure 20 This is an exemplary illustration of multiple electrodes sampling multiple contact points in a cardiac chamber according to an embodiment of this application. Figure 20 In the exemplary embodiment shown, an IC catheter 2024A is inserted into the coronary sinus 2062 of the heart 2034. The catheter preferably includes five reference ECG electrodes 2060 A-2060 E, positioned along the length of the body of the catheter 2024A near the distal end 2065. The reference ECG electrodes 2060 A-2060 E contact multiple tissue points on the surface of the coronary sinus and are used to measure the corresponding reference ECG signals at the multiple contact points.

[0166] Similarly, IC catheter 2024B is inserted into the right ventricle of heart 2034. The catheter includes five mapping electrodes 2070 A-2070 E at a point near the distal end 2075 along the body of catheter 2024B. Mapping electrodes 2070 A-2070 E contact multiple tissue points on the surface of the right ventricle and are used to measure the corresponding mapping ECG signals at the multiple contact points.

[0167] Figure 20 The five reference ECG electrodes 2060 A-2060 E and the multiple mapping electrodes 2070 A-2070 E shown are merely for clarity of concept and not for limiting the scope of the claims. In alternative embodiments, any suitable number of mapping and reference electrodes may be used. For example, instead of a single electrode for unipolar ECG detection, the electrodes may be arranged in pairs for bipolar ECG detection. Any suitable number of mapping and reference catheters may be used in any suitable configuration. Reference catheter 2024A and mapping catheter 2024 B, or any number of catheters, may be navigated to a suitable location within the heart 2034 to perform the functions described herein.

[0168] ECG signal interface, such as processor 1741 ( Figure 17 As shown), it receives and processes signals from surface electrodes, such as surface electrode 1743 ( Figure 17 (As shown), IC ECG reference electrodes 2060 A-2060 E and mapping electrodes 2070 A-2070 E. Cardiac mapping is performed by moving multiple mapping electrodes 2070 A-2070 E across the entire surface of the cardiac tissue and recording the ECG at each contact point and the position of the electrodes while recording the ECG waveform. Alternatively, cardiac mapping may include inserting the reference electrodes and / or mapping electrodes from outside the patient's body, for example through the pleural cavity, to contact the surface of the cardiac tissue.

[0169] The data recorded at each mapping point is also referred to as the mapping annotation. However, because the activation wavefront propagates to different electrodes at different spatial points on the endocardium, the local ECG signal arrives at different points at different times, and therefore at different electrodes. The difference in the arrival time of the ECG signal at a given ECG electrode indicates the velocity of the local activation wavefront. The difference between the measurement time of the ECG signal arriving at a specific mapping electrode for a specific cardiac cycle or heartbeat and a single timing reference indicating the timing of the cardiac cycle is referred to herein as the LAT. Similarly, the single timing reference is referred to as the reference annotation time.

[0170] A typical cardiac mapping includes a LAT mapping at multiple different points on the surface of the heart, a propagation mapping showing the activation wavefront at different times across the heart, and a characteristic voltage of the ECG at the same given point. Users (such as physicians) can then use the changes in the expected activation wavefront and / or voltage shown in the cardiac mapping to detect areas of cardiac dysfunction, such as atrial or ventricular tachycardia. For example, ablation therapy can be used to correct dysfunction.

[0171] In some implementations, processor 1741 uses the acquired reference ECG signal to calculate a reference annotation time from surface electrode 1743 and / or from IC reference electrodes 2060 A-2060 E. IC mapping ECG signals are acquired as the mapping electrodes move across the cardiac tissue, and ECGs are recorded to acquire mapping data points. However, multiple IC reference electrodes are in contact with the cardiac tissue and do not move during medical procedures. Depending on the type of expected pathology, physicians typically specify whether the reference annotation time is calculated from ECGs acquired from surface electrode 1743 or from IC reference electrodes 2060 A-2060 E.

[0172] Example 1 In an exemplary implementation, a machine learning algorithm such as a neural network may use attributes of two heartbeats as input, learn to identify which heartbeat exhibits better features based on predetermined criteria, and output a binary result, as described below relative to... Figures 1 to 2 A more detailed description. 18 and Figure 19 As shown. Exemplary attributes used to identify which heartbeat is better include, but are not limited to, heartbeats with less noise, more pronounced LAT activation, more pronounced hysteresis potential, more stable fragmentation, and less V interference. In one embodiment, the heartbeat is derived from a stable arrhythmia.

[0173] Conventional EP mapping systems either acquire the first heartbeat at each spatial location or use a rule-based system to select the optimal heartbeat for each spatial location. For example, when a physician acquires the first heartbeat at a specific spatial location, subsequent heartbeats are prevented from being acquired even if the first heartbeat exhibits poor characteristics. The system and method disclosed in this example collect multiple heartbeats at a specific spatial location and learn to select the optimal heartbeat for each spatial location via machine learning.

[0174] Figure 18 This is an exemplary embodiment of the CNN 1800 according to this application, which can receive the attributes of two heartbeats and learn to identify which heartbeat is better.

[0175] See Figure 18 In an exemplary embodiment, neural network 1800 is a CNN. CNN 1800 preferably receives input data 1810. In one embodiment, input data 1810 preferably includes a first set of input data related to attributes of a first heartbeat and a second set of input data related to a second heartbeat obtained at the same spatial location as the first heartbeat. The first and second sets of input data may include inputs In-1, In-2, In-3, ..., In-n, where "n" is the last of the plurality of inputs. For example, the first set of input data may include inputs In-1 and In-2, and the second set of input data may include In-3 and In-n. The first and second sets of input data may be input into the CNN during the same period or during separate periods. Input data 1810 may include attributes of the heartbeat, which preferably include, but are not limited to: - Intracardiac EGM signals received by electrodes or bipolar electrode pairs of a catheter (such as a mapping catheter), wherein the EGM signal may be a bipolar or unipolar EGM of the electrodes of the mapping catheter surrounding a reference annotation in the EP mapping. - ECG signals received from the calibration electrodes surrounding the reference annotation; - ECG signals received by one or more body surface electrodes surrounding the reference annotation; - The TPI of the calibration electrode as stated in the reference notes; - The force value detected by the force sensor of the calibration catheter when referring to the annotation (usually measured in grams); - Spatial position or distance vector of the mapping electrode at the reference note, which may or may not have respiratory compensation; - A respiratory state vector surrounding the aforementioned reference annotation; - When monitoring lung movement, the reference notes are positioned within the respiratory cycle; - The difference between the current respiratory cycle length and the previous cycle length; - The ratio of the current respiratory cycle length to the previous cycle length; - The difference between the current respiratory cycle length and the average or median cycle length; - The ratio of the current respiratory cycle length to the average or median cycle length; - Should the physician manually input the heartbeat into the EP mapping system or delete the heartbeat instruction? - The distance between the mapping electrode at the current reference annotation and the spatial position of the same electrode at the previous reference annotation, which can be measured by the actual position or the respiratory-compensated position. In other words, the distance the electrode has moved since the last heartbeat serves as an indicator of catheter stability, for example, based on the actual position or the respiratory-compensated position.

[0176] Input data 1810 is provided to a first hidden layer 1820 comprising nodes HL1-1, HL1-2, HL1-3, ... and optionally, a second or more hidden layers 1830 HL2-n comprising nodes HL2-1, HL2-2, HL2-3, ... are combined to produce output 1840. Hidden layers 1820 and 1830 utilize the data from the input to determine which of the first or second heartbeats has better features, such as, but not limited to, a heartbeat with less noise, more pronounced LAT activation, more pronounced hysteresis potential, more stable fragmentation, less V-interference, etc. Over time, the neural network learns weights to apply to the input data.

[0177] In some implementations, output 1840 is a value indicating, for example, whether the first or second heartbeat in a pair of heartbeats captured at the same spatial location is better. For example, in CNN 1800, all input data 1810 is immediately fed into the neural network to compute output 1840.

[0178] Figure 19 This is an exemplary flowchart illustrating a method according to this application for training a neural network 1900 to obtain the properties of two heartbeats and learning to identify which heartbeat is better.

[0179] At 1910, CNN 1800 preferably receives from a database a pair of deleted and undeleted heartbeats obtained at the same spatial location. The received data is the input data 1810 of CNN 1800, and preferably includes an attribute dataset corresponding to each heartbeat in the heartbeat pair obtained at the same spatial location. For example, the attribute dataset can be any of the input data 1810 described above for each heartbeat.

[0180] At position 1920, the hidden layers 1820 and 1830 of CNN 1800 preferably compare the attribute information sets between deleted and undeleted heartbeats to determine which of the first or second heartbeats has better features based on predetermined criteria, such as, but not limited to, a heartbeat with less noise, more pronounced LAT activation, more pronounced late potentials, more stable fragmentation, less V interference, manual physician omission, etc. For example, if the first heartbeat was manually deleted by a physician and the second heartbeat was manually acquired by a physician, then CNN 1800 considers the second heartbeat to have better features than the first heartbeat. CNN 1800 compares the attribute information associated with each heartbeat to learn the optimal heartbeat features.

[0181] At 1930, the CNN 1800 output indicates whether the undeleted heartbeats have better features than the deleted heartbeats.

[0182] At 1940, CNN 1800 preferably repeats 1910, 1920, and 1930 for another pair of deleted and undeleted heartbeats obtained at the same spatial location until CNN 1800 is trained.

[0183] In one implementation, once the CNN 1800 is trained, it can be used to replace suboptimal heartbeats in the EP mapping system with heartbeats that have better features either in real-time or after procedural processing at the same spatial location. For example, the EP mapping system initially acquires a first heartbeat at spatial locations X, Y, Z. When another heartbeat is recorded at the same spatial location, the system uses the attribute data of each heartbeat as input data to perform a CNN and outputs a determination of which heartbeat has better features. If the first heartbeat is determined to be better, the second heartbeat is discarded from the EP mapping system. If the second heartbeat is determined to be better, the first heartbeat is automatically replaced with the second heartbeat in the EP mapping system.

[0184] The training model described above is preferably supervised by doctors at the hospital or through remote training facilities.

[0185] According to this application, the advantage of using a CNN 1800 to receive attribute data from two heartbeats and learn to identify which heartbeat is better is that the CNN 1800 can learn to simulate a physician's decision by reviewing heartbeats that have been removed by the physician. Furthermore, by determining the optimal heartbeat at a spatial location, physicians can more accurately determine heartbeat characteristics such as LAT value, peak-to-peak bipolar voltage value, ECG signal, and EGM signal.

[0186] In the above implementation, the machine learning takes two heartbeats and determines which heartbeat is better. The system is trained with a pair of heartbeats (a deleted heartbeat and a re-acquired heartbeat at approximately the same spatial location in the heart). Now, we describe another implementation for the same purpose. In this implementation, each heartbeat is assigned a grade. For example, in one implementation, the first and second heartbeats may each be assigned a grade of 0, 0.5, or 1. If the physician deletes the heartbeat, it receives a grade of 0; if the physician acquires a new heartbeat at approximately the same location, it receives a grade of 1; and all other heartbeats receive a grade of 0.5. Unlike the aforementioned implementation that takes two heartbeats and predicts which is better, this implementation takes a set of attributes from a single heartbeat and predicts the grade of that single heartbeat. If the heartbeat has characteristics of a heartbeat that the physician would typically delete, the machine learning will likely output a number of approximately 0. If the heartbeat has characteristics of a heartbeat that the physician would typically re-acquire after deleting a bad heartbeat, the machine learning will likely output a number of approximately 1. In this implementation, the grades are continuous variables (not discrete categories). Although we train the machine learning system with only levels 0, 0.5, and 1, the machine learning system can output any real number. In this implementation, during runtime (in other words, during the inference phase, while the physician is performing electrophysiological studies on the patient), when the system needs to understand which heartbeat is better, it can independently predict the level of each heartbeat, and the heartbeat with the highest level can be considered the best heartbeat.

[0187] Example 2 In an exemplary implementation, a machine learning algorithm such as a neural network may receive EP data representing multiple heartbeats obtained at the same spatial location and learn to select and output the best mapping annotation for the EP mapping system based on any of the above inputs.

[0188] In traditional rule-based EP calibration systems, calibration annotations are typically determined based on either the highest peak voltage 2201 or the highest negative derivative 2202 of the ECG signal surrounding the reference annotation. Figure 22As shown. Physicians can accept the mapping annotation, or move it if the results are unsatisfactory. Examples of reasons why physicians may move mapping annotations include, but are not limited to, significant noise in the heartbeat, less obvious LAT, less obvious late potentials, less stable fragmentation, more V interference, etc.

[0189] According to the subject matter of this application, a neural network can receive attributes of multiple heartbeats at the same spatial location and learn to identify the best labeling annotation based on the aforementioned input data. In one implementation, the heartbeats are derived from stable arrhythmias.

[0190] In one embodiment of the subject matter of this application, initially, a neural network or machine learning system is trained to find the same mapping annotations computed by a rule-based system. The rule-based system uses multiple rules to select actions. These multiple rules can be defined by a human programmer. In some embodiments, calcium is in the form of free ions. For example, initially, many ECG signals and corresponding mapping annotations computed by the rule-based system for each ECG signal are fed into the machine learning system to train it. This is expected to make the machine learning system nearly as accurate as the rule-based system. When physicians at different locations begin using the system, the initial results will be similar to those of a rutile-based system. However, whenever a physician manually fixes the mapping annotations, this information is used to train the machine learning system, which is preferably a self-learning system. In one embodiment, a single model can be maintained for all hospitals, or each hospital or group of hospitals can maintain its own model. After the machine learning system has been trained, its performance can be validated against a gold standard to verify that its accuracy is above a predetermined threshold and / or superior to the accuracy of previous models. Over time, the machine learning system will enable the machine learning algorithm to outperform the rule-based system.

[0191] See Figure 21 In an exemplary embodiment, the neural network is an RNN 2100. The RNN 2100 preferably receives input data 2110. In one embodiment, the input data 2110 may include attribute data for each of a plurality of heartbeats 2110-1, 2110-2, 2110-3, ... obtained from the same spatial location. For example, the attribute data may be derived from... Figure 20 The mapping electrode shown is one of 2070 A-2070 E. The attribute data for each heartbeat can include any of the attributes described herein. Instead of providing attribute data for a single heartbeat, the RNN receives all attribute data associated with each heartbeat per cycle. Furthermore, as referenced... Figure 21 As indicated by arrow 2140, if the accuracy of the newly trained model is equal to or higher than the threshold, or alternatively, if the accuracy of the newly trained model is higher than the accuracy of the previous model, then the model can be used as input to the neural network.

[0192] Input data 2110 is fed into RNN 2100 and combined to produce output 2130, such as a Boolean value or a matching index. As more EGM samples are fed into RNN 2100, output 2130 becomes more accurate.

[0193] In one implementation, to train the model, the physician accepts heartbeats as mapping annotations, which can be input to an RNN 2100, along with other heartbeats obtained from the spatial location. It is expected that the physician accepting the heartbeats will have better characteristics than other heartbeats obtained at the same spatial location, and the RNN 2100 outputs the physician accepting the heartbeats as the optimal label annotation. The RNN 2100 compares the attribute information associated with each heartbeat with the heartbeats accepted by the physician to learn the optimal heartbeat features. Over time, the RNN 2100 learns features of the attributes of the physicians accepting heartbeats to train the model.

[0194] The output of an RNN, such as output 2130, can be used to train an RNN 2100. For example, see reference... Figure 21 As indicated by arrow 2140, if the accuracy of output 2130 is equal to or higher than the threshold, then output 2130 can be used as input data 2110 for neural network 2100.

[0195] For example, as described above, after each training iteration, the trained model (including its output) can be executed against a standard database (such as a gold standard database) to verify its accuracy. In an exemplary implementation, if the accuracy of the newly trained model is below a threshold, or alternatively, if the accuracy of the newly trained model is lower than that of the previous model, the model can be discarded. Similarly, if the accuracy of the newly trained model is equal to or higher than a threshold, or alternatively, if the accuracy of the newly trained model is higher than that of the previous model, the model can be released to a field labeling system.

[0196] In one implementation, once the RNN 2100 is trained, it can be used to improve the annotations in the EP annotation system in real time or after the procedure.

[0197] In one implementation, the training of the neural network 2100 can be supervised at a facility where cardiac procedures are performed (such as a hospital or medical facility) or at a remote location (such as a training center).

[0198] Example 3 In an exemplary implementation, a machine learning algorithm (such as a neural network) may receive data representing any manual corrections (such as corrections performed by a physician) to heart rate and mapping annotations, and learn to compute the mapping annotations of the EP mapping system.

[0199] In traditional mapping and annotation systems, such as rule-based systems, mapping and annotation are determined based on intracardiac ECG activation. ECG activation is typically determined in several ways, such as, but not limited to, the onset of bipolar EGM, the timing of maximum bipolar amplitude, or by combining various features of the ECG signal, such as those described in U.S. Patent Publication 2018 / 00042504. The attending physician needs to review the determined mapping and annotation for accuracy. If the mapping and annotation is inaccurate, the physician must manually move the mapping and annotation to fix it.

[0200] Figure 23 This is an exemplary flowchart illustrating a method for training a machine learning system, such as a neural network, to compute mapping annotations in intracardiac signals, such as EGM signals, according to this application. In one embodiment, the neural network can be any kind of neural network, such as a CNN or RNN, similar to the reference. Figure 18 and Figure 21 Those mentioned above.

[0201] Figure 23 Sections 2310-2316 illustrate preferred systems and methods for training an initial model for a machine learning system according to one embodiment. At 2310, the machine learning system preferably receives input data associated with an EP signal. For example, the input data may include, but is not limited to, one or more of the following: - Bipolar EGM signal around the reference annotation of the mapping catheter; - The mapping catheter shows a unipolar EGM signal around the reference annotation; - ECG signals received by one or more body surface electrodes surrounding the reference annotation; - The starting point of the Window of Interest (WOI); - The end of the WOI; - The spatial location of the mapping electrodes around the reference annotation, such as 3D coordinates; - The calibration electrode surrounds the TPI of the reference annotation; - The force value detected by the force sensor of the calibration catheter when referring to the annotation (usually measured in grams); - A respiratory state vector surrounding the aforementioned reference annotation; - When monitoring lung movement, the reference notes are positioned within the respiratory cycle; - The difference between the current respiratory cycle length and the previous cycle length; - The ratio of the current respiratory cycle length to the previous cycle length; - The difference between the current respiratory cycle length and the average or median cycle length; - The ratio of the current respiratory cycle length to the average or median cycle length; - Should the physician manually input the heartbeat into the EP mapping system or delete the heartbeat instruction? - The distance between the mapping electrode at the current reference annotation and the spatial position of the same electrode at the previous reference annotation, which can be measured by the actual position or the respiratory-compensated position. In other words, the distance the electrode has moved since the last heartbeat serves as an indicator of catheter stability, for example, based on the actual position or the respiratory-compensated position.

[0202] At 2312, a rule-based algorithm is preferably applied to the input data, and at 2314, the labeled annotations are computed using the rule-based algorithm. At 2316, the labeled annotations are output as an initial model for further training of the machine learning system. The initial model is preferably trained to simulate the rule-based algorithm.

[0203] In some implementations, the signal can be shifted to obtain a shifted signal. At 2310, the filtered signal can be used as additional input data. For example, if data corresponding to a bipolar signal and two unipolar signals are used as input data, the signal can be shifted 1 ms to the left, 2 ms to the left, 3 ms to the left, 4 ms to the left, 5 ms to the left, 6 ms to the left, 7 ms to the left, 8 ms to the left, 9 ms to the left, 10 ms to the left, 1 ms to the right, 2 ms to the right, 3 ms to the right, 4 ms to the right, 5 ms to the right, 6 ms to the right, 7 ms to the right, 8 ms to the right, 9 ms to the right, and 10 ms to the right. Thus, in this example, 21 different signals can be obtained. When the shifted signals are fed into the neural network, the neural network can learn that the signal is time-invariant. This allows training to be completed with less input data.

[0204] In some implementations, the signal can be enhanced by adding low-frequency noise and / or high-frequency noise. Low-frequency noise can represent the effects of breathing, and high-frequency noise can represent electromagnetic noise. For example, the following four signals can be obtained from a single signal: the signal itself, a signal with added 1Hz noise, a signal with added 450Hz noise, and a signal with added 1Hz and 450Hz noise. In this way, the convolutional layers of the CNN can learn to ignore low-frequency and high-frequency noise more quickly and effectively.

[0205] In some implementations, a combination of time offset and noise addition can be used. For example, the signal can be shifted left by 1 ms with the addition of 1 Hz noise, shifted left by 2 ms with the addition of 1 Hz noise, and so on.

[0206] Figure 23Sections 2320 to 2328 illustrate an implementation scheme for continuously improving training of a deployment model for a machine learning system, according to the implementation scheme. At 2320, the deployment model is deployed to an EP labeling system in an environment such as a hospital or medical testing facility. In one implementation scheme, the deployment model can be an initial model or a new model.

[0207] In some implementations, the deployed model can be downloaded from a central server. The data downloaded from the central server can be signed with a private key, known only to the central server. The public key corresponding to the aforementioned private key can be known by the medical device downloading the deployed model from the central server. In some implementations, a public / private key mechanism in HTTPS or a similar protocol (such as SFTP) can also be used to ensure the security of the download. In these implementations, the medical device can check the URL address of the central server and the validity of the Secure Sockets Layer (SSL) certificate.

[0208] At point 2322, the EP mapping system uses the deployed model to compute mapping annotations in real time. For example, but not limited to, when the deployed model is an initial model, a rule-based algorithm is used to compute mapping annotations.

[0209] At point 2324, if the mapping annotation is inaccurate, the physician can manually fix the mapping annotation into the EP mapping system. The machine learning system preferably receives data related to the mapping annotation manually fixed by the physician in the EP mapping system. The physician's corrections can be fed into the machine learning system in real time via the cloud or via offline devices such as Memory Stick, CD, or Digital Universal Optical Disc (DVD). In one implementation, the physician's corrections can be manually checked before being fed into the machine learning system. Figure 24 The EGM signal A 2410 in the example shows a calibration annotation fixed by a rule-based system. Figure 24 The EGM signal B 2420 in the image shows an example of a mapping annotation manually fixed by a physician.

[0210] In some implementations, physician corrections may be signed by the medical device and then fed into a machine learning system. The machine learning system can check the physician corrections to ensure they indeed originate from the medical device. In some implementations, private / public keys may be used. For example, the private key may be known to the medical device, and the public key may be known to the machine learning system. In another implementation, the private key may be stored in a temperature-resistant hardware security module on the medical device. Alternatively, the medical device's private key may be stored within a trusted platform module embedded in the motherboard of the medical device. In another implementation, the private / public key method may be combined with an IP whitelist (e.g., private key P1 should come from IP address xxx.x.xxx.x). In some implementations, multiple medical devices contain the same private key. In other implementations, different private keys may be deployed to each medical device. In some implementations, to enhance security, users may be required to provide a username and password when uploading data. The username and password may be combined with the private key method and / or an IP whitelist.

[0211] At 2326, the machine learning system preferably further trains the deployed model by using new EP signals and mapping annotations manually fixed by physicians in the EP mapping system.

[0212] At point 2326, the machine learning system preferably further trains the deployed model by using new EP signals and label annotations fixed by the physician. As mentioned above, initially, the machine learning system will compute the mapping annotations based on the results of a rule-based algorithm. However, when the neural network receives more label annotations manually fixed by the physician, the machine learning system will learn from the corrections made by the physician and compute the label annotations more accurately.

[0213] At 2328, the machine learning system preferably outputs a new model. In one embodiment, the output may further include at least one of a label annotation and a Boolean value indicating whether the label annotation can be computed. For example, the machine learning system may recognize that under certain conditions, such as insufficient force, when the conduit is unstable, etc., the label annotation cannot be computed. The Boolean output may be implemented as a neuron that outputs a real number. Optionally, a logit function may be applied to the output number, wherein the Boolean value is assumed to be true when the number is above a predetermined threshold, and the logit function is considered incorrect when the number is below the predetermined threshold. In one embodiment, a user may set the predetermined threshold, for example, by manipulating a slider in a graphical user interface.

[0214] In an exemplary implementation, after each training iteration, the trained model (including its output) can be executed against a standard database (such as the gold standard database) to verify its accuracy. In an exemplary implementation, if the accuracy of the newly trained model is below a threshold, or alternatively, if the accuracy of the newly trained model is lower than that of a previous model, the model can be discarded. Similarly, if the accuracy of the newly trained model is equal to or higher than a threshold, or alternatively, if the accuracy of the newly trained model is higher than that of a previous model, the model can be released to the field labeling system. Once the new model has been validated, it can be deployed to the EP labeling system at 2320.

[0215] If the accuracy of the newly trained model is worse than that of the previous model, but still above a threshold, the results of the new training model can be sent to human experts. Human experts can then compare the expected results from the gold standard database with the results calculated by the new training model. If the human experts believe that the results of the new training model are superior to the gold standard, they can mark the results of the new training model as the expected results, thus updating the gold standard database.

[0216] In one implementation, once the machine learning system is trained, it can be used to improve the annotations in the EP annotation system in real time or after the procedure.

[0217] In one implementation, the training of the neural network 1800 can be supervised at a facility where cardiac procedures are performed (such as a hospital or medical facility) or at a remote location (such as a training center).

[0218] Although the features and elements have been specifically described above, those skilled in the art will recognize that each feature or element can be used alone or in any combination with other features and elements. Furthermore, while the process steps have been described in a specific order above, these steps can be performed in other desired orders.

[0219] The methods, processes, modules, and systems described herein may be implemented in a computer program, software, or firmware incorporated in a computer-readable medium for execution by a computer or processor. Examples of computer-readable media include electronic signals (transmitted via wired or wireless connections) and computer-readable storage media. Examples of computer-readable storage media include, but are not limited to, read-only memory (ROM), random access memory (RAM), registers, cache memory, semiconductor memory devices, magnetic media (such as internal hard disks and removable disks), magneto-optical media, and optical media (such as CD-ROMs and digital versatile discs (DVDs)). The processor associated with the software may be used to implement a radio frequency transceiver used in a WTRU, UE, terminal, base station, RNC, or any host computer.

[0220] Further embodiments of this document may be constituted by supplementing the embodiment with one or more elements from any one or more other embodiments of this document, and / or by replacing one or more elements in one embodiment with one or more elements from one or more other embodiments of this document.

[0221] Therefore, it should be understood that the subject matter disclosed is not limited to the specific embodiments disclosed, but is intended to cover all modifications that conform to the spirit and scope of this disclosure, as defined in the appended claims, detailed description and / or as shown in the drawings. ***

Claims

1. A system for detecting a heartbeat having optimal characteristics for an electrophysiological EP mapping system, the system comprising: A processor, the processor including a machine learning algorithm configured to perform the following operations: A first heartbeat is received at the identified spatial location of the heart, the first heartbeat including first attribute information corresponding to the first heartbeat; A second heartbeat is received at the identified cardiac spatial location, the second heartbeat including second attribute information corresponding to the second heartbeat; Compare the first attribute information with the second attribute information; as well as Determining which of the first and second heartbeats has the best characteristics based on the compared attribute information includes: determining which of the first and second heartbeats has the best characteristics based on at least one of heartbeats with less noise, heartbeats with more pronounced local activation time (LAT), heartbeats with more pronounced late potentials, heartbeats with more stable fragmentation, and heartbeats with less V interference.

2. The system according to claim 1, wherein, The machine learning algorithm will also output the determination of the heartbeat with the best features to the EP mapping system.

3. The system according to claim 1, wherein, At least one of the first heartbeat, which includes first attribute information, and the second heartbeat, which includes second attribute information, is stored in a database that communicates with the processor.

4. The system according to claim 3, wherein, One of the first heartbeat and the second heartbeat is a heartbeat manually received in the EP mapping system, and the other of the first heartbeat and the second heartbeat is a heartbeat located at the same spatial location up to a threshold tolerance that was removed from the EP mapping system by a physician.

5. The system according to claim 4, wherein, The machine learning algorithm learns to determine which of the first heartbeat and the second heartbeat has the best features based on the results of manually accepted heartbeats in the EP mapping system and heartbeats deleted by the physician in the EP mapping system, wherein the manually accepted heartbeats and the deleted heartbeats are located at the same position up to a threshold tolerance.

6. The system according to claim 1, wherein, The attribute information includes at least one of the following: The intracardiac electrocardiogram signal received by the electrodes of the catheter; Electrocardiogram signals received from mapping electrodes; Electrocardiogram signals received by one or more surface electrodes; The tissue proximity indicator TPI at the time of the reference annotation is indicated by the calibration electrode. The force value detected by the force sensor of the calibration catheter; The spatial location of the calibration electrode at the reference annotation; The breathing state vector surrounding the aforementioned reference annotation; The reference notes are located within the respiratory cycle; The difference between the length of the current respiratory cycle and the length of the previous respiratory cycle; The ratio of the current respiratory cycle length to the previous respiratory cycle length; The difference between the current respiratory cycle length and the average or median respiratory cycle length; The ratio of the current respiratory cycle length to the average or median respiratory cycle length; Whether the physician manually received the heartbeat into the EP mapping system or deleted the heartbeat indication; and The distance between the current reference annotation and the spatial position of the reference electrode at the previous reference annotation.

7. The system according to claim 1, wherein, The determination of whether the first heartbeat or the second heartbeat has the best features based on the compared attribute information is a binary determination.

8. The system according to claim 1, wherein, The machine learning algorithm is a neural network.

9. The system according to claim 8, wherein, The neural network in question is a convolutional neural network.

10. A method for detecting heartbeats with optimal features for an electrophysiological EP mapping system using a machine learning algorithm, the method comprising: Receive first data, the first data including a first heartbeat at an identified cardiac spatial location, the first heartbeat including first attribute information corresponding to the first heartbeat; Receive second data, the second data including a second heartbeat at the identified cardiac spatial location, the second heartbeat including second attribute information corresponding to the second heartbeat; Compare the first data with the second data; as well as Based on the comparison, an output is made determining which of the first and second heartbeats has the best characteristics, the determination including: determining which of the first and second heartbeats has the best characteristics based on at least one of heartbeats with less noise, heartbeats with more pronounced local activation time (LAT), heartbeats with more pronounced late potentials, heartbeats with more stable fragmentation, and heartbeats with less V interference.

11. The method of claim 10, further comprising outputting the determination to the EP calibration system.

12. A system for detecting mapping annotations in an electrophysiological EP mapping system, the system comprising: A processor, the processor including a machine learning algorithm configured to perform the following operations: Receive input data, which includes attribute data for each of multiple heartbeats obtained at the same spatial location; The attribute data for each of the multiple heartbeats is compared with a predefined threshold; as well as Determining which heartbeat to use as the mapping annotation based on the attribute data includes: determining which heartbeat to use as the mapping annotation based on at least one of heartbeats with less noise, heartbeats with more pronounced local activation time (LAT), heartbeats with more pronounced late potentials, heartbeats with more stable fragmentation, and heartbeats with less V interference.

13. The system according to claim 12, wherein, The machine learning algorithm will also output the determination of the heartbeat used as the annotation of the calibration to the EP calibration system.

14. The system according to claim 12, wherein, One of the multiple heartbeats is manually acquired by the physician in the EP mapping system.

15. The system according to claim 14, wherein, The machine learning algorithm learns to determine which of the plurality of heartbeats to use as the mapping annotation based on the heartbeat results obtained by the physician in the EP mapping system.

16. The system according to claim 12, wherein, The attribute data includes at least one of the following: The intracardiac electrocardiogram signal received by the electrodes of the catheter; Electrocardiogram signals received from mapping electrodes; Electrocardiogram signals received by one or more surface electrodes; The obtained local annotation time of heartbeats; The tissue proximity indicator TPI at the time of the reference annotation is indicated by the calibration electrode. The force value detected by the force sensor of the calibration catheter at the spatial position of the calibration electrode at the reference annotation; Respiratory gating status; The breathing state vector surrounding the aforementioned reference annotation; The reference notes are located within the respiratory cycle; The difference between the length of the current respiratory cycle and the length of the previous respiratory cycle; The ratio of the current respiratory cycle length to the previous respiratory cycle length; The difference between the current respiratory cycle length and the average or median respiratory cycle length; The ratio of the current respiratory cycle length to the average or median respiratory cycle length; Whether the physician manually received the heartbeat into the EP mapping system or deleted the heartbeat indication; Indications regarding the presence of any surface activation at the time of that heartbeat; and The distance between the current reference annotation and the spatial position of the reference electrode at the previous reference annotation.

17. The system according to claim 12, wherein, The machine learning algorithm is a neural network.

18. The system according to claim 17, wherein, The neural network in question is a recurrent neural network (RNN).

19. The system according to claim 18, wherein, The heartbeats identified as being used as the annotations are used as inputs for training the RNN.

20. The system according to claim 12, wherein, The determination of the heartbeats used as the annotation for the calibration will be compared with the gold standard database to verify its accuracy.

21. A method for detecting mapping annotations in an electrophysiological EP mapping system using a machine learning algorithm, comprising: Receive input data, which includes attribute data for each of multiple heartbeats obtained at the same spatial location; The attribute data for each of the multiple heartbeats is compared with a predefined threshold; Determining which heartbeat to use as the mapping annotation based on the comparison includes: determining which heartbeat to use as the mapping annotation based on at least one of heartbeats with less noise, heartbeats with more pronounced local activation time (LAT), heartbeats with more pronounced late potentials, heartbeats with more stable fragmentation, and heartbeats with less V interference. as well as The determination is output to the EP calibration system.

22. The method according to claim 21, wherein, One of the multiple heartbeats is manually acquired by the physician in the EP mapping system; and The method further includes training the machine learning algorithm to learn to determine which of the plurality of heartbeats to use as the mapping annotation based on heartbeat results obtained by the physician in the EP mapping system.

Images