Systems and methods for detecting and identifying cardiac pace mapping sites and pacing operations

Abstract

To improve the accuracy and efficiency of identification of sites that are likely to be the origin of cardiac arrhythmia.SOLUTION: Techniques are provided that utilize a training dataset associated with biometrics of patients' hearts, the training dataset including: electrophysiological data associated with cardiac arrhythmia; pace-mapping datasets; and correlation data measuring the degree of correlation between the pace-mapping datasets and the electrophysiological data. Based on the training dataset, a pace-mapping prediction model is trained to predict a degree of correlation between a patient's electrophysiological data and a pace-mapping dataset. Based on the predicted degree of correlation, a cardiac location in the heart of a patient is predicted as the location for the next pace-mapping. Further systems and methods are disclosed for generating a pacing maneuver prediction model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] (CROSS-REFERENCE TO RELATED APPLICATIONS) This application claims the benefit of U.S. Provisional Patent Application No. 63 / 062,715, filed August 7, 2020, the disclosure of which is incorporated herein by reference in its entirety. [Background technology]

[0002] Cardiac arrhythmias, such as atrial fibrillation (AF), ventricular fibrillation, ventricular tachycardia (VT), or atrial flutter, can cause morbidity and mortality. Treating cardiac disorders, such as cardiac arrhythmias, often requires obtaining detailed mapping of the cardiac tissue, chambers, veins, arteries, and / or electrical pathways of the heart. Such mapping helps identify problematic areas of scar tissue or arrhythmia sources (e.g., electrical rotors) as well as healthy areas. For example, as a prerequisite for performing catheter ablation, the spatial origin of a cardiac arrhythmia must be precisely identified within the heart. Identifying the origin of a cardiac arrhythmia may require electrophysiological testing.

[0003] Electrophysiological testing, i.e., electrophysiological (EP) cardiac mapping or cardiac electroanatomical mapping, provides 3D mapping data. The 3D mapping data can be constructed based on electrical potentials measured from signals emitted by a catheter introduced into a cardiac chamber. The 3D mapping data can be based on various modalities, such as local activation time (LAT), electrical activity, unipolar or bipolar voltage, topology, dominant frequency, or impedance. Thus, data corresponding to various modalities can be captured using a catheter inserted into a patient's body. The captured data can be processed and / or visualized on a display for viewing by a medical professional or stored for later processing and / or visualization.

[0004] Myocardial scarring is known to be associated with arrhythmia conduction pathways and foci (e.g., reentrant foci) responsible for VT. Accurate localization of suspected arrhythmogenic foci is necessary to maximize the likelihood of successful catheter ablation. Localization of suspected arrhythmogenic foci can be achieved by pacing, i.e., introducing signals to specific locations (sites) within the ventricle and measuring the corresponding electrical potentials. Pacing at different sites within the ventricle can thus be used to identify likely sites of origin for a patient's VT. A likely site is predicted to be one where pacing produces measured electrical potentials that match those measured during a previous inducible VT in the patient.

[0005] Conventional pace mapping techniques require a skilled technician, such as a physician, to acquire electrical potential signals (i.e., pace mapping data) from multiple points within a cardiac region of interest, such as a cardiac chamber. Typically, electrical activity associated with a point within the heart is generated by first advancing a catheter (including an electrical sensor at or near its distal tip) to contact tissue at that point within the heart, then emitting a signal via the catheter's sensor, which is measured to generate electrical activity associated with that point. This process is repeated at multiple points within the heart, and the data measured at each point is stored in a map (i.e., a pace map) representing the cardiac electrical activity at those points. For example, in clinical practice, it is common to accumulate data from 100 or more cardiac sites to generate a detailed and comprehensive pace map of the cardiac chamber's electrical activity. Pace mapping data associated with a cardiac site is compared with corresponding data, e.g., electrophysiological data generated from an inducible VT, to determine the degree of correlation and thereby the likelihood that the inducible VT originates from the same paced cardiac site. Summary of the Invention [Problem to be solved by the invention]

[0006] Currently, identifying multiple points of interest in cardiac tissue associated with arrhythmia conduction pathways and foci to find pacing sites associated with electrical activity highly correlated with electrical activity associated with inducible VT is difficult and time-consuming, requiring trial and error by a skilled technician, such as a cardiologist. Methods and systems are needed to improve the accuracy and efficiency of identifying likely sites of origin for cardiac arrhythmias. [Means for solving the problem]

[0007] Systems and methods for detecting and identifying cardiac pace mapping sites and pacing maneuvers are disclosed in the present disclosure.

[0008] Aspects disclosed in the present disclosure describe a method for training a pace mapping prediction model. The method includes receiving training datasets related to a patient's heart. For each patient, the training dataset includes electrophysiological data related to the patient's cardiac arrhythmia, pace mapping datasets, each dataset acquired from an electrode when positioned at a cardiac location on the patient's heart, and correlation data measuring a degree of correlation between each of the pace mapping datasets and the electrophysiological data. The method also includes training a pace mapping prediction model based on the training datasets to predict a degree of correlation between the electrophysiological data and the pace mapping datasets related to the new patient.

[0009] Aspects disclosed herein also describe a system for training a pace mapping prediction model. The system includes at least one processor and a memory storing instructions. When executed by the at least one processor, the instructions cause the system to receive training datasets associated with a patient's heart. For each patient, the training dataset includes electrophysiological data associated with the patient's cardiac arrhythmia, pace mapping datasets, each dataset acquired from an electrode when positioned at a cardiac location on the patient's heart, and correlation data measuring a degree of correlation between each of the pace mapping datasets and the electrophysiological data. The instructions then cause the system to train a pace mapping prediction model based on the training datasets to predict a degree of correlation between the electrophysiological data and the pace mapping dataset associated with a new patient.

[0010] Further, aspects disclosed herein describe a non-transitory computer-readable medium including instructions executable by at least one processor to execute a method for training a pace mapping prediction model. The method includes receiving training datasets associated with a patient's heart. For each patient, the training dataset includes electrophysiological data associated with the patient's cardiac arrhythmia, pace mapping datasets, each dataset acquired from an electrode when positioned at a cardiac location on the patient's heart, and correlation data measuring a degree of correlation between each of the pace mapping datasets and the electrophysiological data. The method also includes training a pace mapping prediction model based on the training datasets to predict a degree of correlation between the electrophysiological data and the pace mapping dataset associated with the new patient.

[0011] Aspects disclosed in the present disclosure describe a method for training a pacing maneuver prediction model. The method includes receiving a training dataset associated with a patient's heart. For each patient, the training dataset includes pacing maneuvers, each associated with a pacing location in the patient's heart, and corresponding interval measurements, each associated with a distance between a last pacing pulse and an intrinsic beat since the corresponding pacing maneuver. The method also includes training a pacing maneuver prediction model based on the training dataset to predict interval measurements based on pacing maneuvers associated with the new patient.

[0012] Aspects disclosed in the present disclosure also describe a system for training a pacing maneuver prediction model. The system includes at least one processor and a memory storing instructions. When executed by the at least one processor, the instructions cause the system to receive a training dataset associated with a patient's heart. For each patient, the training dataset includes pacing maneuvers, each associated with a pacing location in the patient's heart, and corresponding interval measurements, each associated with a distance between the last pacing pulse and an intrinsic beat since the corresponding pacing maneuver. The instructions also cause the system to train a pacing maneuver prediction model based on the training dataset to predict interval measurements based on pacing maneuvers associated with a new patient.

[0013] Further disclosed aspects of the present disclosure describe a non-transitory computer-readable medium including instructions executable by at least one processor to execute a method for training a pacing maneuver prediction model. The method includes receiving a training dataset associated with a patient's heart. For each patient, the training dataset includes pacing maneuvers, each associated with a pacing location in the patient's heart, and corresponding interval measurements, each associated with a distance between a last pacing pulse and an intrinsic beat since the corresponding pacing maneuver. The method also includes training a pacing maneuver prediction model based on the training dataset to predict interval measurements based on pacing maneuvers associated with the new patient. [Brief explanation of the drawings]

[0014] A more detailed understanding may be had from the following description, given by way of example in conjunction with the accompanying drawings, in which like reference numerals indicate similar elements and in which: [Figure 1] 1 is a diagram of an exemplary cardiac ablation system, upon which one or more features of the present disclosure may be implemented. [Figure 2] FIG. 2 is a block diagram of an exemplary system deployable by the exemplary cardiac ablation system of FIG. 1, upon which one or more features of the present disclosure may be implemented. [Figure 3] FIG. 2 is a diagram of an exemplary catheter deployable by the exemplary cardiac ablation system of FIG. 1, upon which one or more features of the present disclosure may be implemented. [Figure 4] FIG. 1 is a functional block diagram of an exemplary machine learning system, based on which one or more features of the present disclosure may be implemented. [Figure 5A] An exemplary pace map (FIG. 5A) is shown that records the correlation between a guided ECG signal (FIG. 5B) and a pace mapping ECG signal (FIG. 5C), based on which one or more features of the present disclosure can be implemented. [Figure 5B]An exemplary pace map (FIG. 5A) is shown that records the correlation between a guided ECG signal (FIG. 5B) and a pace mapping ECG signal (FIG. 5C), based on which one or more features of the present disclosure can be implemented. [Figure 5C] An exemplary pace map (FIG. 5A) is shown that records the correlation between a guided ECG signal (FIG. 5B) and a pace mapping ECG signal (FIG. 5C), based on which one or more features of the present disclosure can be implemented. [Figure 6] An example is provided for training a model to predict a complete pace map from an incomplete pace map, based on which one or more features of the present disclosure can be implemented. [Figure 7] Another example is provided for training a model to predict a complete pace map from an incomplete pace map, based on which one or more features of the present disclosure can be implemented. [Figure 8] 1 is a flowchart of an example method for training a pace mapping prediction model, according to which one or more features of the present disclosure may be implemented. [Figure 9] 9 is a flowchart of an exemplary method for applying the trained model of FIG. 8, based on which one or more features of the present disclosure may be implemented. [Figure 10] 10 is a flowchart of another example method for training a pace mapping prediction model, according to which one or more features of the present disclosure may be implemented. [Figure 11] 11 is a flowchart of an exemplary method for applying the trained model of FIG. 10, based on which one or more features of the present disclosure may be implemented. [Figure 12] 1 illustrates exemplary ECG tracings of pacing maneuvers and time caliper measurements taken manually by a physician, based on which one or more features of the present disclosure may be implemented. [Figure 13A]An exemplary flow diagram illustrating a method for training a machine learning model (FIG. 13A) is shown, based on which one or more features of the present disclosure may be implemented. [Figure 13B] An exemplary flow diagram illustrating a machine learning model application (FIG. 13B) is shown, based on which one or more features of the present disclosure may be implemented. [Figure 14] FIG. 1 is a functional block diagram of an exemplary recurrent neural network (RNN) on which one or more features of the present disclosure may be implemented. DETAILED DESCRIPTION OF THE INVENTION

[0015] Systems and methods are provided for detecting and identifying the origin of cardiac arrhythmias through pace mapping and pacing maneuvers, where the detection and identification is based on machine learning models that are trained to predict the likely origin of cardiac arrhythmias and the site of pacing maneuvers.

[0016] FIG. 1 is a diagram of an exemplary cardiac ablation system 100, upon which one or more features of the present disclosure may be implemented. System 100 may include a console 124 operated by a physician 130, a display 127, and a catheter 140. System 100 may be configured to acquire anatomical and electrical measurements taken from an organ of a patient 128, such as a heart 126, and may be configured to perform a cardiac ablation procedure. System 100 may be used to collect data for a training dataset used to train a model and to apply the trained model. One example of system 100 is the Carto® system sold by Biosense Webster.

[0017] Cardiac ablation system 100 may include a catheter 140, further described with reference to FIG. 3 . Catheter 140 may be configured to injure (ablate) tissue regions in an internal organ and / or acquire biometric data, including electrical signals. System 100 may include one or more probes 121 having a shaft 122 that can be navigated by a physician or user 130 to a body part, such as a heart 126, of a patient 128 reclining on a table 129. Physician 130 can insert shaft 122 through sheath 123 while manipulating and / or deflecting the distal end of shaft 122 from sheath 123 using a manipulator near the proximal end of catheter 140. Inset 145 shows an enlarged view of catheter 140 within a chamber of heart 126. As shown, catheter 140 may be attached to the distal end of shaft 122. Catheter 140 can be inserted through sheath 123 in a collapsed state and then expanded within heart 126. The catheter 140 can be configured to ablate a tissue region of a chamber of the heart 126. The catheter 140 can include at least one ablation electrode 147 coupled to the body of the catheter. For example, the ablation electrode 147 can be configured to provide energy to a tissue region of a body organ, such as the heart 126. The energy can be thermal energy and can cause damage to the tissue region that begins at the surface of the tissue region and extends through the thickness of the tissue region. Other elements, such as electrodes or transducers, can be part of the catheter and can be configured to ablate as well as acquire biometric data.

[0018] In one aspect, biometric data obtained by elements of the catheter can represent information related to LAT, electrical activity, topology, unipolar or bipolar voltage, dominant frequency, or impedance. LAT can represent the time at which electrical activity is measured at a particular location. LAT can be 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. Electrical activity can be sensed and / or enhanced (e.g., using a filter to improve the signal-to-noise ratio). Topology can represent the physical structure of a body part or portion of a body part, or may correspond to changes in physical structure between different portions of a body part or between different body parts. Dominant frequency can represent a frequency or range of frequencies that is dominant in a portion of a body part and may differ in different portions of the same body part. For example, the dominant frequency of the pulmonary veins of a heart may be different from the dominant frequency of the right atrium of the same heart. Impedance can represent resistance in a given region of a body part.

[0019] The console 124 of the system 100 may include a processing unit 141, which may include front-end and control components (e.g., a computer with a multi-core processor). The console may also include memory 142, such as, for example, volatile and / or non-volatile memory, and a communication interface circuit 138, for example, to send and receive signals to and from the catheter 140. The console 124 may be configured to receive biometric data, then process the biometric data, and store the data for later processing or transmit the data to another system over a network. In one aspect, the processing component 141 may be external to the console 124, for example, located within the catheter 140, an external device, a mobile device, a cloud-based device, or may be a stand-alone processor. The processing unit 141 may execute software modules programmed to perform the functions of aspects described herein. The software modules may be downloaded to the processing component 141 over a network or from a non-transitory tangible medium, such as magnetic, optical, or electronic memory, external to or local to the console 124.

[0020] System 100 can be modified to implement aspects disclosed herein. The aspects disclosed herein can be similarly applied using other system components and configurations. Furthermore, system 100 may include additional components, such as an element for sensing electrical activity, a wired or wireless connector, a processing unit, or a display device. Console 124 typically includes a real-time noise reduction circuit configured as a field programmable gate array (FPGA), followed by an analog-to-digital (A / D) ECG (electrocardiogram) or EMG (electromyogram) signal conversion integrated circuit. The output of the A / D ECG or EMG circuit may be processed to perform the methods disclosed herein.

[0021] In addition to electrical measurements obtained by catheter 140 (e.g., ECG) or other sensors that measure electrical properties of the heart, in one aspect, system 100 can also obtain anatomical measurements of the patient's heart. The anatomical measurements can be generated by imaging modalities such as ultrasound, computed tomography (CT), or magnetic resonance imaging (MRI). Thus, system 100 can obtain biometric data including anatomical and electrical measurements and can store the biometric data in memory 142 of system 100. The biometric data can be transmitted from memory 142 to processing unit 141. Alternatively or additionally, the biometric data can be transmitted to a server, which can be local or remote to console 124.

[0022] The console 124 may be connected by a cable 139 to body surface electrodes 143, which may include adhesive skin patches affixed to the patient 128. The processing unit 141, in conjunction with a current tracking module, may determine position coordinates of the catheter 140 within a body portion (e.g., the heart 126) of the patient 128. The position coordinates may be based on impedance or electromagnetic fields measured between the body surface electrodes 143 and the electrodes 147 or other electromagnetic components of the catheter 140. Additionally or alternatively, location pads may be attached to the surface of the bed 129.

[0023] During a procedure, the processing unit 141 can facilitate the rendering of the body part 135 on the display 127 for viewing by the physician 130 and can store data representing the body part 135 in the memory 142. In one aspect, the medical professional 130 can manipulate the body part rendering 135 using one or more input devices, such as a touchpad, a mouse, a keyboard, or a gesture recognizer. For example, the input device can be used to change the position of the catheter 140 so that the rendering 135 of the body part 126 is updated. In another example, the display 127 can include an input device (e.g., a touchscreen) that can be configured to accept input from the medical professional 130, for example, to control the rendering of the body part 135. In one aspect, the display 127 can be located at a remote location, such as a separate hospital, or within a separate healthcare provider network.

[0024] 2 is a block diagram of an exemplary system 200 deployable by the exemplary cardiac ablation system of FIG. 1 and based on which one or more features of the present disclosure can be implemented. System 200 may include a monitoring and processing system 205, a local system 280, and a remote system 290. Alternatively, monitoring and processing system 205 may represent console 124 of system 100. Monitoring and processing system 205 may include patient biometric sensors 210, a processor 220, a memory 230, input devices 240, output devices 250, and a transceiver 260, i.e., a transmitter-receiver, that communicates with network 270. System 205 may continuously or periodically monitor, store, process, and transmit various patient biometrics via network 270. Examples of patient biometrics include electrical signals (e.g., ECG signals), anatomical images, blood pressure data, blood glucose data, and temperature data. Patient biometrics can be monitored and transmitted for the treatment of various diseases, such as cardiovascular diseases (e.g., arrhythmias, cardiomyopathies, and coronary artery disease) and autoimmune diseases (e.g., type I and type II diabetes).

[0025] The monitoring and processing system 205 may be internal to the patient's body, for example, the system 205 may be implantable subcutaneously, orally, or surgically inserted via a vein or artery, via endoscopic or laparoscopic surgery. Alternatively, the system 205 may be external to the patient, for example, attached to the patient's skin. In one aspect, the system 205 may include both components internal to the patient's body and components external to the patient's body.

[0026] The monitoring and processing system 205 may represent multiple monitoring and processing systems 205 capable of processing patient biometric data in parallel and / or in communication with each other or with a server over a network. One or more systems 205 may acquire or receive all or a portion of the patient biometric data (e.g., electrical signals, anatomical images, blood pressure, temperature, blood glucose levels, or other biometric data). One or more systems 205 may also acquire or receive additional information related to the acquired or received patient biometric data from one or more other systems 205. The additional information may be, for example, diagnostic information and / or information acquired from a device such as a wearable device. Each monitoring and processing system 205 may process the data it acquires and may also process data received from another system 205.

[0027] The patient biometric sensor 210 may be one or more sensors that may be configured to sense biometric data. For example, the sensor 210 may be an electrode configured to acquire an electrical signal (e.g., a bioelectric signal originating from the heart), a temperature sensor, a blood pressure sensor, a blood glucose sensor, a blood oxygen sensor, a pH sensor, an accelerometer, or a microphone. In one aspect, the system 205 may be an ECG monitor that measures an ECG signal originating from the heart. In such a case, the sensor 210 may be one or more electrodes that may be configured to acquire the ECG signal. The ECG signal may be used to treat various cardiovascular diseases. In one aspect, the patient biometric sensor 210 may also include a catheter with one or more electrodes, a probe, a blood pressure cuff, a weight scale, a bracelet (e.g., a smartwatch biometric tracker), a blood glucose monitor, a continuous positive airway pressure (CPAP) machine, or any other device that provides biometric or other data related to a patient's health.

[0028] The transceiver 260 may include a transmitter component and a receiver component. These transmitter and receiver components may be integrated into a single device or may be implemented separately. The transceiver may provide connectivity between the system 205 and other systems or servers via a communications network 270. The network 270 may be a wired network, a wireless network, or may include a combination of wired and / or wireless networks. The network 270 may be a short-range network (e.g., a local area network (LAN) or a personal area network (PAN)). Information may be transmitted or received over the short-range network using various short-range communication protocols, such as Bluetooth, Wi-Fi, Zigbee, Z-Wave, near-field communication (NFC), Ultraband, Zigbee, or infrared (IR). The network 270 may also be a long-range network (e.g., a wide area network (WAN), the Internet, or a cellular network). Information can be sent or received over long-range networks using a variety of long-range communication protocols, such as TCP / IP, HTTP, 3G, 4G / LTE, or 5G / New Radio.

[0029] The processor 220 may be configured, for example, to process patient biometric data acquired by the sensor 210 and store the biometric data and / or processed biometric data in the memory 230. The processor 220 may also be configured to transmit the biometric data over the network 270 via the transmitter of the transceiver 260. Biometric data from one or more other monitoring and processing systems 205 may be received by the receiver of the transceiver 260. The processor 220 may use machine learning algorithms (e.g., based on neural networks), or the machine learning algorithms may be used by another processor, for example, in the local system 280 or the remote system 290. In aspects, the processor 220 may include one or more CPUs, one or more GPUs, or one or more FPGAs. In these aspects, the machine learning algorithms may be executed on one or more of these processing units. Similarly, the processor 220 may include an ASIC dedicated to performing deep learning calculations (such as an Intel® Nervana™ Neural Network Processor), and the machine learning algorithms may be executed on such dedicated ASICs. The processing unit that runs the machine learning algorithm may be located in the medical procedure room or elsewhere (e.g., in another medical facility or in the cloud).

[0030] The input device 240 of the monitoring and processing system 205 may be used as a user interface. The input device 240 may include, for example, a piezoelectric or capacitive sensor configured to receive user input, such as a tap or touch. Accordingly, the input device 240 may be configured to implement capacitive coupling in response to a user tapping or touching the surface of the system 205. Gesture recognition can be implemented through various capacitive couplings, such as resistive capacitive, surface capacitive, projected capacitive, surface acoustic wave, piezoelectric, or infrared touch. A capacitive sensor may be disposed on the surface of the input device 240, such that tapping or touching the surface activates the system 205. The processor 220 may be configured to selectively respond to different tapping patterns of the capacitive sensor (e.g., a single tap or a double tap on the input device 240) such that different functions of the system 205 (e.g., data acquisition, storage, or transmission) are activated based on the detected pattern. In one aspect, the system 205 may provide audible feedback to the user, for example, when a gesture is detected and recognized.

[0031] In one aspect, a local system 280, which may communicate with the monitoring and processing system 205 via a network 270, may be configured to act as a gateway to a remote system 290 via another network 285 that may be accessible to the local system 280. The local system 280 may be, for example, a smartphone, a smartwatch, a tablet, or other portable smart device. Alternatively, the local system 280 may be a fixed or standalone device. Patient biometric data may be communicated between the local system 280 and the monitoring and processing system 205. In one aspect, the local system 280 may also be configured to display acquired patient biometric data and related information.

[0032] In one aspect, remote system 290 may be configured to receive at least a portion of the monitored patient biometric data and related information via network 285, which may be a long-range network. For example, if local system 280 is a cellular phone, network 285 may be a wireless cellular network, and information may be communicated between local system 280 and remote system 290 via a wireless technology standard, such as any of the wireless technologies described above. Remote system 290 may be configured to present the received patient biometric data and related information to a medical professional (e.g., a physician) visually on a display or audibly through a speaker.

[0033] FIG. 3 is a diagram of an exemplary pace mapping catheter 300 deployable by the exemplary cardiac ablation system of FIG. 1 , based on which one or more features of the present disclosure can be implemented. For example, the catheter 300 may be a mapping and therapy delivery catheter for insertion into a human body, e.g., a heart chamber. The catheter 300 shown in FIG. 3 is exemplary, and many other types of catheters can be used according to aspects of the present disclosure. An electrode 332 may be positioned in the distal portion 334 to measure electrical properties of cardiac tissue. The electrode 332 may also be useful for emitting electrical signals into the heart for diagnostic purposes (e.g., for electrical mapping or to induce VT) or therapeutic purposes (e.g., to ablate defective cardiac tissue). The distal portion 334 of the catheter 300 may further include an array 336 of non-contact electrodes 338 for measuring far-field electrical signals within the heart chamber. The array 336 may be a linear array in that the non-contact electrodes 338 are arranged linearly along the longitudinal axis of the distal portion 334. The distal portion 334 can further include at least one position sensor 340 that generates signals used to determine the position and orientation of the distal tip 318 within the body. In one embodiment, the position sensor 340 is adjacent to the distal tip 318. There is a fixed position and orientation relationship between the position sensor 340, the distal tip 318, and the electrodes 332. The handle 320 of the catheter 300 can include controls 346 for steering or deflecting the distal portion 334 or orienting it as desired.

[0034] The position sensor 340 may be configured to transmit position-related electrical signals to the console 124 via a cable 342 through the catheter 300 (i.e., cable 139 shown in FIG. 1 ) in response to fields that may be generated by the system 100 ( FIG. 1 ). In another alternative, the position sensor 340 in the catheter 300 may transmit signals to the console 124 via a wireless link. For example, a positioning process performed by the processing unit 141, 220 may calculate the position and orientation of the distal portion 334 of the catheter 300 based on the signals transmitted by the position sensor 340. The positioning process may receive, amplify, filter, digitize, and otherwise process signals from the catheter 300. The positioning process may also provide a signal output to the display 127 that may visualize the position of the tip portion 334 and / or distal tip 318 of the catheter 300 relative to a site selected for ablation.

[0035] 4 is a functional block diagram of an exemplary machine learning system 400, based on which one or more features of the present disclosure can be implemented. Various machine learning systems can be used to train and apply the pace mapping prediction model disclosed herein. For example, the machine learning system 400 can be based on artificial neural networks (ANNs) of various architectures, such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs), an example of which is a long short-term memory (LSTM) network. Generally, neural networks are trained to predict information of interest based on observations. The neural networks are trained via a supervised learning process, whereby correlations between exemplary pairs (i.e., observations and corresponding information of interest) are learned.

[0036] In one embodiment, the ANN 410 can be a CNN, which is useful for learning patterns from data provided in a spatiotemporal format, such as the pace map 540 shown in Figure 5A. In general, a CNN can use convolutional operations across several layers using kernels. Each layer in the network, e.g., layer n 440, can process data from an image at its input and produce a processed image that is processed by the next layer, e.g., layer m 450. Convolution kernels applied in earlier layers in the network can integrate information from adjacent map elements more efficiently than convolution kernels applied in later layers in the network. Thus, correlations between image elements that are closely spaced in a map may be better learned in a CNN.

[0037] Typically, a neural network 410 includes nodes ("neurons") that are connected according to a given architecture. For example, in a given architecture, the nodes may be arranged in layers, i.e., the output of a node in one layer, e.g., layer n 440, feeds the input of a node in the next layer connected to it, e.g., layer m 450. j ) node j 455 is usually assigned a specific strength or a specific weight w(m j ,n i ) 460 in layer n 440 (i.e., n i ) is connected to node i445. Therefore, the weights {w(m j ,n i )} ("synaptic weights") parameterize the neural network model. Thus, training a neural network can be viewed as specializing the network by determining the network weights (parameters), i.e., by determining the model parameters 430.

[0038] The way neural network 410 processes data can be described as follows: Input data can be provided to nodes in the first layer of the neural network, such that each node in the first layer receives a weighted combination of the input data (or a weighted combination of a subset of the input data). The input weighted combination of each node is then transformed according to the node's activation function to obtain the node's output data. The output data from each node in the first layer can then be provided to nodes in the second layer of the neural network, such that each node in the second layer receives a weighted combination of the nodes' outputs in the first layer (or a weighted combination of the outputs of a subset of the nodes in the first layer). The input weighted combination of each node is then transformed according to the node's activation function to obtain the node's output data. The output data from the nodes in the second layer is then propagated and processed similarly in other intermediate layers of the network, with the final layer providing the network's output data. Thus, a neural network is typically characterized by the structure of its nodes and the activation functions of these nodes. The weights associated with the inter-node connections (network parameters or model parameters 430) are learned by an iterative training process, such as a back-propagation algorithm, according to training parameters (e.g., a learning rate and a cost function) and based on the training dataset 420.

[0039] The training data set 420, based on which the neural network model 410 can be trained, can include example data pairs, such as observed data (e.g., measurements collected during a surgical procedure) and corresponding information of interest (e.g., the outcome of the surgical procedure) predicted by the model. For example, cardiac temperature data (observed data) can be collected and correlated (by a training process) with the outcome of the cardiac procedure (the predicted information of interest). Once the model parameters are determined by the training process, the model can be applied to predict the information of interest based on new observations. For example, in the cardiac case, based on an input of temperature during the procedure (e.g., 97.7-100.2 degrees Celsius), the output of the model can be a prediction of the outcome of the procedure. Such predictions are based on the correlation between temperature and treatment outcome learned by a neural network model based on a training dataset.

[0040] Aspects of the present disclosure can train a machine learning model (e.g., ANN 410) and apply the trained model to detect and / or identify pace mapping sites. Aspects of the present disclosure can also train a machine learning model and apply the trained model for pacing operations during cardiac pace mapping. The algorithms disclosed herein can be applied to train a model based on a training dataset, where the training dataset includes biometric data measured by various hardware such as those disclosed herein.

[0041] Cardiac arrhythmia, specifically AF, is a common and dangerous condition, especially in the elderly population. In patients with normal sinus rhythm, the heart, including atrial and ventricular excitable conduction tissue, is electrically excited to beat in a synchronous and patterned manner. In patients with cardiac arrhythmia, abnormal regions of cardiac tissue do not follow the synchronous beating cycle associated with normal conductive tissue. Instead, the abnormal regions of cardiac tissue conduct abnormally to adjacent tissue, thereby disrupting the cardiac cycle and resulting in an asynchronous cardiac rhythm. Such abnormal conduction has long been known to occur in various regions of the heart, for example, in the region of the sinoatrial (SA) node, along the conduction pathways of the atrioventricular (AV) node and the bundle of His, or in the myocardial tissue that forms the walls of the ventricular and atrial chambers.

[0042] Cardiac arrhythmias, including atrial arrhythmias, can be multiwavelet reentrant, characterized by multiple asynchronous loops of electrical impulses scattered around the atrium, often self-propagating. Alternatively, or in addition to multiwavelet reentrant, cardiac arrhythmias can also have a localized origin, such as when isolated regions of tissue within the atrium are rapidly and repetitively excited autonomously. VT is a tachycardia (fast heart rhythm) originating from one of the ventricles of the heart. This is a potentially life-threatening arrhythmia, as it can lead to ventricular fibrillation and sudden death.

[0043] AF, a type of arrhythmia, occurs when normal electrical impulses generated by the sinoatrial node are overwhelmed by disorganized electrical impulses generated by the atria and pulmonary veins, resulting in irregular impulses being conducted to the ventricles. Arrhythmias that can result from this condition can last from minutes to weeks, or even years. AF is often a chronic condition and can often lead to an increased risk of death from stroke. Risk increases with age. Approximately 8% of people over the age of 80 have some degree of AF. AF is often asymptomatic and generally not life-threatening in itself, but it can cause palpitations, weakness, fainting, chest pain, and congestive heart failure. The risk of stroke increases during AF because blood can pool in the atria and left atrial appendage, which contract insufficiently, potentially forming clots. The first-line treatment for AF is medication to slow the heart rate or restore normal heart rhythm. Furthermore, patients with AF are often given anticoagulants to reduce the risk of stroke. The use of these anticoagulants carries its own risk of internal bleeding. In some patients, medical therapy is inadequate, and their AF is considered drug-refractory, meaning it cannot be treated with standard pharmacological interventions. Synchronized electrical cardioversion can also be used to convert AF to a normal cardiac rhythm. Alternatively, patients with AF are treated with catheter ablation.

[0044] Catheter ablation-based therapy can involve mapping the electrical properties of cardiac tissue, particularly the endocardium and cardiac volumes, and selectively ablating the cardiac tissue through the application of energy. Cardiac mapping, such as creating a map of the electrical potential of wave propagation along cardiac tissue (e.g., a voltage map) or a map of arrival times to various tissue location points (e.g., a LAT map), can be used to detect local cardiac tissue dysfunction. Ablation, such as cardiac mapping-based ablation, can stop or modify the propagation of unwanted electrical signals from one part of the heart to another.

[0045] The ablation process damages unwanted electrical pathways through the formation of non-conductive lesions. Energy delivery modalities use microwave, laser, and more commonly, radiofrequency energy to create conduction blocks along cardiac tissue walls. In a two-step procedure, mapping followed by ablation, activity at various points within the heart is measured (i.e., mapped), some of which are selected for ablation. Thus, electrical activity at points within the heart can be measured by advancing a catheter (such as catheter 300 of FIG. 3 ) into the heart to acquire data at multiple points, and the acquired data can then be used to select endocardial target regions for ablation, according to embodiments described herein.

[0046] Cardiac ablation and other cardiac electrophysiology procedures are becoming increasingly complex as clinicians treat challenging conditions such as AF and VT. Treatment of complex arrhythmias can now rely on the use of three-dimensional (3D) mapping systems to reconstruct the anatomical structure of the cardiac chamber of interest. For example, cardiologists rely on software such as the Complex Fractionated Atrial Electrode (CFAE) module of the CARTO® 3 3D Mapping System, manufactured by Biosense Webster, Inc. (Diamond Bar, Calif.), to analyze intracardiac EGM signals and determine ablation points for the treatment of a wide range of cardiac conditions, including atypical atrial flutter and VT. The 3D maps can provide multiple measurements of the electrophysiological properties of tissues that represent the anatomical and functional substrates of these challenging arrhythmias.

[0047] In aspects disclosed herein, the systems and methods use machine learning models (e.g., the ANN shown in FIG. 4) that can process input data such as LAT maps, voltage maps, lead ECG signal data, and pace mapping ECG signal data to identify cardiac locations that are likely to be the origin (foci) of a patient's arrhythmia.

[0048] In conventional pace mapping systems, such as that disclosed in U.S. Patent No. 7,907,994, incorporated herein by reference, VT signals are induced in a patient. Pace mapping signals are then acquired from multiple points within the heart chamber, and the acquired pace mapping signals are compared with the stimulated signals. By recognizing a high degree of correlation between the stimulated signals and one or more pace mapping signals, an arrhythmogenic focus can be identified and subsequently ablated. In conventional systems, pace mapping signals are acquired manually by a physician through trial and error. The physician introduces a pacing catheter (or electrodes) into the heart chamber, which allows the physician to apply electrical stimulation pulses to the myocardium at different locations. The resulting electrical activity (i.e., pace mapping ECG signal data) is recorded. This operation is referred to herein as pacing or pace mapping. Typically, numerous points are paced, and only a few are determined to be candidates for ablation. This conventional pace mapping process is tedious and time-consuming, and can lead to inefficiencies as a result of the trial-and-error approach to identifying pace mapping sites.

[0049] Embodiments disclosed herein utilize previously performed pace mapping cases (e.g., provided by the trial and error process described above) to construct a training data set 420. A machine learning model disclosed herein, such as 410, is trained to output data that can be utilized to predict the next cardiac location to be pace mapped by a physician. The input data used for training is data from past pace mapping procedures. For example, the input data may include electrophysiological data (e.g., lead ECG signals) of cardiac arrhythmias, pace mapping data (e.g., pace mapping ECG signals acquired from the catheter when positioned at multiple cardiac locations), LAT maps, or voltage maps. Furthermore, for each of the multiple cardiac locations, the input data used for training may also include data related to a physician's determination of whether the corresponding pace mapping data sufficiently correlates with the electrophysiological data of cardiac arrhythmias to be used as a site for ablation. Once the machine learning model is trained, it can be applied to provide a prediction of the cardiac location a physician will use as the next pacing site. Such a prediction may provide greater certainty compared to the trial and error approach described above.

[0050] FIG. 5A illustrates an exemplary pace map 540 recording the correlation between a guided ECG signal 510 (FIG. 5B) and a pace-mapping ECG signal 520 (FIG. 5C), based on which one or more features of the present disclosure can be implemented. As described above, ECG signals can be recorded while inducing an arrhythmia in a patient. For example, eight-lead guided ECG signals 510.1-510.8 are shown in FIG. 5B. During pacing, multiple pace-mapping ECG signals can be recorded by stimulating the heart 530 at various myocardial locations. For example, the eight-lead pace-mapping ECG signals 520.1-520.8 shown in FIG. 5C correspond to the specific myocardial location 550 shown in FIG. 5A. The degree of correlation between the guided ECG signal 510 and the pace-mapping ECG signal 520 indicates the likelihood that the arrhythmia induced in the patient originated from the location within the heart 550 stimulated to generate the pace-mapping ECG signal 520.

[0051] In one aspect, P i A pace map 540, denoted by P1, P2, ... P2, can be calculated for each patient i. Each element of the matrix can correspond to a cardiac location 550 where the myocardium is stimulated and can represent the correlation between the stimulated ECG signal 510 and the pace mapping ECG signal 520 corresponding to that location 550. In one aspect, multiple pace maps P1, P2, ... P2 are used. n can be calculated for each pace map P i can be represented by a 3-D matrix, with each matrix element corresponding to a 3-D location on the myocardial surface of the patient's heart (e.g., the right ventricular myocardium). Alternatively, the 3-D myocardial surface can be projected onto a 2-D planar surface to generate a pace map P, such as the 2-D matrix 540 shown in FIG. 5A. i The pace map correlation may allow for a 2D matrix representation of the correlation between the lead signal and the pacing signal. In one embodiment, the pace map value 540 may be a percentage indicating the correlation between the lead signal and the pacing signal, as described above. Pace map correlation may be determined using known methods, such as those described in U.S. Patent No. 7,907,994, which is incorporated herein by reference.

[0052] According to aspects disclosed herein, the neural network 410 can be trained to predict a pace map based on the partial maps. To that end, for each pace map P i is replicated M times. The replica is P i1 , P i2 ...P im In each of the replicas, the correlation values ​​in one or more randomly selected regions of the matrix are replaced with a predetermined value, e.g., an out-of-range number such as 999. The replaced value indicates that the correlation within that matrix element is unknown. Next, the neural network is presented with pairs of matrices, and each pair is represented by a complete map P i and the incomplete map P im That is, the training data set contains the example pairs {P 11 ,P1}, {P 12 ,P1}...{P 1m ,P1}, {P 21 ,P2}, {P 22 ,P2}...{P 2m ,P2}...{P n1 ,P n}, {P n2 ,P n}...{P nm ,P n}. Then, the neural network receives the input P i1 , P i2 ...P im The predicted P for each i In this way, the neural network "learns" to predict a perfect map (such as pace map 540) from a given imperfect map.

[0053] 6 illustrates an example for training a model 600 to predict a complete pace map from an incomplete pace map, based on which one or more features of the present disclosure can be implemented. 11 , 610.1 (e.g., pace map 540) and another replica P of P1, 620.2. 12, 610.2 (e.g., pace map 540). As mentioned above, the matrix P 11 and matrix P 12 Elements of were randomly selected and replaced with an out-of-range number, e.g., 999, to indicate that the values ​​of these selected elements are unknown. 11 and P 12 For a given map, the neural network is trained to give the original map P1. The neural network is trained to give the replica P 1m and its pair P1 is optimized to minimize the cost representing the difference between them. Setting the unknown correlation value to a number greater than the valid correlation value may contribute to faster convergence of the neural network. The neural network is trained with as many example pairs 610 and 620 as possible. The trained neural network can be applied to complete the unknown regions of a new pace map according to the "experience" gained from the pace map on which it was trained. Thus, the trained neural network can receive at its input a new, incomplete pace map and provide at its output a predicted, complete pace map.

[0054] FIG. 7 illustrates another example for training model 700 to predict a complete pace map from an incomplete pace map, based on which one or more features of the present disclosure can be implemented. In one embodiment, in addition to creating replica 710.1 or 710.3, as described above, a second matrix 710.2 or 710.4 is created, having categorical element values ​​(e.g., 0 or 1) indicating whether an element corresponds to a replaced element of 710.1 or 710.3 or an unknown element, respectively. The second matrix can cause the neural network to converge faster. Thus, in this embodiment, training example 710 may include a replica matrix (e.g., 710.1 or 710.3) and a categorical matrix (e.g., 710.2 or 710.4) and a corresponding pair 720 (e.g., 720.1 or 720.2). As before, the trained neural network can receive a new incomplete pace map at its input and provide a predicted complete pace map at its output.

[0055] In one aspect, during the pace mapping process performed by a physician, the system 100, 200 can examine the correlation percentages of the pace map recorded so far and treat the remaining elements in the pace map as unknown (e.g., the system sets the unknown elements to out-of-range values). The system 100, 200 can then provide the incomplete pace map (and, optionally, the corresponding category map, as described with reference to FIG. 7 ) as input to the neural network 410. As described above, the neural network, already trained with multiple maps from the training data set, can predict the correlation percentages in the unknown regions. With a predicted pace map fully populated with correlation values, the system 100, 200 can now suggest to the physician a direction or region within the heart for performing the next map pacing. The system's 100, 200 recommended direction or region may result in the location with the highest correlation. The system's recommendation may be indicated by a visual or audio indication pointing to the recommended direction or region. For example, the visual indication may be represented by an arrow, a star, a pin, or any similar visual indication, and may be superimposed on an image of the heart 530 displayed on the system's display.

[0056] In one aspect, instead of indicating the next pacing site that the physician should try, the system can monitor the direction in which the physician is moving the catheter and evaluate how successful that direction will be. Based on that evaluation, the system can provide a success indication as a percentage, color (e.g., green, yellow, red traffic light), brightness, or sound.

[0057] FIG. 8 is a flowchart of an exemplary method 800 for training a pace mapping prediction model, based on which one or more features of the present disclosure can be implemented. The pace mapping model can be based on a neural network, as described with reference to FIG. 4. The method 800 can receive training data sets based on which the pace mapping model is trained, and the training data sets can include data related to pace mapping procedures previously performed on patients. Thus, for each such patient, the method 800 can receive electrophysiological data related to cardiac arrhythmias endured by the patient at step 810. At step 820, the method 800 can receive pace mapping data sets. Each data set is acquired from an electrode (or catheter) when positioned at a cardiac location on the patient's heart. The method 800 can also receive correlation data at step 830 measuring the degree of correlation between each of the pace mapping data sets and the electrophysiological data. Next, at step 840, the pace mapping prediction model is trained based on the received training data sets. The pace mapping prediction model is trained to predict the degree of correlation between the electrophysiological data and the pace mapping data sets of a new patient at a pace mapping site.

[0058] FIG. 9 is a flowchart of an exemplary method 900 for applying a pace mapping prediction model, based on which one or more features of the present disclosure may be implemented. For example, as described with reference to FIG. 8 , the trained pace mapping prediction model may be applied to a new patient undergoing care during a pace mapping procedure. At step 910, the method 900 may receive as input electrophysiological data associated with a cardiac arrhythmia endured by the patient undergoing care. Furthermore, at step 920, the method 900 may receive as input a pace mapping dataset acquired from electrodes when positioned at cardiac locations on the patient's heart. At step 930, the method 900 may then provide the received input data to the trained pace mapping prediction model to obtain a predicted degree of correlation between the electrophysiological data and the pace mapping dataset. The predicted degree of correlation may be used to guide a physician performing the procedure in locating a pace mapping site that is likely to be the origin of the patient's cardiac arrhythmia. In one aspect, at step 940, the method 900 may predict a cardiac location on the patient's heart to be used for the next pace mapping during the procedure based on the predicted degree of correlation. In another aspect, method 900 can track movement of electrodes used in the procedure in step 950. Then, in step 960, method 900 can evaluate whether the movement is in a direction corresponding to an increase in correlation based on the tracked movement and the predicted correlation. In both aspects (step 940 aspect and steps 950 and 960 aspect), method 900 can utilize multiple predictions of correlation corresponding to multiple cardiac locations by performing steps 910-930 multiple times.

[0059] FIG. 10 is a flowchart of another exemplary method 1000 for training a pace mapping prediction model, based on which one or more features of the present disclosure can be implemented. The pace mapping model can be based on a neural network, as described with reference to FIG. 4. Method 1000 can receive a training dataset, based on which the pace mapping model is trained. The training dataset can include pace map pair data associated with pace mapping procedures previously performed on the patient. Thus, for each such patient, method 1000 can receive multiple pace map pairs, each pair including a complete pace map and an incomplete pace map (e.g., pair 610.1 and 620.1 or pair 610.2 and 620.2). Thus, at step 1010, method 1000 can receive complete pace maps, each of which can include a correlation matrix. Each element of the matrix can correspond to a cardiac location in the patient's heart and can represent the degree of correlation between the pace mapping dataset (corresponding to the cardiac location) and the patient's electrophysiological data. At step 1020, method 1000 may receive incomplete pace maps, each of which includes an overlap correlation matrix of the correlation matrix of the corresponding complete pace map, with one or more elements of the randomly selected overlap correlation matrix being set to a predetermined value indicating an unknown value. In one aspect, for each pair of complete and incomplete pace maps (e.g., pair 710.1 and 720.1 or pair 710.3 and 720.2), step 1020 may also receive a category matrix associated with the incomplete pace map in the pair (e.g., 710.2 associated with 710.1 or 710.4 associated with 710.3). Each element value of the category matrix may indicate whether the corresponding element value in the associated incomplete pace map is set to a predetermined value.Next, in step 1030, method 1000 can train a pace mapping prediction model to receive an incomplete pace map of a new patient that includes known and unknown correlation matrix elements, and provide a predicted complete pace map that includes predictions of the unknown correlation matrix elements.

[0060] 11 is a flowchart of another example method 1100 for applying a pace mapping prediction model, based on which one or more features of the present disclosure may be implemented. For example, as described with reference to FIG. 10, a trained pace mapping prediction model may be applied to a patient receiving care during a pace mapping procedure. Method 1100 may receive an incomplete pace map generated during a procedure at step 1110. The incomplete pace map may include a correlation matrix having known and unknown elements. Each known element of the matrix may correspond to a cardiac location in the new patient's heart and may represent a degree of correlation between the new patient's electrophysiological data and pace mapping data acquired from the electrodes when placed at the cardiac location in the new patient's heart. Next, at step 1120, method 1100 may feed the new incomplete pace map into a pace mapping prediction model to obtain a new predicted complete pace map including predictions of the unknown elements. The new predicted pace map may be used to guide a physician performing the procedure in searching for pace mapping sites that are likely to be the origin of the patient's cardiac arrhythmia. In one aspect, method 1100 may predict a cardiac location in the patient's heart to be used for the next pace mapping based on the new predicted complete pace map at step 1130. In another aspect, method 1100 may track electrode movement at step 1140. Next, in step 1150, the method 1100 can evaluate, based on the tracked movement and based on the new predicted complete pace map, whether the movement is in a direction corresponding to an increase in correlation.

[0061] Electrophysiological data related to cardiac arrhythmias endured by the patient can be derived (e.g., as described above with respect to methods 800, 900, 1000, and 1100). For example, the patient may be experiencing VT induced by an arrhythmogenic drug, such as isoproterenol, or by undergoing strenuous activity.

[0062] In one aspect, as described with reference to FIGS. 4-11 , training and application of the machine learning model can be performed in real time on a server at the facility where the cardiac procedure is being performed, such as a hospital or medical facility, or at a remote location, such as the cloud or a training center, by the system 100 ( FIG. 1 ) or 200 ( FIG. 2 ) described herein, which also refer to the CARTO® 3 3D mapping system. In one aspect, a vendor of system 100 ( FIG. 1 ) or 200 ( FIG. 2 ) can provide such a system with a pre-trained pace mapping prediction model. Hospitals can continue to train the system (e.g., to update the pace mapping prediction model based on expanded or new training data sets). In one aspect, a single pace mapping prediction model can be maintained for all hospitals or groups of hospitals, or every hospital can maintain its own pace mapping prediction model.

[0063] In one aspect of the present application, a machine learning model is utilized to identify a sequence of pacing pulses in a pacing procedure workflow and automatically measure the interval between the last pacing pulse (in the pacing sequence) and the first intrinsic beat following the last pacing pulse, where such interval measurements can be obtained from time or voltage calipers associated with a particular ECG signal.

[0064] Some electrophysiology procedures require pacing maneuvers for various arrhythmias (e.g., AF or VT), in which a train of pacing pulses may be generated. Pacing can occur at one or more cardiac locations and can be measured at one or more cardiac locations and on body surface electrodes. The train of pacing pulses may be generated at equal or different time distances. The system operator can then open time or voltage calipers associated with a particular ECG signal and measure the distance from the last pacing pulse to the first intrinsic beat. Pacing maneuvers may be useful for characterizing cardiac tissue, inferring the presence of short pathways, and locating reentrant circuits.

[0065] 12 illustrates an example ECG trace 1200 of pacing maneuvers and time caliper measurements manually obtained by a physician, based on which one or more features of the present disclosure can be implemented. For example, with respect to ablation catheter ABLd 1250, a pacing maneuver can include a pacing sequence including pulses, each having an associated duration, e.g., 1210.1-1210.4. For example, the last pulse has a duration 1210.4 of 340 milliseconds (ms). The pacing maneuver can also include intrinsic pulses, each having associated durations 1210.5-1210.N. A post-pacing interval (PPI) can be defined as the interval 1210.4 extending from the start time of the last pulse 1220 to the time of the first intrinsic beat 1230 (hereafter, the next intrinsic beat). Both the pacing sequences and the intrinsic beats may be associated with different cardiac locations and may have different durations, eg, 1210.1-1210.N.

[0066] In one aspect, a machine learning algorithm can be applied to detect interval measurements (of time and / or voltage calipers) based on pacing maneuvers, which may be accepted, rejected, or modified by a physician, whereby a machine learning model is trained based on a training dataset that includes pacing maneuvers and corresponding interval measurements manually obtained from a physician. In one aspect, the pacing maneuvers and corresponding interval measurements of the training data set are associated with different cardiac locations and have different durations. During training of the neural network (training phase) and application of the trained neural network (inference phase), the interval measurements can include the start and end of a period (or voltage) and can be represented by: 1) post-pacing intervals (time from the last pacing spike to the first intrinsic beat), e.g., 1210.4; 2) pacing train characteristics (regular and irregular time intervals), e.g., 1210.1; 3) tachycardia cycle length; 4) similar measurements on coronary sinus catheter electrodes; 5) similar measurements on electrodes of other catheters; and 6) combinations thereof.

[0067] 13A-13B illustrate exemplary flow charts illustrating a machine learning model training method (FIG. 13A) and a machine learning model application (FIG. 13B), based on which one or more features of the present disclosure may be implemented. Steps 1310-1330 describe the training phase of the machine learning algorithm. Thus, in step 1310, pacing maneuvers are received, each of which may include a pacing sequence associated with a cardiac location and a subsequent intrinsic beat. In step 1320, interval measurements are received, each corresponding to a pacing maneuver. Each received interval measurement 1210.4 may relate to the distance between the last pacing pulse and the next intrinsic beat. Next, in step 1330, a machine learning model can be trained based on the received pacing maneuvers 1310 and the corresponding received interval measurements 1320 to predict interval measurements when a new pacing maneuver is presented. Thus, in one aspect, the machine learning model can be trained based on exemplary pairs of training data, each exemplary pair including a pacing maneuver (including a pacing sequence and a subsequent intrinsic beat) and a corresponding interval measurement. For example, a training pair can include a sequence of 300 ms, 290 ms, 280 ms, and 270 ms and a corresponding interval measurement (e.g., a time caliper) measuring the distance between the last pacing pulse 1220 and the next intrinsic beat 1230, e.g., measured from ±10 ms around the last pacing pulse 1220 to ±10 ms around the next intrinsic beat 1230. The interval measurement may be manually determined by a physician based on each pacing maneuver. Thus, the machine learning model learns the physician's preference for adjusting the caliper on the last pacing pulse or mapping annotation, e.g., in interval steps of ±10 ms.

[0068] Once the machine learning model is trained, as described with reference to FIG. 13A, it can be applied in an inference phase as shown in FIG. 13B. Thus, in step 1340, the machine learning model can receive as input a pacing maneuver, which may include a pacing sequence and a subsequent intrinsic beat. Based on the received pacing maneuvers as trained, the model can output a prediction of an interval measurement in step 1350. For example, if the machine learning model receives a sequence of 301 ms, 289 ms, 282 ms, and 269 ms, the machine learning model can predict an interval measurement (caliper interval) associated with the distance 1210.4 between the last pacing pulse and the next intrinsic beat. The physician can optionally accept, reject, or modify the predicted caliper interval. For example, if accepted by the physician, the predicted caliper interval can be stored and / or used to update the EP cardiac map generated by system 100 (FIG. 1) or 200 (FIG. 2).

[0069] In one aspect, predicted interval measurements (i.e., time or voltage calipers) can be utilized to update the EP map to aid in tissue characterization, identifying the presence of short pathways, locating reentrant circuits, etc. For example, elements of the EP map can represent caliper intervals at corresponding pacing locations of the heart. In one aspect, the EP map may be color-coded to identify any of the foregoing.

[0070] FIG. 14 illustrates a functional block diagram 1400 of an exemplary RNN 1420, based on which one or more features of the present disclosure can be implemented. Various machine learning models can be applied to implement the features described with reference to FIGS. 13A-13B, such as the neural networks described with reference to FIGS. 4 and 14. For example, the RNN 1420 can be used. In one aspect, the RNN 1420 can receive pacing pulses and intrinsic beats as input data 1410. The RNN 1420 is trained to generate output 1430, such as an interval measurement between the last pacing pulse and a mapping annotation (i.e., an intrinsic beat), based on the input data 1410. The more input data 1410 received by the RNN 1420, the more accurate the output 1430 can be. The output of the RNN 1420, such as output 1430, can be used to train the RNN 1400, as indicated by arrow 1440 in FIG. 14. Thus, if the output 1430 is accepted by the physician, the output 1430 can be used as the input 1410 for training the RNN 1420.

[0071] In one aspect, as described with reference to Figures 12-14, the training and application of the machine learning model can be performed in real time on a server at the facility where the cardiac procedure is being performed, such as a hospital or medical facility, or at a remote location, such as the cloud or a training center, by system 100 (Figure 1) or 200 (Figure 2) described herein, which also represent a mapping system.

[0072] Although features and elements are described above in particular combinations, those skilled in the art will understand that each feature or element can be used alone or in combination with other features and elements. Further, although process steps are described above in a particular order, the steps can be performed in any other desired order.

[0073] The methods, processes, modules, and systems described herein may be implemented in a computer program, software, or firmware embodied 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, optical media such as CD-ROM disks, and digital versatile disks (DVDs). A processor in association with software may be used to implement a radio frequency transceiver for use in a WTRU, UE, terminal, base station, RNC, or any host computer.

[0074] Further embodiments of the present specification may be formed by supplementing an embodiment with one or more elements from any one or more other embodiments of the present specification and / or by replacing one or more elements from one embodiment with one or more elements from one or more other embodiments of the present specification.

[0075] It is understood, therefore, that the disclosed subject matter is not limited to the particular embodiments disclosed, but is intended to encompass all modifications that are within the spirit and scope of the present disclosure as defined by the appended claims, the above description, and / or as illustrated by the accompanying drawings.

[0076] [Embodiment] (1) A method for training a pace mapping prediction model, comprising: receiving a training data set related to a patient's heart, for each patient, the training data set comprising: electrophysiological data related to the patient's cardiac arrhythmia; pace mapping data sets, each data set acquired from an electrode when positioned at a cardiac location on the patient's heart; correlation data measuring a degree of correlation between each of the pace mapping data sets and the electrophysiological data; training the pace mapping prediction model based on the training data set to predict a degree of correlation between electrophysiological data associated with a new patient and a pace mapping data set. (2) During a cardiac pace mapping procedure on a new patient, receiving input data, the input data comprising: electrophysiological data related to cardiac arrhythmias of the new patient; and a pace mapping data set obtained from the electrodes when positioned at cardiac locations on the new patient's heart; 2. The method of claim 1, further comprising: providing the input data to the pace mapping prediction model to obtain a predicted degree of correlation between the electrophysiological data of the input data and the pace mapping dataset. (3) The method of embodiment 2, further comprising predicting a cardiac position in the new patient's heart to be used in subsequent pace mapping based on the predicted correlation. (4) tracking the movement of the electrodes; The method of embodiment 2, further comprising: evaluating, based on the tracked movement and based on the predicted correlation, whether the movement is in a direction corresponding to an increase in correlation. (5) the training data set includes, for each patient, a pair of complete and incomplete pace maps; the complete pace map includes a correlation matrix, each element of the matrix corresponding to a cardiac location and representing a degree of correlation between the patient's electrophysiological data and a pace mapping data set among the pace mapping data sets corresponding to the cardiac location; the incomplete pace map includes a duplicate correlation matrix of the correlation matrix of the complete pace map, and one or more randomly selected elements of the duplicate correlation matrix are set to a predetermined value representing an unknown value; 2. The method of claim 1, wherein the training includes training the pace mapping prediction model to receive an incomplete pace map of a new patient and to output a predicted complete pace map.

[0077] (6) The method of embodiment 5 further includes, for each pair of complete and incomplete pace maps, a category matrix is ​​associated with the incomplete pace map in the pair, and each element value of the category matrix indicates whether the corresponding element value in the associated incomplete pace map is set to the predetermined value. (7) During cardiac pace mapping of a new patient, receiving a new incomplete pace map, the incomplete pace map including a correlation matrix having known elements and unknown elements, each of the known elements of the matrix corresponding to a cardiac location on the new patient's heart and representing a degree of correlation between the new patient's electrophysiological data and pace mapping data obtained from an electrode when positioned at the cardiac location on the new patient's heart; 6. The method of claim 5, further comprising: supplying the new incomplete pace map to the pace mapping prediction model to obtain a new predicted complete pace map. (8) The method of embodiment 7, further comprising predicting cardiac positions in the new patient's heart to be used in subsequent pace mapping based on the new predicted complete pace map. (9) tracking the movement of the electrodes; The method of embodiment 7, further comprising: evaluating whether the movement is in a direction corresponding to an increase in correlation based on the tracked movement and based on the new predicted complete pace map. (10) The method described in embodiment 1, wherein the cardiac arrhythmia is ventricular tachycardia.

[0078] (11) The method of embodiment 1, wherein the cardiac arrhythmia is induced by the electrode and the electrophysiological data is induced electrophysiological data. (12) The method of embodiment 1, wherein each of the electrophysiological data and the pace mapping data set includes electrocardiogram signal data. (13) The method of embodiment 12, wherein the electrophysiological data includes lead electrocardiogram signal data. (14) The method of embodiment 1, wherein the pace mapping prediction model is a neural network. (15) The method of embodiment 1, wherein the training data set is obtained from a database.

[0079] (16) The method of embodiment 1, wherein the training data set further includes one of a local activation time map or a voltage map. (17) The method of embodiment 1, wherein the electrode is disposed on a catheter. (18) The method of embodiment 1, wherein the electrodes include a plurality of electrodes positioned at a plurality of cardiac locations, and each of the pace mapping datasets is acquired sequentially or simultaneously from the plurality of electrodes. (19) A system for training a pace mapping prediction model, comprising: at least one processor; a memory storing instructions that, when executed by the at least one processor, cause the system to: receiving a training data set related to a patient's heart, for each patient, the training data set comprising: electrophysiological data related to the patient's cardiac arrhythmia; pace mapping data sets, each data set acquired from an electrode when positioned at a cardiac location on the patient's heart; correlation data measuring a degree of correlation between each of the pace mapping data sets and the electrophysiological data; and a memory storing instructions for training the pace mapping prediction model based on the training dataset to predict a degree of correlation between electrophysiological data associated with a new patient and a pace mapping dataset. (20) A non-transitory computer-readable medium comprising instructions executable by at least one processor to perform a method for training a pace mapping prediction model, the method comprising: receiving a training data set related to a patient's heart, for each patient, the training data set comprising: electrophysiological data related to the patient's cardiac arrhythmia; pace mapping data sets, each data set acquired from an electrode when positioned at a cardiac location on the patient's heart; correlation data measuring a degree of correlation between each of the pace mapping data sets and the electrophysiological data; training the pace mapping prediction model based on the training dataset to predict a degree of correlation between electrophysiological data associated with a new patient and a pace mapping dataset.

[0080] (21) A method for training a pacing operation prediction model, comprising: receiving a training data set related to a patient's heart, for each patient, the training data set comprising: pacing maneuvers, each associated with a pacing location on the patient's heart; corresponding interval measurements, each relating to the distance between the last pacing pulse and an intrinsic beat since the corresponding pacing maneuver; training the pacing maneuver prediction model based on the training data set to predict interval measurements based on pacing maneuvers associated with a new patient. (22) During a cardiac pacing procedure on a new patient, receiving a pacing maneuver associated with a pacing location in the new patient's heart; 22. The method of claim 21, further comprising: supplying the received pacing operation to the pacing operation prediction model to obtain predicted interval measurements. (23) The training data set further includes, for each of the pacing operations, a last pacing pulse and an intrinsic beat detected from the pacing operation; 23. The method of embodiment 22, wherein the pacing operation prediction model is further trained to detect pacing pulses and intrinsic beats from received pacing operations. (24) The method of embodiment 22, wherein the interval measurements of the training data set and the predicted interval measurements measure a time caliper. (25) The method of embodiment 22, wherein the interval measurements of the training data set and the predicted interval measurements measure voltage calipers.

[0081] (26) The method of embodiment 22, wherein the pacing operations of the training data set and the pacing operations associated with the new patient are derived from an electrocardiogram. (27) The method of embodiment 22, further comprising updating an electrophysiological map using the predicted interval measurements. (28) The method of embodiment 22, wherein the pacing operations of the training data set and the pacing operations associated with the new patient each include a pacing sequence and a subsequent intrinsic beat. (29) A system for training a pacing operation prediction model, comprising: at least one processor; a memory storing instructions that, when executed by the at least one processor, cause the system to: receiving a training data set related to a patient's heart, for each patient, the training data set comprising: pacing maneuvers, each associated with a pacing location on the patient's heart; corresponding interval measurements, each relating to the distance between the last pacing pulse and an intrinsic beat since the corresponding pacing maneuver; and a memory storing instructions to train the pacing maneuver prediction model based on the training data set to predict interval measurements based on pacing maneuvers associated with a new patient. (30) A non-transitory computer-readable medium comprising instructions executable by at least one processor to perform a method for training a pacing operation prediction model, the method comprising: receiving a training data set related to a patient's heart, for each patient, the training data set comprising: pacing maneuvers, each associated with a pacing location on the patient's heart; corresponding interval measurements, each relating to the distance between the last pacing pulse and an intrinsic beat since the corresponding pacing maneuver; training the pacing maneuver prediction model based on the training data set to predict interval measurements based on pacing maneuvers associated with a new patient.

Claims

1. 1. A system for training a pace mapping prediction model, comprising: at least one processor; a memory storing instructions that, when executed by the at least one processor, cause the system to: receiving a training data set related to a patient's heart, for each patient, the training data set comprising: electrophysiological data related to the patient's cardiac arrhythmia; pace mapping data sets, each data set acquired from an electrode when positioned at a cardiac location on the patient's heart; correlation data, the correlation data being a value indicative of a correlation between each of the pace mapping data sets and the electrophysiological data, measuring a degree of correlation; and training the pace mapping prediction model based on the training data set to predict the degree of correlation between electrophysiological data associated with a new patient and a pace mapping data set. the training data set includes, for each patient, a pair of complete and incomplete pace maps; the complete pace map includes a correlation matrix, each element of the matrix corresponding to one cardiac location and representing the degree of correlation between the patient's electrophysiological data and one of the pace mapping data sets corresponding to the one cardiac location; the incomplete pace map includes an overlapping correlation matrix of the correlation matrix of the complete pace map, and one or more randomly selected elements of the overlapping correlation matrix are set to a predetermined value representing an unknown value; The training includes training the pace mapping prediction model to receive an incomplete pace map of a new patient and to output a predicted complete pace map.

2. 1. A non-transitory computer-readable medium comprising instructions executable by at least one processor to perform a method for training a pace mapping prediction model, the method comprising: receiving a training data set related to a patient's heart, for each patient, the training data set comprising: electrophysiological data related to the patient's cardiac arrhythmia; pace mapping data sets, each data set acquired from an electrode when positioned at a cardiac location on the patient's heart; correlation data, the correlation data being a value indicative of a correlation between each of the pace mapping data sets and the electrophysiological data, measuring a degree of correlation; training the pace mapping prediction model based on the training data set to predict the degree of correlation between electrophysiological data associated with a new patient and a pace mapping data set; the training data set includes, for each patient, a pair of complete and incomplete pace maps; the complete pace map includes a correlation matrix, each element of the matrix corresponding to one cardiac location and representing the degree of correlation between the patient's electrophysiological data and one of the pace mapping data sets corresponding to the one cardiac location; the incomplete pace map includes an overlapping correlation matrix of the correlation matrix of the complete pace map, and one or more randomly selected elements of the overlapping correlation matrix are set to a predetermined value representing an unknown value; The non-transitory computer-readable medium, wherein the training includes training the pace mapping prediction model to receive an incomplete pace map of a new patient and to output a predicted complete pace map.

3. 1. A method for training a pace mapping prediction model, comprising: receiving a training data set related to a patient's heart, for each patient, the training data set comprising: electrophysiological data related to the patient's cardiac arrhythmia; pace mapping data sets, each data set acquired from an electrode when positioned at a cardiac location on the patient's heart; correlation data, the correlation data being a value indicative of a correlation between each of the pace mapping data sets and the electrophysiological data, measuring a degree of correlation; training the pace mapping prediction model based on the training data set to predict a degree of correlation between electrophysiological data associated with a new patient and a pace mapping data set; the training data set includes, for each patient, a pair of complete and incomplete pace maps; the complete pace map includes a correlation matrix, each element of the matrix corresponding to one cardiac location and representing the degree of correlation between the patient's electrophysiological data and one of the pace mapping data sets corresponding to the one cardiac location; the incomplete pace map includes an overlapping correlation matrix of the correlation matrix of the complete pace map, and one or more randomly selected elements of the overlapping correlation matrix are set to a predetermined value representing an unknown value; The method, wherein the training includes training the pace mapping prediction model to receive an incomplete pace map of a new patient and to output a predicted complete pace map.

4. The instruction is to instruct the system to: During a cardiac pace mapping procedure on a new patient, receiving input data, the input data comprising: electrophysiological data related to cardiac arrhythmias of the new patient; and a pace mapping data set obtained from the electrodes when positioned at cardiac locations on the new patient's heart; 2. The system of claim 1, further comprising: providing the input data to the pace mapping prediction model to obtain the predicted degree of correlation between the electrophysiological data of the input data and the pace mapping data set.

5. The instructions are provided to the system: The system of claim 4 , further comprising predicting cardiac positions in the new patient's heart for use in subsequent pace mapping based on the predicted degree of correlation.

6. The instructions are provided to the system as follows: tracking the movement of the electrodes; 5. The system of claim 4, further comprising: evaluating, based on the tracked movement and based on the predicted degree of correlation, whether the movement is in a direction corresponding to an increase in the degree of correlation.

7. 2. The system of claim 1, further comprising: for each of the pairs of complete and incomplete pace maps, a category matrix is ​​associated with the incomplete pace map in the pair, and each element value of the category matrix indicates whether a corresponding element value in the associated incomplete pace map is set to the predetermined value.

8. The instructions are provided to the system: During cardiac pace mapping of a new patient, receiving a new incomplete pace map, the incomplete pace map including a correlation matrix having known elements and unknown elements, each of the known elements of the matrix corresponding to a cardiac location on the new patient's heart and representing the degree of correlation between the new patient's electrophysiological data and pace mapping data acquired from an electrode when positioned at the cardiac location on the new patient's heart; 2. The system of claim 1, further comprising: feeding the new incomplete pace map to the pace mapping prediction model to obtain a new predicted complete pace map.

9. The instructions are provided to the system:

10. The system of claim 8, further comprising predicting cardiac positions in the new patient's heart for use in subsequent pace mapping based on the new predicted complete pace map.

10. The instructions are provided to the system: tracking the movement of the electrodes; 9. The system of claim 8, further comprising: evaluating whether the movement is in a direction corresponding to an increase in the correlation based on the tracked movement and based on the new predicted complete pace map.

11. The system of claim 1 , wherein the cardiac arrhythmia is ventricular tachycardia.

12. The system of claim 1 , wherein the cardiac arrhythmia is induced by the electrodes and the electrophysiological data is induced electrophysiological data.

13. The system of claim 1 , wherein the electrophysiological data and the pace mapping data set each include electrocardiogram signal data.

14. The system of claim 13 , wherein the electrophysiological data includes lead electrocardiogram signal data.

15. The system of claim 1 , wherein the pace mapping prediction model is a neural network.

16. The system of claim 1 , wherein the training data set is obtained from a database.

17. The system of claim 1 , wherein the training data set further includes one of a local activation time map or a voltage map.

18. The system of claim 1 , wherein the electrodes are disposed on a catheter.

19. The system of claim 1 , wherein the electrodes include a plurality of electrodes positioned at a plurality of cardiac locations, and wherein each of the pace mapping datasets is acquired sequentially or simultaneously from the plurality of electrodes.

20. 1. A system for training a pacing operation prediction model, comprising: at least one processor; a memory storing instructions that, when executed by the at least one processor, cause the system to: receiving a training data set related to a patient's heart, for each patient, the training data set comprising: pacing maneuvers, each associated with a pacing location on the patient's heart; corresponding interval measurements, each relating to the distance between the last pacing pulse and an intrinsic beat since the corresponding pacing maneuver; and a memory storing instructions to train the pacing maneuver prediction model based on the training data set to predict interval measurements based on pacing maneuvers associated with a new patient.

21. 1. A non-transitory computer-readable medium comprising instructions executable by at least one processor to perform a method for training a pacing operation prediction model, the method comprising: receiving a training data set related to a patient's heart, for each patient, the training data set comprising: pacing maneuvers, each associated with a pacing location on the patient's heart; corresponding interval measurements, each relating to the distance between the last pacing pulse and an intrinsic beat since the corresponding pacing maneuver; training the pacing maneuver prediction model based on the training data set to predict interval measurements based on pacing maneuvers associated with a new patient.

22. The instruction: During a new patient's cardiac pacing procedure, receiving a pacing maneuver associated with a pacing location in the new patient's heart; 21. The system of claim 20, further comprising: feeding the received pacing maneuvers into the pacing maneuver prediction model to obtain predicted interval measurements.

23. the training data set further includes, for each of the pacing maneuvers, a last pacing pulse and an intrinsic beat detected from the pacing maneuver; 23. The system of claim 22, wherein the pacing maneuver prediction model is further trained to detect pacing pulses and intrinsic beats from received pacing maneuvers.

24. 23. The system of claim 22, wherein the interval measurements of the training data set and the predicted interval measurements measure a time caliper.

25. 23. The system of claim 22, wherein the interval measurements of the training data set and the predicted interval measurements measure voltage calipers.

26. 23. The system of claim 22, wherein the pacing maneuvers of the training data set and the pacing maneuvers associated with the new patient are derived from an electrocardiogram.

27. The instruction:

23. The system of claim 22, further comprising updating an electrophysiological map with the predicted interval measurements.

28. 23. The system of claim 22, wherein the pacing maneuvers of the training data set and the pacing maneuvers associated with the new patient each include a pacing sequence and a subsequent intrinsic beat.

Images