Improve mapping efficiency by suggesting locations for mapping points

Abstract

The present invention is entitled "Improving Mapping Efficiency by Suggesting Mapping Point Locations." A method and apparatus for improving mapping efficiency by suggesting mapping point locations, the method comprising: receiving data at a machine, the data comprising a plurality of signals received during triangulation to locate a focal tachycardia; generating, by the machine, a predictive model regarding the location of the focal tachycardia; and modifying, by the machine, the predictive model based on additional data received by the machine.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of U.S. Provisional Application No. 63 / 046,948, filed on July 1, 2020, the contents of which are incorporated herein by reference. Technical Field

[0003] The present invention relates to artificial intelligence and machine learning associated with improving mapping efficiency by suggesting mapping point locations. Background Art

[0004] When attempting to locate activation source points using triangulation methods, multiple acquisition points are required and no data exists on how these points need to be spatially distributed. Summary of the invention

[0005] A method and apparatus for improving mapping efficiency by suggesting mapping point locations, the method comprising: receiving data at a machine, the data comprising multiple signals received during triangulation to locate focal tachycardia; generating, by the machine, a predictive model regarding the location of the focal tachycardia; and modifying, by the machine, the predictive model based on additional data received by the machine. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] A more detailed understanding may be obtained from the following detailed description provided by way of example in conjunction with the accompanying drawings, in which like reference numerals indicate like elements, and in which:

[0007] Figure 1 is a block diagram of an exemplary system for remotely monitoring and transmitting patient biometrics;

[0008] Figure 2 is a system diagram of an example of a computing environment in communication with a network;

[0009] Figure 3 is a block diagram of an exemplary apparatus that may implement one or more features of the present disclosure;

[0010] Figure 4 Shows the combination Figure 3 A graphical depiction of an artificial intelligence system of an exemplary apparatus;

[0011] Figure 5 Shown in Figure 4 Methods for implementation in artificial intelligence systems;

[0012] Figure 6 An example of a probability calculated by Naive Bayes is shown;

[0013] Figure 7 An exemplary decision tree is shown;

[0014] Figure 8 An exemplary random forest classifier is shown;

[0015] Fig. 9 An exemplary logistic regression is shown;

[0016] Fig.10 An exemplary support vector machine is shown;

[0017] Fig.11 An exemplary linear regression model is shown;

[0018] Fig.12 An exemplary K-means clustering is shown;

[0019] Fig.13 An exemplary ensemble learning algorithm is shown;

[0020] Fig.14 An exemplary neural network is shown;

[0021] Fig.15 A hardware-based neural network is shown;

[0022] FIG. 16A to FIG. 16D Examples of cardiomyopathies with different etiologies are shown;

[0023] Fig.17 is an illustration of an exemplary system in which one or more features of the disclosed subject matter may be implemented; and

[0024] Fig.18 is an exemplary illustration of triangulation according to an embodiment;

[0025] Fig.19 is an exemplary flow chart of an exemplary method for improving mapping efficiency by suggesting mapping point locations according to an embodiment;

[0026] Fig. 20 A graphical depiction of the system and information flow showing a single query point;

[0027] Fig.21 Shows the use of Fig. 20 system to infer the delineation of the next region of interest;

[0028] Fig. 22 Shown for Fig. 20 The architecture of the transformer network in the system;

[0029] Fig.23 An example Vnet network architecture is shown; and

[0030] Fig.24 Shown as Fig. 20 The integrated network architecture of NN in . DETAILED DESCRIPTION

[0031] While the details of the application will be described herein, simple machine learning (ML) is used to train the system to improve mapping efficiency by guiding the user to collect points in informative locations.

[0032] A method and apparatus for improving mapping efficiency by suggesting mapping point locations, the method comprising: receiving data at a machine, the data comprising multiple signals received during triangulation to locate a focal tachycardia; generating, by the machine, a predictive model regarding the location of the focal tachycardia; and modifying, by the machine, the predictive model based on additional data received by the machine.

[0033] Figure 1 is a block diagram of an exemplary system 100 for remotely monitoring and transmitting patient biometrics (i.e., patient data). Figure 1 In the example shown, the system 100 includes a patient biometric monitoring and processing device 102 associated with a patient 104 , a local computing device 106 , a remote computing system 108 , a first network 110 , and a second network 120 .

[0034] According to an embodiment, the monitoring and treatment device 102 can be a device (e.g., implantable subcutaneously) in the patient's body. The monitoring and treatment device 102 can be inserted into the patient's body via any suitable means, including oral injection, surgical insertion via a vein or artery, endoscopic procedure, or laparoscopic procedure.

[0035] Depending on the embodiment, the monitoring and processing device 102 can be a device that is external to the patient. For example, as described in more detail below, the monitoring and processing device 102 can include an attachable patch (e.g., which is attached to the patient's skin). The monitoring and processing device 102 can also include a catheter with one or more electrodes, a probe, a blood pressure cuff, a weight scale, a bracelet or smart watch biometric tracker, a glucose monitor, a continuous positive airway pressure (CPAP) machine, or virtually any device that can provide input related to the patient's health or biometrics.

[0036] According to one embodiment, the monitoring and treatment device 102 may include components that are internal to the patient and components that are external to the patient.

[0037] exist Figure 1 1. A single monitoring and processing device 102 is shown in FIG. However, the exemplary system may include multiple patient biometric monitoring and processing devices. The patient biometric monitoring and processing device may communicate with one or more other patient biometric monitoring and processing devices. Additionally or alternatively, the patient biometric monitoring and processing device may communicate with a network 110.

[0038] One or more monitoring and processing devices 102 may acquire patient biometric data (e.g., electrical signals, blood pressure, temperature, blood glucose level, or other biometric data) and receive at least a portion of the patient biometric data representing the acquired patient biometrics and additional information associated with the acquired patient biometrics from one or more other monitoring and processing devices 102. The additional information may be, for example, diagnostic information and / or additional information obtained from an additional device such as a wearable device. Each monitoring and processing device 102 may process data including its own acquired patient biometrics and data received from one or more other monitoring and processing devices 102.

[0039] exist Figure 1 In the embodiment of the present invention, the network 110 is an example of a short-range network, such as a local area network (LAN) or a personal area network (PAN). Information can be sent between the monitoring and processing device 102 and the local computing device 106 via the short-range network 110 using any of a variety of short-range wireless communication protocols, such as Bluetooth, Wi-Fi, Zigbee, Z-Wave, near field communication (NFC), ultraband, Zigbee, or infrared (IR).

[0040] The network 120 may be a wired network, a wireless network, or include one or more wired and wireless networks. For example, the network 120 may be a long-range network (e.g., a wide area network (WAN), the Internet, or a cellular network). Information may be sent via the network 120 using any of a variety of long-range wireless communication protocols (e.g., TCP / IP, HTTP, 3G, 4G / LTE, or 5G / New Radio).

[0041] The patient monitoring and processing device 102 may include a patient biometric sensor 112, a processor 114, a user input (UI) sensor 116, a memory 118, and a transmitter-receiver (i.e., transceiver) 122. The patient monitoring and processing device 102 may continuously or periodically monitor, store, process, and transmit any number of various patient biometrics via the network 110. Examples of patient biometrics include electrical signals (e.g., ECG signals and brain biometrics), blood pressure data, blood glucose data, and temperature data. Patient biometrics may be monitored and transmitted to treat any number of various diseases, such as cardiovascular diseases (e.g., arrhythmias, cardiomyopathy, and coronary artery disease) and autoimmune diseases (e.g., type I and type II diabetes).

[0042] The patient biometric sensor 112 may include, for example, one or more sensors configured to sense a type of biometric patient biometric. For example, the patient biometric sensor 112 may include electrodes configured to acquire electrical signals (e.g., heart signals, brain signals, or other bioelectrical signals), temperature sensors, blood pressure sensors, blood glucose sensors, blood oxygen sensors, pH sensors, accelerometers, and microphones.

[0043] As described in more detail below, the patient biometric monitoring and processing device 102 may be an ECG monitor for monitoring ECG signals of the heart. The patient biometric sensor 112 of the ECG monitor may include one or more electrodes for acquiring ECG signals. The ECG signals may be used to treat various cardiovascular diseases.

[0044] In another example, the patient biometric monitoring and processing device 102 can be a continuous glucose monitor (CGM) for continuously monitoring the patient's blood glucose level to treat various diseases, such as type I and type II diabetes. The CGM can include electrodes disposed subcutaneously that can monitor blood glucose levels from the patient's interstitial fluid. The CGM can be, for example, a component of a closed-loop system in which blood glucose data is sent to an insulin pump for calculating insulin delivery without user intervention.

[0045] The transceiver 122 may include a separate transmitter and receiver. Alternatively, the transceiver 122 may include a transmitter and receiver integrated into a single device.

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

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

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

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

[0050] In some embodiments, the remote computing system 108 may be configured to receive at least one of the monitored patient biometric and information associated with the monitored patient via the network 120 as a remote network. For example, if the local computing device 106 is a mobile phone, the network 120 may be a wireless cellular network, and information may be transmitted between the local computing device 106 and the remote computing system 108 via a wireless technology standard such as any of the wireless technologies described above. As described in more detail below, the remote computing system 108 may be configured to provide (e.g., visually display and / or audibly provide) at least one of the patient biometric and the related information to a healthcare professional (e.g., a physician).

[0051] Figure 21 is a system diagram of an example of a computing environment 200 in communication with a network 120. In some cases, computing environment 200 is incorporated into a public cloud computing platform (such as Amazon Web Services or Microsoft Azure), a hybrid cloud computing platform (such as HP Enterprise OneSphere), or a private cloud computing platform.

[0052] like Figure 2 As shown, computing environment 200 includes a remote computing system 108 (hereinafter computer system), which is one example of a computing system on which embodiments described herein may be implemented.

[0053] The remote computing system 108 may perform various functions via a processor 220, which may include one or more processors. Functions may include analyzing monitored patient biometrics and associated information, and providing (e.g., via a display 266) alerts, additional information, or instructions based on physician-determined or algorithm-driven thresholds and parameters. As described in more detail below, the remote computing system 108 may be used to provide a patient information dashboard (e.g., via a display 266) to a healthcare provider (e.g., a physician) such that such information may enable the healthcare provider to identify and prioritize patients with more critical needs than others.

[0054] like Figure 2 As shown, the computer system 210 may include a communication mechanism such as a bus 221 or other communication mechanism for transmitting information within the computer system 210. The computer system 210 also includes one or more processors 220 coupled to the bus 221 for processing information. The processor 220 may include one or more CPUs, GPUs, or any other processors known in the art.

[0055] The computer system 210 also includes a system memory 230 coupled to the bus 221 for storing information and instructions to be executed by the processor 220. The system memory 230 may include computer-readable storage media in the form of volatile and / or non-volatile memory, such as a read-only system memory (ROM) 231 and / or a random access memory (RAM) 232. The system memory RAM 232 may include other dynamic storage devices (e.g., dynamic RAM, static RAM, and synchronous DRAM). The system memory ROM 231 may include other static storage devices (e.g., programmable ROM, erasable PROM, and electrically erasable PROM). In addition, the system memory 230 may be used to store temporary variables or other intermediate information during the execution of instructions by the processor 220. The basic input / output system 233 (BIOS) may include routines for transferring information between elements within the computer system 210, such as during the boot process, which may be stored in the system memory ROM 231. The RAM 232 may contain data and / or program modules that are immediately accessible to and / or currently being operated by the processor 220. System memory 230 may additionally include, for example, operating system 234 , application programs 235 , other program modules 236 , and program data 237 .

[0056] The illustrated computer system 210 also includes a disk controller 240 that is coupled to the bus 221 to control one or more storage devices for storing information and instructions, such as a hard disk 241 and a removable media drive 242 (e.g., a floppy disk drive, an optical drive, a tape drive, and / or a solid-state drive). Storage devices may be added to the computer system 210 using an appropriate device interface (e.g., small computer system interface (SCSI), integrated device electronics (IDE), universal serial bus (USB), or FireWire).

[0057] The computer system 210 may also include a display controller 265 coupled to the bus 221 to control a monitor or display 266, such as a cathode ray tube (CRT) or a liquid crystal display (LCD), to display information to a computer user. The illustrated computer system 210 includes a user input interface 260 and one or more input devices, such as a keyboard 262 and a pointing device 261, for interacting with a computer user and providing information to the processor 220. The pointing device 261 may be, for example, a mouse, a trackball, or a pointing stick for communicating directional information and command selections to the processor 220 and for controlling cursor movement on the display 266. The display 266 may provide a touch screen interface that may allow input to supplement or replace the communication of directional information and command selections by the pointing device 261 and / or the keyboard 262.

[0058] In response to processor 220 executing one or more sequences of one or more instructions contained in a memory such as system memory 230, computer system 210 may perform a portion or each of the functions and methods described herein. Such instructions may be read into system memory 230 from another computer-readable medium such as hard disk 241 or removable media drive 242. Hard disk 241 may contain one or more data repositories and data files used by the embodiments described herein. The data repository contents and data files may be encrypted to increase security. Processor 220 may also be employed in a multi-processing arrangement to execute one or more sequences of instructions contained in system memory 230. In an alternative embodiment, hard-wired circuitry may be used in place of or in combination with software instructions. Therefore, embodiments are not limited to any particular combination of hardware circuitry and software.

[0059] As described above, the computer system 210 may include at least one computer-readable medium or memory for storing instructions programmed according to the embodiments described herein and for containing data structures, tables, records, or other data described herein. The term computer-readable medium as used herein refers to any non-temporary tangible medium that participates in providing instructions to the processor 220 for execution. Computer-readable media can take many forms, including but not limited to non-volatile media, volatile media, and transmission media. Non-limiting examples of non-volatile media include optical disks, solid-state drives, magnetic disks, and magneto-optical disks, such as hard disks 241 or removable media drives 242. Non-limiting examples of volatile media include dynamic memory, such as system memory 230. Non-limiting examples of transmission media include coaxial cables, copper wires, and optical fibers, including wires that constitute bus 221. Transmission media can also take the form of sound waves or light waves, such as those generated during radio wave and infrared data communications.

[0060] The computing environment 200 may also include a computer system 210 that operates in a networked environment using logical connections to the local computing device 106 and one or more other devices, such as a personal computer (laptop or desktop), a mobile device (e.g., a patient mobile device), a server, a router, a network PC, a peer device, or other public network node, and typically includes many or all of the elements described above with respect to the computer system 210. When used in a networked environment, the computer system 210 may include a modem 272 for establishing communications over a network 120, such as the Internet. The modem 272 may be connected to the system bus 221 via a network interface 270 or via another appropriate mechanism.

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

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

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

[0064] Storage device 306 includes fixed or removable storage devices, such as hard disk drives, solid-state drives, optical disks, or flash drives. Input device 308 includes, but is not limited to, a keyboard, a keypad, a touch screen, a touchpad, a detector, a microphone, an accelerometer, a gyroscope, a biometric scanner, or a network connector (e.g., a wireless LAN card for transmitting and / or receiving wireless IEEE 802 signals). Output device 310 includes, but is not limited to, a display, a speaker, a printer, a tactile feedback device, one or more lights, an antenna, or a network connection (e.g., a wireless LAN card for transmitting and / or receiving wireless IEEE 802 signals).

[0065] The input driver 312 communicates with the processor 302 and the input device 308 and allows the processor 302 to receive input from the input device 308. The output driver 314 communicates with the processor 302 and the output device 310 and allows the processor 302 to send output to the output device 310. It is noted that the input driver 312 and the output driver 314 are optional components, and if the input driver 312 and the output driver 314 are not present, the device 300 will operate in the same manner. The output driver 316 includes an accelerated processing device ("APD") 316 coupled to the display device 318. The APD accepts computational commands and graphics rendering commands from the processor 302, processes those computational and graphics rendering commands, and provides pixel outputs to the display device 318 for display. As described in further detail below, the APD 316 includes one or more parallel processing units to perform calculations according to a single instruction multiple data ("SIMD") paradigm. Thus, while various functions are described herein as being performed by or in conjunction with APD 316, in various alternatives, the functions described as being performed by APD 316 are additionally or alternatively performed by other computing devices having similar capabilities that are not driven by a host processor (e.g., processor 302) and that provide graphical output to display device 318. For example, it is contemplated that any processing system that performs processing tasks according to the SIMD paradigm can perform the functions described herein. Alternatively, it is contemplated that computing systems that do not perform processing tasks according to the SIMD paradigm can perform the functions described herein.

[0066] Figure 4 Shows the combination Figure 3 4. A graphical depiction of an artificial intelligence system 200 of an exemplary apparatus of FIG. 400 includes data 410, a machine 420, a model 430, a plurality of results 440, and underlying hardware 450. The system 400 operates in the following manner: the machine 420 is trained using data 410, while the model 430 is constructed to enable prediction of a plurality of results 440. The system 400 may operate relative to the hardware 450. In such a configuration, the data 410 may be associated with the hardware 450 and may originate from, for example, the device 102. For example, the data 410 may be data being generated, or output data associated with the hardware 450. The machine 420 may operate or be associated with it as a controller or data collection associated with the hardware 450. The model 430 may be configured to model the operation of the hardware 450 and to model the data 410 collected from the hardware 450 in order to predict the results achieved by the hardware 450. Using the predicted results 440, the hardware 450 may be configured to provide a certain expected result 440 from the hardware 450.

[0067] Figure 5 Shown in Figure 4Method 500 is performed in an artificial intelligence system of FIG. Method 500 includes collecting data from hardware at step 510. The data may include data currently collected from the hardware, historical data from the hardware, or other data. For example, the data may include measurements during a surgical procedure and may be associated with the results of the procedure. For example, the temperature of the heart may be collected and associated with the results of a cardiac procedure.

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

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

[0070] At step 540, method 500 includes predicting an outcome of a model associated with the hardware. Such prediction of an outcome may be based on a trained model. For example, with respect to the heart, if a temperature between 97.7-100.2 during a procedure produces a positive outcome by a procedure, then an outcome may be predicted in a given procedure based on the temperature of the heart during the procedure. While this model is rudimentary, it is provided for exemplary purposes and to increase understanding of the present invention.

[0071] The systems and methods of the present invention are used to train machines, build models, and predict results using algorithms. These algorithms can be used to solve trained models and predict results associated with hardware. These algorithms can generally be divided into classification, regression, and clustering algorithms.

[0072] For example, classification algorithms are used to classify dependent variables (which are variables being predicted) into multiple categories and predict the category (dependent variable) given the input. Thus, classification algorithms are used to predict outcomes from a fixed set of predefined outcomes. Classification algorithms may include Naive Bayes, Decision Trees, Random Forest Classifiers, Logistic Regression, Support Vector Machines, and k Nearest Neighbors.

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

[0074] Figure 6An example of a probability calculated by Naive Bayes is shown. The probabilistic approach to Bayes' theorem essentially means that the algorithm has a set of prior probabilities for each class of the target, rather than jumping directly into the data. After inputting the data, the Naive Bayes algorithm can update the prior probabilities to form the posterior probabilities. This is given by the following formula:

[0075]

[0076] This Naive Bayes algorithm, and Bayesian algorithms in general, can be useful when you need to predict whether your input belongs to a given list of n categories. Probabilistic methods can be used because the probabilities of all n categories will be fairly low.

[0077] For example, Figure 6 As shown, a person plays golf, which depends on factors including the external weather shown in the first data set 610. The first data set 610 shows the weather in the first column and the golfing results associated with the weather in the second column. In the frequency table 620, the frequency of certain events occurring is generated. In the frequency table 620, the frequency of a person playing golf or not playing golf in each weather condition is determined. From this, a likelihood table for generating initial probabilities is compiled. For example, the probability that the weather is cloudy is 0.29, while the general probability of playing golf is 0.64.

[0078] Posterior probabilities can be generated based on likelihood table 630. These posterior probabilities can be configured to answer questions about weather conditions and whether to play golf in those weather conditions. For example, the probability of it being sunny outside and playing golf can be explained by the Bayesian formula:

[0079] P(yes|sunny)=P(sunny|yes)*P(yes) / P(sunny)

[0080] According to likelihood table 630:

[0081] P(sunny|yes)=3 / 9=0.33,

[0082] P(clear) = 5 / 14 = 0.36,

[0083] P(yes)=9 / 14=0.64.

[0084] Therefore, P(yes|sunny) = 0.33*0.64 / 0.36 or approximately 0.60 (60%).

[0085] Generally speaking, a decision tree is a tree structure similar to a flowchart, where each external node represents a test on an attribute and each branch represents the result of that test. The leaf nodes contain the actual predicted labels. The decision tree starts at the root, where attribute values ​​are compared until a leaf node is reached. Decision trees can be used as classifiers when dealing with high-dimensional data and when little time has been spent after data preparation. Decision trees can take the form of simple decision trees, linear decision trees, algebraic decision trees, deterministic decision trees, random decision trees, non-deterministic decision trees, and quantum decision trees. Below is a list of examples of decision trees that can be used to classify data. Figure 7 An exemplary decision tree is provided in .

[0086] Figure 7 A decision tree for deciding whether to play golf is shown, which follows the same structure as the Bayesian example above. In the decision tree, the first node 710 checks the weather, giving sunny 712, cloudy 714, and rainy 716 as options to progress down the decision tree. If the weather is sunny, then follow the legs of the tree to the second node 720 that checks the temperature. In this example, the temperature at node 720 can be high 722 or normal 724. If the temperature at node 720 is high 722, a prediction of "no" (do not play) 723 golf appears. If the temperature at node 720 is normal 724, a prediction of "yes" (play) 725 golf appears.

[0087] Furthermore, starting from the first node 710, the result cloudy 714 appears, then "yes" (play) 715 golf.

[0088] Starting from the first node weather 710, the result of rainy day 716 leads to a third node of (again) checking the temperature 730. If the temperature at the third node 730 is normal 732, then "yes" (play) 733 golf. If the temperature at the third node 730 is low 734, then "no" (do not play) 735 golf.

[0089] According to the decision tree, if the weather is cloudy 715, sunny with normal temperature 725, and rainy with normal temperature 733, the golfer plays golf, and if it is sunny with high temperature 723 or rainy with low temperature 735, the golfer does not play golf.

[0090] A random forest classifier is a committee of decision trees where each decision tree has been fed a subset of the attributes of the data and makes predictions based on that subset. The mode of the actual predicted values ​​of the decision trees is considered to provide the final random forest answer. A random forest classifier generally mitigates overfitting present in individual decision trees, resulting in a more robust and accurate classifier.

[0091] Figure 8An exemplary random forest classifier for classifying the color of clothing is shown. Figure 8 As shown, the random forest classifier includes five decision trees 8101, 8102, 8103, 8104, and 8105 (collectively or generally referred to as decision trees 810). Each tree is designed to classify the color of clothing. A discussion of each tree and the decisions made is not provided because each individual tree is generally used as a Figure 7 In this example, three of the five trees (8101, 8102, 8104) determine that the garment is blue, while one tree determines that the garment is green (8103), and the remaining tree determines that the garment is red (8105). The random forest takes these actual predictions of the five trees and calculates the mode of these actual predictions to provide the random forest answer that the garment is blue.

[0092] Logistic regression is another algorithm used for binary classification tasks. Logistic regression is based on the logistic function (also known as the sigmoid function). This sigmoid curve can take any real-valued number and map it between 0 and 1, asymptotically approaching those limits. Logistic models can be used to model the probability of a certain class or event such as pass / fail, win / lose, alive / dead, or healthy / sick existing. This can be extended to modeling several classes of events, such as determining whether an image contains a cat, dog, lion, etc. Each object detected in the image will be assigned a probability between 0 and 1, and the sum of the probabilities is 1.

[0093] In a logistic model, the log odds (log of the odds) of a value labeled "1" is a linear combination of one or more independent variables ("predictors"); these independent variables can each be a binary variable (two categories, encoded by an indicator variable) or a continuous variable (any real value). The corresponding probability of a value labeled "1" can vary between 0 (definitely the value "0") and 1 (definitely the value "1"), so it is labeled; the function that converts log odds to probabilities is the logistic function, hence this name. The unit of measurement for the log odds scale is called the logit, from the logistic unit, hence this alternative name. Similar models with different sigmoid functions instead of logistic functions, such as the probability model, can also be used; the defining characteristic of the logistic model is that increasing one of the independent variables multiplicatively scales the odds of a given outcome at a constant rate, where each independent variable has its own parameter; for a binary dependent variable, this summarizes the odds ratio.

[0094] In a binary logistic regression model, the dependent variable has two levels (categorically). Outputs with more than two values ​​are modeled by multinomial logistic regression, and if the multiple categories are ordered, by ordinal logistic regression (e.g., proportional odds ordinal logistic model). The logistic regression model itself simply models the probability of the output given the input, and does not perform statistical classification (it is not a classifier), but it can be used to act as a classifier, for example, by selecting a cutoff value and classifying inputs with probabilities greater than the cutoff value into one category and inputs with probabilities less than the cutoff value into another category; this is a common way to make a binary classifier.

[0095] Fig. 9 An exemplary logistic regression is shown. This exemplary logistic regression enables prediction of outcomes based on a set of variables. For example, based on an individual's grade point average, the outcome of being accepted to a school can be predicted. The past history of grade point average and the relationship to acceptance enables prediction to occur. Fig. 9 The logistic regression enables the analysis of the GPA variable 920 to predict an outcome 910 defined by 0 to 1. At the low end 930 of the S-curve, the GPA 920 predicts a non-accepted outcome 910. While at the high end 940 of the S-curve, the GPA 920 predicts an accepted outcome 910. Logistic regression can be used to predict home values, customer life span values ​​in the insurance industry, and the like.

[0096] Support Vector Machines (SVM) can be used to classify data with a margin between two classes that are as far apart as possible. This is called the maximum margin. SVM can take the support vectors into account when drawing the hyperplane, unlike linear regression, which uses the entire data set for this purpose.

[0097] Fig.10 An exemplary support vector machine is shown. In the exemplary SVM 1000, data can be classified into two different categories represented as squares 1010 and triangles 1020. The SVM 1000 operates by drawing a random hyperplane 1030. The hyperplane 1030 is monitored by comparing the distance between the hyperplane 1030 and the nearest data point 1050 from each category (shown by line 1040). The data points 1050 closest to the hyperplane 1030 are called support vectors. The hyperplane 1030 is drawn based on these support vectors 1050, and the best hyperplane has the maximum distance from each of these support vectors 1050. The distance between the hyperplane 1030 and the support vector 1050 is called the margin.

[0098] The SVM 1000 may be used to classify data by using the hyperplane 1030 such that the distance between the hyperplane 1030 and the support vector 1050 is maximized. For example, such an SVM 1000 may be used to predict heart disease.

[0099] K nearest neighbors (KNN) refers to a group of algorithms that generally make no assumptions about the underlying data distribution and perform a relatively short training phase. Generally speaking, KNN uses many data points divided into several categories to predict the classification of a new sample point. Operationally, KNN specifies an integer N with a new sample. The N entries closest to the new sample in the system's model are selected. The most common classification of these entries is determined and the classification is assigned to the new sample. KNN generally requires storage space that increases as the training set increases. This also means that the estimation time increases proportionally with the number of training points.

[0100] In a regression algorithm, the output is a continuous quantity, so a regression algorithm can be used when the target variable is a continuous variable. Linear regression is a general example of a regression algorithm. Linear regression can be used to estimate the true quality (house cost, number of auctions, all buyout transactions, etc.) based on one or more consistent variables. The connection between the variables and the results is created by fitting a best line (thus fitting a linear regression). This best fit line is called the regression line and is expressed by the direct condition Y=a*X+b. Linear regression is best used in methods involving low dimensional numbers.

[0101] Fig.11 An exemplary linear regression model is shown. In this model, a predictor variable 1110 is modeled relative to a measured variable 1120. Clusters of instances of the predictor variable 1110 and the measured variable 1120 are plotted as data points 1130. The data points 1130 are then fitted with a best fit line 1140. The best fit line 1140 is then used for subsequent predictions given the measured variable 1120, which is used to predict the predictor variable 1110 for that instance. Linear regression can be used for modeling and forecasting in financial portfolios, income forecasting, real estate, and transportation to estimated arrival times.

[0102] Clustering algorithms can also be used to model and train datasets. In clustering, inputs are assigned into two or more clusters based on feature similarities. Clustering algorithms typically learn patterns and useful insights from data without any guidance. For example, clustering viewers into similar groups based on their interests, age, geography, etc. can be performed using unsupervised learning algorithms such as K-means clustering.

[0103] K-means clustering is often considered a simple unsupervised learning method. In K-means clustering, similar data points can be grouped together and bound in the form of clusters. One method for binding data points together is by calculating the centroid of the group of data points. When determining valid clusters, in K-means clustering, the distance between each point and the centroid of the cluster is evaluated. Based on the distance between the data point and the centroid, the data is assigned to the nearest cluster. The goal of clustering is to determine the intrinsic groupings in a set of unlabeled data. The "K" in K-means represents the number of clusters formed. The number of clusters (basically the number of categories that can be classified into for new data instances) can be determined by the user. For example, this determination can be performed using feedback during training and looking at the size of the cluster.

[0104] K-means is mainly used in cases where the dataset has distinct and well-spaced points, otherwise, the modeling may give inaccurate clusters if the clusters are not spaced. Also, K-means can be avoided in cases where the dataset contains a lot of outliers or the dataset is non-linear.

[0105] Fig.12 K-means clustering is shown. In K-means clustering, data points are plotted and assigned K values. For example, for Fig.12 , K=2, the data points are plotted, as depicted at 1210. The points are then assigned to similar centers at step 1220. The cluster centroids are identified, as depicted at 1230. Once the centroids are identified, the points are reassigned to clusters to provide the minimum distance between the data points to the respective cluster centroids, as depicted at 1240. The new centroids of the clusters can then be determined, as depicted at 1250. As the data points are reassigned to the clusters, new cluster centroid formation, iteration, or a series of iterations can occur to minimize the size of the clusters and determine the centroid of the best centroid. Then, when a new data point is measured, the new data point can be compared to the centroid and the cluster to be identified with the cluster.

[0106] Ensemble learning algorithms can be used. These algorithms use multiple learning algorithms to achieve better predictive performance than can be achieved from any of the component learning algorithms alone. Ensemble learning algorithms perform the task of searching through a hypothesis space to find suitable hypotheses that will make good predictions for a particular problem. Even if the hypothesis space contains hypotheses that are very suitable for a particular problem, it can be very difficult to find a good hypothesis. Ensemble algorithms combine multiple hypotheses to form a better hypothesis. The term ensemble is usually reserved for methods that use the same base learner to generate multiple hypotheses. The broader term multiple classifier system also covers mixtures of hypotheses that are not induced by the same base learner.

[0107] Evaluating the predictions of an ensemble generally requires more computation than evaluating the predictions of a single model, so ensembles can be thought of as a way to compensate for a poor learning algorithm by performing a lot of extra computation. Fast algorithms such as decision trees are often used in ensemble methods, such as random forests, although slower algorithms can also benefit from ensemble techniques.

[0108] Ensembles are themselves supervised learning algorithms, in that they can be trained and then used to make predictions. Thus, a trained ensemble represents a single hypothesis. However, that hypothesis is not necessarily accommodated within the hypothesis space of the model from which it is constructed. Thus, it can be shown that ensembles have greater flexibility in the functions they can represent. In theory, this flexibility can enable these ensembles to fit the training data better than a single model, but in practice, some ensemble techniques (particularly bagging) tend to reduce the problems associated with overfitting the training data.

[0109] As a rule of thumb, ensemble algorithms tend to produce better results when there is significant diversity between models. Therefore, many ensemble methods attempt to promote diversity between the models they combine. Although non-intuitive, it is possible to use more random algorithms (such as randomized decision trees) to produce stronger ensembles than very deliberate algorithms (such as entropy-reduced decision trees). However, using multiple brute-force learning algorithms has been shown to be more effective than using techniques that attempt to discard models to promote diversity.

[0110] The number of component classifiers in the ensemble has a large impact on the accuracy of the predictions. Predetermining the ensemble size and volume for large data streams as well as speed makes this even more important for online ensemble classifiers. The theoretical framework suggests that there is an ideal number of component classifiers for the ensemble such that having more or less than this number will reduce the accuracy. The theoretical framework suggests that using the same number of independent component classifiers as class labels gives the highest accuracy.

[0111] Some common types of ensembles include Bayesian best classifiers, bootstrap aggregation (bagging), boosting, Bayesian model averaging, Bayesian model combination, model banking, and stacking. Fig.13 An exemplary ensemble learning algorithm is shown where bagging is performed in parallel 1310 and boosting is performed sequentially 1320.

[0112] A neural network is a network or circuit of neurons, or in the modern sense, an artificial neural network consisting of artificial neurons or nodes. The connections of biological neurons are modeled as weights. Positive weights reflect excitatory connections, while negative values ​​represent inhibitory connections. The inputs are modified by the weights and summed using linear combinations. An activation function controls the amplitude of the output. For example, the acceptable range of outputs is usually between 0 and 1, or the range can be between -1 and 1.

[0113] These artificial networks can be used for predictive modeling, adaptive control and applications, and can be trained via data sets. Self-learning from experience can occur within the network, which can draw conclusions from complex and seemingly unrelated sets of information.

[0114] For the sake of completeness, biological neural networks consist of one or more groups of chemically connected or functionally related neurons. A single neuron may be connected to many other neurons, and the total number of neurons and connections in a network can be extensive. Connections (called synapses) are usually formed from axons to dendrites, but dendritic synapses and other connections are also possible. In addition to electrical signals, there are other forms of signals caused by the diffusion of neurotransmitters.

[0115] Artificial intelligence, cognitive modeling, and neural networks are information processing paradigms induced by the way biological nervous systems process data. Artificial intelligence and cognitive modeling attempt to simulate some of the properties of biological neural networks. In the field of artificial intelligence, artificial neural networks have been successfully applied to speech recognition, image analysis, and adaptive control in order to build software agents or autonomous robots (in computer and video games).

[0116] A neural network (NN) is a group of interconnected natural or artificial neurons that use mathematical or computational models for information processing based on computational connection methods, in terms of artificial neurons known as artificial neural networks (ANN) or simulated neural networks (SNN). In most cases, ANNs are adaptive systems that change their structure based on external or internal information flowing through the network. A more practical term neural network is a nonlinear statistical data modeling or decision making tool. These terms neural networks can be used to model complex relationships between inputs and outputs or to find patterns in data.

[0117] Artificial neural networks involve networks of simple processing elements (artificial neurons) that can exhibit complex global behaviors determined by the connections between the processing elements and the element parameters.

[0118] A classic type of artificial neural network is the recurrent Hopfield network. The utility of artificial neural network models is that they can be used to infer a function from observations and also to use that function. Unsupervised neural networks can also be used to learn representations of inputs that capture salient features of the input distribution, and more recently deep learning algorithms can implicitly learn distribution functions of observed data. Learning in neural networks is particularly useful in applications where the complexity of the data or task makes it impractical to manually design such functions.

[0119] Neural networks can be used in different fields. The tasks to which artificial neural networks are applied tend to fall into the following broad categories: function approximation or regression analysis, including time series forecasting and modeling; classification, including pattern and sequence recognition, novelty detection, and sequential decision making; and data processing, including filtering, clustering, blind signal separation, and compression.

[0120] Application areas for ANNs include nonlinear system identification and control (vehicle control, process control), game playing and decision making (backgammon, chat, tournaments), pattern recognition (radar systems, facial recognition, object recognition), sequence recognition (gesture, speech, handwritten text recognition), medical diagnostics, financial applications, data mining (or knowledge discovery in databases, "KDD"), visualization, and email spam filtering. For example, a semantic feature map of user interests can be created from pictures trained for object recognition.

[0121] Fig.14 An exemplary neural network is shown. In the neural network, there is an input layer represented by multiple inputs such as 14101 and 14102. The inputs 14101, 14102 are provided to a hidden layer depicted as including nodes 14201, 14202, 14203, 14204. These nodes 14201, 14202, 14203, 14204 are combined to produce output 1430 in the output layer. The neural network performs simple processing via the hidden layer (nodes 14201, 14202, 14203, 14204) of simple processing elements, which can exhibit complex global behaviors determined by the connections between the processing elements and element parameters.

[0122] Fig.14 Neural networks can be implemented in hardware. Fig.15 , a hardware-based neural network is shown.

[0123] Arrhythmias, specifically atrial fibrillation, have always been common and dangerous medical conditions, especially in the elderly population. For patients with normal sinus rhythm, the heart, which is composed of atria, ventricles, and excitatory conduction tissue, beats in a synchronous, patterned manner under the action of electrical stimulation. For patients with arrhythmias, abnormal areas of cardiac tissue do not follow the synchronous beating cycle associated with normal conduction tissue like patients with normal sinus rhythm. Instead, abnormal areas of cardiac tissue conduct abnormally to adjacent tissues, disrupting the cardiac cycle into an asynchronous rhythm. It was previously known that such abnormal conduction occurs in various regions of the heart, such as in the sinoatrial (SA) node region, in the conduction pathways along the atrioventricular (AV) node and the bundle of His, or in the myocardial tissue that forms the walls of the ventricular and atrial heart chambers.

[0124] Arrhythmias, including atrial arrhythmias, may be of the multi-wavelet reentrant type, characterized by multiple asynchronous loops of electrical impulses dispersed around the atrial chambers and generally self-propagating. Alternatively, or in addition to the multi-wavelet reentrant type, arrhythmias may also have a focal origin, such as when an isolated area of ​​tissue within the atrium beats autonomously in a rapid, repetitive manner. Ventricular tachycardia (V-tach or VT) is a tachycardia or rapid heart rhythm originating from one of the ventricles. This is a potentially life-threatening arrhythmia as it can lead to ventricular fibrillation and sudden death.

[0125] Atrial fibrillation, a type of arrhythmia, occurs when the normal electrical impulses produced by the sinus node are overwhelmed by chaotic electrical impulses originating in the atria and pulmonary veins, which cause irregular impulses to be transmitted to the ventricles. Irregular heartbeats result and can last from minutes to weeks, or even years. Atrial fibrillation (AF) is usually a chronic condition that slightly increases the risk of death, usually from stroke. The risk increases with age. About 8% of people over the age of 80 have some degree of AF. Atrial fibrillation is usually asymptomatic and generally not life-threatening on its own, but it can cause palpitations, weakness, fainting, chest pain, and congestive heart failure. The risk of stroke increases during AF because blood can pool and form clots in the poorly contracting atria and left atrial appendage. The first line of treatment for AF is medication that can slow the heart rate or restore the heart rhythm to normal. In addition, people with AF are often given anticoagulants to prevent them from being at risk for stroke. The use of such anticoagulants comes with its own risk of internal bleeding. For some patients, drug therapy is not enough and their AF is considered medically refractory, meaning it cannot be cured with standard medical interventions. Synchronized electrical cardioversion can also be used to restore AF to a normal rhythm. Alternatively, AF patients are treated with catheter ablation.

[0126] Therapies based on catheter ablation can include mapping the electrical properties of cardiac tissue (particularly the endocardium and cardiac volumes), and selectively ablating cardiac tissue by applying energy. Cardiac mapping, such as creating a map of electrical potentials propagating along the cardiac tissue (voltage map) or a map of arrival times to various tissue locations (local activation time (LAT) map), can be used to detect localized cardiac tissue dysfunction. Ablation, such as those based on cardiac mapping, can stop or modify the propagation of unwanted electrical signals from one part of the heart to another.

[0127] The ablation process damages unwanted electrical pathways by forming non-conductive lesions. A variety of energy delivery modalities have been disclosed for forming lesions, and include the use of microwaves, lasers, and more commonly radiofrequency energy to form conduction blocks along the walls of cardiac tissue. In a two-step procedure of mapping followed by ablation, electrical activity at various points in the heart is sensed and measured, typically by inserting a catheter containing one or more electrical sensors (or electrodes) into the heart and acquiring data at multiple points. This data is then used to select a target area of ​​the endocardium to be ablated.

[0128] Cardiac ablation and other cardiac electrophysiology procedures are becoming increasingly complex as clinicians treat increasingly challenging conditions such as atrial fibrillation and ventricular tachycardia. Treatment of complex arrhythmias currently relies on the use of three-dimensional (3D) mapping systems to reconstruct the anatomy of the cardiac chamber of interest.

[0129] For example, cardiologists rely on software such as Biosense Webster, Inc. (Diamond Bar, Calif.) The 3D Mapping System's Complex Fractionated Atrial Electrogram (CFAE) module analyzes intracardiac EGM signals and identifies ablation sites for the treatment of a wide range of cardiac conditions, including atypical atrial flutter and ventricular tachycardia.

[0130] 3D maps provide multiple pieces of information about the electrophysiological properties of the tissue, representing the anatomical and functional substrates of these challenging arrhythmias.

[0131] Cardiomyopathies of varying etiologies (hypoxic, dilated (DCM), hypertrophic cardiomyopathy (HCM), arrhythmogenic right ventricular dysplasia (ARVD), left ventricular noncompaction (LVNC), etc.) have an identifiable matrix characterized by areas of unhealthy tissue surrounded by functional cardiomyocytes.

[0132] FIG. 16A to FIG. 16D Examples of cardiomyopathies with different etiologies are shown. As a first example, Fig.16A and Fig. 16B An exemplary rendering of a heart 1600 with post-ischemic ventricular tachycardia (VT) is shown, characterized by an endo-epicardial low or intermediate voltage region 1602 where signal conduction is slowed. This shows that measuring any prolonged electrical potential within or around dense scar regions can help identify potential isthmuses that sustain VT. Fig.16A Post-ischemic VT is shown to be characterized by endo-epicardial low or intermediate voltage areas where signal conduction is slowed. This shows that measuring any prolonged electrical potential within or around dense scar areas can help identify potential isthmuses that sustain VT. Fig.16AThe bipolar signal amplitude (Bi) variations in various sectors of the heart 1600 are shown. Fig.16A A Bi range of 0.5 mV to 1.5 mV is shown. Fig. 16B The Shortex Composite Interval (SCI) variations in various sectors of the heart are shown. For example, the SCI ranges from 15.0 ms to 171.00 ms, with the SCI range of interest being between 80 ms and 170 ms.

[0133] Fig. 16C and Fig.16D An exemplary rendering of a heart 1610 experiencing left ventricular noncompaction cardiomyopathy is shown. More specifically, Fig. 16C shows an epicardial voltage map, and Fig.16D A potential duration map (PDM) is shown. Fig. 16C and Fig.16D In FIG. 16 , the three black circles in 1612 are marked as abnormally prolonged potentials, such as potentials above 200 milliseconds.

[0134] Abnormal tissue is usually characterized by low-voltage EGMs. However, initial clinical experience in endocardial-epicardial mapping indicates that low-voltage areas are not always present as the only arrhythmogenic mechanism in such patients. In fact, areas of low or medium voltage can exhibit EGM fragmentation and prolonged activity during sinus rhythm, which corresponds to the key isthmus identified during sustained and organized ventricular arrhythmias, for example, only for non-tolerant ventricular tachycardia. In addition, in many cases, EGM fragmentation and prolonged activity are observed in areas that show normal or near-normal voltage amplitudes (>1-1.5mV). Although the latter areas can be evaluated based on voltage amplitude, they cannot be considered normal based on the intracardiac signal and therefore represent the true arrhythmogenic substrate. 3D mapping is able to locate the arrhythmogenic substrate on the endocardial and / or epicardial layers of the right / left ventricle, which can vary in distribution depending on the extension of the primary disease.

[0135] The substrate associated with these cardiac disorders is associated with the presence of fragmented and elongated EGMs in the endocardial and / or epicardial layers of the ventricular chambers (right and left). 3D mapping systems, such as 3. Ability to locate potential arrhythmogenic substrates of cardiomyopathy in terms of abnormal EGM detection.

[0136] Electrode catheters have been commonly used in medical practice for many years. They are used to stimulate and map electrical activity in the heart, as well as to ablate sites of abnormal electrical activity. When used, the electrode catheter is inserted into a major vein or artery, such as the femoral artery, and then guided into the heart chamber of interest. A typical ablation procedure involves inserting a catheter having at least one electrode at its distal end into a heart chamber. A reference electrode, usually taped to the patient's skin, is provided, or a second catheter disposed in or near the heart can be used to provide the reference electrode. RF (radio frequency) current is applied to the tip electrode of the ablation catheter, and the current flows to the reference electrode through the surrounding medium (i.e., blood and tissue). The distribution of the current depends on the amount of contact between the electrode surface and the tissue compared to the blood, which has a higher conductivity than the tissue. Due to the resistance of the tissue, heating of the tissue occurs. The tissue is heated sufficiently to cause cell destruction in the heart tissue, resulting in the formation of non-conductive ablation lesions in the heart tissue. In this process, heating of the electrode also occurs due to conduction from the heated tissue to the electrode itself. If the electrode temperature becomes high enough, perhaps above 60°C, a thin transparent coating of dehydrated blood proteins may form on the surface of the electrode. If the temperature continues to rise, this dehydrated layer may become increasingly thicker, causing blood coagulation on the electrode surface. Because the dehydrated biomaterial has a higher electrical resistance than the endocardial tissue, the impedance to the flow of electrical energy into the tissue also increases. If the impedance increases sufficiently, impedance rise occurs, and the catheter must be removed from the body and the tip electrode cleaned.

[0137] Fig.17 is an illustration of an exemplary system 1720 that may implement one or more features of the presently disclosed subject matter. All or part of the system 1720 may be used to collect information for a training data set, and / or all or part of the system 1720 may be used to implement a trained model. The system 1720 may include a component, such as a catheter 1740, configured to damage a tissue region of an organ in the body. The catheter 1740 may also be further configured to obtain biometric data. Although the catheter 1740 is shown as a pointed catheter, it should be understood that any shape of catheter including one or more elements (e.g., electrodes) may be used to implement the embodiments disclosed herein. The system 1720 includes a probe 1721 having an axis that can be navigated by a physician 1730 to a body part of a patient 1728 lying on a table 1729, such as a heart 1726. According to an embodiment, multiple probes may be provided, however, for the sake of brevity, a single probe 1721 is described herein, but it should be understood that the probe 1721 may represent multiple probes. As shown in FIG. Fig.17As shown, a physician 1730 may insert the shaft 1722 through the sheath 1723 while manipulating the distal end of the shaft 1722 using a manipulator 1732 near the proximal end of the catheter 1740 and / or deflecting from the sheath 1723. As shown in inset 1725, the catheter 1740 may be mounted at the distal end of the shaft 1722. The catheter 1740 may be inserted through the sheath 1723 in a collapsed state and may then be deployed within the heart 1726. As further disclosed herein, the catheter 1740 may include at least one ablation electrode 1747 and a catheter needle 1748.

[0138] According to an exemplary embodiment, the catheter 1740 may be configured to ablate a tissue region of a chamber of the heart 1726. Inset 1745 shows the catheter 1740 in a magnified view within the chamber of the heart 1726. As shown, the catheter 1740 may include at least one ablation electrode 1747 coupled to the body of the catheter. According to other exemplary embodiments, multiple elements may be connected via a strip that forms the shape of the catheter 1740. One or more other elements (not shown) may be provided, which may be any element configured to ablate or obtain biometric data, and may be an electrode, a transducer, or one or more other elements.

[0139] According to embodiments disclosed herein, ablation electrodes such as electrode 1747 can be configured to provide energy to a tissue region of an internal organ such as heart 1726. The energy can be thermal energy and can cause damage to the tissue region starting from the surface of the tissue region and extending into the thickness of the tissue region.

[0140] According to exemplary embodiments disclosed herein, biometric data may include one or more of LAT, electrical activity, topology, bipolar mapping, dominant frequency, impedance, and the like. Local excitation time may be a time point corresponding to a threshold activity of local excitation calculated based on a normalized initial starting point. Electrical activity may be any applicable electrical signal that can be measured based on one or more thresholds and can be sensed and / or enhanced based on a signal-to-noise ratio and / or other filters. Topology may correspond to a physical structure of a body part or a portion of a body part, and may correspond to changes in the physical structure relative to different portions of a body part or relative to different body parts. Dominant frequency may be a frequency or range of frequencies that is prevalent at a portion of a body part, and may be different 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 may be a resistance measurement at a given area of ​​a body part.

[0141] like Fig.17As shown, the probe 1721 and the catheter 1740 may be connected to a console 1724. The console 1724 may include a processor 1741 (such as a general purpose computer) having a suitable front end and interface circuit 1738 for transmitting signals to and receiving signals from the catheter, as well as for controlling other components of the system 1720. In some embodiments, the processor 1741 may be further configured to receive biometric data, such as electrical activity, and determine whether a given tissue area is conductive. According to one embodiment, the processor may be located outside the console 1724, and may be located, for example, in a catheter, in an external device, in a mobile device, in a cloud-based device, or may be an independent processor.

[0142] As noted above, processor 1741 may include a general purpose computer that may be programmed with software to perform the functions described herein. The software may be downloaded to the general purpose computer in electronic form, such as over a network, or may alternatively or additionally be provided and / or stored on a non-transitory tangible medium, such as magnetic, optical, or electronic storage. Fig.17 The exemplary configuration shown can be modified to implement the embodiments disclosed herein. The embodiments disclosed in the present invention can be applied similarly using other system components and settings. In addition, the system 1620 can include additional components, such as elements for sensing electrical activity, wired or wireless connectors, processing and display devices, etc.

[0143] According to one embodiment, the display connected to the processor (e.g., processor 1741) can be located at a remote location such as at a separate hospital or within a separate healthcare provider network. In addition, system 1720 can be part of a surgical system configured to obtain anatomical measurements and electrical measurements of a patient's organ (such as the heart) and to perform a cardiac ablation procedure. An example of such a surgical system is sold by Biosense Webster. system.

[0144] System 1720 may also and optionally use ultrasound, computed tomography (CT), magnetic resonance imaging (MRI), or other medical imaging techniques known in the art to obtain biometric data, such as anatomical measurements of the patient's heart. System 1720 may use a catheter, electrocardiogram (EKG), or other sensor that measures the electrical properties of the heart to obtain electrical measurements. Fig.17 As shown, the biometric data, including the anatomical measurements and the electrical measurements, may then be stored in a memory 1742 of the mapping system 1720. The biometric data may be transmitted from the memory 1742 to the processor 1741. Alternatively or in addition, the biometric data may be transmitted using a network 1662 to a server 1760, which may be local or remote.

[0145] The network 1762 can be any network or system known in the art, such as an intranet, a local area network (LAN), a wide area network (WAN), a metropolitan area network (MAN), a direct connection or a series of connections, a cellular telephone network, or any other network or medium capable of facilitating communication between the mapping system 1720 and the server 1760. The network 1662 can be wired, wireless, or a combination thereof. A wired connection can be achieved using Ethernet, a universal serial bus (USB), RJ-11, or any other wired connection known in the art. A wireless connection can be achieved using Wi-Fi, WiMAX and Bluetooth, infrared, cellular networks, satellite, or any other wireless connection method known in the art. In addition, several networks can work alone or communicate with each other to facilitate communication in the network 1762.

[0146] In some cases, server 1762 may be implemented as a physical server. In other cases, server 1762 may be implemented as a public cloud computing provider (e.g., Amazon Web Services ) virtual server.

[0147] The console 1724 may be connected to body surface electrodes 1743 via cables 1739, which may include adhesive skin patches attached to the patient 1730. The processor, in conjunction with the current tracking module, may determine the position coordinates of the catheter 1740 within the patient's body part (e.g., heart 1726). The position coordinates may be based on the impedance or electromagnetic field measured between the body surface electrodes 1743 and the electrodes 1748 or other electromagnetic components of the catheter 1740. Additionally or alternatively, a positioning pad may be located on the surface of the bed 1729 and may be detachable from the bed 1729.

[0148] Processor 1741 may include real-time noise reduction circuitry, typically 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. Processor 1741 may pass signals from the A / D ECG or EMG circuit to another processor and / or may be programmed to perform one or more functions disclosed herein.

[0149] The console 1724 may also include an input / output (I / O) communication interface that enables the console to transmit signals from and / or to the electrodes 1747 .

[0150] During the procedure, the processor 1741 may facilitate the presentation of a body part rendering 1735 to the physician 1730 on the display 1727 and store data representing the body part rendering 1735 in the memory 1742. The memory 1742 may include any suitable volatile and / or non-volatile memory, such as a random access memory or a hard drive. In some embodiments, the medical professional 1730 may be able to manipulate the body part rendering 1735 using one or more input devices (such as a touch pad, a mouse, a keyboard, a gesture recognition device, etc.). For example, the input device may be used to change the position of the catheter 1740 so that the rendering 1735 is updated. In an alternative embodiment, the display 1727 may include a touch screen that may be configured to accept input from the medical professional 1730 in addition to presenting the body part rendering 1735.

[0151] As described herein, during the triangulation process, the acquired data indicates where the activation source point of the focal tachycardia location may be and which points are most informative for validating and finding the source point in the most effective data. The algorithm learns from previous maps, anatomy, and mapping sequences in order to recommend the best course of action at each point.

[0152] Fig.18 1800 is an exemplary illustration of triangulation 1800 according to an embodiment. A prediction of the LV prediction curve is calculated. The LV prediction curve includes curves 1810, 1820, 1830. The intersection 1840 (circled) of the curves 1810, 1820, 1830 predicts the wavefront source point 1850. The wavefront propagation is graphically represented by the progression from the earliest prediction to the latest prediction. The site of successful ablation is considered to be the true wavefront source point 1850. As shown, the wavefront source point 1850 is located within the intersection circle 1840.

[0153] In cases where the physician assumes focality (single source not a fibrillatory arrhythmia), triangulation 1800 uses several mapping points (in this illustration, curves 1810, 1820, 1830) to locate the focal arrhythmia source (wavefront source point 1850). This mapping results in a shortened procedure time for such cases. The algorithm receives a FAM grid and several LAT points. By guessing the propagation of the LAT on the FAM, the intersection 1840 of several propagation curves (in this illustration, curves 1810, 1820, 1830) is used to estimate the source of the propagation (wavefront source point 1850). Disadvantages of triangulation 1800 include that the physician has little knowledge of where to obtain the informative LAT, and does not take into account geometric / electrical noise or tissue conduction velocity differences.

[0154] Fig.19is an exemplary flow chart of an exemplary method 1900 for improving mapping efficiency by suggesting mapping point locations according to an embodiment. In step 1910, data is received at a machine / system for machine learning (ML). The data includes multiple signals received during the performance of triangulation to locate a focal tachycardia. The received data is used to generate a predictive model for the location of the focal tachycardia (step 1920). The predictive model is based on additional data received by the machine (step 1930).

[0155] When trying to determine focal propagation to determine a lesion for ablation, some effects in the myocardium are stable and some are unstable. Therefore, local activation points (LAT) are used to estimate the location of the lesion point in the triangle. Data is, for example, anatomical (FAM / CT) data and LAT points acquired by, for example, a Carto machine or via CartoNet. In addition, ablation locations indicated by lesion / termination by a physician or coherent mapping can be used.

[0156] For grid representation, a grid CNN or a dense volume network (3DCNN) can be used. These points can be given as a list, or as an RNN, or as a volume sample (where each voxel represents the volume of one or more LATs that are also 3D convolutions). Use a dense coding layer to combine these two modes to provide a regression output of the next sample point and an estimate of the lesion point. At each stage, the input data is described by adding a single LAT point (position and activation) to the input, and the system / network output is the position of the next best point sampled on the anatomical structure, which can be expressed as a spatial coordinate and a current estimate of the lesion point. The system model will receive the input LAT point by point, and will minimize the cost of the item containing the minimum number of input points and a good estimate of the lesion point (by self-estimation and by using the original triangulation algorithm or coherent or doctor ablation points). In addition, the convolution filter weights, fully connected layers, and GRU layers are all trainable. The output of the model represents a suggestion for the next point to collect LAT and estimate the lesion point. Since there may be statistical similarities between patients, similarities can be learned and used to further provide more accurate predictions for physicians. Additionally, arrhythmias may differ based on the specific disease, which may also be used for prediction.

[0157] Despite advances in EP technology, localizing the source of focal arrhythmias (e.g., VT, focal atrial tachycardia) in the ventricles or atria can be time consuming, and activation mapping continues to have inherent limitations. Fig.19The following system and architecture use the position and activation timing information in two pairs of sampling points to predict the source point of the focal wavefront. The system can be incorporated into an electroanatomical mapping (EAM) system to evaluate its accuracy in a three-dimensional clinical environment. The system described below includes an ML-based algorithm that integrates the knowledge of pre-recorded cases obtained to provide guidance to physicians in real time. The system can overcome the above shortcomings and provide a faster and more robust solution. The system takes advantage of the situation where the mapping map is dense and the lesion point is easy to estimate, so that the system can operate on a portion of the input, and use this data to estimate which unvisited area is the most likely area for finding the source of the lesion. The source of the lesion can be known by accessing the complete data.

[0158] A triangular mesh defines a 3D surface in space. Formally, it is a space v€R 3 A surface is a set of vertices V and faces F in , where each face is a triple of vertices (f: (v1, v2, v3). A map on a mesh is a function that assigns values ​​(real, complex, vector) to faces or vertices.

[0159] The anatomical activation map can be represented in a uniform manner. Assume that the input space is contained in 10 3 (cm 3 ) grid; each contains 2 3 (mm 3 ) volume. Each voxel may contain a feature vector describing the electrical activation signal measured inside that volume. The feature vector may be learned to represent the reference annotations and the LAT annotations. The anatomical surface (the boundary between the blood pool and the tissue) is represented by a symbolic Euclidean distance (also called a distance transform). For each voxel center, this measures the distance to the nearest point on the anatomical surface. If the voxel is outside the anatomical structure, it is positive, and if it is inside the anatomical structure, it is negative. A feature vector summarizing the anatomical structure may be learned.

[0160] Fig. 20 A graphical depiction of the system and information flow 2000 for a single query point is shown. The system 2000 provides anatomical structure and signal representations, learns a map between these representations, and assumes focal activation propagation of a single focus.

[0161] Inputs 2010 to the system 2000 may include recorded ECG signals and locations 2010.1, Carto annotations 2010.2a, and anatomical surfaces (eg, FAM, mFAM, CT) 2010.3. Input 2010 may include a FAM grid 2010.3, acquired LAT points (position in space (mm) 2010.2 and ECG (signal matrix described below) 2010.1, REF and LAT values ​​(milliseconds), estimated lesion point from the algorithm, and a series of ablation points that terminate the arrhythmia. This is used to verify the "true" lesion point. The LAT / coherence map (including the LAT map) is used to specify all chambers with activation time values ​​for each vertex, while the coherence map also indicates the conduction velocity vector and the non-conducting area (for each vertex). The lesion point can be estimated on a sufficiently densely mapped LAT / coherence map. The acquired points can all be on the FAM. Verification is possible if there are ablation points near the lesion that terminates the arrhythmia. ECG signal 2010.1, both the inner surface of the heart and the body surface can be used. The ECG signal 2010.1 can provide a signal for each point x in space The matrix of , sampled in time t (ECG sampling rate) for all recorded channels c. The input 2010 may include parameters about the patient, such as age, gender, physical dimensions (of body and / or heart and atria), medical history including medication use, and type of arrhythmia.

[0162] In order to make the system 2000 useful for the anatomy and activation propagation of the patient, the system 2000 can learn to extract uniform, compact and meaningful information about the anatomy and ECG signals. According to an embodiment, two networks can perform this operation in an unsupervised manner. This information can be in the latent space R k (where K is fixed and relatively small), which scales the input from its original large vector dimension to a more compact vector dimension and reconstructs the input using only the latent representation. The latent space can be tuned by capturing important features by itself or by additional tasks and constraints. Denoising / Variational Autoencoders, Deep Belief Networks, and Generative Adversarial Networks (GANs) are examples of networks that can perform this task.

[0163] The recorded ECG signals and positions 2010.1 and Carto annotations can be combined in the first NN 2020. The input may also include other inputs 2010. The transformer network 2020 can be used to encode the ECG signal into a feature vector. The transformer is a prior art of sequence-to-sequence networks. Each ECG signal set 2010.1 is collected at some positions in the space and continues for a fixed time interval. This produces a signal matrix for each acquisition point. The transformer takes the input sequence and tries to predict the next (or several) time steps of the given sequence. This is done in an unsupervised manner by only looking at the training ECG signal set 2010.1, which can be taken from any case of the same chamber. The task of annotating reference annotations and LAT can be performed as an additional training option to fine-tune the potential space. An encoder-decoder network can be used to perform interpretable transformations. Semantic autoencoders for zero-shot learning can also be used. The transformer itself is constructed by an encoder and a decoder, so the potential vector is defined as the input of the encoder.

[0164] Fig. 22 Shown is the architecture of a transformer network 2200 for use in system 2000. On the left side of network 2200 is an encoder, and on the right side of network 2200 is a decoder.

[0165] The input 2220 of the transformer 2200 may be a sequence of tokens, where a token is a vector that selects a member of a discrete set. To tokenize the ECG signal 2230, one or more discrete quantities may be used to represent or estimate (amount or series) the measured signal. For example, the measured voltage in the ECG signal may be in the interval [0, V MAX ] are measured within, and these signals can be represented by a set of discrete values. In one embodiment, 768 discrete values ​​are used. When the signals are represented discretely, for each ECG signal, the nearest value from the discrete space is selected. In another embodiment, the discrete space can be divided on a logarithmic scale.

[0166] Each input has now been converted into a vector, with one token in the correct "token" and zero otherwise ("one hot encoding"). Multiple of these vectors 2240, such as n = an amount covering approximately 300ms to 400ms based on the estimated tachycardia time, can be used to teach the network to predict the value of the next sample. According to one embodiment, temporal resampling of the vectors can be performed to improve efficiency. The value can be predicted by using softmax 2210 on the 768-dimensional output probability vector. The decoder receives the shifted encoder output 2250, allowing regression to be applied to multiple samples 2260, learning to predict the entire sequence. The network internal representation is the input embedded, transformed and operated via the encoder's attention layer and projection layer. The final representation size can be selected by the specific implementation. The transformation of the entire sequence in n blocks (with or without overlap) can be performed to produce the final representation vector of the signal 2270 by cascading the block vectors.

[0167] The second NN 2030 can operate on the anatomical surface information 2010.3 and other inputs 2010. The anatomical surface can be represented using a Vnet network. The network can be a fully convolutional network with residual connections. The network accepts an input volume of a predefined size (50×50×50 in our case) and returns an output volume of the same size. The first part performs 3D convolutions in each layer, followed by non-linearity (RELU) and downsampling (max pooling). Each 3D convolution has a filter of size 50×50×50×C, where C is the number of channels. The weights of the first three dimensions represent spatial locations and are shared, while C is the current dimension of the input or feature data. The inner part (the vector of size 512 of the network in the figure) represents the addition (via cascading) of the latent space of each residual link. The latent vector of the cascade is A n €R d , where d is the dimension of the new potential space (512 plus each residual connection). The left part performs upsampling via transposed convolution, and then performs additional convolution and RELU for each level until the initial volume size is reached. Both input voxels and output voxels represent the distance transformation (minimum distance) to the anatomical surface. The network is prevented from learning the identity function by using random inactivation (Dropout), weight regularization, and optionally masking the different connection areas of the input. In addition, the higher layers of the residual connection may be disconnected for further regularization. Data can be collected from the case of having CT or high-resolution FAM mapping.

[0168] Fig.23An example Vnet network architecture 2300 is shown. As shown, the Vnet architecture 2300 may include 323, but may be adapted to other resolutions. The Vnet architecture 2300 shows an example of a Vnet network, where actual numbers in each layer such as the initial volume size and the number of filters are optimized. Since we can have several acquisition points in the same voxel, we only take the average potential vector to represent the voxel.

[0169] return Fig. 20 , the output of the first NN 2020 includes the spatial ECG feature variations 2040 as described. The output of the second NN 2030 includes the spatial shape representation 2050 as described.

[0170] The spatial ECG feature variations 2040 and the spatial shape representation 2050 may be combined in a third NN 2060 along with the query points 2070 to produce an output. The output may include a vector 2080 to source and a confidence level 2090.

[0171] The vector to source 2080 is a vector pointing in the direction of the source (on the surface geometry), and the confidence level 2090 represents the confidence of the vector to source 2080 for the current acquisition point. The system 2000 can also output one or more points that can be sampled by the physician to improve the estimation of the lesion point and the estimation of the lesion point using both the classical triangulation algorithm and the network.

[0172] The vector direction 2080 to the source (close to the local inverse gradient). The confidence level 2090 measured by the local consistency of the signal representation and the amount of meaningful data. By using the network to sample points on the surface of the anatomical structure, the output can include points in space indicating the next most informative area to sample. These points can be from areas with high confidence values ​​2090 and points in space that approximate new source points using the intersection of the vector with the source or using a classical triangulation algorithm.

[0173] System 2000 can be combined with the following data stream. Currently known signals and anatomical structures can be input 2010. Input 2010 can be converted into feature structures 2040, 2050 via one or more NNs (shown as NN 2020, 2030). System 2000 can operate in NN 2060 on all voxels containing anatomical surface and no adjacent ECG signal, and query the estimated confidence level of each voxel. Region is a defined area of ​​adjacent voxels. Region 2080 with the highest average confidence 2090 can be selected as the next location of the suggestion of query 2070. System 2000 can also output the current estimation of the lesion point using the classical triangulation algorithm.

[0174] Fig.24An integrated network architecture 2400 is shown for use as the NN 2060. The network 2400 may receive an encoded representation of the ECG 2040 and anatomical structure 2050, along with a query point 2070, and return a source vector 2080 and a confidence level 2090. Fig.23 The network described in encodes the ECG signal in each voxel into a vector. An additional network can be used to convert this set of vectors into a single vector representation summarizing all spatial information. This can be performed using the left part of Vnet2410, instead of using the volume 50×50×50×K (where the first 3 dimensions are spatial and K is the latent representation) to convert it into a latent vector E€R u The output of the Vnet half 2410 may be combined with the anatomical structure representation network 2420, and the query point location 2070 (which may be in spatial coordinates or voxel indices) is input to a fully connected multilayer perceptron 2430. The multilayer perceptron 2430 may be a network that performs a nonlinearity after performing several matrix multiplications on the input. The output of the network 2430 is R 3 The source vector in , which points to the lesion source 2080 and the confidence level c€[0,1]2090.

[0175] Alternatively, the system 2000 may use weighted majority voting to provide an estimate using the intersection of the source vectors for all currently known points. Fig.21 As shown, there is a depiction 2100 of inferring the next region of interest using the system 2000. Similar to Fig. 20 , the recorded ECG signal and position 2010.1 and the anatomical surface 2010.3 may be provided as input into the system 2000. The system 2000 may query each vertex of the anatomical surface that does not have information depicted as a heart 2110. The system 2000 may output the next region to obtain data 2120 and a vector 2080 to the source.

[0176] The unsupervised encoding networks 2020, 2030, 2060 can be trained on a large number of cases. A network can be created for each chamber, since each chamber differs in anatomical structure and ECG signal phenomena. The second step in training the integrated network is to consider the LAT / coherence / triangulation case, where only the activations of a portion of the chamber are used as input to the system at a time. Then, the source vector 2080 and confidence 2090 are calculated for each of the remaining points on the anatomical structure. This can occur multiple times in each case, requiring a different initial set of points each time.

[0177] The confidence score 2090 is calculated as a function of the following considerations: how close the point is to the actual source, how diverse the vectors to the source are in the neighborhood of the point - the lower the better, how diverse the potential representations of the ECG in the neighborhood of the point are - the lower the better as long as there are enough measurements, and how much the estimate of the lesion point is improved (by using the current input point) and by adding the point - the main considerations.

[0178] The exact function that combines this can vary. Results with scores between 0 (no improvement in the point) and 1 (excellent take point) may be beneficial.

[0179] As briefly described above, data from previous patient cases is used to train the system 2000. For each such case, the available data may include the patient id, data used as system input, data used to calculate the desired output. The "gold standard", i.e., the data used to calculate the desired output, may include a surface mesh from a FAM or other model and a LAT / coherence map. A LAT / coherence map with sufficient acquisition points (determined experimentally) may be selected as a training input. The earliest activation point may be indicated as an input lesion point. Cases where lesion points can be verified by ablation data may be identified in cases where they have a higher impact (via weighting) on ​​the training process.

[0180] A few points (At least 3) sets can be sampled as starting points. Each point is associated with a series of ECG signals in some radial neighborhood of the point. A new point can then be selected, and the vector to the source, confidence, and new lesion estimate can be calculated for the additional point. Repeat this process for all points to determine the point P* that best improves the lesion estimate. This process can generate training data containing pairs (S, p*). New S sets can be identified, and the process is repeated multiple times for each given situation. Conduction velocity vectors (coherence) can be used to provide additional clues to the input points, which should not deviate too far from the inverse vector to the source.

[0181] Based on, for example, CARTO (R) The calculated "ablation index" is used to weight each case for training ranking. Points can be compromised for ablation points with low procedure quality (e.g., point instability, insufficient ablation depth, or not using enough power), large area ablation, and multiple ablation sites. The data set of a specific patient can be weighted according to the survival duration of the outcome of the procedure. For example, cases with arrhythmia recurrence 3 months after the procedure may have a lower weight than cases with arrhythmia recurring only 12 months after the procedure, while cases without ablation verification receive the lowest weight. The quality of the estimated lesion points and the mapping density from different point sets can also be included.

[0182] The goal of this process is to allow the algorithm to be trained more on cases where the lesions are clearly and clinically verified and the map quality is high.

[0183] Typically, the dataset should be divided into training, validation, and test sets. The first two sets will be used during system development, while the test set is only used to evaluate the accuracy of the system. Additionally, cross-validation can be used to improve performance.

[0184] Patient parameters, such as age, gender, and type of arrhythmia, can affect the results. A separate model can be trained based on some of these parameters, and these parameters can be introduced into some of the layers in the NN architecture to help learn differences based on these parameters.

[0185] The proposed method combines information in an unsupervised manner, the encoding of signals and anatomical structures is completely unsupervised, while for activation maps, pre-existing maps can be exploited.

[0186] Although the network is relatively simple, the same knowledge representation approach may be useful for other queries and more complex arrhythmias. The integrated framework combines the information available in our standard Carto cases including ECG, maps, and activation maps.

[0187] In some embodiments, a complete CT / MRI does not exist at the beginning of the case. The NN may be able to use the acquired catheter positions to estimate the anatomical shape during the traversal. For example, Vnet can achieve this estimation by replacing the input.

[0188] Even after the system is ready and deployed in a hospital, additional data can be accumulated. This additional data can be added to the training data set, and the system can be retrained to continually improve the accuracy of the system. Specifically, data from additional operations can provide feedback by considering the success rating of ablations performed according to the system's recommendations.

[0189] Although the features and elements are described above in detail, it will be appreciated by those of ordinary skill in the art that each feature or element may be used alone or in any combination with other features and elements. In addition, the methods described herein may be implemented in a computer program, software, or firmware incorporated in a computer-readable medium for execution by a computer or processor. Examples of computer-readable media include electronic signals (transmitted via a wired or wireless connection) and computer-readable storage media. Examples of computer-readable storage media include, but are not limited to, read-only memory (ROM), random access memory (RAM), registers, cache memory, semiconductor memory devices, magnetic media (such as internal hard disks and removable disks), magneto-optical media, and optical media (such as CD-ROM disks and digital versatile disks (DVDs)). A processor associated with software may be used to implement a radio frequency transceiver used in a WTRU, UE, terminal, base station, RNC, or any host computer.

[0190] ***

Claims

1. A system for locating a lesion, the system comprising: Multiple inputs; a first converter that converts at least a first portion of the plurality of inputs into a spatial ECG feature vector; a second transformer that transforms at least a second portion of the plurality of inputs into a spatial shape representation; A neural network operates on the spatial ECG feature vector and the spatial shape representation to produce a plurality of outputs.

2. The system of claim 1, wherein the input comprises at least one or more of: a FAM grid, LAT points, an ECG signal, an estimated lesion point, and a series of ablation points.

3. The system of claim 1, wherein the output comprises a vector to a source and a confidence level. The system of claim 1 , wherein the output comprises a next region of interest.

5. The system of claim 1, wherein the first transformer is a first neural network.

6. The system of claim 5, wherein the first neural network is a transformer network.

7. The system of claim 1, wherein the second transformer is a second neural network.

8. The system of claim 7, wherein the second neural network is a Vnet network.

9. The system of claim 1, wherein the neural network inputs a query point.

10. The system of claim 1, wherein at least one of the plurality of outputs is a confidence level c€[0,1].

11. The system of claim 1, wherein at least one of the plurality of outputs is R 3 The source vector pointing to the lesion source.

Images