Neural network intracardiac EGM annotation

Abstract

A method for medical diagnostics, the method comprising: receiving electrophysiology data, the electrophysiology data comprising a first intracardiac electrogram (IEGM) signal acquired by a first pair of electrodes having a first intracardiac electrode. The data includes a second IEGM signal acquired by a second pair of electrodes having a second intracardiac electrode proximate to the first intracardiac electrode. And inputting the first IEGM signal and the second IEGM signal into a first convolutional layer of a neural network and a second convolutional layer parallel to the first convolutional layer in the neural network to generate a corresponding first intermediate result and a corresponding second intermediate result. And inputting the first intermediate result and the second intermediate result into one or more common layers of the neural network. The method includes, in response to inputting the first intermediate result and the second intermediate result, receiving, from an output layer of the neural network, indications of local activation times at locations of the first and second intracardiac electrodes.

Description

Technical Field

[0001] The present disclosure relates generally to electrocardiograms (EGMs), and in particular to annotation of intracardiac EGM signals. Background Art

[0002] Annotations of EGM signals are labels that point to specific locations within the signal. Methods for annotating EGM signals are known in the art, some of which are referenced below.

[0003] U.S. Patent Application No. 2021 / 0386355 to Ravuna et al. describes a method for detecting mapping annotations of an electrophysiology (EP) mapping system. The method includes using a processor having a machine learning algorithm configured to receive a first heartbeat at an identified cardiac spatial location.

[0004] U.S. Patent Application No. 2022 / 0005198 to Goldberg et al. describes a method provided by an ablation neighbor engine executed by a processor. The ablation neighbor engine is said to utilize manual annotations of training data.

[0005] Chinese patent application CN111096735A by Ou Feng et al. describes an electrocardiogram analysis system capable of iterative updating. The system is said to have a back-end processing center, which may include an annotation data expansion module.

[0006] European application EP3865060A1 by Amaury et al. describes a computer-implemented method for generating an annotated photoplethysmographic (ppg) signal. The method is said to include recording an ECG signal and annotating segments of the ECG signal algorithmically or expert-based.

[0007] PCT application WO2020 / 214094A1 by Bhograj et al. describes a system for facilitating data processing of physiological information. The system is said to include a device operable to collect a data set of physiological information from a person, and a server operable to receive the data set from the device and classify the data set into at least one preliminary group of medical conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The foregoing and other features and advantages of the invention will be apparent from the following more particular description of preferred embodiments of the invention, as illustrated in the accompanying drawings.

[0009] Figure 1 is a schematic illustration of an electrocardiogram (EGM) signal acquisition system used in cardiac diagnostics;

[0010] Figure 2is a schematic block diagram illustrating the overall structure and operation of an artificial neural network;

[0011] Figure 3 is a schematic diagram of an example of a layer group in an artificial neural network;

[0012] Figure 4 is a schematic diagram of an alternative example of the layer group;

[0013] Figure 5 is a schematic diagram of an example of another layer group;

[0014] Figure 6 yes Figure 5 A schematic diagram of an alternative example of the layer group;

[0015] Figure 7 is a flowchart of the algorithm used to generate the sample corpus; and

[0016] Figure 8 It is used for training Figure 2 Flowchart of the artificial neural network algorithm. DETAILED DESCRIPTION

[0017] Overview

[0018] In order to map the flow of electrical waves as they pass through a patient's beating heart, a physician may insert a catheter with electrodes into the heart during a diagnostic investigation of the heart. As the waves pass through the heart, they cause expansion and contraction of the area they pass through, causing the heart to beat. The inserted electrodes detect the passage of the waves as an intracardiac electrocardiogram voltage versus time signal, and the point in time that the waves pass through a given area of ​​the heart, called the local activation time (LAT), is a parameter that physicians can use when diagnosing heart conditions.

[0019] Typically, the catheter has two electrodes, and the voltage mentioned above may be the voltage between the two electrodes, referred to herein as a bipolar voltage or signal. Additionally, the patient may have electrodes, such as patch electrodes attached to the patient's skin, which may be used as a reference, or a WCT (Wilson Center Terminal) reference may be used. Using one of these references, the voltage mentioned above may be the voltage measured relative to the reference, referred to herein as a unipolar voltage or signal. It should be understood that there is a unipolar voltage for each of the two electrodes.

[0020] A recognized method for automatically finding the LAT of a region is to use a wavefront annotation algorithm, such as described in U.S. Pat. No. 11,058,342 to Botzer et al. The algorithm determines the LAT by finding the time point at which the unipolar voltage signal has its steepest drop. This time, which can be adjusted using the bipolar signal and other parameters of the signal, is typically set to the LAT. Other methods for automatically finding the LAT of a region are also known in the art.

[0021] However, automatic determination of LAT is not completely straightforward. For example, the analyzed signal may have more than one dip, there may be differences in the shape of the signal for different subjects (which the algorithm used may not be able to account for), and / or there may be differences in the shape of the signal for different regions of one subject. Errors caused by incorrect LAT determination may be corrected manually, usually by the physician performing the diagnostic investigation, and ignoring the LAT value automatically assigned based on his / her experience.

[0022] Catheters currently used for diagnosis may have many electrodes, often more than a hundred, each of which generates at least a monopolar signal and often also a bipolar signal. For these catheters, manual calibration of the large number of signals becomes cumbersome and expensive, as it relies on the physician's expertise.

[0023] The examples of the present disclosure improve upon the prior art methods of finding the LAT by using a trained artificial neural network with multiple layer groups in parallel to analyze the signals collected by a pair of electrodes. In one example, there are four layer groups in parallel, each dedicated to analyzing two monopolar signals, a bipolar signal, and a combination of monopolar and bipolar signals. The results of each of these layer groups are combined in the network to give an overall estimate of the LAT of the signal.

[0024] For training, the network requires a very large number (on the order of one million or more) of signals to form a corpus of training data. Although in theory the corpus for training could be collected from healthy patients, in practice most of the available data comes from unhealthy patients. The data is available from unhealthy patients because they have already undergone diagnostic investigations; healthy patients do not typically undergo diagnostic investigations. The problem with using data from unhealthy patients is that the automatically discovered LAT may be caused by unhealthy areas in the heart (e.g., patient pathology such as areas of fibrosis) or by errors in the operation of the algorithm that determines the LAT.

[0025] Examples of the present disclosure provide a method for generating a corpus of the required large amount of data. Intracardiac EGM (IEGM) signals are collected from healthy animals (usually pigs). An algorithm such as the wave annotation algorithm mentioned above is used to automatically calculate the LAT of each signal, and the resulting data can be presented on a cardiac map, colored according to the time of the LAT. This method of presenting the LAT is well known in the field of cardiology.

[0026] Because the collected data is from healthy animals, there is no LAT problem mentioned above, that is, the observed signal may come from an unhealthy heart area. Therefore, because in healthy animals, the LAT value changes in a smooth and gradual manner across the heart, the LAT color presented in the map should also change in a smooth and gradual manner, and any abnormalities in this gradual color change are easily visible. For example, in the green area of ​​the LAT map, there may be blue dots. Such points correspond to signals where the algorithm for calculating LAT has found an abnormal LAT that is different from the LAT values ​​around it. However, because the signal comes from a healthy animal, the different colors must be due to algorithmic errors.

[0027] The examiner of the map can correct the abnormal LAT of the blue dot by simply selecting the dot, where the processor can automatically reassign the LAT of the dot to a value corresponding to the surrounding LAT. Alternatively, the examiner can manually reassign the LAT. In addition, upon selection, the processor can save the signal and the reassigned value of the LAT, and the signal and its LAT can be added to the corpus of training data and can be used to train the network of the present disclosure. It should be understood that the examiner does not need the expertise of a physician to select the signal to be corrected.

[0028] System Description

[0029] Reference now Figure 1 , which is a schematic illustration of an electrocardiogram (EGM) signal acquisition system 20 used in cardiac diagnostics. In the illustrated example, a physician 22 is performing a diagnostic procedure using the system 20. The physician 22 is performing a procedure on a heart 52 of a subject 24 using a catheter probe 26, the distal end 28 of which includes a plurality of substantially similar electrodes 30 arranged along the length of the distal end. In the present disclosure, electrodes 30 are distinguished by appending suffixes A, B to the reference numerals as necessary.

[0030] The system 20 includes a processor 32, a memory 31, and an EGM module 46. The EGM module 46 is coupled to EGM electrodes 50 via cables 48, which are attached to the skin of the subject 24 to collect body surface EGM signals from the subject. The EGM module 46 is also configured to collect intracardiac EGM signals from the electrodes 30 on the distal end 28 and analyze these signals using an artificial neural network (ANN) 34 incorporated into the EGM module. The operation of the ANN 34 will be described in more detail below.

[0031] The processor 32 and the module 46 typically include a programmable processor that is programmed in software and / or firmware to perform the functions described herein. Alternatively or in addition, the processor may include hard-wired and / or programmable hardware logic circuits that perform at least some of these functions. Although the processor 32 and the module 46 are shown in the drawings, for simplicity, as separate monolithic functional blocks, in practice some of these functions may be combined in a single processing and control unit.

[0032] The processor 32 and the EGM module 46 typically reside within the console 40. The console 40 includes input devices 42, such as a keyboard and mouse, that are operated by the physician 22. A display screen 44 is positioned proximate to the console 40 and may be used to present a map 68 of the heart 52. The display screen 44 may optionally include a touch screen, thereby providing another input device.

[0033] The system 20 may include other modules for system operation that typically reside within the console 40. For example, the system may include a tracking module 60 that is coupled to one or more electromagnetic position sensors 61 in the distal end 28. In the presence of an external magnetic field generated by a magnetic field generator 62, the electromagnetic position sensors output signals that vary with the position of the sensors. Based on these signals, the tracking module 60 can determine the position of the electrode 30 in the heart 52.

[0034] Position tracking using an external magnetic field is performed in the CARTO TM The system is implemented in the present invention and is described in detail in U.S. Patent Nos. 5,391,199, 6,690,963, 6,484,118, 6,239,724, 6,618,612 and 6,332,089, PCT Patent Publication No. WO 96 / 05768, and U.S. Patent Application Publication Nos. 2002 / 0065455 Al, 2003 / 0120150 Al and 2004 / 0068178 Al.

[0035] Alternatively or in addition, module 60 may use a tracking system based on the current passed through or the resistance experienced by the electrode 30. In such a system, module 60 estimates the position of a given electrode 30 in response to the current or resistance between the given electrode and a plurality of surface electrodes 63 coupled to the skin of subject 24. The Advanced Current Localization (ACL) system manufactured by Biosense-Webster (Irvine, Calif.) described in U.S. Pat. No. 8,456,182 is one such tracking system.

[0036] An external electrode 65 or "return patch" may additionally be attached externally to subject 24 (typically on the skin of the subject's torso) to serve as a neutral or reference electrode. Alternatively, a reference electrode, such as a Wilson center terminal (WCT) electrode, may be formed by appropriately positioning electrode 50.

[0037] Figure 2 is a schematic block diagram illustrating the overall structure and operation of ANN 34. In the illustrated figure, after training, ANN 34 is configured to receive as input signals 36 unipolar and bipolar signals acquired by electrodes 30A and 30B, which are measured relative to each other and relative to a neutral reference electrode (by way of example, assumed to include patch 65, and also referred to herein as reference R). Electrodes 30A and 30B are typically separated from each other by about 1 mm.

[0038] The signal 36 is acquired in a preset window of interest (WOI) and includes a plurality of values. In the example described herein, there are 401 digitized elements of the signal, each element being acquired at a corresponding acquisition time point within the WOI. Typically, the signal time points are about 1 ms apart and have a "width" of about 1 ms. It will be understood that the number of elements and their spacing is by way of example, and other examples may have a greater or lesser number of digitized elements of the signal and corresponding spacings.

[0039] The ANN 34 is formed by layers of artificial neurons, and it is assumed hereinafter that each of the layers comprises a rectified linear unit (ReLU) neuron. However, the layers of the ANN 80A may be composed of other neurons, such as derivatives of ReLU neurons, hyperbolic tangent neurons, and / or sigmoid neurons, and one of ordinary skill in the art will be able to apply the present disclosure to layers having other such neurons with necessary changes.

[0040] In the first input layer 100, there are 401 neurons, corresponding to the number of elements in the input signal 36, i.e. the number of time points. The input layer 100 is followed by hidden layers 102, which are formed as a plurality of layer groups 80, 84, 88 and 92 in parallel, each layer group comprising at least one convolutional layer. Figure 1Prior to the process of , the network 34 including all layer groups has been trained, and the method of training each of these layer groups and the structure of each group are described below.

[0041] Specifically, group 80 receives a first unipolar signal measured between electrode 30A and reference R from input layer 100. 30A V R , group 84 receives from the input layer a second unipolar signal measured between electrode 30B and reference R 30B V R , group 88 receives the bipolar signal measured between electrode 30A and electrode 30B 30A V 30B , and group 92 receives the signal 30A V R , 30B V R , 30A V 30B Combination V c The output of each of these groups of layers is combined in a cascade layer 96 included within the hidden layer 102. The cascade layer 96 feeds into a fully connected output layer 144, which provides a final estimate of the LAT of the signals acquired from the electrodes 30A, 30B.

[0042] Figure 3 is a schematic diagram of an example of a group 80 (referred to herein as group 80A) configured to receive as input signals from an input layer 100 the signals collected by electrodes 30A. 30A V R As mentioned above, in the input layer 100, there are 401 neurons corresponding to the input signal 30A V R The number of elements in , that is, the number of time points. 30A V R The acquired signal can be considered as a 1-dimensional (1D) matrix in the form of a 1×n matrix, where n is a positive integer corresponding to the number of time points in the signal. In the example here, the matrix is ​​a 1×401 matrix.

[0043] Group 80A is included in hidden layer 102 and is assumed herein to include at least one convolutional layer. In the illustrated example, the first hidden layer of the group includes a 1D convolutional layer 104, which is composed of at least one filter or kernel, configured to perform convolution of the layer by scanning the values ​​derived from the input layer 100. Each kernel is a 1D filter, and the illustrated example depicts a layer 104 including a first 1D kernel 108 and a second 1D kernel 112, each kernel having five elements formed as a 1×5 matrix. Typically, each kernel of the layer may have fewer or more elements than five, and in one example, the kernel has three elements. In addition, although the illustration assumes that the convolutional layer has two kernels, there are typically more than two kernels.

[0044] The kernel in a convolutional layer, such as layer 104, is typically configured to filter or isolate features of the data being analyzed through its convolution operation. Typical features isolated by the kernel are regions that include negative slopes or regions that include high values. The kernel operates by sliding along the presented data set in a stepwise manner with a preset step size, and after each step forming a convolution of the data segment "covered" by the kernel. Typically, padding elements are added to the initial and terminal elements of the convolutional layer to ensure that the convolution of the kernel on these elements is correct.

[0045] The convolution layer 104 is typically followed by a 1D pooling layer 116 configured to receive the output of the layer 104. The pooling layer 116 in turn includes at least one 1D filter associated with the layer configured to reduce the data generated by each of the kernels of the convolution layer 104. By way of example, assume that the layer 116 has two 1D filters 120, 122, each with two elements, formed as a 1 x 2 matrix applied to the data received from the convolution layer 104. The filters of the pooling layer do not perform convolutions, but are configured to reduce the data from the layer 104, for example by taking the maximum value of the filtered data, in which case the pooling layer 116 is a max pooling layer.

[0046] Hidden layers 102 of group 80A may include additional convolutional layers followed by pooling layers, such as illustrated convolutional layer 124 followed by pooling layer 132. Convolutional layer 124 is structured and operates substantially as convolutional layer 104; pooling layer 132 is structured and operates substantially as pooling layer 116.

[0047] The data from the convolutional layers of the group 80A are transmitted to one or more fully connected layers within the hidden layer 102, which are assumed herein to include a fully connected output layer 140 of the group. The fully connected layer 140 is in turn connected to a fully connected cascade layer 96, so that the results 80AR (referred to herein as intermediate results) from the layer 140 are transmitted to the cascade layer. The output layer 140 typically includes the same number of neurons as the input layer 100, i.e., 401.

[0048] In the examples of the present disclosure, layer groups 84 and 88 have corresponding example groups 84A and 88A, which are substantially similar in structure to example 80A and operate in a substantially similar manner. As described herein, the three layer groups are trained using different input signals, i.e., using 30A V R To train group 80, use 30B V R To train group 84, and use 30A V 30B To train group 88. Thus, while the input signals for the three groups may have the same structure of a 1×401 matrix in response to training (and will also have the same LAT), the three groups typically have different weights associated with their neurons and kernels. Moreover, in response to training, there may be different numbers and sizes of layers and kernels in the three groups. Groups 84A and 88A have corresponding intermediate results 84AR and 88AR. It should be understood that the intermediate results 80AR, 84AR, 88AR are typically different.

[0049] By adding identifier 80A, 30A V R , 80AR, corresponding identifier in brackets 84A, 30B V R , 84AR and 88A, 30A V 30B , 88AR, Figure 3 The similarities in structure and operation of the three groups are illustrated in .

[0050] Figure 4 80B. The operation of group 80B is similar to that of group 80A ( Figure 3 ) are generally similar in operation, and elements indicated by the same reference numerals in the two groups are generally similar in construction and operation.

[0051] In contrast to group 80A, group 80B does not use a pooling layer; instead, data from a convolutional layer (herein assumed to include two such layers 104) is passed to a 1D flattening layer 150. As for group 80A, group 80B includes a fully connected layer 140, which receives data from the flattening layer 150, and which in turn is connected to a fully connected cascade layer 96, so that the intermediate results 80BR from layer 140 are transferred to the cascade layer.

[0052] In an alternative example of the present disclosure, groups 84 and 88 have corresponding example groups 84B and 88B, which are substantially similar in structure to example 80B and operate in a substantially similar manner, except for the differences mentioned above with respect to examples 80A, 84A, and 88A. Groups 84B and 88B have corresponding intermediate results 8BAR and 88BR.

[0053] By adding identifier 80B, 30A V R ,80BR, the corresponding identifier in brackets 84B, 30B V R , 84BR and 88B, 30A V 30B , 88BR, in Figure 4 The similarities in structure and operation of the three groups are illustrated in .

[0054] The above description of the three layer groups 80, 84 and 88 illustrates that their structures are substantially identical. While this is generally the case for layer group 92, there are differences between group 92 and the three groups 80, 84 and 88, as explained below.

[0055] Figure 5 92A. The operation of group 92A is similar to that of group 80A ( Figure 3 ) are generally similar in operation, and elements indicated by the same reference numerals in the two groups are generally similar in construction and operation.

[0056] However, in contrast to groups 80, 84, and 88 which have as their inputs signals in the form of 1×n matrices, group 92A is configured such that input layer 100 receives signals V in the form of a two-dimensional (2D) 3×n matrix. C , where n=401 in the example described herein. Signal V C As 30A V R , 30B V R as well as 30A V 30B Group 92A includes hidden layer 162, which is generally similar to hidden layer 102 and includes at least one convolutional layer. Signal Vc The first hidden layer of the group 92A is input, and in the illustrated example, the first hidden layer of the group comprises a 2D convolutional layer 164, which is composed of at least one filter or kernel configured to perform a convolution of this layer by scanning the values ​​derived from the input layer 100.

[0057] Each kernel is a 2D filter, and the illustrated example depicts layer 164 as including a first 2D kernel 168 and a second 2D kernel 172, each having five elements formed as a 3×5 matrix. In general, each kernel of the layer may have fewer or more elements as its width while having a depth of three, and in one example, the kernel has three elements. In addition, while the illustration assumes that the convolutional layer has two kernels, there are typically more than two kernels.

[0058] By way of example, the convolution layer 164 is followed by a 2D pooling layer 176 that is configured to receive the output of the layer 164. The pooling layer 176 includes at least one 2D filter associated with the layer that is configured to reduce the data generated by each of the kernels of the convolution layer 164. Each filter of the pooling layer has a depth of three for the associated convolution layer kernel. By way of example, assume that the layer 176 has two 2D filters 178, 182, each filter having two elements, formed as a 3×2 matrix applied to the data received from the convolution layer 164.

[0059] Hidden layer 162 may include additional 2D convolutional layers followed by 2D pooling layers, such as illustrated convolutional layer 180, which has a 2D kernel of depth three. As described, the convolutional layers are followed by pooling layers, such as illustrated pooling layer 188, which has a 2D filter of depth three. Convolutional layer 180 is structured and operates substantially as convolutional layer 164; pooling layer 188 is structured and operates substantially as pooling layer 176.

[0060] The data from the convolutional layer is transmitted to one or more fully connected layers within the hidden layer 162, which are assumed herein to include a fully connected output layer 196. The output layer 196 typically includes the same number of neurons as the input layer 100, i.e., 401, each of which is configured to accept inputs of depth three. The intermediate results 92AR from the output layer 196 are transmitted to the cascade layer 96.

[0061] Figure 6 92B. The operation of group 92B is similar to that of group 92A ( Figure 5 ) are generally similar in operation, and elements indicated by the same reference numerals in the two groups are generally similar in construction and operation.

[0062] Group 92B does not use a pooling layer; instead, data from a convolutional layer (herein assumed to include two such layers 164) is passed to a 1D flattening layer 204. As for group 92A, group 92B includes a fully connected output layer 196 that receives data from the flattening layer 204 and provides an intermediate result 92BR to the cascade layer 96.

[0063] Back to Figure 2 , the intermediate results from layer groups 80, 84, 88 and 92 are all input to the cascade layer 96, as described above. The cascade layer 96 feeds into a fully connected output layer 144, which has 401 neurons, each neuron corresponding to a time point, and the value provided by the neuron generally provides a probability that the time point corresponds to the LAT of the input signal. The output derived from the output layer can be expressed as a single hot value, and there is generally a single single hot value 1, which corresponds to the time point of the LAT assigned to the input signal by the ANN 34.

[0064] As mentioned above, in the Figure 1 Prior to the illustrated process, the sets 80, 84, 88, and 92 were trained using stored intracardiac EGM (IEGM) signals and corresponding values ​​of LAT associated with the signals. Figure 7 is a flow chart illustrating a process for generating such signals and LAT values, and the steps of the flow chart are described below.

[0065] Figure 7 is a diagram of a flow chart of an algorithm for generating a sample corpus that can be used to train the artificial neutral network of the present disclosure. A large number of sample IEGM signals are required in order for the training to be effective, and each signal requires the correct LAT. As described above, there is a problem in using most of the available signals because these signals are often from unhealthy patients and such data may have incorrect LAT values. The algorithm described below addresses this problem.

[0066] By way of example, suppose you use Figure 1 The algorithm is implemented using the elements illustrated in (e.g., processor 32, memory 31, probe 26, tracking module 60, and position sensor 61). However, it should be understood that the algorithm may be implemented using any other convenient processor, memory, probe, and elements for tracking the probe.

[0067] In an initial step 220, a healthy animal (usually a pig) is selected and a catheter probe 26 is inserted into the animal's heart so as to contact the wall of one of the heart chambers. The catheter is assumed to have two electrodes 30 contacting the wall and a neutral electrode such as a patch 65 is attached to the animal's skin.

[0068] In a mapping step 224, the catheter is moved within the cardiac chamber and the tracking module 60 processes the signals from the sensors 61 to generate a map of the cardiac chamber.

[0069] In an acquisition step 228, which may be implemented simultaneously with step 224, the processor acquires the monopolar IEGM signals from the two catheter electrodes measured relative to the neutral electrode and the bipolar signals between the electrodes. At the same time, the processor uses the tracking module 60 and the sensor 61 to determine the position of the electrodes. The signals and their electrode positions are stored in the memory. It should be understood that these signals correspond to the three types of signals mentioned above, namely, 30A V R , 30B V R and 30A V 30B .

[0070] In an automatic processing step 232, the processor implements an algorithm for automatically analyzing each signal and assigning a corresponding value of the LAT associated with each signal based on the analysis. One algorithm that may be implemented is the wavefront annotation algorithm described in U.S. Patent No. 11,058,342, but the processor may use any other suitable algorithm.

[0071] In a presentation step 236, the processor presents on display screen 44 a map of the cardiac chambers, such as map 68, overlaid with the LAT values ​​that have been calculated in step 232. The overlaid LAT values ​​are typically displayed as a spectrum of different colors; for example, a LAT of 100 ms may be displayed in red, and a LAT of 300 ms may be displayed in blue.

[0072] In a correction step 240, the displayed map is examined by an algorithm operator, who typically has much less expertise than the physician 22. Because the IEGM signal is collected from a healthy animal, the LAT of the animal's heart typically changes in a smooth, gradual manner. Therefore, the color of the LAT presented in the displayed map should also change in a smooth, gradual manner, and any deviation from such a gradual color change should be easily visible to the algorithm operator.

[0073] For example, there may be orange dots within the red area of ​​the map; such dots correspond to signals where the algorithm calculating LAT has found a LAT different from the LAT values ​​surrounding it. However, because the signals are from healthy animals, the different colors are anomalies that must be caused by algorithmic error.

[0074] The algorithm operator may select an outlier point. Once selected, correction of the LAT of the point may be applied by any convenient method, which may be manual or automatic. For example, if the screen 44 is configured as a touch screen, the operator may touch the outlier point and the processor 32 may be configured to automatically reallocate the LAT of the point to a value corresponding to the surrounding LAT. Alternatively, the operator may input a reallocated LAT value for the point. The processor may then save the signal generating the LAT and the reallocated LAT value.

[0075] In a storage step 244, once the algorithm operator has completed step 220, the operator may store all of the signals and their corresponding LAT values ​​for the map. It will be appreciated that the stored signals and LAT values ​​contain a corpus that includes those signals for which the operator reallocated LAT values ​​as well as signals for which the operator did not reallocate LATs (and their automatically found LAT values). It will be appreciated that, considering the set of all analyzed signals, the subset of signals that have reallocated LATs and the subset of signals that do not have reallocated LATs complement each other.

[0076] The stored signals may then be used in a training step 248 to train the ANN 34 .

[0077] Figure 8 34 is a flow chart of an algorithm for training ANN 34. Training is an iterative process in which the parameters of the network (such as the weights of the network neurons, the size and weights of the filters of the convolutional layers, and the number and type of different layers) are adjusted in order to optimize the output of the network. It is assumed that the training is performed using processor 32, but any other suitable processor may be used.

[0078] In an initial step 260, the Figure 7 The corpus of IEGM signals generated by the algorithm is input to the input layer 100. Due to the large number of signals in the corpus, the input of the signals is usually implemented as batch processing.

[0079] In an evaluation step 264, the processor records the LAT value calculated and output by the network, and in a cost calculation step 268, the processor uses the known LAT value of the input batch signal to calculate the cost relating the known LAT value to the LAT value output by the network.

[0080] Step 268 may use any cost function known in the art, such as a quadratic cost function or a cross entropy cost function.

[0081] In decision step 272, the processor determines whether the cost calculated in step 268 is low enough to be acceptable. If the cost is too high, i.e., the cost is unacceptable, so that the decision returns negative, then in adjustment step 276, the processor changes the parameters of the network using any suitable optimization algorithm (e.g., a gradient descent algorithm such as an Adam optimizer).

[0082] As described above, and as illustrated by arrow 280 , the training process is an iterative process such that the processor returns from step 276 to the initial step 260 .

[0083] If decision step 272 returns positive, ie the cost calculated in step 268 is sufficiently low, then in a save step 284 the processor saves the calculated network parameters and uses them in the ANN 34 .

[0084] As used herein, the term "about" or "approximately" for any numerical value or range of entities indicates a suitable dimensional tolerance that allows the entity to function as intended as described herein. More specifically, "about" or "approximately" may refer to a range of ±10% of the value of the recited value, for example, "about 80%" may refer to a range of values ​​from 72% to 88%.

[0085] Example

[0086] Embodiment 1. A method for medical diagnosis, comprising:

[0087] receiving electrophysiological data including a first intracardiac electrogram (IEGM) signal acquired by a first pair of electrodes including a first intracardiac electrode and a second IEGM signal acquired by a second pair of electrodes including a second intracardiac electrode proximal to the first intracardiac electrode;

[0088] Inputting the first IEGM signal and the second IEGM signal into a first convolutional layer of a neural network and a second convolutional layer in the neural network parallel to the first convolutional layer, respectively, to generate a corresponding first intermediate result and a second intermediate result;

[0089] inputting the first intermediate result and the second intermediate result together into one or more common layers of the neural network; and

[0090] In response to inputting the first intermediate result and the second intermediate result, an indication of local activation times at locations of the first intracardiac electrode and the second intracardiac electrode is received from an output layer of the neural network.

[0091] Embodiment 2. The method of embodiment 1, wherein the first pair of electrodes and the second pair of electrodes have a common neutral electrode.

[0092] Embodiment 3. The method of Embodiment 2, wherein at least one of the first signal and the second signal comprises a monopolar signal measured relative to the neutral electrode.

[0093] Embodiment 4. A method according to embodiment 2, wherein the first signal includes a combination of a first unipolar signal generated between the first intracardiac electrode and the neutral electrode, a second unipolar signal generated between the second intracardiac electrode and the neutral electrode, and a bipolar signal generated between the first intracardiac electrode and the second intracardiac electrode.

[0094] Embodiment 5. A method according to embodiment 1, wherein the first signal is in the form of a one-dimensional (1D) matrix, and the first convolutional layer is one-dimensional, wherein at least one 1D kernel is configured to slide along the first convolutional layer.

[0095] Embodiment 6. A method according to Embodiment 5, wherein the neural network includes at least one 1D convolutional layer after the first convolutional layer and before the one or more common layers.

[0096] Embodiment 7. A method according to Embodiment 5, wherein the neural network includes at least one 1D fully connected layer after the first convolutional layer and before the one or more common layers.

[0097] Embodiment 8. A method according to embodiment 1, wherein the neural network includes at least one flat layer after the first convolutional layer and before the one or more common layers.

[0098] Embodiment 9. A method according to embodiment 1, wherein the first signal is in the form of a two-dimensional (2D) matrix, and the first convolutional layer is two-dimensional, wherein at least one 2D kernel is configured to slide along the first convolutional layer.

[0099] Embodiment 10. The method of embodiment 9, wherein the neural network comprises at least one 2D convolutional layer after the first convolutional layer and before the one or more common layers.

[0100] Embodiment 11. A method according to Embodiment 9, wherein the neural network includes at least one 2D fully connected layer after the first convolutional layer and before the one or more common layers.

[0101] Embodiment 12. The method of Embodiment 1, wherein the one or more common layers include at least one fully connected layer.

[0102] Embodiment 13. A method comprising:

[0103] estimating the corresponding local activation time (LAT) of the first set of intracardiac electrocardiogram signals;

[0104] generating a three-dimensional (3D) map based on the estimated LAT;

[0105] receiving an indication of an abnormality in the 3D map;

[0106] generating training data using the anomaly and a region of the 3D map that does not include the anomaly;

[0107] Using the training data to train an artificial neural network; and

[0108] The artificial neural network is applied to estimate the LAT of a second set of intracardiac electrocardiogram signals.

[0109] Embodiment 14. The method according to embodiment 13, wherein the first set of intracardiac electrocardiogram signals are generated from a healthy animal.

[0110] Embodiment 15. The method of embodiment 13, wherein estimating the corresponding LATs of the first group is automatically performed by a computer processor.

[0111] Embodiment 16. The method of embodiment 13, wherein the indication is provided by manual inspection of the 3D map.

[0112] Embodiment 17. The method of embodiment 16, and comprising: in response to the manual inspection, a computer processor automatically reassigning a corresponding LAT to the signal corresponding to the anomaly.

[0113] Embodiment 18. The method of embodiment 16, and comprising: in response to the manual inspection, manually reassigning a corresponding LAT to the signal corresponding to the anomaly.

[0114] Embodiment 19. The method according to embodiment 13, wherein the training data includes the first set of signals, the LAT corresponding to the abnormality, and the LAT corresponding to the area of ​​the 3D map that does not include the abnormality.

[0115] Embodiment 20. A device for medical diagnosis, comprising:

[0116] a probe (26) comprising a first intracardiac electrode (30A) proximal to a second intracardiac electrode (30B);

[0117] A neural network (34), the neural network comprising:

[0118] A first convolutional layer (80) and a second convolutional layer (84), wherein the second convolutional layer is parallel to the first convolutional layer;

[0119] one or more common layers; and

[0120] Output layer; and

[0121] A processor (32), the processor being configured to:

[0122] receiving electrophysiological data, the electrophysiological data comprising a first intracardiac electrogram (IEGM) signal acquired by a first pair of electrodes including the first intracardiac electrode and a second IEGM signal acquired by a second pair of electrodes including the second intracardiac electrode;

[0123] Inputting the first IEGM signal and the second IEGM signal into the first convolution layer and the second convolution layer respectively to generate a corresponding first intermediate result and a second intermediate result;

[0124] inputting the first intermediate result and the second intermediate result together into the one or more common layers; and

[0125] In response to inputting the first intermediate result and the second intermediate result, an indication of local activation times at locations of the first intracardiac electrode and the second intracardiac electrode is received from the output layer.

[0126] Embodiment 21. The device of Embodiment 20, and comprising a neutral electrode common to the first pair of electrodes and the second pair of electrodes.

[0127] Embodiment 22. The apparatus of Embodiment 21, wherein at least one of the first signal and the second signal comprises a monopolar signal measured relative to the neutral electrode.

[0128] Example 23. An apparatus according to Example 21, wherein the first signal includes a combination of a first unipolar signal generated between the first intracardiac electrode and the neutral electrode, a second unipolar signal generated between the second intracardiac electrode and the neutral electrode, and a bipolar signal generated between the first intracardiac electrode and the second intracardiac electrode.

[0129] Embodiment 24. An apparatus according to embodiment 21, wherein the first signal is in the form of a one-dimensional (1D) matrix, and the first convolutional layer is one-dimensional, wherein at least one 1D kernel is configured to slide along the first convolutional layer.

[0130] Embodiment 25. An apparatus according to Embodiment 24, wherein the neural network includes at least one 1D convolutional layer after the first convolutional layer and before the one or more common layers.

[0131] Embodiment 26. An apparatus according to Embodiment 24, wherein the neural network includes at least one 1D fully connected layer after the first convolutional layer and before the one or more common layers.

[0132] Embodiment 27. An apparatus according to Embodiment 21, wherein the neural network includes at least one flat layer after the first convolutional layer and before the one or more common layers.

[0133] Embodiment 28. An apparatus according to embodiment 21, wherein the first signal is in the form of a two-dimensional (2D) matrix, and the first convolutional layer is two-dimensional, wherein at least one 2D kernel is configured to slide along the first convolutional layer.

[0134] Embodiment 29. An apparatus according to Embodiment 28, wherein the neural network includes at least one 2D convolutional layer after the first convolutional layer and before the one or more common layers.

[0135] Embodiment 30. An apparatus according to Embodiment 28, wherein the neural network includes at least one 2D fully connected layer after the first convolutional layer and before the one or more common layers.

[0136] Embodiment 31. The apparatus of Embodiment 21, wherein the one or more common layers include at least one fully connected layer.

[0137] Embodiment 32. A device comprising:

[0138] Display (44);

[0139] Artificial neural networks (34); and

[0140] A processor (32), the processor being configured to:

[0141] estimating the corresponding local activation time (LAT) of the first set of intracardiac electrocardiogram signals;

[0142] generating a three-dimensional (3D) map based on the estimated LAT and presenting the 3D map on the display;

[0143] receiving an indication of an abnormality in the 3D map;

[0144] generating training data using the anomaly and a region of the 3D map that does not include the anomaly;

[0145] Using the training data to train the artificial neural network; and

[0146] The artificial neural network is applied to estimate the LAT of a second set of intracardiac electrocardiogram signals.

[0147] Embodiment 33. The device of embodiment 32, wherein the first set of intracardiac electrocardiogram signals are generated from a healthy animal.

[0148] Embodiment 34. The apparatus of Embodiment 32, wherein estimating the corresponding LATs of the first group is automatically performed by the processor.

[0149] Embodiment 35. The apparatus of embodiment 32, wherein the indication is provided by manual inspection of the 3D map.

[0150] Embodiment 36. The apparatus of embodiment 35, and comprising: in response to the manual inspection, the processor automatically reassigns a corresponding LAT to the signal corresponding to the anomaly.

[0151] Embodiment 37. The apparatus of embodiment 35, and comprising: in response to the manual inspection, manually reassigning a corresponding LAT to the signal corresponding to the anomaly.

[0152] Embodiment 38. The apparatus of embodiment 32, wherein the training data comprises the first set of signals, the LAT corresponding to the abnormality, and the LAT corresponding to the region of the 3D map that does not include the abnormality.

[0153] It should be understood that the above examples are cited by way of example, and the present disclosure is not limited to what is specifically shown and described above. On the contrary, the scope of the present disclosure includes combinations and sub-combinations of the various features described above, as well as variations and modifications thereof, which should occur to those skilled in the art when reading the above description and which are not disclosed in the prior art.

Claims

1. A method for medical diagnosis, comprising: receiving electrophysiological data including a first intracardiac electrogram (IEGM) signal acquired by a first pair of electrodes including a first intracardiac electrode and a second IEGM signal acquired by a second pair of electrodes including a second intracardiac electrode proximal to the first intracardiac electrode; Inputting the first IEGM signal and the second IEGM signal into a first convolutional layer of a neural network and a second convolutional layer in the neural network parallel to the first convolutional layer, respectively, to generate a corresponding first intermediate result and a second intermediate result; Inputting the first intermediate result and the second intermediate result together into one or more common layers of the neural network; as well as In response to inputting the first intermediate result and the second intermediate result, an indication of local activation times at locations of the first intracardiac electrode and the second intracardiac electrode is received from an output layer of the neural network.

2. The method according to claim 1, wherein: The first pair of electrodes and the second pair of electrodes have a common neutral electrode.

3. The method according to claim 2, wherein: At least one of the first signal and the second signal comprises a unipolar signal measured relative to the neutral electrode.

4. The method according to claim 2, wherein: The first signal includes a combination of a first unipolar signal generated between the first intracardiac electrode and the neutral electrode, a second unipolar signal generated between the second intracardiac electrode and the neutral electrode, and a bipolar signal generated between the first intracardiac electrode and the second intracardiac electrode.

5. The method according to claim 1, wherein: The first signal is in the form of a one-dimensional (1D) matrix, and the first convolutional layer is one-dimensional, wherein at least one 1D kernel is configured to slide along the first convolutional layer.

6. The method according to claim 5, wherein: The neural network includes at least one 1D convolutional layer after the first convolutional layer and before the one or more common layers.

7. The method according to claim 5, wherein: The neural network includes at least one 1D fully connected layer after the first convolutional layer and before the one or more common layers.

8. The method according to claim 1, wherein: The neural network includes at least one flat layer after the first convolutional layer and before the one or more common layers.

9. The method according to claim 1, wherein: The first signal is in the form of a two-dimensional (2D) matrix, and the first convolutional layer is two-dimensional, wherein at least one 2D kernel is configured to slide along the first convolutional layer.

10. The method according to claim 9, wherein: The neural network includes at least one 2D convolutional layer after the first convolutional layer and before the one or more common layers.

11. The method according to claim 9, wherein: The neural network includes at least one 2D fully connected layer after the first convolutional layer and before the one or more common layers.

12. The method according to claim 1, wherein: The one or more common layers include at least one fully connected layer.

13. A method comprising: estimating the corresponding local activation time (LAT) of the first set of intracardiac electrocardiogram signals; generating a three-dimensional (3D) map based on the estimated LAT; receiving an indication of an abnormality in the 3D map; generating training data using the anomaly and a region of the 3D map that does not include the anomaly; training an artificial neural network using the training data; as well as The artificial neural network is applied to estimate the LAT of a second set of intracardiac electrocardiogram signals.

14. The method according to claim 13, wherein: The first set of intracardiac electrocardiogram signals is generated from a healthy animal.

15. The method according to claim 13, wherein: Estimating the respective LATs of the first group is automatically performed by a computer processor.

16. The method according to claim 13, wherein: The indication is provided by manual inspection of the 3D map.

17. A method according to claim 16, and comprising: In response to the manual inspection, a computer processor automatically reassigns a corresponding LAT to the signal corresponding to the anomaly.

18. The method according to claim 16, and comprising: In response to the manual inspection, a corresponding LAT is manually reassigned to the signal corresponding to the anomaly.

19. The method of claim 13, wherein the training data includes the first set of signals, the LAT corresponding to the abnormality, and the LAT corresponding to the region of the 3D map that does not include the abnormality.

20. A device for medical diagnosis, comprising: a probe comprising a first intracardiac electrode proximal to a second intracardiac electrode; A neural network, the neural network comprising: A first convolutional layer and a second convolutional layer, wherein the second convolutional layer is parallel to the first convolutional layer; one or more common layers; and Output layer; and A processor, the processor being configured to: receiving electrophysiological data, the electrophysiological data comprising a first intracardiac electrogram (IEGM) signal acquired by a first pair of electrodes including the first intracardiac electrode and a second IEGM signal acquired by a second pair of electrodes including the second intracardiac electrode; Inputting the first IEGM signal and the second IEGM signal into the first convolution layer and the second convolution layer respectively to generate a corresponding first intermediate result and a second intermediate result; inputting the first intermediate result and the second intermediate result together into the one or more common layers; and In response to inputting the first intermediate result and the second intermediate result, an indication of local activation times at locations of the first intracardiac electrode and the second intracardiac electrode is received from the output layer.

21. Apparatus according to claim 20 and comprising a neutral electrode common to the first and second pairs of electrodes.

22. The device according to claim 21, wherein At least one of the first signal and the second signal comprises a unipolar signal measured relative to the neutral electrode.

23. The device according to claim 21, wherein The first signal includes a combination of a first unipolar signal generated between the first intracardiac electrode and the neutral electrode, a second unipolar signal generated between the second intracardiac electrode and the neutral electrode, and a bipolar signal generated between the first intracardiac electrode and the second intracardiac electrode.

24. The device according to claim 21, wherein The first signal is in the form of a one-dimensional (1D) matrix, and the first convolutional layer is one-dimensional, wherein at least one 1D kernel is configured to slide along the first convolutional layer.

25. The device according to claim 24, wherein: The neural network includes at least one 1D convolutional layer after the first convolutional layer and before the one or more common layers.

26. The device according to claim 24, wherein: The neural network includes at least one 1D fully connected layer after the first convolutional layer and before the one or more common layers.

27. The device according to claim 21, wherein The neural network includes at least one flat layer after the first convolutional layer and before the one or more common layers.

28. The device according to claim 21, wherein The first signal is in the form of a two-dimensional (2D) matrix, and the first convolutional layer is two-dimensional, wherein at least one 2D kernel is configured to slide along the first convolutional layer.

29. The device according to claim 28, wherein The neural network includes at least one 2D convolutional layer after the first convolutional layer and before the one or more common layers.

30. The device according to claim 28, wherein The neural network includes at least one 2D fully connected layer after the first convolutional layer and before the one or more common layers.

31. The device according to claim 21, wherein The one or more common layers include at least one fully connected layer.

32. An apparatus comprising: monitor; Artificial neural networks; and A processor, the processor being configured to: estimating the corresponding local activation time (LAT) of the first set of intracardiac electrocardiogram signals; generating a three-dimensional (3D) map based on the estimated LAT and presenting the 3D map on the display; receiving an indication of an abnormality in the 3D map; generating training data using the anomaly and a region of the 3D map that does not include the anomaly; Using the training data to train the artificial neural network; as well as The artificial neural network is applied to estimate the LAT of a second set of intracardiac electrocardiogram signals.

33. The device according to claim 32, wherein: The first set of intracardiac electrocardiogram signals is generated from a healthy animal.

34. The device according to claim 32, wherein: Estimating the respective LATs of the first group is automatically performed by the processor.

35. The device according to claim 32, wherein: The indication is provided by manual inspection of the 3D map.

36. The apparatus of claim 35, and comprising: In response to the manual inspection, the processor automatically reassigns a corresponding LAT to the signal corresponding to the anomaly.

37. The apparatus of claim 35, and comprising: In response to the manual inspection, a corresponding LAT is manually reassigned to the signal corresponding to the anomaly.

38. The apparatus of claim 32, wherein: The training data includes the first set of signals, the LAT corresponding to the abnormality, and the LAT corresponding to the region of the 3D map that does not include the abnormality.

Images