Systems and methods for recommending ablation lines

Abstract

A system (10) for improving a cardiac ablation procedure includes a recommendation unit (14) configured to provide an initial recommendation of at least one proposed ablation line (4) for an ablation procedure performed on a patient's anatomy. The system displays the at least one proposed ablation line on an anatomical map of the anatomical structure. The recommendation unit includes a first trained machine learning model and a second trained machine learning model. Both the first trained machine learning model and the second trained machine learning model have the same structure.

Description

Technical Field

[0001] The present invention generally relates to systems for treating atrial fibrillation. Background Art

[0002] Atrial fibrillation (AF) is the most commonly diagnosed sustained arrhythmia, characterized by rapid and irregular atrial activation. Atrial fibrillation can be paroxysmal, lasting 7 days or less, with or without intervention, or can last for more than 7 days (persistent, PS-AF) or more than 12 months (long-term, LS-PS-AF). Persistent AF is a term used for long-term persistent AF, when any attempts to restore sinus rhythm have been abandoned or have proven impossible. (See, e.g., Kaba, RA, Momin, A. & Camm, J. "Persistent atrial fibrillation: The role of left atrial posterior wall isolation and ablation strategies". J. Clin. Med. 10, (2021)).

[0003] Treatment strategies vary depending on the degree and duration of AF. For example, paroxysmal AF is usually managed with wide antral circumferential ablation (WACA), i.e., delivering radiofrequency energy in a point-by-point manner along a predefined line in the atrium. However, a large proportion of AF patients have PS-AF or LS-PS-AF, which are more difficult to treat, probably due to the more extensive diffusion of arrhythmogenic triggers and substrates outside the pulmonary veins. The posterior wall of the left atrium is a common site of these changes and has been the target of ablation strategies to treat these more resistant forms of AF. Other common sites are the left and right carina and apex. Summary of the invention

[0004] Therefore, according to a preferred embodiment of the present invention, a system for improving a cardiac ablation procedure is provided, the system comprising a recommendation unit. The recommendation unit provides an initial recommendation for at least one proposed ablation line for an ablation procedure performed on an anatomical structure of a patient, and displays the at least one proposed ablation line on an anatomical map of the anatomical structure. The recommendation unit comprises a first trained machine learning model and a second trained machine learning model, both of which have the same structure.

[0005] Furthermore, according to a preferred embodiment of the present invention, the system comprises an ablation line editor enabling the physician to modify a selected ablation line of the at least one proposed ablation line.

[0006] Additionally, in accordance with a preferred embodiment of the present invention, the system includes a therapy determiner that determines a recommended energy delivery for each segment of the selected ablation line.

[0007] Still further, according to a preferred embodiment of the present invention, the system comprises a guiding unit which displays a next ablation site on the anatomical structure based at least on the selected ablation line and the catheter position in the anatomical structure.

[0008] In addition, according to a preferred embodiment of the present invention, the recommendation unit includes a map segmenter, an ablation line trainer, and an ablation line proposer. The map segmenter segments the anatomical map to output a portion of the anatomical structure, and the map segmenter uses a first trained machine learning model. The ablation line trainer trains a second trained machine learning model to output a proposed ablation line. The ablation line proposer uses the second trained machine learning model to propose at least one proposed ablation line for the anatomical structure.

[0009] Additionally, according to a preferred embodiment of the present invention, the same structure is a graph convolutional neural network (GCN) or a classifier.

[0010] Still further, according to a preferred embodiment of the present invention, the recommendation unit recommends which of the at least one proposed ablation line is most suitable for the procedure for the patient.

[0011] Furthermore, according to a preferred embodiment of the present invention, the at least one proposed ablation line is a wide peri-antral ablation (WACA) line.

[0012] In addition, according to a preferred embodiment of the present invention, the system includes a key point filter for comparing the actual ablation points of the training case with the intersection points or lines of adjacent parts on the segmented mapping map of the training case, and selecting those ablation points closest to the intersection points as key points.

[0013] According to a preferred embodiment of the present invention, there is also provided a computer-implemented method for improving a cardiac ablation procedure. The method includes providing an initial recommendation for at least one proposed ablation line for an ablation procedure performed on an anatomical structure of a patient. The providing includes displaying the at least one proposed ablation line on an anatomical map of the anatomical structure. The initial recommendation utilizes a first trained machine learning model and a second trained machine learning model, both of which have the same structure.

[0014] Furthermore, according to a preferred embodiment of the present invention, the method comprises enabling the physician to modify a selected ablation line of the at least one proposed ablation line.

[0015] Additionally in accordance with a preferred embodiment of the present invention, the method includes determining a recommended energy delivery for each segment of the selected ablation line.

[0016] Still further in accordance with a preferred embodiment of the present invention, the method includes displaying a next ablation site on the anatomical structure based at least on the selected ablation line and the catheter position in the anatomical structure.

[0017] In addition, according to a preferred embodiment of the present invention, the provision includes: segmenting the anatomical mapping map to output a portion of the anatomical structure, the segmentation utilizing the first trained machine learning model; training the second trained machine learning model to output a proposed ablation line; and proposing at least one proposed ablation line for the anatomical structure, the proposal utilizing the second trained machine learning model.

[0018] Additionally, according to a preferred embodiment of the present invention, the same structure is a graph convolutional neural network (GCN) or a classifier.

[0019] Still further in accordance with a preferred embodiment of the present invention, the providing includes recommending which of the at least one proposed ablation line is most suitable for the procedure of the patient.

[0020] Furthermore, according to a preferred embodiment of the present invention, the at least one proposed ablation line is a wide peri-antral ablation (WACA) line.

[0021] Finally, according to a preferred embodiment of the present invention, the method includes comparing the actual ablation points of the training case with the intersection points or lines of adjacent parts on the segmented map of the training case, and selecting those ablation points closest to the intersection points as key points. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The subject matter which is regarded as the invention is particularly pointed out and distinctly claimed in the concluding portion of the specification. However, the organization and method of operation of the invention, together with objects, features, and advantages thereof, may be best understood by reference to the following detailed description when read in connection with the accompanying drawings, in which:

[0023] Figure 1A and Figure 1B is a schematic diagram of an exemplary left atrium in a posterior-anterior view without and with marked ablation lines on the left atrium, respectively;

[0024] Figure 2 is a block diagram illustration of a deep learning system for drawing recommended ablation lines on anatomical structures, constructed and operative in accordance with a preferred embodiment of the present invention;

[0025] Figure 3 It is available for Figure 2 A block diagram of the deep learning architecture in the system;

[0026] Figure 4 The left atrium has an ablation line. Figure 1A A schematic diagram of an exemplary left atrium is provided to help understand Figure 2 the operation of the system;

[0027] Figure 5 It is a grid with indicated points on the left atrium Figure 1A A schematic diagram of an exemplary left atrium is provided to help understand Figure 2 the operation of the system;

[0028] Figure 6 There is an ablation point on the left atrium Figure 1A A schematic diagram of an exemplary left atrium is provided to help understand Figure 2 the operation of the system;

[0029] Figure 7 is a block diagram illustration of an alternative deep learning system for drawing recommended ablation lines on anatomical structures, constructed and operative in accordance with an alternative preferred embodiment of the present invention; and

[0030] Figure 8 It is available for Figure 7 Schematic diagram of the operation of the key point filter in the system.

[0031] It should be understood that for simplicity and clarity of illustration, the elements shown in the figures are not necessarily drawn to scale. For example, for clarity, the size of some of the elements may be exaggerated relative to other elements. In addition, where deemed appropriate, reference numerals may be repeated in the accompanying drawings to indicate corresponding or similar elements. DETAILED DESCRIPTION

[0032] In the following detailed description, numerous specific details are set forth in order to provide a comprehensive understanding of one example of the present invention. However, those skilled in the art will appreciate that the present invention may be practiced with numerous other examples or embodiments within the scope of the claims as amended, without conforming to the specific details of the particular illustrative examples provided herein. In other cases, well-known methods, procedures, and components have not been described in detail in order to avoid obscuring the present invention.

[0033] Applicants have recognized that deep learning systems can be used to plan ablation lines, such as wide circumsinus ablation (WACA) lines, and to check the proximity of the lines actually performed on the left atrium to the planned lines. In addition, applicants have recognized that the problem of defining ablation lines is a segmentation problem. Therefore, the input to such a system can be an anatomical map of the patient's heart, and the same type of deep learning structure can be used to automatically segment the anatomical map and for defining the location of the ablation line.

[0034] Reference now Figure 1A and Figure 1B , Figure 1A illustrates an exemplary left atrium 2 in a posteroanterior (PA) view, Figure 1B An exemplary ablation line 4 is shown marked on the left atrium 2. The left atrium 2 has multiple regions, such as a top wall 3 and a posterior wall 5. On the right side of the left atrium are the right superior pulmonary vein (RSPV) 6 and the right inferior PV (RIPV) 7, and on the left side of the left atrium are the left inferior PV (LIPV) 8 and the left superior PV (LSPV) 9.

[0035] Reference now Figure 2 , which illustrates a deep learning system 10 for drawing recommended ablation lines on anatomical structures. The deep learning system 10 includes three parts: an ablation line trainer 12, which builds trained deep learning units for an ablation line proposer 13; an ablation line recommender 14, which uses the ablation line proposer 13 to generate ablation line recommendations; and an ablation line guide 16, which provides ablation line guidance during a procedure.

[0036] The ablation line trainer 12 may receive an anatomical map of the heart, which may be obtained by using a suitable medical imaging system 20 or by using CARTO TM The system may be generated using the Fast Anatomical Mapping (FAM) technology available in the system, or using any other suitable technology or using any suitable combination of the above to generate an anatomical map image. The map / model of the heart onto which the recommended ablation line can be drawn may include any suitable type of three-dimensional (3D) anatomical map generated using any suitable technology. The map may be derived from an intracardiac ECG (electrocardiogram) and may include position information of the catheter moved around the atrium and ECG measurements (such as voltage at each location).

[0037] The ablation line trainer 12 may include an automatic map segmenter 22 and an ablation line neural network trainer 24, wherein the automatic map segmenter is used to segment the 3D map of the heart received from the imaging system 20 into its various parts, such as the posterior wall and anterior wall, top wall, inferior wall, left side wall, septum, mitral valve, left atrial appendage (LAA) and four pulmonary veins (LSPV, LIPV, RSPV, RIPV); the ablation line neural network trainer is used to investigate thousands of ablation cases on different maps received from the imaging system 20 and their different ablation strategies to generate a trained deep learning unit for the ablation line proposer 13.

[0038] The ablation line recommender 14 (including the ablation line proposer 13) gives an initial recommendation for a typical ablation line during a procedure, wherein the recommended line is drawn on an anatomical structure (e.g., a model of the heart segmented by the automatic map segmenter 22 and / or an anatomical map) so that the physician can ablate as close as possible to the recommended line, thereby producing an ablation line that will be as close as possible to the ablation line preferred by a key opinion leader (KOL) in the field. However, it should be noted that the ablation line proposer 13 can display the recommended ablation line described herein using the ablation line display 25 on any digital depiction of the anatomical structure, whether it is an anatomical map or an ultrasound image, including 4D ultrasound images, etc. Therefore, as used herein, a "map" of an anatomical structure refers to any such digital depiction onto which an ablation line can be projected.

[0039] The ablation line proposer 13 allows the physician to specify the desired ablation line; it can be a known ablation line, such as, for example: WACA, top line, carina line (right, left), posterior line, anterior line, inferior line, mitral valve line, posterior wall debulking, anterior wall debulking, inferior wall debulking, left side debulking, LAA isolation, etc.

[0040] The ablation line recommender 14 may also include an ablation line editor 27 to enable the physician to review and modify the selected ablation line according to their requirements, preferences and findings during the case. For example, the physician may want to edit the initial ablation line so that the initial ablation line will contain the trigger observed in the posterior wall.

[0041] The ablation line proposer 13 may segment the WACA circle into multiple parts and may recommend energy delivery for each segment, i.e., a recommended ablation index (radio frequency (RF) or irreversible electroporation (IRE)) and / or a recommended distance between adjacent points, such as may be specific to a particular ablation machine (such as Carto TM The ablation line may be predefined on the machine and / or may be configurable by the physician prior to performing the ablation and / or may be for each type of ablation line.

[0042] The ablation line editor 27 may also allow the physician to update the recommended therapy delivery for each segment.

[0043] The ablation line recommender 14 may require the physician to approve the plan once he has reviewed all parameters.

[0044] The ablation line guide 16 provides ablation line guidance during the procedure. The ablation line guide may receive an anatomical map of the heart from a medical imaging system 20 during the procedure and may include an automatic map segmenter 22, an ablation line guidance unit 30, and a performance reporter 32 to indicate various elements of the heart during the procedure.

[0045] During the procedure, based on the catheter position and the associated ablation line from the approved plan; the ablation line guidance unit 30 can present a dynamic display of the next ablation site based on the previous site and the recommended ablation line (approved or edited by the physician). The next ablation site can then be displayed based on a predefined distance (e.g., 4 mm) from the previous site, with a minimum distance from the suggested line and in the direction of advancement of the line.

[0046] The goal of an ablation procedure is to create effective lesions that can interrupt abnormal electrical pathways that cause arrhythmias. To this end, the ablation wire guidance unit 30 can calculate an ablation index that indicates the quality of each radiofrequency (RF) ablation lesion during the procedure. The ablation index is a function of at least contact force, time, and power, and therefore increases as the ablation operation continues. The instantaneous ablation index provides real-time feedback to the physician, enabling them to monitor and adjust their technique during the procedure.

[0047] Since the physician can define a "required ablation index" for a given ablation line and for a given segment within the line, the ablation line guidance unit 30 can present both the "required ablation index" and the instantaneous ablation index of the current site. This can enable the physician to obtain the desired ablation index.

[0048] The performance reporter 32 may provide a report to the physician based on the physician's performance during the procedure relative to the ablation plan generated by the ablation line recommender 14. The report may include data such as, for example:

[0049] Maximum and median deviation from plan for each ablation line [in mm];

[0050] Deviations from the planned location; including the number and percentage of ablation points with distances greater (or less) than the planned distance; and

[0051] deviations from the ablation planned for each ablation line;

[0052] NOTE: This report is particularly useful when training less experienced physicians.

[0053] The ablation line recommender 14 may be configured to always draw the recommended WACA (left and right) ablation lines.

[0054] As mentioned above, applicants have recognized that the same type of deep learning architecture can be utilized and the same type of input can be provided to both the automatic map segmenter 22 and the ablation line neural network trainer 24. Figure 3 The deep learning structure discussed in more detail below is shown, and the input is indicated to be the anatomical map discussed above and the adjacency map discussed below. However, Figure 3Two types of output are shown, segmentation codes for the automatic map segmenter 22 and ablation line codes for the ablation line neural network trainer 24 .

[0055] The segmentation codes may be a set of codes indicating a portion of the heart or a portion of the atrium. For example, there may be 13 different portions of the atrium. The ablation line codes may be a set of codes indicating the type of ablation line, as well as codes for all areas where ablation lines are missing. As discussed below, each training case may include an input map and an output map, with each location labeled with one of the associated codes, a segmentation code for training the automatic map segmenter 22, or an ablation line code for training the ablation line neural network trainer 24.

[0056] The initial data set may consist of a sufficient number of paroxysmal and persistent AF cases (e.g., 5000 or more cases) along with an anatomical map of each case and at least one ablation line performed in that case. The ablation line is typically one of the following: (1) left WACA, (2) right WACA, (3) ostium PVI, (4) top line, (5) right carina, (6) left carina, (7) posterior line, (8) inferior line, (9) mitral isthmus line, (10) anterior line, (11) tricuspid isthmus (CTI) line, (12) superior vena cava division (SVCI), (13) LAA division.

[0057] In order to prepare an initial data set to be used as input for training and testing the automatic map segmenter 22 and the ablation line neural network trainer 24, an expert physician (e.g., a KOL in the field) may initially add manual annotations of ablation points on the anatomical map of the case generated by the imaging system 20. Each point is associated with at least one of the ablation lines (1) to (13).

[0058] In the next step, data pre-processing can be performed, whereby each ablation point can be drawn on the anatomical map using a specific radius (e.g., a 5 mm radius in this example). Figure 4 , to which reference will now be briefly made. Figure 4 A posterior view of an exemplary left atrium 40 is shown after projection of a 5 mm radius ablation point. Figure 4 A left WACA line 42 and a right WACA line 44 are shown.

[0059] In another step, each map is investigated and maps with gaps and / or blurred lines can be filtered out from the training and testing sets.

[0060] After this, further data preprocessing can be performed, among which:

[0061] a. The atria are converted to fixed dimensions (e.g., by generating a constant grid file with a fixed number of points (such as, for example, 2000 points) and a 2000 x 2000 adjacency matrix, where all values ​​are configurable, such as values ​​1000, 2000, 4000, 10000, etc.).

[0062] b. Feature extraction for 3D general segmentation: In order to build a deep learning framework for 3D segmentation, the graph data is characterized by improving the local features of each node. This can be achieved by using the 1-triangle neighboring space and structural features of each mesh face, such as Figure 5 Described in Figure 5 A mesh surface is shown formed by a point 46 having a position (x, y, z) and a vector (N x ,N y ,N z ) represents its normal N.

[0063] In a 1-triangle mesh, each face contains at most three neighboring faces. 3D mesh data is characterized by M = {V, E, F} of vertices, edges, and faces.

[0064] Therefore, the 1-triangle mesh is converted into a graph G = {features, adjacency}. The face {F} on the mesh M has three corresponding nodes (vertices) on the graph {G}. In this case, the adjacency matrix {adjacency} is a 6000x6000 matrix. The adjacency matrix is ​​based on {F} and represents the connections between vertices in M. In order to enhance the perception of the local area of ​​the graph node, spatial and structural geometric features are used based on a single vertex of the simplified mesh (consisting of 6000 vertices in this example). The feature space is a 6000x9 matrix, and for each vertex (a triangle consisting of three vertices with 3 adjacent faces), the preprocessing uses the following features:

[0065] pi — normalized (xi,yi,zi) position of vertex Vi in Euclidean space, where i represents the three indices of the vertex;

[0066] spi—spherical coordinates of vertex i in normalized by π and θ normalized by 2π; and

[0067] ni — Normal of vertex i defined by (xni,yni,zni).

[0068] c. Network Architecture: The automatic map segmenter 22 can utilize a machine learning model consisting of two layers, the first layer consisting of six graph convolutional neural network units, and the second layer is a decision layer consisting of another GCN with a Softmax unit. Each GCN unit gives a prediction for a given vertex associated with each of the target categories (anatomical structures or ablation lines).

[0069] like Figure 3 As shown, the anatomical map and the adjacency matrix are used as inputs to all units in the network. Two types of GCN units are used: 1) Approximate Personalized Propagation of Neural Predictions (APPNP) and 2) Autoregressive Moving Average Layer (ARMA), which is efficient as a SoftMax layer for improving the overall performance of the network during the training and cross-validation phases, and provides a more flexible frequency response, is more robust to noise, and better captures the global graph structure. Gasteiger, J., Bojchevski, A. & Unnemann, SG¨. "Predict then propogate: Graph neural networks meet personalized PageRank". Int. Conf. Learn. Represent. (2018) describes personalized graph neural networks. Maria Bianchi, F., Grattarola, D., Livi, L. & Alippi, C. "Graph Neural Networks with Convolutional ARMA Filters" (2018) has described a graph neural network with CNN ARMA filters. The APPNP unit models the relationship between the GCN network and the PageRank algorithm (a link analysis algorithm originally designed by Google that assigns numerical weights to each node based on the number and quality of links pointing to each node) and extends it to personalized PageRank. It uses information from a large, adjustable neighborhood to classify each node. The model is computationally efficient and outperforms several prior art methods for semi-supervised classification on multiple graphs. The APPNP has 13 channels, with a multilayer perceptron (MLP) having two hidden layers with 128 and 64 neurons in each layer, respectively.

[0070] The network weights are trained using a weighted class cross entropy loss function. The automatic map segmenter 22 may provide segmentation codes to its segmentation loss function, and the ablation line neural network trainer 24 may provide ablation line codes to its ablation line loss function.

[0071] d. In an exemplary embodiment, the training set contains 5000 maps. Three matrices (6000x9 feature matrix, 6000x6000 adjacency matrix and 6000x1 target matrix) are used to represent each map, one matrix for segmentation codes and one matrix for ablation codes. For ablation codes, zero indicates no line, and non-zero indicates an ablation line ID.

[0072] e. Once the neural network of the automatic map segmenter 22 is trained, the automatic map segmenter 22 may segment the anatomical map of each atrium and may provide the segmented map to the ablation line display 25 .

[0073] f. From Segments to Lines - The ablation line proposer 13 can form a concentrated line with potential locations for point-by-point ablation on the anatomical map. For example, Figure 3 The mapping map is Figure 6 is depicted with point-by-point ablations indicated by points 50 .

[0074] G. Point-by-Point Ablation "Points" may be updated on the fly based on the actual ablations performed by the physician during each case. The mechanism for updating points may involve minimizing the distance between the planned line and the next point in the advancing direction of the ablation line.

[0075] As discussed, the system 10 can use the mapping data and the automatic segmentation data to recommend which ablation line is more appropriate for a given case. The recommendation can be set based on instructions from a specific research hospital or a specific physician (e.g., a KOL in the field of cardiac ablation) or based on data from multiple physicians and / or research hospitals.

[0076] In an alternative embodiment, the structure for both the automatic map segmenter 22 and the ablation line trainer 24 may include a classifier that operates on data from an intracardiac ECG. During an intracardiac ECG recording, a catheter is inserted into the heart to measure the electrical activity of the heart. The electrical signal at each anatomical location is represented as a voltage amplitude.

[0077] In this alternative embodiment, the classifier may be any suitable classifier, such as a random forest classifier, a support vector machine (SVM) classifier, or a deep learning classifier, such as described in the following article:

[0078] Breiman L(2001)."Random Forests".Machine Learning.45(1):5–32.Bibcode:2001;

[0079] Ho, Tin Kam (1995). "Random Decision Forests" (PDF). Proceedings of the 3rd International Conference on Document Analysis and Recognition, Montreal, QC, August 14-16, 1995, pp. 278-282. Archived from the original (PDF) on April 17, 2016. Retrieved on June 5, 2016; and

[0080] Cortes, Corinna; Vapnik, Vladimir (1995). “Support-vector networks” (PDF). Machine Learning. 20(3):273–297. CiteSeerX 10.1.1.15.9362.doi:10.1007 / BF00994018.S2CID 206787478.

[0081] Each case in the training dataset may be pre-processed with an indication of whether one of a given number of ablation lines (e.g., 14) was performed by a key opinion leader (KOL), so the target is a matrix of size Nx14, where N represents the number of cases in the training dataset.

[0082] iii. Input Data: ECG data from intracardiac ECG is saved as an additional feature matrix of 6000x22. The following trigger conditions and substrate maps from intracardiac ECG data are used as feature space:

[0083] 1) The focused source at each vertex (6000x1) of "Finder" represents the number of S waves per second;

[0084] 2) “Finder” RAP – each vertex is represented by the number of rotation patterns per second;

[0085] 3) CL (cycle length measurement) - each vertex is represented by the cycle length in milliseconds;

[0086] 4) STD CL (Standard Deviation of Cycle Length) – Each vertex is represented by the standard deviation of cycle length per second;

[0087] 5) Spatiotemporal dispersion - each vertex represents the percentage of spatiotemporal dispersion (in records ranging from 10 to 30 seconds);

[0088] 6) Periodic space-time dispersion - each vertex represents the percentage of periodic space-time dispersion (in records between 10 and 30 seconds);

[0089] 7) Voltage map - each vertex represents the bipolar amplitude in mV;

[0090] 8) CLM ROI – each vertex indicates the presence of a loop length region of interest, based on the cutoffs of loop length < minimum CL + 15 ms and STD CL < 15 ms;

[0091] 9) Segmented one-hot encoding is the output of the automatic map segmenter 22: a matrix of size (6000xk), where K represents the number of anatomical structures (13 for LA) (output from the segmentation algorithm). Ablation line encoding is the output of the ablation line trainer 24;

[0092] 10) Global voltage mV per segment area (6000x1); and

[0093] 11) Low voltage areas, which are areas within the left atrium where the recorded voltage amplitude is relatively low compared to surrounding areas. For example, these areas may be areas where the voltage is between 0.2mV and 0.5mV. Typically, scars may be defined as areas where the voltage is below 0.2mV. It should be understood that these thresholds may be different based on operator decisions in the protocol.

[0094] The classifier may receive as input a feature matrix, which may have dimensions of 6000xM, where 6000 represents the number of vertices on the map and M represents the number of features (focal, substrate-related, ripple, frequency, anatomical location, etc.).

[0095] The target vector has dimensions Kx1, where K indicates the number of ablation methods or IDs. For this embodiment, the data may come from successful cases of the procedure, specifically those cases that have been free of any arrhythmias for 12 months. For each of these cases, the Kx1 vector will be set to 1 if a particular ablation method was performed during the procedure.

[0096] The neural network of the classifier can utilize a set of dense nets, where the last layer will consist of a SoftMax layer with K neurons in the output layer. This modification enables the prediction of the K ablation method.

[0097] It should be understood that the systems and methods described herein may exist in a variety of local or distributed computers (e.g., cloud-based processing systems), including but not limited to the forms described by U.S. Patent No. 11,198,004 or U.S. Patent Publication No. 20200352652A1, which are owned by the applicant and incorporated herein by reference.

[0098] Applicants have recognized that actual ablation line data may be noisy and the ablation lines they indicate may not be ideal lines. Figure 7 In another embodiment shown in , the system 10' may additionally include a key point filter 23 for examining the noisy ablation line data and, based on the segmented map received from the automatic map segmenter 22, finding those ablation points in the ablation line data that may be defined as key points (i.e., important points along the ablation line that should not be missed) defined by an expert (such as a KOL) associated therewith. Typically, these key points may be points at the intersection of two parts of the atrium. This is Figure 8 , to which reference is now made. The input 100 of the key point filter 23 can be an image or grid of the left atrium with actual ablation lines, which may have been segmented (102) by the automatic map segmenter 22. The key point filter 23 can detect key points 104 by comparing the ablation points of each case with the intersection points on the segmented map 102 for the case and selecting those ablation points that are closest to the intersection points. The key point filter 23 can combine key points from multiple cases to generate a resulting ablation line 106 for training, which can be thinner and better defined than the ablation lines of other cases. The key point filter 23 can provide the resulting ablation line 106 to the ablation line neural network trainer 24 for training.

[0099] Unless specifically noted otherwise, it is apparent from the foregoing discussion that throughout the specification, discussions using terms such as "processing," "computing," "calculating," "determining," etc. refer to the actions and / or processes of any type of general purpose computer, such as a client / server system, a mobile computing device, a smart device, a cloud computing unit, or a similar electronic computing device, that operates on and / or transforms data within registers and / or memory of a computing system into other data within the memory, registers, or other such information storage, transmission, or display device of the computing system.

[0100] Embodiments of the present invention may include a device for performing the operations herein. The device may be specially constructed for the desired purpose, or the device may include a computing device or system generally having at least one processor and at least one memory, selectively activated or reconfigured by a computer program stored in a computer. When indicated by software, the resulting device may turn a general-purpose computer into an inventive element as discussed herein. Instructions may define inventive devices that operate with a desired computer platform. Such computer programs may be stored in a computer-readable storage medium, such as, but not limited to, any type of disk, including optical disks, magneto-optical disks, read-only memories (ROMs), volatile memories, and non-volatile memories, random access memories (RAMs), electrically programmable read-only memories (EPROMs), electrically erasable and programmable read-only memories (EEPROMs), magnetic cards or optical cards, flash memory, disks on keys, or any other type of medium suitable for storing electronic instructions and capable of being connected to a computer system bus. Computer-readable storage media may also be implemented in cloud storage devices.

[0101] Some general purpose computers may include at least one communication element to enable communication with a data network and / or a mobile communication network.

[0102] The process and display provided herein are not inherently related to any particular computer or other device. Various general purpose systems can be used together with the program according to the teaching of this article, or it is also proved to be convenient to construct a more specialized device to perform the desired method. The desired structure for various such systems will appear from the following description. In addition, embodiments of the present invention are not described with reference to any specific programming language. It should be understood that a variety of programming languages ​​can be used to implement the teaching of the present invention described herein.

[0103] While certain features of the invention have been illustrated and described herein, numerous modifications, substitutions, changes, and equivalents will now occur to those skilled in the art. It is therefore to be understood that the appended claims are intended to cover all such modifications and changes that fall within the true spirit of the invention.

Claims

1. A system for improving a cardiac ablation procedure, the system comprising: a recommendation unit configured to provide an initial recommendation of at least one proposed ablation line for an ablation procedure performed on an anatomical structure of a patient, the system displaying the at least one proposed ablation line on an anatomical map of the anatomical structure; Wherein, the recommendation unit includes a first trained machine learning model and a second trained machine learning model, and wherein the first trained machine learning model and the second trained machine learning model both have the same structure. 2 . The system of claim 1 , further comprising an ablation line editor configured to enable a physician to modify a selected ablation line of the at least one proposed ablation line. 3 . The system of claim 1 , further comprising a treatment determiner configured to determine a recommended energy delivery for each segment of the selected ablation line. 4 . The system according to claim 1 , further comprising a guiding unit configured to display a next ablation site on the anatomical structure based at least on a selected ablation line and a catheter position in the anatomical structure.

5. The system according to claim 1, and wherein: The recommendation unit includes: a map segmenter for segmenting the anatomical map to output portions of the anatomical structure, the map segmenter utilizing the first trained machine learning model; an ablation line trainer for training the second trained machine learning model to output proposed ablation lines; and An ablation line proposer that utilizes the second trained machine learning model to propose the at least one proposed ablation line for the anatomical structure.

6. The system according to claim 5, wherein: The same structure is a graph convolutional neural network (GCN).

7. The system according to claim 5, wherein: The same structure is a classifier.

8. The system according to claim 1, wherein: The recommendation unit is further configured to recommend which of the at least one proposed ablation line is most suitable for a procedure for the patient.

9. The system according to any one of claims 1 to 8, wherein: The at least one proposed ablation line is a wide peri-antral ablation (WACA) line.

10. The system according to claim 5 further comprises a key point filter, which is used to compare the actual ablation points of the training case with the intersection points or lines of adjacent parts on the segmented mapping map of the training case, and select those ablation points closest to the intersection points as key points.

11. A computer-implemented method for improving a cardiac ablation procedure, the method comprising: providing an initial recommendation of at least one proposed ablation line for an ablation procedure performed on an anatomical structure of a patient, including displaying the at least one proposed ablation line on an anatomical map of the anatomical structure; Wherein the initial recommendation utilizes a first trained machine learning model and a second trained machine learning model, and wherein both the first trained machine learning model and the second trained machine learning model have the same structure. 12 . The method of claim 11 , further comprising enabling a physician to modify a selected ablation line of the at least one proposed ablation line.

13. The method of claim 11, further comprising determining a recommended energy delivery for each segment of the selected ablation line.

14. The method of claim 11, further comprising displaying a next ablation site on the anatomical structure based at least on the selected ablation line and the catheter position in the anatomical structure.

15. The method according to claim 11, and wherein: The provision includes: segmenting the anatomical map to output portions of the anatomical structure, the segmenting utilizing the first trained machine learning model; training the second trained machine learning model to output proposed ablation lines; and The at least one proposed ablation line is proposed for the anatomical structure, the proposal utilizing the second trained machine learning model.

16. The method according to claim 15, wherein: The same structure is a graph convolutional neural network (GCN).

17. The method according to claim 15, wherein: The same structure is a classifier.

18. The method according to claim 11, wherein: Providing a protocol further includes recommending which of the at least one proposed ablation line is most appropriate for the patient.

19. The method according to any one of claims 11 to 18, wherein: The at least one proposed ablation line is a wide peri-antral ablation (WACA) line.

20. The method of claim 15, further comprising comparing actual ablation points of a training case with intersection points or lines of adjacent portions on a segmented map of the training case, and selecting those ablation points closest to the intersection points as key points.

Images