Computer system and method for enhancing atrial fibrillation treatment

Abstract

The presently disclosed subject matter includes computer methods and computer systems that enable selection and provision of suitable treatments for AF patients that increase the likelihood of long term amelioration of AF conditions, thereby enhancing treatment for AF patients. The present disclosure provides methods and systems for analyzing conditions of AF patients and customizing personalized treatment regimens and determining treatments selected from at least only PVI and PVI + based on various AF characteristics obtained from these patients.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of U.S. Provisional Patent Application No. 63 / 524,712 filed on July 3, 2023 and U.S. Provisional Patent Application No. 63 / 611,303 filed on December 18, 2023, which are incorporated herein by reference. Technical Field

[0003] The presently disclosed subject matter relates to computer systems and computer-implemented methods for treating atrial fibrillation. Background Art

[0004] 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 persist for more than 7 days (persistent, PS-AF) or more than 12 months (long-term, LS-PS-AF). Persistent AF is the 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., and Camm, J., Persistent atrial fibrillation: The role of left atrial posterior wall isolation and ablation strategies J. Clin. Med., 10, (2021).

[0005] Treatment strategies vary depending on the extent and duration of AF. Pulmonary vein isolation (PVI) is a medical procedure aimed at restoring normal heart rhythm and reducing or eliminating symptoms associated with AF. However, in severe cases of AF, PVI alone may not be sufficient, and long-term failures of "PVI alone" have been observed. Summary of the invention

[0006] Evidence shows a consistent link between enlargement of the left atrium's size and worsening severity and chronic nature of atrial fibrillation (AF) in affected patients. The shape of the atrium is also known to change as the disease progresses, from a tubular to a spherical shape.

[0007] Figure 1a and Figure 1b The imaging system output of the left atrium of two AF patients is shown.In this example, the dilation value (in the present case, the Euclidean dilation value) is calculated by dividing the width of the posterior wall by its length.

[0008] The methods and systems disclosed herein enable selection and provision of appropriate therapy for AF patients, which increases the likelihood of long-term improvement of AF conditions, thereby enhancing treatment for AF patients.

[0009] Disclosed herein are changes (e.g., dilation) in different subcompartments (or "chambers") of the heart observed and measured in specific anatomical locations of the different subcompartments of the heart that can be used to predict the long-term success or failure of PVI-only therapy and / or PVI plus therapy in AF patients.

[0010] Taking the left atrium as an example, the inventors have found that geometric transformations (e.g., dilation) in each of its anatomical regions (including the posterior wall, anterior wall, left side wall, right side wall, top wall, and inferior wall) have predictive quality regarding the outcome of PVI-only treatment. According to the inventors' findings, certain geometric properties (or specific combinations of geometric properties) of the left atrium observed in AF patients indicate whether long-term success or long-term failure is expected after PVI-only treatment, and therefore help select the appropriate treatment to be administered to the patient, choosing between PVI-only and PVI+.

[0011] According to some examples, an index (and novel techniques utilizing the index) is presented that indicates the likelihood of long-term success of a "PVI only" protocol in subjects with atrial fibrillation. The index can be used to select patients for PVI only or one of the different PVI+s. The index is generated based on cardiac-related characteristics, including geometric properties and possibly other cardiac-related characteristics, such as voltage-related characteristics and cycle length. Additional details of the statistical analysis of data obtained from AF patients supporting the above principles are provided in the "Examples" section below.

[0012] One challenge of using the geometric properties of sub-compartments of the heart relates to the fact that the imaging output of cardiac imaging systems is often impaired or incomplete and includes unmapped regions. This shortcoming creates difficulties and, in some cases, precludes the ability to determine changes in the overall volume and / or shape of the heart and / or atria, thus creating technical deficiencies in analyzing cardiac pathology and selecting appropriate treatments in patients with AF.

[0013] The subject matter disclosed herein includes a new method for predicting the success of PVI treatment that is effective even in the case of incomplete imaging output. According to the method, a set of anatomical prediction parameters (or "success prediction parameters") can be used, each parameter being calculated based on geometric data collected from a corresponding anatomical part of the heart. The multiplicity and diversity of the parameters and the fact that different anatomical prediction parameters are calculated based on different corresponding anatomical parts provide redundancy in the data, so that even if the imaging output of the heart is incomplete, the mapped region of the heart provides sufficient output that can be used to calculate the parameters and provide the desired prediction based on these parameters.

[0014] The subject matter disclosed herein also includes computer systems and methods specifically for recommending / selecting appropriate ablation methods for AF patients using a machine learning (ML) model. A machine learning classifier is trained using a training data set including multiple features determined for different patients in an AF patient database (hereinafter referred to as "AF-related features") to associate various combinations of AF-related features with corresponding success probabilities of a PVI-only procedure and / or a PVI+ procedure.

[0015] According to one approach, the training data includes information indicating whether PVI-only or PVI+therapy was administered to each patient and the observed long-term success of the AF therapy. AF-related features are labeled according to the observed success or failure of the therapy, thereby enabling the ML model to make associations between the AF-related features and the recommended AF therapy.

[0016] According to another example, the training data includes information indicating PVI treatments (ablation methods) recommended by physicians treating patients, rather than actual observed treatment outcomes. As further explained below, the approach is based on the wisdom of crowds (WoC) principle, enabling the ML model to make associations between AF-related features and treatment options selected by physicians. This helps overcome technical challenges caused by a lack of sufficient data on observed AF treatment outcomes.

[0017] The present disclosure also relates to the field of personalized medicine. More specifically, the present disclosure provides a method and system for analyzing the condition of an AF patient and customizing a personalized treatment selected from at least PVI only and PVI+ based on various AF characteristics obtained from the patient. Summary of the invention

[0019] According to a first aspect of the disclosed subject matter, there is provided a computer-implemented method of selecting atrial fibrillation (AF) treatment for a patient, the method comprising:

[0020] obtaining a 3D anatomical map of at least one subcompartment of the patient's heart;

[0021] segmenting the anatomical (3D) map to obtain a plurality of anatomical locations of the at least one subcompartment;

[0022] determining at least one corresponding dimension for at least one of the plurality of anatomical sites;

[0023] determining one or more anatomical prediction parameters, wherein each anatomical prediction parameter is determined based on each anatomical prediction parameter calculated based on geometric data collected from an anatomical region of the patient's heart;

[0024] Based on the one or more anatomical prediction parameters, a prediction is generated indicating a likelihood of long-term success of a pulmonary vein isolation (PVI)-only procedure and / or a likelihood of long-term success of a PVI+procedure applied to the patient's heart.

[0025] In addition to the features described above, the method according to this aspect of the disclosed subject matter may optionally include one or more of the features i to xi listed below in any technically possible combination or arrangement:

[0026] i. The method includes: generating a recommendation for AF treatment for the patient based on the prediction, the recommendation selected from a pulmonary vein isolation (PVI) only protocol and / or a PVI+ protocol; and generating data indicating the recommendation to be provided to a healthcare provider.

[0027] ii. wherein generating the prediction comprises:

[0028] Obtain AF-related features of the patient, where the AF-related features include cardiac-related features characterizing various cardiac-related attributes of the patient, and the cardiac-related features include at least the following types:

[0029] a. the one or more anatomical prediction parameters; and

[0030] b. one or more voltage-related characteristics;

[0031] applying a machine learning (ML) model to the AF-related feature of the patient; wherein the ML model has been trained using a training dataset, the training dataset comprising a plurality of data points, each data point corresponding to an AF patient associated with the AF-related feature and an observed long-term success of an AF therapy administered to the patient, the AF therapy comprising at least only PVI, whereby the ML model is trained to provide data indicative of a recommended AF therapy for a given patient based on the AF-related feature of the given patient during inference; and

[0032] A machine learning model output is generated that indicates a recommended AF treatment selected from a group including at least PVI-only and PVI+treatment.

[0033] iii. Wherein the machine learning model output includes an indication of the importance of different output features in determining the recommended

[0034] Information on the corresponding significance in AF treatment.

[0035] iv. Wherein an ML model is trained to provide data indicative of the probability of long-term success of a pulmonary vein isolation (PVI)-only procedure.

[0036] v. Wherein an ML model is trained to provide data indicating the probability of long-term success of a pulmonary vein isolation (PVI)+ procedure.

[0037] vi. wherein an ML model is trained to provide data indicating the probability of long-term success of a pulmonary vein isolation (PVI)-only procedure as well as a PVI+procedure.

[0038] vii. The ML output including a recommendation to apply the PVI+ protocol to the patient also includes information indicating the specific type of PVI+ protocol that is recommended.

[0039] viii. Wherein the AF-related features further include one or more of the following: cycle length, patient's medical history, and patient's demographics.

[0040] ix. The method includes comparing each of the one or more anatomical prediction parameters to a corresponding threshold value, and determining a prediction based on the comparison output.

[0041] x. The method includes determining anatomical prediction parameters by applying mathematical operations between different dimensions of the anatomical part.

[0042] xi. The method includes performing one or more procedures on the patient to obtain heart-related characteristics from the patient.

[0043] According to another aspect of the presently disclosed subject matter, a computer system includes at least one processing circuit configured to select an atrial fibrillation (AF) treatment for a patient, the at least one processing circuit configured to:

[0044] obtaining a 3D anatomical map of at least one subcompartment of the patient's heart;

[0045] segmenting the anatomical (3D) map to obtain a plurality of anatomical locations of the at least one subcompartment;

[0046] determining at least one corresponding dimension for at least one of the plurality of anatomical sites;

[0047] determining one or more anatomical prediction parameters, wherein each anatomical prediction parameter is determined based on each anatomical prediction parameter calculated based on geometric data collected from an anatomical region of the patient's heart;

[0048] Based on the one or more anatomical prediction parameters, a prediction is generated that indicates the likelihood of long-term success of a pulmonary vein isolation (PVI)-only procedure and / or a PVI+procedure applied to the patient's heart.

[0049] The presently disclosed subject matter also contemplates a non-transitory program storage device readable by a computer, tangibly embodying a program of instructions executable by the computer to perform the method disclosed according to the first aspect above.

[0050] The subject matter disclosed herein also contemplates a computer-implemented method and computer system (and corresponding non-transitory computer storage device) for inducing long-term improvement in a patient with atrial fibrillation (AF), the method comprising performing the operations as described in the first aspect above and administering to the patient PVI-only therapy or PVI+ as prescribed by the recommendation.

[0051] The subject matter disclosed herein also contemplates a computer-implemented method and computer system (and corresponding non-transitory computer storage device) for predicting a response to pulmonary vein isolation (PVI) therapy administered to a patient with atrial fibrillation (AF), the method comprising operations according to the first aspect described above.

[0052] The presently disclosed subject matter also includes a computer-implemented method and system (and corresponding non-transitory computer storage device) dedicated to training a machine learning (ML) model to select or recommend a treatment for AF (atrial fibrillation) by providing data indicating a recommended treatment selected between pulmonary vein isolation (PVI) and PVI+therapy; the method comprising:

[0053] Generate a training dataset comprising a plurality of data points, each data point corresponding to an AF patient associated with an AF-related feature and an observed long-term success of an AF therapy administered to the patient, the AF therapy comprising at least PVI only and possibly also PVI+; use the generated dataset to train an ML model to associate specific AF-related features with corresponding probabilities of long-term success of the AF therapy, thereby enabling the ML model to provide data indicative of a recommended AF therapy for a given patient based on the AF-related features of the given patient during inference.

[0054] According to another aspect of the disclosed subject matter, there is provided a computer program product stored on a (non-transitory) computer storage device, the computer program product comprising software executable by a computer, the software comprising a machine learning (ML) model for recommending AF (atrial fibrillation) treatment to an AF patient, wherein the software is generated by a process comprising the steps of:

[0055] Generate a training dataset comprising a plurality of data points, each data point corresponding to an AF patient associated with an AF-related feature and an observed long-term success of an AF therapy administered to the patient, the AF therapy comprising at least PVI only and possibly also PVI+; train an ML model using the generated training dataset to associate specific AF-related features with corresponding probabilities of long-term success of the AF therapy, thereby enabling the ML model to provide data indicative of a recommended AF therapy for a given patient based on the AF-related features of the given patient during inference.

[0056] According to other aspects of the subject matter disclosed in the present invention, a computer system and a computer-implemented method (and corresponding non-transitory computer storage device) for determining personalized treatment for AF patients are provided, which method includes the steps / operations described above according to the above-mentioned first aspect.

[0057] The systems, methods, non-transitory program storage devices and computer program products disclosed above according to various aspects of the subject matter disclosed by the present invention may optionally include one or more of the features (i) to (xi) listed above in any technically relevant combination or arrangement, mutatis mutandis. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to understand the invention and to see how the subject matter may be carried out in practice, various embodiments may be described by way of non-limiting examples with reference to the accompanying drawings, in which:

[0059] Figure 1a and Figure 1b is an illustration of a dilated left atrium in a patient with AF;

[0060] Figure 2 is a block diagram of a computer system for enhanced treatment of AF patients according to some examples of the presently disclosed subject matter;

[0061] Figure 3a is a flow chart illustrating operations performed as part of enhanced recommendations and treatments for AF patients in accordance with some examples of the presently disclosed subject matter;

[0062] Figure 3b is a flow chart illustrating operations performed as part of enhanced machine learning-based recommendations and treatments for AF patients in accordance with some examples of the presently disclosed subject matter;

[0063] Figure 4 shows various stages of computer analysis according to some examples of the presently disclosed subject matter;

[0064] Figure 5 are images showing four anchor points of a rear wall according to some examples of the presently disclosed subject matter;

[0065] Figure 6 is a flow chart illustrating operations performed to train multiple machine learning models according to the WoC method according to some examples of the presently disclosed subject matter;

[0066] Figure 7 schematically illustrates inference using an ensemble of classifiers according to some examples of the presently disclosed subject matter;

[0067] Figure 8is a block diagram of a computer system according to some examples of the presently disclosed subject matter that combines the functionality of system 200 with additional functionality related to the identification of ablation lines;

[0068] Fig. 9 is a graph showing specific examples of correlations between right side wall or septum (RLAT) width parameters and long-term success of PVI-only treatment according to some examples of the presently disclosed subject matter; and

[0069] Fig.10 Graphs showing the probability of survival of AF patients over time according to some examples of the disclosed subject matter are presented. DETAILED DESCRIPTION

[0070] In the following detailed description, many specific details are set forth in order to provide a comprehensive understanding of the present invention. However, it should be understood by those skilled in the art that the subject matter disclosed in the present invention can be practiced without these specific details. In other cases, well-known methods, procedures, components and circuits are not described in detail to avoid obscuring the subject matter disclosed in the present invention.

[0071] Unless otherwise specifically noted, it will be apparent from the following discussion that throughout the specification, discussions utilizing terms such as "obtain," "apply," "generate," "provide," etc. refer to the actions and / or processes of a computer that manipulates data and / or transforms data into other data, where the data is represented as physical (such as electronic) quantities and / or where the data represents physical objects.

[0072] The terms "computer", "computer system", "computer device", "computerized device" and the like should be interpreted expansively to include any kind of hardware-based electronic device having one or more data processing circuits. Each processing circuit may include, for example, one or more processors operably connected to a computer memory loaded with executable instructions for performing operations, as further described below.

[0073] The one or more processors referred to herein may represent, for example, one or more general-purpose processing devices, such as a microprocessor, a central processing unit, etc. More specifically, a given processor may be one of the following: a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor implementing other instruction sets, or a processor implementing a combination of instruction sets. The one or more processors may also be one or more special-purpose processing devices, such as an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), a graphics processing unit (GPU), a network processor, etc.

[0074] The memory referred to in this document may include, for example, one or more of the following: internal memory, such as processor registers and cache, main memory, such as read-only memory (ROM), flash memory, dynamic random access memory (DRAM), such as synchronous DRAM (SDRAM) or Rambus DRAM (RDRAM), etc.

[0075] Some operations in accordance with the teachings herein may be performed by a computer device specially constructed for the desired purposes, or by a general-purpose computer specially configured for the desired purposes via a computer program stored in a computer-readable storage medium.

[0076] Figure 2 and Figure 7 An overall schematic block diagram of a system according to some examples of the presently disclosed subject matter is shown. Figure 2 and Figure 7 The different components in may be comprised of any combination of software, hardware and / or firmware that performs the functions as defined and explained herein. Figure 2 and Figure 7 The components in the system may be centralized in one location and / or device, or distributed across more than one location or device. Figure 2 and Figure 7 Computer system 220 and computer 210 are shown as two separate computers, but in other examples, the two computer systems can be integrated into one computer system. In various examples of the subject matter disclosed in the present invention, the system can include a computer system that is compatible with the computer system. Figure 2 and Figure 7 Fewer, more, and / or different modules than those shown.

[0077] Figure 3a , Figure 3b and Figure 6 is a flow chart illustrating operations performed according to an example of the subject matter disclosed herein. According to different examples of the subject matter disclosed herein, operations performed according to Figure 3a , Figure 3b ,and Figure 6 According to various examples of the subject matter disclosed herein, Figure 3a , Figure 3b ,and Figure 6 One or more of the stages shown in may be performed in a different order, and / or one or more groups of stages may be performed simultaneously for multiple commands.

[0078] It should be noted that any references in the specification to the left atrium are made by way of non-limiting example only, and the subject matter disclosed herein also contemplates applying the same principles in a similar manner to other sub-compartments of the heart, including, for example, the right atrium, left ventricle, right ventricle, and blood vessels connected to the heart, such as the pulmonary veins.

[0079] Reference now Figure 2 , showing a schematic block diagram of a computer system according to some examples of the subject matter disclosed herein. System 200 includes a computer 210 (also referred to as an "AF therapy recommendation and treatment computer system"), which is configured to analyze a cardiac anatomical map of an AF patient and provide a recommendation regarding a preferred method of PVI treatment. Figure 2 The computer system 210 shown may include at least one processing circuit 230, which includes one or more computer processors configured to execute several functional modules according to computer-readable instructions implemented on a non-transitory computer-readable storage medium. These functional modules are hereinafter referred to as being included in the processing circuit 230. It is worth noting that although the system 210 shows a single processing circuit, it can include more than one processing circuit in other ways, such as having different circuits including different functional modules.

[0080] An anatomical map (or model) of the heart may be generated by computer 210 or received from another computer system. An anatomical map of the heart may be generated using an anatomical image of the heart received from a suitable medical imaging (computer) system (220). Generating the anatomical map may involve, for example, using the Fast Anatomical Mapping (FAM) technology provided in the CARTO™ system produced by Biosense Webster Inc. (Irvine, California), or using any other suitable technology, or using any suitable combination of the above technologies. The cardiac map may include any suitable type of three-dimensional (3D) anatomical map produced using any suitable technology. The map may be derived from an intracardiac ECG (electrocardiogram) and may include position information of a catheter moved around the atria and ECG measurements (such as voltage at each location).

[0081] In some examples, the processing circuit 230 includes an automatic map segmentation module 221 that is configured to automatically segment the 3D map of the heart into its various sub-compartments, and further segment one or more of the sub-compartments into corresponding anatomical sites, such as the posterior and anterior walls, left lateral wall, top wall, inferior wall, septum, mitral valve, left atrial appendage (LAA), and four pulmonary veins (LSPV, LIPV, RSPV, RIPV). It is worth noting that while all of the anatomical sites listed above may be provided during segmentation, not all anatomical sites need be considered during the analysis and treatment recommendation process.

[0082] The processing circuit 230 may also include: a map dimension reduction module 223, which is configured to transform the 3D map into a 2D representation of the map; a heart subcompartment size calculator 225, which is configured to measure or calculate the size of various subcompartments of the heart, for example, various anatomical locations of the left atrium; a geometric parameter calculator 227, which is configured to calculate at least one parameter (also referred to herein as an "anatomical prediction parameter"); and a treatment recommendation module 229, which is configured to determine and provide a recommendation as to whether PVI-only treatment is likely to be successful if administered to the patient. The computer system 210 may also include a computer interface, which includes, for example, a keyboard and mouse and a computer display, where the data can be displayed to, for example, a healthcare provider using an appropriate graphical user interface program.

[0083] Go to Figure 3a , which is a flow chart of some examples of the subject matter disclosed herein, illustrating operations performed as part of the analysis and treatment recommendation process. For ease of understanding and by way of non-limiting example only, reference is made to the components shown in system 200 for description. Figure 3a Operations disclosed in .

[0084] The process described below can be applied to analyze the heart of a patient suffering from AF and determine whether a PVI-only procedure may be the right choice for treatment of the patient, for example, if it is administered to the patient, it may provide long-term success. According to one example, the term "long-term success" is intended to include a period of absence of AF and any arrhythmias ("AF- and arrhythmia-free period") after treatment, lasting 3 to 18 months. According to one specific example, long-term success refers to a period of absence of AF and arrhythmias of at least 3 months after treatment. According to another specific example, long-term success refers to a period of absence of AF and arrhythmias of at least 6 months after treatment. According to another specific example, long-term success refers to a period of absence of AF and arrhythmias of at least 9 months after treatment. According to another specific example, long-term success refers to a period of absence of AF and arrhythmias of at least 12 months after treatment. According to another specific example, long-term success refers to a period of absence of AF and arrhythmias of at least 15 months after treatment. According to another specific example, long-term success refers to a period of absence of AF and arrhythmias of at least 18 months after treatment.

[0085] The terms "improve," "augment," or "enhance" AF therapy refer to an AF therapy determined to result in longer AF- and arrhythmia-free periods compared to alternative AF therapies.

[0086] The term "PVI only" (or "PVI alone") refers to an ablation procedure applied only to the pulmonary veins of the left atrium, where PVI+ refers to additional ablation applied to additional anatomical sites of the heart. Additional sites targeted in PVI+ include, for example, one or more of the following:

[0087] Top line; left carina; right carina; posterior line; inferior line; mitral line; anterior line; LAA isolation; IVC-tricuspid isthmus; RAA isolation; IVC isolation; SVC isolation; stromal modification; stromal isolation; fragmentation; focal; rotor; LA_posterior wall; coronary sinus; RA_free wall; LA_anterior wall; LA_lateral wall; LA_inferior wall; RA_septum; RA_crista terminalis; inferior mitral line; and LA_septum.

[0088] At block 301, a 3D map of a patient's heart is obtained. As described above, the 3D map may be generated based on images of the heart received from a suitable imaging system (220). If the left atrium is being examined, the 3D map includes at least a portion of the heart including the left atrium.

[0089] At block 303, an automatic map segmentation process is applied to the 3D map to identify multiple anatomical locations of one or more sub-compartments of the heart (e.g., the left atrium). In some examples, the 3D map is received by an automatic map segmentation module 221, which is configured to perform an automatic segmentation process to identify different portions of the heart in the 3D map. In some examples, the 3D image received by computer 210 is transformed into a 3D map using, for example, appropriate software implemented as part of automatic map segmentation module 221.

[0090] In some examples, the automatic map segmentation process includes applying a machine learning module on the 3D map that is configured to automatically segment the heart or atria (and possibly other sub-compartments of the heart) and provide data indicative of different anatomical locations (or structures) of the heart or atria.

[0091] According to some examples, the automatic map segmentation process is the same process described in U.S. Patent Application No. 18 / 207854 (Publication No. US2023-0397950), entitled "SYSTEMS AND METHODS FOR RECOMMENDING ABLATION LINES," filed by the applicant on June 9, 2023 and published on December 14, 2023. Application No. 18 / 207854 claims priority to U.S. Provisional Patent Application No. 63 / 350,983, filed on June 10, 2022.

[0092] U.S. Patent Application No. 18 / 207,854 discloses an automatic map segmenter (22) configured to segment an anatomical map and output data identifying anatomical sites. The map segmenter utilizes a machine learning model (implementing a neural network and described in detail with reference to FIG. 3 in 18 / 207,854) that receives a 3D map of the heart as input and provides a code set indicating anatomical sites of the heart (e.g., anatomical sites of the atria) as output. The machine learning model is trained using a training data set that includes a sufficient number of paroxysmal and persistent AF cases (e.g., 5,000 or more cases) and anatomical maps for each case. A more detailed description of the training and execution of a machine learning model specifically for segmenting the heart is disclosed in 18 / 207,854.

[0093] In some examples, the automatic map segmenter 22 of the present invention is configured in the same manner as the automatic map segmenter 22 disclosed in 18 / 207854. Figure 2 In other examples, it may retain some, but not all, features of the automatic map segmenter 22. For example, the machine learning model described in 18 / 207854 is trained to provide both segmentation of the heart or left atrium and ablation lines. In some examples, the automatic map segmentation module 221 disclosed herein may be trained to provide segmentation of anatomical regions of the heart and / or left atrium, but not ablation lines.

[0094] At box 305, the 3D map (now segmented) undergoes dimensionality reduction to obtain a corresponding 2-dimensional (2D) representation. In some examples, these operations are performed by applying principal component analysis (PCA) that projects the 3D map onto a 2D plane, as is known in the art (e.g., by map dimensionality reduction module 223).

[0095] At block 307, different dimensions of the 2D representation of various anatomical regions of one or more subcompartments of the heart are measured to obtain corresponding dimension values. The measured dimensions include, but are not limited to, the height and width of the corresponding region.

[0096] Figure 4An image of the left atrium is shown on the left (a). A front view of the atrium with an anterior design line (indicated by dots) is shown in the middle (b), and its two-dimensional representation after PCA projection is shown on the right (c). In some examples, the anterior width (or height, when measured along the y-axis) is determined as the length in millimeters (mm) of the maximum point minus the minimum point along the x-axis (or y-axis). Mathematically, the width (or height) of an anatomical part (structure) is the length (mm) of the maximum point minus the minimum point along the x-axis (or y-axis) of the points in the 2D plane. This calculation assumes that each anatomical part of the heart has constant properties, which means that given two different dimensions of the atrium, one dimension is always greater than the second dimension. For example, it can be expected that the anterior length of the atrium is greater than its width.

[0097] In addition, according to some examples, the back wall dimensions are calculated based on four anchor points, such as Figure 5 shown.

[0098] - Anchor point 1 - is the junction between the LSPV, top wall and posterior wall.

[0099] - Anchor point 2 - is the junction between the RSPV, top wall and posterior wall.

[0100] - Anchor point 3 - is the junction between the RIPV, the inferior wall and the posterior wall.

[0101] - Anchor point 4 - is the junction between the LIPV, the inferior wall and the posterior wall.

[0102] Based on these anchor points, 8 types of left atrial dimension measurements were calculated:

[0103] a. Back Width 1 - Geodesic and Euclidean distances between anchor point 1 and anchor point 2.

[0104] b. Back width 2 - geodesic and Euclidean distances between anchor points 3 and 4.

[0105] c. Post-height 1 - Geodesic and Euclidean distances between anchor points 1 and 4.

[0106] d. Post-height 2 — Geodesic and Euclidean distances between anchor points 2 and 3.

[0107] At box 309, one or more parameters (also referred to herein as "anatomical prediction parameters") are determined that indicate the likelihood of long-term success (or long-term failure) of a PVI-only protocol administered to an AF patient. Examples of these parameters are listed in Table 1 in the "Examples" section below.

[0108] In addition to singular parameters ("singular anatomical prediction parameters" or "singular geometric parameters") determined based on measurements of a single dimension of a single anatomical site of the heart (e.g., RLAT width, RLAT length, POST width, POST height, etc.), additional parameters are determined based on two or more singular parameters (also referred to herein as "complex anatomical prediction parameters" or "complex geometric parameters"). This can be achieved, for example, by applying mathematical operations (e.g., multiplication) between pairs or triplets of singular parameters to obtain corresponding values. In some examples, the products of all singular value pairs are calculated to obtain the additional parameters. In some examples, the products of all singular value triplets are calculated (in addition to or in place of the products of the pairs) to obtain the additional parameters. Table 2 in the "Examples" section below shows a list of complex anatomical parameters generated by multiplying pairs of singular parameters. It is noteworthy that the subject matter disclosed in the present invention is not limited to pairs and triplets of singular values, and in some examples, a larger number of singular values ​​can be used to calculate complex anatomical prediction parameters.

[0109] The above calculations may be performed, for example, by a geometric parameter calculator 227 in the system 210, which may be configured to determine singular geometric parameters and calculate complex geometric parameters. It is worth noting that in the case where the anatomical prediction parameters are based on a single dimension, the dimension value determined in the previous stage may be used in a straightforward manner. However, if the anatomical prediction parameters are based on more than one dimension, corresponding calculations are performed.

[0110] In some examples, once one or more anatomical prediction parameters have been calculated, these parameters are compared to corresponding predefined values, for example, thresholds set to distinguish long-term success from long-term failure (box 311). For example, taking into account the width of the right lateral wall of the left atrium (RLAT), based on the results described below, the threshold indicating long-term success may be equal to a value above 42.6 mm in order to achieve a 75% sensitivity for detecting long-term failure. Based on the results of this comparison, the long-term success or long-term failure of the PVI-only procedure can be determined, and a corresponding recommendation can be generated, wherein in the case of predicted long-term success, PVI-only treatment is recommended, otherwise, PVI+ is recommended. It is worth noting that although Figure 3a Determination of predictions is described with respect to PVI only, but it is also contemplated that appropriate AF-related features and thresholds are used to predict long-term success of PVI+ as well. In some examples, a different set of anatomical prediction parameters are used to increase the robustness of the recommendations.

[0111] In some examples, the process continues by administering the appropriate treatment to the patient according to the recommendation (block 313). If long-term success of the PVI-only protocol is predicted, the patient is administered the PVI-only treatment, and if long-term failure of the PVI-only protocol is predicted, the patient is administered a different treatment (e.g., PVI+). Figure 8 An example of a computer system that combines the functionality of system 200 with additional functionality, including facilitating and guiding appropriate PVI treatment, is shown. In some examples, once a recommended treatment is determined, the system can recommend appropriate ablation lines, display these lines on the corresponding map, and guide the physician through the process.

[0112] The presently disclosed subject matter also contemplates a machine learning (ML) model specifically for selecting an appropriate ablation method (ie, PVI or PVI+) for AF patients. Figure 3b 1 is a flow chart of some examples of the subject matter disclosed herein, illustrating operations performed as part of a machine learning based analysis and treatment recommendation process. The model is trained to correlate between AF-related features of AF patients and recommended treatments.

[0113] During the training phase, the ML model is trained using a training data set that includes a plurality of features determined for a number of different AF patients (hereinafter referred to as "AF-related features"). According to one example, the model is trained to associate between various combinations of AF-related features and corresponding probabilities of success for a PVI-only procedure. In another example, the model is trained to associate between various combinations of AF-related features and corresponding probabilities of success for a PVI-only procedure and / or a PVI+ procedure.

[0114] A training data set of AF-related features of many different patients is obtained (block 301b). The AF-related features include at least one type of cardiac-related features related to cardiac parameters, which are properties that characterize the patient's heart. The AF-related features may also include other AF-related features that are not related to the heart.

[0115] One example of a cardiac-related feature is an anatomical prediction parameter (including both singular and complex geometry parameters, e.g., pairs and triplets). Additional examples of cardiac-related features that may be used during training and execution of the ML model include voltage-related features and cycle length.

[0116] Voltage-related AF features are based on voltage measurements obtained from a patient. Voltage measurements include, for example, measurements made at low voltage zones (LVZ) regions, low voltage splits, etc. Voltage measurements can be applied in various ways. For example, the LVZ can be measured at each anatomical site (e.g., posterior wall, anterior wall, left side wall, right side wall, top wall, and lower wall). Voltage measurements can include, for example, one or more of an absolute low voltage value measured over the entire area of ​​the corresponding anatomical site and a relative low voltage value representing the ratio between the measured low voltage and the voltage over the entire area of ​​the anatomical site (e.g., posterior wall).

[0117] Additional voltage-related features include global voltage-related parameters, such as different percentiles of the voltage distribution (voltage pattern) in a portion of the heart or in its entirety, which may be based on different thresholds, e.g., the 25th percentile, the 50th percentile, the 90th percentile, etc.

[0118] Voltage-related features can be measured, for example, using the CARTO system of the Biosense Webseter described above. FAM technology is designed to quickly create a high-resolution 3D anatomical map of a cardiac chamber of interest, such as the left atrium. This is achieved by moving a mapping catheter along the endocardial surface while continuously recording its position and corresponding electrical activity. FAM can be combined with voltage mapping to identify the LVZ within a rapidly acquired anatomical map. It is worth noting that in some examples, the global voltage can be used alone as a unique feature to determine PVI only or PVI+.

[0119] As described above, certain voltage measurements may be applied to multiple separate anatomical locations of the heart and / or atria, thereby obtaining multiple different voltage-related signatures from a single patient.

[0120] Cycle length refers to any type of measurement of the time interval between detected electrical activations. For example, activations measured on a CS (coronary sinus) catheter, HRA (high right atrium) catheter, locally placed catheter, or other catheter.

[0121] Non-cardiac AF-related features include, for example, demographic data of the patient, including various details characterizing the patient's personal information, such as gender, age, place of residence, etc.

[0122] Note that, although various examples of AF-related feature types are mentioned above, in different examples, only a subset of these features are used during the training and execution of the machine learning model. Note also that the examples of AF features disclosed above should not be construed as limiting, and other types of AF-related features are also contemplated to be within the scope of the present disclosure.

[0123] To obtain AF-related features for training the model, data from multiple AF patients (e.g., from thousands to tens of thousands of patients or more) are collected, and corresponding parameters are calculated from the data of each patient, which are then used as features in the training data set. AF-related features can be obtained, for example, during an atrial fibrillation ablation procedure. If available, a database of AF patients including parameters obtained during catheterization and / or ablation can be used.

[0124] The training data set is used to train the machine learning model (block 303b). Different types of classifiers may be generated. These classifiers include, for example: a classifier that provides data indicating the long-term probability of success of a pulmonary vein isolation (PVI)-only procedure (e.g., generated using a training data set that includes data obtained from AF patients treated with a PVI-only procedure), a classifier that provides data indicating the long-term probability of success of a PVI+procedure (e.g., generated using a training data set that includes data obtained from AF patients treated with a PVI+procedure), and a classifier that provides data indicating the long-term probability of success of both PVI-only and PVI+ (e.g., generated using a training data set that includes data obtained from a set of AF patients treated with either a PVI-only or a PVI+procedure).

[0125] According to some examples, the machine learning model is implemented as a supervised model, where patients (data points) in the training dataset are labeled according to their corresponding observed long-term success or long-term failure of the PVI-only protocol, representing the ground truth. The method requires valid information to be obtained from a sufficiently large group of AF patients, which indicates the observed long-term success or failure of the PVI-only protocol. In other examples, the data points in the training dataset are labeled according to their corresponding observed long-term success or long-term failure of the PVI-only protocol and / or the PVI+ protocol.

[0126] The training may be performed, for example, by computer 210 or by another computer remotely connected to computer 210 (e.g., by ML classifier training module 231, which may be configured to receive a training data set and train a model). The trained ML model may be stored in a computer data repository (235) and may be used to select an appropriate ablation method for an AF patient (block 305b).

[0127] The above approach can be implemented if such a group of AF patients is indeed available. However, it also creates a technical challenge, namely that if such information is not available, the implementation of such a machine learning model will be precluded. To overcome this technical disadvantage, this article proposes an alternative approach in which the machine learning model is trained based on decisions made by physicians who examine AF patients and recommend ablation methods (and possibly also administering the treatment) rather than actual observations of the treatment, and thus overcomes technical obstacles caused by insufficient training data. The alternative approach is based on assumptions consistent with the "wisdom of crowds" (WoC) principle. According to this principle, in large-scale situations (involving, for example, hundreds of thousands or millions of unrelated cases or more), the majority of individuals make the correct decision.

[0128] The use of physician recommendations in training ML models addresses the technical issue of the paucity of data related to actual outcomes of AF treatment and leverages expert clinical judgment as a reliable surrogate for outcome data, thereby facilitating the robust and efficient generation of ML models despite limited availability of outcome data.

[0129] In this context, the "individual" is the physician, and the current "decision" involves a recommendation whether to administer PVI-only or PVI+procedure, and in some examples, which type of PVI+procedure to administer. Thus, the ground truth here is the recorded medical recommendation rather than the observed success or failure, but it is believed to be a good predictor of success or failure that can be used to select an appropriate ablation method.

[0130] According to some examples, a patient database that meets certain requirements is used for this purpose. These requirements include, for example:

[0131] Size: The database needs to be large enough to provide a statistically feasible analysis, storing information about many different AF patients, such as thousands or tens of thousands of AF patients and many different physicians.

[0132] Diversity: The database includes many AF patients with various paroxysmal and PS-AF cases, characterized by different features. These features may include varying atrial shapes, different voltage patterns (i.e., the distribution of low voltages over different areas of the atria), and different ages and genders. Known statistical methods can be used to determine the diversity of the database.

[0133] Independent: The database contains many patients with similar AF characteristics, who are treated independently by different doctors at different medical institutions. The similarity between different AF patients in the database can be determined by applying a clustering mechanism to the data, such as Gaussian mixture model (GMM), spectral clustering, K-means, etc.

[0134] According to one approach, training can be based on actual opinions presented by a group of physicians to different AF patients to obtain the opinions of many physicians on each case. This approach can use binary classification or multi-class classification. This approach can be used if the data includes recommendations made by many physicians on specific "real" patients.

[0135] However, since such data are often not available, different approaches to overcome the data deficiencies have been proposed. Since the database is characterized by a high diversity of recorded cases on the one hand and similarities between cases on the other hand, it actually includes groups (or categories) of patients with similar AF characteristics, where each group can be considered to correspond to a certain AF patient ("virtual patient"). The inventors found that independent medical opinions made by many different physicians on similar cases (all assigned to the same group) can be regarded as made for the same AF patient. Therefore, for the purpose of training a machine learning model, the data can be regarded as if many different physicians had given medical opinions to the same patient. As described above, in large-scale situations, it can be assumed that the majority vote of the medical opinions made by the physicians for each group of patients may be the ideal choice for a particular patient.

[0136] According to the WoC approach, the machine learning model can be implemented as, for example, a supervised model, where patients in the training dataset are labeled according to the corresponding medical opinions made by different physicians recommending PVI-only or PVI+treatment, representing the ground truth.

[0137] Whether the first approach is used (probability of success is based on observed long-term success of PVI alone or PVI + treatment) or the second approach (WoC) (recommendation of treatment is based on treatment choices made by the physician), a training dataset is used to train a machine learning classifier to associate various combinations of AF-related features with corresponding predictions. The training dataset can be used to train a suitable classifier machine learning model, such as a support vector machine (SVM), random forest, or logistic regression. If the training dataset is large enough, a deep neural network can be used for classification, such as a convolutional network with dense layers.

[0138] The ML model may be trained to provide one or more training outputs. According to one example, the training ML model may be trained to provide a binary classification that provides a recommendation indicating which approach (PVI alone or PVI+) is more likely to be effective in treating AF in a particular patient.

[0139] When the ML model is implemented according to the WoC approach, according to one example, each patient is individually labeled according to the actual treatment administered to the patient or according to the recommended treatment, e.g., 0 for PVI only and 1 for PVI+.

[0140] In the case of training a multi-class classifier (including a PVI-only class and several PVI+ classes), each class corresponds to a different type of additional ablation lines (e.g., additional ablation lines in anatomical locations of the heart other than the left atrium pulmonary veins), which can be similarly labeled based on each patient's individual treatment and / or recommendations.

[0141] Based on the example where patients in a training dataset are clustered into groups (using the clustering mechanism described above) according to the similarities between their respective AF signatures, each group represents a type of patient (“virtual patient”) that can be labeled according to the collective distribution of decisions made by all physicians for different virtual patients.

[0142] Consider a simplified example, assuming that there are two classes of patients (virtual patients), each class corresponding to patients with common AF-related characteristics, and there are two physicians with the following recommendation distributions:

[0143]

[0144] According to the collective labeling method, the labeling of the patient is determined based on the combined information from all physicians (e.g., by calculating a weighted average). This collective labeling method utilizes the combined expertise of each physician to produce a more balanced and robust labeling of the patient population, reflecting a consensus or average decision.

[0145] According to some examples, the machine learning model is implemented as an ensemble of supervised machine learning classifiers, where each classifier in the ensemble represents a virtual physician that provides a corresponding recommendation. In some cases, the classifiers in the ensemble can all be the same type of ML model, while in other cases, they can be different types of models (e.g., random forests, XGBoost, etc.). The training set can be divided into groups (e.g., randomly), where a subset of the data points (patients) in the group is used to train each classifier. During inference, each classifier provides a corresponding recommendation, and the final recommendation is determined based on the majority vote principle.

[0146] If the training dataset is large enough and contains enough recommendations made by different physicians, laboratories, or other medical institutions (occupying multiple physicians), the training dataset can be grouped according to physician or laboratory. This allows the creation of separate ML models, each representing a "virtual physician" associated with a real-world physician or laboratory.

[0147] Figure 6 is a flow chart illustrating an example of training of multiple models according to the WoC method. It is worth noting that the same type of ensemble can be used using a cohort of patients that document the observed success or failure of PVI treatment, if such data is available.

[0148] In block 601, a training data set of AF patients is generated; each patient is associated with a corresponding AF-related feature and a PVI treatment recommendation. The patients are divided into subgroups, where in some examples, each subgroup includes patients with similar AF features (block 603). As explained above, in some examples, each subgroup includes patients for whom AF therapy has been recommended by a certain physician or laboratory.

[0149] The patients in each subgroup are labeled according to the observed recommendations made by the corresponding physician or laboratory (block 605). Individual or collective labeling methods can be implemented.

[0150] These subgroups are then used to train corresponding classifiers for each physician-related or laboratory-related patient subgroup (block 607). This results in multiple classifiers, each corresponding to the decision-making process of a particular physician or laboratory, effectively acting as a virtual physician providing predicted recommendations for patients.

[0151] This approach allows for personalized and context-specific predictions, as each model is fine-tuned based on the medical preferences and opinions of individual physicians or laboratories. By leveraging these specialized models, the system can enhance the accuracy and relevance of its recommendations, ultimately leading to better patient outcomes and more efficient healthcare delivery, while overcoming the aforementioned technical challenges caused by the paucity of data on completed AF treatments.

[0152] Furthermore, in both approaches (Observed Success and WoC), additional outputs of the model include data indicating the significance of different AF-related features in predicting the final ML output associated with the recommended treatment. In the context of the ML model, individual features or any combination of AF-related features (e.g., specific single geometric parameters, specific complex geometric parameters, combinations of two or more parameters of any type, voltage AF-related features, etc.) may be critical in determining the recommended treatment. The ML model disclosed herein can be used to identify the most significant one or more features when providing a recommendation. Providing this information to the physician will help the physician rationalize the recommendations made by the model, thereby further assisting the physician in making the final decision on how to treat the patient. This can be achieved by applying feature importance analysis. For example, in the case of using a decision tree or random forest model, the Gini index can be used to quantify the importance of a feature.

[0153] Back to Figure 3b , in the execution phase (inference), the trained ML model can be applied to determine the recommended treatment for a particular AF patient. It is worth noting that although the training phase and the execution phase are shown sequentially, as is known in the art, the two phases can be performed asynchronously.

[0154] During execution, AF-related features obtained for one or more AF patients are applied to the ML model to obtain a corresponding recommendation for each patient regarding whether to administer PVI-only or PVI+therapy to the patient.

[0155] The execution of the ML model can be used as a reference above Figure 3aAssuming that a certain AF patient is a candidate for an ablation approach, and the physician wishes to receive a recommendation as to which ablation approach is more suitable for the patient, the physician can use system 220 to apply the ML model for this purpose.

[0156] To this end, AF-related features of the patient are obtained (307b). AF-related features (e.g., cardiac-related features) can be obtained by applying an appropriate procedure to the patient specifically for measuring relevant values, and performing further processing on the measured values ​​if necessary. For example, anatomical prediction parameters can be obtained by applying the process described in blocks 301 to 309 in reference FIG. 3.

[0157] Multiple anatomical prediction parameters (which may be single or complex) may be calculated (e.g., by a geometric parameter calculator 227) to provide multiple AF-related features, which are recorded as part of an input dataset of AF-related features for the ML model.

[0158] Other types of AF-related features can be obtained by applying additional procedures to the patient or related data. For example, various types of voltage-related features can be obtained by performing different voltage measurements on the patient's heart and adding the obtained voltage values ​​to the input data set. Voltage and cycle length measurements can be performed using appropriate medical instrumentation, such as using an intracardiac electrocardiogram operably connected to system 200. For example, the patient's demographic data can be obtained from the patient's medical records.

[0159] The ML model is applied to an input data set of AF-related features obtained from the patient, and an ML model output is obtained (box 309b). An ML model output is provided, which includes a recommendation indicating whether to administer PVI or PVI+ to the patient (box 311b). As further explained above, in some examples, the ML model provides additional information indicating which additional treatment or treatments are the recommended treatments for the patient for the PVI+ recommendation. As explained above, in some examples, the ML model also provides information indicating which input features have the greatest contribution to the output of the model. In some examples, the process continues by administering an appropriate treatment to the patient according to the recommendation (box 313b).

[0160] Figure 7 Inference using an ensemble of "virtual physician" classifiers is schematically illustrated, wherein an ensemble of ML models including classifiers is used to generate treatment recommendations for AF patients. Each of the machine learning models in the ensemble is applied to the AF-related features of the patient to obtain a corresponding output. Once the outputs from all classifiers in the model are obtained, the majority vote principle is applied and the appropriate recommendation is determined accordingly.

[0161] In some examples, the above Figure 3b The operations described in blocks 309b and 311b in are performed by the treatment recommendation module 229. The treatment recommendation module 229 may include, for example, an ML classifier module configured to apply an ML classifier (or an ensemble of classifiers) to the AF-related features of the patient, as described above.

[0162] It should also be noted that U.S. patent application Ser. No. 18 / 207,854 describes (for example, reference Figure 2 3 ) for presenting a recommended ablation line on an anatomical structure (hereinafter referred to as the "presentation system"). In some examples, the system 200 disclosed herein can be incorporated as part of the presentation system, providing the presentation system with the ability to generate a recommendation as to whether PVI treatment alone is appropriate for the patient. In other examples, the system 200 can be implemented as a stand-alone system separate from the presentation system described in 18 / 207854. Regardless, in some examples, the system 200 and the presentation system can share certain components. For example, in 18 / 207854 Figure 2 The medical imaging system 20 and the automatic mapping segmentation device 22 shown in the figure correspond to the present application respectively. Figure 2 A medical imaging system 220 and an automatic mapping segmentation module 221 are provided.

[0163] Figure 8 is a schematic diagram of a computer system that combines the functionality of the system 200 disclosed herein and the deep learning system 10 disclosed in 18 / 207854 and 63 / 350,983. Figure 8 In the present application, reference numerals with a prime symbol indicate elements disclosed in 18 / 207854 and 63 / 350,983, while reference numerals without a prime symbol indicate corresponding elements disclosed in the present application. As explained in detail in 18 / 207854, Figure 8 The system in includes three parts: an ablation line trainer 12, which builds a trained deep learning unit for an ablation line recommender 13; an ablation line recommender 14, which uses the ablation line recommender 13 to generate ablation line recommendations; and an ablation line guide 16, which provides ablation line guidance during a procedure. The ablation line trainer 12 can receive an anatomical map of the heart on which the recommended ablation line can be presented, and trains a machine learning model utilized by the ablation line recommender 13 to output the recommended ablation line according to the patient's cardiac anatomy.

[0164] In some examples, the ablation line recommender 14 receives data indicating the recommended ablation method from the treatment recommendation module 229 and uses this information when recommending an ablation line, wherein the recommended ablation line is selected to accommodate the recommended ablation method. For example, in the case of PVI-only treatment, the recommended ablation is adapted to PVI-only treatment. The ablation line guide 16 provides ablation line guidance to the physician during the procedure. It can receive an anatomical map of the heart from the medical imaging system 20 during the procedure, and can include an automatic map segmenter 22 to indicate various elements of the heart during the procedure.

[0165] Example

[0166] Table 1 shows the results of a statistical analysis performed by the inventors, which supports the correlation between certain geometric properties of the (left) atrium and the prediction of long-term success or long-term failure of a PVI-only procedure. The data include 95 AF patients who underwent a PVI-only procedure. Long-term failure of the PVI-only procedure was defined based on a combination of 3 outcomes: 1) repeat ablation; 2) hospitalization for direct current cardioversion; or 3) AF-related hospitalization. Patients who developed any of the above three conditions within up to 12 months after PVI-only treatment were classified into the long-term failure subgroup. Based on the above definition, 16 of the 95 subjects exhibited long-term failure.

[0167] Table 1 shows basic atrial parameters (singular parameters) and their relationship to long-term success prediction. Table 1 (and Table 2) shows the P value indicating the difference between the long-term success value and the long-term failure value for each parameter. For example, the posterior wall height and posterior wall width based on geodesic distance are significant in distinguishing long-term success from long-term failure in subjects with persistent atrial fibrillation (PS-AF) based on PVI only.

[0168] Table 1

[0169]

[0170]

[0171] Geo=Geodesic

[0172] Euc = Euclid

[0173] ANT=Anterior wall

[0174] POST = Back wall

[0175] RLAT = right side wall or septum

[0176] LLAT = Left side wall

[0177] postWidth—back wall width

[0178] Different parameters (characteristics) of left atrial dilatation are listed in the left column of the table. Each parameter corresponds to a certain anatomical part of the heart and is calculated based on one or more dimensions of the anatomical part. The correlation between the values ​​of the different parameters (obtained before treatment) and the long-term success or failure of the PVI-only procedure was examined. The parameters listed include: anterior wall (ANT) length and width, posterior left atrial wall (POST) length and width, top wall length and width, right side wall (RLAT) length and width, inferior wall length and width, LLAT width and length.

[0179] Table 2 shows the complex anatomical prediction parameters generated by multiplying the singular parameter pairs .

[0180]

[0181]

[0182] Based on this study, the most prominent singular parameter was the right side wall width, which increased significantly (p value < 0.005) from 37.5 [34.3-42.6] mm in patients with a successful PVI procedure to 43.2 [39.2-47.4] mm in patients with a failed PVI procedure. Fig. 9 , which shows a comparison between RLAT width in patients who presented with long-term (LT) failure of a PVI-only procedure and those who presented with long-term success.

[0183] The inventors also found that for subjects who failed the PVI procedure, the height and width of the posterior wall increased, and the width of the top wall increased.By multiplying each pair of statistically significant parameters, the inventors obtained a new set of parameters that were more significant in distinguishing long-term failure from long-term success.

[0184] Experiments to validate WoC-based ML models

[0185] Experiments were conducted to validate the efficacy of the ML model trained using the WoC approach. A real cohort of AF patients who had received either PVI-only or PVI+ treatment was associated with their outcomes (specifically AF epochs and arrhythmia-free epochs). The ML model was then applied to the AF-related features of this cohort of patients, and the corresponding recommendations (PVI or PVI+) were obtained from the model. Based on the recommendations of the ML model and the actual treatments received by the patients, the patients were divided into four groups:

[0186] 1. Patients for whom the ML model recommended PVI and who received PVI.

[0187] 2. The ML model recommended PVI but patients received PVI+.

[0188] 3. The ML model recommended PVI+ and the patients who received PVI+ were treated.

[0189] 4. Patients for whom the ML model recommended PVI+ received PVI.

[0190] Fig.10 A graph showing the probability of survival of AF patients over time, defined as the AF and arrhythmia-free period, is presented. Table 3 below details the probability of each group of patients achieving a 1-year AF and arrhythmia-free period based on the observed outcomes.

[0191]

[0192] The analysis revealed that alignment between the ML model's recommendations and the actual treatment administered was associated with improved outcomes in terms of freedom from AF and arrhythmias at 1 year. Specifically:

[0193] Patients who underwent PVI and for whom the ML model recommended PVI showed the highest 1-year probability of AF without arrhythmia (70.5%).

[0194] Patients who received PVI+ but for whom the model recommended PVI had a slightly lower 1-year probability of AF being arrhythmia-free (66.1%).

[0195] ● PVI+ is recommended and patients treated with PVI+ showed a 1-year AF-free arrhythmia probability of 50.4%.

[0196] ● Patients who were recommended PVI+ but received PVI had only the lowest 1-year probability of AF and arrhythmia freedom (46.1%).

[0197] Comparison results :

[0198] Group 1 vs. Group 2: Patients who were recommended PVI and received PVI (Group 1) had a higher 1-year AF and arrhythmia-free probability (70.5%), compared to 66.1% for patients who were recommended PVI but received PVI+ (Group 2). This suggests that following the ML model's recommendation for PVI produces better outcomes than deviating from that recommendation and performing PVI+.

[0199] Group 3 vs. Group 4: Patients recommended PVI+ and received PVI+ (Group 3) had a higher 1-year AF and arrhythmia-free probability (50.4%) compared to 46.1% for patients recommended PVI+ but received PVI (Group 4). This suggests that adhering to the ML model's recommendation for PVI+ provides better outcomes than performing PVI.

[0200] like Fig.10These results, shown in the survival probability plots in Figure 4 and detailed in Table 3, indicate that the ML model's recommendations are strongly associated with treatment success. Adherence to the model's guidance can enhance clinical outcomes, emphasizing the critical importance of following the ML model's recommendations to achieve optimal clinical outcomes.

[0201] The various illustrative logic blocks, modules, and algorithmic steps described in conjunction with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or a combination of the two. In order to clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, and steps have been generally described above with respect to their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the entire system. The described functionality may be implemented in different ways for specific applications, but such specific implementation decisions should not be interpreted as causing any departure from the scope of the present disclosure.

[0202] It will also be understood that the system according to the present disclosure can be implemented at least in part on a suitably programmed computer. Likewise, the present disclosure contemplates a computer program readable by a computer for performing the method of the present disclosure. The present disclosure also contemplates a non-transitory computer-readable memory tangibly embodying an instruction program executable by a computer for performing the method of the present disclosure.

[0203] A person skilled in the art will readily appreciate that various modifications and changes may be applied to the examples of the invention as described above, without departing from the scope of the invention as defined by the appended claims.

Claims

1. A computer-implemented method of selecting a treatment for atrial fibrillation (AF), the method comprising: obtaining a 3D anatomical map of at least one subcompartment of the patient's heart; segmenting the anatomical (3D) map to obtain a plurality of anatomical locations of the at least one subcompartment; determining at least one corresponding dimension for at least one of the plurality of anatomical sites; determining one or more anatomical prediction parameters, wherein each anatomical prediction parameter is determined based on each anatomical prediction parameter calculated based on geometric data collected from an anatomical region of the heart of the patient; generating a prediction indicating a likelihood of long-term success of a pulmonary vein isolation (PVI)-only procedure and / or a likelihood of long-term success of a PVI+procedure applied to the heart of the patient based on the one or more anatomical prediction parameters; as well as Based on the prediction, a recommendation for AF treatment for the patient is generated, the recommendation selected from a pulmonary vein isolation (PVI) only protocol and a PVI+ protocol.

2. The method according to claim 1, wherein: Generating the prediction includes: Obtaining AF-related features of the patient, wherein the AF-related features include cardiac-related features characterizing various cardiac-related attributes of the patient, and the cardiac-related features include at least the following types: a. said one or more anatomical prediction parameters; and b. one or more voltage-related characteristics; applying a machine learning (ML) model to the AF-related features of the patient; wherein the ML model has been trained using a training dataset, the training dataset comprising a plurality of data points, each data point corresponding to an AF patient associated with an AF-related feature and an observed long-term success of an AF therapy administered to the patient, the AF therapy comprising at least only PVI, whereby the ML model is trained to provide data indicative of a recommended AF therapy for a given patient based on the AF-related features of the given patient during inference; and A machine learning model output is generated that indicates a recommended AF treatment selected from a group including at least PVI-only and PVI+treatment.

3. The method according to claim 2, wherein: The machine learning model output includes information indicating the respective significance of different AF-related features in determining the recommended AF treatment.

4. The method according to claim 2, wherein: The ML model is trained to provide data indicative of the probability of long-term success of a pulmonary vein isolation (PVI)-only procedure.

5. The method according to claim 2, wherein: The ML model was trained to provide data indicative of the probability of long-term success of a pulmonary vein isolation (PVI)+ procedure.

6. The method according to claim 2, wherein: The ML model was trained to provide data indicating the probability of long-term success of a pulmonary vein isolation (PVI)-only procedure and a PVI+procedure.

7. A method according to any one of the preceding claims, wherein: The ML output including a recommendation to apply the PVI+ protocol to the patient also includes information indicating the specific type of PVI+ protocol that is recommended.

8. The method of any of the preceding claims, comprising performing a procedure on the patient to obtain the heart-related characteristics.

9. The method according to any one of claims 2 to 8, wherein: The AF-related features also include one or more of: cycle length, patient's medical history, and patient's demographics.

10. The method of claim 1, further comprising comparing each of the one or more anatomical prediction parameters to a corresponding threshold value, and determining the prediction based on an output of the comparison.

11. A method according to any one of the preceding claims, comprising: The one or more anatomical prediction parameters are determined by applying mathematical operations between different dimensions of the anatomical site.

12. A computer program product, comprising a computer-readable storage medium storing an instruction program, wherein when the instruction program is read by a computer processor, the computer processor is caused to execute the method according to any one of claims 1 to 11.

13. A computer system comprising at least one processing circuit configured to perform the operations of the method of enhancing atrial fibrillation (AF) treatment according to any one of claims 1 to 11.

Images