Heart mapping for cardiac ablation treatment process using artificial intelligence

Abstract

The heart mapping for cardiac ablation treatment process using five artificial intelligence sets with different functions and objectives. Starting from receiving atrial intracardiac electrogram data from the patient to identify the severity level of the complex fractionated atrial electrograms abnormalities (CFAEs Score) and patterns of complex fractionated atrial electrograms (CFAEs Pattern) and sending both results along with the data of the intracardiac electrogram signal measurement location to identify the location of the heart for the first ablation. Next, intracardiac electrogram signals are measured at the ablation location to collect Ablation distal (ABLd) and Ablation proximal (ABLp) signals. Then, bringing the two results of data, the data of the ablation location, and the Coronary Sinus Cycle Length (CS-CCL) value into the analysis and prediction process until the output is the parameter values for ablation. The process will loop until the output is that no abnormal locations are found in the heart.

Description

[0001]. , ., , . , , ,. , ., , . , . , - , , , ,. , ., .. , - ,. , . . , . . , , . . , . , - -- - . ' - . , . ,, - , . . . - . . . . . ,, . . , . , .. , .., . . . ., , ., ' ' . . , . . . . ,. . ' - , . . - . , , ,. . . . , . . . , . , , ,. ,. , . . . , , , - , ., , . , . . ,. , . . . . . , , , . ,. . , -, - .. , ., .. . . . , ' . . : : , , , , ,- . , . , : . , : . . , : - . , , . : . , . , : . , , . , , , . , . . , . ,, . , , . ,- , . . , . , ,- .., , . - . , ,,, .,. , , , . . -,.,- . , , , . , . -. -. , . . , , . , . , - , - , , . , . . , - . . - . , . , , ,- ., .. , - . . . , . . , , . . , . . , . . . , , . . . , . ., . . - , ., . - . , ., . . - , , . , ,. , .,, . , , - . - , . , - . , . ,, , , , , , , , , , . ,. . ., , , . ,, ' , ,- . . , , , . . . , - ,. . , .. - . . - . . - . . . , ,- .,. . ,, , . ., , , - . - , ' . , . . - . . . . - . . - , , . , . , , , . . . . , . . - - ... , , ^^^^^^^^" ^^^^ ^^^^^^^^^^^^^^,^^^^^ ^^ ^^^^^^^^^^^^^^ ^ ^^^ , , . ,, , , , , . , - - ., . . , . , . , , - . . . ,,, - . , . . . .. . . .

Claims

AMENDED CLAIMS received by the International Bureau on 25 October 2025 (25.10.2025)1. The heart mapping for cardiac ablation treatment process using artificial intelligence consists of 5 main processing units or artificial intelligence sets with different functions and objectives as follows:• Artificial intelligence set 1 (2) receives atrial intracardiac electrogram data from the patient (1), filters the signals (201), applies Bayesian statistical denoising (202), and processes them through a Convolutional Neural Network (CNN) comprising a convolution signal process (203) with not less than three operation blocks, each employing not less than a 3 X 1 filter / kernel configuration and not less than 8 and up to 512 sets of filters per block , with 2x 1 max-pooling, followed by additional convolutional processing with not less than 8 and up to 512 sets, and a deep learning network (204) comprising not less than two fully connected layers with dropout technique (234) to identify the severity level of complex fractionated atrial electrograms abnormalities (CFAEs Score) (3) classified into exactly six levels (0-5) based on medical specialist criteria.• Artificial intelligence set 2 (4) receives atrial intracardiac electrogram data from the patient (1), filters the signals (401), applies Bayesian statistical denoising (402), and processes them through a different Convolutional Neural Network (61, architecture with not less than two operation blocks(611) comprising a ID convolutional layer with not less than 8 and up to 512 filters, batch normalization layer, activation function layer, and max pooling layer, followed by Deep Learning section (404) comprising not less than two hidden layers (612) and softmax activation function (613) to classify complex fractionated atrial electrograms patterns (CFAEs Pattern) (5) into five specific patterns: No CFAE, CFAE-1, CFAE-2 (cycle length <120ms), CFAE-3 (cycle length >120ms), and Rapid Firing (duration <50ms, amplitude >0.15, <4 peaks).• Artificial intelligence set 3 (6) will receive the severity level of the complex fractionated atrial electrograms abnormalities (CFAEs Score) (3), the pattern of the complex fractionated atrial electrograms (CFAEs Pattern) (5), which are the outputs of artificial intelligence set 1 (2) and artificial intelligence set 2 (4), as well as the location of the intracardiac electrogram signal measurement (Location of Measurement) (7) from thedevice, which locates the location of abnormalities in the cardiac electrical system. The three groups of data will undergo normalization before being fed into a mathematical calculation process and a probability distribution function-based decision. The decision results the top 3 locations with the highest probability for Ablation (8).• Artificial intelligence set 4 (15) receives CFAEs Score and CFAEs Pattern from artificial intelligence sets 1 and 2, location data, and coronary sinus cycle length (CS-CCL), and consists of three regression-based deep learning networks (156) that calculate, analyze and predict three parameter values (power value, duration value, and contact force value (16)), wherein these parameter values are adjusted using reinforcement learning in conjunction with an Ablation Index value calculated according to the formula:where CF represents contact force in newtonmeters, P represents power output in watts, and t represents duration in seconds.• Artificial intelligence set 5 (22) receives post-ablation data and identifies the next cardiac ablation location using not less than two classification-based Deep Learning Network (DLN) models, wherein the first DLN model outputs a heart zone and the second DLN model identifies a specific location within that zone, and the process iterates between artificial intelligence sets 4 and 5 until no abnormal locations are found within the heart.

2. The heart mapping for cardiac ablation treatment process using artificial intelligence according to claim 1, wherein atrial intracardiac electrogram data from the patient (1) is received from the multipolar catheter and two sets of intracardiac electrogram signals (Ablation distal (ABLd) and Ablation proximal (ABLp) signals) (11) are received from the electrophysiology catheter.

3. The heart mapping for cardiac ablation treatment process using artificial intelligence according to claim 1, wherein the data of the intracardiac electrogram signal measurement location is measured from the device for finding the location of abnormalities in the cardiac electrical system (such as CARTO system).

4. The heart mapping for cardiac ablation treatment process using artificial intelligence according to claim 1, wherein Artificial intelligence set 1 (2) comprises a Convolutional Neural Network (CNN) with a convolution signal process (203) having not less than three operation blocks, each block configured with a 3 X 1 filter / kernel and not less than 8 and up to 512 sets of filters per block, and a 2x 1 max-pooling layer, comprising three operation blocks: Block 1 (231) configured with not less than 8 and up to 512 sets of filters, Block 2 (232) configured with not less than 16 and up to 512 sets of filters, Block 3 (233) configured with not less than 32 and up to 512 sets of filters, each employing 2x 1 max-pooling, followed by additional convolution processing using3x 1 filter / kemel with not less than 8 and up to 512 sets of filters, wherein signals are preprocessed using Bayesian statistical methods (202) for denoising.

5. The heart mapping for cardiac ablation treatment process using artificial intelligence according to claim 1, wherein Artificial intelligence set 2 (4) comprises signal filtering (401), Bayesian statistical denoising (402), and a Convolutional Neural Network (61) not less than two Operation Blocks (611) comprising a ID convolutional layer for feature extraction with not less than 8 and up to 512 filters, batch normalization layer, activation function layer with ReLU as main activation function, and max pooling layer, followed by Deep Learning section (404) comprising not less than two and up to 512 optimized hidden layers (612) and softmax activation function (613) for highest probability selection (405), wherein the five CFAEs patterns are: No CFAE (irregular signals lacking CFAE characteristics), CFAE-1 (indistinguishable cycle length with low voltage variations), CFAE-2 (cycle length <120 milliseconds with high voltage variations), CFAE-3 (cycle length >120 milliseconds), and Rapid Firing (duration <50 milliseconds, amplitude >0.15, <4 peaks, interval shorter than CS-CCL signal).

6. The heart mapping for cardiac ablation treatment process using artificial intelligence according to claim 1, wherein artificial intelligence set 3 (6) uses a Classification-Based Deep Learning Network (602) comprising not less than two hidden layers with softmax activation function (603) that calculates and distributes probability of each cardiac location out of 6 total locations and presents the top 3 heart locations (8) with highest probabilities for ablation.

7. The heart mapping for cardiac ablation treatment process using artificial intelligence according to claim 1 , wherein Artificial intelligence set 4 (15) uses reinforcement learning with an environment (157) that evaluates reward (159) based on the calculated ablation index compared to a reference value, assigning low reward if higher than reference and high reward if lower than or close to reference, wherein the Regression-based Deep Learning Networks (156) adjust parameters until optimal values with high reward are obtained.

8. The heart mapping for cardiac ablation treatment process using artificial intelligence according to claim 1, wherein Artificial intelligence set 5 (22) employs a hierarchical processing system where DLN network 1 classifies and decides on a cardiac zone, and a second DLN network set is determined according to the identified zone to classify and decide on a specific location within that zone, with output in one-hot encoded format indicating either a new or previously treated location.

9. The heart mapping for cardiac ablation treatment process using artificial intelligence according to any one of claims 1 to 8, wherein the data for training the artificial intelligence is obtained from analysis by expert physicians and actual treatment cases of expert physicians.

10. The heart mapping for cardiac ablation treatment process using artificial intelligence according to claim 1, wherein the CFAEs Score is classified into exactly six levels or equivalent categorical scales defined by medical or Al-based models, including: Score 0 (Low signal quality), Score 1 (Normal - No CFAE pattern and no abnormal patterns), Score 2 (Abnormal No Ablation - No CFAE pattern but abnormal patterns present), Score 3 (Borderline - Abnormal with less than 50% CFAE pattern present), Score 4 (Very Abnormal Ablation - Inconsistent CFAE present 50% to 80%), and Score 5 (Primary Treatment Target - consistent CFAE present 80% to 100%).

11. The heart mapping for cardiac ablation treatment process using artificial intelligence according to claim 1, wherein the artificial intelligence set 2 identifies CFAEs patterns including: No CFAE (normal signals lacking CFAE characteristics), CFAE-1 (indistinguishable cycle length with low voltage variations), CFAE-2 (cycle length less than 120 milliseconds with high voltage variations), CFAE-3 (cycle length more than 120 milliseconds), and Rapid Firing (duration less than 50 milliseconds, amplitude greater than 0.15, and less than 4 peaks with interval shorter than CS-CCL signal).

12. The heart mapping for cardiac ablation treatment process using artificial intelligence according to claim 1, wherein the performance of artificial intelligence sets 1, 2, 3, and 5 is measured using Fl -score equation or equivalent precision-recall-based performance metrics, while the performance of artificial intelligence set 4 is measured using the MSE (Mean Square Error) equation or equivalent regression error metrics such as MAE (Mean Absolute Error).

Images