Artificial intelligence trained with optical mapping to improve detection of cardiac arrhythmia sources

1 . A computer-implemented method for determining a location of a source of cardiac arrhythmia in a patient's heart, comprising at least the steps of:

a) receiving electrogram signals acquired from a region of the patients' heart using a first set of electrodes; and

b) applying a pre-trained artificial intelligence (AI) model to predict the location of the cardiac arrhythmia source by using the received signals;

wherein pre-training of the AI model comprises:

acquiring electrogram signals from explanted human hearts, said signals are generated by a second set of electrodes assembled into an electrode array that covers at least a part of the explanted human heart, and acquiring co-registered functional and/or structural imaging data in the part of the explanted human heart covered with the electrode array, wherein said functional and/or structural imaging data provide location of at least one source of cardiac arrhythmia;

processing said electrogram signals and functional and/or structural imaging data to learn characterizing features that will be used in the AI model;

assigning learned features characterizing electrogram signals generated from at least one electrode on the electrode array to corresponding features characterizing functional and/or structural imaging data generated adjacent to said at least one electrode; and

classifying features characterizing electrogram signals generated from at least one electrode on the electrode array where functional and/or structural imaging data were acquired as corresponding to a source or to a non-source.

2 . The method of claim 1 , wherein the source of cardiac arrhythmia is a driver of atrial fibrillation.

3 . The method of claim 1 , wherein the AI model comprises an AI algorithm selected from the following group: supervised machine learning binary and multiclass classification and regression algorithm, chosen from k-Nearest Neighbors model, Support Vector Machine model, Boosting algorithm, Logistic Regression, or Random Forest; neural network, chosen from fully-connected neural network, convolution neural network, or recurrent neural network; or a unsupervised clustering algorithm.

4 . The method of claim 1 , wherein during pre-training of the AI model and before assigning, the characterizing features are selected by a user for Machine Learning algorithms or automatically for neural networks or any unsupervised models.

5 . The method of claim 1 , wherein assigning occurs for features that characterize electrogram signals generated from 3*3 matrix of electrodes located on the electrode array.

6 . The method of claim 1 , wherein processing of electrogram signals and functional and/or structural imaging data comprises the following steps: generating Fourier transformed electrogram signals and Fourier transformed imaging signals, normalizing signals, band-pass filtering of signals.

7 . The method of claim 1 , wherein acquiring functional imaging data comprises optical mapping with voltage sensitive dyes, with calcium sensitive dyes, or with fluorescent proteins.

8 . The method of claim 1 , wherein acquiring structural imaging data comprises performing magnetic resonance imaging, x-ray computed tomography, optical computed tomography, ultrasound imaging.

9 . A method for providing a cardiac arrhythmia ablation treatment plan, comprising:

a) receiving electrogram signals acquired from a region of the patients' heart using a first set of electrodes; and

b) applying a pre-trained artificial intelligence (AI) model to predict the location of the cardiac arrhythmia source by using the received signals;

wherein pre-training of the AI model comprises:

acquiring electrogram signals from explanted human hearts, said signals are generated by a second set of electrodes assembled into an electrode array that covers at least a part of the explanted human heart, and acquiring co-registered functional and/or structural imaging data in the part of the explanted human heart covered with the electrode array, wherein said functional and/or structural imaging data provide location of at least one source of cardiac arrhythmia;

processing said electrogram signals and functional and/or structural imaging data to learn characterizing features that will be used in the AI model;

assigning learned features characterizing electrogram signals generated from at least one electrode on the electrode array to corresponding features characterizing functional and/or structural imaging data generated adjacent to said at least one electrode;

classifying features characterizing electrogram signals generated from at least one electrode on the electrode array where functional and/or structural imaging data were acquired as corresponding to a source or to a non-source; and

providing a cardiac arrhythmia ablation treatment plan that includes an ablation of the located source as at least a portion of said cardiac arrhythmia treatment plan.

10 . The method of claim 1 , wherein acquiring electrogram signals from explanted human hearts comprises simultaneously acquiring electrogram signals and co-registered functional imaging data comprising near-infrared optical mapping from the same explanted human heart.

11 . The method of claim 9 , wherein acquiring electrogram signals from explanted human hearts comprises simultaneously acquiring electrogram signals and co-registered functional imaging data comprising near-infrared optical mapping from the same explanted human heart.