Predicting atrial fibrillation recurrence after pulmonary vein isolation using simulations of patient-specific magnetic resonance imaging models and machine learning

A computer-implemented method, comprising:

• receiving clinical late gadolinium enhanced cardiac magnetic resonance imaging (LGE-MRI) images of at least one first patient;
• performing segmentation of epicardial and endocardial surfaces on the LGE-MRI images to create a three-dimensional (3D) model for the at least one first patient;
• identifying normal tissue regions and atrial fibrosis regions in the 3D model;
• assigning first cell and tissue properties to the normal tissue regions according to a model of human cardiac myocyte and based on conduction velocities;
• assigning second cell and tissue properties to the atrial fibrosis regions according to a model of diffuse fibrosis;
• performing simulations on the normal tissue regions and the atrial fibrosis regions, based on the first cell and tissue properties and the second cell and tissue properties, to generate simulation results;
• extracting first features from the simulation results based on the 3D model and based on a simulation protocol;
• processing, by the device, at least the first features, with a machine learning model, to select a feature that is predictive of atrial fibrillation recurrence;
• and
• utilizing, by the device, the feature to augment the simulation protocol for a second patient.

A system comprising:

• one or more processors;
• and
• one or more non-transitive computer readable media, communicatively coupled to the one or more processors, and storing instructions that, when executed by the one or more processors, configure the one or more processors to perform operations comprising:
• receiving clinical late gadolinium enhanced cardiac magnetic resonance imaging (LGE-MRI) images of at least one first patient;
• performing segmentation of epicardial and endocardial surfaces on the LGE-MRI images to create a three-dimensional (3D) model for the at least one first patient;
• identifying normal tissue regions and atrial fibrosis regions in the 3D model;
• assigning first cell and tissue properties to the normal tissue regions according to a model of human cardiac myocyte and based on conduction velocities;
• assigning second cell and tissue properties to the atrial fibrosis regions according to a model of diffuse fibrosis;
• performing simulations on the normal tissue regions and the atrial fibrosis regions, based on the first cell and tissue properties and the second cell and tissue properties, to generate simulation results;
• extracting first features from the simulation results based on the 3D model and based on a simulation protocol;
• processing, by the device, at least the first features, with a machine learning model, to select a feature that is predictive of atrial fibrillation recurrence;
• and
• utilizing, by the device, the feature to augment the simulation protocol for a second patient.