Predicting Atrial Fibrillation Treatment Outcome with Siamese Multi-modal Fusion and Digital Twins

Alexander M. Zolotarev, Abbas Khan Rayabat Khan, Gregory Slabaugh, Caroline Roney

Keywords: Electrophysiology, Digital twins, Multi-modal fusion, Personalized medicine

Abstract:

Atrial fibrillation, the most common heart rhythm disorder, presents challenges in treatment due to difficulty pinpointing the patient-specific regions of abnormal electrical activity.
While biophysical simulations of cardiac electrophysiology create a digital twin of atrial electrical activity based on CT or MRI scans, testing various treatment strategies on them is time-consuming and impractical on clinical timescales.
Our proposed pipeline, incorporating Siamese architecture, fuses latent representations of multi-modal features extracted from atrial digital twin before any therapy and predicts the outcomes of several treatment strategies.
A large in-silico dataset of 1000 virtual patients, generated from clinical data, was utilized to provide the biophysical simulations before (used for feature extraction) and after (used for calculating ground truth labels depending on whether atrial fibrillation terminates or not) various treatment strategies.
By accurately predicting freedom from atrial fibrillation, our pipeline paves the way for personalized atrial fibrillation therapy with a fast and precise selection of optimal treatments.

Submission Number: 241