Elucidating 3D Constructs of Reentrant Circuits via a Novel Noninvasive Hybrid-AI System

Rochester Institute Of Technology, Rochester NY

Investigators

Linked publications & trials

Abstract

Project Summary Scar-mediated ventricular tachycardia (VT) is responsible for over 350,000 sudden cardiac deaths each year in the US. It is often caused by a reentrant circuit, maintained by narrow channels of surviving tissue (critical isthmus) protected within fibrotic tissue in the heart. While catheter ablation is used to treat VT by ablating the critical isthmus, its current decision support is based on a 2D interpretation of the reentrant circuit on one ventricular surface. This oversimplifies the 3D morphology of a reentrant circuit across the ventricular wall, leading to suboptimal ablation strategies and leaving a critical void of noninvasive decision support for ablation targeting that elucidates the 3D construct of a reentrant circuit beneath surfaces. A pivotal barrier to progress is the minute signal of reentrant circuit compared to its subsequent ventricular activity as manifested on body- surface electrocardiograms (ECGs), rendering it an easily-missed needle in a haystack. This proposal aims to fill this void by a hybrid neural-physics AI system that models and identifies the reentrant circuit as an unknown cause (modeled by neural networks) of a well-known effect (modeled by known physiology). It builds on significant progress from our last award, especially preliminary data on 1) the feasibility of extracting from ECGs signatures about the intramural depth of a reentrant circuit, 2) a hybrid-AI foundation prototyped in various complex systems including the heart, and 3) a unique multi-site capability to create an unprecedented in-silico and in-vivo dataset about 3D reentrant circuits for training and testing the AI system. Our long-term goal is to deliver a one-of-its-kind hybrid-AI system that can easily fit into the clinical workflow of scar-mediated VT ablation to inform clinicians with possible 3D constructs of an observed clinical VT. Our specific aims are: 1) To develop and evaluate hybrid-AI for elucidating intramural exit sites in stable VTs. 2) To develop and evaluate hybrid-AI for elucidating 3D critical isthmus in stable VTs. 3) Retrospective internal and external clinical evaluation in stable and unstable VTs. The multidisciplinary investigating team includes a technical core with expertise in AI and virtual-heart simulation, and clinical core with expertise in imaging as it relates to arrhythmia and VT ablation, endo-epi mapping of 3D VT constructs, and AI adoption in VT ablation. The outcome of this research will improve clinical practice of VT ablation by an unprecedented decision support to simplify procedural planning, shorten procedural time, reduce trial-and-errors, and improve procedural outcomes. The in-silico and in-vivo database generated will provide unparalleled capabilities for scientific discovery about the mechanisms of scar-mediated VT. Finally, the proposed hybrid-AI system will advance the AI foundation to be less demanding on data annotation while being more interpretable and trustworthy, contributing to better AI adoption in medical practice.