CardioTwin - Precision Cardiology based on Digital Twins

Wider research context / theoretical framework: The treatment of cardiovascular diseases relies increasingly on device-based therapies that need to be further improved to enhance efficacy by tailoring these to the pathophysiology of a given patient. The grand challenge in achieving such personalized precision therapies is to identify optimal device and deployment options. Physics law based virtual-heart technology that replicates a patient’s anatomy and electrophysiology (EP), referred to as digital twin model, shows high promise as a transformative tool addressing this challenge. However, fitting of parameters to patient data and generating reliable predictions remains challenging.

Hypotheses / research questions / objectives: The overall objective is to develop uncertainty-aware virtual-heart technology for the automated generation of mechanistic high fidelity ventricular digital twins (VDTs) of human ventricular EP from routine non-invasive clinical data. VDTs will incorporate personalized representations of the ventricular conduction system enabling these to replicate electrocardiograms (ECGs) during intrinsic activation, and to predict ECGs and electrograms (EGMs) in response to pacing therapies.

Approach / methods: Developments include an automated workflow for the generation of VDTs, a real-time in silico forward EP model for ECG and EGM computations, a multi-fidelity uncertainty quantification (UQ) framework to quantify the impact of observational uncertainty, and identification techniques for inferring probabilistic EP model parameters. These will be applied to calibrate virtual cohorts of healthy subjects, patients undergoing ablation of infarct-mediated ventricular tachycardia (VT) or cardiac resynchronization therapy (CRT). Calibrated models will be validated, and their predictive power evaluated. UQ will provide estimates of confidence in predictions.

Level of originality / innovation: The main innovation of the project is the development of a systematic and automated approach for generating EP VDTs from clinical data. Importantly, the project proposes a paradigmatic shift from pointwise parameters estimates towards probabilistic modeling approaches. Demonstrating the ability to personalize EP function and to predict therapeutic responses with statistical bounds will boost credibility and expand the scope of modeling toward a precision cardiology tool. VDTs will also play a pivotal role in transforming medical device development in industry, as devices can be developed faster at cheaper costs.

Primary researchers involved: The projects build on the expertise of Dr. Plank in modeling cardiac EP and in building workflows for generating VDTs from clinical data, and Prof. Krause and Dr. Pezzuto in parameter identification techniques for cardiac EP problems, multi-fidelity methods and UQ. The development of the core parameter inference technologies is based on the expertise of Dr. Pock in probabilistic modeling, optimization, and machine learning.