From Anatomy to Risk: Fast Digital Twins for LAA Thrombosis Risk in Atrial Fibrillation

About the Project

Atrial fibrillation (AF) significantly increases the risk of stroke, yet current risk scores such as CHA₂DS₂ VASc rely solely on comorbidities and demographics, and ignore patient specific LAA anatomy and haemodynamics. However, growing evidence shows that LAA morphology (length, eccentricity, bending, lobes, trabeculae) and the resulting local flow stasis play a central role in thrombus formation. High fidelity fluid–structure interaction (FSI) models are capable of capturing these effects but remain too computationally demanding for clinical use.

This PhD proposes to transform FSI based thrombosis assessment by developing machine learning surrogate models trained on simulations generated from a uniquely rich parametric LAA model, enabling fast and generalisable prediction of haemodynamics across the full spectrum of anatomical variability. This work directly supports more accurate AF stroke risk stratification by incorporating personalised morphological features that CHA₂DS₂ VASc does not capture.

Aims

To build a machine learning–accelerated modelling framework that predicts thrombosis related LAA haemodynamics in seconds, enabling anatomy aware risk assessment for AF patients.

Objectives

1. Create a parametric FSI dataset spanning full anatomical variability
2. Develop machine learning surrogate models for rapid haemodynamic prediction
3. Quantify how anatomy shapes thrombosis risk