Atrial fibrillation (AFib) is a prevalent cardiac arrhythmia with severe complications that impact millions of individuals globally. Current detection and diagnosis methods depend on expert interpretation of standardized ECG recordings taken during paroxysmal AFib episodes and risk scores from patients' electronic health records (EHRs), offering limited accuracy.

This project aims to develop novel approaches to predict the time-to-onset of AFib accurately. It will generate patient trajectories using a hybrid AI approach that incorporates features derived from a singular ECG or multiple follow-up ECGs, the patient's EHR, biomedical knowledge, biomarkers, and mechanistic simulation models of the heart. This exceeds the current state-of-the-art for AFib
prediction which is limited to binary classification over a pre-specified time horizon.

The generated trajectories will be extended with new interpretability techniques and uncertainty quantification to identify precursors of future AFib, whose shapelets are unknown in the medical domain.

In summary, this project seeks to develop innovative solutions for detecting, screening, providing personalized treatment, and remote monitoring of AFib and other heart arrhythmias in a medical context by fusing new, advanced machine learning techniques with expert medical knowledge to improve patient outcomes.