Multifeature signal encoding for sLBBB detection via ECG-fingerprint

Abstract

Accurately diagnosing strict Left Bundle Branch Block (sLBBB) from ECGs is crucial for optimizing Cardiac Resynchronization Therapy (CRT) in heart failure patients, as sLBBB indicates beneficial ventricular dyssynchrony. However, current visual interpretation is often subjective due to subtle QRS morphology dif-ferences. We developed and validated a new deep learning model using the “ECG-Fingerprint,” a unique multidimensional ECG representation. This approach cap-tures morphology, temporal dynamics, and lead interrelationships beyond human perception. Our model achieved an AUC-ROC of 0.8331 and an AUC-PR of 0.8761, showing robust performance. It delivered high sensitivity (0.8667) and F1-Score (0.8387) for sLBBB detection, proving stable and generalizable. This work offers an objective and reproducible solution for sLBBB classification, address-ing a key clinical need. Our model can significantly improve CRT patient selec-tion, optimizing outcomes and resource allocation, pushing forward AI-assisted precision cardiology.