ECG-Based Screening of Chagas Disease Using Deep Residual Networks and Feature-Based Machine Learning

Chagas disease affects millions worldwide and can progress to chronic cardiomyopathy. Early detection is essential, but access to serological tests remains limited in endemic regions. Since Chagas-related abnormalities can be detectable on electrocardiograms (ECGs), we developed two automated approaches for disease screening in the framework of the Detection of Chagas Disease from the ECG: The George B. Moody PhysioNet Challenge 2025. The first is a deep learning method based on a residual neural network (ResNet) applied directly to ECG waveforms. The second is a machine learning pipeline that extracts fiducial features (intervals, slopes, and amplitudes) and classifies them using Gradient Boosting. On internal cross-validation, the ResNet achieved a score of 0.723 and the machine learning pipeline 0.486. In the official phase, the ResNet achieved 0.061, while the ML pipeline achieved 0.088; on the final test set, the ML pipeline achieved 0.076, placing team Chaguys 40th overall. These results illustrate the challenges of robust ECG-based Chagas screening, as the official and test phases evaluate true generalization from diverse, strongly labeled clinical sources.