Detection of Chagas Disease Using Tensor Decomposition and Wavelet Scattering Transform of ECG Signals

Shivnarayan Patidar
National Institute of Technology Goa


Abstract

Early detection of Chagas disease is critical for en- abling timely medical intervention and preventing cardio- vascular complications ahead. Electrocardiogram (ECG) signals offer crucial insights into the disease's progres- sion, underscoring the need for advanced signal process- ing and machine learning-based diagnostic tools. This study proposes a novel framework for accurate and effi- cient identification of Chagas disease using standard 12- lead ECG signals. The methodology begins with signal preprocessing to ensure uniformity and reduce noise. Four categories of features are extracted: (a) statistical fea- tures derived from heart rate variability, (b) statistical fea- tures from all 12 ECG leads, and (c) wavelet scattering transform (WST) features computed from leads II, V1, and V3 and (d) patient metadata. These features are input to a RUSBoost classifier, designed to address class im- balance while performing binary classification. The pro- posed framework was evaluated on the PhysioNet/CinC Challenge 2025 training dataset and achieved a Challenge score of 0.266 on the hidden test set under the team name Medics.