Wavelet-Derived Entropy and Complexity Biomarkers for ECG-Based Detection of Chagasic Cardiomyopathy

Abstract

As part of the Detection of Chagas Disease from the ECG: The George B. Moody PhysioNet Challenge 2025, our team CompleXformers developed an interpretable ECG-based approach to identify Chagasic cardiomyopathy (CCM) from standard 12-lead ECGs. The method constructs a principal-component beat (PCB) through R-peak alignment and singular-value decomposition, followed by a wavelet-based entropy–complexity analysis using the Daubechies-6 wavelet. From the relative wavelet energy distribution, two biomarkers are computed: the normalized Shannon entropy (H) and the wavelet statistical complexity (C), which together characterize morphological and spectral variability in the PCB. These features are used to train a Random-Forest classifier for binary CCM detection.Our approach achieved an official Challenge score of 0.119 (rank 36) on the hidden test set. Five-fold cross validation on the public training data confirmed stable entropy–complexity patterns across folds, demonstrating low sensitivity to alignment errors and noise. The proposed pipeline emphasizes physiological interpretability, reproducibility, and low computational cost, facilitating its integration into diagnostic workflows in resource-limited environments.