Chagas disease, endemic to Latin America, affects millions and remains a major cause of cardiac morbidity and sudden death. ECG abnormalities can appear in early stages, making ECG a valuable non-invasive tool for diagnosis and risk assessment, especially in settings with limited access to serological tests.
CompleXformers present a novel methodology to distinguish patients with Chagas cardiomyopathy from unaffected individuals using ECG signal analysis. The approach extracts wavelet entropy (H) and statistical complexity (C) based on Jensen–Shannon divergence. These measures are derived from the relative energy distribution across wavelet scales, obtained via the continuous wavelet transform. For each ECG, H and C were computed from QRS complexes and aggregated at the patient level to obtain mean indicators of H and C. A training pipeline was implemented to extract features and assign diagnostic labels from individual ECG. To maximize data utilization, Random Forest classifiers were trained on subsets of records grouped by feature availability. This strategy achieved a performance score of 0.187 on the validation set, as determined by the official evaluation metric of the 2025 Challenge.
As part of future improvements, the methodology will be extended to jointly analyze QRS complexes and T waves, focusing on the frequency band from 2 to 30 Hz. Additional indicators will be incorporated, including the minimum, maximum, and standard deviation of H and C. Furthermore, alternative classification models will be explored to optimize performance.
These findings suggest that the proposed biomarkers have strong potential as effective tools for diagnosing Chagas cardiomyopathy, particularly in low-resource clinical settings.