Automated detection of Chagas disease from ECG signals using wavelet scattering transform and RUSBoost classifier.

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Early diagnosis of Chagas disease is crucial for effective treatment and mitigating severe cardiovascular complications.
• Electrocardiogram (ECG) signals offer valuable insights into cardiac health and disease progression.
• Advanced signal processing and machine intelligence can enhance the accuracy of Chagas disease diagnosis.

Purpose of the Study:

• To develop an automated system for early Chagas disease detection using 12-lead ECG recordings.
• To leverage machine learning and signal processing techniques for accurate classification of Chagas disease.
• To evaluate the system's performance on a benchmark dataset for clinical applicability.

Main Methods:

• Preprocessing of ECG signals, including standardization and QRS complex detection.
• Feature extraction using wavelet scattering transform (WST), heart rate variability (HRV) statistical descriptors, and patient metadata.
• Classification using the RUSBoost algorithm to handle imbalanced data for binary Chagas vs. non-Chagas classification.

Main Results:

• The proposed framework achieved an accuracy of 90.53% on the hidden test set of the PhysioNet/CinC Challenge 2025 dataset.
• Performance metrics included F1 Chagas = 10.73%, AUROC = 63.67%, AUPRC = 11.96%, and a Challenge Score of 20.5%.

Conclusions:

• The study demonstrates the potential of ECG analysis combined with signal processing and machine learning for scalable, non-invasive, and cost-effective early detection of Chagas disease.
• The findings support improved clinical decision-making and the development of preventive healthcare strategies for Chagas disease.
• This automated system offers a promising tool for enhancing diagnostic capabilities in resource-limited settings.