Título: Automated Detection of Chagas Disease via CatBoost
Autores: Rogério Rodrigues Ferreira Júnior, João Gabriel Soares Ferreira, João Samuel Maciel de Sales, Marcio Barros Oliveira de Souza, Gisela V. Clemente, Pedro Joás Freitas Lima & João Paulo do Vale Madeiro
Resumo: Chagas disease, caused by Trypanosoma cruzi, is a neglected tropical disease that continues to pose a significant health threat in Latin America and is increasingly found in non-endemic regions due to migration. The chronic cardiac form of the disease (chronic Chagasic cardiomyopathy) often develops silently, making early detection difficult and effective treatment challenging. In this work, a machine learning approach based on the CatBoost algorithm is proposed for the detection of Chagas disease using electrocardiogram signals. A large-scale dataset from multiple sources was collected and processed, enabling the extraction of 425 features encompassing demographic and advanced ECG waveform characteristics. Severe class imbalance was addressed through targeted evaluation metrics and segmentation by age. The CatBoost model demonstrated robust performance, especially for individuals aged equal to or younger than 45 years, achieving a ROC-AUC of 0.89 and balanced accuracy of 0.81. Performance decreased to ROC-AUC of 0.81 and balanced accuracy of 0.73 for older individuals ( > 45 years). Despite these favorable metrics, the model presented notable limitations, particularly a low Matthews Correlation Coefficient (approximately 0.18) and low precision in predicting the minority class due to severe class imbalance, leading to a substantial number of false positives. This highlights the need for further methodological refinements to improve minority class detection. These findings reinforce the clinical potential of AI-empowered ECG analysis for scalable Chagas screening, particularly in resource-constrained settings.
Palavras-chave: Chagas disease; ECG; CatBoost; classification.
Páginas: 8
Código DOI: 10.21528/CBIC2025-1189924
Artigo em PDF: CBIC_2025_paper1189924.pdf
Arquivo BibTeX:
CBIC_2025_1189924.bib