02334nas a2200205 4500000000100000008004100001260002800042653001900070653001400089653001700103100001500120700001400135700001300149700001100162700001400173700001600187245008200203520181800285022002502103 2026 d c10/2026bIOP Publishing10aChagas disease10aScreening10achest x-rays1 aCardenas D1 aRibeiro E1 aSoares Q1 aDias F1 aKrieger J1 aGutierrez M00aExploring the potential of DL for Chagas disease screening using chest x-rays3 a

Objective:

Chagas disease remains an important public health problem, particularly in Latin America, where chronic infection frequently leads to cardiac involvement and associated structural changes. Recent advances in deep learning (DL) have enabled automated image analysis of chest X-ray (CXR) in several clinical applications, yet little is known about its applicability to Chagas disease. However, the potential of these methods for Chagas disease screening remains largely unexplored.

Approach:

In this study, we investigate the feasibility of DL models to identify patients with Chagas disease from CXR images. A private dataset of CXRs was used to train and evaluate several convolutional neural network architectures.

Main results:

Among the evaluated models, EfficientNet-B4 achieved the best overall performance metrics, with accuracy of 0.655 ± 0.026, precision of 0.653 ± 0.041, recall of 0.700 ± 0.157, F1-score of 0.664 ± 0.076, and AUC of 0.727 ± 0.015.

Significance:

These results indicate that CXRs contain subtle imaging features associated with chronic Chagas disease that can be partially captured by DL models, although the overall discriminative performance remains moderate. While CXR-based DL models are unlikely to serve as a standalone diagnostic tool, they may have potential as complementary screening or triage tools, particularly in settings where chest radiographs are routinely acquired and access to specialized diagnostic testing is limited.. These findings highlight both the feasibility and the limitations of applying artificial intelligence to support Chagas disease screening using routinely acquired radiographic images.

 a0967-3334, 1361-6579