Abstract
Chagas disease is a neglected tropical infection caused by Trypanosoma cruzi, affecting approximately eight million people. Although serology is the diagnostic standard, its cost limits population-wide screening. The inexpensive 12-lead electrocardiogram (ECG) offers a promising alternative, but existing deep learning approaches typically rely on centralized data pooling, which may conflict with privacy regulations. We present a controlled benchmark of centralized and federated learning for ECG-based Chagas disease detection, progressively partitioning patient-level data across clients while keeping the architecture, preprocessing, loss, optimizer, and basic hyperparameters fixed while using predefined centralized and federated training schedules. On CODE-15%, federated learning closely matches centralized performance with two clients (AUPRC = 0.150, 7.6× prevalence lift), but degrades with increasing fragmentation: at 100 clients, AUROC falls from 0.836 to 0.664 and the precision–recall lift to 1.7×. The associated increase in false positives depends on the selected operating threshold. The results are consistent with reduced exposure to positive examples contributing to degradation, although the present design does not isolate this mechanism from other effects of increased client fragmentation. Source code: https://github.com/tail-unica/federated-learning-chagas.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/HemaRAI_005.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=R41ObpCHt7
BibTex
@InProceedings{BosThi_Comparing_MICCAISAT2026,
author = { Bossy, Thierry AND Loddo, Andrea AND Marras, Mirko AND Pala, Vittoria AND Perniciano, Alessandra},
title = { { Comparing Centralized and Federated Learning for ECG-Based Chagas Disease Detection } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17252},
month = {pending},
page = {pending}
}