CONCORDAT: Trustworthy multimodal federated learning for autism research under incomplete, heterogeneous and privacy-constrained health data
Abstract
Abstract Multimodal brain-health research combines imaging-derived measurements with demographic, phenotypic and behavioural information, but these data are often incomplete, heterogeneous across sites, and subject to governance constraints that prevent subject-level pooling. CONCORDAT is a cross-silo federated framework for autism spectrum disorder (ASD) research that integrates these complementary feature blocks while keeping subject-level records at their source institution. Across the 20-site ABIDE-I cohort (n = 1035), adding demographic, phenotypic and Social Responsiveness Scale blocks to imaging-derived quality-control features improved 5-fold, site-stratified AUC by approximately 0.04 (0.610 to 0.648), and federated training reproduced the corresponding centralised results to within 0.001 AUC. We then stress-tested this multimodal pipeline under real acquisition incompleteness, comparing eight missing-modality scenarios; balanced accuracy fell despite nearly unchanged AUC when phenotypic information was unavailable, showing why multimodal deployment must monitor operating-point performance rather than discrimination alone. We further quantify site-harmonisation choices under different disclosure models, a client-level differential-privacy sweep (ε ∈ {1, 2, 5, 10, ∞}), clip-norm sensitivity, and a fairness intervention that raised female-ASD sensitivity by 0.532 (95% CI 0.419–0.661) for a 0.039 loss in balanced accuracy (95% CI 0.009–0.064). An independent…
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.