Abstract
Training AI across the brain-health data ecosystem is limited less by algorithms than by infrastructure: scans are sensitive, sit behind institutional and jurisdictional barriers, and cannot easily be pooled. We present NeuroFL, a production federated-learning (FL) platform that brings model training to brain-health data rather than moving the data. Built on the Flower AI framework, NeuroFL connects hospitals, universities, and research centres through a hub-and-spoke design with a human-in-the-loop approval gate, a containerized microservice backend, a six-phase orchestration flow, a three-layer security and governance model (physical data isolation; TLS with per-client authentication; role-scoped access and audit), and web interfaces for modelers and data sites. Because heterogeneous sites contribute unequally to a global model, NeuroFL exposes a graphical aggregation interface and uses client-contribution scoring, GTG-Shapley and reputation scoring, for fair, quality-weighted aggregation, having also evaluated Shapley values and cosine-similarity alignment as alternatives. We demonstrate the platform end-to-end on an open, reproducible brain-age benchmark built from two public structural-MRI cohorts treated as independent sites. Trained without any voxel data leaving its site, the federated model reaches a test mean-absolute-error (MAE) of 17.0 years (R^2 ≈ 0.21), substantially outperforming both single-site models (MAE 21.6 and 39.8) and recovering much of the centralized upper bound (11.3). Finally, we outline the clinical research pilots being initiated and the planned platform roadmap, differential privacy and secure aggregation, vertical and multi-modal federation, and heterogeneity-aware aggregation. NeuroFL is a step toward collaborative brain-health AI that preserves data sovereignty.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/DeCaF_016.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=8oIkZjcNej
BibTex
@InProceedings{TagShi_NeuroFL_MICCAISAT2026,
author = { Taghian Alamooti, Shirin AND Demir, Ali Cagri AND Sinhmar, Kajal AND Jeanson, Francis},
title = { { NeuroFL: A Production Federated Learning Platform for Privacy-Preserving Brain-Health Research } },
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}
}