FSF3A: Federated Spatial Feature Alignment and Adaptive Aggregation for Heterogeneous Brain Tumor Segmentation


Peketi Divya (Indian Institute of Technology Hyderabad), C Krishna Mohan (Indian Institute of Technology Hyderabad), Sobhan Babu (Indian Institute of Technology Hyderabad), Sumanth Yenduri (Sam Houston State University)
The 35th British Machine Vision Conference
PDFPosterVideo (Right click to download)Supplementary

Abstract

Deep learning has the potential to identify complex patterns in brain tumors, but privacy concerns hinder centralized approaches. Federated learning (FL) addresses privacy concerns by enabling collaborative model training without exchanging local client data. However, managing data heterogeneity in FL is challenging due to the varying distributions and data scales at the local clients. To address these problems, we propose FSF3A, a novel federated framework with two approaches: \textbf{(i)} \textit{client localized structural loss (CLSL)}, a local training loss function that can handle varying data distributions by effectively finding a trade-off between spatial localization of local data, using modified dice loss, and the similarity between model representations to correct the local training of individual clients through our localized contrastive penalty and \textbf{(ii)} \textit{adaptive scale-aware weighted aggregation (ASWA)} method which facilitates weighted aggregation at the global model, prioritizing clients based on the data scales to ensure fairness in representation. The efficacy of FSF3A is demonstrated through comprehensive experiments on UCSF PDGM, UPenn-GBM, and MSD datasets for brain tumor segmentation.

Citation

@inproceedings{Divya_2025_BMVC,
author    = {Peketi Divya and C Krishna Mohan and Sobhan Babu and Sumanth Yenduri},
title     = {FSF3A: Federated Spatial Feature Alignment and Adaptive Aggregation for Heterogeneous Brain Tumor Segmentation},
booktitle = {36th British Machine Vision Conference 2025, {BMVC} 2025, Sheffield, UK, November 24-27, 2025},
publisher = {BMVA},
year      = {2025},
url       = {https://bmva-archive.org.uk/bmvc/2025/assets/papers/Paper_461/paper.pdf}
}


Copyright © 2025 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection