Federated Fine-tuning of SAM-Med3D for MRI-based Dementia Classification
This provides practical insights for deploying foundation models in decentralized clinical settings, though it's an incremental benchmarking study.
This study systematically evaluated how classification head architecture, fine-tuning strategy, and aggregation method affect federated learning performance for dementia diagnosis using brain MRI data, finding that classification head design substantially influences results, freezing the encoder matches full fine-tuning, and advanced aggregation outperforms standard methods.
While foundation models (FMs) offer strong potential for AI-based dementia diagnosis, their integration into federated learning (FL) systems remains underexplored. In this benchmarking study, we systematically evaluate the impact of key design choices: classification head architecture, fine-tuning strategy, and aggregation method, on the performance and efficiency of federated FM tuning using brain MRI data. Using a large multi-cohort dataset, we find that the architecture of the classification head substantially influences performance, freezing the FM encoder achieves comparable results to full fine-tuning, and advanced aggregation methods outperform standard federated averaging. Our results offer practical insights for deploying FMs in decentralized clinical settings and highlight trade-offs that should guide future method development.