FedSynthCT-Brain: A Federated Learning Framework for Multi-Institutional Brain MRI-to-CT Synthesis

2024-12-09

Computers in Biology and Medicine

AI Summary

Multi-center brain MRI-to-CT synthesis suffers from poor generalizability of single-center models and risks of cross-institutional data privacy leakage. Method: We propose a privacy-preserving federated learning framework that integrates, for the first time, a 3D conditional generative adversarial network (cGAN) with a hierarchical federated aggregation mechanism. It incorporates adaptive gradient clipping and a cross-institutional modality alignment loss to enable collaborative training across heterogeneous medical institutions without sharing raw data. Contribution/Results: Evaluated on real clinical data from four hospitals, our method achieves a PSNR of 28.6 dB on synthesized CT volumes—3.2 dB higher than local training—and inference latency of <1.2 seconds per volume. The framework delivers scalable, high-fidelity, and privacy-compliant multi-center medical imaging modeling, establishing a new paradigm for cross-institutional collaboration in radiological AI.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Security and Privacy: Security and privacy of machine learning and AI applications

Problem

• Enhancing MRI-to-CT synthesis generalizability across diverse clinical settings
• Addressing privacy concerns in multi-center medical data collaboration
• Improving radiotherapy planning accuracy with federated learning

Innovation

• Federated Learning for MRI-to-CT synthesis
• Cross-silo horizontal FL with U-Net model
• Multi-institutional collaboration preserving privacy