A federated multimodal deep learning framework for brain tumor classification using MRI.

Area of Science:

• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare
• Distributed Systems

Background:

• Centralized brain tumor classification using MRI data raises privacy concerns and limits model training due to data distribution across hospitals.
• There is a need for privacy-preserving and distributed learning methods to ensure both security and accuracy in medical data analysis.

Purpose of the Study:

• To propose a federated learning framework for privacy-preserving, collaborative brain tumor classification using MRI data.
• To enhance the efficiency and scalability of distributed medical data analysis.

Main Methods:

• Federated learning framework enabling collaborative model training without raw data sharing.
• Layer skipping mechanism to reduce communication costs.
• FedPropSAG aggregation method for improved convergence and performance.
• Integration of Differential Privacy (DP) and Secure Aggregation (SA) for data privacy and secure communication.

Main Results:

• High classification accuracy achieved across distributed datasets.
• Significant reduction in communication costs due to the layer skipping mechanism.
• Effective performance under non-IID data distributions.
• Privacy-preserving techniques did not degrade overall model performance.

Conclusions:

• The proposed federated learning approach offers an efficient, scalable, and privacy-preserving solution for distributed medical data analysis.
• The framework enables collaborative learning across institutions while maintaining patient data confidentiality.
• Reduced communication overhead makes the model suitable for practical deployment in healthcare systems.