Research on Medical Image Data Privacy Protection and Collaborative Diagnosis Based on Federated Learning

Deep learning has shown strong capabilities in handling visual tasks such as medical imaging. However, the privacy protection characteristics of medical data have led to an extreme scarcity of data available for training deep models, which limits their further development. Based on the federated learning framework, this paper proposes a medical image classification algorithm FL-RA that integrates the attention mechanism, aiming to solve the problems of medical data privacy protection and data silos. The experiments used four public medical image datasets (COVID-19, PneumoniaMNIST, OrganSMNIST, and OCTMNIST). The results show that the performance of the model varies across different datasets, but in most cases, the recognition accuracy for positive samples is high. The results of the ablation experiment show that after introducing the attention mechanism, the classification accuracy of the model is improved by 1%-2%, and the accuracy of the model trained jointly using federated learning is 4%-8% higher than that of local training alone. This study provides an effective technical solution for privacy protection and collaborative diagnosis in medical image diagnosis, and has important practical significance and broad application prospects.