Auxiliary Diagnosis of COVID-19 Based on 5G-Enabled Federated Learning

Abstract

The development of 5G and artificial intelligence technologies have brought novel ideas to the prevention, control, and diagnosis of disease. Due to the limitation of the privacy protection of medical big data, releasing the data of patients is not allowed. However, as COVID-19 spreads globally, it is urgent to develop a robust diagnostic model to serve as many institutions as possible. Therefore, we propose a 5G-enabled architecture of auxiliary diagnosis based on federated learning for multiple institutions and central cloud collaboration to realize the sharing of diagnosis models with high generalization performance. In order to exchange model and parameters between the central and distributed nodes, a framework of diagnosis model cognition is constructed for sharing and updating the model adaptively. The severity classification experiments of COVID-19 were carried out on the central cloud and three edge cloud servers to verify the effectiveness of the proposed architecture and model cognition strategy. At the same time, the aggregated model shows good performance with accuracy rates of 95.3, 79.4, and 97.7 percent on distributed nodes, and the recognition results can assist doctors in executing auxiliary diagnosis. Finally, the open issues of model fusion of multimodal data in the 5G network architecture are discussed.