Cross paradigm fusion of federated and continual learning on multilayer perceptron mixer architecture for incremental thoracic infection diagnosis

Abstract

Abstract Medical imaging is essential in the study of chest virus infections. Due to data sovereignty issues in healthcare, it is essential to employ federated learning to overcome these obstacles. However, obtaining all relevant data at once is challenging, as it is often acquired incrementally. Therefore, addressing continual learning is imperative. To this end, we combined federated learning with continual learning to construct a transnational infectious disease prediction model. This model was applied to the COVID-XRAY and COVID-CT datasets using 3 and 6 clients, respectively, implementing 4 continual learning algorithms across ten different models. Notably, we integrated MLP-Mixer with Learning without Forgetting (LwF) techniques, achieving an accuracy of 54.34%. This demonstrates the effectiveness of our approach in the early detection, sensing, and timely warning of infectious diseases and ultimately builds a multicentre prediction system for future infectious diseases.