Class Imbalanced Medical Image Classification Based on Semi-Supervised Federated Learning

In recent years, the application of federated learning to medical image classification has received much attention and achieved some results in study semi-supervised problems, but there are problems such as lack thorough labeled data, serious model degradation case small batches face data category imbalance problem. this paper, we propose a method using combination regularization constraints pseudo-label construction, where framework consists central server local clients containing only unlabeled passed from each client take part training. We first extracted class factors participate training achieve label constraints, secondly fused with at construct augmented samples, looped through generate pseudo-labels. The purpose combining these two methods is select fewer classes higher probability, thus providing an effective solution problem improving sensitivity network data. experimentally validated our on publicly available set consisting 10,015 images Our improved AUC by 7.35% average 1.34% compared state-of-the-art methods, which indicates that maintains strong capability even unbalanced trained models.