Federated Contrastive Learning for Volumetric Medical Image Segmentation

Supervised deep learning needs a large amount of labeled data to achieve high performance. However, in medical imaging analysis, each site may only have limited and labels, which makes ineffective. Federated (FL) can help this regard by shared model while keeping training local for privacy. Traditional FL requires fully-labeled training, is inconvenient or sometimes infeasible obtain due labeling cost the requirement expertise. Contrastive (CL), as self-supervised approach, effectively learn from unlabeled pre-train neural network encoder, followed fine-tuning downstream tasks with annotations. when adopting CL FL, diversity on client federated contrastive (FCL) In work, we propose an FCL framework volumetric image segmentation More specifically, exchange features pre-training process such that diverse are provided effective raw private. Based exchanged features, global structural matching further leverages similarity align remote ones unified feature space be learned among different sites. Experiments cardiac MRI dataset show proposed substantially improves performance compared state-of-the-art techniques.