Federated contrastive learning for decentralized unlabeled medical images

Conference item

A label-efficient paradigm in computer vision is based on self-supervised contrastive pre-training on unlabeled data followed by fine-tuning with a small number of labels. Making practical use of a federated computing environment in the clinical domain and learning on medical images poses specific challenges. In this work, we propose FedMoCo, a robust federated contrastive learning (FCL) framework, which makes efficient use of decentralized unlabeled medical data. FedMoCo has two novel modules: metadata transfer, an inter-node statistical data augmentation module, and self-adaptive aggregation, an aggregation module based on representational similarity analysis. To the best of our knowledge, this is the first FCL work on medical images. Our experiments show that FedMoCo can consistently outperform FedAvg, a seminal federated learning framework, in extracting meaningful representations for downstream tasks. We further show that FedMoCo can substantially reduce the amount of labeled data required in a downstream task, such as COVID-19 detection, to achieve a reasonable performance.

Authors: Dong, N; Voiculescu, ID

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: Springer
• Host title: Medical Image Computing and Computer Assisted Intervention – MICCAI 2021.
• Pages: 378-387
• Series: Lecture Notes in Computer Science
• Series number: 12903
• Publication date: 2021-09-21
• Acceptance date: 2021-06-11
• Event title: 24th International Conference on Medical Image Computing and Computer Assisted Intervention: MICCAI 2021
• Event location: Virtual event
• Event website: https://miccai2021.org/
• Event start date: 2021-09-27
• Event end date: 2021-10-01
• DOI: 10.1007/978-3-030-87199-4_36
• EISBN: 978-3-030-87199-4
• ISBN: 978-3-030-87198-7
• Language: English
• Keywords: FFR; federated learning; contrastive representation learning