Federated Learning-assisted Self-supervised CNN for Monkeypox Diagnosis

Monkeypox (Mpox) is a contagious viral illness that affects both humans and animals, and its early diagnosis is critical for the effective management and prevention of this disease. This work proposes a federated learning-assisted self-supervised convolutional neural network (CNN) for Mpox identification from skin lesion images. The demand for domain experts for ground truth annotations is reduced with the help of self-supervised learning, as it can process unlabeled data. The self-supervision is driven by a framework called simple contrastive learning for representations (SimCLR), which extracts discriminative patterns of Mpox from the skin lesion images. Federated learning (FL), on the other hand, enables a privacy-preserved collaborative training approach to build the Mpox classifier on vast diverse datasets from several healthcare institutions, remotely. Thorough experimental study on a publicly available benchmark dataset, the proposed approach achieves competitive performances.