SPIE Digital LIbrary, CAFES: chest x-ray analysis using federated self-supervised learning for pediatric Covid-19 detection

The study "CAFES: Chest X-ray Analysis Using Federated Self-Supervised Learning for Pediatric COVID-19 Detection" presents an AI-driven approach to enhance the accuracy of COVID-19 diagnosis in pediatric patients through chest X-rays (CXRs). Utilizing a Federated Self-Supervised Learning (FSSL) framework, the research employs Vision Transformers (ViTs) pre-trained on adult CXR data to improve binary classification of COVID-19 versus non-COVID-19 cases in children. Implemented on the Rhino Health Federated Computing Platform, this method ensures data privacy by keeping patient information localized while enabling collaborative model training across institutions. The FSSL-pre-trained ViT achieved an area under the precision-recall curve (AUPR) of 0.952, outperforming fully supervised models by 0.231 points, indicating significant improvements in detecting COVID-19 from pediatric CXRs. This research underscores the potential of combining federated learning and self-supervised techniques to advance diagnostic tools in pediatric healthcare.