CAFES: Chest X-ray Analysis using Federated Self-supervised Learning for Pediatric COVID-19 Detection

Area of Science:

• Artificial Intelligence in Medical Imaging
• Radiology and Diagnostic Imaging
• Computational Pathology

Background:

• Chest X-rays (CXRs) are crucial for assessing pediatric lung conditions, including those related to COVID-19.
• Accurate and rapid COVID-19 diagnosis in children is essential for timely clinical management.
• Existing AI models may require large, centralized datasets, posing privacy and scalability challenges.

Purpose of the Study:

• To develop and evaluate an AI-driven solution for binary COVID-19 versus non-COVID-19 classification in pediatric CXRs.
• To enhance the performance of Vision Transformers (ViTs) for COVID-19 detection using a Federated Self-Supervised Learning (FSSL) framework.
• To ensure a privacy-conscious and scalable approach for medical image analysis.

Main Methods:

• Implementation of a Federated Self-Supervised Learning (FSSL) framework utilizing Vision Transformer (ViT) architecture.
• Self-supervised pre-training of ViT on adult CXR data, followed by fine-tuning on pediatric CXR datasets.
• Deployment on the Rhino Health Federated Computing Platform (FCP) for privacy-preserving distributed learning.
• Development of the Chest X-ray Analysis using Federated SSL (CAFES) model.

Main Results:

• The FSSL-pre-trained ViT model (CAFES) demonstrated improved COVID-19 detection accuracy in pediatric CXRs compared to a fully supervised model.
• The CAFES model achieved an area under the precision-recall curve (AUPR) of 0.952.
• This represents a significant improvement of 0.231 points over the fully supervised model for pediatric COVID-19 diagnosis.

Conclusions:

• Leveraging ViTs with FSSL pre-training offers an effective strategy for COVID-19 diagnosis from pediatric CXRs.
• Distributed federated learning-based self-supervised pre-training enhances diagnostic performance while maintaining data privacy.
• This privacy-conscious approach, aligning with HIPAA guidelines, shows promise for broader applications in medical imaging AI.