Learning the Unlabeled: Advancing Self-Supervised Learning for Diagnosing Underrepresented Pathologies in Medical Imaging

This research aims to improve how artificial intelligence (AI) analyzes medical images to better detect underrepresented diseases. Current AI models often miss uncommon conditions because they rely on large labeled datasets and struggle with imbalanced data. The project proposes new semantically aware self-supervised learning (SSL) methods that learn from unlabeled images, preserving important medical details and introducing a “re-weighting” system that helps the model focus on rare or difficult cases. The goal is to develop more accurate and equitable diagnostic AI tools that support radiologists and improve healthcare access in underserved regions