CoralBay: A Self-Supervised CT Foundation Model
Researchers have developed CoralBay, a novel self-supervised learning framework for 3D medical imaging, specifically CT scans. This method extends the DINO framework with a 3D Swin backbone and self-distillation techniques to capture rich spatial representations. CoralBay demonstrates effective transfer learning across various radiological tasks and contributes to the open-source \eva framework with a new 3D radiology leaderboard. AI
IMPACT Advances self-supervised learning for 3D medical imaging, potentially improving diagnostic accuracy and efficiency.