Mutual Distillation of Dual-Foundation Models for Semi-Supervised PET/CT Segmentation
Researchers have developed a novel semi-supervised learning framework called MuDuo for segmenting organs in PET/CT scans. This approach leverages dual-foundation models, SAM-Med3D for CT and SegAnyPET for PET, to distill knowledge into a more lightweight student network. MuDuo effectively utilizes unlabeled data to achieve state-of-the-art performance on the AutoPET dataset with minimal labeled cases, eliminating the need for manual prompts. AI
IMPACT This research could significantly reduce the annotation burden for medical imaging tasks, accelerating the development and deployment of AI-powered diagnostic tools.