SAMora: Enhancing SAM through Hierarchical Self-Supervised Pre-Training for Medical Images

The introduction of SAMora marks a significant advancement in medical image segmentation, building on the capabilities of the Segment Anything Model (SAM). By leveraging hierarchical self-supervised learning, SAMora captures valuable medical knowledge at multiple levels, enhancing its performance in both few-shot and fully supervised settings. Experimental results indicate that SAMora outperforms existing SAM variants, achieving state-of-the-art results while drastically reducing fine-tuning epochs by 90%. This is particularly important in the medical field, where the availability of labeled data is often limited, and efficient segmentation can greatly impact diagnostic processes. The framework has been tested on prominent datasets like Synapse, LA, and PROMISE12, showcasing its versatility and effectiveness. The code for SAMora is publicly available, encouraging further research and development in this critical area of artificial intelligence.