# Towards Segment Anything Model (SAM) for Medical Image Segmentation: A Survey

**Authors:** Yichi Zhang, Rushi Jiao  
**arXiv:** [2305.03678](https://arxiv.org/abs/2305.03678)  
**Date:** 2023-05-05  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2305.03678v3/

## Key points

- Foundation models, such as the Segment Anything Model (SAM), have become powerful tools in natural language processing and image generation.
- SAM extends the prompt-driven paradigm to image segmentation tasks, showing promise in natural image segmentation but facing challenges in medical image segmentation due to differences between the two types of images.
- Recent research efforts focus on adapting SAM for medical image segmentation through empirical benchmarking and methodological adaptations.
- Challenges with multi-modal and multi-target medical datasets have been addressed, providing valuable insights for developing foundation models tailored for medical image analysis.
- A comprehensive survey by Yichi Zhang and Rushi Jiao discusses recent advancements in using SAM for medical image segmentation and emphasizes the need for specialized approaches to handle unique characteristics of medical images.
- The authors maintain an active repository (SAM4MIS) containing relevant papers and open-source projects related to SAM for Medical Image Segmentation to support ongoing research efforts.

## Summary

In recent years, foundation models have emerged as a powerful tool in natural language processing and image generation. Thanks to their flexibility in responding to prompts, these models have made significant advancements in various applications. One such model is the Segment Anything Model (SAM), which extends the prompt-driven paradigm to image segmentation tasks. While SAM has shown promise in natural image segmentation, its applicability to medical image segmentation remains uncertain due to inherent differences between the two types of images. To address this gap, recent research efforts have focused on adapting SAM for medical image segmentation tasks. These endeavors involve empirical benchmarking and methodological adaptations to enhance SAM's performance in analyzing medical images. Despite initial challenges with multi-modal and multi-target medical datasets, these efforts have provided valuable insights that can guide future research in developing foundation models specifically tailored for medical image analysis. A comprehensive survey conducted by Yichi Zhang and Rushi Jiao summarizes these recent advancements and discusses potential directions for further exploration of SAM in the context of medical image segmentation. The survey highlights the need for specialized approaches that can effectively handle the unique characteristics of medical images, such as varying modalities and complex target structures. To support ongoing research in this area, the authors maintain an active repository containing an up-to-date list of relevant papers and open-source projects related to SAM for Medical Image Segmentation (SAM4MIS). Overall, while challenges persist in applying SAM directly to medical image segmentation tasks, the ongoing efforts outlined in the survey offer promising avenues for advancing the field and developing more sophisticated models tailored to the specific requirements of medical imaging applications.