Segment anything model for medical image segmentation: Current applications and future directions.

Area of Science:

• Artificial Intelligence
• Medical Image Analysis
• Computer Vision

Background:

• Foundation models, particularly the Segment Anything Model (SAM), are transforming NLP and computer vision with prompt-driven capabilities.
• SAM's expansion into image segmentation offers new possibilities, but its direct application to medical imaging is challenging due to domain-specific differences.

Purpose of the Study:

• To comprehensively review recent efforts adapting SAM for medical image segmentation.
• To empirically benchmark SAM's performance and explore methodological adaptations.
• To identify future research directions for SAM in medical image analysis.

Main Methods:

• Empirical benchmarking of SAM on medical image segmentation tasks.
• Methodological adaptations to enhance SAM's efficacy for medical data.
• Review of existing research and open-source projects related to SAM in medical imaging.

Main Results:

• Direct application of SAM to multi-modal and multi-target medical datasets has not yet yielded satisfactory results.
• Significant insights have been gained regarding SAM's limitations and potential in the medical domain.
• Ongoing research is actively adapting and evaluating SAM for diverse medical imaging applications.

Conclusions:

• While direct application is currently limited, SAM's adaptation holds potential for advancing medical image segmentation.
• Insights from current research guide the development of future foundational models for medical image analysis.
• A curated repository of research and open-source projects is maintained to support the community.