Application of SAM Model in Zero-shot Segmentation of Skin Wound Images

Quantitative evaluation of skin wounds is important in tracking the healing process and planning the treatment. For non-invasive and automatic analysis of the wound, segmentation of the wound area from images is a primary step. This study evaluates the performance of the Segment Anything Model (SAM) in the task of segmenting skin wound images. SAM supports zeroshot segmentation, thus eliminating the need to build large training datasets. The adaptability and potential of SAM in this task is fully demonstrated through a variety of different prompting strategies. The results of the study indicate that SAM has a good ability to segment skin wound images and can accurately segment wounds with few prompts. This study provides practical guidance for the clinical application of the SAM model and suggests ways to further optimize the results, ultimately aiming to improve the efficiency and accuracy of skin wound analysis.