A Modified SAM-Based Skin Cancer Segmentation Pipeline: SKIN-SA Model

Accurate skin lesion segmentation is a pivotal task in dermatology with significant implications for early diagnosis and treatment of skin conditions, including skin cancers. In this research paper, we introduce Skin-SA, a novel skin lesion segmentation model that leverages Meta's Segment Anything Model (SAM) as the cornerstone of a comprehensive pipeline. Through an extensive exploration of contemporary approaches, we provide a review of both traditional techniques and modern deep neural networks-based approaches, despite the substantial progress achieved with CNNs in recent years, we take a different path by adopting a vision transformer-based architecture, capitalizing on its ability to capture intricate spatial information within skin lesion images, the method devised surpasses the state-of-the-art in skin lesion segmentation with improvements of 5%-15% in segmentation evaluation metrics.