Advancements in Semantic Skin Lesion Segmentation: A Comparative Study with YOLOv8 for Accurate Skin Lesion Segmentation

Acquiring labeled data for multi-type skin lesions poses a significant challenge due to the high cost associated with the involvement of dermatologists, given the intricate nature of small pixel variations and irregularly shaped lesion characteristics. Moreover, the presence of multiple types of skin lesions exhibiting these characteristics often leads to overlapping or nearby occurrences, further complicating the labeling process. Numerous models have been developed to accurately segment identified pixels in skin lesion datasets, with a focus on semantic segmentation and instance-based segmentation approaches. This paper aims to compare the most effective semantic segmentation techniques with the recently introduced YOLO V8 model, evaluating their performance and suitability for accurately segmenting skin lesions. Due to its superior mean average precisions (mAPs) and faster inference speed on the COCO dataset, YOLOv8 is considered the new state-of-the-art model, making it a promising candidate for accurate skin lesion segmentation.