YOLO-SCSA: Enhanced YOLOv8 with Spatially Coordinated Shuffling Attention Mechanisms for Skin Cancer Detection

Jinyoon Kim, Tianjie Chen, Hien Nguyen, Md Faisal Kabir

Abstract: Skin cancer is one of the most prevalent and deadliest diseases worldwide. Traditional detection methods, relying on visual examination and biopsy, are time-consuming. Early detection is crucial, as delays can significantly risk patients' lives. Advances in machine learning, particularly in computer vision, have enabled faster and more accurate detection of skin cancer. YOLO (You Only Look Once) is a state-of-the-art model for object detection, known for its high accuracy and speed. In 2023, Ultralytics released the latest version, YOLOv8. This research proposes the YOLO-SCSA model, which enhances performance by integrating both general and domain-specific attention modules. Our SCSA attention module combines mechanisms from previous attention modules and introduces a new branch for richer feature understanding. Additionally, the Center Weighted Masking module improves focus on crucial parts of the feature map, enhancing performance on the skin cancer dataset within the YOLOv8 architecture.