SkinDet-YOLO: a context- and boundary-aware yolov8-based framework for skin disease detection

Skin diseases are among the most common human disorders, and accurate localization of suspicious lesions in clinical or dermoscopic images is a critical first step towards computer-aided diagnosis. However, skin lesion detection remains challenging due to extreme scale variation, fuzzy lesion boundaries, and high inter-class visual similarity.

We propose SkinDet-YOLO, a YOLOv8-based framework with a Context-Aware Multi-scale Feature Fusion network (CAMF) and an Adaptive Boundary-Aware Detection Head (ABADH).

SkinDet-YOLO achieves validation mAP@0.5 of 0.992 and mAP@0.5:0.95 of 0.900, outperforming strong baselines under a 7:2:1 train/validation/test split.

Ablation and sensitivity analyses show complementary gains from CAMF and ABADH and support the framework's potential for clinical skin lesion detection.