A Machine Learning-Based Model for the Detection of Skin Cancer Using YOLOv10

Melanoma, the most dangerous form of skin cancer originates in pigment cells (melanocytes) and poses a significant threat to patients. Therefore, individuals must have any suspicious lesions or moles promptly evaluated for early detection to improve survival rates. The following study presents a Machine learning-based model for detecting skin cancer using YOLOv10 technology. The process involves preprocessing, augmentation, and training on two datasets: the ISIC 2017 dataset (with classes melanoma, nevus, and seborrheic keratosis) annotated based on YOLOv8 versions via Roboflow; and another dataset from Roboflow containing three classes-malignant tumors, benign growths, and healthy skin. Preprocessing techniques were utilized alongside augmentation methods to enhance the model's performance before fine-tuning the pre-trained YOLOv10 model as per requirements. This approach yielded promising results with achieved precision of 1, recall of 0.88, and
$mAP50$
of 0.94 on the ISIC2017 dataset; precision of 1, recall of 0.95, mAP50 of 0.79, and F-score score of 82% on the Skin Detection Dataset respectively.