Search for a command to run...
Skin cancer remains a major global health burden, and early identification is critical for reducing mortality. Traditional diagnostic methods rely heavily on visual assessment, which is subjective, inconsistent, and limited by clinician availability. This work introduces an AI-powered dermatological analysis system designed to detect malignant and benign lesions from dermatoscopic images using deep learning. The study utilizes the ISIC dataset, consisting of diverse lesion images collected between 2018–2020, containing metadata such as lesion type, anatomical site, and clinical diagnosis. Preprocessing steps include resizing, normalization, illumination correction, and extensive augmentation to address class imbalance and imaging variability. Multiple architectures were implemented, including MobileNetV2, ResNet50, a custom CNN, and an optimized YOLOv8 variant for real-time lesion localization and classification. Model performance was evaluated using accuracy, precision, recall, F1-score, and mAP. The YOLOv8m model achieved the best results, yielding 93% mAP@0.5 and high sensitivity for malignant lesions, demonstrating clinical reliability. Overall, the system significantly improves detection accuracy and reduces diagnostic dependency on manual expertise, highlighting its potential to enhance early screening.