A Deep Learning-Based Skin Cancer Detection with ResNet50V2

Skin cancer is a prevalent global health concern, with melanoma being particularly dangerous due to its potential to metastasise. Researchers have harnessed deep learning techniques, specifically transfer learning, to create an automated classification system for skin lesions. This system can be especially valuable in areas with limited medical resources. The study leverages pre-trained convolutional neural networks (CNNs) and transfer learning. Using Google Collab and a dataset attached via Google Drive, skin lesion images were analysed. The dataset from the International Skin Imaging Collaboration (ISIC) was divided into eight classes, with “Melanocytic Nevus” being the most common (51% of data). Augmentation was applied to address overfitting. Comparing the sequential CNN and pre-trained ResNet50V2 models, both achieved over 60% accuracy. While ResNet50V2 had slightly better accuracy, the sequential model exhibited greater stability in validation results. Notably, segmentation techniques were not employed due to image-specific challenges. This research contributes to improving skin cancer diagnosis and underscores the potential of AI in healthcare. Doctors can benefit from this system, enhancing patient care and treatment decisions.