A Deep Learning and Transfer Learning-Based Application for Skin Cancer Classification

Skin cancer is the most prevalent type of cancer worldwide, accounting for approximately one-third of all cancer cases, with its incidence steadily rising in recent years. This is why early and precise identification of skin cancer is crucial for effective treatment and enhancing survival rates. Leveraging advancements in deep learning and transfer learning, we developed a comprehensive system to address this challenge. In this study, we employed the HAM10000 dataset, a benchmark resource for skin cancer research, to train and evaluate deep learning models. Two architectures, DenseNet201 and ResNet50, were fine- tuned and modified to enhance their accuracy in classifying skin lesions as benign or malignant. Rigorous preprocessing, data augmentation, and model fine-tuning were utilized to address challenges such as dataset imbalance and improve classification accuracy. Our experiments revealed that the modified fine-tuned DenseNet201 model achieved superior performance compared to fined-tuned ResNet50 and other leading models reviewed in the literature with an accuracy of 95.04%. This model was then integrated into a mobile application, 'SkinSafe,' to provide an accessible and user-friendly tool for real-time skin cancer classification.