An Interpretable Skin Cancer Classification Using Optimized Deep Transfer Learning Method

The skin, the largest organ in the body, acts as a vital layer of defense but can also develop several diseases, including skin cancer, one of the top three cancers. There are various forms of skin cancer, each of which poses multiple challenges for identification and treatment. This study investigates the effectiveness of deep learning approaches, namely convolutional neural networks (CNNs), for the rapid and exact diagnosis of skin cancer types. Using the ISIC Skin Cancer Challenge 2019 dataset, our research sheds insight into CNNs ability to distinguish diagnostic accuracy for skin cancer diagnosis. We used 12,295 skin lesion images from this dataset, where three forms of skin cancer: nevus, melanoma, and basal cell carcinoma are focused. During the preprocessing step, the photographs were augmented, normalized, and resized. The MobileNetV2 transfer learning model is evaluated against DenseNet121, InceptionV3, ResNet152V2, CNN and comparative analysis with the the current state of the art. After testing, the model achieved the greatest accuracy at 97.48%. Our MobileNetV2 transfer learning model surpasses earlier models in terms of dependability and robustness. Beyond providing an in-depth assessment of current difficulties, it further promotes critical conversation and collaboration in medical image analysis and healthcare innovation.