An Advanced Deep Learning Framework for Skin Cancer Classification

One of the most prevalent cancers in humans, skin cancer is typically identified by visual inspection. Early detection of this kind of cancer is essential. Consequently, one of the most difficult aspects of designing and implementing digital medical systems is coming up with an automated method for classifying skin lesions. Convolutional Neural Network (CNN) models, enabled by thermoscopic pictures, are being used by an increasing number of individuals to automatically differentiate benign from malignant skin tumors. The classification of skin cancer through the use of deep learning and machine learning techniques may have a significant positive impact on patient diagnosis and care. These approaches’ significant computational cost means that their capacity to extract highly nonlinear properties needs to be improved. Using fewer learnable parameters, this work aims to enhance model convergence and expedite training by classifying early-stage skin cancer. Combining the VGG19 and network-in-network (NIN) architectures, the VGG-NIN model is a strong and scale-invariant deep model. The exceptional nonlinearity of this model simplifies the task of capturing complex patterns and features. Additionally, by adding NIN to the model, additional nonlinearity is introduced, improving classification performance. Based on samples of skin cancer, both Benign and Malignant, our model has an outstanding 90% accuracy with the fewest possible trainable parameters. As part of our research, we used a publicly accessible Kaggle dataset to do a benchmark analysis to assess the performance of our suggested model. The processed photos from the ISIC Archive, notably the HAM10000 Skin Cancer dataset, made up the dataset used in this study. One well-known source for dermatological photos is the ISIC Archive. The suggested model effectively uses computer resources and performs more accurately than cutting-edge techniques.