Abstract
Skin cancer is a deadly kind of cancer that grows rapidly and produces life-threatening issues within six weeks from the initial stage. Accurate analysis is needed for observing both the malignant and benign skin lesions that are more complicated to ensure the essential treatment for the individuals. The best solution to perform automatic skin lesion categorization is to employ deep learning systems that are trained from a huge amount of benchmark data. Moreover, melanoma is referred to as a crucial skin cancer that badly affects entire human health and also their life. In this case, early-stage melanoma diagnosis is more complicated due to the presence of color changes in the same lesion. Hence, to overcome these complications in the traditional techniques, a novel automated skin lesion segmentation model is required to provide effective diagnosis results. The collected images are applied to the preprocessing stage to enhance quality. The pre-processed images are given to the segmentation stage by utilizing an Adaptive Layer-based Visual Transformer with UNet (AL-VTransUNet), where the variables are tuned with Improved Random Parameter-based Galactic Swarm Optimization (IRP-GSO). Ultimately, the segmented images are subjected to the Dilated DenseNet with a Multi-Head Attention mechanism (DD-MHA) for classifying the outcomes. Here, the variable is also tuned with the support of the same IRP-GSO algorithm for improving classification effectiveness. Thus, the efficiency of the recommended framework is examined using existing optimization and classification approaches.