Novel Deep Learning Framework for Melanoma Diagnosis Using Enhanced Super Resolution Generative Adversarial Network and Convolutional Neural Network

Melanoma, a more aggressive class of skin cancer, develops from melanocytes, pigment-producing cells in the skin. Late diagnosis increases the risk of cancer spreading to other parts of the body, thereby decreasing the patient’s survival rate. An early diagnosis of melanoma cancer not only increases the survival rate but also improves the disease prognosis. Reduced medical costs incurred during cancer treatment are one of the major advantages of early cancer diagnosis. This research study proposes and implements a novel framework that leverages the enhanced super resolution generative adversarial network (ESRGAN) for pre-processing. The segmentation of the region of interest in preprocessed high-resolution images is carried out by density-based clustering algorithm with the Salp optimization technique. In the segmentation phase, the Salp Optimization algorithm computes the optimum cluster center for enhancing the performance of the density-based clustering algorithm. The classifier adopted in this study is a custom-built convolutional neural network model. The proposed convolutional neural network model with set parameters demonstrates a high accuracy and recall rate, deferring the need to adopt time-consuming and complex deep learning models. The experimental study evaluates the performance of the proposed methodology and emphasizes that the implemented diagnosis tool shows accuracy of 94.5% in melanoma diagnosis. The peak signal noise ratio (PSNR) and structural similarity index metrics are evaluated for the ESRGAN-generated images, showing significant increase in image resolution with the method adopted.