EOSA-Net: A deep learning framework for enhanced multi-class skin cancer classification using optimized convolutional neural networks

Abstract

Most cancer diagnoses are inevitably fatal, and an increasing number of genetic and metabolic abnormalities are being identified by experts as the disease’s cause. Every organ in the body is susceptible to the invasion and spread of cancerous cells, which pose a major threat to health. So, considering the frequency of skin cancer is rising quickly, it is recommended to undergo a screening. The multi-class approach of deep learning (DL) may provide more comprehensive data and insights into the underlying causes and risk factors of all different skin cancer classes, which can aid in developing fresh therapies or preventative measures. DL is one of the best methods for rapidly and reliably detecting skin cancer. The ISIC 2018 and 2019 public datasets were used in the present research to identify and detect multi-class skin cancer by employing an Improved canny edge for detection and an optimized deep learning technique convolution neural network (CNN) for classification. To enhance the detection of the affected area from the input image, the pre-processing stage is conducted using an improved canny edge detector. The control parameters for the canny edge detector are selected using the coronavirus optimization algorithm (CVOA), and the proposed model has three discrete hidden layers, each with a corresponding channel size (16, 32, and 64). A learning version of the Ebola Optimization Search Algorithm (EOSA) with a learning rate of 0.001 is included in the suggested model. Both the mathematical and propagation models were modified to create the new metaheuristic technique. The proposed EOSA algorithm has been brought towards as a robust mechanism for achieving a balance between exploration and exploitation to determine the optimal solution to problems. With an accuracy of 99%, the proposed model performs the best when classifying the skin lesions in the ISIC dataset into eight categories. The results obtained are superior to the present system of skin cancer categorization.