An Improved CNN Architecture to Diagnose Skin Cancer in Dermoscopic Images Based on Wildebeest Herd Optimization Algorithm

Skin cancer is one of the most common types of cancers that is sometimes difficult for doctors and experts to diagnose. The noninvasive dermatoscopic method is a popular method for observing and diagnosing skin cancer. Because this method is based on ocular inference, the skin cancer diagnosis by the dermatologists is difficult, especially in the early stages of the disease. Artificial intelligence is a proper complementary tool that can be used alongside the experts to increase the accuracy of the diagnosis. In the present study, a new computer-aided method has been introduced for the diagnosis of the skin cancer. The method is designed based on combination of deep learning and a newly introduced metaheuristic algorithm, namely, Wildebeest Herd Optimization (WHO) Algorithm. The method uses an Inception convolutional neural network for the initial features' extraction. Afterward, the WHO algorithm has been employed for selecting the useful features to decrease the analysis time complexity. The method is then performed to an ISIC-2008 skin cancer dataset. Final results of the feature selection based on the proposed WHO are compared with three other algorithms, and the results have indicated good results for the system. Finally, the total diagnosis system has been compared with five other methods to indicate its effectiveness against the studied methods. Final results showed that the proposed method has the best results than the comparative methods.