Skin cancer is known as one of the most usual malignant cancer in the human body. Statistics show that each year above one million people are added to the who has this cancer. There are different types of skin cancer, where, the main difference is on the type of cell that is developing cancer. The best way to treat this cancer is to diagnose it early and to prevent the lesion from spreading with surgery. Early detection and treatment of skin cancer from skin images can significantly reduce mortality. Although skin cancer is very dangerous, early diagnosis and appropriate treatment, in most cases, prevent death. The present study introduces a new diagnostic technique for skin cancer based on deep learning and metaheuristics. At first, a pre-trained modified AlexNet based on batch normalization layers is used to train the skin dermoscopy images. Afterward, the last several layers are substituted by an Extreme Learning Machine (ELM). For providing higher efficiency in the ELM network, a newly amended metaheuristic, called Fractional-order Red Fox Optimization (FORFO) Algorithm is used. The final results of the proposed technique are compared with some various techniques and the results showed the effectiveness of the suggested method.
Keywords: AlexNet; Extreme learning machine; Fractional-order Red Fox Optimization (FORFO) algorithm; Skin cancer; computer-aided.