Convolution Neural Network Method for Skin Cancer Diagnosis: Comparison and Improvement

Abstract

Skin cancer, the most common cancer in the world, has many detection steps and the detection process is easy to make mistakes. A detection method based on convolutional neural network (CNN) is proposed to assist doctors in the detection. Based on the development of CNN in the classification and diagnosis of skin cancer in recent years, this paper compares and summarizes the development of each step in this process. After reviewing previous papers, it can be concluded that the classification process is roughly divided into four parts. In addition, the evaluation indicators of the model are further analyzed. AUC Sen and SPE are the most basic evaluation indicators in the model evaluation. As a skin classifier, CNN improves the accuracy of classification and diagnosis results to a great extent. CNN model has also made progress in "lightweight" and "concise". However, there are few evaluation indicators available for different CNN methods, and the evaluation latitude is relatively single. In the future, the evaluation indicators should develop in more aspects, it will enable to better understand the personality of a CNN model.