Skin Cancer Diagnosis and Medical Service System Based on Deep Learning Models

Abstract

Skin cancer is a kind of common disease that seriously does harm to peoples’ life quality, and its diagnosis is of great importance. This paper aims to apply a Convolutional Neural Network (CNN) model to diagnose skin cancer automatically to provide a reference for doctors and an application containing a website, a chatbot and a speech synthesis to provide patients with an electronic diagnosis service. We used image normalization to preprocess the data, as well as data augmentation which overcomes the overfitting problem to some extent. Our proposed model has a convolutional layer and a MaxPooling layer. Then a Dropout layer follows to reduce overfitting. After that, there are a Flatten layer and two Dense layers. Adam is used as the optimizer and the loss function is sparse categorical cross entropy. As a result, our model reached a training accuracy of 95.56% and a validation accuracy of 90.95% in diagnosing different kinds of skin diseases. The training loss is 0.1099 and the validation loss is 0.3356. For our website application, it realized functions of storage of users’ information, chatting service and voice announcement of diagnose results. In conclusion, it can diagnose skin diseases accurately and provide convenience for automatically modern medical services.