Improvement of Skin Cancer Detection Performance Using Deep Learning Technique

Determination of cancer on time and accurately could increase the probability of successful treatment of it. Inside the cancer types malignant skin cancer is the most dangerous cancer type which is seen widely in the world and frequency of seeing it increases with time. Computer aided diagnosis applications are used for the classification of skin cancers. In this study, dermoscopic images acquired from ISIC archieve are used to create a dataset which has two classes, and this dataset is used to classify benign and malignant cancer types. And as a result of early diagnosis, the classification score is aimed to be increased. To do this, some image preprocessing operations like color clarification, edge detection, and noise extraction on dermoscopic images obtained from dataset are applied. After this processing operation, InceptionV2 deep learning network is used to classify these processed images. As a result of this study, it is seen that the preprocessing operation increases the accuracy ratio by 3.33 point, and an accuracy rate of 88.66% is get.