Classification of skin lesion images using machine learning and deep learning techniques

Cancer is the uncontrolled division and spread of cells. Skin cancer, like other types of cancer, is a very important type of cancer for early diagnosis and treatment. Reasons such as the expertise of the people who examine the dermoscopy images used in the diagnosis phase and various artifacts in the images can reduce the diagnostic accuracy when the images are examined manually. Examination of images with computer aided applications gives more precise and faster results. In addition, it eliminates the factors that reduce the accuracy of diagnosis during the classification and segmentation of skin lesions. In this study, segmentation and classification operations were performed on 2 open data sets. First, various pre-processes such as resizing and DullRazor algorithm were applied to dermoscopy images. The preprocessed images are segmented with Otsu Threshold or Active Contour methods. Then, 13 different features of the segmented images, such as texture, shape and color, were extracted and the features were classified by machine learning methods such as k-NN, SVM, easy ensemble classifier, RUSBoost Classifier, Balanced Bagging Classifier. In addition, it was classified with deep learning architectures ResNet, EfficientNet and CoAtNet, and segmentation operations were carried out with U-Net, SegNet and DeepLabv3+.