Detection of Skin Diseases from Dermoscopy Image Using the combination of Convolutional Neural Network and One-versus-All

Abstract

The use of deep learning in the field of image processing is increasing. In this study, a new method based on Convolutional Neural Network is proposed to detect skin diseases automatically from Dermoscopy images. Skin cancer is one of the common diseases in the community. In this article, the skin images were taken from the data HAM10000 dataset prepared by Philipp Tschandl. There are seven classes in the skin disease data set; Actinic keratoses and intraepithelial carcinoma, Basal cell carcinoma, Benign keratosis, Dermatofibroma, Melanoma, Melanocytic type and Vascular lesions. To classify the skin diseases automatically, two different methods have been proposed: i) Alone Convolutional Neural Network model, and ii) the combination of Convolutional Neural Network and one- versus- all. In the proposed method, we have not used any pre-processing method to classify them. The raw dermatology images taken from the dataset have been given to the input of Convolutional Neural Network and then trained and tested by these images. In the second proposed method, seven different models having two-classes have been composed and then combined with the one-versus-all approach. While alone, Convolutional Neural Network obtained 77% classification accuracy in the detection of skin disease with seven classes, the combination of Convolutional Neural Network and one-versus-all approach achieved 92.90% accuracy. The obtained results have shown that the proposed method is very promising in the classification of skin disease from Dermoscopy images.