A novel ai-based approach for detecting skin diseases from histopathological images

Skin diseases are highly prevalent, affecting billions of people worldwide. Diagnosing skin diseases involves photographing the relevant area and taking tissue samples. Pathologists can examine tissue samples and make a diagnosis, which can be time-consuming. In this thesis, we used machine learning and deep learning methods to classify two different datasets for Lichen Sclerosus, Morphea, and Cutaneous Small Vessel Vasculitis, using images obtained from the Department of Pathology at Düzce University Faculty of Medicine. The unique characteristics of the datasets make this a pioneering study in the literature. Reinhard normalization, Gaussian filtering, CLAHE histogram equalization, and data augmentation were applied to the datasets. Texture features were then highlighted using Gray Level Co-occurrence Matrix (GLCM) and histogram-based statistical methods. In the study, the first dataset consists of images obtained with a digital microscope, and when transfer learning methods were applied to this dataset, the VGG16 model achieved an accuracy of 97% and the ResNet50 model achieved an accuracy of 98%. The second dataset consists of images from the Full Slide Imaging system, which enables high-quality image merging of tissue samples, used for the creation of Lichen Sclerosis and Morphea. Further processing on the second dataset using machine learning methods resulted in up to 97% accuracy using Artificial Neural Networks and Support Vector Machines. The studies have resulted in an approach that will support pathologists in making accurate and rapid diagnoses.