The use of deep learning-based methods in the diagnosis of dermatological diseases

Psoriasis is an important skin disease with an increasing prevalence that has a significant impact on human life and reduces living standards. There are various types of this skin disease and accurate diagnosis of its types is of great importance in terms of treatment and improving the quality of life of patients. In this study, a new method was developed in addition to the classification of psoriasis in the literature. With this method, it is possible to classify psoriasis disease with a higher success rate. In this new method, three deep learning models (DenseNet-121, EfficientNetB0 and ResNet50), which are frequently used in the literature, were trained with a clinically unique dataset by combining multiple models. The three models were combined to create an ensemble learning model. With this model, psoriasis disease types were classified better and successfully. The dataset used in the study was created from the images of patients treated at the Dermatology Outpatient Clinic of Fırat University Hospital. The dataset contains images of four types of psoriasis, namely Generalized, Guttate, Plaque and Pustular Psoriasis, labeled by expert doctors. Among the three deep learning models that were initially trained and tested separately, the most successful results were obtained with the DenseNet-121 model. The ensemble learning model, which is obtained by combining the last layer features of the three models, outperformed the other three models in most performance metrics and achieved 93% accuracy. In this study, a dataset with four classes was identified with an accuracy rate of 93%.