Automated detection and classification of psoriasis types using deep neural networks from dermatology images

Abstract

Psoriasis is a chronic inflammatory disease that significantly affects the patient's living standard. Dermatologists examine the skin lesions visually to determine the condition. Guttate psoriasis, flexural or inverse psoriasis, pustular psoriasis, erythrodermic psoriasis, psoriatic arthritis, etc., are the most common skin conditions affecting people worldwide. Dermoscopic images are used for clinical diagnosis of these conditions, which require a specialized setup with various medical equipment for better understanding. Patients with skin conditions in Pakistan don't bother going to clinics; they might have serious skin conditions. In this study, a light-weighted deep neural network (DNN)-based model has been developed with fewer but learnable parameters. The results obtained with the proposed DNN model have been compared with the state-of-the-art pre-trained models, i.e., Googlenet, InceptionV3, and VGG-19. Benchmarked, publicly available datasets have been used for experiments. The datasets included in the investigation contain RGB images and are converted into YCbCr for better classification. Standard evaluation parameters, i.e., accuracy, specificity, sensitivity, and area under the curve (AUC), have been used to evaluate the proposed DNN-based model. The proposed model for all psoriasis types achieves the best classification performance. The highest one is the case of psoriatic arthritis, where these measures are (accuracy = 99.89%, specificity = 99.08%, sensitivity = 99.0%, and AUC = 0.99). Using the proposed model, it is demonstrated that YCbcr is the best color space for identifying psoriasis and its types.