Computer-Aided Diagnosis Based on DenseNet201 Architecture for Psoriasis Classification

Psoriasis is identified as a common, chronic, and inflammatory skin disease. The progression of the psoriasis could lead to significant complications, which makes early detection very crucial. It is possibly to recognize the psoriasis by visually inspecting the skin lesions. However, due to the appearance similarity of psoriasis with its variants and atypical cases, it is more difficult to diagnosis the psoriasis accurately. Hence, a computer-aided diagnosis (CAD) system has the potential to help dermatologists improve the quality of the diagnosis and treatment. State-of-the-art methods for the skin lesion diagnosis focus on convolutional neural network (CNN). It has shown promising performances for medical image recognition problem. Therefore, in this chapter, we propose a classification system of psoriasis lesions based on DenseNet201 model. Extensive experiments are conducted on a collection of 4212 images from DermNet, including psoriasis, eczema, rosacea, and healthy skin images. The obtained results demonstrate that the DenseNet201 outperforms the InceptionResNetV2, NasNetMobile, ResNet101, EfficientNetB4, and EfficientNetB6 models. It yields a sensitivity of 97.48%, a specificity of 99.42%, a precision of 97.27%, a recall of 97.01%, and an accuracy of 97.14%. The achieved findings can contribute in classifying the severity of psoriasis.