A Comparative Study of Convolutional Neural Network Architecture for Efficient Classification of Psoriasis Disease

According to the statistics, 125 million individuals globally have some kind of skin illness. Psoriasis is a skin disorder that often affects the physical appearance of a person. The advancement of machine learning approaches in psoriasis diagnosis and classification has aided dermatologists in diagnosis, classification, and treatment. Convolutional neural networks (CNNs), one of the deep learning models, are capable of extracting middle- and top-level characteristics from input and have been shown to be efficient in acquiring specific features from images. The fundamental benefit of CNN is its ability to extract robust features that are both distortion- and position-invariant for image categorization. A class of CNN models known as EfficientNet-B3 achieves superior performance through effective model scaling. This paper presents the classification of 495 images taken from the publically available dataset of skin images belonging to nine classes of psoriasis types by basic CNN and EfficientNet architecture. A comparison of the two architectures reveals that the accuracy achieved with EfficientNet-B3 for classification is significantly greater than that of the basic CNN architecture.