Automatic Psoriasis Lesion Segmentation from Raw Color Images using Deep Learning

Psoriasis is a chronic skin disease having raised, red, and scaly patches over the different regions of the human body. The segmentation of psoriasis lesions is a pre-requisite step for the psoriasis diagnosis. In current practice, the lesion segmentation of psoriasis is done manually by visual inspection, which is a time-consuming and subjective process. The automatic segmentation of psoriasis lesions is a challenging task in the field of computer vision. This is due to the presence of the complex background and challenging surroundings in the photographic color images. In this paper, a fully automated approach has been proposed for psoriasis lesion segmentation from the color images. A customized and compact U-Net architecture based deep learning model is implemented for the lesion segmentation task. The psoriasis dataset having 350 digital images acquired from about 100 patients is used in the proposed work. The segmentation performance of the proposed model is validated with a five-fold cross-validation technique and achieves average Dice Similarity Index and Jaccard Index of 0.9102 and 0.8371 respectively. The encouraging results will help researchers to extend the work with the larger dataset and hence to adopt for developing the real-time computer-aided diagnostic system for the psoriasis lesion area assessment and its severity assessment.