Dermatological expert system implementing the ABCD rule of dermoscopy for skin disease identification

Abstract

Doctors and radiologists generally follow the standard ABCD rule of dermoscopy for differentiating the malignant and benign skin lesions. The estimation of the dermoscopic score by visual inspection only, may lead to the inaccurate diagnosis of the disease at an early stage. In this work, the ABCD attributes have been improvised and quantified in a dermatological expert system (DermESy) for the differentiation of malignant and benign lesions. DermESy, a rule based expert system has been developed by implementing dermatologist’s knowledge with proper quantification of the dermoscopic findings. Using DermESy, the dermoscopic images have been categorized as malignant, benign and suspicious lesions based on the estimated total dermoscopic score (TDS), similar to the findings of an expert. To estimate the TDS, shape, brightness and color variations are considered to modify the ‘A’ score. The color information extraction algorithm is introduced to extract significant color regions to quantify the ‘C’ score. To find the appropriate ‘D’ score of a skin lesion, dermoscopic structures segmentation algorithms have been introduced. In this work, the ABCD rule of dermoscopy has been improvised by considering the spatial properties of dermoscopic structures for improved identification of malignant lesions. An explanatory subsystem is implemented in DermESy to assist the dermatologist with proper in-detail visualization. DermESy has differentiated the benign and malignant skin lesions with 97.69% sensitivity, 97.97% specificity and 97.86% accuracy. The TDS evaluated by DermESy is verified and compared against expert dermatologist’s TDS scores of same dermoscopy images to establish the reliability and robustness of the proposed system.