Abstract

Diagnosis of skin cancer acts as a vital part of the early and accurate detection of skin lesions using dermoscopic imaging. However, the automatic skin lesion diagnosis becomes hard owing to the existence of artifacts, indistinct boundary, poor contrast, and distinct size and shape of the lesion. This study introduces a novel Zernike Moments (ZM) with Logistic Regression Classifier (LRC), named ZM-LRC model for skin lesion diagnosis using dermoscopic images.  The presented ZM-LRC model comprises four stages namely pre-processing, segmentation, feature extraction, and classification. The ZM-LRC model performs image segmentation using Shannon’s Entropy with Brainstorm optimization (BSO) algorithm. Besides, the ZM based feature extraction and LRC oriented classification processes are carried out. Elaborative experimental analysis was carried out on ISIC dataset and the obtained results are investigated with respect to distinct performance measures. The attained simultaion values signified the effective diagnostic outcome of the ZM-LRC model with a higher sensitivity of 95.78%, specificity of 97.86%, and accuracy of 98.56%.