Comparison of distance metric measures to differentiate non-dermoscopic skin lesions

Abstract

The ABCD rule of dermoscopy is a way for a clinical diagnosis of melanoma that is based on low-level features such as asymmetry, border, color, and differential structures. They are currently used for solving various tasks related to the diagnosis of skin lesions. In this study, we are focusing on the diameter of the skin lesions (D criterion) as a tool to identify the most effective distance metric measures for the evaluation of pigmented lesions. The following distance measures were used: Euclidian and quasi-Euclidian, city block, chessboard distances. They were compared with a manual distance (as a ground truth data) for accurate results.  We sorted 140 skin lesion images into two categories (nevi and melanomas) based on diameter and diagnosis. To validate and assess the clinical utility of the selected distance metrics, a receiver operating characteristic (ROC) curve and AUC (area under the curve) together with Youden were employed. Data from the ROC and AUC analysis suggest the chessboard distance as the sensitivity measure. It has higher values of sensitivity, specificity, and accuracy, for both nevi and melanoma. However, the diameter criterion (D) cannot be a stand-alone feature for melanoma detection but can be a valuable criterion and of increased suspicion for melanoma.