Analysis of surface non-uniformity for melanomas using irregularity and asymmetry of virtual contours

Abstract

Surface non-uniformity of skin tumors is an important hint in the early diagnosis of malignant melanomas in clinic. A virtual contour modeling the surface inhomogeneity is proposed in the paper to evaluating region inhomogeneity of skin tumors. A novel feature cluster, combined irregular and asymmetric descriptors of virtual contour, is formed to describing 2D surface unevenness. The surface unevenness of intensity distributions of a skin tumor is firstly converted to the virtual contour's irregularity. Then features on irregularity and asymmetry are extracted. Furthermore, the features are selected and combined by Hausdorff Distance metrics to implement the classification of skin tumors using the irregularity and asymmetry of the virtual contour. The experiments show that the surface inhomogeneity in the 2D space could be effectively described through a modified virtual contour model in the 1D space of object contours. Meanwhile, the classifier performances are improved based on jointed irregular and asymmetric descriptors. An effective way of analyzing the non-uniform for target surfaces is proposed.