Enhancement of Skin Tilt Pattern for Lesion Classification

Y. Ding, L. Smith, M. Smith, J. Sun, and R. Warr (UK)


Skin tilt pattern disruption, preprocessing, Gaussian smoothing, postprocessing, anisotropic nonlinear diffusion


Skin texture pattern analysis has drawn considerable interests over the last 10 years particularly motivated by the observation that the skin texture tends to be disrupted by Malignant Melanoma (MM) and not by Benign Lesions (BL). The authors’ previous work has demonstrated useful MM indicators can be extracted from 3D skin texture, in the form of tilt and slant directions of surface normals, so called skin tilt patterns and skin slant patterns. This paper proposes an effective approach to enhance the skin tilt pattern (disruptions) for lesion classification. This is achieved using both a preprocessing Gaussian filter and a postprocessing step with anisotropic nonlinear diffusion. The experiments are carried out on a total of 33 lesions of which 10 are MM. Experimental results have shown an improvement from 0.74 to 0.80 on Receiver-Operating Characteristic (ROC) area, and a larger inter-class distance between MM and BLs.

Important Links:

Go Back