Description of Boundary Irregularity On Multi-Scale Local FD for Melanomas

Boundary irregularity of melanomas is one of most important clinical indicators to discriminate from benign moles. Conventional fractal dimension (FD) is utilized to describe boundary complexity with weaker criminative power. The objective of the paper is to explore new boundary irregularity descriptors at different scales. In the paper, there are two novel measures of boundary´s structural irregularities and multiscale features related to boundary roughness are proposed using an integrated quantitative model combining Gaussian filtering and local FDs. There are two key elements in the scheme: (1) Local fractal distributions are taken as a tool to describe local irregularity along boundaries at different scales with meaningful diagnosis features; (2) multi-scaled boundaries generated by Gaussian filters could produce boundary´s outlines important to measure of structural irregularity. The advantage of the approach is that feature clusters of boundary irregularity are formed at different scales, so that extracted measurements from multi-scales could provide effective descriptors compared to the conventional FD and other boundary based methods. Experiments show that the proposed irregular descriptors not only enhance finer expressions of boundary complexity but also could effectively discriminate melanomas among moles using statistical features of boundary roughness at different scales.