基于局部分形维的黑色素瘤轮廓不对称性和不规则性分析

Analysis of Irregularity and Asymmetry for Melanoma′s Boundaries Based on Local Fractals

针对计算机辅助早期诊断黑色素瘤中皮肤肿瘤轮廓的结构不规则性和不对称性检测问题,本研究提出了多尺度局部分形维算法。在不断去除轮廓纹理不规则部分的同时提取多尺度下轮廓不规则程度的特征量,建立了具有重要诊断意义的结构不规则性度量的新方法;同时提出基于局部分形维描述轮廓不对称的特征度量。试验表明所提取的特征较全局分形维和基于轮廓空间不规则性度量的方法在多尺度上具有显著的不规则性信息;轮廓不对称度量使用局部分形维能显著地表现出黑色素瘤轮廓的不对称程度,可提供多种不对称信息描述,增强了基于轮廓不规则性和不对称性特征鉴别黑色素瘤的能力。

In order to tackle the problem of detecting structural irregularity and asymmetry of lesion boundaries in the early CAD diagnosis of melanomas, a multiscaled local fractal algorithm was proposed in the paper. The sequent removing textural irregularity from skin lesion boundaries was performed and features of boundary irregularity under muhiscale were extracted. A novel method of structural irregularity measurements, as a key diagnostic factor, was given and boundary asymmetry measurements using local fractals were presented as well. Experiments showed that the features extracted by the proposed method had advantages over global fractal and geometric measures in multiscaled boundary irregular information. The local fractal based asymmetry measures obviously strengthened asymmetric degrees of melanoma lesions to provide multiple irregularity descriptors and enhanced the capability of discriminating skin tumors on characteristics of boundary irregularity and asymmetry.