基于显著小波子带的轮廓结构不规则性检测

摘要

为解决皮肤肿瘤轮廓结构的不规则性表达和特征提取问题,提出一种在重构肿瘤轮廓结构分量上利用局部分形维(LFD)提取轮廓不规则特征的方法.在神经网络中实现肿瘤分类,使用小波分解和Hausdroff Distance确定肿瘤轮廓结构分量所处的频带(显著小波子带),根据重构轮廓结构分量、LFD派生轮廓的不规则性特征对黑色素瘤进行分类.实验结果表明,该方法具有较高的分类准确率、敏感度和特异度.%With respect to visual character descriptions and extractions of boundary structural irregularity, a novel method is proposed in the paper on a reconstructed structural component of a tumour contour using Local Fractal Dimension(LFD). A neural network is utilized to implement skin tumour classifications on boundary structural irregularity, then by using wavelet decomposition and Hausdroff Distance analysis, frequency bands for structural components of tumour boundaries(significant wavelet sub-bands) are determined, and based on the structural components of reconstructed boundaries, features of irregularity which are extracted from LFD are used for melanomas classification. Experimental results show that the method proposed in the paper has advantages in classification accuracy, specialty and sensitivity.