磁共振增强图像三维纹理分析对乳腺良恶性病变的鉴别诊断价值

Three-dimensional texture analysis of contrast enhanced magnetic resonance imaging in the differential diagnosis of benign and malignant breast lesions

目的:探讨基于磁共振增强图像的三维纹理分析方法对乳腺良恶性病变的鉴别诊断价值。方法:回顾性分析经手术病理证实的16例乳腺癌和18例乳腺良性病变患者的临床和MRI资料。磁共振检查在术前1周内完成。采用MaZda软件对早期动态增强图像进行三维纹理分析,提取整个病变的纹理参数,使用Fisher系数、交互信息(MI)、分类错误概率联合平均相关系数(POE+ACC)三种方法获得30个最优纹理参数,进一步对这些纹理参数进行分类分析,方法包括原始数据分析(RDA)、主成分分析(PCA)、线性分类分析(LDA)和非线性分类分析(NDA)。采用SPSS 16.0统计软件比较乳腺良恶性病变的30个纹理参数有间的差异,采用MedCalc 15.8统计软件,对具有统计学意义的纹理参数进行受试者工作特征(ROC)曲线分析。结果:基于早期动态增强图像的三维纹理特征,采用非线性分类分析(NDA)的误判率最低,其中POE+ACC联合非线性分类分析(NDA)的误判率最低,误判率为5.88%。在乳腺病变的30个最优纹理参数中,有10个纹理参数在良恶性组间的差异具有统计学意义(P<0.05),相应的ROC曲线下面积(AUC)为0.717～0.755。结论:磁共振增强图像三维纹理分析方法对乳腺良恶性病变的鉴别诊断具有良好的临床应用价值。

Objective: The purpose of this study was to explore the value of 3D texture analysis of enhanced magnetic resonance imaging in differential diagnosis of benign and malignant breast lesions. Methods: The clinical and MRI data of 16 cases of breast cancer and 18 cases of benign breast lesions were retrospectively analyzed.All cases were confirmed by surgical pathology and magnetic resonance imaging was completed within one week before surgery.MaZda software was used to analyze the texture of early dynamic enhanced 3d images and extract the texture parameters of the entire lesion.30 optimal texture parameters were obtained by use of the methods of Fisher coefficient,mutual information (MI),classification error probability and average correlation coefficient (POE+ACC).Classification analysis was further performed by use of methods including raw data analysis (RDA),principal component analysis (PCA),linear discriminant analysis (LDA) and nonlinear discriminant analysis (NDA).SPSS 16.0 software was used to compare the texture features of benign and malignant breast lesions.MedCalc 15.8 statistical software was used to analyze the receiver operator characteristic curve of texture parameters which were statistically significant. Results: The nonlinear discriminant analysis (NDA) had the lowest misclassification rate,of which POE+ACC combined with non-linear classification analysis (NDA) had the lowest misclassification rate and misclassification rate was 5.88%.Among the 30 optimal texture parameters for benign and malignant breast lesions,10 were found with statistically significant difference ( P <0.05),and the area under the curve (AUC) was 0.717~0.755,respectively. Conclusion: Texture analysis of three-dimensional image of contrast enhanced magnetic resonance has good clinical value in the differential diagnosis of benign and malignant breast lesions.