常规MRI纹理分析鉴别乳腺良、恶性病变的价值初探

Differentiation of benign and malignant breast lesions using texture analysis of conventional MRI：a preliminary study

目的 探讨常规MRI纹理分析鉴别乳腺良、恶性病变的价值.方法 回顾性分析经手术病理证实,且在术前2周内行双侧乳腺平扫T1WI、T2WI及延时增强T1WI扫描的乳腺病变患者69例的临床资料,其中恶性病变36例,良性病变33例.选取病灶最大层面图像,在MaZda软件中手动勾画病灶ROI,提取病变的纹理特征参数.通过MaZda软件提供的纹理特征选择方法选择最具有鉴别乳腺良、恶性病变价值的纹理特征参数,并采用纹理特征分类分析统计方法对所选的纹理特征参数的鉴别诊断效果进行评估.鉴别诊断结果以误判率的形式表示.MaZda软件提供的纹理特征选择方法包括交互信息（MI）、Fisher系数、分类错误概率联合平均相关系数（POE＋ACC）及3种方法联合（Fisher＋POE＋ACC＋MI联合法,FPM）.MaZda软件提供的纹理特征分类分析统计方法包括原始数据分析（RDA）、主要成分分析（PCA）、线性分类分析（LDA）和非线性分类分析（NDA）.结果 平扫T1WI、T2WI及延时增强T1WI序列中,鉴别乳腺良、恶性病变的纹理特征主要来自T2WI序列,误判率最小为4.35%（3/69）.纹理特征参数选择方法中,MI、Fisher系数和POE＋ACC 3种方法鉴别乳腺良、恶性病变的误判率接近,MI为15.94%~56.52%,Fisher系数为17.39%~56.52%,POE＋ACC为17.39%~56.52%,FPM选择的纹理特征参数鉴别两种病变的误判率最低,为4.35%~53.62%.纹理特征分类分析方法中,NDA区分两种病变的误判率（4.35%~27.54%）明显低于RDA（33.33%~56.52%）、PCA（33.33%~53.62%）和LDA（15.94%~44.93%）3种方法,具有最优的鉴别诊断效果.结论 常规MRI纹理分析可用于鉴别乳腺良、恶性病变,能为乳腺良、恶性病变鉴别诊断提供可靠的依据.

Objective To investigate the diagnostic value of texture analysis derived from conventional MR imaging in differentiating benign and malignant breast lesions. Methods Thirty-six patients with malignant breast lesion and 33 patients with benign breast lesion were retrospectively analyzed in our study. All patients underwent conventional MR imaging including axial T1WI, T2WI, and contrast-enhanced T1WI before surgery. Texture features were calculated from manually drawn ROIs by using MaZda software. The feature selection methods included mutual information （MI）, Fishers coefficient, classification error probability combined with average correlation coefficients （POE ＋ ACC） and the combination of the above three methods（FPM）. These methods were used to identify the most significant texture features in discriminating benign breast lesion from malignant breast lesion. The statistical methods including raw data analysis （RDA）, principal component analysis （PCA）, linear discriminant analysis （LDA） and nonlinear discriminant analysis （NDA） were used to distinguish malignant breast lesion from benign breast lesion. The results were shown by misclassification rate. Results In the three kinds of sequences, the texture features for differentiating malignant breast lesion and benign breast lesion were mainly from T2WI which had the lowest misclassification rate 4.35%（3/69）. The misclassification rates of the feature selection methods were similar in MI, Fisher coefficient and POE＋ACC （15.94%to 56.52%for MI;17.39%to 56.52%for Fisher coefficient and 17.39%to 56.52%for POE＋ACC）. However, the misclassification rate of the combination of the three methods （4.35%to 53.62%for FPM） was lower than that of any other kind of method. In the statistical methods, NDA （4.35% to 27.54%） had lower misclassification rate than RDA （33.33% to 56.52%）, PCA （33.33% to 53.62%） and LDA （15.94% to 44.93%）. Conclusion Texture analysis of conventional MR imaging can provide reliably objective basis for differentiating benign from malignant breast lesions.