磁共振弥散加权成像的纹理特征鉴别良恶性单纯微钙化型乳腺隐匿性病灶的价值

目的: 探索使用磁共振弥散加权成像（DWI）的纹理特征鉴别良恶性单纯微钙化型乳腺隐匿性病灶的价值。方法: 回顾性分析2012年10月至2015年12月于浙江省肿瘤医院进行乳腺X射线金属丝三维定位的61例（38例恶性肿瘤、23例良性病变）单纯微钙化型乳腺隐匿性病灶患者的DWI影像。对每个感兴趣区域提取6个直方图特征和16个灰度共生矩阵纹理特征。利用随机森林算法进行特征选择并构建分类器，使用留一法进行交叉验证。采用ROC曲线评价分类器的性能。结果: 筛选获得的性能最佳的特征组合共包含6个特征，分别为直方图中的均值、方差、偏度、熵，以及灰度共生矩阵的0°方向对比度、45°方向相关性，其中，直方图中的均值、方差、偏度和熵在良性病变与恶性肿瘤组间差异有统计学意义（均P < 0.05）。由上述特征构建的分类器鉴别诊断单纯微钙化型乳腺隐匿性病灶良性病变与恶性肿瘤的AUC达到0.76，诊断准确率、敏感度、特异度分别为77.05%、84.21%、65.21%。结论: DWI的纹理特征分析能在一定程度上提高良恶性单纯微钙化型乳腺隐匿性病灶诊断的准确性，直方图中的均值、方差、偏度和熵是极具潜力的影像学标志物。
Abstract: Objective: To evaluate the application of MR diffusion-weighted imaging(DWI) texture features in differentiation of malignant from benign nonpalpable breast lesion for patients with microcalcifications-only in mammography. Methods: The clinical and MR-DWI data of 61 patients with microcalcifications, who underwent three-dimensional positioning of breast X-ray wire from October 2012 to December 2015 in Zhejiang Cancer Hospital, were retrospectively analyzed, including 38 patients with malignant lesions and 23 patients with benign lesions. Two radiologists independently drew the regions of interest (ROI) on DWI for image segmentation, and 6 histogram features and 16 grayscale symbiosis matrix (GLCM) texture features were extracted on each ROI. The random forest algorithm was applied to select the features and built the classification model. The leave-one-out cross-validation (LOOCV) was used to validate the classifier, and the performance of the classifier was evaluated by ROC curve. Results: Six features were selected, including histogram features of mean, variance, skewness, entropy, as well as contrast (0°) and correlation (45°) in GLCM. The histogram features of mean, variance, skewness and entropy were significantly different between the benign and malignant breast lesions (all P < 0.05). The AUC of the model was 0.76, and the diagnostic accuracy, sensitivity and specificity were 77.05%, 84.21% and 65.21%, respectively. Conclusions: The texture feature analysis of DWI can improve the diagnostic accuracy of differentiating benign and malignant breast nonpalpable lesions with microcalcifications-only in mammography. Histogram features of mean, variance, skewness, entropy of DWI may be used as important imaging markers. Key words: Breast neoplasms/diagnosis Breast diseases/diagnosis Diffusion magnetic resonance imaging Diagnosis, differential