探讨扩散加权成像序列(DWI)图像纹理分析鉴别诊断乳腺良恶性肿瘤的价值。

回顾性分析泰州市人民医院手术病理证实的28例乳腺良性肿瘤与28例恶性肿瘤的DWI图像的影像特征及纹理特征。采用MaZda软件提取所有患者DWI图像中肿瘤病灶的直方图与灰度游程矩阵参数，包括均值(mean)、方差(variance)、偏度(skewness)、峰度(kurtosis)和第1、10、50、90、99百分位数(Pere.1%、Pere.10%、Pere.50%、Pere.90%、Pere.99%)及短游程因子(SRE)、长游程因子(LRE)、灰度不均匀度(GLNU)、游程长不均匀度(RLNU)、游程中的图像分数(fraction)(包括水平、垂直、45dgr、135dgr 4个方向)，采用独立样本t检验(正态分布数据)或非参数检验Mann-Whitney U检验(偏态分布数据)分析良恶性肿瘤病灶DWI图像纹理参数的差异，提取差异有统计学意义的纹理特征参数，使用ROC曲线分析有统计学意义的纹理参数鉴别良恶性肿瘤的诊断效能，运用多变量Logistic回归分析对差异有统计学意义的纹理参数进行建模并绘制ROC曲线评价模型效能。

直方图参数中的方差(variance)与灰度游程矩阵参数中的游程长不均匀度(RLNU)(包括水平、垂直、45dgr、135dgr 4个方向)在两组间的差异有统计学意义(P<0.05)，其中游程长不均匀度水平方向(HRLNU)以447.5517为阈值时诊断效能最佳，对应的AUC、灵敏度和特异度分别为0.874、85.71%、78.58%；通过对差异有统计学意义的纹理特征参数建立多参数Logistic回归诊断模型，对应的AUC、灵敏度及特异度为0.940、96.40%、82.10%。

DWI图像纹理分析鉴别诊断乳腺良恶性肿瘤具有良好的应用价值。

To explore the value of texture analysis of diffusion-weighted imaging(DWI)sequence in differential diagnosis of benign and malignant breast tumors.

The image and texture features of DWI images of 28 cases of benign breast tumors and 28 cases of malignant breast tumors confirmed by surgery and pathology in Taizhou People′s Hospital of Jiangsu Province were retrospectively analyzed.MaZda software was used to extract histogram and gray-scale run-length matrix parameters of tumors in DWI images of all patients, including mean, variance, skewness, kurtosis, Pere.1%, Pere.10%, Pere.50%, Pere.90%, Pere.99% and short run-length emphasis(SRE), long run-distance factor(LRE), grey-level non-uniformity(GLNU), run-length non-uniformity(RLNU), image fraction(including horizontal, vertical, 45dgr and 135dgr directions)in run-length were tested by independent sample t test(normal distribution data)or non-parametric Mann-Whitney U test(skewed distribution data). The difference of texture parameters in DWI images of benign and malignant tumors was analyzed, and the texture characteristic parameters with statistical significance were extracted.The diagnostic efficiency of texture parameters was identified by using receiver operating characteristic(ROC)curve analysis.The texture parameters with statistical significance were modeled by using multivariate logistic regression analysis and ROC curve was drawn to evaluate the effectiveness of the model.

The variance of histogram parameters and RLNU(including horizontal, vertical, 45dgr and 135dgr)of gray run-length matrix parameters had significant differences between the two groups(P<0.05). The diagnostic efficiency of horizontal run-length non-uniformity(HRLNU)was the best when the threshold value was 447.5517, and the corresponding area under the curve(AUC), sensitivity and specificity scores were 0.874, 85.71%, 78.58%.The diagnostic model of multi-parameter logistic regression was established through texture feature parameters with statistical significance, and the corresponding AUC, sensitivity and specificity were 0.940, 96.40% and 82.10%.

DWI image texture analysis has good application value in differential diagnosis of breast benign and malignant tumors.