CT纹理特征对5~10 mm磨玻璃肺结节侵袭性的诊断价值

目的探讨CT纹理特征对5~10 mm磨玻璃肺结节侵袭性的诊断价值。 方法回顾性分析2010年1月至2019年11月扬州大学附属苏北人民医院收治的CT表现为直径5~10 mm的肺结节患者67例。经手术病理证实微浸润性腺癌（MIA）36例,浸润性腺癌（IAC）31例。评估患者肺结节CT影像学特征（实性成分、毛刺、边界）,运用MaZda 4.6软件在薄层CT图像病灶最大层面勾画感兴趣区自动提取肺结节4种纹理特征值,包括灰度直方图、绝对梯度、游程矩阵及共生矩阵。比较MIA患者与IAC患者肺结节CT影像学特征及纹理特征值的差异。运用WeKa 3.8软件对CT纹理特征进行重要性从高到低排序后,依次纳入训练多变量Logistic回归模型,得到纳入不同特征个数对应的曲线下面积（AUC）,并选择最佳AUC值构建接受者操作特征（ROC）曲线,评估该模型区分肺结节侵袭性的诊断效能,采用十折交叉验证防止过度拟合。 结果2组患者肺结节边界、含实性成分以及毛刺征等CT影像学特征差异均有统计学意义。重要性排名前12的CT纹理特征建立多变量Logistic回归模型后诊断效能最佳,模型示其AUC为0.845,特异度、敏感度分别为80.6%、74.2%。对该模型进行十折交叉验证,模型准确性为77.6%。 结论基于CT平扫的纹理分析能有效鉴别5~10 mm磨玻璃肺结节侵袭性。

Diagnostic value of CT texture characteristics in the invasiveness of 5-10 mm ground glass pulmonary nodules

ObjectiveTo investigate the diagnostic value of CT texture features in the invasiveness of 5-10 mm ground glass pulmonary nodules. MethodsA total of 67 patients with 5-10 mm diameter pulmonary nodules confirmed by operation and pathology were analyzed retrospectively from January 2010 to November 2019 in Northern Jiangsu People's Hospital Affiliated to Yangzhou University. All the patients were divided into two groups according to the pathological results (36 cases of microinvasive adenocarcinoma and 31 cases of invasive adenocarcinoma). The image morphology of pulmonary nodules (solid composition, burr, boundary) was evaluated and counted. Four kinds of texture eigenvalues of pulmonary nodules, including gray histogram, absolute gradient, run-length matrix and co-occurrence matrix, were extracted automatically by using MaZda 4.6 software to draw the region of interest at the maximum level of lesions in thin-slice CT images. The differences of morphological and texture features between the two groups of pulmonary nodules were compared. WeKa 3.8 software was used to rank the importance of features from high to low and then incorporated into the training multivariate logistics regression model. The area under the curve (AUC) values corresponding to different number of features were obtained and the best AUC values were selected to construct the receiver operating characteristic (ROC) curve to evaluate the diagnostic effectiveness of the model in distinguishing the invasiveness of pulmonary nodules. 10-fold cross-validation was used to prevent over-fitting. ResultsThere were statistically significant differences in CT imaging characteristics such as pulmonary tubercle boundary, solid component and burr sign between the two groups. The diagnostic efficiency of the top 12 texture features was the best after the establishment of multivariate logistics regression model, the AUC was 0.845, and the specificity and sensitivity were 80.6% and 74.2% respectively. The model was cross-verified by 10% discount, and the accuracy of the model was 77.6%. ConclusionThe texture analysis based on chest CT scan can effectively identify the invasiveness of 5-10 mm ground glass lung nodules.