| 黄栎有,王延花,高先聪.基于CT平扫图像纹理分析鉴别浸润性肺腺癌与非钙化结核球[J].中国医学影像技术,2020,36(4):545~549 |
| 基于CT平扫图像纹理分析鉴别浸润性肺腺癌与非钙化结核球 |
| Identification of invasive lung adenocarcinoma and non-calcified lung tuberculoma on plain CT images based on texture analysis |
| 投稿时间：2019-08-21 修订日期：2020-01-15 |
| DOI：10.13929/j.issn.1003-3289.2020.04.015 |
| 中文关键词: 肺肿瘤 诊断 人工智能 体层摄影术,X线计算机 纹理分析 影像组学 |
| 英文关键词:lung neoplasms diagnosis artificial intelligence tomography, X-ray computed texture analysis radiomics |
| 基金项目: |
|
| 摘要点击次数: 3048 |
| 全文下载次数: 876 |
| 中文摘要: |
| 目的 探讨基于CT平扫图像纹理分析鉴别诊断浸润性肺腺癌与非钙化结核球的可行性。方法 回顾性分析52例经病理证实的单发肺结节患者的平扫CT资料，其中31例浸润性肺腺癌，21例非钙化结核球。采用MaZda软件于2种病灶各提取300个纹理特征，之后以费希尔参数法（Fisher）、最小分类误差与最小平均相关系数法（POE+ACC）、相关信息测度法（MI）分别筛选出10个最佳纹理特征，并将其合并得到3种方法联合的最佳纹理特征组合（MPF）。采用线性判别分析（LDA）和非线性判别分析（NDA）对4组最佳纹理特征进行分类，LDA及NDA分别以K-近邻分类器（K-NN）及人工神经网络（ANN）进行分类。分析4组纹理特征鉴别2种病变的最小错误率，比较2组病变间30个最佳纹理特征的差异，并绘制其鉴别2种病变的ROC曲线，计算AUC，评价其诊断效能。结果 对于单组最佳纹理特征，NDA/ANN-Fisher法的错误率最低，为7.69%（4/52）；对于MPF，NDA/ANN-MPF法的错误率最低，为5.77%（3/52）；而NDA/ANN-Fisher法的错误率与NDA/ANN-MPF法差异无统计学意义（χ2=0.15，P>0.05）。2种病变间存在10个纹理特征差异有统计学意义，其中差异熵S（1，1）、差方差S（1，1）及梯度方差的诊断效能较好（AUC=0.71、0.71、0.70），3者间AUC差异无统计学意义（P均>0.05）。结论 基于CT平扫图像纹理分析可较好地区分浸润性肺腺癌和非钙化肺结核球，为鉴别诊断提供可靠的客观依据。 |
| 英文摘要: |
| Objective To investigate the feasibility of differential diagnosis of invasive lung adenocarcinoma and non-calcified lung tuberculoma on CT plain images based on texture analysis. Methods Data of plain CT images of 52 patients with single pulmonary nodules confirmed pathologically were retrospectively analyzed, including 31 cases of invasive lung adenocarcinoma and 21 cases of non-calcified lung tuberculosis. Totally 300 texture features of each kind of lesions were extracted with MaZda software, then 10 optimized texture parameters were selected for texture analysis with fisher coefficient (Fisher), minimization of both probability of classification error and average correction coefficient (POE+ACC), mutual information coefficients (MI) methods, respectively, and the optimal texture features combination combined with three methods (MPF) was obtained. The four groups of optimal texture characteristics were classified using linear discriminant analysis (LDA) and nonlinear discriminant analysis (NDA), while classification of LDA and NDA were performed using K-nearest neighbor classifier (K-NN) and artificial neural network (ANN), respectively. The minimum error probability of 4 groups of texture features in differential diagnosing of 2 kinds of lesions was analyzed, the differences of 30 optimal texture features were compared between 2 kinds of lesions, their ROC curves for identifying 2 kinds of lesions were drawn, and then AUC of the curves were calculated to evaluate their diagnostic performance.Results For single group of optimal texture features, NDA/ANN-Fisher method had the lowest error rate (7.69%[4/52]), while for MPF, the error rate of NDA/ANN-MPF was the lowest (5.77%[3/52]). There was no statistical difference of error rate between NDA/ANN-Fisher and NDA/ANN-MPF method (χ2=0.15, P>0.05). Statistical differences of 10 optimal texture features were noticed between 2 kinds of lesions, among which difference entropy S(1,1), difference variance S(1,1) and gradient variance had good diagnostic efficacy (AUC=0.71, 0.71, 0.70), and their AUC were not statistically different (all P>0.05).Conclusion Based on texture analysis of plain CT images, invasive lung adenocarcinoma and non-calcified lung tuberculosis can be well distinguished, providing objective and reliable basis for differential diagnosis of these two lesions. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
