平扫CT特征联合纹理分析鉴别经治乳腺癌患者肺部单发转移癌与原发腺癌北大核心CSCD

目的评估基于平扫CT特征及纹理分析联合模型鉴别经治乳腺癌患者单发肺内病灶为乳腺癌孤立性肺转移(SPMBC)或原发性肺腺癌(PLA)的价值。方法回顾性分析111例经治乳腺癌伴术后病理证实的SPMBC(SPMBC组,n=54)或PLA(PLA组,n=57)患者,对比其肺内病灶CT表现。采用MaZda软件提取并筛选CT所示病灶最佳纹理参数,以集成算法为分类器,针对组间差异有统计学意义的CT表现、最佳纹理参数或联合二者分别构建CT特征模型、纹理特征模型及联合模型,以鉴别SPMBC与PLA;以5折交叉验证法及受试者工作特征(ROC)曲线评估各模型的诊断效能。结果CT特征模型、纹理特征模型及联合模型鉴别经治乳腺癌患者伴SPMBC与PLA的平均曲线下面积分别为0.64±0.08、0.82±0.07及0.85±0.05。CT特征模型的平均分类准确率(0.60±0.09)低于纹理特征模型及联合模型(0.78±0.09、0.82±0.08,t=-3.14、-4.06,P均<0.05),纹理特征模型平均分类准确率与联合模型差异无统计学意义(t=-0.66,P>0.05)。结论平扫CT特征联合纹理分析有助于鉴别诊断经治乳腺癌患者SPMBC与PLA。

Plain CT features combined with texture analysis for differentiating solitary pulmonary metastasis and lung adenocarcinoma inpatients with breast cancer after treatments

Objective To explore the value of the combined model based on plain CT features and texture analysis for differentiating solitary pulmonary metastasis of breast cancer (SPMBC) or primary lung adenocarcinoma (PLA) in patients with breast cancer after treatments. Methods Totally 111 patients with breast cancer after treatments and pathologically confirmed SPMBC (SPMBC group, n=54) or PLA (PLA group, n=57) were retrospectively analyzed. CT findings of solitary lung nodules were compared between groups. The best CT texture parameters of lesions were extracted and screened with MaZda software. Taken integrated algorithm as the classifier, CT features model, texture features model and combined model were constructed based on CT features with significant difference between groups, the best CT texture parameters as well as their combination, respectively, to differentiate SPMBC and PLA. Five-fold cross-validation and receiver operating characteristic (ROC) curves were used to evaluate the diagnostic performances of each model. Results The average area under the curve of CT feature model, texture feature model and combined model in differentiating SPMBC and PLA was 0.64±0.08, 0.82±0.07 and 0.85±0.05, respectively. The average classification accuracy of CT feature model (0.60±0.09) was lower than that of texture feature model and combined model (0.78±0.09, 0.82±0.08, t=-3.14, -4.06, both P<0.05), while there was no significant difference of the average classification accuracy between texture feature model and combined model (t=-0.66, P>0.05). Conclusion Plain CT features combined with texture analysis were helpful to differentiating SPMBC and PLA in patients with breast cancer after treatments.