标题：HRCT纹理分析在肺腺癌组织学分化程度预测评估中的应用

      作者：曾伟胜，黄育鑫，林杰    普宁市人民医院影像中心，广东 普宁 515300

      卷次： 2022年33卷14期

      【摘要】 目的 探究高分辨率 CT(HRCT)纹理分析在肺腺癌组织学分化程度预测评估中的应用价值。方法 回顾性分析 2020年 1月至 2021年 5月普宁市人民医院收治的 60例肺腺癌患者的影像资料，所有患者均在术前进行双期增强扫描与胸部HRCT平扫，经后期数据处理获得骨和软组织重建算法平扫肺窗图像，平扫、动脉期和静脉期软组织重建算法纵隔窗共计5组图像。通过软件对感兴趣区(ROI)进行勾画，对病变的纹理特征参数进行提取，主要选择Fisher系数、分类错误概率联合平均相关系数(POE＋ACC)、交互信息(MI)及上述3种方法的联合法(FPM)为主要纹理特征参数。结果 肺腺癌病变的最低误判率为6.67% (4/60)，最低误判率分别出现在软组织重建算法的肺窗与静脉期纵隔窗图的FMP+NDA分类方法上；在平扫纵隔窗图像上提取纹理特征进行鉴别最小误判率为11.67% (7/60)；在所选的纹理特征参数中，病变鉴别最低的错判率在FPM法的纹理特征中，Fisher系数、POE+ACC、MI及FPM鉴别的误判率分别为 10.00%~45.00%、10.00%~51.67%、10.00%~41.67%及 6.67%~51.67%；NDA对两种病变区分的误判率为 6.67%~16.67%，明显低于 RDA的 21.67%~48.33%、PCA的 16.67%~51.67%和LDA的8.33%~38.33%，差异均有统计学意义(P<0.05)。结论 基于HRCT图像的纹理分析可以为肺腺癌的诊断提供有效的信息，有助于提高临床医生对肺腺癌患者的预测评估准确性。
      【关键词】 肺腺癌；肺部肿瘤；高分辨率CT；分化程度；预测评估
      【中图分类号】 R734.2 【文献标识码】 A 【文章编号】 1003—6350（2022）14—1844—04

Application of high resolution CT texture analysis in predicting and evaluating histological differentiation of lungadenocarcinoma.
ZENG Wei-sheng, HUANG Yu-xin, LIN Jie. Imaging Center, Puning People's Hospital, Puning 515300,Guangdong, CHINA
【Abstract】 Objective To explore the application value of high resolution CT (HRCT) texture analysis in theprediction and evaluation of histologic differentiation of lung adenocarcinoma. Methods The imaging data of 60 pa-tients with lung adenocarcinoma admitted in Puning People's Hospital from January 2020 to May 2021 were analyzedretrospectively. All patients underwent two-phase enhanced scanning and chest HRCT plain scanning before operation.After data post-processing, five groups of images, including plain scanning lung window image of bone and soft tissuereconstruction algorithm, and mediastinal window image of plain scanning, arterial and venous soft tissue reconstructionalgorithm were obtained. The Region of Interest (ROI) was outlined by software, and the texture feature parameters of le-sions were extracted. Fisher coefficient, joint average correlation coefficient of classification error probability (POE +ACC), interactive information (MI), and the joint method (FPM) of the above three methods were mainly selected as themain texture feature parameters. Results The lowest false positive rate of lung adenocarcinoma was 6.67% (4/60), whichappeared in the lung window of soft tissue reconstruction algorithm and venous mediastinal window of FMP+NDA classi-fication method of lung window respectively. Minimum misjudgment rate of texture feature extraction from plain scan me-diastinal window images was 11.67% (7/60). Among the selected texture feature parameters, the wrong judgment rate of le-sion identification was the lowest in the texture features of FPM method. The wrong judgment rates of Fisher coefficient,POE+ACC, MI, and FPM identification were 10.00%-45.00%, 10.00%-51.67%, 10.00%-41.67%, and 6.67%-51.67%, re-spectively. The wrong judgment rate of NDA in distinguishing the two lesions was 6.67%-16.67%, which was significant-ly lower than 21.67%-48.33% of RDA, 16.67%-51.67% of PCA, and 8.33%-38.33% of LDA, and the differences werestatistically significant (P<0.05). Conclusion Texture analysis based on HRCT images can provide effective informa-tion for the diagnosis of lung adenocarcinoma, which is helpful to improve the accuracy of clinicians in predicting andevaluating patients with lung adenocarcinoma.
      【Key words】 Lung adenocarcinoma; Lung tumor; High resolution CT; Differentiation degree; Predicting andevaluating