国际肿瘤学杂志

国际肿瘤学杂志››2023,Vol. 50››Issue (11): 655-660.doi:10.3760/cma.j.cn371439-20230704-00124

• 论著 •上一篇下一篇

基于CT影像学特征的恶性肺纯磨玻璃结节患者病理分型预测模型构建

陈郁, 许华, 刘海, 陈士新()

  1. 三二〇一医院医学影像科，汉中 723000
  • 收稿日期:2023-07-04修回日期:2023-10-30出版日期:2023-11-08发布日期:2024-01-11
  • 通讯作者:陈士新 E-mail:1724035899@qq.com

Construction of pathological classification prediction model for malignant pulmonary pure ground-glass nodule patients based on CT imaging

Chen Yu, Xu Hua, Liu Hai, Chen Shixin()

  1. Department of Medical Imaging，3201 Hospital，Hanzhong 723000，China
  • Received:2023-07-04Revised:2023-10-30Online:2023-11-08Published:2024-01-11
  • Contact:Chen Shixin E-mail:1724035899@qq.com

RichHTML

8

PDF

27

摘要：

目的构建恶性肺纯磨玻璃结节（pGGN）患者基于CT影像学特征的病理分型预测模型并分析其预测效能。方法回顾性纳入2018年1月—2022年12月于三二〇一医院行手术治疗并具有病理组织学检查结果的肺pGGN患者193例217个病灶，根据是否为浸润性腺癌划分为浸润性腺癌组（68例患者，73个病灶）和非浸润性腺癌组（125例患者，144个病灶）；比较两组患者的临床特征资料及CT影像学参数指标；采用logistic多因素回归分析恶性肺pGGN确诊为浸润性腺癌的危险因素；构建恶性肺pGGN病理分型logistic预测模型并利用受试者操作特征（ROC）曲线分析其预测效能。结果浸润性腺癌组与非浸润性腺癌组毛刺征比例分别为34.25%（25/73）和5.56%（8/144）；内部血管征比例分别为93.15%（68/73）和18.75%（27/144）；空气支气管征分别为67.12%（49/73）和12.50%（18/144），差异均有统计学意义（χ2=30.93，P<0.001；χ2=108.95，P<0.001；χ2=67.72，P<0.001）；浸润性腺癌组结节平扫最大CT值（-527.82±72.95）HU显著高于非浸润性腺癌组（-592.79±86.47）HU，差异有统计学意义（t=-5.50，P<0.001）。多因素分析结果显示，毛刺征（OR=8.93，95%CI为1.99~39.97，P=0.004）、空气支气管征（OR=8.16，95%CI为2.91~22.86，P<0.001）、内部血管征（OR=48.39，95%CI为14.81~158.07，P<0.001）、平扫最大CT值（OR=1.01，95%CI为1.00~1.02，P=0.001）均是恶性肺pGGN确诊为浸润性腺癌的独立影响因素。ROC曲线分析显示，利用毛刺征、空气支气管征、内部血管征、平扫最大CT值、logistic回归模型P值对于恶性肺pGGN病理分型情况进行预测，最佳截断值分别为0.50、0.50、0.50、-547.23 HU、0.46，曲线下面积分别为0.64、0.77、0.87、0.69、0.96，敏感性分别为34.25%、67.12%、93.15%、82.19%、89.04%，特异性分别为94.44%、87.50%、81.25%、46.53%、92.36%，约登指数分别为28.69%、54.62%、74.40%、28.72%、81.40%。结论恶性肺pGGN患者中合并毛刺征、空气支气管征、内部血管征及平扫最大CT值较大者确诊为浸润性腺癌的风险较高；基于毛刺征、空气支气管征、内部血管征及平扫最大CT值构建的预测模型在辅助恶性肺pGGN病理分型鉴别诊断中显示出良好的预测效能。

关键词:肺肿瘤,体层摄影术，X线计算机,病理学,预测

Abstract:

ObjectiveTo construct the pathological classification prediction model for malignant pulmonary pure ground-glass nodule（pGGN） patients based on CT imaging and to analyze the predictive efficacy.MethodsA total of 193 pulmonary pGGN patients with histopathological findings who underwent surgical treatment in 3201 Hospital from January 2018 to December 2022 were retrospectively included，with 217 lesions. All patiens were divided into invasive adenocarcinoma group （68 patients，73 lesions） and non-invasive adenocarcinoma group （125 patients，144 lesions） based on whether they were invasive adenocarcinoma； The clinical feature data and CT imaging parameters were compared between the two groups； Multivariate logistic regression analysis was used to analyze the risk factors of malignant lung pGGN diagnosed as invasive adenocarcinoma； A logistic prediction model for pathological classification of malignant lung pGGN was constructed to analyze its predictive efficacy using receiver operator characteristic （ROC） curves.ResultsThe percentages of burr signs in invasive adenocarcinoma group and non-invasive adenocarcinoma group were 34.25% （25/73） and 5.56% （8/144），respectively； The proportion of internal vascular signs was 93.15% （68/73） and 18.75% （27/144），respectively； The air bronchial signs were 67.12% （49/73） and 12.50% （18/144），respectively，with statistically significant differences （χ2=30.93，P<0.001；χ2=108.95，P<0.001；χ2=67.72，P<0.001）. The maximum CT value of nodular plain scan in invasive adenocarcinoma group （-527.82±72.95）HU，was significantly higher than that in non-invasive adenocarcinoma group （-592.79±86.47）HU，with a statistically significant difference （t=-5.50，P<0.001）. The results of multivariate analysis showed that spicule sign （OR=8.93，95%CI： 1.99-39.97，P=0.004），air bronchial sign （OR=8.16，95%CI： 2.91-22.86，P<0.001），internal vascular sign （OR=48.39，95%CI： 14.81-158.07，P<0.001） and the maximum CT value of plain scan （OR=1.01，95%CI： 1.00-1.02，P=0.001） were independent factors for the diagnosis of malignant pulmonary pGGN as invasive adenocarcinoma. Using burr sign，air bronchogram sign，internal vascular sign，maximum CT value of plain scan，and logistic regression modelP-value to predict the pathological classification of malignant lung pGGN，the optimal cutoff values were 0.50，0.50，0.50，-547.23 HU，0.46，and the area under the curve was 0.64，0.77，0.87，0.69 and 0.96，respectively. The sensitivity was 34.25%，67.12%，93.15%，82.19% and 89.04%，and the specificity was 94.44%，87.50%，81.25%，46.53% and 92.36%，respectively，with the Jordan index being 28.69%，54.62%，74.40%，28.72% and 81.40%.ConclusionPatients with malignant pulmonary pGGN who have concomitant spicule sign，air bronchial sign，internal vascular sign，and maximum CT value on plain scan have a higher risk of being diagnosed with invasive adenocarcinoma； The predictive model constructed based on spicule sign，air bronchial sign，internal vascular sign，and maximum CT value on plain scan has shown good predictive performance in assisting the differential diagnosis of malignant pulmonary pGGN pathological classification.

Key words:Lung neoplasms,Tomography technology，X-ray computer,Pathology,Prediction