基于高分辨率CT扫描的孤立性肺结节高危预测模型的建立与验证  

Establishment and validation of a prediction model for high-risk patients with solitary pulmonary nodules based on high-resolution CT scans

作　　者：张蜜 谈炎欢 钱芯 吴志超 耿悦 邵宇[1] 俞阳 李俊晨[1] ZHANG Mi;TAN Yanhuan;QIAN Xin;WU Zhichao;GENG Yue;SHAO Yu;YU Yang;LI Junchen(Department of Radiology,Changshu Hospital Affiliated to Nanjing University of Chinese Medicine,Changshu 215500,China;Department of Thoracic Surgery,Changshu Hospital Affiliated to Nanjing University of Chinese Medicine,Changshu 215500,China)

机构地区：[1]南京中医药大学常熟附属医院放射科,江苏常熟215500 [2]南京中医药大学常熟附属医院胸外科,江苏常熟215500

出　　处：《中国医学物理学杂志》2023年第2期190-195,共6页Chinese Journal of Medical Physics

基　　金：常熟市卫生和计划生育委员会科技计划资助性青年项目(cswsq201804)。

摘　　要：目的:筛选孤立性肺结节(SPN)高危病变的高分辨率CT危险特征因素,建立SPN高危预测模型,并对模型进行内部验证。方法:回顾性分析317例SPN患者的胸部影像学征象,采用随机分组的方法按约6:4比例分为训练集和验证集,通过单因素及多因素分析Logistic逐步回归法筛选出与高危SPN相关的独立危险因素,建立预测模型;通过验证集数据对建立的模型进行验证,绘制受试者工作特征曲线(ROC),评估模型预测价值。结果:单因素分析显示影像学征象位置、结节最大径、CT值、毛刺征、空泡征、分叶征与SPN是否高危存在统计学差异(P<0.05),边界与SPN是否高危无统计学差异(P>0.05)。多因素分析显示位置、结节最大径、CT值、分叶征4个因素是SPN高危的独立预测因子(P<0.05)。通过训练集构建的预测模型为P=ex/(1+ex),其中e为自然对数,x=-7.767+(2.821×位置)+(0.391×结节最大径)-(0.003×CT值)+(3.576×分叶)。训练集受试者工作特征曲线下面积(AUC)为0.932,95%CI为0.892~0.973,最佳截点值为T=0.208,敏感度为91.7%,特异性为85.6%。验证集AUC为0.911,95%CI为0.847~0.975,最佳截点值为T=0.268,敏感度为78.8%,特异性为93.1%。结论:位置、结节最大径、CT值、分叶征是判断SPN高危的独立影响因素,其建立的数学预测模型准确性较高,可在临床进行推广应用。Objective To screen out the high-resolution CT sign of high-risk lesions of solitary pulmonary nodules(SPN) for establishing a prediction model for high-risk SPN, and to conduct an internal validation of the model. Methods The chest imaging signs of 317 patients with SPN were retrospectively analyzed, and the patients were randomly divided into training set and validation set in a ratio of about 6:4. The independent risk factors associated with high-risk SPN were screened out using univariate and multivariate Logistic stepwise regressions, and a prediction model was established. The established model is verified using validation set, and the receiver operating characteristic(ROC) curve is drawn to evaluate the predictive value of the model. Results Univariate analysis showed that there were significant differences between high-and low-risk SPN in the following imaging signs: location, maximum diameter, CT value, spicule sign, vacuole sign, lobulation sign(P<0.05), and that the proportion of clear or unclear boundary in high-risk SPN group was similar to that in low-risk SPN group(P>0.05). Multivariate analysis showed that 4 factors including location, maximum diameter, CT value and lobulation sign were independent predictors for high-risk SPN(P<0.05). The prediction model constructed based on training set was as follow: P=ex/(1+ex), where the e is the natural logarithm, x=-7.767+(2.821×location) +(0.391×maximum diameter)-(0.003×CT value)+(3.576×lobulation). The area under the ROC curve(AUC), 95%CI, the optimal cutoff value,sensitivity, and specificity were 0.932, 0.892-0.973, 0.208, 91.7%, and 85.6% in training set, and 0.911, 0.847-0.975, 0.268,78.8%, and 93.1% in validation set. Conclusion Location, maximum diameter, CT value, and lobulation sign are independent influencing factors for high-risk SPN, and the mathematical prediction model established by these independent influencing factors has high accuracy and is worth to be popularized in clinical practice.

关 键 词：预测模型 孤立性肺结节 CT 列线图 ROC曲线 

分 类 号：R734.2[医药卫生—肿瘤] R814.42[医药卫生—临床医学]