基于HRCT构建影像组学模型评估IPF患者的GAP分期

A Radiomics Model Based on HRCT to Evaluate GAP Stage of IPF Patients

目的 探究基于高分辨率CT(HRCT)提取影像组学特征构建无侵袭性模型预测特发性肺纤维化(IPF)患者性别-年龄-生理学指标分期(GAP)高低分级。方法 回顾性纳入2016年6月至2021年12月于河南中医药大学第一附属医院就诊的174例IPF稳定期患者为研究对象。严格按照IPF诊断标准进行诊断。所有患者均接受吡非尼酮治疗,部分患者加用本院特色治疗方案隔药物饼灸。搜集所有入组患者治疗前的临床信息指标,肺功能指标,以及治疗前的GAP分级;所有患者均在治疗前行HRCT扫描。对所有患者HRCT影像的“蜂窝状”区域进行手动勾画,并提取影像组学特征,构建影像组学标签(Radscore),并联合临床指标以及肺功能指标构建联合模型。结果 共计入组174例,其中GAP低级别患者73例,GAP高级别患者101例,所有患者按照7∶3比例进行随机分层进入训练组和测试组,IPF患者在训练组和测试组中,GAP高级别患者的6分钟步行距离(6MWD)均高于GAP低级别患者的6MWD,且差异具有统计学意义。取惩罚系数Logλ=0.051构建影像组学标签Radscore时,共保留13个特征。继而纳入训练组不同GAP级别患者间具有差异的临床特征进行临床模型以及联合模型构建,训练组中,联合模型诊断效能[曲线下面积(AUC)=0.818]高于Radscore(AUC=0.801),高于临床模型(AUC=0.607);测试组中,联合模型诊断效能(AUC=0.7798)高于Radscore(AUC=0.717),高于临床模型(AUC=0.635),决第曲线分析法(DCA)分析显示联合模型的临床收益高于临床模型的收益。结论 基于HRCT提取纹理特征参数构建Radscore联合肺功能参数第1秒用力呼气量(FEV1)可协助临床预测患者的GAP分级。

Objective To explore the non-invasive model based on radiomics features extracted from High resolution computed tomography(HRCT)to predict GAP grade in patients with IPF.Methods A total of 174 patients with stable IPF in the First Affiliated Hospital of Henan University of Chinese Medicine from June 2016 to December 2021 were retro-spectively enrolled.The diagnosis is performed strictly according to the IPF diagnostic criteria.All patients were treated with pirfenidone,and some patients were treated with drug cake moxibustion,the characteristic treatment plan of our hospital.The clinical information index,pulmonary function index and GAP grade before treatment were collected.All patients underwent high-resolution CT imaging before treatment.The honeycomb-like regions of HRCT images of all patients were manually de-lineated,Radiomics features were extracted,Radiomics signature(Radscore)was constructed,and combined with clinical indicators and pulmonary function indicators,a joint model was constructed.Results A total of 174 patients were included in this study,including 73 patients with low grade GAP and 101 patients with high grade GAP.All patients were randomly stratified into the training group and the test group according to the ratio of 7:3.The FEV,of IPF patients with low grade GAP was higher than that of GAP patients with high grade.And the difference was statistically significant.When the penalty coefficient logλ=0.064 was used to construct radiomics tag Radscore,a total of 9 features were retained.Then,the clinical characteristics with differences between patients with different GAP levels in the training group were included to construct the clinical model and the combined model.In the training group,the combined model diagnostic performance(AUC=0.872)was higher than the Radscore(AUC=0.823)and the clinical model(AUC=0.714).In the test group,the diagnostic performance of the combined model(AUC=0.768)was higher than that of the Radscore(AUC=0.730)and the clinical model(AUC=0.693).DCA analysis showed that the clinical benefit of the combined model was higher than that of the clinical model.Conclusion The Radscore combined with lung function parameter FEV,based on texture feature parameters extracted from HRCT can help to predict the GAP grade of patients.