基于^(18)F-FDG PET/CT的影像组学模型对霍奇金淋巴瘤预后的预测价值  

The predictive value of^(18)F-FDG PET/CT based radiomics model for the prognosis of Hodgkin lymphoma

作　　者：罗与 高子涵 王兵兵 王梅云 LUO Yu;GAO Zihan;WANG Bingbing;WANG Meiyun(Department of Medical Imaging,People’s Hospital of Zhengzhou University,Zhengzhou 450003,China;Department of Medical Imaging,Henan Provincial People’s Hospital)

机构地区：[1]郑州大学人民医院医学影像科,郑州450003 [2]河南省人民医院医学影像科

出　　处：《国际医学放射学杂志》2023年第6期639-645,共7页International Journal of Medical Radiology

基　　金：河南省医学科技攻关计划项目(SBGJ202101002)。

摘　　要：目的探讨从霍奇金淋巴瘤(HL)病人基线^(18)F-FDG PET/CT中提取的肿瘤代谢参数、影像组学特征及临床危险因素构建的联合模型对于预测病人3年无进展生存期(PFS)的预后价值。方法回顾性收集治疗前行^(18)F-FDG PET/CT检查的HL病人78例,中位年龄为43.5(26.3,58.5)岁,按7∶3的比例将病人随机分为训练集(55例)和验证集(23例)。从经过预处理的PET和CT影像中提取影像组学特征,采用最小绝对值收敛和选择算子(LASSO)算法结合Cox回归降维筛选最佳的预后特征,计算相应的影像组学评分(Rad-score)。通过单因素和多因素Cox回归分析确定与PFS相关的临床因素,然后分别构建临床、影像组学和联合模型。采用C指数来评价3种模型的预测效能,基于最优模型构建列线图,并采用校准曲线来评估模型的拟合优度。以预测模型计算的风险评分的中位数为截断值将病人分为高风险组和低风险组,采用Kaplan-Meier分析和log-rank检验比较2组病人生存曲线的差异。结果经单因素和多因素Cox回归分析显示Ann Arbor分期、全身肿瘤代谢体积(TMTV)是与PFS相关的临床独立预后因素,以此构建临床模型。经降维筛选出12个最优组学特征构建影像组学模型。此外还构建包含临床危险因素和组学特征在内的联合模型,其在训练集(C指数:0.872,95%CI:0.781~0.963)和验证集(C指数:0.814,95%CI:0.692~0.936)中比单独的临床模型和影像组学模型具有更高的预后效能。将联合模型以列线图可视化,校准曲线显示联合模型预测的3年PFS在训练集和验证集中均与实际的3年PFS有较高的一致性。Kaplan-Meier分析显示高风险组PFS显著低于低风险组,log-rank检验显示均P<0.05。结论结合肿瘤代谢参数、临床因素和影像组学特征的联合模型提高了预测HL病人无进展生存期的预后效能。Objective To investigate the prognostic value of combined model of tumor metabolic parameters and radiomic features extracted from baseline ^(18)F-FDG PET/CT combined with clinical risk factors in predicting 3-year progression-free survival(PFS)in patients with Hodgkin lymphoma(HL).Methods All of 78 patients with HL who underwent ^(18)F-FDG PET/CT examination before treatment were retrospectively collected,with a median age of 43.5(26.3,58.5)years.The patients were randomly divided into a training set(55 cases)and a validation set(23 cases)according to a ratio of 7∶3.Radiomics features were extracted from preprocessed PET and CT images,and the least absolute shrinkage and selection operator(LASSO)algorithm combined with Cox regression dimension reduction was used to select the best prognostic features,and the corresponding radiomics score(Rad-score)was calculated.Univariate and multivariate Cox regression analyses were used to identify clinical factors associated with PFS,and then clinical,radiomics,and combined models were constructed,respectively.The C-index was used to evaluate the prediction efficiency of the three models,the nomogram was constructed based on the optimal model and calibration curves were used to evaluate the goodness of fit of the model.Patients were divided into high-risk and low-risk groups using the median risk score calculated by the prediction model as the cut-off value.Kaplan-Meier analysis and log-rank test were used to compare the difference in survival curves between the two groups.Results Univariate and multivariate Cox regression analysis showed that Ann Arbor stage and total metabolic tumor volume(TMTV)were clinically independent prognostic factors related to PFS,and clinical model was constructed.A total of 12 optimal radiomics features were selected by dimensionality reduction and the radiomics model was constructed.In addition,combined model including clinical risk factors and radiomics features was constructed,which had higher prognostic performance in the training set(C-ind

关 键 词：霍奇金淋巴瘤 正电子发射体层成像 体层摄影术 X线计算机 影像组学 

分 类 号：R733.4[医药卫生—肿瘤] R814.4[医药卫生—临床医学]