摘要
目的:探究瘤内和不同瘤周影像组学特征预测肺腺癌患者间变性淋巴瘤激酶(ALK)突变状态的价值。方法:回顾性收集温州医科大学附属第五医院2016年1月至2023年9月及嘉兴大学附属第一医院2018年1月至2023年9月经病理证实为肺腺癌的患者病例资料共362例,所有病例均在治疗前进行胸部平扫CT检查。采取7:3的比例将温州医科大学附属第五医院223例患者随机分为训练集(156例)和内部测试集(67例),将嘉兴大学附属第一医院139例患者作为外部测试集。使用人工智能辅助诊断建模分析软件提取瘤内(GTV)、瘤周0~3 mm(PTV_(0~3mm))、瘤周-3~3 mm(PTV_(-3~3mm))、瘤周0~6 mm(PTV_(0~6mm))、瘤内联合瘤周0~3 mm(GTV+PTV_(0~3mm))和瘤内联合瘤周0~6 mm(GTV+PTV_(0~6mm))的影像组学特征,依次采用Z-score、t检验、最大相关最小冗余算法和最小绝对收缩和选择算子回归进行降维,筛选出具有鉴别性的特征子集,极端梯度上升算法构建组学模型;利用多因素Logistic回归筛选临床信息中的独立危险因素,以此构建临床模型;以验证集中AUC最高的分类器作为最佳瘤周组学模型,计算相应的影像组学评分(Rad-score),并联合临床特征构建综合模型。采用AUC、灵敏度、特异度、准确度评价模型的效能。结果:筛选后保留GTV中24个特征、PTV_(0~3mm)中26个特征、PTV_(-3~3mm)中26个特征、PTV_(0~6mm)中7个特征、GTV+PTV_(0~3mm)中12个特征、GTV+PTV_(0~6mm)中29个特征。在组学模型中,PTV_(-3~3mm)模型表现出最佳性能,在训练集、内部测试集、外部测试集中的AUC分别为0.830、0.785、0.807。临床分期、结节密度、胸膜凹陷征是预测ALK状态的临床危险因素。由上述因素构成的临床模型在训练集、内部测试集、外部测试集中的AUC分别为0.712、0.692、0.714。联合Radscore和临床危险因素构建综合模型,其在训练集、内部测试集、外部测试集中的AUC分别为0.874、0.855、0.
Objective:To investigate the value of intratumoral and different peritumoral radiomic features in predicting the mutation status of anaplastic lymphoma kinase(ALK)in patients with lung adenocarcinoma.Methods:A total of 362 patients pathologically confirmed with lung adenocarcinoma were retrospectively collected from January 2016 to September 2023 in the Fifth Affiliated Hospital of Wenzhou Medical University and from January 2018 to September 2023 in the First Hospital of Jiaxing University,all subjected to chest scanning CT examination before treatment.A 7:3 ratio was adopted to randomly divide 223 patients in the Fifth Affiliated Hospital of Wenzhou Medical University into a training set(156 cases)and an internal testing set(67 cases),while 139 patients in the First Hospital of Jiaxing University served as the external testing set.The radiomic features of gross target volume(GTV),peritumor volume of 0-3 mm(PTV_(0~3mm)),PTV_(-3-3mm),PTV_(0-6mm),GTV+PTV_(0~3mm),and GTV+PTV_(0-6mm)were extracted using artificial intelligence-assisted diagnosis modeling software,and dimensionality reduction was performed with Z-score,t-test,maximum relevance minimal redundancy,least absolute shrinkage and selection operator to screen a subset of discriminative features,and extreme gradient boosting was used to construct the radiomic model.The independent risk factors in the clinical information were screened out by multifactorial logistic regression to construct the clinical model.The classifier with the highest AUC in the testing set was used as the best peri-tumor radiomic model,the corresponding radiomic score(Rad-score)was calculated,and a comprehensive model was constructed in conjunction with clinical risk features.The efficacy of the model was evaluated in terms of the AUC,sensitivity,specificity,and accuracy.Results:After screening,24 features in GTV,26 features in PTV_(0~3mm),26 features in PTV_(-3-3mm),7 features in PTV_(0-6mm),12 features in GTV+PTV_(0~3mm),and 29 features in GTV+PTV_(0-6mm)retained.Of all the radiomic m
作者
冯烨
陈炜越
钱旭升
洪晨晨
刘柏君
赖林强
李燕珺
陈敏江
涂建飞
纪建松
FENG Ye;CHEN Weiyue;QIAN Xusheng;HONG Chenchen;LIU Bojun;LAI Linqiang;LI Yanjun;CHEN Minjiang;TU Jianfei;JI Jiansong(Zhejiang Key Laboratory of Imaging and Interventional Medicine,the Fifth Affiliated Hospital of Wenzhou Medical University,Lishui Municipal Central Hospital,Lishui 323000,China;Suzhou Institute of Biomedical Engineering and Technology,Chinese Academy of Sciences,Suzhou 215163,China;Department of Intervention,the Fifth Affiliated Hospital of Wenzhou Medical University,Lishui Central Hospital,Lishui 323000,China;Department of Radiology,the First Hospital of Jiaxing University,Jiaxing 314000,China)
出处
《温州医科大学学报》
CAS
2024年第12期996-1003,共8页
Journal of Wenzhou Medical University
基金
国家卫生健康委员会科研基金项目(WKJ-ZJ-2452)。
关键词
肺腺癌
影像组学
瘤周
间变性淋巴瘤激酶
lung adenocarcinoma
radiomics
peritumoral
anaplastic lymphoma kinase
