基于MRI的放射组学评分和临床病理影像参数预测垂体瘤复发的Nomogram模型研究  

Nomogram model for predicting pituitary adenoma recurrence based on radiomics score and clinicopathological-imaging parameters of multi-sequence MRI

作　　者：杨洪安 张景润 王家兴 黄继蓝 谭永明[1] 吴主强[2] YANG Hong-an;ZHANG Jing-run;WANG Jia-xing(Department of MRI,the First Affiliated Hospital of Nanchang University and Jiangxi Children's Hospital,Nanchang 330006,China)

机构地区：[1]南昌大学第一附属医院MRI室,南昌330006 [2]江西省儿童医院,南昌330006

出　　处：《放射学实践》2023年第7期853-862,共10页Radiologic Practice

基　　金：江西省医学影像临床研究中心项目(20223BCG74001);国家自然科学基金项目(81460329);江西省自然科学基金项目(20192ACBL20039)。

摘　　要：目的:本研究拟联合多序列MRI的放射组学评分和临床、病理学等危险因素,构建预测垂体腺瘤术后复发的模型。方法:回顾性分析2012年6月-2017年6月128例确诊为垂体腺瘤(复发58例,未复发70例,随访5~10年)的临床病理和术前MRI资料。对术前图像进行分割并提取特征。经过降维后,最终选取6个与复发相关的组学特征构建linear-SVM、rbf-SVM、KNN、LR、RF和XGBoost机器学习模型及R-score。随后将R-score与病理、影像学变量相结合建立联合Nomogram模型。评价和比较模型的预测性能并通过校准曲线和决策曲线分析其临床价值。结果:在临床病理学及影像学特征中,Ki-67和肿瘤最大直径是垂体腺瘤复发的独立预测因子,linear-SVM是性能最好机器学习模型。相较于单一模型而言,联合Nomogram模型则在训练集(AUC=0.907,95%CI:0.843~0.972)和测试集(AUC=0.883,95%CI:0.769~0.996)中表现出了更好的预测性能。决策曲线也显示联合Nomogram显示出更好的预测性能和临床应用价值。结论:联合Nomogram模型在预测垂体腺瘤复发具有良好性能,有助于临床个性化术前治疗决策和提前规划术后辅助治疗。Objective:This study intends to combine the radiomic score of multi-sequence MRI with clinical and pathological risk factors to build a model for predicting postoperative recurrence of pituitary adenoma.Methods:From June 2012 to June 2017,the clinicopathological and preoperative MRI data of 128 patients diagnosed with pituitary adenoma(58 cases with recurrence and 70 cases without recurrence,followed up for 5~10 years)were retrospectively analyzed.Preoperative images were segmented and features were extracted.After dimensionality reduction,six relapse-related omics features were selected to construct linear SVM,rbf-SVM,KNN,LR,RF and XGBoost machine learning models and R-score.Then a nomogram model was established by combining R-score with pathological and imaging variables.The predictive performance of the models was evaluated and compared,and their clinical value was analyzed by calibration and decision curves.Results:Ki-67 and maximum tumor diameter were independent predictors of pituitary adenoma recurrence in clinicopathological and imaging features.Linear SVM was the best machine learning model.Compared with a single model,the combined nomogram model showed better predictive performance in both the training set(AUC=0.907,95%CI:0.843~0.972)and the test set(AUC=0.883,95%CI:0.769~0.996).The decision curve also showed that the combined nomogram showed better predictive performance and clinical application value.Conclusion:The combined nomogram model has good performance in predicting the recurrence of pituitary adenoma,which is helpful for personalized preoperative treatment decision and postoperative adjuvant treatment planning in advance.

关 键 词：放射组学 机器学习 垂体腺瘤 磁共振成像 诺谟图 

分 类 号：R445.2[医药卫生—影像医学与核医学] R736.4[医药卫生—诊断学]