基于CT影像组学特征的机器学习模型对重型颅脑创伤患者颅内压相关参数的预测效能  

The predictive value of CT radiomics features for intracranial pressure-related parameters in patients with severe traumatic brain injury

作　　者：朱俊[1] 刘珈杞 单颖驰 高国一 Zhu Jun;Liu Jiaqi;Shan Yingchi;Gao Guoyi(Department of Neurosurgery,Shanghai General Hospital,Shanghai Jiao Tong University School of Medicine,Shanghai 201600,China;Neurosurgery Center,Beijing Tiantan Hospital,Capital Medical University,Beijing 100070,China)

机构地区：[1]上海交通大学医学院附属第一人民医院神经外科,上海201600 [2]首都医科大学附属北京天坛医院神经外科学中心,北京100070

出　　处：《中华神经外科杂志》2024年第5期436-440,共5页Chinese Journal of Neurosurgery

基　　金：国家自然科学基金(81971699)。

摘　　要：目的探讨基于CT影像组学特征的机器学习模型对重型颅脑创伤患者压力波幅相关性指数(RAP)和压力反应性指数(PRx)的预测价值。方法回顾性分析2019年1月至2020年12月上海交通大学医学院附属第一人民医院神经外科收治的36例重型颅脑创伤患者的临床和影像学资料。纳入患者的入院格拉斯哥昏迷评分(GCS)[M(范围)]为6(3~8)分。所有患者均行手术治疗、持续颅内压监测及有创动脉压监测,并采集术后1 h内的RAP和PRx。于患者术后1 h的颅脑CT图像中选择1个感兴趣体积(VOI)区域并提取93个影像组学特征。运用递归特征消除法分别筛选出预测PRx和RAP的最优特征子集,然后使用随机森林算法训练分类器预测PRx及RAP,构建基于CT影像组学特征的预测模型。分别评估模型的准确率、精确率、召回率、F1评分和受试者工作特征(ROC)曲线的曲线下面积(AUC),以判断CT影像组学特征的预测性能。结果预测PRx和RAP的最优特征数量分别为12个和15个。通过CT影像组学特征预测PRx的准确率为72%,精确率为85%,召回率为68%,F1评分为0.61,AUC为0.79。通过CT影像组学特征预测RAP的准确率为63%,精确率为78%,召回率为63%,F1评分为0.61,AUC为0.80。结论基于CT影像组学特征建立的模型能够有效预测重型颅脑创伤患者的PRx和RAP,有助于指导治疗方案及评估患者的预后。Objective To explore the application value of a machine learning model based on CT radiomics features in predicting the pressure amplitude correlation index(RAP)and pressure-reactivity index(PRx)in patients with severe traumatic brain injury(TBI).Methods The clinical and imaging data of 36 patients with severe TBI admitted to the Department of Neurosurgery of Shanghai General Hospital,Shanghai Jiao Tong University School of Medicine from January 2019 to December 2020 were retrospectively analyzed.The admission Glasgow Coma Scale(GCS)score[M(range)]of the included patients was 6(3 to 8)points.All patients underwent surgical treatment,continuous intracranial pressure monitoring,and invasive arterial pressure monitoring.The RAP and PRx were collected within 1 h after surgery.Then one volume of interest(VOI)was selected from the craniocerebral CT images of patients within 1 h after surgery,and a total of 93 radiomics features were extracted from the VOI for predicting RAP and PRx.The recursive feature elimination method was used for feature selection to obtain the optimal feature subset.The random forest algorithm was used to train the classifier to predict PRx and RAP respectively,and a prediction model was constructed based on CT radiomics features.The accuracy,precision,recall rate,F1 score,and receiver operating characteristic(ROC)curve and area under curve(AUC)of models were used to evaluate the predictive performance of CT radiomics features.Results The optimal number of features for predicting PRx and RAP was 12 and 15,respectively.The accuracy of predicting PRx by CT radiomics features was 72%,the precision was 85%,the recall rate was 68%,the F1 score was 0.61,and the AUC was 0.79.The accuracy of predicting RAP by CT radiomics features was 63%,the precision was 78%,the recall rate was 63%,the F1 score was 0.61,and the AUC was 0.80.Conclusion The prediction model based on CT radiomics features can effectively predict PRx and RAP in patients with severe TBI,which could help guide treatment and assess the patien

关 键 词：脑损伤 创伤性 颅内压 机器学习 电子计算机断层扫描 压力波幅相关性指数 压力反应性指数 

分 类 号：R651.15[医药卫生—外科学]