文瑶,刘丹,喻媛,李鑫娅,苏丽平,王芳.基于临床联合CT影像组学特征列线图预测急性胰腺炎预后[J].中国医学影像技术,2022,38(11):1675~1679
基于临床联合CT影像组学特征列线图预测急性胰腺炎预后
Nomogram based on clinical and CT radiomics features for predicting prognosis of acute pancreatitis
投稿时间:2022-07-15  修订日期:2022-09-28
DOI:10.13929/j.issn.1003-3289.2022.11.018
中文关键词:  胰腺炎  预后  影像组学  列线图
英文关键词:pancreatitis  prognosis  radiomics  nomogram
基金项目:
作者单位E-mail
文瑶 重庆医科大学附属永川医院放射科, 重庆 402160  
刘丹 重庆医科大学附属永川医院放射科, 重庆 402160 5677676@qq.com 
喻媛 重庆医科大学附属永川医院放射科, 重庆 402160  
李鑫娅 重庆医科大学附属永川医院放射科, 重庆 402160  
苏丽平 重庆医科大学附属永川医院放射科, 重庆 402160  
王芳 上海联影智能有限公司, 上海 200232  
摘要点击次数: 1265
全文下载次数: 410
中文摘要:
      目的 观察基于临床联合CT影像组学特征构建的联合模型列线图预测急性胰腺炎(AP)预后的价值。方法 回顾性分析203例临床首诊AP患者的临床及上腹部CT资料,按7 : 3比例将其分为训练集(n=142)和验证集(n=61),基于增强静脉期CT提取并筛选最优影像组学特征,计算影像组学评分Radscore;根据预后分为预后良好组(n=114)与预后不良组(n=89)。将临床变量(包括临床及CT表现)及Radscore纳入单因素和多因素逻辑回归分析,筛选影响AP预后的独立危险因素;构建临床、影像组学和联合模型,以受试者工作特征(ROC)曲线评价其预测AP预后的效能;以联合模型预测指标构建列线图,结合校准曲线评估其校准度。结果 共筛选出7个最优影像组学特征用于计算Radscore,其中C反应蛋白、糖尿病史和Radscore为影响AP预后的独立危险因素。联合模型预测训练集及验证集AP预后的AUC (0.84、0.82)均高于临床模型(0.71、0.66,Z=3.12、2.71,P均<0.05);其预测训练集的AUC (0.84)高于影像组学模型(0.76,Z=2.39,P=0.02),预测验证集的AUC (0.82)与影像组学模型(0.81)差异无统计学意义(Z=0.08,P>0.05)。校正曲线显示联合模型列线图的校准度良好。结论 基于临床联合CT影像组学特征列线图可有效预测AP预后。
英文摘要:
      Objective To observe the value of nomogram based on clinical and CT radiomics for predicting prognosis of acute pancreatitis (AP). Methods Clinical and upper-abdominal CT data of 203 first-diagnosed AP patients were retrospectively analyzed. The patients were divided into training set (n=142) and validation set (n=61) at the ratio of 7:3. The optimal radiomics features based on enhanced venous phase CT were extracted and screened, the Radscores were calculated. The patients were divided into good prognosis group (n=114) and poor prognosis group (n=89). Clinical relevant variables, i.e. clinical data and CT findings and Radscores were included for univariate and multivariate logistic regression analysis to screen the independent risk factors of the prognosis. Then clinical, radiomics and combined models were constructed, respectively. Receiver operating characteristic (ROC) curve was used to evaluate the efficacy of each model for predicting the prognosis of AP. The combined model prediction indexes were used to construct a nomogram, and the calibration curve was used to evaluate its calibration degree. Results Totally 7 optimal radiomics features were screened to calculate Radscores, among which C reactive protein, history of diabetes and Radscores were independent risk factors for the prognosis of AP. The area under the curve (AUC) of the combined model for predicting AP prognosis in training set and validation set (0.84, 0.82) were both higher than those of clinical model (0.71, 0.66, Z=3.12, 2.71, both P<0.05). AUC of the combined model in training set (0.84) was higher than that of radiomics model (0.76, Z=2.39, P=0.02), while AUC of the combined model in validation (0.82) was not significant different with that of radiomics model (0.81, Z=0.08, P>0.05). The calibration curve showed that the combined model nomogram model was well calibrated. Conclusion Nomogram based on clinical and CT radiomics features could effectively predict the prognosis of AP.
查看全文  查看/发表评论  下载PDF阅读器