基于CT平扫影像组学模型对早期急性坏死性胰腺炎的鉴别诊断价值

Early detection of acute necrotizing pancreatitis based on noncontrast CT radiomics model

目的:基于急性胰腺炎(AP)首次发作时CT平扫图像,比较间质水肿性AP与出血坏死性AP的影像组学特征差异,探讨基于CT平扫影像组学模型对早期坏死性胰腺炎的鉴别诊断价值。方法:搜集2017年1月-2020年6月被诊断为AP的300例患者,回顾性分析AP患者腹痛症状发作24h内首次CT平扫图像和临床一般资料。以72~96h内增强CT及改良CT严重指数(MCTSI)评分为评判标准,将300例AP患者分为间质水肿性AP(230例)和出血坏死性AP(70例)。在联影科研平台下,手动逐层分割整个胰腺形态轮廓,得到感兴趣区体积(VOI),提取特征参数,按照7:3的比例将影像资料随机分为训练组与验证组,通过随机森林(RF)模型进行训练,采用五折交叉法对其进行验证,采用ROC曲线分析模型对出血坏死性AP的诊断效能,并计算准确度、敏感度和特异度。结果:性别、年龄在间质水肿性AP与出血坏死性AP患者间差异无统计学意义。通过构建RF组学模型,训练组及验证组中区分间质水肿性AP与出血坏死性AP的ROC曲线下面积分别为0.979和0.936。结论:基于CT平扫影像组学特征结合机器学习随机森林模型建立的鉴别诊断模型,能够在早期准确判别急性坏死性胰腺炎。

Objective:To compare the difference in the radiomics characteristics of acute interstitial edematous pancreatitis and hemorrhagic necrotizing pancreatitis based on noncontrast CT images at onset,and to explore the value of radiomics model for the early detection of necrotizing pancreatitis.Methods:A total of 300 patients with acute pancreatitis(AP) were reviewed from January 2017 to June 2020.The first noncontrast CT images and clinical data within 24 hours of onset of abdominal pain were retrospectively analyzed.They were divided into acute interstitial edematous(n=230) and hemorrhagic necrotizing pancreatitis(n=70) according to enhanced CT and modified CT severity Index(MCTSI) scores within 72 to 96 hours.On the United Imaging scientific research platform,the pancreatic contour was manually segmented layer by layer to obtain the volume of region(VOI),and the characteristic parameters were extracted.The image data were randomly divided into training and validation groups in a ratio of 7:3.Then through random forest(RF) model for training,5-fold cross validation method was adopted to carry out verification.ROC curve was used to analyze the diagnostic performance for detection of hemorrhagic necrotizing pancreatitis.Results:There was no significant difference in gender and age between interstitial edematous AP and hemorrhagic necrotizing AP group.By constructing the RF radiomics model,the AUC values of distinguishing between interstitial edematous AP and hemorrhagic necrotizing AP in the training group and the validation group were 0.979 and 0.936,respectively.Conclusion:The model based on noncontrast CT radiomics features combined with RF model of machine learning can accurately identify acute necrotizing pancreatitis in the early stage.