CT radiomics based machine-learning model predicts portal vein-superior mesenteric vein involvement in pancreatic ductal adenocarcinoma
10.3760/cma.j.cn113884-20220205-00052
- VernacularTitle:基于CT影像组学的机器学习模型预测胰腺癌门静脉-肠系膜上静脉侵犯的研究
- Author:
Fangming CHEN
1
;
Shuanglin ZHANG
;
Yue CHENG
;
Xiumin QI
;
Yongping ZHOU
;
Lei ZHANG
;
Zhuiyang ZHANG
Author Information
1. 南京医科大学附属无锡第二医院影像科，无锡 214002
- Keywords:
Pancreatic neoplasms;
Machine learning;
Vascular invasion;
Radiomics
- From:
Chinese Journal of Hepatobiliary Surgery
2022;28(7):525-530
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the value of machine learning-based computed tomography (CT) images radiomics analysis in preoperative evaluation of surgical portal vein-superior mesenteric vein (PV-SMV) invasion in patients with pancreatic ductal adenocarcinoma (PDAC).Methods:The retrospective study was conducted with 156 consecutive PDAC patients who were underwent surgery at the Affiliated Wuxi No.2 People's Hospital of Nanjing Medical University between January 2010 and July 2021. There were 95 males and 61 females, with the age of (65.7±8.2) years. Patients were randomly split into training set and validation set by a ratio of 3∶2. Minimum redundancy maximum relevance was used to select radiomic features, which were extracted from contrast-enhanced CT images. Five machine learning classifiers were developed, and those models' area under the curve (AUC) values were compared with the conventional radiologic-feature-based evaluation.Results:Ninety-four and 52 patients were included into the training set and validation set, respectively. Their PV-SMV invasion rates were confirmed by intraoperative exploration with 31.9%(30/94) and 40.3%(25/61), respectively. Five models: LASSO regression, random forest, support vector machine, k-nearest neighbor and Naive Bayesian, were established based on ten features from CT images radiomics, and LASSO regression model achieved the highest AUC value compared with the other four models (all P<0.05). Compared with the conventional radiologic evaluation, the LASSO regression model had higher AUC (0.920 vs. 0.752) and sensitivity (92.0% vs. 86.5%)(both P<0.05). Conclusion:Machine learning-based CT images radiomics analysis can be used to evaluate PV-SMV invasion status preoperatively in PDAC. The LASSO regression model showed better performance than the conventional radiologic evaluation.
