Prediction of pancreatic fistula after pancreaticoduodenectomy by multi-phase enhanced CT radiomics model
10.3969/j.issn.1002-1671.2025.04.014
- VernacularTitle:多期增强CT影像组学模型预测胰十二指肠切除术后胰瘘的研究
- Author:
Tianxin CHENG
1
;
Hongwei WU
;
Zhixiang WANG
;
Piao YAN
;
Xiaoyang LI
;
Zhenhao LIU
;
Kuinan TONG
;
Kun LIU
;
Hui XU
;
Zhenghan YANG
Author Information
1. 首都医科大学附属北京友谊医院放射科,北京 100050
- Publication Type:Journal Article
- Keywords:
pancreaticoduodenectomy;
pancreatic fistula;
multi-phase enhanced computed tomography;
radiomics
- From:
Journal of Practical Radiology
2025;41(4):603-607
- CountryChina
- Language:Chinese
-
Abstract:
Objective To compare the ability of single-phase,dual-phase,and triphasic models in forecasting postoperative pancreatic fistula(POPF)after pancreaticoduodenectomy(PD)using radiomics based on triphasic enhanced CT.Methods A total of 181 patients who underwent multi-phase enhanced CT prior to PD were retrospectively selected,and the collection phase included non-contrast,arterial phase(AP),and equilibrium phase(EP).3D Slicer software was utilized to segment the region of interest(ROI)for the postoperative pancreatic remnant on each phase.Radiomics feature extraction was performed using R software,followed by feature selection through least absolute shrinkage and selection operator(LASSO)regression with five-fold cross-validation to prevent model overfitting.The effective features selected were combined in a weighted linear manner to obtain a Radiomics score(Radscore).The patients were divided into training set and test set in a 7︰3 ratio.Logistic regression was employed to construct seven POPF prediction models(three single-phase,three dual-phase,and one triphasic models)based on different phase combinations.The diagnostic performance of the models was evaluated using the area under the curve(AUC)of receiver operating characteristic(ROC)curve,accuracy(ACC),sensitivity(SEN),and specificity(SPE).The DeLong test was applied to compare the differences in AUC among different models.Results After LASSO regression,24 effective features associated with POPF were selected from different phases.In the test set,the triphasic model exhibited the highest AUC and ACC(AUC=0.76,ACC=0.808).The calibration curve demonstrated the strongest agreement between the estimated probabilities and observed probabilities for the triphasic model.The decision curve analysis(DCA)curve indicated that the triphasic model had the largest threshold range with a higher net benefit.Conclusion Compared with single-phase and dual-phase models,the triphasic model based on enhanced CT provides better prediction of POPF after PD,aiding clinical decision-making and improve prognosis.
