Construction and evaluation of a radiomics model for predicting perineural invasion in intrahepatic cholangiocarcinoma
10.3760/cma.j.cn113884-20250512-00153
- VernacularTitle:肝内胆管癌神经侵犯的影像组学预测模型的构建与评估
- Author:
Kai ZHANG
1
;
Gengping ZHOU
;
Yang XU
;
Chenxi XIE
;
Pengyu CHEN
;
Yangyang WANG
;
Taiyang CHEN
;
Qingshan LI
;
Bo MENG
;
Haibo YU
Author Information
1. 郑州大学人民医院肝胆胰腺外科，郑州 450003
- Publication Type:Journal Article
- Keywords:
Cholangiocarcinoma;
Perineural invasion;
Radiomics;
Prediction model
- From:
Chinese Journal of Hepatobiliary Surgery
2025;31(11):817-822
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To construct and evaluate a radiomics model for predicting perineural invasion in patients with intrahepatic cholangiocarcinoma (ICC).Methods:Clinical data of 144 patients with ICC undergoing surgery in the People’s Hospital of Zhengzhou University ( n=113) and the Affiliated Cancer Hospital of Zhengzhou University ( n=31) from January 2018 to June 2023 were retrospectively analyzed, including 80 males and 64 females, aged (58.8±10.1) years. The patients were randomly divided into a training set ( n=100) and a test set ( n=44) at a ratio of 7: 3. The former set was used to build the model for predicting perineural invasion, and the latter was used to evaluate the model. Enhanced CT images and clinical data of the patients were collected, and features related to perineural invasion were screened. A light gradient boosting machine was used to construct an imaging genomics model. The model was evaluated using the receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA). Results:Univariate and multivariate logistic regression analysis showed that none of the clinical features were associated with neural invasion in ICC patients (all P>0.05). Six, 25, 32, and 37 radiomics features were obtained by screening the intratumoral, 2 mm peritumoral, 5 mm peritumoral, and 8 mm peritumoral regions, respectively. The area under the ROC curve for predicting perineural invasion in ICC patients was 0.849 (95% CI: 0.774-0.923) in the training set and 0.745 (95% CI: 0.597-0.894) in the test set for the intratumoral model, 0.966 (95% CI: 0.938-0.995) and 0.750 (95% CI: 0.604-0.896) for the 5mm peritumoral model, 0.936 (95% CI: 0.892-0.980) and 0.792 (95% CI: 0.644-0.939) for the 2mm peritumoral model, and 0.961 (95% CI: 0.929-0.992) and 0.689 (95% CI: 0.526-0.853) for the 8mm peritumoral model. The area under the ROC curve, accuracy, sensitivity, and specificity of the combined intratumoral and 5mm peritumoral model for predicting perineural invasion were 0.927 (95% CI: 0.878-0.976), 88.0%, 84.5%, and 89.8% in the training set, and 0.849 (95% CI: 0.737-0.960), 77.3%, 85.2%, and 72.0% in the test set, respectively. The calibration curve showed a deviation between the calibration curve of the combined intratumoral and 5mm peritumoral model and the ideal line, but it could achieve basic consistency. DCA showed that when the threshold was between 0.18 and 0.70, the combined intratumoral and 5mm peritumoral model could bring clinical net benefit to patients when predicting neural invasion. Conclusion:The intratumoral and 5mm peritumoral imaging genomics model based on enhanced CT features can effectively predict neural invasion and offer clinical benefits in patients with ICC.
