Establishment and validation of a risk prediction model combined CT-radiomics and clinical features for lymph node metastasis in hilar cholangiocarcinoma
10.3760/cma.j.cn112149-20230829-00137
- VernacularTitle:基于CT影像组学和临床特征的肝门部胆管癌淋巴结转移风险预测模型的建立和验证
- Author:
Pengchao ZHAN
1
;
Keyan LIU
;
Xing LIU
;
Hanyu JIANG
;
Peijie LYU
;
Jianbo GAO
Author Information
1. 郑州大学第一附属医院放射科，郑州 450052
- Keywords:
Cholangiocarcinoma;
Tomography, X-ray computed;
Lymph node metastasis;
Radiomics;
Nomogram
- From:
Chinese Journal of Radiology
2024;58(4):409-415
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To establish and validate a clinical and CT radiomics combined model for predicting lymph node metastasis (LNM) risk in patients with hilar cholangiocarcinoma (HCCA).Methods:This was a case-control study. Data from 158 pathologically confirmed HCCA patients between January 2016 and January 2022 at the First Affiliated Hospital of Zhengzhou University were retrospectively analyzed. Using stratified random sampling, the patients were randomly divided into a training set ( n=95) and an internal validation set ( n=63) at a 6∶4 ratio. According to postoperative pathology, 31 LNM-positive cases and 64 LNM-negative cases were in the training set, and 22 LNM-positive cases and 41 LNM-negative cases were in the internal validation set. A cohort of 50 HCCA patients was retrospectively collected from West China Hospital of Sichuan University between October 2018 and June 2021 as an external validation set, including 21 LNM-positive and 29 LNM-negative cases. Clinical features were selected by multivariate logistic regression analysis to establish a clinical model. Radiomics features were extracted from portal venous phase CT images using 3D Slicer software. A radiomics model was developed using the least absolute shrinkage and selection operator regression algorithm. A clinical-radiomics model was constructed by integrating clinical features and Radscore, and a nomogram was developed. The prediction performance of models was evaluated by the area under the receiver operating characteristic curve (AUC). The AUC values were compared using the DeLong test. Calibration curves and decision curves were plotted to assess calibration and clinical net benefit. Results:Clinical N (cN) staging was an independent risk factor for LNM ( OR=6.86, 95% CI 2.70-18.49, P<0.001). Totally 12 optimal features were selected to construct the radiomics model, and the clinical-radiomics nomogram model was constructed by combining cN staging and Radscore. In the external validation set, the AUC (95% CI) of the clinical model, radiomics model, and clinical-radiomics nomogram were 0.706 (0.576-0.836), 0.768 (0.637-0.899), and 0.803 (0.680-0.926), respectively. The nomogram achieved higher AUC than clinical and radiomics models with statistical significance ( Z=2.01, 2.21; P=0.044, 0.027). The calibration and decision curves demonstrated good model fit, providing clinical net benefits for patients. Conclusion:The clinical-radiomics nomogram model combining cN staging and CT radiomics features can effectively predict LNM risk in HCCA patients.
