Development and validation of early diagnostic model using radiomics for distinguishing benign and malignant pancreatic tumors
10.3760/cma.j.cn113884-20241126-00347
- VernacularTitle:基于放射组学的胰腺良恶性肿瘤诊断模型的构建与验证
- Author:
Chengxu DU
1
;
Yilin LI
;
Bin ZHANG
;
Wenfeng FENG
;
Ang LI
;
Fengshan LI
;
Haitao LYU
;
Weihong ZHAO
;
Dongrui LI
Author Information
1. 河北医科大学第二医院肝胆胰腺外科，石家庄 050000
- Publication Type:Journal Article
- Keywords:
Pancreatic neoplasms;
Benign tumor;
Diagnosis;
Radiomics
- From:
Chinese Journal of Hepatobiliary Surgery
2025;31(8):597-602
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To develop and validate a diagnostic model for pancreatic benign and malignant tumors using radiomics technology.Methods:A retrospective analysis was conducted on the clinical data of 113 patients with pancreatic tumors who underwent surgical treatment at the Hepatobiliary and Pancreatic Surgery Departments of the Second Hospital and the First Hospital of Hebei Medical University from January 2020 to December 2022. There were 59 male and 54 female patients, aged (55.3±16.8) years. Preoperative enhanced thin-slice computed tomography (CT) data and postoperative pathological diagnosis results were collected. Data from 74 patients at the Second Hospital were selected, and according to the random classification principle of 7∶3, the data of 52 patients were determined as the training set for model construction, while the remaining 22 patients' data served as the internal validation set. Data from 39 patients at the First Hospital of Hebei Medical University were used as the external validation set to assess the generali-zability of the established model. The region of interest in the lesions on CT images was analyzed using three-dimensional radiomics feature extraction, and the top 5 features were selected using feature selection methods. Radiomics models were established for the selected features using 17 classifiers. The performance of the models was evaluated using the area under curve (AUC) of the receiver operating characteristic (ROC).Results:Two hundred and fifty-five models were established with 15 feature selection methods and 17 classifiers. 7 models with the AUC greater than 0.7 were selected, among which the best one was LASSO-K neighbors classifier model, constructed using the LASSO feature selection method and the k-nearest neighbors algorithm, achieving AUC values of 0.933 (95% CI: 0.859-0.984) in the training set, 0.973 (95% CI: 0.896-1.000) in the validation set, and 0.774 (95% CI: 0.624-0.908) in the external validation set, with satisfactoryclassification and generalization ability. Conclusion:The radiomics-based diagnostic model for pancreatic benign and malignant tumors can effectively distinguish the benignancy and malignancy of tumors. The LASSO-K neighbors classifier model demonstrated high accuracy and reliability in this study.
