Predictive value of CT radiomics model for radioresistance in patients with esophageal squamous cell carcinoma
10.3760/cma.j.cn113030-20240131-00039
- VernacularTitle:CT影像组学模型对食管鳞癌患者放射抵抗的预测价值
- Author:
Mengyu HAN
1
;
Yu ZHANG
;
Linrui LI
;
Liting QIAN
Author Information
1. 中国科学技术大学附属第一医院放射治疗科，合肥 230001
- Publication Type:Journal Article
- Keywords:
Esophageal squamous cell carcinoma;
Radioresistance;
Radiomics;
Machine learning
- From:
Chinese Journal of Radiation Oncology
2025;34(2):136-143
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the predictive value of machine learning-based CT radiomics model for radioresistance in patients with esophageal squamous cell carcinoma (ESCC).Methods:Clinical data of 185 patients with ESCC treated with radical radiotherapy in the First Affiliated Hospital of Anhui Medical University from December 2015 to July 2022 were retrospectively analyzed, and all patients were randomly divided into a training set ( n=129) and a validation set ( n=56) at a ratio of 7 : 3. The radiomics parameters of the primary lesion of esophageal cancer and the surrounding 5 cm region in the patients' CT arterial phase images were extracted, and 6 machine learning methods were used to screen the optimal radiomics model to obtain the optimal radiomics score (Radscore). Independent prognostic predictors of radioresistance in ESCC were obtained by univariate and multivariate logistic regression analyses, which was used as the basis for constructing the nomogram. The predictive performance of different models was compared by the area under the receiver operating characteristic (ROC) curve (AUC). The predictive efficacy and clinical value of the combined model were evaluated using calibration curve, decision curve analysis and clinical impact curve, net reclassification improvement (NRI) and integrated discrimination improvement (IDI). Results:The combined intratumoral and peritumoral radiomics model based on naive Bayesian classifier yielded the optimal prediction performance, with AUC of 0.859 and 0.936 in the training set and validation set, respectively. Multivariate logistic regression analysis showed that Radscore and T stage were the independent prognostic predictors of radioresistance in ESCC patients, and the AUC of the combined model constructed based on these predictors in the training and validation sets were 0.942 and 0.959, respectively. Calibration curve, decision curve analysis and clinical impact curve, net reclassification improvement (NRI) and integrated discrimination improvement (IDI) all indicated higher clinical benefit and more consistent predictive efficacy of the combined model.Conclusions:Machine learning-based CT radiomics model is useful for the prediction of radioresistance in ESCC. The nomogram of radiomics and clinical parameters can further improve the prediction accuracy and provide novel reference for individualized treatment of patients with ESCC.
