ABSTRACT
Objective:To construct a predictive nomogram incorporating pretreatment CT-based radiomics for radiation pneumonitis (RP) in esophageal cancer (EC) patients treated with intensity-modulated radiotherapy (IMRT), and to evaluate the value of CT radiomics in predicting RP.Methods:Clinical data of 267 EC patients sequentially treated with IMRT in Quanzhou First Hospital affiliated to Fujian Medical University from January 2019 to December 2021 were prospectively analyzed. Among them, the first 206 patients were assigned into the training cohort and the last 61 patients were enrolled in the validation cohort. Radiomics features of bilateral lungs were extracted by radiotherapy CT simulation. Univariate analysis was performed to screen the potential predictive variables for symptomatic RP. Machine learning algorithms, such as least absolute shrinkage and selection operator (LASSO), extreme gradient boosting (XGboost), and support vector machine (SVM), were performed for radiomic features selection, respectively. The best classifier was chosen to construct a radiomic signature (RS). Clinical, radiomics and combined nomogram predictive model were developed, respectively. The predictive efficiency and clinical benefits of three models were compared by calculating the area under the receiver operating characteristic (ROC) curve (AUC), calibration curve and decision curve analysis (DCA), and then validated in the validation cohort. Multivariate logistic regression analysis was conducted. Different ROC curves were compared by Delong test.Results:Cardiovascular disease, minimum internal diameter of esophagus and adjuvant chemotherapy and RS were the independent related factors of RP. The AUC of clinical, radiomics and combined models were 0.772, 0.745, 0.842 in the training cohort, and 0.851, 0.811, 0.901 in the validation cohort, respectively. DCA showed that combined radiomic model yielded better clinical benefits compared with clinical model.Conclusion:Radiomics features from pretreatment CT have the potential of improving the efficiency of RP prediction models for EC patients treated with IMRT.