CT-based radiomics combined with deep learning for predicting radiation pneumonitis in patients with esophageal...

This study aims to develop a multimodal prediction model combining radiomics and deep learning techniques to assess the risk of radiation pneumonia (RP) in esophageal cancer patients undergoing radiotherapy.

This retrospective study enrolled esophageal squamous cell carcinoma (ESCC) patients who received conventional fractionated radiotherapy at two hospitals in China between September 2018 and September 2023. The patients were divided into a training cohort (Hospital I, 116 cases) and a validation cohort (Hospital II, 41 cases). The intraclass correlation coefficient (ICC), Least Absolute Shrinkage and Selection Operator (LASSO), and Boruta were used as feature selection methods, while a support vector machine (SVM) was employed for model construction. Six models were constructed based on a combination of clinical features, radiomic features, and deep learning features. Model performance was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC), calibration curve, and decision curve analysis (DCA). Four different machine learning algorithms were used to assess the best model, with the optimal classifier selected.

The hybrid model combining clinical features, radiomics, and deep learning achieved an AUC of 0.902 in the training cohort and 0.857 in the validation cohort, significantly outperforming the single-modality models. Among various machine learning algorithms, the random forest method demonstrated the best performance in external validation with an AUC of 0.859. Based on DCA and calibration curve analysis, the model showed good net clinical benefit and fit.

The multi-modal predictive model, which integrates radiomics and deep learning techniques, effectively predicts the risk of radiation-induced pneumonitis in esophageal cancer patients after radiotherapy. This approach provides a novel pathway for the early identification and prevention of RP.