A machine learning approach for radiation pneumonitis prediction in elderly esophageal cancer patients by integrating baseline computed tomography radiomics, dosiomics, and clinical characteristics.

Background and purpose: To integrate clinical characteristics, radiomics, and dosiomics to provide accurate and individualized prediction of radiation pneumonitis (RP) in elderly patients aged 70 and over with esophageal cancer receiving radiotherapy. Materials and methods: Based on a phase III clinical study (NCT02979691) that included elderly patients with esophageal squamous cell carcinoma (ESCC) who received definitive radiotherapy, we selected a total of 229 patients with available computed tomography (CT) and dose images. Radiomic and dosiomics features were extracted from both lungs. The patients were randomly assigned to either the training group (N = 161) or the test group (N = 68) in a 7:3 ratio. In the training set, logistic regression (LR) was applied to calculate the radiomic score (R score) and dosiomic score (D score). The constructed multivariate LR and ridge regression prediction models were evaluated using the test set. The endpoint of the predictive model is defined as a grade ≥ 2 RP. Discrimination and prediction were assessed by calculating the area under curve (AUC) of the receiver operating characteristic curve and plotting calibration and decision curve analyses (DCA). Results: The hybrid LR model integrating R score, D score and clinical characteristics had the best clinical applicability. The hybrid model demonstrated superior predictive performance on the test set, achieving an area under the curve (AUC) of 0.76, while the combined clinical and DVH model achieved an AUC of 0.70. Conclusions: A hybrid model combining radiomics and dosiomics with clinical characteristics showed the best performance for predicting RP.