2101 - Subregional Radiomics and Dosiomics Analysis for Predicting Radiation Pneumonitis in Lung Cancer: A Multi-Omics Nomogram Study

Purpose/Objective(s): Radiation pneumonitis (RP) is a common complication of thoracic radiotherapy. This study aimed to develop a nomogram based on subregional radiomics and dosiomics for predicting RP in lung cancer patients and to guide individualized intensity-modulated radiation therapy.

Materials/Methods: We retrospectively enrolled 110 lung cancer patients from our hospital to form the primary and validation cohorts. The synthetic minority over-sampling technique (SMOTE) was implemented to address class imbalance. To identify the intratumoral heterogeneity of the tumor, the lesions identified on pretreatment CT scans were subdivided into phenotypically consistent subregions by automatic clustering on the patient-level and population-level. Radiomics features were extracted separately from the entire tumor region and subregions, while dosiomics features and dose-volume histogram (DVH) metrics were derived from radiotherapy plans. Key predictors were selected through intraclass correlation coefficient (ICC) analysis and least absolute shrinkage and selection operator (LASSO) regression. A radiomics signature (RS), dosiomics risk score (D-score), and clinical parameters were integrated into a nomogram using multivariable logistic regression. Area under the receiver operating characteristic (ROC) curve (AUC) was calculated to assess the models.

Results: Seven radiomics features and four dosiomics features (belonging to the textural and shape-based feature category) exhibited significant correlations (P<0.05) with RP were identified to develop the RS and D-score. The nomogram incorporating the RS, D-score, and clinical parameters achieved the best discriminative ability with AUCs ranging from 0.70 to 0.85 in SMOTE-augmented datasets, and good calibration. Decision curve analysis suggested that the combined nomogram was clinically useful.

Conclusion: Subregional radiomics analysis has potential value for predicting RP. The proposed CT-based nomogram, integrating radiomics, dosiomics, and clinical parameters, provides a clinically actionable tool for personalized risk stratification in lung cancer patients receiving radiotherapy.