Tumor habitat-derived radiomics features in pretreatment CT scans for predicting concurrent chemoradiotherapy responses in nasopharyngeal carcinoma: a retrospective study.

Background: Nasopharyngeal carcinoma (NPC) is a highly heterogeneous malignancy, characterized by significant variability in its biological and clinical features, which contribute to diverse treatment responses among patients. This study aimed to investigate intratumoral heterogeneity (ITH) in pretreatment computed tomography (CT) scans and test its performance for predicting responses to simultaneous chemoradiotherapy treatment in NPC patients. Methods: Pretreatment CT scans of 113 NPC patients were retrospectively analyzed at our center from March 2012 to September 2022. Radiomics features were selected from tumor and habitat regions to establish models. Both univariate and multivariate analyses were conducted to identify clinical risk indices related to treatment responses. Significant variables, including clinical variables, radiomics features, and habitat radiomics (H-Rad) features, were integrated into a joint predictive model, with its performance assessed using the area under the receiver operating characteristic (ROC) curve (AUC). Results: A total of ten prediction models were constructed, including six radiomics models [support vector machine (SVM), random forest, extra trees, extreme gradient boost (XGBoost), light gradient boosting machine (LightGBM), and habitat model] and one joint predictive model. The ExtraTrees model performed exceptionally well, resulting in AUCs of 0.969 and 0.894 in the training and testing cohorts, respectively. This indicates its strong ability to effectively predict between treatment responses. In the training cohort, the joint model demonstrated superior predictive accuracy with the highest AUC of 0.961. Additionally, the HabitatMean model showed excellent performance, with an AUC of 0.944. Overall, the joint model demonstrated robustness and superior integration of various features for predictive analysis, with the highest AUCs of 0.961 and 0.861 in the training and testing cohorts, respectively. Conclusions: A model that integrates conventional radiomics (C-Rad), a quantitative CT-based measure of ITH, and clinical variables has shown significant accuracy in predicting treatment response to chemoradiotherapy in NPC patients.