Application of delta radiomics based on cone-beam computed tomography in predicting radiotherapy efficacy for nasopharyngeal carcinoma.
Authors
Affiliations (3)
Affiliations (3)
- Department of Oncology, Sichuan Provincial People's Hospital, University of Electronic Science and Technology of China, 610072, Sichuan Chengdu, China.
- Department of Radiation Oncology, Sun Yat-sen University Cancer Center, 510060, Guangdong Guangzhou, China.
- Department of Oncology, Sichuan Provincial People's Hospital, University of Electronic Science and Technology of China, 610072, Sichuan Chengdu, China. [email protected].
Abstract
To investigate the value of cone-beam computed tomography (CBCT)-based delta radiomics for predicting short-term radiotherapy (RT) response in nasopharyngeal carcinoma (NPC). A total of 132 pathologically confirmed NPC patients receiving RT were retrospectively enrolled. Serial CBCT images during weeks 1-4 were collected. Patients were grouped by therapeutic response and randomly divided into training and test sets (7:3). Radiomic features from fractional CBCTs were extracted via Pyradiomics. Temporal delta-radiomic features were derived from interfraction differences. After applying feature normalization and dimensionality reduction, optimal features were selected using analysis of variance (ANOVA), recursive feature elimination, relevant features, and Kruskal-Wallis tests. Ten classifiers, including logistic regression (LR), were trained with 5‑fold cross-validation strategy. Predictive performance was evaluated by receiver operating characteristic (ROC) curves, decision curve analysis (DCA), and the DeLong's test. The LR model based on the CBCT<sub>1st-3rd</sub> temporal interval achieved the optimal predictive performance (balanced accuracy 0.73, area under the curve [AUC] 0.74, sensitivity 0.64, specificity 0.81) in the cross-validation set. DeLong's tests revealed no statistically significant differences (P > 0.05) in AUC values within the cross-validation set between the CBCT<sub>1st-3rd</sub> model and models based on CBCT<sub>1st-4th</sub> or CBCT<sub>2nd-4th</sub> intervals. DCA indicated that the LR model based on CBCT<sub>1st-3rd</sub> temporal interval provided the highest net clinical benefit within threshold probabilities ranging from 0.2 to 0.4 and exceeding 0.65. The CBCT-based delta radiomics models can dynamically assess short-term RT response in NPC patients. This approach offers potential as an early-warning indicator during the RT course and provides a novel approach to guiding personalized precision radiotherapy for NPC.