Predicting chemotherapy response in pediatric lymphoma using pre-treatment contrast-enhanced CT radiomics: A hypothesis-generating study.

[PURPOSE] To preliminarily evaluate the predictive value of baseline contrast-enhanced CT (CECT) radiomics for assessing chemotherapy response in pediatric lymphoma.

[METHODS] This retrospective study included 92 pediatric patients with lymphoma (72 males, 20 females). Patients were classified as responders (n = 74) and non-responders (n = 18) based on treatment outcomes. The cohort was randomly stratified into a training set (n = 65, 70%) and a test set (n = 27, 30%). A total of 960 radiomics features were extracted from venous-phase baseline CECT images of target lesions. Feature selection was performed, and a logistic regression model was developed for response classification using the Synthetic Minority Over-sampling Technique (SMOTE). To evaluate model robustness, the entire radiomics pipeline was repeated across 10 independent randomized train-test splits. Model performance was evaluated using receiver operating characteristic (ROC) curve analysis, reporting the area under the ROC curve (AUC), 95% confidence intervals (CIs), and accuracy.

[RESULTS] Eight radiomics features were selected for the final model, including four filter-transformed first-order features and four filter-transformed texture features. The SMOTE model achieved an AUC of 0.883 (95% CI: 0.799-0.967) and an accuracy of 0.800 in the training set. In the test set, the SMOTE model achieved an AUC of 0.809 (95% CI: 0.606-1.000) and an accuracy of 0.741. In repeated validation, the SMOTE model showed mean AUCs of 0.915 (training) and 0.767 (test) across 10 splits.

[CONCLUSION] This hypothesis-generating study demonstrates that baseline CECT radiomics shows promise for predicting chemotherapy response in pediatric lymphoma.