MRI-based intratumoral and peritumoral radiomics predicting neoadjuvant chemotherapy response in osteosarcoma

To evaluate the predictive performance of a nomogram that integrates intratumoral and peritumoral MRI-based radiomics with clinical variables for assessing the efficacy of neoadjuvant chemotherapy (NAC) in patients with osteosarcoma (OS).

This retrospective study included 93 patients with pathologically confirmed OS who underwent standard NAC. Intratumoral regions were manually segmented on axial T2-weighted fat-suppressed (T2WI-FS) images using ITK-SNAP, and peritumoral regions were generated semi-automatically by isotropic expansions of 2 mm, 4 mm, and 6 mm. Random forest classifiers were constructed separately for intratumoral, peritumoral, and combined intratumoral-peritumoral radiomics features. The optimal radiomics model was incorporated with significant clinical predictors to build an individualized nomogram. Model performance was assessed through the F1 score, Delong's test and receiver operating characteristic (ROC) curve analysis. Decision curve analysis (DCA) was applied to assess the model's clinical utility.

Multivariate logistic regression identified alkaline phosphatase (ALP) (OR = 1.003, 95% CI: 1.000 ~ 1.006, P = 0.031) and pathological fracture (PF)(OR = 2.575, 95% CI: 1.036 ~ 6.401, P = 0.042) as independent predictors of NAC response. Among all radiomics models, the Model_rad-intra + peri^2mm combination model demonstrated the best performance, achieving AUCs of 0.888 in the training set and 0.765 in the test set. The integrated nomogram further improved predictive accuracy, with AUC of 0.990 and 0.815 in the training and test sets, respectively.

We developed and validated a nomogram that combines intratumoral and peritumoral MRI radiomics with clinical variables for predicting NAC efficacy in OS. The model demonstrated robust performance and may support early, individualized treatment evaluation and clinical decision-making in patients undergoing NAC.