Construction and validation of a multimodal MRI-based quantitative feature prediction model for the prognosis of...

Accurate prognosis assessment of non-metastatic primary osteosarcoma is essential for treatment decisions. This study aimed to develop and validate a pre-treatment predictive model using multimodal magnetic resonance imaging (MRI) quantitative parameters.

This retrospective study included patients with non-metastatic primary osteosarcoma who received treatment at our hospital. Patients were divided into good or poor prognosis groups based on Response Evaluation Criteria in Solid Tumors at 30-month follow-up. We analyzed multimodal MRI data of these patients. Evaluated parameters included intramedullary extension measured by T2-weighted imaging, pure diffusion coefficient (D value), pseudo-diffusion coefficient (D* value), and apparent diffusion coefficient (ADC value) from diffusion-weighted imaging, and the contrast agent back-flux rate constant (Kep) from dynamic contrast-enhanced MRI. All these parameters were assessed as pre-treatment.

The training cohort included 169 good prognosis and 52 poor prognosis patients. Good prognosis patients showed significantly higher Kep (1.32 ± 0.24 vs 1.21 ± 0.21, P = 0.006), D value (0.95 ± 0.13 vs 0.84 ± 0.25, P = 0.003), D* value (19.78 ± 5.45 vs 17.34 ± 4.34, P = 0.004), and ADC value (1.11 ± 0.17 vs 1.01 ± 0.16, P<0.001), but lower intramedullary extension (9.62 ± 1.22 vs 10.50 ± 2.33, P = 0.012) compared to those with poor prognosis. The area under the curve (AUC) of the multivariate model integrating these features was 0.836. External validation confirmed the model's discriminatory ability (AUC = 0.812) and reproduced significant differences in Kep, intramedullary extension, D value, D* value, and ADC value between groups.

This study developed a predictive model based on multimodal MRI quantitative features that effectively identified poor prognosis in non-metastatic primary osteosarcoma, providing a non-invasive assessment tool to optimize treatment strategies.