Differentiating medulloblastoma and pilocytic astrocytoma in children based on multimodal MRI radiomics model.

purposeMedulloblastoma (MB) and pilocytic astrocytoma (PA) are common in pediatric brain tumors and difficult to distinguish. To establish and evaluate a radiomics model based on multimodal MRI to distinguish medulloblastoma from pilocytic astrocytoma.

methodsRetrospective collection of magnetic resonance images from 113 patients with MB and 74 patients with PA. Radiomics analysis was performed on Dynamic Contrast-Enhanced T1 weighted imaging (DCE-T1WI), T2 weighted imaging (T2WI), and Apparent diffusion coefficient (ADC) images, respectively. A multimodal MRI combined radiomics model incorporating DCE-T1WI, T2WI, and ADC sequence features was developed by extraction of valuable features, and the radiomics nomogram was generated to evaluate its diagnostic capability. The DeLong test was used to compare the diagnostic performance of DCE-T1WI, T2WI, ADC single sequence model, and multimodal MRI radiomics model.

resultsThe combined model showed the highest performance among all models, with an area under the curve (AUC) of 0.999 on the primary cohort and maintained an AUC of 0.994 during the validation cohort. For the single-sequence model, the ADC sequence model performs better in the primary cohort, with an AUC of 0.996; the T2WI sequence model slightly outperforms in validation cohort, with an AUC of 0.985. The DCE-T1WI sequence model performed slightly worse than ADC and T2WI in the validation cohort, with an AUC of 0.951. Overall, the combined model performed best in the differential diagnosis of MB and PA.

conclusionMultimodal MRI-based radiomics analysis is effective in differentiating MB from PA and radiomics imaging may have important clinical significance in the preoperative detection of posterior fossa brain tumors in children.