Radiomics-Based Decision Support Tool for Pediatric Medulloblastoma Risk Stratification Open Access
Reddy, Kartik (Spring 2026)
Abstract
Background: Medulloblastoma (MB) is the most common malignant pediatric brain tumor and remains associated with substantial morbidity and treatment failure despite multimodality therapy. Current risk stratification relies largely on postoperative clinical, histologic, and molecular features and remains limited for individualized prognostication. Radiomics offers a noninvasive method to extract quantitative imaging features from routine MRI and may improve preoperative risk assessment. We evaluated whether multicompartment radiomic features from pretreatment multiparametric MRI could predict molecular subgroup, histology, progression-free survival (PFS), and overall survival (OS) in pediatric MB.
Methods: In this retrospective single-center study, children and young adults with pathologically confirmed MB and available presurgical MRI were identified. Radiomic features were extracted from tumoral and peritumoral compartments on post-contrast T1-weighted, T2-weighted, and ADC images after expert segmentation and preprocessing. Molecular subgroup classification was dichotomized as sonic hedgehog (SHH) versus non-SHH, and histology as anaplastic versus non-anaplastic. Classification models were developed using random forests after correlation filtering and minimum redundancy maximum relevance feature selection. Survival models for PFS and OS were developed using Cox proportional hazards modeling. Performance was assessed on held-out test sets using area under the receiver operating characteristic curve (AUC), concordance index (C-index), Brier score, and Kaplan-Meier risk-group separation.
Results: Classification analyses included 105 patients; survival analyses included 75 patients for OS and 76 for PFS. For SHH versus non-SHH classification, the combined tumoral and peritumoral model achieved the best held-out discrimination (AUC 0.83), compared with 0.79 for the peritumoral model and 0.70 for the tumoral model. For anaplastic versus non-anaplastic histology, the combined model also performed best (AUC 0.78). For survival prediction, the combined model achieved the highest held-out C-index for both PFS (0.77) and OS (0.78), with significant Kaplan-Meier separation for both endpoints.
Discussion: Pretreatment multiparametric MRI radiomic features were associated with molecular subgroup, histology, and survival outcomes in pediatric MB. Combined tumoral and peritumoral models generally outperformed single-compartment models, suggesting added prognostic value from tissue beyond the visible tumor. These exploratory findings support further validation of multicompartment radiomics for preoperative decision support in pediatric MB.
Table of Contents
1.Introduction
2.Materials and Methods
2.1 Patient Selection
2.2 Outcome Definitions
2.3 Clinical and Imaging Data
2.4 Segmentation and Radiomic Feature Extraction
2.5 Feature Screening and Model Development
2.6 Statistical Evaluation
3. Results
3.1 Cohort Summary
3.2 Molecular Subgroup and Histology Classification
3.3 Survival Modeling
4.Discussion
5.Conclusion
About this Master's Thesis
| School | |
|---|---|
| Department | |
| Degree | |
| Submission | |
| Language |
|
| Research Field | |
| Keyword | |
| Committee Chair / Thesis Advisor | |
| Committee Members |
Primary PDF
| Thumbnail | Title | Date Uploaded | Actions |
|---|---|---|---|
|
|
Radiomics-Based Decision Support Tool for Pediatric Medulloblastoma Risk Stratification () | 2026-04-30 10:39:31 -0400 |
|
Supplemental Files
| Thumbnail | Title | Date Uploaded | Actions |
|---|