Radiomic-based prognostic score for survival risk-stratification in pediatric medulloblastoma tumors: A...

Medulloblastoma (MB) is the most common malignant brain tumor in children and is known for substantial heterogeneity. MB is classified into low/average- and high-risk; however, improving risk-stratification remains one of the biggest challenges in MB. Enriching risk-stratification offers potential treatment intensification for high-risk MB while decreasing treatment sequalae in low-risk MB through treatment de-escalation. This work presents a new radiomics-based medulloblastoma prognostic (RaMP) score that quantifies the tumor heterogeneity by analyzing routine clinical imaging. The hypothesis is that radiomic (computational) features can analyze the complex behavior and heterogeneity of MB tumors, allowing for segregating high-risk tumors from the ones with low risk.

One hundred and nineteen MRI scans for MB patients were collected from 3 institutions (Site 1: n = 42, Site 2: n = 47, Site 3: n = 30). Following segmentation of the tumor habitat (edema, enhancing lesion, and non-enhancing tumor + cystic core) and pre-processing, features capturing (1) structural deformations from the normal regions around the tumor, (2) textural attributes, and (3) morphological changes, were extracted. RaMP was created by combining the features then feeding them to Cox proportional hazards models, categorizing the subjects into low and high risks.

RaMP outperformed clinical standard-of-care assessment when employed for risk-stratification. For instance, when using Sites 1 and 3 for training and Site 2 for testing, significant differences were observed across edema and the tumor habitat with P = .02, .013, and C-indices of 0.7, 0.63, respectively, between the risk groups.

Our work shows the promise of radiomic analysis in risk-stratification and predicting outcomes in pediatric MB.