Abstract Purpose Despite the recent improvements in survival rates for pediatric medulloblastoma (MB), disease survivors may get severely affected by late sequalae due to overtreatment, while patients with high risk could be potentially undertreated. The current risk-stratification (Chang’s, molecular subgrouping), while guiding clinical trials, still lacks the ability to recognize the extensive heterogeneity in MB. Development of complementary prognostic tools for risk-stratification can help develop tailored treatment regimens with dose escalation/de-escalation for eligible patients. This work conducts a histomorphometric analysis (extracting computational attributes from histopathology slides) on MB for risk-stratification. Our rationale is that histomorphometry can analyze the highly cellular microscopic structure of MB and its histology variants, hence provide additional prognostic insights. Methods Hematoxylin and eosin slides for 80 patients (3–21 years old) diagnosed with MB were collected from the Children’s Brain Tumor Network. Data was split into training and test sets (70:30 ratio). Quality control followed by nuclei segmentation were conducted on the Whole Slide Images (WSIs). Features were then extracted, including shape/morphology features (angularity, roundness), Haralick textural features (capturing variations in chromatin staining), and architectural graph-based features at both nuclei-level and centroids-of-nuclei cluster-level, as the graph nodes. Finally, the WSI representation for every patient was acquired via the statistics (mean, median, kurtosis, skewness) of features across all the WSI patches, obtaining the final feature vector. The representations were fed to Cox proportional hazards models with different penalties (LASSO, ridge) and cross-validation schemes within the training set to perform survival analysis. Results The models were able to risk-stratify patients on the training set into 2 groups (average/high) (p-value0.0001, concordance-index=0.78) using LASSO regularization. Applying the model’s parameters on the test set yielded significant differences between the risk groups (p-value=0.04, concordance-index=0.71). Conclusion Histomorphometric analysis may be used complementary to existing MB risk assessment, offering additional prognostic insights.
Ismail et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: