SAGE-Based Habitat Analysis for Distinguishing Meningiomas, Metastases, and Gliomas

In a multicenter study, SAGE-derived habitat analysis with RBF-SVM differentiated meningiomas, brain metastases, and gliomas with internal macro-AUC 0.700±0.086. External validation macro-AUC was 0.629. A simpler model using mean features alone gave external macro-AUC 0.849.

• K=3 habitat clusters provided the most parsimonious and reproducible representation.
• A simpler model (mean-only features + linear SVM) unexpectedly outperformed habitat-based models on external validation (macro-AUC 0.849 vs. 0.629).
• SHAP analysis confirmed that both global mean and habitat-derived features contributed to classification.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.