2200 Predicting Regions of High-Grade Glioma Progression: A Matched Comparison of Radiomics versus Magnetic-Resonance Spectroscopy Machine Learning Models

View at publisher Summarise with VIRAT
Authors: Cameron A. Rivera, Santiago Ceped, Vratko Himič, Alan Ho, Venkat Uppalapati, Maxon V. Knott, Anshul Ratnaparkhi, Ricardo J. Komotar, Ashish H. Shah, Michael E. Ivan
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

INTRODUCTION: The prediction of high-grade (WHO Grade III & IV) glioma growth using machine learning (ML) tools has been a growing field of interest in neuro-oncology. Two previously reported successful methodologies include the use of radiomics or whole-brain magnetic resonance spectroscopy (WB-MRS). While both methods have demonstrated utility, their performance has not been directly compared. METHODS: Two adult patients with histologically confirmed Grade IV gliomas and known progression within 6 months of resection were selected. Neither patient was included in the training or development of the radiomics or WB-MRS-based models. The radiomics model was trained on extracted features from T1, T1-weighted contrast-enhanced, T2, FLAIR, and ADC sequences, while the WB-MRS model incorporated voxel-wise supervised classification based on five extracted metabolites: choline, n-acetyl-aspartate, creatinine, myo-inositol, and glutamine/glutamate. For each patient, both models generated voxel-level probability maps of “future tumor” versus “healthy” tissue. Predictions were compared with true progression delineated on follow-up MRI imaging. Accuracy, sensitivity, and specificity were calculated. RESULTS: Both models successfully classified tissue into either healthy or future tumor with a whole-brain accuracy above 99%. The radiomics model had an average sensitivity of 65.84% and average specificity of 99.39%. The WB-MRS model showed an average sensitivity of 33.49% and average specificity of 96.08%. CONCLUSIONS: This two-case study demonstrates that radiomics- and WB-MRS–derived ML models can prospectively highlight regions destined for HGG progression up to 6 months before they are radiographically apparent. While the sample size limits conclusions on model superiority, future work will include development of hybrid models that integrate radiomic and metabolic features to leverage the complementary strengths of both modalities.

Fields of Study & Subjects

Radiomics and Machine Learning in Medical Imaging Glioma Diagnosis and Treatment Brain Tumor Detection and Classification

Publication details

Journal / source
Neurosurgery
Volume
72
Issue
Supplement_1
Pages
301–301
Publication date
26 March 2026
Language
EN