Hyperpolarized Magnetic Resonance Imaging, Nuclear Magnetic Resonance Metabolomics, and Artificial Intelligence to Interrogate the Metabolic Evolution of Glioblastoma.

Abstract

Glioblastoma (GBM) is a malignant Grade VI cancer type with a median survival duration of only 8-16 months. Earlier detection of GBM could enable more effective treatment. Hyperpolarized magnetic resonance spectroscopy (HPMRS) could detect GBM earlier than conventional anatomical MRI in glioblastoma murine models. We further investigated whether artificial intelligence (A.I.) could detect GBM earlier than HPMRS. We developed a deep learning model that combines multiple modalities of cancer data to predict tumor progression, assess treatment effects, and to reconstruct in vivo metabolomic information from ex vivo data. Our model can detect GBM progression two weeks earlier than conventional MRIs and a week earlier than HPMRS alone. Our model accurately predicted in vivo biomarkers from HPMRS, and the results inferred biological relevance. Additionally, the model showed potential for examining treatment effects. Our model successfully detected tumor progression two weeks earlier than conventional MRIs and accurately predicted in vivo biomarkers using ex vivo information such as conventional MRIs, HPMRS, and tumor size data. The accuracy of these predictions is consistent with biological relevance.