Radiogenomic analysis of hypoxia pathway is predictive of overall survival in Glioblastoma

Niha Beig, Jay Patel, Prateek Prasanna, Virginia Hill, Amit Gupta, Ramón Correa, Kaustav Bera, Salendra Singh, Sasan Partovi, Vinay Varadan, Manmeet S. Ahluwalia, Anant Madabhushi, Pallavi TiwariScientific 2018 · DOI

Abstract

Hypoxia, a characteristic trait of Glioblastoma (GBM), is known to cause resistance to chemo-radiation treatment and is linked with poor survival. There is hence an urgent need to non-invasively characterize tumor hypoxia to improve GBM management. We hypothesized that (a) radiomic texture descriptors can capture tumor heterogeneity manifested as a result of molecular variations in tumor hypoxia, on routine treatment-naïve MRI, and (b) these imaging-based texture surrogate markers of hypoxia can discriminate GBM patients as short-term (STS), mid-term (MTS), and long-term survivors (LTS). 115 studies (33 STS, 41 MTS, 41 LTS) with gadolinium-enhanced T1-weighted MRI (Gd-T1w), T2-weighted (T2w), and FLAIR MRI protocols and the corresponding RNA sequences were obtained. After expert segmentation of necrotic, enhancing, and edematous/nonenhancing tumor regions for every study, 30 radiomic texture descriptors were extracted from every region across every MRI protocol. Using the expression profile of 21 hypoxia-associated genes, a hypoxia enrichment score (HES) was obtained for the training cohort of 85 cases. Mutual information score was used to identify a subset of radiomic features that were most informative of HES within 3-fold cross-validation to categorize studies as STS, MTS, and LTS. When validated on an additional cohort of 30 studies (11 STS, 9 MTS, 10 LTS), our results revealed that the most discriminative features of HES were also able to distinguish STS from LTS (p = 0.003).

Introduction

Glioblastoma (GBM) remains the most aggressive primary brain malignancy, characterized by a dismal median survival of approximately 14 months. A primary driver of treatment resistance and poor prognosis is tumor hypoxia—regions of low oxygen tension within the tumor mass. Hypoxic cells are notoriously resistant to standard-of-care chemotherapy and radiation, often surviving treatments that effectively target well-oxygenated regions.

Identifying hypoxia currently relies on invasive biopsies or specialized PET imaging, neither of which are standard for routine GBM management. This lack of non-invasive characterization hinders clinical trials for anti-angiogenic therapies, which often fail due to the inability to identify patients with specific hypoxic profiles. This study explores a "radiogenomic" approach, leveraging routine MRI to identify sub-visual texture markers that correlate with genetic hypoxia signatures and patient survival.

Methods

The researchers developed a multi-step radiogenomic framework utilizing data from 115 patients in The Cancer Imaging Archive (TCIA), which provides paired multi-parametric MRI and RNA-seq data.

Multi-Parametric MRI Processing

The study utilized three standard MRI protocols: gadolinium-enhanced T1-weighted (Gd-T1w), T2-weighted (T2w), and Fluid-Attenuated Inversion Recovery (FLAIR). These images were co-registered and intensity-standardized to ensure cross-patient comparability.

Multi-parametric MRI (T1w and T2w) used for feature extraction

Expert Segmentation and Feature Extraction

Radiologists manually segmented three distinct tumor habitats: the necrotic core, the enhancing tumor, and peritumoral edema. From these regions, 30 texture descriptors—including Laws energy, Gabor filters, and Haralick features—were extracted across all MRI protocols, totaling 270 features per patient.

FLAIR MRI showing tumor segmentation

Genomic Integration and Classification

A Hypoxia Enrichment Score (HES) was calculated from the expression of 21 hypoxia-associated genes. Mutual Information (MI) was then used to select radiomic features most predictive of HES. Finally, a Random Forest classifier was trained on 85 cases to categorize patients into survival groups: Short-Term (STS, <7 months), Mid-Term (MTS, 7–16 months), and Long-Term (LTS, >16 months).

Results

The analysis identified eight key radiomic features, primarily Laws and Haralick descriptors, that were highly informative of the tumor's hypoxia profile. On an independent validation set of 30 cases, the model successfully distinguished STS from LTS with a significant p-value of 0.003.

When combining radiomic features with clinical markers (age and KPS score), the model achieved a Concordance Index (C-index) of 0.83 for STS vs. LTS discrimination. Notably, the most predictive features were concentrated in the enhancing tumor and edematous regions, suggesting that the tumor-brain interface is a critical site for hypoxia-driven morphological changes.

Discussion

This study demonstrates that computer-extracted texture features from routine MRI can serve as non-invasive surrogates for molecular hypoxia. While promising, the findings are limited by the relatively small sample size and the retrospective nature of the TCIA dataset. Future work should focus on prospective validation in larger, multi-institutional cohorts and the integration of advanced imaging modalities like Perfusion MRI. Furthermore, expanding the radiogenomic map to include other biological pathways, such as tumor infiltration, could further refine survival prediction and personalized treatment strategies for GBM.

Summary updated

Abstract

Hypoxia, a characteristic trait of Glioblastoma (GBM), is known to cause resistance to chemo-radiation treatment and is linked with poor survival. There is hence an urgent need to non-invasively characterize tumor hypoxia to improve GBM management. We hypothesized that (a) radiomic texture descriptors can capture tumor heterogeneity manifested as a result of molecular variations in tumor hypoxia, on routine treatment naïve MRI, and (b) these imaging based texture surrogate markers of hypoxia can discriminate GBM patients as short-term (STS), mid-term (MTS), and long-term survivors (LTS). 115 studies (33 STS, 41 MTS, 41 LTS) with gadolinium-enhanced T1-weighted MRI (Gd-T1w) and T2-weighted (T2w) and FLAIR MRI protocols and the corresponding RNA sequences were obtained. After expert segmentation of necrotic, enhancing, and edematous/nonenhancing tumor regions for every study, 30 radiomic texture descriptors were extracted from every region across every MRI protocol. Using the expression profile of 21 hypoxia-associated genes, a hypoxia enrichment score (HES) was obtained for the training cohort of 85 cases. Mutual information score was used to identify a subset of radiomic features that were most informative of HES within 3-fold cross-validation to categorize studies as STS, MTS, and LTS. When validated on an additional cohort of 30 studies (11 STS, 9 MTS, 10 LTS), our results revealed that the most discriminative features of HES were also able to distinguish STS from LTS (p = 0.003).
BibTeX
@inproceedings{beig2018radiogenomic,
  author = {Niha Beig and Jay Patel and Prateek Prasanna and Virginia Hill and Amit Gupta and Ramón Correa and Kaustav Bera and Salendra Singh and Sasan Partovi and Vinay Varadan and Manmeet S. Ahluwalia and Anant Madabhushi and Pallavi Tiwari},
  title = {{Radiogenomic analysis of hypoxia pathway is predictive of overall survival in Glioblastoma}},
  booktitle = {Scientific},
  year = {2018},
  doi = {10.1038/s41598-017-18310-0},
}
Figure