Intratumoral Heterogeneity of Glioblastoma Infiltration Revealed by Joint Histogram Analysis of Diffusion Tensor Imaging

Chao Li, Shuo Wang, Jiun‐Lin Yan, Rory J. Piper, Hongxiang Liu, Turid Torheim, Hyunjin Kim, Jingjing Zou, Natalie R. Boonzaier, Rohitashwa Sinha, Tomasz Matys, Florian Markowetz, Stephen J. PriceNeurosurgery 2018 · DOI

TL;DR

This paper introduces a novel method using joint histogram analysis of Diffusion Tensor Imaging (DTI) parameters (DTI-p and DTI-q) to better understand the complex, heterogeneous nature of glioblastoma infiltration. This approach reveals distinct tumor subregions that are more invasive and can predict patient survival, offering a more nuanced view beyond traditional imaging.

What Problem Does This Paper Address?

Glioblastoma, a highly aggressive brain tumor, is notoriously difficult to treat due to its profound intratumoral heterogeneity. This means the tumor isn't uniform; it contains diverse cell populations with varying characteristics and behaviors. This heterogeneity leads to inconsistent responses to treatment and makes complete surgical removal challenging, as the tumor often infiltrates surrounding healthy brain tissue in unpredictable ways. Current imaging techniques, like contrast-enhanced T1-weighted MRI, primarily show areas where the blood-brain barrier is compromised, but they struggle to accurately map the extent of tumor infiltration into healthy tissue. Diffusion Tensor Imaging (DTI) offers a way to probe the microstructure of the brain by measuring water molecule movement, but interpreting its complex data for tumor infiltration has been challenging. This paper aims to bridge this gap by developing a more interpretable and prognostic method using DTI.

Methods

The core innovation here is the application of joint histogram analysis to DTI data. Standard DTI analysis often focuses on scalar metrics like Fractional Anisotropy (FA) and Mean Diffusivity (MD). However, this paper decomposes the DTI tensor into two key components:

  • DTI-p (isotropic component): Represents the diffusion of water molecules in all directions, often related to cell density.
  • DTI-q (anisotropic component): Represents the directional diffusion of water molecules, reflecting the integrity of white matter tracts.

The researchers then constructed univariate histograms for DTI-p and DTI-q to visualize their distribution within tumor regions. More importantly, they created joint histograms that plot the co-occurrence of DTI-p and DTI-q values across voxels. This allows for the identification of distinct "voxel groups" based on combinations of increased or decreased DTI-p and DTI-q.

These voxel groups were analyzed separately for both contrast-enhancing and non-enhancing tumor regions. Eight joint histogram features (representing the proportion of each voxel group within these regions) were extracted for each patient. These features were then correlated with tumor progression and patient survival using statistical models, including Cox proportional hazards regression and receiver operating characteristic (ROC) curves.

Datasets

The study utilized data from 115 primary glioblastoma patients prospectively recruited between July 2010 and August 2015. Patients underwent preoperative MRI scans, including DTI. The study also incorporated clinical data such as IDH-1 mutation status, MGMT methylation status, sex, age, and extent of resection.

Key Insights and Ideas

  • DTI-p and DTI-q are more than just scalars: Decomposing the DTI tensor into isotropic (p) and anisotropic (q) components provides a richer picture of tissue microstructure than single scalar values like FA or MD.
  • Joint histograms reveal distinct tumor habitats: The co-occurrence of DTI-p and DTI-q values within tumor regions creates distinct "voxel groups" that appear to represent different microstructural environments within the tumor.
  • Non-enhancing regions are key to invasiveness: While contrast enhancement highlights areas of blood-brain barrier breakdown, the non-enhancing regions, often considered infiltrative, showed specific diffusion patterns (decreased DTI-p and increased DTI-q) that were strongly linked to worse survival. This suggests these regions are more aggressive.
  • Heterogeneity is prognostic: The proportion of these different voxel groups within the tumor, not just their presence, carries significant prognostic information.

Key Results

  • Voxel Group IV (increased DTI-p/decreased DTI-q) was the most prevalent group in both contrast-enhancing and non-enhancing regions.
  • A higher proportion of Voxel Group II (decreased DTI-p/increased DTI-q) in non-enhancing regions was significantly associated with worse progression-free survival (hazard ratio = 1.08, p < 0.001) and overall survival (hazard ratio = 1.11, p < 0.001).
  • In multivariate analysis, specific joint histogram features, particularly Voxel Group II (p↓/q↑) in non-enhancing regions, were significant independent predictors of both PFS and OS. For example, Voxel Group II in non-enhancing regions had a hazard ratio of 1.08 for PFS (p < 0.001) and 1.36 for OS (p < 0.001).
  • The joint histogram features significantly improved the predictive models for survival when added to baseline clinical and molecular factors (e.g., IDH-1 mutation, MGMT methylation).

Limitations and Future Work

  • Single-center study: The findings were derived from a single institution, and the authors acknowledge the need for validation in an independent cohort.
  • Lack of direct biological validation: While previous studies have validated DTI-p and DTI-q correlates with image-guided biopsies, this specific study did not include such direct histological correlation.
  • Future Directions: The authors suggest that this approach could be applied to a broader imaging research field and could inform targeted surgical resection and radiation therapy by identifying more invasive subregions. Further research could explore combining DTI with other imaging modalities and validating these findings with detailed histopathological analysis.

Summary updated

Abstract

Joint histogram features of DTI showed incremental prognostic values over clinical factors for glioblastoma patients. The nonenhancing tumor subregion with decreased isotropic diffusion and increased anisotropic diffusion may indicate a more infiltrative habitat and potential treatment target.
BibTeX
@inproceedings{li2018intratumoral,
  author = {Chao Li and Shuo Wang and Jiun‐Lin Yan and Rory J. Piper and Hongxiang Liu and Turid Torheim and Hyunjin Kim and Jingjing Zou and Natalie R. Boonzaier and Rohitashwa Sinha and Tomasz Matys and Florian Markowetz and Stephen J. Price},
  title = {{Intratumoral Heterogeneity of Glioblastoma Infiltration Revealed by Joint Histogram Analysis of Diffusion Tensor Imaging}},
  booktitle = {Neurosurgery},
  year = {2018},
  doi = {10.1093/neuros/nyy388},
}

Loading related content...

Figure