Introduction

Glioblastoma (GBM) is the most common form of central nervous system malignancy in adults, and carries a particularly poor prognosis with median expected survival of just over a year.¹ Standard protocols for assessing therapy response and progression utilize approximations of tumor burden based on bi-dimensional measurements of contrast-enhancing tissue and fluid-attenuated inversion recovery (FLAIR) signal. FLAIR signal is presumed to represent tumor cell infiltration into healthy tissues, causing edema. Current prognostic methods use only volumetric changes to determine progression, however a voxel-wise approach call the Parametric Response Map (PRM) has been shown to improve sensitivity to changes in quantitative imaging maps, especially in cases of small localized or heterogeneous response.² We propose that application of PRM to normalized FLAIR imaging may provide an early indication of progression risk, before gross anatomical changes can be distinguished.

Methods

Subjects (N=52) were recruited as a single-site prospective trial including patients with diffuse high-grade glioma (WHO grade III and IV) to be treated with radiation therapy (IR).³ Treatment response was assessed at approximately 10 weeks using the Macdonald criteria, and a subset of responders (N=26; partial response, PR, or stable disease, SD) was monitored for progression. Standard magnetic resonance imaging (MRI) was performed, including T2-weighted fluid-attenuated inversion recovery (T2/FLAIR) acquisitions. T2/FLAIR images were normalized to the mean signal of frontal white matter, resulting in maps of the relative FLAIR signal, or rFLAIR. Hyperintense regions on T2/FLAIR were manually delineated by an experienced radiologist for quantitative analysis of relative change from baseline (dV_FLAIR). Images were spatially registered serially using open source software (Elastix⁴). Briefly, a multi-resolution b-spline optimization was used over the whole brain with mutual information describing the cost function along with a bending energy penalty. Parametric Response Maps (PRM) were generated on the aligned rFLAIR images within the union of FLAIR VOIs using a threshold of +/-0.25, based on the 95% confidence interval found for normal brain tissue far from the lesion. This resulted in relative volumes of increased (PRM_rFLAIR+), decreased (PRM_rFLAIR-), or unchanged (PRM_rFLAIR0) signal. Evaluation of PRM_rFLAIR and dV_FLAIR was performed using receiver-operator curve (ROC) analysis to assess overall performance (using the area under the curve, AUC) and find optimal stratification between early and late overall survival (OS) and progression-free survival (PFS). Kaplan-Meier curves were then generated and log rank significance was assessed at p-values < 0.05.

Results

PRM_rFLAIR+ and dV_FLAIR were both found to be predictive of both OS and PFS when assessed at approximately 11 weeks post-IR (Figure 1). Optimal cutoffs for stratification of early or late survival using volume changes in this time period were 10.3% or 15.8% increases for OS and PFS, respectively. Cutoffs for PRM_rFLAIR+ were 15.3% and 11.5%, respectively. An early time point of approximately 20 weeks post-IR (~10 weeks post-response assessment) was selected for evaluation of progression risk (Figure 2). Here, dV_FLAIR was not found to be predictive of clinical outcomes, while PRM_rFLAIR+ significantly stratified early/late survival with high specificity and sensitivity. The time to PRM_rFLAIR+ progression was then evaluated using an empirically-derived threshold of 10% relative volume. ROC analysis found an optimal stratification cutoff of 5.6 months to exceed 10% PRMrFLAIR+, resulting in a significant stratification of OS and PFS (Figure 3).

Discussion

The goal of this study was to highlight PRM of T2/FLAIR imaging as a potentially valuable early indicator of patient progression risk. FLAIR imaging is currently incorporated in standard protocols for monitoring GBM, allowing for implementation of this method without affecting standard clinical workflow. PRM_rFLAIR was shown to be sensitive to therapeutic response as seen by a significant stratification at 11 weeks post-IR, comparable to the current RANO standard criteria. When evaluated between weeks 11 and 20 post-IR, PRM_rFLAIR was predictive of survival even while volume change was not. Further, when evaluated as a progression marker with a threshold of 10% relative volume, PRM_rFLAIR was able to significantly stratify recurrence risk nearly 10 months earlier than the standard RANO criteria.

Conclusion

It is now known through exhaustive research that assessment of tumor-associated edema, visualized using FLAIR imaging, is critical for the accurate depiction of tumor burden as well as radiation treatment planning. Despite continual improvements in computer systems and analysis software, advanced image analysis tools for the assessment of treatment response and recurrence remain underutilized. PRM of normalized FLAIR MRI can be readily implemented in clinical workflows, and provides a significant early indication of recurrence risk.

Acknowledgements

No acknowledgement found.