Predicting overall survival in glioblastoma patients from DTI, DCE and DSC MRI data acquired prior to surgery and post-chemoradiotherapy
Lawrence Kenning1, Martin Lowry2, Martin D Pickles1, Chris Roland Hill3, Shailendra Achawal3, and Chittoor Rajaraman3

1Centre for MR Investigations, Hull York Medical School at University of Hull, Hull, United Kingdom, 2Hull York Medical School at University of Hull, Hull, United Kingdom, 3Hull and East Yorkshire Hospitals NHS Trust, Hull, United Kingdom

Synopsis

DTI, DCE and DSC MRI parameters obtained pre-surgery and post-chemoradiotherapy were used to predict overall survival in a cohort of patients with glioblastoma multiforme. Results suggest that preoperative diffusivity measurements contain prognostic information about survival. Following chemoradiotherapy, Ktrans, ve, rCBV and tumour volumes were found to have significant prognostic value with higher values associated with shorter overall survival. Cox regression analysis identified 2 volumes and 2 MR parameters, confirming the Kaplan-Meier findings that preoperative DTI and post-chemoradiotherapy DCE parameters have added prognostic value to more traditional prognostic features such as tumour volume.

Introduction

The aim of this study was to investigate whether functional MR parameters obtained pre-surgery and post-chemoradiotherapy, could be used to predict overall survival in a cohort of patients with glioblastoma multiforme. Furthermore, the study aimed to investigate if certain parameters provided greater prognostic information at specific time points. This work may be of interest to both clinicians and clinical scientists.

Methods

Multiparametric MR data was acquired from 30 patients with histologically proven glioblastoma multiforme prior to surgery (TP1) and 2 weeks post-chemoradiotherapy (TP2) (mean scan interval = 99±11 days). All data was acquired using a 3.0T GE 750 Discovery system and eight channel phased array head coil. Morphological imaging was acquired with the addition of DTI (32 directions), T1 DCE (tdel=5sec) with 5 pre-contrast flip angles, and T2* DSC (tdel=2sec).

Motion correction and registration was applied using FSL1, 2. Data was processed using in-house developed software. Calculated DTI parameters were: mean diffusivity (MD), fractional anisotropy (FA), anisotropic component of diffusion (q), longitudinal (LD), and radial diffusivity (RD). Pharmacokinetic modelling using a two compartment Tofts-Kety model was applied to the DCE-MRI data transformed to contrast concentration using T1 values calculated from multi-flip angle data (R1). DSC-MRI was processed using a contrast agent extravasation correction model3 and normalised to global white matter (rCBV). Parametric volumes were created by registering all maps into a single 4D [x, y, z, parameter] volume2.

Volumes of interest (TUM) were manually contoured using morphological imaging (T2 abnormality + T1 post-contrast abnormality – necrosis/cyst – haemorrhage). Mean values were calculated for each parameter. Gaussian mixture modelling (limited to 2 populations) was applied to the VOI of each parameter map (Figure 1), generating a further two means, sorted in ascending order, and labelled P0 and P1 respectively. Parameters were dichotomised using the median value of each measure prior to Kaplan-Meier survival analysis with Log rank tests used to calculate significant survival differences between groups. Cox regression survival analysis was subsequently implemented using a forward Wald methodology to evaluate interactions between time and MR parameters.

Results

At the time of censoring, 19/30 patients were deceased with a median follow up time of 902 days (range, 322-1563 days). Median survival was 508 days (range, 124-1563 days). Age was not a significant factor in overall survival (P=0.718). From the univariate pre-surgical MRI, 5/9 diffusivity, 1/3 perfusion and 2/3 volume measurements were significantly related to overall survival (P<0.05). Post-chemoradiotherapy MRI found 3/9 pharmacokinetic, 2/3 perfusion and 3/3 volume measurements to significantly associated with overall survival. R1 values were not significant predictors of overall survival at either time point, however, 2/3 post-chemoradiotherapy measurements had P-values < 0.1. Measurements of anisotropic diffusion (FA and q) showed no association with overall survival.

Discussion

The results suggest that preoperative diffusivity measurements contain useful prognostic information relating to overall survival, with lower diffusivity values (more cellular/rapidly proliferating tumours) leading to shorter survival intervals. Anisotropic diffusion parameters showed no associated significance, suggesting that mean diffusivity or even ADC calculated from DWI may be sufficient. Gaussian mixture modelling appears useful for sampling rCBV maps, with the preoperative rCBV P1 subpopulation reaching significance (P=0.035) even when the mean rCBV (TP1) did not (P=0.096). High rCBV values were associated with shorter overall survival. The ability to automatically determine regions of increased perfusion within the total volume of abnormality could have clinical utility and reduce intra-user variability. Interestingly the total volume of tumour abnormality (TUM) and volume of non-enhancing tumour (TUM P0) were significant prognosticators, yet the volume of enhancing tumour on the preoperative scans was not. This may be related to the surgical target and the potential extent of resection.

Following chemoradiotherapy, Ktrans, ve, rCBV and tumour volumes were found to have significant prognostic value (P<0.05) with higher values associated with shorter overall survival. In our cohort, DTI had no prognostic value at this time point following chemoradiotherapy. Significant vascular parameters at this time point were more numerous than at the pre-surgical scan, and may be useful in distinguishing treatment related pseudoprogression/reactive changes from residual tumour and/or progressive disease.

Cox regression analysis identified TP1 MD P0, TP2 Ktrans P0, TP2 TUM and TP2 TUM P1 as being key parameters related to survival and confirms the Kaplan-Meier findings that preoperative DTI and post-chemoradiotherapy DCE parameters have added prognostic value to more traditional prognostic features such as tumour volume.

Conclusions

The results from this study suggest that preoperative DTI, DSC and tumour volume measurements all have significant prognostic value prior to treatment in glioblastoma patients. Following chemoradiotherapy, DCE, DSC and tumour volume measurements are better prognostic predictors. Cox regression analysis suggests that preoperative DTI and post-treatment DCE MRI are the most useful sequences in addition to traditional tumour volume measurements.

Acknowledgements

This work is kindly funded by Yorkshire Cancer Research.

References

1. Jenkinson M, Beckmann CF, Behrens TE, Woolrich MW, Smith SM. FSL. NeuroImage 2012; 62(2): 782-90.

2. Kenning L, Lowry M, Turnbull LW. A Novel Scheme for Producing Multi-Parametric Volumes Proc Intl Soc Mag Reson Med; 2013 2013; 2013.

3. Boxerman JL, Schmainda KM, Weisskoff RM. Relative cerebral blood volume maps corrected for contrast agent extravasation significantly correlate with glioma tumor grade, whereas uncorrected maps do not. American Journal of Neuroradiology 2006; 27(4): 859-67.

Figures

TUM VOI overlaid on the T2 FLAIR (left) with the same VOI overlaid on the registered rCBV map (centre). The GMM results are displayed on the same rCBV map (right) with the yellow VOI being the lower rCBV subpopulation (P0) and the red VOI being the higher rCBV subpopulation (P1).

Table showing the Kaplan-Meier results based on the dichotomisation of MR parameters measured pre-surgery (TP1) and post-chemoradiotherapy (TP2). P0 and P1 denote the Gaussian mixture model subpopulations sorted in ascending order. Red cells denote P-values <0.05 whilst yellow values showed P-values<0.1 but >0.05.

Table showing the Cox regression model results based on the dichotomisation of MR parameters measured pre-surgery (TP1) and post-chemoradiotherapy (TP2). P0 and P1 denote the Gaussian mixture model subpopulations sorted in ascending order. The hazard ratios show the risk associated with being dichotomised into the higher group for each parameter.

Kaplan-Meier plots of the 4 MR parameters obtained pre-surgery (TP1) and post-chemoradiotherapy (TP2) that Cox Regression analysis identified as having significant predictive value. P0 and P1 denote the Gaussian mixture model subpopulations sorted in ascending order. Green represents values above the median parameter value and blue is lower than median.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
4347