Assessment of the prognostic value of spatial heterogeneity of hypoxia in FMISO PET images of glioma patients

Objectives: Hypoxia is associated with resistance to radiotherapy and chemotherapy in malignant gliomas. Hypoxia quantification in gliomas involves a simple tissue-to-blood ratio (T:B), for images of a tracer that is trapped when the level of O2 is insufficient for cellular respiration. Previous results for FMISO PET in brain cancer patients imaged before RT showed T:Bmax and Hypoxic Volume (HV=voxels with T:B>threshold) were associated with poor survival and time-to-progression [Spence 2008]. However, in a recently completed multi-center clinical trial (ACRIN 6684), traditional parameters of T:Bmax and HV were not as effective prognostically as SUVpeak for FMISO in the neighborhood of the voxel with greatest uptake [Gerstner 2016]. The present analysis considered both of these study cohorts with a view to reconciling key differences and also examining the potential utility of texture features and spatial heterogeneity as further prognostic variables for brain FMISO images.

Methods: We evaluated a set of radiomic measures assessing spatial heterogeneity and texture characteristics of the hypoxia distribution in 71 brain cancer patients from multiple imaging centers (29 UW and 42 ACRIN 6684 patients) who had FMISO PET studies with concurrent blood sampling. Patients were scanned at a clinically relevant time between diagnostic surgery/biopsy and before RT/ChX. Quantification of tissue hypoxia included T:Bmax, HV, measures of SUV (mean, max, peak) and radiomics measures of texture. An open-source software implementation in R was used to determine 1st and 2nd order statistics of the FMISO uptake distribution (e.g., contrast, entropy, homogeneity) following thresholding of the tumor boundary. A multivariate Cox regression analysis was used to assess the added prognostic benefit of these and other established predictors of clinical outcome based on FMISO imaging.

Results: Multivariate analysis showed that HV was an independent predictor of outcome in the combined data set (p<0.01); SUVpeak failed to reach overall significance in this setting (p=0.11). For the UW cohort, T:Bmax and HV remained independent predictors. While texture (2nd order entropy) was an independent predictor of survival in the ACRIN cohort, it did not contribute to prognosis in the UW set. In neither cohort did spatial heterogeneity of FMISO uptake prove to be prognostic. The sub groups had differences with the degree of tumor burden, where the UW cohort had significantly greater HV than the ACRIN cohort (t-test, p=0.024).

Conclusion: In the entire series, HV, SUVpeak and 2nd order entropy were jointly independent prognostic variables. FMISO hypoxia parameters predictive of survival differ between single center (UW) and multicenter (ACRIN 6684) cohorts. For the ACRIN cohort, entropy and SUVpeak played a strong role, but their prognostic value was not apparent in the UW set. The sub groups had clinical differences due to selection by a single oncologist in the UW cohort for patients with higher tumor burden, while the ACRIN cohort had relaxed selection criteria for tumor volume. Furthermore, the ACRIN cohort had many patients with alternative therapies during and following conventional therapy, where the UW cohort had fewer. Notable differences between the early survival characteristics of the two cohorts, the degree of tumor burden and the inclusion of alternative therapies may partially explain these findings. . <b>Refs:</b> Spence AM, et al. Regional Hypoxia in Glioblastoma Multiforme Quantified with [18F]Fluoromisonidazole Positron EmissionTomography before Radiotherapy: Correlation with Time to Progression and Survival. Clin Canc Res. 14:2623-2630, 2008. Gerstner ER, et al. ACRIN 6684: Assessment of Tumor Hypoxia in Newly Diagnosed Glioblastoma Using 18F-FMISO PET and MRI. Clin Canc Res 22:5079-5086. 2016. Research Support: Supported by NIH/NCI P01-CA042045, U01-CA079778, U01-CA080098 and Science Foundation Ireland grant PI 11/1027.