Bayesian Estimation of Microstructural Parameters in Glioma Patients and Comparison with Genetic Analysis
Elias Kellner1, Marco Reisert1, Ori Staszewski2, Bibek Dhital1, Valerij G Kiselev1, Karl Egger3, Horst Urbach3, and Irina Mader3

1Department of Radiology, Medical Physics, University Medical Center Freiburg, Freiburg, Germany, 2Freiburg, Germany, 3Department of Neuroradiology, University Medical Center Freiburg, Freiburg, Germany

### Synopsis

In a recent study, we proposed a method for fast and direct estimation of mictrostructural tissue parameters such as intra/extraaxonal volume fraction and diffusivities based on multishell DWI. In this study, we report the first method application to human gliomas and demonstrate connections of microstructural parameters with genetic markers IDH and 1p19q in a group of 32 patients.

### Purpose

In another abstract [1], we propose a method for direct estimation of the tissue mictrostructural parameters such as intra/extraaxonal volume fraction and diffusivities based on multishell DWI. The major novelty of the method lies in the fast and robust estimation of the volume fractions for multi-compartmental tissue models with minimal constraints and without the detour via the cumulants such as the diffusion and kurtosis tensors. In the present work, we report the application to 32 human gliomas and investigate the connection of microstrucural changes with the tumor genetics.

### Method

The microstructure determination method is explained in detail in Abstract [1]. In brief, the method relies on a machine learning algorithm based on features derived directly from the DWI signal. The chosen features are rotationally invariant, such that they are independent of fiber orientation and dispersion. With this method, parameters of the tissue microstructure such as intra / extra axonal volume fractions and diffusivities can be obtained within seconds. The presently used microstructural model includes an axonal, an extraaxonal compartments and a fraction of CSF (Figure 1). 32 patients (20 female), median age: 55.5 years (range: 29 – 77 years) were prospectively investigated before stereotactic biopsy. Included were seven low grade astrocytomas WHO grade 2 (A°II), five oligodendrogliomas WHO grade 2 (OD°II), eight anaplastic astrocytomas WHO grade 3 (A°III), three oligodendrogliomas WHO grade 3 (OD°III), and nine Glioblastoma multiforme WHO grade 4 (GBM). Genetic characterization of IDH mutation, 1p19q codeletion and MGMT status were obtained from the neuropathological reports. MRI protocol included MPRAGE contrast-enhanced and flair scans. The parameters for the diffusion protocol were resolution $(2.5mm)^3$, Matrix 84x84x59, b=1000, 2000 $s/mm^2$, 61 directions per shell. The changes inside the tumorous region as indicated by the non-contrast-enhancing FLAIR lesion was for all microstructural maps visually rated in an ordinal scale of “normal”, “changed”, “strongly changed” and these measures were correlated with the genetic characterizations.

### Results

In Figure 2 - 5, we show examples for the microstructure maps, together with MPRAGE, FLAIR and standard DTI maps of mean diffusivity (ADC) and fractional anisotropy. For the correlation with genetics (Table 1), non-parametric testing for the distribution of the median between groups and cross tables and chi-squared test were applied. The intraaxonal volume fraction was not evenly distributed between the IDH wildtype and mutant group. It was “strongly reduced” in 13/14 IDH mutants, whereas it was only “strongly reduced” in 9/18 wildtypes (pcorr= 0.03). Extraaxonal radial diffusivity was not evenly distributed between the IDH wildtype and mutant group. It was “strongly increased” in 10/14 IDH mutants, whereas it was “increased” or “normal” in 15/18 wildtypes (pcorr= 0.005). Concerning the 1p19q codeletion, extraaxonal radial diffusivity was not evenly distributed between the 1p19q codeletion and normal group. It was “strongly increased” in 5/6 cases with 1p19q-codeletion, whereas it was “increased” or “normal” in 15/20 cases without 1p19q-codeletion, (puncorr= 0.03, pcorr= 0.09).

### Discussion and Conclusions

Tumor heterogeneity is a well-known problem. With the present method, contributions of the different tissue compartments can be separated to show more details than standard DTI (Figure 2). Further, in some cases, the tumor is not visible in the standard DTI ADC, but clearly shows up in the extraaxonal volume fractions (Figures 2-4). In general, the volume fractions are most reliable and show the best contrast, whereas the estimates for the diffusivities are less reliable as discussed in abstract submission [1]. Although the results look promising, we warn against over-interpretation: Before clarification in further research, the microscopic diffusivities should be treated as biomarkers with yet not fully understood microstructural content and possible mutual interdependencies. According to the new WHO classification of brain tumors, it is important that the new genetic characterization of the brain tumors according to IDH mutation or wildtype and to the presence or absence of 1p19q codeletion is reflected by these new contrasts. The combination of a reduced intraaxonal volume fraction with an increased radial diffusivity may hint on a dominance of interstitial edema in IDH mutant tumors. A similar trend seems to be visible for an increased radial diffusivity in cases with 1p19q-codeletion. As patients with both or only one of the mutations have a better prognosis than tumors without, it would be one aim for the future to mirror tumor genetics by MR imaging [2,3] to have early prognostic information. Summarizing, the resulting maps for the intra and extraaxonal volume fractions show a great contrast in the tumorous regions and reveal tumor heterogeneities. The comparison to genetic analyses indicates significant connections between the intraaxonal volume fraction and extraaxonal radial diffusivity with IDH mutation.

### Acknowledgements

German Research Foundation (DFG) grant number KI1089/3-2
German Research Foundation (DFG) grant number RE3286/2-1

### References

1. Submitted to this conference, submission # 2120.

2. Diehn M, Nardini C, Wang DS, McGovern S, Jayaraman M, Liang Y, Aldape K, Cha S, Kuo MD. Identification of noninvasive imaging surrogates for brain tumor gene-expression modules. Proc Natl Acad Sci U S A. 2008;105(13):5213-8. doi: 10.1073/pnas.0801279105. Epub 2008 Mar 24.

3. Jamshidi N, Diehn M, Bredel M, Kuo MD. Illuminating radiogenomic characteristics of glioblastoma multiforme through integration of MR imaging, messenger RNA expression, and DNA copy number variation. Radiology. 2014;270(1):1-2. doi: 10.1148/radiol.13130078. Epub 2013 Oct 28.

### Figures

Figure 1. Model for the tissue microstructure, consisting of an intraaxonal compartment $$v_i$$ with diffusion only parallel to the fibers, extraaxonal compartment $$v_e$$, and a free-water compartment $$v_f$$ to account for partial volume and csf, with fixed $$D_f=3\mathrm{\mu m^2/ms}$$. With the proposed method, the parameters can directly be obtained.

Figure 2. In this patient, an Astrocytoma Grade III was diagnosed. The intraaxonal volume fraction v_intra clearly shows a reduced axonal fraction inside the tumorous region. Comparison of v_extra, v_intra and v_csf shows heterogeneity of the tumor.

Figure 3. In this patient, an Astrocytoma Grade III was diagnosed. The tumor is hardly visible in any of the standard DTI measures, whereas it clearly shows up in the extraaxonal volume fraction (white arrow head in v_extra), probably due to extracellular edema.

Figure 4. In this patient, an Astrocytoma Grade III was diagnosed. The tumor is hardly visible in any of the standard DTI measures, whereas it clearly shows up in the extraaxonal volume fraction (white arrow head in v_extra) probably due to extracellular edema.

Figure 5. In this patient, a Glioblastoma was diagnosed. The intraaxonal Diffusivity Dax_intra is increased in the tumorous region. The reason might be imprecise parameter determination for very low intraaxonal fraction.

Table 1. Summary of genetic analysis, ordered by WHO classification. n/a = not assessed.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
1358