Eduard Pogosbekian^{1}, Artem Batalov^{1}, Alexander Turkin^{1}, Igor Pronin^{1}, and Ivan Maximov^{2}

Primary brain gliomas are very a widespread type of intra-axial brain tumours. Brain gliomas have different cellular origins and can be differentiated by WHO 2016 classification. The most aggressive type of glioma is the glioblastoma (GB). Thus, a correct and reliable grading of gliomas, in particular, the GB, is critical for a patient treatment and prognosis. Diffusion kurtosis imaging (DKI) has been applied to glioma validation in order to perform non-invasive evaluation. DKI allowed one to obtain information about microstructure and inhomogeneity differentiation. In the present work we demonstrate advantages of generalised DKI approach for GB detection and evaluation.

We scanned two confirmed GB patients on a 3T GE Signa HDxt scanner (GE
Healthcare) equipped with a 8 channel head coil. Spin-echo EPI sequence was
used, voxel size was 3 mm^{3}, FOV 240 mm^{2}, b-values were 0,
1000 and 2500 s/mm^{2} with 60 diffusion gradient directions for each
non-zero b-value. Informed consent was obtained from the legal representatives
of the patients before any study-related procedures. The study was approved by
the local ethical committee. The generalised DKI signal model using the
cumulant expansion gives^{5}:

$$ ln[S(b,α)] = ln[S_0] - bD + \frac{K}{6} · D^2b^2 + (α-1) \frac{K^2}{54} · D^3b^3 + O(b^4)$$

In a
special case α = 1, Eq. (1) reduces to the conventional DKI signal model^{6}.
For the generalised DKI model we used the α = 2/7. Estimation of the
conventional kurtosis metrics was performed using a linear weighted algorithm^{6}
and generalised kurtosis metrics were estimated by a non-linear constrained optimisation
algorithm exploiting the in-house Matlab scripts (MathWorks, Natick, MA USA).
Three regions of interest (ROI) were manually segmented by the trained
neuroradiologist for each patient: tumour, edema, and contralateral normal
appeared white mater (NAWM) (see Fig. 1).

[1] Maximov et al., Phys. Med. 40 (2017) 24.

[2] Lous et al., Acta Neuropathol. 131 (2016) 803.

[3] Hempel et al., J Neuroradiol. (2017) S0150-9861 (17) 30172.

[4] Jiang et al., Oncotarget 6 (2015) 42380.

[5] Jensen et al., Proc. ISMRM (2017) 1731.

[6] Veraart et al., Neuroimage 81 (2013) 335.

Figure 1.
Manually segmented regions of interests for two GB patients. Colours of ROIs
are red – tumour, green – edema, blue – NAWM.

Figure
2. The estimated MK values obtained for
the conventional (α = 1) and generalised
(α = 2/7) diffusion signal models. The results are presented for two GB
patients.

Figure
3. The scatter plots for the estimated diffusion scalar metrics obtained from
the conventional (CSM) and generalised (GSM) kurtosis models. Colour encodes
the ROI locations: red colour is a tumour, green colour is an edema, and blue
colour is NAWM. The diagonal black line is a linear correlation. The colour
lines are estimated correlations.

Table 1. Mean values and standard deviations of the diffusion scalar
metrics estimated for two GB patients.

Table 2. Statistical comparison of the diffusion metrics estimated from
generalised and conventional DKI signal models. In the Table we presented the
results for the Mann-Whitney tests and evaluated p-values for a rejection of
the H0 hypothesis.