The Diffusion Kurtosis Imaging (DKI) model can successfully characterize non-Gaussian diffusion. In turn, the White Matter Tract Integrity (WMTI) model is proposed to be based on the DKI model to further characterize the intra- and extra-axonal compartments. However, the accuracy with which model parameters reflect the underlying tissue characteristics has not been tested. Here, we compared two MRI model metrics using a unique combined post-mortem MRI and histopathology approach. Preliminary results show that AWF, $$$D_{e,\perp}$$$and MD from MRI correlate strongly with myelin fraction and that $$$D_{e,\parallel}$$$ from MRI correlates strongly with astrocyte fraction.
DKI model: the diffusion signal is modeled as:
$$\frac{S(b)}{S(0)}=e^{-bD+\frac{1}{6}b^2D^2K+O(b^3)}$$
The model is extended to characterize the non-Gaussian diffusion pattern by introducing symmetric fourth order diffusion kurtosis tensor.
WMTI model is a model that relates DKI metrics directly to WM microstructure. This model applies to highly aligned fiber bundles and partitions water into two compartments, the intra- and extra-axonal space, with the metrics of AWF, $$$\bf D_a$$$, and $$$\bf D_e$$$ based on the following relationships derived by Fieremans2.$$f=\frac{K_{max}}{K_{max}+3(1-\frac{\sqrt{K_{max}}(D_a-D_{a,min})}{D_1\sqrt{K_1}+D_2\sqrt{K_2}+D_3\sqrt{K_3}})^2}, D_{e,i}=D_i(1+\sqrt{\frac{K_if}{3(1-f)}}), D_{a,i}=D_i(1-\sqrt{\frac{K_i(1-f)}{3f}}) (i=1,2,3)$$ where $$$f$$$ is axon water fraction (AWF), $$$K_{max}$$$ is the maximum kurtosis over all directions, and $$$D_a$$$ is the intra-axonal diffusion coefficient with $$$D_{a,min}$$$ being a lower bound for $$$D_a$$$. $$$D_i$$$, $$$K_i$$$, $$$D_{e,i}$$$, and $$$D_{a,i}$$$ with $$$i=1,2,3$$$ are the overall diffusivity, kurtosis, extra-axonal diffusivity and intra-axonal diffusivity along the axis directions of a chosen reference frame. And the extra-axonal axial $$$D_{e,\parallel}$$$ and radial $$$D_{e,\perp}$$$ diffusivities are then derived from the eigenvalues $$$\lambda_{e,1}$$$, $$$\lambda_{e,2}$$$, and $$$\lambda_{e,3}$$$ of $$$\bf D_e$$$ as $$$D_{e,\parallel}=\lambda_{e,1}$$$ and $$$D_{e,\perp}=\frac{\lambda_{e,2}+\lambda_{e,3}}{2}$$$. The intra-axonal diffusivity $$$D_a$$$ is calculated as $$$D_a=tr(\bf D_a$$$$$$)$$$.
Materials: A formalin fixed single hemisphere brain of a 62-year-old male patient with schizophrenia was obtained from China Brain Bank (Zhejiang University School of Medicine).
MRI: dMRI data was acquired on a 3T MAGNETON Prisma (Siemens Healthcare, Erlangen, Germany), b values of 1000, 2000, 4000, 6000, 8000, and 10000 $$$\tt {s/mm^2}$$$ over 30 gradient directions, with 18 non-diffusion-weighted data set acquired after every 10 different-weighted data sets, and an isotropic resolution of 1.8 mm. DKI-derived mean diffusivity (MD) and mean kurtosis (MK) were calculated using the DKE software while WMTI-derived metrics were calculated in MATLAB.
Histology: After MRI, the brain was cut into 5 mm thick coronal slices using a 5 mm deep cutting panel from midpoint of the third ventricle to the anterior shown in Figure. 1(a). Then the tissue blocks were dissected from corpus callosum and cut into 6 µm thick sections for immunohistochemistry. The tissue sections were stained with antibodies against myelin proteo-lipid protein (PLP; AT11024; SIGMA; 1:200), and glial fibrillary acidic protein (GFAP; AB5804; MILLIPORE; 1:1000) for astrocytes. Also Luxol fast blue (LFB) stains were used for myelin. Digitization of the stained sections was performed on an OLYMPUS VS120 (Japan) using x10 magnifying objective, leading to a resolution of 0.69 µm/pix. Myelin fraction and astrocyte fraction in corpus callosum were measured in PLP-stained/LFB stained and GFAP-stained sections using ImageJ. Staining signal above threshold levels were compared between the ROI on each slice of the brain.
Statistical analysis: For each slice, correlations between MR measures (MD, MK, etc) and histological measures (myelin and astrocyte area fraction) were evaluated using one tailed Spearman's correlation coefficient implemented in SPSS.