Diffusion tensor imaging (DTI) is a widely used tool to investigate microstructural properties of biological tissue in vivo. However, the contribution of different tissue microstructural components to DTI contrast is poorly understood. Mainly, the approaches to relate DTI to real tissue microstructure have used two-dimensional (2D) histological sections, 3D histological analyses appearing only very recently¹. In this work, we developed a 3D Fourier analysis to quantify anisotropy and orientation in a 3D stack of serial block-face scanning electron microscopy (SBEM) images². SBEM produces a 3D visualization of all diffusion barriers in the tissue in high resolution. Our aim was to analyze 3D tissue microstructure in SBEM, and compare its contribution to DTI contrast in normal and traumatic brain injury (TBI) rat brains.

Severe lateral fluid percussion injury was performed on male adult Sprague-Dawley rats (n=2). One sham operated rat (n=1) was used as a control. TBI model was used in order to create robust microstructural changes in brain tissue.

DTI was acquired in vivo with 7 T Bruker scanner using segmented spin-echo EPI (TR=2.5 s, TE=30 ms, matrix: [96 64], 14 slices, resolution: 0.22x0.22x0.5 mm, b-value=3000 s/mm²) six months after injury (Fig. 1A). After the scans, the animals were perfused and scanned ex vivo with 9.4 T Agilent scanner using segmented spin-echo EPI (TR=1 s, TE=35 ms, matrix = [128 96 96], resolution: 0.15x0.15x0.15 mm, b-value=2000 s/mm²) (Fig. 1B).

After ex vivo DTI, brains were sectioned and processed for histology and EM. SBEM imaging was done on samples taken from the corpus callosum and perilesional cortex both ipsilaterally and contralaterally to the lesion (Fig. 1C). The imaging was performed using 0.050x0.050x0.050 μm resolution (4096x2048x1000 voxels) in the corpus callosum, and 0.046x0.046x0.050 μm resolution in the perilesional cortex (2048x2048x1400 voxels). Additionally, cortical stacks and one of the stacks in the corpus callosum were also imaged using higher in-plane resolution of 0.014x0.014x0.050 μm or 0.018x0.018x0.050 μm (1024x1024x1400 voxels) (Fig. 2).

The 3D Fourier analysis is an extension of an existing 2D analysis, where Fourier transform of an image is used to obtain information from structure orientation in the image³. For that, we placed volumes-of-interest (VOIs) of 50x50x50 μm in the low magnification data sets, and 14x14x20 μm in the high magnification data sets. We computed anisotropy (AI) and orientation (orthogonal vectors V1, V2, and V3 comparable to DTI diffusivity orientation vectors) characteristics on each of the VOIs. Different SBEM stack orientations were taken into account and vectors rotated accordingly.

The localization of SBEM stacks on DTI images (Fig. 3A, 3B) was done using semi-thin histological sections and anatomical landmarks in them (Fig. 3C, 3D). A DTI voxel was chosen for each SBEM stack for comparison of DTI metrics with Fourier derived metrics from EM. AI matched with in vivo and ex vivo fractional anisotropy (FA) in cortical VOIs (Table 1). In the corpus callosum, AI was similar to in vivo FA, whereas ex vivo FA was much higher (Table 1). Orientations in SBEM stacks and their respective DTI voxels matched well, with angles between corresponding vectors in SBEM stacks and DTI ranging from 0.04 to 0.57 radians for ex vivo and 0.10 to 0.53 radians for in vivo (Table 1).

We also compared anisotropy values of high magnification stack to values in the corresponding low magnification stack in seven VOIs to investigate the effect of resolution. Change in in-plane resolution of SBEM stacks did not affect the AI or orientation, the values in high magnification stacks compared to low magnification stacks were very similar (Table 2).

We were able to demonstrate that 3D Fourier analysis is able to derive anisotropy and orientation values from 3D SBEM image stacks. We also showed that anisotropy and orientation of SBEM images reflects the DTI anisotropy and orientation. Furthermore, our data indicate that 3D Fourier analysis is very robust to changes in SBEM in-plane resolution.

Anisotropy calculated from SBEM might be affected by EM contrast, tissue alterations due to fixation and staining procedures, while FA from DTI might suffer from partial volume effects in vivo and ex vivo, as well as tissue fixation in ex vivo DTI. These possible variables affecting the results need further investigation.

In this study, we focused on the corpus callosum and cortex, however, we will perform 3D Fourier analysis in brain areas with different microstructural morphologies to interpret the DTI contrast in more detail in 3D. These results will pave way for more complete understanding of DTI data and the underlying microstructural tissue in normal brain and during pathology.