MRI of post-mortem tissue is gaining attention because of achievable high spatial resolution and subsequent tractography results^{1, 2}. Here, we demonstrate validation of diffusion tractography from a whole, post-mortem human brain acquired at 7T against polarized light imaging (PLI) data in the same brain. PLI is an optical imaging method that can provide high-resolution estimates of white matter fibre orientation in mounted tissue slices based on the birefringence of myelin.Data were acquired of a post-mortem human brain with no known neuropathology, a post-mortem interval <72 hours, and scan interval of 6 weeks. The brain was submerged in fluorinert for tissue susceptibility matching and imaged using a Siemens 7T whole body scanner with a 32 channel receive head coil. Diffusion data were acquired with DW-SSFP at 0.5mm, 1mm, and 2mm isotropic resolutions covering 90 directions. The protocol was repeated using two different transmit voltages (corresponding to 33˚ or 98˚ flip angle at the centre of the brain), chosen to improve uniformity of CNR across the entire brain despite B1 inhomogeneity³. DW-SSFP has a well defined q-value, but not b-value. Here, we used q=300 cm^-1 to achieve an effective diffusion contrast of b_eff=5150 s/mm² (i.e. the signal attenuation was equivalent to this b-values) in white matter. Each 90 direction dataset required 5 days of continuous scanning. A version of the FSL tool DTIFIT modified to include the DW-SSFP signal model⁴ was used to estimate the principal diffusion directions and other tensor parameters. Deterministic tractography was performed with Trackvis to visualize local fibre bundles. After MRI scanning, the same brain was sectioned into blocks, selecting three characteristic regions for inspection: the corpus callosum (a well ordered white matter pathway), a gyrus in V1 (fanning architecture where fibres enter gray matter) and the pons (highly interdigitated pathways). For PLI, a Leica DM4000 microscope was upgraded with a linear polarizer, quarter wave plate and rotatable analyzer. Images at 4x4x60 μm resolution were acquired at 18 different polarization orientations (10˚ increments from 0-170˚). PLI-based fiber orientation maps were calculated using custom MATLAB code⁵.

Figure 1 demonstrates fibre anatomy across the corpus callosum at high resolution. Tractography results show commissural fibres predominantly running left/right; however, the traced tracts also intertwine or braid with one another, particularly at the mid-line. PLI data in the same region presents a striking similarity, showing a disordered appearance at the mid-line, with striated appearance on the lateral aspects.

Figure 2 demonstrates the orientation estimate at the three different MRI resolutions in a single gyrus. At lower resolutions (Figs 2a,b), orientation estimates in gyri suffer from a bias due to fiber bending within a voxel, with the net effect being a subsequent bias in connectivity to the gyral crown. The presence of orientation estimates (Fig 2c) reflecting the fanning of fiber bundles as they intersect the length of the grey/white matter interface increases with resolution. PLI results in V1 (Fig 3b) also depict the fanning anatomy of white matter tracts along the grey/white matter interface. Tractography performed on this dataset (Fig 3a) resolves this anatomy, as well.

Results from the pons are shown in Figure 4. At lower resolutions (Figs 4a,b), the larger relative fraction of corticospinal tract (CST) compared with transverse pontine fibre results in an estimate of the orientation that favors the CST. PLI through the same region of the pons (Fig 5a) depicts the CST and pontine fibres forming interdigitating bundles at a spatial scale approaching that of the high-resolution diffusion data. These bundles are visible in both the tensor-based orientation estimates and streamline tractography (Fig 5b). These results suggest that the improved resolution either resolves these pontine bundles, or at least results in voxel neighborhoods that are predominated by them.

This work compares very high-resolution diffusion MRI data with PLI from the same brain. A number of features seen in the diffusion tractography were corroborated by PLI data. Sutble twisting in the corpus callosum was indicated by both modalities; if not captured accurately, this could lead to incorrect cortical-cortical connectivity estimates. At conventional spatial resolution, diffusion tractography of fibres entering a gyrus often fail to correctly identifying sharp bends into the gyral banks⁶. Our data suggest this bias could be overcome with improved spatial resolution or more sophisticated models, with data of this kind being invaluable for model validation. Finally, measurements in the pons demonstrate the exquisitely complicated fibre architecture that remains difficult or impossible to identify with existing methods, but which may be resolvable with the next major leap forward in spatial resolution.