Histological validation of high-resolution DTI in human post mortem tissue

Abstract

Diffusion tensor imaging (DTI) is amongst the simplest mathematical models available for diffusion magnetic resonance imaging, yet still by far the most used one. Despite the success of DTI as an imaging tool for white matter fibers, its anatomical underpinnings on a microstructural basis remain unclear. In this study, we used 65 myelin-stained sections of human premotor cortex to validate modeled fiber orientations and oft used microstructure-sensitive scalar measures of DTI on the level of individual voxels. We performed this validation on high spatial resolution diffusion MRI acquisitions investigating both white and gray matter. We found a very good agreement between DTI and myelin orientations with the majority of voxels showing angular differences less than 10°. The agreement was strongest in white matter, particularly in unidirectional fiber pathways. In gray matter, the agreement was good in the deeper layers highlighting radial fiber directions even at lower fractional anisotropy (FA) compared to white matter. This result has potentially important implications for tractography algorithms applied to high resolution diffusion MRI data if the aim is to move across the gray/white matter boundary. We found strong relationships between myelin microstructure and DTI-based microstructure-sensitive measures. High FA values were linked to high myelin density and a sharply tuned histological orientation profile. Conversely, high values of mean diffusivity (MD) were linked to bimodal or diffuse orientation distributions and low myelin density. At high spatial resolution, DTI-based measures can be highly sensitive to white and gray matter microstructure despite being relatively unspecific to concrete microarchitectural aspects.

Keywords: diffusion microstructure; diffusion tensor imaging; fiber orientations; gray matter; histological validation.

Figures

Figure 1

Histological fiber orientation distribution with fitted von Mises mixture for two exemplary voxels. Left: Normalized histograms of structure tensor orientations of all pixels within a voxel (gray line), smoothed with a Gaussian window (blue line) and fitted using a mixture of three von Mises probability density functions (red line). The central values of the von Mises components were the resulting histological orientations in a voxel (circles). Middle: Same histograms as polar plots. Right: High amplitude voxel orientations (corresponding to red circles) visualized on corresponding histological tile. (A) Example of 1-orientation voxel, (B) 2-orientation voxel.

Figure 2

Histogram of angular differences between MRIDT and HistST orientations over all analyzed voxels for white and gray matter. Dashed line: Histogram fit with generalized Pareto distribution.

Figure 3

Exemplary histological section with MRIDT and HistST orientations. Diffusion tensors are coded by oriented green rectangles, where rectangle aspect ratio indicates fractional anisotropy (more elongated = higher FA). HistST orientations are coded by bars. The color of the bars indicates the respective orientation (lateral-medial = red; inferior-superior = blue). Size of DT rectangles and ST bars is proportional to DT projection length. Scale bar: 3 mm. (A) Voxels in which myelin orientations are parallel to the cortical surface, DT orientations orthogonal. (B) Fiber crossing. FA is low and the DT orientation is in between the primary and secondary histological orientation. (C) 1-orientation fiber pathway. FA is high, the secondary histological orientations are very small, and there is a good match of primary histological and DT orientation. (D) Area (a) without DT and ST information. Arrows indicate the tangential fibers determining the ST orientations. (E) Classification of cortical layers in an area corresponding to area (a) in a neighboring Nissl-stained section.

Figure 4

MRIDT and HistST orientations in one exemplary section. (A) Diffusion tensor orientations projected into the sectioning plane. Orientation angle is coded as hue (with the full color spectrum to optimize orientation contrast), projection length as brightness in HSB space. (B) Primary structure tensor orientations, same color code as (A). (C) Mapping of angular difference between (A,B) in axis angles (ranging from 0 to 90°). (D) Mapping of projection length of diffusion tensors into the sectioning plane (ranging from 0 to 1). Note that a short projection length is a possible but not the only source of high angular difference in (C). (E) Original section with stained area emphasized. In (A–D), the parts without stained fiber material are displayed in gray. Size (A–D), 1 image pixel equals 1 voxel (340 μm); (E), scale bar = 5 mm.

Figure 5

Average difference between main MRIDT and HistST orientations in gray (dark line) and white (light line) matter as a function of FA. Marker size linearly represents the sizes of the voxel subsamples of the FA bins. FA bins with less than 100 voxels are not displayed. Confidence interval: 1.96* standard error.

Figure 6

Differences between MRIDT and HistST orientations displayed on an FA map. MRIDT and HistST orientations are plotted in a color range from green (for orientations that perfectly agree) to red (for orientations that differ by 90°). Bar length reflects the projection length of the diffusion tensor in the sectioning plane. There is a generally good histology-DTI agreement for high FA values, especially in white matter. For lower FA, particularly when entering gray matter, there are three cases: (A) Good agreement in case of predominant radial orientations, (B) Moderate to weak agreement, (C) Perpendicular histology-DT orientedness. Here, voxels are located well within gray matter and run in bands parallel to the cortical surface. Scale bar: 5 mm.