While methods for estimating neuronal fiber orientation distributions (FODs) in dMRI are becoming increasingly popular, the correspondence between FOD measures and realistic biological fiber architectures has been unclear. Using detailed manual recordings of individual myelin stained fiber orientations in ex vivo rat brain tissue we have shown that tomographic dMRI FOD estimates provide accurate assays of the underlying myeloarchitecture, even in regions with complex multi-directional crossing fiber architectures.
In this study, FODs were quantitatively validated using a voxel-wise approach against empirical FOD estimates derived from registered myelin stained images. As evident in , QSI-FODs provided an accurate characterization of the underlying myelinated fiber orientation distribution in regions of both uniform (r>0.9) and crossing fiber (r>0.8) architectures. A subsequent evaluation of the angular error of the QSI-FOD peaks in ROI-2 demonstrated an average angular error of approximately 5–6°, with an average (acute) fiber crossing angle of approximately 73° (). Because it seems reasonable to assume that some of this error is likely due to image registration and stereological errors (e.g., through-plane fiber contributions and sampling limitations), the actual intrinsic angular error of the QSI-FOD peaks is probably even less than 5–6°.
It should be noted that the FODs in this paper were computed from QSI data with high b-values and a large number of diffusion measurements (b-max
, 515 diffusion measurements, see “Materials and Methods
”). Thus, it remains to be determined to what extent similarly accurate assays of myeloarchitecture can be derived using more conservative spherical acquisition protocols as often employed on clinical 1.5T and 3.0T scanners for in vivo
applications. However, numerous simulation studies 
have demonstrated that FODs (with the ability to resolve crossing fibers down to about 45°) can be derived from spherical acquisition protocols using moderate b-values (b~2000–4000 s/mm2
) and reasonable scan times (SNR~30, scan time<10 min). These simulation studies suggest that accurate FODs can be achieved in regions with complex architectures with use of reduced encoding spherical acquisition protocols. Hence, the validation benchmarks established in this paper will likely also have a high degree of translational value for the clinical situation.
The histological FODs were based on manually traced myelin fibers sampled from high-resolution digital images of 50 µm thick sections using a systematic random approach. Sampling occurred both within (using sampling bins) and across planes (one focal plane in two sections spaced at 200 µm) to ensure that a representative fraction of the myeloarchitecture was recorded per voxel volume. It should also be noted that water diffusion is influenced by the complete tissue microarchitecture and not only myelinated fibers (for review, see Ref 
), which will bias the dMRI measurements relative to the histological measurements. However, the high correlations measured suggest that the contribution of other tissue elements is relatively small in the regions investigated.
The QSI-FODs showed a remarkable consistency across animal specimens (). We thus chose to restrict the quantitative validation to extensive, in depth anatomical analyses of fiber architectures in one specimen. In these analyses, FOD estimates within each voxel were derived from several hundred manually traced histological measurements across multiple coronal sections and several hundred QSI diffusion measurements. Thus, although the total number of voxels compared was relatively small (4 for ROI-1 and 12 for ROI-2), the statistical correlations were based on measurements derived from extremely high dimensional datasets. It should further be noted that the statistical correlations were only used to provide a quantitative metric of similarity at each voxel, and not as a statistical test of consistency or generalizability across animal specimens.
The voxel-wise histological validations were conducted in paraformaldehyde fixed tissue that had been immersed in contrast enhancing Magnevist® liquid (see “Materials and Methods
”). Because this treatment is known to reduce the ADC 
, some care should be exercised when extrapolating these results to the in vivo
case. However, it has been shown that fixation has relatively small effects on the overall amount of diffusion anisotropy, as the ADC is reduced equally in all directions 
. Therefore, the fixation process itself is not likely to have influenced the general orientation structure of the QSI-FOD measurements in this study.
We conclude that fiber orientation distributions derived from high dimensional diffusion MRI data provide accurate assays of the underlying myeloarchitecture, even in regions with complex crossing fiber architectures. These results have important implications for both clinical and research studies investigating structural aspects of biological tissues using estimates of the fiber orientation distribution. Furthermore, this study provides an important biological benchmark for further improvement of fiber orientation reconstruction and tracking methods.