Purpose

The non-invasive mapping of neuronal connections in cortical gray matter is challenging with diffusion MRI because the cortex has relatively low diffusional anisotropy, presumably due to an abundance of intravoxel fiber crossings. Here we explored the ability of diffusion MRI to detect multiple fiber directions in mouse cortex by using diffusional kurtosis imaging (DKI). Robust detection of multiple intravoxel fiber directions would demonstrate the potential of DKI-based tractography of neuronal networks in mouse cortex.

Methods

Conventional DKI data from a 24 month old male C57bl/6 mouse were collected using a Bruker BioSpec 7T MRI system with 64 diffusion encoding directions and b = 0, 1000, 2000 s/mm². Coronal images were acquired with an in-plane resolution of 0.23 x 0.23 mm² and a slice thickness of 0.4 mm. A T2 TurboRARE dataset with the same slice thickness was obtained as an anatomical reference. Diffusion and kurtosis tensors were estimated using the Diffusional Kurtosis Estimator¹ (DKE, Medical University of South Carolina, Charleston, USA, http://academicdepartments.musc.edu/cbi/dki/dke.html). Cortical voxels were determined by thresholding FA<0.25 with subsequent semi-automatic removal of deep gray matter. To explore the diffusion patterns, we calculated, in each voxel, a kurtosis diffusion orientation distribution function² (dODF) using a radial weighting power of four. Gaussian dODFs, which encode the same information as the traditional diffusion ellipsoid, were also calculated from just the diffusion tensor (constructed with only b = 1000 s/mm²), to visualize the difference between resolving for single (Gaussian dODF) or multiple fibers (kurtosis dODF) in the cortex. Fiber directions were estimated from dODF maxima.

Results

To illustrate fiber crossing detection with DKI, a single coronal slice with superimposed kurtosis dODFs is depicted in Fig. 1. Throughout the mouse cortex, the kurtosis dODF consistently resolves multiple fiber directions. As a reference, also in Fig. 1 are the corresponding Gaussian dODFs, which reflect the ability of conventional diffusion tensor imaging to detect a single, dominant fiber direction but not fiber crossings. A histogram for the number of detected fiber directions in each voxel with the kurtosis dODF from both cortex and white matter is shown in Fig. 2. Whereas a fiber crossing was detected in about 50% of voxels in white matter, in cortical gray matter over 80% contained a fiber crossing. The average number of fiber directions per cortical voxel was 2.3. The distribution of angular differences calculated between the various fiber directions within a given cortical voxel are plotted in Fig. 3. The angular distribution for each pair of fiber directions is reasonably well approximated (average R² of 0.97) by a half-normal distribution with a peak at 90^o and standard deviations of 34^o, 30^o, and 31^o between directions primary-secondary, primary-tertiary, and secondary-tertiary, respectively, where the directions have been ordered according to the magnitude of the corresponding dODF maxima.

Discussion

In cortical gray matter, the kurtosis dODFs constructed using DKI detect multiple intravoxel fiber directions in the majority of voxels. On average, we detect more than two fiber directions per cortical voxel. White matter, in contrast, is found to consist half of voxels with only a single fiber direction (Fig. 2), which is similar to previously reported results for human brain². The distribution of angular differences between the fiber directions within individual voxels is peaked at approximately 90^o (Fig. 3), which is in rough accordance to the grid-like geometrical organization proposed by Wedeen, et al³. A similar fiber crossing angle distribution has previously been found with DKI for white matter in human brain². These observations, together with the qualitative results from Fig. 1, support the feasibility of performing fiber tractography in mouse cortex using DKI. Note that hippocampal gray matter tractography has already been demonstrated in mouse with the alternative approach of constrained spherical deconvolution⁴.

Acknowledgements

This work was supported in part by National Institutes of Health research grants T32DC0014435, T32GM008716, DC014021 and The Litwin foundation.