Diffusion properties of human brain cortex in-vivo. Part 2
We investigated the cortical depth dependence of diffusion properties with high resolution over the whole human brain in vivo establishing a first reference for future cortical diffusion tensor imaging studies. All brain areas showed similar cortical depth dependence of diffusion invariants with gradually decreasing and increasing patterns of fractional anisotropy (FA) and mean diffusivity (MD) from white matter to cerebrospinal fluid. At the white matter – cortex interface a drop in FA and a peak in MD is observed. The structure of these profiles is likely reflecting fiber-bending effects at the white matter – cortex interface and fiber density reduction at the cortex – cerebrospinal fluid border.
Magnetic resonance diffusion weighted imaging (MR DWI) of the human brain has mainly focused on white matter due to its high structural anisotropy and limited resolution of the measurement. The ability to determine cortical diffusion properties promises new in vivo insights into the brain’s cytoarchitecture, which are not only of neuroscientific interest but also may lead to new biomarkers for various brain disorders. Since the cortical thickness is only a few millimetres, DWI is challenging and most studies of the cortex were carried out in animal brain [1, 2] and ex vivo .
High resolution diffusion tensor imaging (DTI) has shown to allow segmentation of the human cortex in vivo based on fractional anisotropy (FA), and tensor directionality [4, 5]. Multishot DTI data were acquired in ten contiguous axial slices at the top of the brain. In particular, FA demonstrated diffusion anisotropy in the cortical grey matter, whereas most of the cortical grey matter regions showed primarily radial diffusion orientation (i.e., orthogonal to the cortical surface). In addition, FA depended on the cortical depth characterized by a band of low FA in the deep cortical lamella adjacent to the grey-white matter interface and a band of higher FA in the middle cortical lamella. The cortical profiles of the axial and radial diffusivity did not show any notable local maximum or minimum, whereas the cortical profiles of the radiality, defined as the scalar product between the major diffusion eigenvector and the cortical surface normal, showed a maximum in the middle cortical lamella.
Somewhat contradicting findings to the study of Truong et al. [4, 5] were reported by McNab et al. , who showed that the main diffusion direction in the cortex is oriented either normally or tangentially to the surfaces of cortical folds. They reported a sharp transition from radial to tangential diffusion orientation at the border between primary motor and somatosensory cortex in human in vivo data with isotropic resolution and partial brain coverage. They also showed evidence of tangential diffusion in secondary somatosensory cortex and primary auditory cortex.
One possible way to improve imaging veracity and increase the spatial resolution without sacrificing signal to noise ratio (SNR) is to move to high magnetic field strength. However, the method of choice for diffusion imaging – echo planar imaging (EPI) – faces several challenges related to the use of high resolution and high field strength, for example distortions and image blurring. To address such challenges, Heidemann et al.  introduced readout-segmented EPI in conjunction with parallel imaging, which significantly reduced artefacts by decreasing the effective echo spacing. With high resolution the authors identified radial anisotropy in the cortex.
Although this had been observed in previous studies, such a uniform anisotropy in a large region of cortex had not been described. Later the same author introduced an adapted EPI sequence with a combination of zoomed imaging and partially parallel acquisition (ZOOPPA)  which also improved the diffusion image quality of high resolution single-shot EPI with partial brain coverage. The method consistently resolved fiber orientation as well as layered structure in the cortex at submillimetre resolution. Recently Vu et al.  used multi-band acceleration to acquire diffusion data covering the whole brain in vivo. The authors recognized dark bands of FA predominantly in the deepest layers of grey matter. The dark bands of FA were reported to be strongest along sulcal banks and weakest in gyral crowns.
The goal of the current work is to resolve complex fiber topology at the interface between white and grey matter and microstructure within the cortex. The DTI data were analysed to depict and quantify microstructure in the entire human brain cortex and may serve as a reference atlas for future cortical DTI studies.
MATERIAL AND RESEARCH METHODS
In vivo human MRI data were downloaded from open database db.humanconnectome.org. Images were acquired on a 3T Siemens scanner (Siemens, Erlangen, Germany) with a 32-channel RX head coil. Twelve slabs of brain DWIs were acquired in 12 noncollinear and noncoplanar directions (b = 1000 s/mm2) and interleaved three nonDWI (b = 0 s/mm2) using the single-refocused Stejskal-Tanner spin-echo sequence [10, 11]. For excitation of 3 slices simultaneously multi-band radio frequency (MB-RF) pulse was used. Additionally, a set of reversed phase-encoded gradient b = 0 s/mm2 images were acquired for correction of geometrical distortions. Diffusion gradient was characterized with diffusion time Δ = 22 ms, diffusion gradient duration δ = 6 ms. No cardiac gating was used and the total scan time for all measurements was about 55 minutes including MP-RAGE and 3D-GE acquisitions for anatomical imaging [12–14].
Data processing and analysis
Initially, after brain cortex was segmented and parcellated from the anatomical images, cortical lamellae were generated [15, 16]. Quantitative DWI properties were determined in the coregistered anatomical regions. Further details are described in the following.
Processing of anatomical images
After B1 bias field correction of the MP-RAGE volume and brain masking [13, 14], segmentation of grey matter cortex (GM CTX), white matter (WM) and cerebrospinal fluid (CSF) was executed. According to the Desikan-Killiany and Destrieux atlases [17, 18], the cortical ribbon was parcellated and local cortical thickness and curvature were calculated. These processes were done by FreeSurfer  without any down-sampling.
It is difficult to fully appreciate the cortical diffusion tensor properties relative to the notably curved geometry of the cortex using two-dimensional orthogonal or slice representations. Therefore, the data were analysed according to their cortical depth. A surface expansion method was applied to create surfaces between the WM-CTX and CTX-CSF interfaces (defined as cortical depth of 0% and 100%, respectively) with cortical depth increments of 25% together with two additional parallel surfaces in WM (–25%) and in CSF (125%).
Processing of DWIs
DWI data corrected for susceptibility- and eddy-currents-induced geometric distortions and T2* blurring were used and each volume was registered to the anatomical image before combination and DTI metrics such as fractional anisotropy (FA), mean diffusivity (MD) and principal DT eigenvectors were derived [19, 20]. All image registrations and DT calculations were performed with the linear image registration tool (FLIRT) (Jenkinson, Smith, 2001) and FDTI in FSL. The local cortex curvature was determined by differentiation of the WM-CTX surface. Straight cortex areas were defined as banks (curvature within [–0.15, 0.15] 1/mm2), while gyri and sulci were characterized by larger curvature.
Desikan-Killiany (Fig.1) and Destrieux (Fig.2) atlases were defined by ‘aparc.annot’ available in FreeSurfer. For these atlases averaged invariants (FA, MD) and cortical thickness and their standard deviations for one subject’s brain were calculated (Fig.3).
Fig.4 demonstrates the cortical depth dependence of FA and MD from WM to CSF at each Desikan-Killiany atlas regions (left and right hemispheres combined). All areas present a similar FA profile, as shown in Fig.4a. The highest (0.38±0.11) and lowest (0.11±0.10) values FA were measured in WM and CSF, respectively and a pronounced drop of FA value (0.15 ±0.15) was observed at the WM-CTX interface. However, the values were higher than those in CSF. In MD, a gradual raise of up to 0.92 [µm2/ms] was observed from CTX to CSF (Fig.4b). In contrast to FA (Fig.4a), a significant increase of MD was seen at the WM-CTX interface and the pattern was very similar over the entire cortex (Fig.4b). The MD value of 0.87±0.36 [µm2/ms] at the WM-CTX interface is slightly smaller than in CSF 0.92±0.37 [µm2/ms]. The same analysis was performed for Destrieux atlas (Fig.5) which qualitatively similar to previous.
The spatial distribution of FA (>0.45) overlaid onto the inflated brain surface (Fig.6a) indicates that high FA regions, in particular at the WM-CTX interface are localized primarily in sulci.
The spatial distribution of MD (>1[µm2/ms]) mapped onto the inflated brain surface (Fig.6b) demonstrates that the highest MD, particularly in the middle lamellae of the cortex are localized mainly in gyri.
DISCUSSION AND CONCLUSION
In this study we have analysed full brain diffusion tensor data. The DTI EPI data are corrected for geometric distortions caused by diffusion direction dependent eddy currents and local susceptibility variations. This allowed precise matching of the very thin cortical lamellae and the diffusion invariants, i.e. FA, MD and diffusion eigenvectors.
Our study has demonstrated substantial sensitivity of MRI to depict patterns of tissue cytoarchitecture. This is shown by lamellae analysis of DT measures and investigation of local directionality of major eigenvectors. Although this was observed in previous studies to different extent (i.e., not for the whole brain, or not with uniform voxel size), our results verify depth dependent cortical profiles of DT measures in all areas of the brain proving to be a robust marker of cortical microarchitecture and establishing a first reference of cortical DTI. The obtained quantitative values (Fig.3) are in accordance with the range reported for cortical thickness [21, 22] and for DTI measures .
With the high resolution and geometric fidelity of the data a detailed analysis of invariant diffusion properties and the relation between local structural orientation and the main diffusion direction is feasible, in particular in the thin cortical ribbon, extending diffusion tensor imaging from a white matter modality into the cortex. The analysis showed that diffusion properties are very similar throughout the entire human cortex. In particular, the cortical depth dependence of FA and MD with a gradual reduction / increase from white matter to CSF and a pronounced drop / peak at the WM-CTX interface seem to be ubiquitous features throughout the brain.
The consistency of the results leads us to hypothesise that the WM-CTX interface behaviour is based on the bending of fibres that enter the cortex. In WM FA is highest due to densely packed fibres characterized by well-expressed anisotropy (Fig.4, 5).
At the WM-CTX interface FA shows a strong decrease together with a local rise in MD that may reflect fibres that bend sharply into the cortex . Thus "pseudoisotropy" appears because different fibre orientations exist in these voxels. The return of FA to higher and MD to lower values within the cortex reflects the preferential radial arrangement of neuronal connections in cortex, which, however, is lower than in WM due to tangential within-cortex connections. We have also shown that this effect may be stronger in straight segments (banks) of the cortex where the fibres seem to turn by approximately 90° compared to the tips of sulci/gyri where they enter cortex in a straighter manner. In parcellated cortex regions this is obviously averaged but nevertheless the "pseudoisotropy" at the WM-CTX interface remains clearly visible.
Interestingly a similar lamella-based analysis performed by McNab et al.  did not show any noticeable local variation of FA and MD at the WM-CTX interface. The authors reported a gradual decrease of FA and increase of MD between WM-CTX and CTX-CSF interfaces. This may be due to lower sensitivity at a lower main magnetic field and approximately double the voxel volume. Partial volume effects in diffusion images with non-isotropic voxel size at lower main field may also be the cause for the difference in radial and axial diffusivity demonstrated by Truong et al. [4, 5]. Although FA profiles are very similar with those we obtained, the drop of FA was significantly shifted in cortical depth and is interpreted as turning fibers in areas away from the WM-CTX interface.
Our results derived from DTI are supported by histologic brain anatomy studies . Cell stained images of excised tissue show how the neuronal dendrites extend radially toward deep cortical layers where they spread out tangentially along the cortical surface. The highest bifurcation can be observed in the superficial CTX lamina, showing a very strong branching. The adjacent external and deep lamellae show mainly radial orientation due to fibers that are extending from the WM radially into the cortex. In the adjacent WM, the fibers run mainly parallel to the CTX-WM boundary.
In conclusion this study shows that quantitative DT measures can reveal patterns of architectural organization of the human cerebral cortex in vivo. The local depth dependent results qualitatively agree with histologic brain anatomy. High resolution cortical DTI may allow characterization of cytoarchitecture and improved understanding of the pathophysiology of diseases associated with changes in cerebral cortex. This can extend MR diffusion tensor imaging from a white matter modality to a cortex modality opening new clinical and research applications. ■
Declaration of Competing Interest. The author declares that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.