0736

Post-mortem mapping of cortical layers using combined multicompartmental relaxometry and diffusometry at ultra-high field (7T and 11.7T)
Justine Beaujoin1,2,3, Christophe Destrieux4, Fabrice Poupon5, Ilyess ZEMMOURA4, Jean-François Mangin2,3,5,6, and Cyril Poupon1,2,3

1UNIRS, CEA/DRF/Neurospin, Gif-sur-Yvette, France, 2Université Paris-Saclay, Orsay, France, 3FLI / Noeud Paris-Sud, Orsay, France, 4Université François-Rabelais de Tours, INSERM, Imagerie et Cerveau UMR 930, Tours, France, 5UNATI, CEA/DRF/Neurospin, Gif-sur-Yvette, France, 6http://cati- neuroimaging.com/, Gif-sur-Yvette, France

Synopsis

The human cerebral cortex has a laminar structure presenting distinct cell organization and myelination. In this work, we compare the lamination observed from multicompartmental quantitative relaxometry- or diffusometry-based microstructural ultra-high field MRI on a post-mortem human brain visual cortex sample. This study reveals that the quantitative intra-neuritic volume fraction map provides a better delineation of the line of Gennari in the primary visual cortex whereas multicompartment T1 relaxometry provides a better segmentation of the superfical layers. Thus, this study strongly plays in favour of a combination of relaxometry and diffusometry methods to robustly map the cortex lamination in humans.

Introduction

The human cerebral cortex has a laminar structure presenting distinct cell organization and myelination. Cortical thickness has been shown to be associated with neurodegenerative diseases such as Parkinson's or Alzheimer's1,2, but pattern of cortical damage at the laminar level remains unclear. There is therefore a growing interest in imaging the brain at a mesoscopic scale to better map its cortical layers. Complementary to the use of ultra high-field (UHF) MRI to reach this scale, quantitative relaxometry and diffusometry MRI methods are also a promising way to enhance the resolution through the use of multicompartmental models that proved their effectiveness to discriminate cortical layers3,4. In this work, we go further and compare the delineations obtained using multicompartmental diffusometry and relaxometry MRI methods at the mesoscopic scale.

Material and methods

Acquisition- A formalin-fixed post-mortem human brain sample containing primary and secondary visual cortex was soaked in a phosphate-buffered saline solution for rehydratation before being scanned with a preclinical 11.7T (for anatomical/diffusion MRI) and 7T (for relaxation) Bruker MRI systems. The binding of the sample within its container was guaranteed by an electrical insulating gel (Figure1). A dedicated MRI protocol was tuned with the following parameters: anatomical scan: isotropic 150μm resolution, TE/TR=12/6000ms; diffusion-weighted (DW) scans using Pulsed Gradient Spina Echo (PGSE) sequences: isotropic 300μm resolution, b=2000/4000/6000s/mm2 , 125/125/125directions, TE/TR=30.6/4500ms, δ=5.5ms, Δ=20ms; T1-mapping scans using 3D inversion-recovery (IR) rapid acquisition with relaxation enhancement (RARE) sequences: isotropic 300μm resolution, TE/TR=5.6/3000ms and 122 inversion times spread between 100 and 2900ms.

T1-mapping- T1 analysis was done on a voxel-by-voxel basis by fitting the entire IR dataset to an exponential decay using Powell's NEWUOA5 algorithm.

Microstructure mapping- The 3-shell DW acquisitions were used to infer the neurite density using the NODDI6 model. Fixing the local principal direction estimated from the jointly estimated DTI model and the diffusivities to d// =1.5.10 -10m2 /s and diso nul, it provided quantitative maps of the intracellular volume fraction representing the neurite density (fintra ). fintra maps were then deblurred with a non-local means agorithm7,8.

Clustering- A Gaussian mixture model was used to fit the whole sample T1 and fintra distributions using a Bayesian Index Criterion (BIC) to define the optimum number of components. Labels were then associated to each voxel, based on the minimal distance to the resulting clusters. Labels corresponding to white matter (WM) were merged to enhance the visualization of cortical layers.

Results and discussion

Figure2 depicts the anatomy, fintra and T1 quantitative maps. The anatomical scan, which perfectly visualized the primary visual cortex, was used as a reference. The fintra map, more sensitive to high neurite density than T1-relaxometry, finely discriminated the line of Gennari, which appeared as an hypointense line paralell to the cortex surface.

Gaussian mixture fits are shown in Figure3. Seven clusters were found from the analysis of the T1 mapping whereas nine clusters were found from the analysis of the intracellular volume fraction. Low T1 values were associated to the WM, as well as high fintra values (deep blue labels). WM presented distinct modes corresponding to various fiber configurations (crossings, various fiber densities). Gray cortical matter was represented by five labels in both approaches, allowing to delineate cortical layers, as depicted in Figure4. The outer layers (I, II and III) were associated to a single mode of the intracellular volume fraction map (deep red colour), whereas two labels have been identified from the T1 custering approach (deep red and green colours). These layers are characterized by low neurite densities and mainly correspond to two types of cells (granular in layer II and pyramidal in layer III), thus yielding different T1 values. This result is in agreement with Kleinnijenhui4 showing a poor intracellular volume fraction in the superficial layers. Stripes of Baillarger, in layers IV and V, have a high neurite density, explaining the increased intracellular volume fractions. The deep and light blue zones respectively correspond to voxels containing only WM and voxels containing both WM and gray matter (GM). White and pink arrows highlighted regions where the fintra-based clustering seems to provide patterns of cortical layers more consistent with the anatomical scan than T1-based clustering. Green line folllows the line of Gennari (white arrow) and the limit between WM and GM is well established (pink arrow).

Conclusion

In this work, we demonstrated that the association of UHF-MRI relaxometry and diffusion MR microscopy allows to robustly detect the cortical layers in a complementary way exploiting the magnetic and diffusive properties of their cellular environment. Clustering based on the neurite density map accurately delineates the stria of Gennari, but displays layers I, II and III as a whole since they have a poor neurite density, whereas clustering based on T1-relaxometry delineates boundaries in superficial layers.

Acknowledgements

This work was partially funded by the European FET Flagship ‘Human Brain Project’ (SP2) FP7-ICT-2013-FET-F/604102.

The authors acknowledge the donor who gave his body to science for this study.

References

1. Fjell, Anders M., et al. "What is normal in normal aging? Effects of aging, amyloid and Alzheimer's disease on the cerebral cortex and the hippocampus." Progress in neurobiology 117 (2014): 20-40.

2. Jubault, Thomas, et al. "Patterns of cortical thickness and surface area in early Parkinson's disease." Neuroimage 55.2 (2011): 462-467.

3. Lifshits, Shlomi, et al. "Resolution considerations in imaging of the cortical layers." NeuroImage (2017).

4. Kleinnijenhui et al, "Detailed laminar characteristics of the human neocortex revealed by NODDI." Proc.ISMRM (Abstract #0837), 2013.

5. Powell, Michael JD. "The NEWUOA software for unconstrained optimization without derivatives." Large-scale nonlinear optimization. Springer US, 2006. 255-297.

6. Zhang, Hui, et al. "NODDI: practical in vivo neurite orientation dispersion and density imaging of the human brain." Neuroimage61.4 (2012): 1000-1016.

7. Buades, Antoni, Bartomeu Coll, and Jean-Michel Morel. "Non-local means denoising." Image Processing On Line 1 (2011): 208-212.

8. Condat, Laurent. "A simple trick to speed up and improve the non-local means." (2010).


Figures

Figure 1. Sample and container. The blue square indicates the scanner field of view, at the level of the occipital lobe.

Figure 2. (a.) : anatomical scan at 150µm. Red arrows show the borders of the primary visual cortex (V1) which contains the stria of Gennari. (b.) : T1 quantitative map at 300µm. The green arrow highlights a slightly darker line with lower T1 : the line of Gennari. (c.) and (d.) : intracellular volume fraction maps at 300 µm from NODDI model before (c.) and after (d.) deblurring using non-local means algorithm. Yellow arrows point at a light line with higher neurite density : the line of Gennari, easier to discriminate than on the T1 map.

Figure 3. T1 (a.) and fintra (b.) histograms of the whole sample, revealing 7 clusters for T1 and 9 for fintra. Each cluster is associated with a colored label, which remains the same in the Figure 4.

Figure 4. Layers revealed by clusters obtained with the fintra map and the T1 map.The outer layers were associated to a single mode in the fintra map (deep red colour), whereas two labels were identified from the T1 clustering (deep red and green colours).Stripes of Baillarger, in layers IV and V, have a high neurite density, explaining the increased intracellular volume fractions.White and pink arrows highlighted regions where the fintra-based clustering provides patterns of cortical layers more consistent with the anatomical scan than T1-based clustering.Green line folllows the line of Gennari (white arrow) and the limit between WM and GM is well established (pink arrow).

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)
0736