Ultra-high-resolution Mapping of Cortical Layers 3T-Guided 7T MRI

Abstract

7T MRI provides unprecedented resolution for examining human brain anatomy in vivo. For example, 7T MRI enables deep thickness measurement of laminar subdivisions in the right fusiform area. Existing laminar thickness measurement on 7T is labor intensive, and error prone since the visual inspection of the image is typically along one of the three orthogonal planes (axial, coronal, or sagittal view). To overcome this, we propose a new analytics tool that allows flexible quantification of cortical thickness on a 2D plane that is orthogonal to the cortical surface (beyond axial, coronal, and sagittal views) based on the 3D computational surface reconstruction. The proposed method further distinguishes high quality 2D planes and the low-quality ones by automatically inspecting the angles between the surface normals and slice direction. In our approach, we acquired a pair of 3T and 7T scans (same subject). We extracted the brain surfaces from the 3T scan using MaCRUISE and projected the surface to the 7T scan’s space. After computing the angles between the surface normals and axial direction vector, we found that 18.58% of surface points were angled at more than 80° with the axial direction vector and had 2D axial planes with visually distinguishable cortical layers. 15.12% of the surface points with normal vectors angled at 30° or lesser with the axial direction, had poor 2D axial slices for visual inspection of the cortical layers. This effort promises to dramatically extend the area of cortex that can be quantified with ultra-high resolution in-plane imaging methods.

1. INTRODUCTION

The laminar organization in the cerebral cortex has been a subject of interest in the research of structural and functional neuroanatomy¹⁶. Several studies have been conducted to characterize the laminar structure of the cerebral cortex and to study the relationship between laminar organization and brain function¹⁵. The different layers in the laminar structure can be identified by their cell density, variations in thickness and other histological features¹⁷. 7T MRI offers an excellent ultra-high-resolution, non-invasive approach to image the cortex laminar arrangements in vivo¹. Visualization of the cortical lamination patterns using 7T MRI has been instrumental for research in several sub fields of neuroscience including the pathological study of neurodegenerative diseases^2,3,4,5,6,8.

However, the arbitrary direction of 2D planes in 7T lead to new challenges since the 7T brain MRI scans are typically non-isotropic⁷. In other words, the scans obtained in the 7T MRI scanner usually have the highest resolution in the direction that is orthogonal to the slice direction. Herein. the proposed method compares the quality of retrieved ultra-high resolution 2D slices that are orthogonal to the cortical surface, yet at different surface locations, from 3D 7T MRI scans. Specifically, the method identifies slices where the slice direction is orthogonal to the surface of the cortex thus providing the best resolution for viewing them. MaCRUISE^9,10,11 is used to achieve the 3D cortical surface automatically. Then, we further identify the high quality and low-quality slices (due to the non-isotropic nature of brain MRI) by first computing normal vectors from the surface cells using VTK and computing the angle between the unit normal vectors and the slice direction. From our experiment, we found that 18.58% of surface points that were angled at more than 80° showed visually distinguishable cortical layers. 15.12% of the surface points were angled at 30° with the slice direction where the quality of the slice was poor for visual inspection of the cortical layers.

2. METHODS

2.1. Data

Two MRI images were obtained for a subject, one from 3T MRI scanner and the other from a 7T MRI scanner, as part of a research study by the Translational Neurosciences Core of Vanderbilt Kennedy Center at the Vanderbilt University Medical Center (VUMC) site. We used the T1 scan from the 3T session and a T2* scan from the session acquired with the 7T scanner for this experiment.

The T1 scan was acquired at 1.0mm isotropic resolution with an echo time (TE) and repetition time (TR) of TE/TR=0.004605/0.009062s on a 3T Philips Achieve dStream scanner. The T2* scan was acquired at a resolution of 0.2 × 0.2 × 1.1 mm³ with an echo time (TE) and repetition time (TR) of TE/TR=0.023689/0.844196s on a 7T Philips Achieva scanner. The scans were obtained in the DICOM format and were converted to NIFTI format. An axial slice from the T1 scan, axial slice from T2* scan and the brain surface are shown in Figure 1.

Figure 1.

We acquired a T1 scan from a 3T scanner and a T2* scan from a 7T scanner. We extracted brain surfaces from the T1 scan using the consistent cortical reconstruction and multi-atlas brain segmentation tool, MaCRUISE.

2.2. Preprocessing

Due to the lack of an available tool to reconstruct cortical surfaces from 7T scans (Figure 1), we employ a 3T MRI scan from the same patient to achieve precise cortical surfaces using our existing image processing pipeline that is built for extracting brain surfaces. Briefly, the surfaces of the brain were extracted from the T1 scan with the help of MaCRUISE^9,10 and the whole brain was separated into the left and right hemispheres¹¹. The surfaces obtained were in the VTK polydata format.

To use the surface points from the VTK files for our experiment, we first aligned the surface files to the correct orientation and then aligned the surfaces from the VTK files with their respective voxels on the 3T T1 scan. The next step was to register the T1 scan to the T2* scan using affine registration from ANTs toolkit¹³. The transformation matrix thus obtained for the 3T to 7T scan registration was used to align the surface files to 7T space. This process has been summarized in Figure 2.

Figure 2.

The surface VTK file when overlaid on the T1 shows misalignment and a flip on the axial plane. We first aligned the surfaces to the correct orientation; after which we aligned the surface to T1 scan voxel space using a custom Python script. The surface aligned with the T1 scan can still be seen misaligned with the 7T scan. Applying the forward affine transform of T1 to the 7T T2* scan on the surface aligns the surface with the 7T scan.

2.3. Implementation

The first step for extraction of the high-resolution orthogonal slices from the T2* scan was implemented by aligning the VTK surface files with the 7T volume. We first tessellated the surface using VTK’s Adaptive Subdivision filter¹². To ensure that the tessellation covered all the voxels, we set maximum edge length of the tessellation function to be smaller than the lowest voxel dimension, 0.2mm. The tessellated surface provided about 10 times the number of points on the surface than the original surface file.

Following the tessellation, we computed the normal vectors on the tessellated cells using VTK’s libraries for computing vertex normal. VTK’s polydata normal approximates the normals on each of the cells and averages the values at shared points. We converted the millimeter coordinates of the points on the surface and the corresponding normals to NumPy format for ease of computation. To compute the angle between the normal vector and the slice direction, we had to first construct unit normal vectors. Following this, we were able to perform the dot product on the unit normal vectors with unit vectors in the slice direction and convert the values to angles. The angles have been normalized to be between 0° and 90° since the angles in the range 90° and 180° just show the surface normal vectors in the opposite direction. The surface points were converted into a NIFTI image using TriMeshPy¹⁴ with the voxel intensities were equal to the angle computed from the respective surface points. The NIFTI image thus obtained is overlaid as a heat map on the 7T T2* scan in Figure 3.

Figure 3.

The angles computed from a surface point has been coded into an image by assigning the numerical value of the angle as intensity of the voxel in the corresponding coordinates. This figure shows a colormap visualization of the NIFTI image generated with numerical value of angle as intensity overlaid on the 7T T2* scan.

3. RESULTS

Our hypothesis for this experiment was that the regions on the surface that are obtained orthogonal to the axial slice direction will provide a higher resolution plane where it can be easier to demarcate the cortical layers visually. Therefore, we assessed three categories of surface points: surface points where the normal vectors are at an angle greater than 80° with the slice direction (hypothesized to be high quality slides where it is possible to observe cortical layers), surface points where normal vectors are at an angle close to 60° with the slice direction (provides lesser information about the cortical layers than the points with angle 80° and above), surface points where the normal vectors are at an angle less than 30° with the slice direction (poor quality imaging ordination and an inability to observe cortical layers). The histogram of the angles between 0° and 90° are reported in Figure 4.

Figure 4.

A histogram of the angles shows that close to 20% of the points have an angle of acquisition with the slice direction above 80° and about 15% of the points are surfaces that are almost along the slice direction. If we set a threshold for points with a surface normal angle of 60° and above to provide usable slices, then about 50% of the points are usable to observe the cortical layers.

From the histogram of the angles between the gray matter surface normal and the slice direction, we see 18.58% of the points have an angle greater than 80°. 30.31% of the points have a normal vector that angles greater than 60° but lesser than 80° with the slice direction. 15.12% of the points have a normal vector that angles lesser than 30° with the slice direction. If we set a threshold of usable slices to be above 60°, then we have 48.79% of the points where the axial slice will show contrastive differences between different cortical layers.

Figure 5. shows slices that correspond to three categories of points: 1) angled over 80°, 2) angled close to 60°, and 3) angled less than 30° with the slice direction. Axial slices taken at coordinates where the angles between the surface normal vectors and slice direction are greater than 80° (Figure 5a.) show clear demarcation between different cortical layers. The high contrast between the different layers clearly defines the boundary of each layer, making them visually differentiable. Axial slices taken at coordinates with angles close to 60° (Figure 5b.) show a drop in contrast between the cortical layers. In these slices we can observe slight differences between cortical layers, but their boundaries are not well defined. Labeling the cortical layers on these slices may be inaccurate since it is difficult to determine the extent of each layer. The quality of the axial slices taken at coordinate points where the angle between the slice direction and the surface normal vectors are lesser than 30° (Figure 5c.), provide no information that would allow us to visually differentiate between the different layers in the cortex. The contrast between the layers is exceptionally low and the boundaries are unclear. Therefore, we observe that if the angle between the surface normal vectors and the slice direction are closer to 90°, the contrast between the cortical layers is more pronounced and the boundaries can be visually defined.

Figure 5.

The three sections correspond to the three broad classifications of the points based on the angles that the normal vectors make with the slice direction. The first category(a) is a surface point that corresponds to an angle of 85.7° and has an exceptionally good axial slice for observing contrast between the three laminar subdivisions (deep, middle, and superficial). The second point of interest (b) has an angle of 60.38° with the acquisition angle and shows an axial slice where the contrast between the cortical layers has dropped and the laminar subdivisions are not clear. The third point of interest (c) has an angle of 27.79° with the acquisition direction and makes for an axial slice that shows extremely poor contrast between the different cortical layers and makes it impossible to obtain any information about the laminar subdivisions.

4. DISCUSSION AND CONCLUSIONS

Observing the different anatomical characteristics of each of the cortical layers such as myelination or the density can help investigate the relationship between the laminar structure and function. From an ultra-high-resolution scan acquired using a 7T MRI scanner, we can non-invasively observe different layers in the cortex. However, as we have established before, the manual process for measuring the thickness of the cortical layers is time intensive since it involves visual inspection of all the imaging planes for impartial volumes. In this work, we built an analytical tool that will greatly reduce this effort by identifying cortical surface points that have been acquired impartially in the ultra-high-resolution plane. From our experiment, we identified 20% of the cortical surface points where the axial slices taken from their coordinates showed distinctive contrast between the cortical layers.

By reserving our efforts for slices that provide an ultra-high-resolution image of the cortical layers, we can eliminate the tedious manual inspection of the slices. Additionally, automatic extraction of the ultra-high-resolution slices will reduce the bias of characterizing the relationship between the laminar structure and brain functionality. In the future, we could extend our work towards automatically labelling the different cortical layers by learning the individual features that are visually observable in the ultra-high-resolution slices.

The primary limitation of our work is that the surface extraction is performed on the 3T scan and is then transferred to a 7T scan acquired on the same subject. This calls for two separate scanning sessions on a subject which limits the amount of data available. Furthermore, we use tessellation of the surface to ensure coverage of all voxels on the 7T scan with a VTK polydata cell. This does not account for minute curvature changes on the surface of the brain from the 7T scan. In this experiment, we used tools and techniques that our group has prior experience with, and we did not find a reliable method to extract brain surfaces directly from the 7T scans. We could potentially use steps from our current work to inform us about improving brain surface extraction directly from a 7T scan.