High-Resolution Motion-corrected 7.0-T MRI to Derive Morphologic Measures from the Human Cerebellum in Vivo

cerebello-cortical


Whole-Cerebellar Cortex Sequence
The whole-cerebellar cortex sequence consisted of a magnetization-prepared 2 rapid gradient-echo (MP2RAGE) sequence (FOV, 210 × 120 × 60 mm 3 ; 5.65/1.88;inversion time 1/inversion time 2, 1000/2900 msec; flip angle for inversion times 1 (up to 200 μm).When the voxel size becomes that small, the effective spatial resolution is limited by involuntary patient motion (7).We therefore interleaved our MRI sequences with rapidly sampled whole-head fat images (8), with which we prospectively corrected participant motion.We aimed to examine the utility of our high-resolution, motion-resistant method in noninvasive imaging of the cerebellar cortex in a clinically applicable time.

Participants
Following approval of this prospective study by the local ethical committee, included participants provided written informed consent.The participants were screened before the experiments to ensure MRI compatibility (Fig 1).Nine healthy participants were scanned in a 7.0-T MRI scanner (Achieva, Philips Healthcare) using a two-channel transmit and 32-channel receive whole-head coil (Nova Medical) with a sequence aimed at imaging layers within the cerebellar cortex.Additionally, nine healthy participants completed a sequence that visualized the whole cerebellar cortex (three overlapping participants between groups).The images were visually inspected (N.P., with 7 years of MRI acquisition experience) for image quality (eg, B 1 inhomogeneity artifacts) and rejected if necessary.The study was conducted between February 2021 and July 2022.

Cerebellar Cortical Layers
The cerebellar cortical depth sequence consisted of a high-resolution in-plane T2*-weighted ■ In a prospective study of nine healthy participants, a susceptibility difference between the inner and outer cerebellar cortex visualized the granular and molecular cerebellar cortical layers in all participants.
■ The motion-corrected magnetization-prepared 2 rapid gradient-echo, or MP2RAGE, sequence with an isotropic in-plane spatial resolution of 0.4 mm resolved the cerebellocortical fissures in all participants; the median cortical surface area was 1.8 times larger and the median cortical thickness 5 times thinner compared with T1-weighted sequences with a spatial resolution greater than 0.75 mm.

Image Analysis
For the FLASH image, the tissue phase was derived following Laplacian unwrapping and background-field elimination with use of vSHARP, or the varying spherical-kernel sophisticated harmonic artifact reduction for phase data method (13) (Fig 2A).Six degrees-of-freedom transforms were calculated between the FLASH and MP2RAGE sequences with use of the Advanced Normalization Tools software, version 2.1 (Penn Image Computing and Science Laboratory) (14).
In the magnitude FLASH image, a sample cortical branch was manually selected through three successive sections (Fig 3A -3D), and voxel classification to WM, GM, and cerebrospinal fluid was performed using a signal intensity gradient magnitude plot (Segmentator [15]).The FLASH data were upsampled to 0.1 × 0.1 mm 2 in plane, and the initial voxel classifications were entered into the cortical reconstruction using implicit surface evolution, or CRUISE, algorithm (Nighres version 1.3 [16,17]) to extract non-selfintersecting WM and GM segmentations.Within these segmentations, the magnitude and phase FLASH intensity values and the T1 values from the MP2RAGE sequence were extracted at nine cortical depths, as defined with an equidistant criterion (17).For comparison purposes, an area of interest was drawn in the occipital cortex where prominent intracortical striations exist (Fig 3A -3D).
The T1-weighted images derived from the MP2RAGE sequence were corrected for residual intensity bias (N4 algorithm) and denoised using a spatially adaptive filter (Advanced Normalization Tools software [18]) (Fig 2A).Downsampled versions of the initial 0.4 × 0.4 × 0.4-mm 3 images were created at 0.75 × 0.75 × 0.75 mm 3 and 1 × 1 × 1 mm 3 .A signal intensity gradient magnitude was used to extract an initial probability distribution for WM, GM, and cerebrospinal fluid, and non-self-intersecting segmentations were derived as described earlier.The mid GM was estimated with an equidistant criterion, and the output was densely tessellated with a fast-marching algorithm to a mesh (approximately 4.6 million vertices; approximately 25 times the typical FreeSurfer neocortex reconstruction).The resulting surface was computationally unfolded (17,19).

Statistical Analysis
For the MRI sequences acquired with and without prospective motion correction, the frame-to-frame motion distribution variances were compared with an F test.The average edge strength ratio (an image sharpness measure) was calculated (20) following brain masking with the FSL Brain Extraction Tool (FMRIB Analysis Group).A paired t test was performed to evaluate if improved sharpness was achieved (empirical evidence suggested that the data were similar to a normal distribution).For the healthy young participants, a repeated-measures analysis of variance was performed to examine if the T2* magnitude, T2* phase, and T1 values were differentially dependent on the cerebellar cortical depth.Similarly, for the healthy young participants, a repeated-measures analysis of variance was used to examine if downsampling the initial data affected the derived cortical measures (surface area, thickness, and WM and GM volume).Significant main effects were investigated pairwise with paired t tests after Bonferroni adjustment.All tests were performed by N.P. using the R Project software (version 3.5.1,The R Foundation).Statistically significant difference was established at P < .05.

Participant Characteristics
The cohort of young participants who underwent imaging with the FLASH sequence consisted of seven participants, with a median age of 36 years (IQR, 25-38 years) (five women and two men) (Table ).Our cohort of young participants who underwent imaging with the MP2RAGE sequence consisted of seven participants, with a median age of 36 years (IQR, 34-38 years) (four women and three men).In addition, two healthy older men (aged 57 and 62 years) were recruited and completed both the MP2RAGE and FLASH sequences.In two male participants who underwent imaging with the MP2RAGE sequence, B 1 inhomogeneities were evident in the cerebellum (one was excluded from analysis).

Image Quality and Motion Correction Efficacy
The prospective motion correction preserved the high-resolution image features (Figs 2B-2F, S3), resulting in improved

Cerebellar Cortical Layers
In all participants, we visually observed a negative frequency shift in the WM and a positive shift in the deep GM (up to for statistical parameters).This frequency shift extended across sections, followed the cerebellar foliation, and was confirmed to be related to susceptibility due to its dependence on B 0 orientation (Figs 3H, S5).The myelin-rich striation of the neocortical occipital lobe visually showed a small negative frequency shift and a T1 reduction (Fig 3C ,  3D).In sum, signal variations consistent with the granular and molecular layers of the cerebellar cortex were visualized across all individuals.

Cerebellar Segmentations
We next extended our approach to image and segment the entire cerebellar cortex.Our high-spatial-resolution MP2RAGE sequence (0.4 mm isotropic; voxel volume, 0.064 mm 3 ) allowed

Cerebellar Cortical Surface
The resulting surface showed the characteristic transverse fissures of the cerebellar cortex, as well as other anatomic details, including splitting folds and lobular folds angulated toward the predominant direction (Fig 4D , 4E).The high-fidelity cerebellar surface allowed us to unfold the transverse fissures of the cerebellar cortex to reveal the continuous cortical sheet (Fig 4F , Movie 2).As an example application, we projected T1 estimates (a myelin-sensitive measure) on the inflated surface (Fig 4F).Downsampling the data to current state-of-the-art MRI acquisitions reduced the visibility of the cerebellar anatomic features, such as folia (Fig S6 ), and significantly lowered the estimated cortical thickness (Tables S4-S6).The median cerebellar cortical surface area was estimated at 949 cm 2 (IQR, 825-1021 cm 2 ), 84% (21) and 60% (1) of the only two previous high-fidelity surface area estimates (both ex vivo and requiring several hours of MRI  S7).The median cerebellar cortical thickness was estimated at 0.88 mm (IQR, 0.81-0.93mm) in agreement with ex vivo reports (0.7-0.8 mm) (22) and four to five times smaller than the current typical imaging-based in vivo estimates.Our median WM volume estimate (64 cm 3 [IQR, 62-69 cm 3 ]) agreed with the reference ex vivo study (63 cm 3 [1]).

Older Participants
The older participants showed visible cortical thinning in the cerebellum at visual inspection (Fig 6A , 6B).The cerebellar cortical thickness and GM T1 values of both older participants were more than 1.5 times the IQR below the first quartile of the young cohort distribution, suggesting a deviation of these data points from the younger cohort (Fig 6C -6E, Table S8).

Discussion
To date, methods to image the human cerebellum in vivo with fidelity to the level of individual foliations remain lacking.This hinders examining the role of the cerebellum in various disease processes.In this study, we demonstrated that motion-resistant 7.0-T MRI can be used to image the cerebellar cortical layers and to reconstruct and unfold the cerebellar cortical surface while resolving individual foliations.This allows for calculation of quantitative measures, such as the cerebellar cortical surface area and thickness, and examination of correlates of the cerebellar myeloarchitecture, such as T1 values, on the continuous cerebellar cortical sheet.
To achieve between-layer contrast in the cerebellar cortex, we used the higher magnetic susceptibility difference of the deep, iron-rich granular layer and the less neuronally dense superficial molecular layer (9) at 7.0 T. This resulted in consistent visualization of an inner and outer layer-like structure in the cerebellar cortex across participants, likely relating to the granular and molecular layers and in agreement with a previous pilot study, confirmed with histologic examination (9).Spatial resolution limits prohibit the visualization of the middle Purkinje layer.The cerebellar layers are differentially affected in diseases like multiple sclerosis (where extensive demyelination of the molecular layer is observed [2]) or spinocerebellar ataxia type 6, where the Purkinje and molecular layers atrophy while the granular layer is spared (23).The visualization of the cerebellar cortical layers may therefore provide disease markers for prognosis or intervention.

9
The derivation of clinical cortical measures, such as cortical volume or thickness, relies on accurate GM/WM segmentations to the level of individual folia.In current clinical cerebellar research, the GM/WM segmentations pragmatically rely on anatomic templates to reduce the sensitivity to inadequate spatial or contrast resolution (1).This implicitly smooths over the sublobular anatomy, including the functionally important mediolateral folds.Measures that are commonly used in cerebral clinical research, such as volume or cortical thickness, become unreliable when applied in the cerebellum with typical spatial resolution (≥0.8 mm) (1).Herein, we extracted WM/GM segmentations that distinguished several, though not all, individual folia.There are few ex vivo reference studies (Table S7), but our measures (cortical surface area, cortical thickness, and WM volume) were closer to these ex vivo references compared with previous in vivo studies.
This further allowed us to unfold the cerebellar cortical surface.Cerebellar foliation-level unfolding has not been attempted before in vivo, although it was recently demonstrated ex vivo (and after laborious effort [1]).Despite the highly regular, cylindrical nature of the cerebellar foliation, the curvature of the cerebellar lobules produces a surface with higher intrinsic curvature than that of the cerebrocortical surface (1).We limited the inflation to the lobular level to avoid incurring large distortions (1) but unfolded the mediolateral folds and revealed the deeper folia.The unfolded view may facilitate examining continuous myelination or functional activation differences in clinical conditions, similar to the cerebral cortex (24).
Our study has limitations.First, our study relied on the high signal-to-noise ratio and contrast-to-noise ratio of 7.0-T MRI.Such a field strength is currently rare, although it is becoming increasingly available to clinical researchers.Second, our sample size, while typical for MRI technical development studies (25), was limited, and the method was evaluated in healthy individuals.Third, our 0.19 × 0.19 × 0.5-mm 3 layer visualization sequence did not include the whole cerebellar cortex to limit scan time.Fourth, there is a current dearth of ex vivo reference studies from which to derive reference data.Finally, cerebellum 7.0-T MRI frequently suffers from B 1 inhomogeneities, which were particularly evident in two participants in our study.Image quality can be further enhanced in future implementations using plug-and-play parallel transmit approaches (12).
In summary, 7.0-T MRI with a nonisotropic in-plane resolution of up to 0.19 mm with motion correction provided in vivo visualization of the cerebellar cortical layers and cerebellar surface.

Figure 3 :
Figure 3: (A) Example fast low-angle shot sections of cerebellar (top) and cerebrocortical (bottom) intracortical contrast.Boxes indicate zoomed portions.(B) T2*weighted magnitude image in the cerebellum and cerebrum.The cortical depth was sampled from white matter (WM) (red) to the cerebrospinal fluid (yellow).(C) Phase image.Contrast in the cerebellar cortex can be seen (arrows).(D) T1 map.(E-G) Box and whisker plots show group cortical depth signal profiles for cerebellar (dark gray) and cerebral (light gray) cortex for the T2*-weighted magnitude images (E), phase images (F), and T1 maps (G).Boxes indicate the IQR, midlines show the median, and whiskers represent the minimum and maximum values.CSF = cerebrospinal fluid.(H) Successive sections of the extracted phase image of the cerebellar cortex.The intracortical contrast was highly consistent across successive sections.An identification of the granular (GL) and molecular (ML) layers was made.The intracortical contrast was more pronounced perpendicular (white dashed line) rather than parallel (black dashed line) to the B 0 orientation, suggesting a susceptibility origin.

Figure 4 : 7 20- 25
Figure 4: (A) Whole-cerebellar cortex T1-weighted image (0.4 mm isotropic).Boxes indicate zoomed portions.The vertical band is the 0.19 × 0.19 × 0.5-mm partial field of view T2*-weighted fast low-angle shot sequence, for reference.At 0.4 mm, individual folia (arrows) can still be resolved.(B) White matter segmentations.(C) Pial surface segmentations.(D) High-fidelity segmentations allow cerebellar surfaces that resolve the characteristic transverse fissures (arrows).(E) Splitting folds (wavy arrows) and angulated folds compared with neighboring lobules (straight arrows) can also be resolved.(F) Projection of the T1 map cortical ribbon to the surface and surface unfolding (anterior and posterior view).Arrows indicate the projection, rotation, and unfolding processes.

Figure 5 :
Figure 5: Individual-value plots show estimates compared with reference values (see Table S7 for extended data).Blue dots indicate our in vivo results; red dots, ex vivo literature results; green dots, in vivo literature results.(A) Cerebellar cortical surface area estimates.(B) Cerebellar cortical thickness estimates.(C) White matter volume estimates.(D) Gray matter volume estimates.

Figure 6 :
Figure 6: (A, B) Example cerebellar sections (increasing in participant age from left to right) from the T1 map (A) and T2*-weighted (T 2 -w) magnitude image (B).For the older individuals, cerebellar cortical thinning can be seen visually as increased cerebrospinal fluid space (arrows).a.u.= arbitrary units, y.o.= years old.(C-E) Individual-value plots show cerebellar measures across cohorts for cortical thickness (C), cerebellar gray and white matter volume (D), and T1 values (E).

Demographic Characteristics of the Study Participants
* Data are numbers of participants.