Imaging the deep cerebellar nuclei: A probabilistic atlas and normalization procedure
Research Highlights
► Identification of deep cerebellar nuclei in the human cerebellum using MRI at 7 T. ► Probabilistic atlas of dentate, interposed and fastigial nucleus. ► Improved spatial normalization of nuclei for functional imaging.
Introduction
A key anatomical feature of the cerebellum is the deep cerebellar nuclei (DCN), which relay all output from the cerebellar cortex to cortical and subcortical targets. The fastigial nuclei receive the output from the vermis, the interposed nuclei (in humans comprised of the emboliform and globose nuclei) from the paravermis, and the large dentate nuclei from the lateral cerebellar hemispheres.
For studies of human participants with focal lesions it is therefore very important to know, whether and to what degree the DCN are involved. Furthermore, while most functional neuroimaging studies focus on the cerebellar cortex (for a review see: Stoodley and Schmahmann, 2009), the BOLD signal here mostly reflects the input to the cerebellum (Thomsen et al., 2009). Therefore, reliable measurement of activity in the DCN, which should relate more to the cerebellar output, could provide useful functional insights. Indeed, a number of studies have provided evidence that reliable functional activation within the dentate nucleus can be observed (Dimitrova et al., 2006a, Habas, 2009, Kim et al., 1994).
The small size of the cerebellar nuclei poses a challenge for human MRI studies. While the size of the dentate nucleus in itself is substantial (~ 13 × 19 × 14 mm3), one has to keep in mind that it is composed of a number of subdivisions with unique functions. In the monkey, the dorso-rostral part of the dentate is relaying somatotopic organized output to the motor cortex, while the ventro-caudal part provides outputs to the parietal (area 7b) and prefrontal cortices (Dum and Strick, 2003, Middleton and Strick, 1997). Thus, the functional subdivisions of the dentate nuclei, as well as the interposed and fastigial nuclei are very small. This causes problems for traditional MRI group analyses. For example, it has been shown that after normalization to the MNI152 template, the dentate nucleus overlaps maximally only for 71% of the participants, with the average overlap being substantially poorer (Dimitrova et al., 2006b). Thus, with common normalization methods the chance that functionally corresponding areas of the deep cerebellar nuclei superimpose across participants is quite small.
An additional challenge for functional imaging of the deep cerebellar nuclei is posed by the high iron content of the DCN, which leads to signal losses due to magnetic susceptibility artifacts (Aoki et al., 1989, Gans, 1924, Maschke et al., 2004). For T2*-weighted EPI sequences, the mean signal of voxels in the deep cerebellar nuclei is typically only 2/3 of the surrounding white and 1/2 of the adjacent gray matter. This is problematic, as both functional activations and noise in fMRI data scale roughly proportional with the mean voxel signal (Diedrichsen and Shadmehr, 2005), meaning that the functional signal from the deep cerebellar nucleus is quite small compared to the surrounding noise signals. When one now applies a 6–10 mm smoothing kernel to account for poor overlap across participants, a small functional signal is averaged with much more variable signal from the surrounding tissue. This makes it likely that true activations of individual compartments of the DCN are missed (Type I error). Furthermore, there is a substantial danger that reported activations of the DCN may be due to activation from surrounding gray matter structures (Type II error).
To address these shortcomings we provide here three related improvements. First, we used ultrahigh (7 T) MRI to collect anatomical data of the cerebellar nuclei in the submillimeter range (Gizewski et al., 2007). We used a susceptibility-weighted imaging sequence (Haacke et al., 2009, Haacke et al., 2004), which is very sensitive to the high iron content of the DCN, allowing us to identify the dentate, interposed, and fastigial nucleus in most participants. This technique will be useful to determine the disruption of the DCN for individual patients with focal cerebellar lesions.
Secondly, we provide a probabilistic atlas of the location of the nuclei in a common atlas space. Because probabilistic maps are dependent on the normalization method that is used to bring individual brains into a common reference frame (Diedrichsen, 2006), we systematically compared a number of normalization techniques. Older, yet still commonly-used normalization techniques, are outperformed substantially by more modern techniques such as concurrent segmentation and normalization (Ashburner and Friston, 2005) or cerebellar-only normalization (SUIT, Diedrichsen, 2006). The resulting maps allow for the valid assignment of structural impairment and functional activations to specific cerebellar nuclei, and have been integrated into an existing probabilistic atlas of the human cerebellum (Diedrichsen et al., 2009).
The evaluated anatomical normalization methods all rely on T1-weighted images, on which the deep cerebellar nuclei are not visible. Assuming that the surrounding structures are brought into perfect alignment through normalization, the variability of the nuclei after normalization therefore provides a quantitative measure of the anatomical variability of the location of the cerebellar nuclei in relationship to these surrounding structures. This remaining variability, however, poses a problem when trying to study the functional specialization of different compartments of the DCN.
We therefore thirdly propose a new technique for the group analysis of functional data from the deep cerebellar nuclei. This method incorporates the information from the hypo-intensity of the dentate nucleus, a piece of information that is readily available from a standard T2*-weighted image (such as the mean EPI image), into the normalization process. This effectively ensures near perfect superposition of the dentate nuclei across different participants, thereby eliminating the remaining anatomical variability. The approach combines the high anatomical accuracy of ROI-based analyses with the ease and information richness of map-wise comparisons. The normalization routines are made freely available online as an addendum to the cerebellar normalization toolbox (Diedrichsen, 2006).
Section snippets
Subjects
MR images were collected in 28 healthy participants. Five subjects were excluded, because of movement artifacts (n = 3) or incidental finding of MRI abnormalities (n = 2: cerebellar cavernoma, large subarachnoidal cyst in the posterior fossa). The final analysis included data from 23 subjects (9 males, 14 females; mean age 35.1, SD 13.1 years, 21–61 years). All subjects except three were right-handed: one subject was left-handed and two were ambidextrous based on the Edinburgh handedness inventory (
Individual anatomy
The dentate nuclei could be identified in all of the 23 subjects. The interposed nucleus was identified in 21 subjects, and fastigial nuclei in 17 subjects. Fig. 2 shows three horizontal slices of an individual SWI image, on which the nuclei are visible as hypo-intensities. For direct comparison we superimposed the ROI drawings for the individual nuclei on the same slices in a second row. In the most ventral slice (a) the dentate nucleus (D; indicated in red) with its corrugated thin walls and
Discussion
In sum, the paper provides a number of innovations that will be useful for the study of the deep cerebellar nuclei in humans. First, we showed the feasibility of depicting the deep cerebellar nuclei using susceptibility-weighted imaging in high-field MRI (7 T). This technique is useful for detecting the deep cerebellar nuclei in single participants for detailed patient and functional imaging studies. Secondly, we provide a probabilistic atlas for the nuclei in the standard atlas space of the MNI
Acknowledgments
The work was supported by a grant from the National Science Foundation (BSC 0726685) to JD, and a grant from the German Research Foundation to DT, MEL and ERG (TI 239/9-1). We thank Marc Guitart Masip for helpful comments.
References (48)
- et al.
Unified segmentation
Neuroimage
(2005) - et al.
Visualization of the deep cerebellar nuclei using quantitative T1 and rho magnetic resonance imaging at 3 Tesla
Neuroimage
(2007) A spatially unbiased atlas template of the human cerebellum
Neuroimage
(2006)- et al.
A probabilistic MR atlas of the human cerebellum
Neuroimage
(2009) - et al.
Detecting and adjusting for artifacts in fMRI time series data
Neuroimage
(2005) - et al.
Activation of cerebellar nuclei comparing finger, foot and tongue movements as revealed by fMRI
Brain Res. Bull.
(2006) - et al.
MRI atlas of the human cerebellar nuclei
Neuroimage
(2002) - et al.
Probabilistic 3D MRI atlas of the human cerebellar dentate/interposed nuclei
Neuroimage
(2006) - et al.
A critique of functional localisers
Neuroimage
(2006) - et al.
fMRI at 7 T: whole-brain coverage and signal advantages even infratentorially?
Neuroimage
(2007)
Improved segmentation of deep brain grey matter structures using magnetization transfer (MT) parameter maps
Neuroimage
The assessment and analysis of handedness: the Edinburgh inventory
Neuropsychologia
Divide and conquer: a defense of functional localizers
Neuroimage
Advances in functional and structural MR image analysis and implementation as FSL
Neuroimage
Functional topography in the human cerebellum: a meta-analysis of neuroimaging studies
Neuroimage
Cerebellum and precerebellar nuclei
FNIRT — FMRIB's non-linear image registration tool
Organ. Hum. Brain Mapp.
The human cerebellum
An atlas of Gross Topography in Serial Sections
Normal deposition of brain iron in childhood and adolescence: MR imaging at 1.5 T
Radiology
The human brain in figures and tables
A Quantitative Handbook
Cerebellar dentate nucleus
Organization, Cytology, and Transmitters
Anatomie des Centres Nerveux. II
An unfolded map of the cerebellar dentate nucleus and its projections to the cerebral cortex
J. Neurophysiol.
The human brain stem and cerebellum
Surface, Structure, Vascularization, and Three-dimensional Sectional Anatomy with MRI
Cited by (276)
Methods for cerebellar imaging analysis
2023, Current Opinion in Behavioral SciencesAn autonomic mode of brain activity
2023, Progress in NeurobiologyAn anatomical and connectivity atlas of the marmoset cerebellum
2023, Cell ReportsA multimodal submillimeter MRI atlas of the human cerebellum
2024, Scientific Reports