Automated MRI cerebellar size measurements using active appearance modeling. Neuroimage. 2014;103:511-521. [PMID: 25192657 DOI: 10.1016/j.neuroimage.2014.08.047]

CerebNet: A fast and reliable deep-learning pipeline for detailed cerebellum sub-segmentation

Quantifying the volume of the cerebellum and its lobes is of profound interest in various neurodegenerative and acquired diseases. Especially for the most common spinocerebellar ataxias (SCA), for which the first antisense oligonculeotide-base gene silencing trial has recently started, there is an urgent need for quantitative, sensitive imaging markers at pre-symptomatic stages for stratification and treatment assessment. This work introduces CerebNet, a fully automated, extensively validated, deep learning method for the lobular segmentation of the cerebellum, including the separation of gray and white matter. For training, validation, and testing, T1-weighted images from 30 participants were manually annotated into cerebellar lobules and vermal sub-segments, as well as cerebellar white matter. CerebNet combines FastSurferCNN, a UNet-based 2.5D segmentation network, with extensive data augmentation, e.g. realistic non-linear deformations to increase the anatomical variety, eliminating additional preprocessing steps, such as spatial normalization or bias field correction. CerebNet demonstrates a high accuracy (on average 0.87 Dice and 1.742mm Robust Hausdorff Distance across all structures) outperforming state-of-the-art approaches. Furthermore, it shows high test-retest reliability (average ICC >0.97 on OASIS and Kirby) as well as high sensitivity to disease effects, including the pre-ataxic stage of spinocerebellar ataxia type 3 (SCA3). CerebNet is compatible with FreeSurfer and FastSurfer and can analyze a 3D volume within seconds on a consumer GPU in an end-to-end fashion, thus providing an efficient and validated solution for assessing cerebellum sub-structure volumes. We make CerebNet available as source-code (https://github.com/Deep-MI/FastSurfer).

Gender-related neuroanatomical differences in alcohol dependence: findings from the ENIGMA Addiction Working Group

•

We tested gender differences in brain volumes of alcohol dependent vs control groups.

•

Group differences in brain volumes emerged as gross and widespread.

•

Group-by-gender effects emerged in selected brain regions (cerebellum, amygdala)

•

In dependent users, greater alcohol use predicted smaller amygdala and larger cerebellum GM volume.

•

Our results highlight the need to account for gender differences in MRI studies of alcohol dependence.

Gender-related differences in the susceptibility, progression and clinical outcomes of alcohol dependence are well-known. However, the neurobiological substrates underlying such differences remain unclear. Therefore, this study aimed to investigate gender differences in the neuroanatomy (i.e. regional brain volumes) of alcohol dependence. We examined the volume of a priori regions of interest (i.e., orbitofrontal cortex, hippocampus, amygdala, nucleus accumbens, caudate, putamen, pallidum, thalamus, corpus callosum, cerebellum) and global brain measures (i.e., total grey matter (GM), total white matter (WM) and cerebrospinal fluid). Volumes were compared between 660 people with alcohol dependence (228 women) and 326 controls (99 women) recruited from the ENIGMA Addiction Working Group, accounting for intracranial volume, age and education years. Compared to controls, individuals with alcohol dependence on average had (3–9%) smaller volumes of the hippocampus (bilateral), putamen (left), pallidum (left), thalamus (right), corpus callosum, total GM and WM, and cerebellar GM (bilateral), the latter more prominently in women (right). Alcohol-dependent men showed smaller amygdala volume than control men, but this effect was unclear among women. In people with alcohol dependence, more monthly standard drinks predicted smaller amygdala and larger cerebellum GM volumes. The neuroanatomical differences associated with alcohol dependence emerged as gross and widespread, while those associated with a specific gender may be confined to selected brain regions. These findings warrant future neuroscience research to account for gender differences in alcohol dependence to further understand the neurobiological effects of alcohol dependence.

Automatic cerebellum anatomical parcellation using U-Net with locally constrained optimization

The cerebellum plays a central role in sensory input, voluntary motor action, and many neuropsychological functions and is involved in many brain diseases and neurological disorders. Cerebellar parcellation from magnetic resonance images provides a way to study regional cerebellar atrophy and also provides an anatomical map for functional imaging. In a recent comparison, a multi-atlas approach proved to be superior to other parcellation methods including some based on convolutional neural networks (CNNs) which have a considerable speed advantage. In this work, we developed an alternative CNN design for cerebellar parcellation, yielding a method that achieves the leading performance to date. The proposed method was evaluated on multiple data sets to show its broad applicability, and a Singularity container has been made publicly available.

Cerebellar Morphometry and Cognition in the Context of Chronic Alcohol Consumption and Cigarette Smoking

BACKGROUND

Cerebellar atrophy (especially involving the superior-anterior cerebellar vermis) is among the most salient and clinically significant effects of chronic hazardous alcohol consumption on brain structure. Smaller cerebellar volumes are also associated with chronic cigarette smoking. The present study investigated effects of both chronic alcohol consumption and cigarette smoking on cerebellar structure and its relation to performance on select cognitive/behavioral tasks.

METHODS

Using T1-weighted Magnetic Resonance Images (MRIs), the Cerebellar Analysis Tool Kit segmented the cerebellum into bilateral hemispheres and 3 vermis parcels from 4 participant groups: smoking (s) and nonsmoking (ns) abstinent alcohol-dependent treatment seekers (ALC) and controls (CON) (i.e., sALC, nsALC, sCON, and nsCON). Cognitive and behavioral data were also obtained.

RESULTS

We found detrimental effects of chronic drinking on all cerebellar structural measures in ALC participants, with largest reductions seen in vermis areas. Furthermore, both smoking groups had smaller volumes of cerebellar hemispheres but not vermis areas compared to their nonsmoking counterparts. In exploratory analyses, smaller cerebellar volumes were related to lower measures of intelligence. In sCON, but not sALC, greater smoking severity was related to smaller cerebellar volume and smaller superior-anterior vermis area. In sALC, greater abstinence duration was associated with larger cerebellar and superior-anterior vermis areas, suggesting some recovery with abstinence.

CONCLUSIONS

Our results show that both smoking and alcohol status are associated with smaller cerebellar structural measurements, with vermal areas more vulnerable to chronic alcohol consumption and less affected by chronic smoking. These morphometric cerebellar deficits were also associated with lower intelligence and related to duration of abstinence in sALC only.

SUITer: An Automated Method for Improving Segmentation of Infratentorial Structures at Ultra-High-Field MRI

BACKGROUND AND PURPOSE

The advent of high and ultra-high-field MRI has significantly improved the investigation of infratentorial structures by providing high-resolution images. However, none of the publicly available methods for cerebellar image analysis has been optimized for high-resolution images yet.

METHODS

We present the implementation of an automated algorithm-SUITer (spatially unbiased infratentorial for enhanced resolution) method for cerebellar lobules parcellation on high-resolution MR images acquired at both 3 and 7T MRI. SUITer was validated on five manually segmented data and compared with SUIT, FreeSurfer, and convolutional neural networks (CNN). SUITer was then applied to 3 and 7T MR images from 10 multiple sclerosis (MS) patients and 10 healthy controls (HCs).

RESULTS

The difference in volumes estimation for the cerebellar grey matter (GM), between the manual segmentation (ground truth), SUIT, CNN, and SUITer was reduced when computed by SUITer compared to SUIT (5.56 vs. 29.23 mL) and CNN (5.56 vs. 9.43 mL). FreeSurfer showed low volumes difference (3.56 mL). SUITer segmentations showed a high correlation (R² = .91) and a high overlap with manual segmentations for cerebellar GM (83.46%). SUITer also showed low volumes difference (7.29 mL), high correlation (R² = .99), and a high overlap (87.44%) for cerebellar GM segmentations across magnetic fields. SUITer showed similar cerebellar GM volume differences between MS patients and HC at both 3T and 7T (7.69 and 7.76 mL, respectively).

CONCLUSIONS

SUITer provides accurate segmentations of infratentorial structures across different resolutions and MR fields.

Convergence of three parcellation approaches demonstrating cerebellar lobule volume deficits in Alcohol Use Disorder

Recent advances in robust and reliable methods of MRI-derived cerebellar lobule parcellation volumetry present the opportunity to assess effects of Alcohol Use Disorder (AUD) on selective cerebellar lobules and relations with indices of nutrition and motor functions. In pursuit of this opportunity, we analyzed high-resolution MRI data acquired in 24 individuals with AUD and 20 age- and sex-matched controls with a 32-channel head coil using three different atlases: the online automated analysis pipeline volBrain Ceres, SUIT, and the Johns Hopkins atlas. Participants had also completed gait and balance examination and hematological analysis of nutritional and liver status, enabling testing of functional meaningfulness of each cerebellar parcellation scheme. Compared with controls, each quantification approach yielded similar patterns of group differences in regional volumes: All three approaches identified AUD-related deficits in total tissue and total gray matter, but only Ceres identified a total white matter volume deficit. Convergent volume differences occurred in lobules I-V, Crus I, VIIIB, and IX. Coefficients of variation (CVs) were <20% for 46 of 56 regions measured and in general were graded: Ceres<SUIT<Hopkins. The most robust correlations were identified between poorer stability in balancing on one leg and smaller lobule VI and Crus I volumes from the Ceres atlas. Lower values of two essential vitamins-thiamine (vitamin B1) and serum folate (vitamin B9)-along with lower red blood cell count, which are dependent on adequate levels of B vitamins, correlated with smaller gray matter volumes of lobule VI and Crus I. Higher γ-glutamyl transferase (GGT) levels, possibly reflecting compromised liver function, correlated with smaller volumes of lobules VI and X. These initial results based on high resolution data produced with clinically practical imaging procedures hold promise for expanding our knowledge about the relevance of focal cerebellar morphology in AUD and other neuropsychiatric conditions.

Comparing fully automated state-of-the-art cerebellum parcellation from magnetic resonance images

The human cerebellum plays an essential role in motor control, is involved in cognitive function (i.e., attention, working memory, and language), and helps to regulate emotional responses. Quantitative in-vivo assessment of the cerebellum is important in the study of several neurological diseases including cerebellar ataxia, autism, and schizophrenia. Different structural subdivisions of the cerebellum have been shown to correlate with differing pathologies. To further understand these pathologies, it is helpful to automatically parcellate the cerebellum at the highest fidelity possible. In this paper, we coordinated with colleagues around the world to evaluate automated cerebellum parcellation algorithms on two clinical cohorts showing that the cerebellum can be parcellated to a high accuracy by newer methods. We characterize these various methods at four hierarchical levels: coarse (i.e., whole cerebellum and gross structures), lobe, subdivisions of the vermis, and the lobules. Due to the number of labels, the hierarchy of labels, the number of algorithms, and the two cohorts, we have restricted our analyses to the Dice measure of overlap. Under these conditions, machine learning based methods provide a collection of strategies that are efficient and deliver parcellations of a high standard across both cohorts, surpassing previous work in the area. In conjunction with the rank-sum computation, we identified an overall winning method.

Cerebellum and neurodegenerative diseases: Beyond conventional magnetic resonance imaging

The cerebellum plays a key role in movement control and in cognition and cerebellar involvement is described in several neurodegenerative diseases. While conventional magnetic resonance imaging (MRI) is widely used for brain and cerebellar morphologic evaluation, advanced MRI techniques allow the investigation of cerebellar microstructural and functional characteristics. Volumetry, voxel-based morphometry, diffusion MRI based fiber tractography, resting state and task related functional MRI, perfusion, and proton MR spectroscopy are among the most common techniques applied to the study of cerebellum. In the present review, after providing a brief description of each technique’s advantages and limitations, we focus on their application to the study of cerebellar injury in major neurodegenerative diseases, such as multiple sclerosis, Parkinson’s and Alzheimer’s disease and hereditary ataxia. A brief introduction to the pathological substrate of cerebellar involvement is provided for each disease, followed by the review of MRI studies exploring structural and functional cerebellar abnormalities and by a discussion of the clinical relevance of MRI measures of cerebellar damage in terms of both clinical status and cognitive performance.

CerebroMatic: A Versatile Toolbox for Spline-Based MRI Template Creation

Brain image spatial normalization and tissue segmentation rely on prior tissue probability maps. Appropriately selecting these tissue maps becomes particularly important when investigating "unusual" populations, such as young children or elderly subjects. When creating such priors, the disadvantage of applying more deformation must be weighed against the benefit of achieving a crisper image. We have previously suggested that statistically modeling demographic variables, instead of simply averaging images, is advantageous. Both aspects (more vs. less deformation and modeling vs. averaging) were explored here. We used imaging data from 1914 subjects, aged 13 months to 75 years, and employed multivariate adaptive regression splines to model the effects of age, field strength, gender, and data quality. Within the spm/cat12 framework, we compared an affine-only with a low- and a high-dimensional warping approach. As expected, more deformation on the individual level results in lower group dissimilarity. Consequently, effects of age in particular are less apparent in the resulting tissue maps when using a more extensive deformation scheme. Using statistically-described parameters, high-quality tissue probability maps could be generated for the whole age range; they are consistently closer to a gold standard than conventionally-generated priors based on 25, 50, or 100 subjects. Distinct effects of field strength, gender, and data quality were seen. We conclude that an extensive matching for generating tissue priors may model much of the variability inherent in the dataset which is then not contained in the resulting priors. Further, the statistical description of relevant parameters (using regression splines) allows for the generation of high-quality tissue probability maps while controlling for known confounds. The resulting CerebroMatic toolbox is available for download at http://irc.cchmc.org/software/cerebromatic.php.

Robust Machine Learning-Based Correction on Automatic Segmentation of the Cerebellum and Brainstem

Automated segmentation is a useful method for studying large brain structures such as the cerebellum and brainstem. However, automated segmentation may lead to inaccuracy and/or undesirable boundary. The goal of the present study was to investigate whether SegAdapter, a machine learning-based method, is useful for automatically correcting large segmentation errors and disagreement in anatomical definition. We further assessed the robustness of the method in handling size of training set, differences in head coil usage, and amount of brain atrophy. High resolution T1-weighted images were acquired from 30 healthy controls scanned with either an 8-channel or 32-channel head coil. Ten patients, who suffered from brain atrophy because of fragile X-associated tremor/ataxia syndrome, were scanned using the 32-channel head coil. The initial segmentations of the cerebellum and brainstem were generated automatically using Freesurfer. Subsequently, Freesurfer's segmentations were both manually corrected to serve as the gold standard and automatically corrected by SegAdapter. Using only 5 scans in the training set, spatial overlap with manual segmentation in Dice coefficient improved significantly from 0.956 (for Freesurfer segmentation) to 0.978 (for SegAdapter-corrected segmentation) for the cerebellum and from 0.821 to 0.954 for the brainstem. Reducing the training set size to 2 scans only decreased the Dice coefficient ≤0.002 for the cerebellum and ≤ 0.005 for the brainstem compared to the use of training set size of 5 scans in corrective learning. The method was also robust in handling differences between the training set and the test set in head coil usage and the amount of brain atrophy, which reduced spatial overlap only by <0.01. These results suggest that the combination of automated segmentation and corrective learning provides a valuable method for accurate and efficient segmentation of the cerebellum and brainstem, particularly in large-scale neuroimaging studies, and potentially for segmenting other neural regions as well.

Ageing shows a pattern of cerebellar degeneration analogous, but not equal, to that in patients suffering from cerebellar degenerative disease

Ageing generally leads to impairments in cognitive function and the ability to execute and learn new movements. While the causes of these impairments are often multi-factorial, integrity of the cerebellum in an elderly population is an important predictive factor of both motor function and cognitive function. A similar association between cerebellar integrity and function is true for cerebellar patients. We set out to investigate the analogies between the pattern of cerebellar degeneration of a healthy ageing population and cerebellar patients. We quantified cerebellar regional volumes by applying voxel-based morphometry (VBM) to a publicly available dataset of MR images obtained in 313 healthy subjects aged between 18 and 96 years and a dataset of MR images of 21 cerebellar patients. We observed considerable overlap in regions with the strongest loss of cerebellar volume in the two datasets. In both datasets, the anterior lobe of the cerebellum (lobules I-V) and parts of the superior cerebellum (primarily lobule VI) showed the strongest degeneration of cerebellar volume. However, the most significant voxels in cerebellar patients were shifted posteriorly (lobule VII) compared to the voxels that degenerate most with age in the healthy population. The results showed a pattern of significant degeneration of the posterior motor region (lobule VIIIb) in both groups, and significant degeneration of lobule IX and X in the healthy population, but not in cerebellar patients. Furthermore, we saw strong volumetric degeneration of functionally defined cerebellar regions associated with cerebral somatomotor function in both groups. Predominance of degeneration in the anterior lobe and lobule VI suggests impairment of motor function in both groups, while we suggest that the posterior shift of degeneration in cerebellar patients would be associated with relatively stronger impairment of higher motor function and cognitive function. Thus, these results may explain the specific symptomology associated with cerebellar degeneration in ageing and in cerebellar patients.