Skip to main content
NCBI home page
Human Brain Mapping logoLink to Human Brain Mapping
. 2014 Apr 28;35(10):5026–5039. doi: 10.1002/hbm.22529

Rapid automatic segmentation of the human cerebellum and its lobules (RASCAL)—Implementation and application of the patch‐based label‐fusion technique with a template library to segment the human cerebellum

Katrin Weier 1,2, Vladimir Fonov 1, Karyne Lavoie 3, Julien Doyon 4, D Louis Collins 1,2,✉
PMCID: PMC6869487  PMID: 24777876

Abstract

Reliable and fast segmentation of the human cerebellum with its complex architecture of lobes and lobules has been a challenge for the past decades. Emerging knowledge of the functional integration of the cerebellum in various sensori‐motor and cognitive‐behavioral circuits demands new automatic segmentation techniques, with accuracies similar to manual segmentations, but applicable to large subject numbers in a reasonable time frame. This article presents the development and application of a novel pipeline for rapid automatic segmentation of the human cerebellum and its lobules (RASCAL) combining patch‐based label‐fusion and a template library of manually labeled cerebella of 16 healthy controls from the International Consortium for Brain Mapping (ICBM) database. Leave‐one‐out experiments revealed a good agreement between manual and automatic segmentations (Dice kappa = 0.82). Intraclass correlation coefficients (ICC) were calculated to test reliability of segmented volumes and were highest (ICC > 0.9) for global measures (total and hemispherical grey and white matter) followed by larger lobules of the posterior lobe (ICC > 0.8). Further we applied the pipeline to all 152 young healthy controls of the ICBM database to look for hemispheric and gender differences. The results demonstrated larger native space volumes in men then women (mean (± SD) total cerebellar volume in women = 217 cm3 (± 26), men = 259 cm3 (± 29); P < 0.001). Significant gender‐by‐hemisphere interaction was only found in stereotaxic space volumes for white matter core (men > women) and anterior lobe volume (women > men). This new method shows great potential for the precise and efficient analysis of the cerebellum in large patient cohorts. Hum Brain Mapp 35:5026–5039, 2014. © 2014 Wiley Periodicals, Inc.

Keywords: cerebellum, segmentation, MRI, lobules

INTRODUCTION

The human cerebellum is a fascinating brain structure not only in terms of its anatomy, but also in terms of its integrating role in multiple processes and its focus in neuro‐psychiatric research. In comparison to the cerebrum, it is a small‐sized structure which occupies only about 10% of the whole brain volume; however it shows a complex architecture of tightly folded cortical grey matter comprising far more neurons (105,000 × 106) than the cerebrum [Andersen et al., 1992]. The grey matter covers a layer of white matter pathways in the so‐called foliae, branching from the medullary core where the deep cerebellar nuclei (i.e., the dentate, emboliform, globose, and fastigii) are embedded [Voogd, 2003]. On a larger scale the fissures between the foliae help to organize the cerebellum into the different lobules of each cerebellar hemisphere, connected via the midline vermis. A further anatomico‐functional distinction of the cerebellar hemispheres is often made in the anterior and posterior direction.

The cerebellum is known to play an important role in motor control and coordination. On the one hand it receives input from afferent sensory systems from the spinal cord, brainstem and cerebral cortical regions, and integrates these inputs to fine tune motor activity [Manto et al., 2013]. Efferent feedback projections on the other hand connect the cerebellar cortex via the deep cerebellar and thalamic nuclei with cortical regions such as the motor or prefrontal cortex [Stoodley and Schmahmann, 2010]. Disturbance of this loop through lesions, e.g. caused by ischemic stroke [Kase et al., 1993], multiple sclerosis [Compston and Coles, 2008; Davie et al., 1995] or neurodegenerative disorders [Klockgether, 2008], may lead to impairment of balance, gait and limb ataxia, dysarthria and occulomotor dysfunction. Furthermore the cerebellum has also been suggested to be part of non‐motor circuits responsible for cognitive and affective processes, especially attention, emotion, and behavior [Leiner et al., 1993; Levisohn et al., 2000; Timmann et al., 2010]. Anatomical and functional imaging studies have shown a topographic organization of the human cerebellum with activation of the anterior lobe and dorsolateral dentate during sensorimotor tasks, whereas posterior lobe and ventrocaudal dentate activation was observed during language, spatial, executive, working memory tasks and affective stimulation [Kuper et al., 2012; Stoodley and Schmahmann, 2010]. Schmahmann and Sherman [1998] subsumed the non‐motor symptoms of the cerebellum to the so called “cerebellar cognitive affective syndrome” (CCAS) and could further localize it to the “limbic cerebellum” which includes the vermis and fastigial nucleus [Schmahmann, 1991; Schmahmann et al., 2007]. Schizophrenic patients for example show a reduction in size of the vermis [Loeber et al., 2001]. However studies in autism have presented controversial results regarding vermal volume [Courchesne et al., 1988, 1994; Piven et al., 1997], which is to some extent easily explained through numerous and partly inconsistent existing nomenclatures of lobules; this is especially true for past definitions of the vermis. A major step to improve the analysis of the cerebellum was taken with the publication of the MR atlas of Schmahmann et al. [1999] showing a T1‐weighted scan of the cerebellum of a single male adult coregistered to the MNI305 template. Schmahmann et al. presented a consensus about the terminology on the basis of previously published data [Angevine et al., 1961; Bolk, 1906; Larsell and Jansen, 1972], helping to standardize the field. In addition to correct identification of cerebellar lobules, accurate volumetric measurements of the latter are essential to better investigate the role of the cerebellum in the healthy individual as well as in various diseases.

In the literature, different approaches have been described to segment the cerebellum, however only a few focused on the individual lobules [Schmahmann et al., 1999]. Manual segmentation of the cerebellum is a highly reliable [Loeber et al., 2001; Raz et al., 1998], but also, especially when considering the complex architecture, a far too time‐consuming method, which may take up to 20 h to segment one subject. This is impractical if large numbers of subjects are involved. Different semi‐automated segmentation methods have been introduced to reduce user input and at the same time preserve a high reliability, but most of these methods only refer to whole cerebellum segmentation or larger‐scale parcellations of the cerebellar hemispheres [Pierson et al., 2002; Weier et al., 2012]. To overcome the problems of intense user input and inter‐ and intra‐rater variability, automated, robust and accurate segmentation techniques become desirable. Concerning cortical or subcortical structures of the cerebrum, various different automated methods like region growing models [Chupin et al., 2007], appearance‐based models [Hu and Collins, 2007], as well as atlas or template‐warping techniques [Aljabar et al., 2009; Collins et al., 1995b; Fischl et al., 2002] have been developed. Regarding the automatic segmentation of the cerebellum in particular, Diedrichsen developed a high‐resolution, spatially unbiased atlas template of the human cerebellum (known as SUIT; [Diedrichsen, 2006]) that is based on the anatomy of 20 young healthy individuals and on the atlas of Schmahmann et al. [1999]. In comparison to a whole brain template atlas, this work improved the appearance of the cerebellum region, especially in terms of the individual fissure alignment. This atlas is used for atlas‐warping based cerebellar segmentation [Balsters et al., 2010]. Further improvement in the general field of atlas‐warping methods was achieved through the introduction of a template library rather than relying on a single template only, which already has been tested on smaller brain regions like the hippocampus [Barnes et al., 2008]. Whereas the structure of the hippocampus and its anatomical variations are relatively simple, Diedrichsen et al. [2009] discovered in their work on a probabilistic MR atlas of the cerebellar lobules, that the human cerebellum shows a high individual variability in healthy subjects (with more variability in the smaller, more caudally located lobules). This sometimes results in systematic mislabeling between lobules after the alignment to the MNI template space, thus underscoring an additional challenge in segmentation of the human cerebellum. Regarding segmentation of the hippocampus, Collins and Pruessner [2010] introduced an extension of the previous described automatic nonlinear image matching and anatomical labeling (ANIMAL) technique [Collins and Evans, 1997a; Collins et al., 1995b] with a template library incorporating a label fusion method to combine segmentations of the “n” most similar templates of the library yielding an optimal median Dice Kappa of 0.886. This approach was further adapted to a more local operating region with the so called patch‐based label‐fusion method, achieving comparable results in terms of Dice Kappa values [Coupe et al., 2011]. This work has inspired the methods presented here.

The aim of the current project was to develop a highly reliable automatic segmentation procedure to allow analyses of large patient cohorts and yet gain detailed information on the volumes of the individual cerebellar lobules. The manuscript describes the development and validation of a patch‐based label‐fusion strategy used with a library of cerebella from MRIs of sixteen healthy controls from the International Consortium for Brain Mapping database (ICBM; [Mazziotta et al., 1995]) data base, manually segmented using the protocol of Schmahmann et al. [1999]. The procedure was then applied to the MRIs of the 152 subjects that make up the ICBM data base [Mazziotta et al., 1995] to examine left–right and male–female differences.

METHODS

Subject Information, MRI Data, and Manual Labeling Procedure of the Template Library

The data, which were used for the template library of manually painted label volumes, were derived from a randomly selected group of sixteen young, neurological healthy individuals (7 women, mean age 25.5 ± 5.1 years) acquired in the context of the ICBM database (Mazziotta et al., 1995). All subjects underwent the same acquisition protocol including T1‐weighted (T1w), T2‐weighted (T2w) and Proton density‐weighted (PDw) sequences, but only T1w MRI data of these subjects were used in this study. Images were acquired at the Montreal Neurological Institute (MNI) on a Philips Gyroscan 1.5 T MRI scanner (Best, Netherlands) using a high‐resolution T1‐weighted, spoiled gradient echo sequence (sagittal acquisition, 140–160 contiguous 1‐mm thick slices, TR = 18 ms, TE = 10 ms, flip angle 30°, rectangular field of view of 256 × 204 mm2). The Ethics Committee of the Montreal Neurological Institute gave approval and informed consent was obtained from all participants. The MRI raw data was pre‐processed to correct for image nonuniformity [Sled et al., 1998] and then stereotaxically registered into the MNI coordinate system [Collins et al., 1994]. For further details see next paragraph on the automatic segmentation procedure.

Manual segmentation of the cerebellum was performed by two trained raters (K.L. and K.W.) using two software packages and the Schmahmann et al. atlas [Schmahmann et al., 1999] as an anatomical guide. K.L. used the Display software, a program developed at the MNI that permits simultaneous viewing and painting in all three planes (axial, sagittal, and coronal) to ensure 3D continuity of segmentations. K.W. used ITKsnap [Yushkevich et al., 2006] to verify and correct the segmentations. On each individual data set we defined twelve lobules for each cerebellar hemisphere according to the atlas of Schmahmann et al. [1999]. To avoid substructures, which may be to small for a reliable automatic segmentation, we decided not to segment separate vermal lobules. The vermal lobules VIIAf and VIIAt were included in their corresponding hemispherical lobules Crus I and II. Cerebellar hemispheres were divided according to the midsagittal plane of the cerebellum, verified on the coronal and axial planes.

Manual segmentation was initiated with the identification of cerebellar fissures on both hemispheres beginning from the mid‐sagittal plane (Fig. 1). Individual voxels within the center of the fissures were labeled with arbitrary color‐coded, numerical labels in all three planes in order to segment the lobules and folia as previously described [Schmahmann et al., 1999]. Second, each cerebellar lobule was labeled using the three cardinal planes to ensure a consistent 3D segmentation. We also added labels for cerebellar white matter as well as for the medial cerebellar peduncles (MCP) and superior cerebellar peduncles (SCP) on both sides. Note that the MCP label included the inferior cerebellar peduncles. The borders defining the cerebellar peduncles were chosen arbitrarily and are presented in Figure 2. To differentiate the MCP from the medullary core, the coronal plane intersecting the anterior‐most point of the vermis was chosen. The rostral border of the MCP was defined as the connection between the emerging trigeminal as well as vestibulocochlear nerve (both well‐defined anatomical landmarks) to the fourth ventricle (Fig. 2B). The cerebellar nuclei were not included in the manual segmentation, as the T1, T2, and PD contrasts did not enable their visualization. Whenever the term “grey matter” is used in this manuscript, it refers only to cortical grey matter.

Figure 1.

Figure 1

Open in a new tab

Manual segmentation was performed in different steps. The axial, sagittal and coronal planes (columns from left to right) ensure a consistent 3D segmentation (images in the upper row). First the fissures are traced and delineated according to the MRI‐atlas of Schmahmann et al. [1999] (middle row). These further help to separate each lobule, which are also labeled with color and numerical codes in a second step (lower row).

Figure 2.

Figure 2

Open in a new tab

(A) From left to right: Demonstration of the separating coronal plane between cerebellar peduncles and white matter core. The coronal plane is chosen which shows the anterior‐most point of the vermis, the dashed white line represents the axial plane cutting through the anterior‐most point. Next, the axial and sagittal plane showing the posterior limitation of the peduncles (non‐dashed line). (B) Display of anterior boundaries of the peduncles as the extension of the trigeminal (1) and vestibulocochlear (2) nerve (black dashed lines).

Automatic Segmentation Using the Template Library and Patch‐based Label Fusion Technique

The automatic segmentation algorithm presented here is an improved version of the methods published previously in Coupe et al. [2011], and incorporates ideas from [Landman et al., 2012]. The main differences from Coupe include a different strategy for intensity normalization, majority voting to account for multiple labels, and the use of nonlinear registration. The method consists of the following four stages: (i) image preprocessing, (ii) nonlinear registration, (iii) customized template library preparation and (iv) structure segmentation. All steps are described below. Manual segmentations of cerebellar subregions were performed using images that passed the pre‐processing stage.

  1. Image preprocessing stage. All steps applied to the T1w MRI scan P

    1. T1w MRI scans of the whole head were corrected for intensity non‐uniformity using the N3 algorithm [Sled et al., 1998]. The intensity range of the whole image is linearly normalized to the range 0.0–100.0 by histogram matching towards the ICBM 2009c template [Fonov et al., 2011; Mazziotta et al., 1995] in order to account for gross inter‐scanner and inter‐subject variability.

    2. Linear nine‐parameter registration to the ICBM 2009c template (shift, rotations and scale in x,y,z) was performed using normalized mutual information as part of the ANIMAL software [Collins and Evans, 1997b; Collins et al., 1995a].

    3. Nonlinear registration using the ANIMAL software and ICBM 2009c template was applied to obtain a rough brain mask, which was further used to perform a second linear intensity normalization using histogram matching, this time only inside the brain ROI. The brain mask is not used in any subsequent analysis.

    4. Next, a region of interest (ROI) around the subject's cerebellum was extracted as defined by the ICBM 2009c template.

    5. Adaptive patch‐based noise reduction was conducted [Manjon et al., 2010].

    6. Finally, a third linear histogram‐based intensity normalization was performed to reduce residual intensity differences within the cerebellar mask.

  2. Nonlinear registration stage with an output of forward (F) and inverse (I) non‐linear transformations.

    1. Nonlinear registration using the ANTs software [Avants et al., 2008] was performed in the cerebellum ROI toward the ICBM2009c template with the following parameters: number of iterations 100 × 100 × 50, cross‐correlation metric with four voxel local support radius, Gaussian regularization of gradient field with sigma of 1.0 and deformation field sigma of 1.0, Symmetric Normalization (SyN) transformation model with 0.25 time‐discretization step.

  3. Customized template library generation stage in order to create a subject‐specific segmentation library consisting of pairs of intensity template images Pt' and discrete segmentation images St'

    1. For each entry of the template library (Pt,Ft,It,St) the nonlinear transformation, mapping it to match the subject, was calculated by concatenating F and It :
      Tt=F·It.

    2. The template intensity image Pt was then warped using the inverse of Tt by 2nd order b‐spline interpolation, producing Pt'.

    3. Finally the template segmentation image St is warped by applying the inverse of Tt to each of the labels separately, using 2nd order b‐spline interpolation followed by voxel‐by‐voxel majority voting across all the label volumes, in order to obtain discrete labels, while reducing aliasing artefacts, producing St'.

  4. Segmentation stage to generate a discrete segmentation map S

    1. The algorithm described in [Coupe et al., 2011] is implemented using P as the input image and subject specific template library consisting of pairs (Pt',St'). To account for multiple discrete labels used in the template library, fuzzy segmentation results for each of the labels are compared in a majority voting scheme, assigning label with highest weight to a given voxel.

Template Library Construction

All scans that were manually segmented were processed with the 1st and 2nd stages of the algorithm described above. In addition, a left‐right flipped version of each anatomical sample was computed to double the number of samples in the database.

Segmentation Algorithm Parameter Optimization

To find the best parameters for the patch‐based segmentation (search area and patch size) leave‐one‐out experiments were performed, excluding one subject (and its left‐right flipped version) from the training library. The segmentation algorithm described above was applied and the overlap with the manual segmentation results was measured using generalized kappa metric [Crum et al., 2006]. The results are shown in Figure 3. A patch size of 9 × 9 × 9 voxels and search area of 5 × 5 × 5 voxels was chosen as providing the best trade‐off between segmentation accuracy (with best generalized dice kappa), while avoiding unnecessary processing.

Figure 3.

Figure 3

Open in a new tab

Segmentation algorithm parameter optimization. A number of combinations yielded equivalent results with median kappa of ∼0.84. A patch size of 9 × 9 × 9 voxels and a search area of 5 × 5 × 5 voxels (highlighted on the graph) were chosen as best parameters.

Further, intraclass correlation coefficient (ICC) model 3 (mixed‐effect model) was calculated for reliability measurements of volumes [Shrout and Fleiss, 1979].

Application to Healthy Control Data Base

The proposed automatic segmentation method was applied on 150 T1w data sets of the previously mentioned ICBM database that included 152 neurologically healthy subjects aged 18–44 years (mean 24.9 ± 4.9 years; 66 women; 129 right‐handed; ethnic background: 130 caucasian, 15 asian and eight others). Two subjects (one woman and one man) were excluded due to motion artifacts in their T1w images. Mean volumes and standard deviation for cerebellar grey matter lobules, medullary core and peduncles of both hemispheres were calculated for the whole cohort. A mixed model ANOVA was used to analyse gender (between‐subject) and hemispherical (inter‐subject) differences with cerebellar volumes as dependent variables and age as a covariate. All statistics were analysed using SPSS 21 (IBM, New York). P values of <0.05 were considered significant.

RESULTS

Manual Segmentations

Manual segmentation of the 16 randomly selected healthy cerebella of the ICBM data base revealed a mean total grey matter volume (CGV) of 117 cm3 ± 11 (mean ± standard deviation, range 101–139 cm3) and mean total white matter volume (CWV), including the peduncles, of 16 cm3 ± 1.8 (12–19 cm3) in native space. Mean values and standard deviations as well as the range for each individual lobule of either hemisphere are shown in Table 1. Comparison of both hemispheres did not reveal any significant difference except for crus II, which was significantly larger on the left side (mean = 6.6 cm3) on average compared to the right hemisphere (mean = 5.4 cm3; P = 0.05). A high anatomical variation was found between individual subjects resulting in low overlap of lobules, predominantly for the anterior lobe (lobules I–V) but also lobule X, Crus II and VIIIa bilaterally. Figure 4 displays nicely the areas of lowest overlap.

Table 1.

Demonstration of both manual and automatic segmented volumes for both hemispheres of n = 16 healthy subjects from the ICBM data base

Manual segmentation (n = 16) Automatic segmentation (n = 16) t ICC 95% CI
Minimum Maximum Mean Std. deviation Minimum Maximum Mean Std. deviation Lower bound Upper bound
TCV 112791 157621 133244 12243 113042 164236 134600 13564 2.29* 0.992 0.976 0.997
CGV 100743 138531 116948 10771 98686 144979 117834 12165 1.41 0.988 0.966 0.996
CWV 12048 19090 16296 1779 13055 19258 16766 1607 2.44* 0.946 0.844 0.981
Total hemi‐spherical GM Right 50586 69816 58730 5965 49301 72651 58918 6102 0.45 0.981 0.944 0.993
Left 49951 68714 58218 5090 49386 72327 58915 6101 1.56 0.974 0.926 0.991
Total hemi‐spherical WM Right 6167 9690 8173 921 6735 9887 8554 847 3.06** 0.914 0.753 0.970
Left 5880 9812 8123 926 6320 9371 8212 796 0.40 0.941 0.831 0.979
Anterior lobe Right 5138 9428 7442 1501 5694 9073 7384 905 −0.21 0.762 0.319 0.917
Left 5316 10034 7699 1207 5672 9588 7831 1024 0.47 0.656 0.016 0.880
Posterior lobe Right 44211 60388 51289 5224 43222 63579 51534 5415 0.73 0.984 0.953 0.994
Left 42192 60094 50519 5088 42939 63213 51085 5433 1.97 0.988 0.966 0.996
Lobule I,II Right 50 140 92 27 49 103 77 16 −3.8** 0.859 0.597 0.951
Left 46 194 83 39 48 103 78 16 −0.64 0.454 −0.562 0.809
Lobule III Right 425 1248 776 235 535 1078 777 181 0.03 0.773 0.350 0.921
Left 458 1053 757 179 647 1297 826 160 1.17 0.039 −1.750 0.664
Lobule IV Right 1179 4596 2665 817 1887 3589 2689 445 0.12 0.414 −0.677 0.795
Left 1298 4598 2710 1022 1870 3714 2856 578 0.69 0.645 −0.016 0.876
Lobule V Right 2620 5260 3909 943 2903 5360 3841 641 −0.34 0.685 0.098 0.890
Left 2289 6192 4149 1194 3038 4985 4071 510 −0.28 0.385 −0.761 0.785
Lobule VI Right 6725 12046 9278 1374 7882 11124 9175 862 −0.40 0.746 0.273 0.911
Left 6682 12083 8771 1435 7391 10969 8892 995 0.52 0.832 0.518 0.941
Crus I Right 10288 18319 13934 1916 11576 17382 14018 1650 0.33 0.912 0.749 0.969
Left 10122 18441 13163 2012 11271 16918 13763 1596 1.96 0.872 0.634 0.955
Crus II Right 3209 9128 5438 1671 3729 6816 5529 1020 0.25 0.626 −0.071 0.869
Left 4250 8846 6608 1570 4459 7025 5639 953 −2.80* 0.605 −0.132 0.862
Lobule VIIb Right 5875 12568 9710 1977 7716 11698 9523 1394 −0.46 0.703 0.149 0.896
Left 4998 12600 9173 2129 7197 12602 9476 1523 0.60 0.58 −0.203 0.853
Lobule VIIIa Right 3203 7269 5092 1009 4123 7204 5422 819 2.34* 0.897 0.704 0.964
Left 3571 7811 5088 1035 4036 6992 5457 750 2.00 0.8 0.426 0.930
Lobule VIIIb Right 2727 5443 3830 791 3074 5574 3737 646 −0.74 0.864 0.611 0.952
Left 1789 4829 3598 775 2906 4774 3791 573 1.12 0.656 0.016 0.880
Lobule IX Right 2585 5980 3833 986 3021 5867 3989 824 1.67 0.956 0.874 0.985
Left 2616 5827 3977 1010 2894 5386 3913 772 −0.61 0.942 0.835 0.980
Lobule X Right 97 274 174 63 101 194 141 26 −2.7* 0.674 0.066 0.886
Left 80 206 142 42 99 205 153 29 0.89 0.224 −1.220 0.729
MCP Right 1952 4462 3105 599 2324 3680 3121 410 0.16 0.813 0.464 0.935
Left 2155 4298 3135 615 2359 3655 3131 385 −0.02 0.634 −0.048 0.872
SCP Right 100 247 192 39 142 234 190 30 −0.19 0.484 −0.477 0.820
Left 117 243 191 36 147 241 189 28 −0.24 0.237 −1.183 0.734
Cerebellar peduncles Right 2188 4618 3296 597 2466 3885 3310 434 0.14 0.835 0.528 0.942
Left 2347 4490 3326 613 2506 3896 3320 402 −0.04 0.661 0.031 0.882
WM core Right 3979 6362 4876 698 4269 6049 5244 464 2.27 0.574 −0.220 0.851
Left 3533 6115 4797 740 3815 5599 4893 430 0.60 0.626 −0.070 0.869
Open in a new tab

Minimum, maximum, mean and standard deviation are displayed in native space. Paired t tests were calculated in order to test whether automatic volumes were larger (positive t value) or smaller (negative t value) than manual segmented volumes: P level: * = 0.05, ** = 0.01. Intraclass correlation coefficient (ICC‐model 3) as well as the 95% confidence interval (CI) are presented for comparison of methods.

Figure 4.

Figure 4

Open in a new tab

Display of major overlap of manual segmented lobules of the 16 healthy cerebella from the template library before nonlinear registration in sagittal (x = −5, 11, 32), axial (z = −20, −34, −52) and coronal (y = −56, −78, −82) planes. The different shades of gray correspond to the level of overlap and where white color shows 100% overlap.

Leave‐One‐Out Validation of Segmentation Method

Leave‐one‐out validation of the automatic labels against the gold standard manual labels yielded a generalized Dice kappa (k) of = 0.82. Individual kappa values are displayed in Figure 5 showing better results (k > 0.85) for the larger lobules VI, Crus I and IX as well as for the white matter core and medial cerebellar peduncles. The lowest kappa (k = 0.60) was found for the small lobule I_II bilaterally.

Figure 5.

Figure 5

Open in a new tab

Box‐ and whisker plots of Dice kappa of individual lobules. The boxplots display the median (bold line), the minimum (lower T‐line) and maximum (upper T‐line) as well as the first quartile (lower part of the box) and third quartile (upper part of the box) of kappa.

Direct comparison of mean volumes (see Table 1 for native space volumes and Table 1b in the Supporting Information for volumes in stereotaxic space) revealed an almost perfect agreement of the global measures like the total volumes TCV, CGV, and CWV with ICC values of 0.992, 0.988, and 0.946, respectively, as well as for hemispherical GM and WM total volumes (avg ICC of 0.98 and 0.93, respectively). Furthermore, strong to excellent agreement was found for the posterior and part of the anterior lobe volumes as well as the major lobules VI, Crus I, lobule VIII a + b and IX with (left+right) average ICC values of 0.789, 0.892, 0.845, 0.760, and 0.949, respectively. Depending on the anatomical variability and size of the remaining structures agreement varied between moderate to poor. Both manual and automatic segmentation roughly calculated the same lobule volumes. Only for TCV, CWV, right total WM and right lobule VIIIa did the automatic method slightly overestimate volumes. For right lobule I/II, left Crus II, and right lobule X the automatic segmentation slightly underestimated volumes.

Application to the ICBM Cohort

Automatic segmentation of 150 healthy subjects of the ICBM database was also performed. Minimum and maximum volume, mean values and standard deviation for the cerebellar lobules of the right and left hemisphere in standard stereotaxic space as well as native space are displayed in Table 2. These volumes were analyzed with paired t tests and an ANOVA.

Table 2.

Demonstration of mean cerebellar volumes (and standard deviation (SD)) in standard stereotaxic as well as native space for both hemispheres of n = 150 healthy subjects from the ICBM data base

Group statistics (n = 150) Standard stereotaxic space Native space
Cerebellar sub‐ structure Hemisphere Mean SD t Sig. (2‐tailed) Mean SD t Sig. (2‐tailed)
Total grey matter Right 82860 6256 0.626 0.532 61104 6571 0.442 0.659
Left 82415 6053 60772 6427
Anterior lobe grey matter Right 10711 1207 −1.513 0.131 7910 1159 −1.111 0.268
Left 10927 1262 8056 1115
Posterior lobe grey matter Right 72149 5721 1.020 0.308 53194 5754 0.722 0.471
Left 71488 5499 52716 5713
Lobule I,II Right 116 26 0.300 0.764 86 20 0.361 0.718
Left 115 29 85 22
Lobule III Right 1167 244 0.513 0.608 859 190 0.372 0.710
Left 1153 247 851 204
Lobule IV Right 3892 592 −1.722 0.086 2878 533 −1.415 0.158
Left 4017 668 2968 566
Lobule V Right 5536 794 −1.085 0.279 4087 689 −0.830 0.407
Left 5642 891 4153 675
Lobule VI Right 12596 1669 0.590 0.556 9269 1273 0.545 0.586
Left 12481 1715 9187 1336
Crus I Right 20179 2298 2.117 0.035 14878 2024 1.728 0.085
Left 19605 2395 14463 2136
Crus II Right 7672 1262 −2.307 0.022 5656 1028 −2.103 0.036
Left 8017 1327 5913 1089
Lobule VIIb Right 13202 1739 0.215 0.830 9731 1449 0.235 0.814
Left 13158 1776 9692 1428
Lobule VIIIa Right 7622 963 0.590 0.556 5626 872 0.477 0.634
Left 7559 886 5579 818
Lobule VIIIb Right 5297 646 1.466 0.144 3909 580 1.179 0.239
Left 5190 613 3831 562
Lobule IX Right 5385 804 1.261 0.208 3979 717 1.023 0.307
Left 5265 839 3893 744
Lobule X Right 197 46 −3.141 0.002 145 35 −3.035 0.003
Left 213 41 157 34
Total cerebellar white matter Right 12137 987 3.436 0.001 8962 1094 2.458 0.015
Left 11742 1008 8661 1020
Total cerebellar peduncles Right 4745 469 0.013 0.990 3508 501 0.067 0.947
Left 4744 490 3500 488
White matter core Right 7392 658 5.172 0.000 5454 663 4.019 0.000
Left 6997 665 5148 620
Open in a new tab

T tests show hemispherical asymmetries, with positive values for bigger volumes on the right side. Significant P level of < 0.05.

With paired t tests, we found hemispheric asymmetries of stereotaxic volumes in Crus I (right > left; t = 2.117, P = 0.035), Crus II (right < left; t = −2.307; P = 0.022), lobule X (right < left; t = −3.141, P = 0.002), the white matter core (right > left; t = 5.172, P < 0.001) and total WM (right > left; t = 3.436, P = 0.001).

Except for both sides of lobule IX, and the right anterior lobe, group t tests showed women had overall larger stereotaxically normalized cerebellar volumes than men. However this was not the case for native space volumes that showed a significantly smaller TCV (mean ± standard deviation: women = 217 ± 26, men = 259 ± 29; P < 0.001), CGV (women = 119 ± 12, men = 135 ± 12; P < 0.001) as well as CWV (women = 17 ± 2, men = 18 ± 2; P < 0.001) in women than men.

The mixed model ANOVA for the stereotaxic cerebellar volumes showed a significant hemisphere by gender interaction for the total WM volume (incl. peduncles) (F(1,147) = 7.3, P = 0.008, η p 2 = 0.047), pointing out a significantly larger hemispherical difference in men (right > left) than in women. Furthermore there was a significant group effect of gender (F(1,147) = 16.01, P < 0.001) as well as age (F(1,147) = 6.6, P = 0.01) on the total cerebellar WM volume. Another significant hemisphere by gender interaction was observed for anterior lobe volume (F(1,147) = 4.5, P = 0.036, η p 2 = 0.03) demonstrating a significant greater hemispherical difference in women (left > right) than men. However in this case no significant group effect was found for the anterior lobe volume. In the case of the total GM, the posterior lobe, and its individual lobules, no significant hemisphere by gender interaction was found. Analysis of main effects showed a significant effect of gender showing larger volumes in females for total GM (F(1,147) = 7.8, P = 0.006;), posterior lobe volume (F(1,147) = 8.8, P = 0.004), lobule VI (F(1,147) = 13.2, P < 0.001), Crus I (F(1,147) = 3.9, P = 0.049), and lobule VII B (F(1,147) = 6.6, P = 0.01), however there was no significant group effect for Crus II, lobule VIII A + B, IX and X (F < 1, P > 0.3).

The performance of the mixed model ANOVA on the native space volumes revealed no significant hemisphere by gender interaction for any of the analyzed volumes. However the main effect of gender was significant over all volumes (P ≤ 0.001), demonstrating larger volumes in men than women.

DISCUSSION

The present study describes the further development of a previously established automatic segmentation procedure for the hippocampus and its application to the human cerebellum using a template library of 16 manual segmented cerebella combined with a patch‐based label‐fusion technique in a large healthy control data base. In a leave‐one‐out validation procedure, the method yielded robust and accurate results for segmentation of the cerebellum, total WM, total GM and posterior lobe as demonstrated by high ICC and Kappa values. Automatic segmentation of the smaller anterior lobe lobules resulted in lower ICC and Kappa values, probably due to the smaller size of these structures, combined with their higher inter‐subject anatomical variability.

Manual segmentation of each lobule of the human cerebellum is a time consuming procedure, which may easily take from 20 to 35 h per subject. The manual segmented cerebella demonstrated that there is a large variability in how the fissures occur, separating each lobule. The highest variability was found for lobules I‐III, VIIIa and b, X as well as in the structure of crus II. Schmahmann et al. pointed out in their atlas an asymmetry of the ansoparamedian fissure, showing that Crus II laterally on the left side is not adjacent to lobule VIIb [Schmahmann et al., 1999]. Whereas Dietrichsen et al. [2009] reported in their work of a probabilistic atlas that only 3/50 hemispheres show this variant, we could find it in 25/32 hemispheres. These variabilities make automatic segmentation challenging.

To develop a technique that requires no user input (when segmenting a new subject) while obtaining highly accurate segmentations, we have chosen to use a template library of manual segmented cerebella together with a patch‐based label‐fusion technique. This technique was shown to be highly accurate in segmenting the hippocampus, which is a comparably small structure, embedded between CSF spaces and surrounding grey matter tissue, a situation similar to the cerebellar lobules [Collins and Pruessner, 2010; Coupe et al., 2010]. While other template‐based registration schemes use non‐linear registration to multiple templates before fusing labels [e.g., Collins and Pruessner, 2010; Wang and Yushkevich, 2013], this strategy did not improve segmentation accuracy with our data for the cerebellum. However it considerably extended the processing time, which is why we did not follow this approach.

When evaluated against the gold standard of manual labels of the cerebellar lobules, the technique presented here yielded a median Dice Kappa of 0.82. To the best of our knowledge, this is the most accurate result achieved for automatic segmentation of the cerebellar lobules. Previous commonly mis‐segmented areas like the venous sinus are no longer present. The technique works well, especially in segmenting the global volumes as well as in major lobules, e.g., like VI or VII. Still, the smaller substructures and lobules showing high anatomical variability remain challenging and leave room for improvement. One possibility for improvement would be most likely to increase the number of manual segmented templates in the library, a manually demanding task that will be completed in the future.

The literature predominantly reports on semi‐automated segmentation approaches, reporting results either only on partial manual labelling or results from an ensemble of merged lobules. In either case, the inter‐ and intra‐observer agreement is expected to be artificially high: semi‐automatic methods are biased towards the automatic boundaries as it takes effort to change them, and with a smaller surface to volume ratio, larger regions have proportionally less surface voxels to make errors on. Makris et al. introduced a semi‐automated method (using Cardviews and Freesurfer) with a detailed parcellation of the cerebellum [Makris et al., 2003, 2005]. Very similar to our approach, they follow the anatomical fissures (according to the MRI atlas of Schmahmann et al. [1999]), separating each individual lobule. This is however performed on the basis of an inflated cortical map or flattened surface generated by Freesurfer. They identify the outer trace of the fissure on the surface, and then trace the full fissure in 3D directly in the MRI. In contrast to our approach a final automated parcellation in the anterior–posterior direction is used, dividing the cerebellar hemispheres into medial and lateral zones as well as dividing the vermis from the hemispheres. In total, 64 regions (32 per hemisphere) are obtained. In their work from 2005 (Makris et al., 2005) two experts applied the mentioned method to 10 healthy cerebella. The inter‐ and intrarater reliability was tested using intraclass correlation coefficient of absolute agreement and showed an average ICC of 0.81 (range −0.23 to 0.99). In particular, the smaller regions lacked good reliability. Therefore, the authors chose to cluster the original parcellation units according to anatomical lobes and could show an improved ICC of 0.99 in anterior and posterior lobes and ICC of 0.74 in the flocculonodular lobes. The pattern of better performance in the larger scale lobules/lobes is similar to ours. In contrast to Makris method, Tiemeier et al. explicitly disclaim a detailed segmentation technique [Tiemeier et al., 2010]. They follow another semi‐automated method, starting with a rough parcellation approach for the cerebellum introduced by Pierson et al. [2002], as they believe that changes cannot reliably be detected in volumes of <10 mL. They report similar good reliability (ICC > 0.8) measures for their method. Okugawa et al. report of excellent (ICC > 0.95) reliability measures of their manual approach to trace the outer contours of the vermis and hemispheres [Okugawa et al., 2002]. As they identify only large compartments, their technique is less specific and far less time‐consuming. Since our method used to segment the template library is completely manual, and very time consuming (25–30 h per cerebellum), we decided not to redo the segmentations of multiple datasets to estimate intra‐ and interrater reliability measures for the templates.

The segmented cerebellar grey and white matter volumes in native space are generally in agreement with previous reported values ranging between 130 and 155cm3 for total cerebellar volume and 114–131cm3 for total cerebellar grey matter volume in native space [Diedrichsen et al., 2009; Dimitrova et al., 2008; Luft et al., 1998; Weier et al., 2012]. The differences are mainly explained by different segmentation techniques as well as whether white matter and/or peduncles were included.

When looking at the group as a whole, we found significant hemispheric asymmetry of the total white matter (right > left), but not for total grey matter, in line with Giedd et al. where they found no volumetric differences of the cerebellar hemispherical grey matter in the developing brain of a fairly large sample of children and adolescents [Giedd et al., 1996]. On a smaller scale we found asymmetries only in the following substructures: Crus I and the white matter core were larger on the right whereas Crus II and lobule X seem to be bigger on the left. Interpretation of these asymmetries should be made with caution, as they are relatively small in magnitude and Crus II as well as lobule X have been shown to be difficult to be segmented reliably (as demonstrated by a relatively low ICC). One may argue that part of the asymmetries (Crus I and white matter core) might be explained by lateralization of speech and motor function, which in our cohort would be the right hemisphere, because of the predominantly seen right‐handed subjects. Dietrichsen et al. [2009] describe in their small cohort of 20 subjects a left‐right asymmetry for the cerebellar hemispheres (left > right), for anterior lobules III and IV (right > left) as well as lobule VI (left > right). This was not the case in our cohort. These differences may be explained by the analysis of a different, larger sample here.

In terms of gender differences, the results show significantly larger native space volumes in men than women, which is in line with the literature [Diedrichsen et al., 2009; Dimitrova et al., 2006; Giedd et al., 2012]. Further significant gender‐by‐hemisphere interactions were found for white matter and the anterior lobe, pointing out a larger white matter core on the right for men compared to women and vice versa a larger anterior lobe GM volume in women when compared to men. However, these findings were only significant for the stereotaxic volumes but not for native space volumes. This may be due to the large inter‐subject variability. Cosgrove et al. [2007] nicely summarized findings on sexual dimorphism of brain structures in humans, an increasing focus of neuroimaging studies over the past 20 years. Tiermeier et al. [2010] have shown in their study on cerebellum development during childhood and adolescence that the growth of the cerebellum peaks during puberty in girls earlier than in boys (girls 11.8 years and boys 15.8 years). For adults, due to the various methods and different cohort sizes, there has been inconsistency of gender differences in brain structures in general and of these, only a small number have focused on the cerebellum. As for the cerebrum and its substructures, most of these studies showed larger native space volumes in men than in women and almost equal volumes when normalizing for intracranial cavity volume, height or weight [Diedrichsen et al., 2009; Dimitrova et al., 2006; Giedd et al., 1996; Good et al., 2001]. It is however worthwhile mentioning that in our cohort, the cerebellum in women is larger in relation to the overall brain and intracranial cavity volume when compared with men, a result supported by the data from Dimitrova et al. [2006].

Overall these gender differences in brain structures are still not well understood. Peri‐ and postnatal processes may play a role as the full size of the adult cerebellum is reached in puberty [Tiemeier et al., 2010]. Furthermore the proof of gender‐specific differences in gene expression, such as the specific sex chromosome‐linked gene XIST, which is highly represented in the female cerebellum (Vawter et al., 2004), are one of various attempts to explain these differences.

Because age has also a known effect on morphologic changes of the brain, especially the cerebellum [Raz et al., 2005], we corrected for age in our ANOVA models. Except for white matter, we did not find any significant effect of age on the brain volumes. This may be explained by the limited age range in the young ICBM cohort, in which the effect growth has stopped and atrophy associated with old age has not yet started [Raz et al., 2005].

In summary, the implementation of the patch‐based label‐fusion technique in combination with a template library shows great potential for the automatic segmentation of the cerebellum and its substructures, yielding more detailed results in a very efficient way compared to other atlases available. This will enable its application in MRI studies with large numbers of subjects and will help to better understand the human anatomy of the cerebellum and its disorders.

Supporting information

Supplementary Information Table

Click here for additional data file. (154.2KB, docx)

ACKNOWLEDGMENTS

The authors acknowledge the use of the MRI data from the ICBM study. All authors report no conflict of interest regarding this work.

REFERENCES

  1. Aljabar P, Heckemann RA, Hammers A, Hajnal JV, Rueckert D (2009): Multi‐atlas based segmentation of brain images: Atlas selection and its effect on accuracy. Neuroimage 46:726–738. [DOI] [PubMed] [Google Scholar]
  2. Andersen BB, Korbo L, Pakkenberg B (1992): A quantitative study of the human cerebellum with unbiased stereological techniques. J Comp Neurol 326:549–560. [DOI] [PubMed] [Google Scholar]
  3. Angevine JB, Mancall EL, Yakovlev PI (1961): The Human Cerebellum: An Atlas of Gross Topography in Serial Sections. Boston: Little, Brown. [Google Scholar]
  4. Avants BB, Epstein CL, Grossman M, Gee JC (2008): Symmetric diffeomorphic image registration with cross‐correlation: Evaluating automated labeling of elderly and neurodegenerative brain. Med Image Anal 12:26–41. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Balsters JH, Cussans E, Diedrichsen J, Phillips KA, Preuss TM, Rilling JK, Ramnani N (2010): Evolution of the cerebellar cortex: The selective expansion of prefrontal‐projecting cerebellar lobules. Neuroimage 49:2045–2052. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Barnes J, Foster J, Boyes RG, Pepple T, Moore EK, Schott JM, Frost C, Scahill RI, Fox NC (2008): A comparison of methods for the automated calculation of volumes and atrophy rates in the hippocampus. Neuroimage 40:1655–1671. [DOI] [PubMed] [Google Scholar]
  7. Bolk L (1906): Das Cerebellum der Sauegetiere. Jena: Bohn and Fischer. [Google Scholar]
  8. Chupin M, Mukuna‐Bantumbakulu AR, Hasboun D, Bardinet E, Baillet S, Kinkingnehun S, Lemieux L, Dubois B, Garnero L (2007): Anatomically constrained region deformation for the automated segmentation of the hippocampus and the amygdala: Method and validation on controls and patients with Alzheimer's disease. Neuroimage 34:996–1019. [DOI] [PubMed] [Google Scholar]
  9. Collins DL, Evans A (1997a): ANIMAL: Validation and applications of non‐linear registration‐bsed segmentations. Int J Pattern Recogn Artif Intell 11:1271. [Google Scholar]
  10. Collins DL, Evans AC (1997b): Animal: Validation and applications of nonlinear registration‐based segmentation. Int J Pattern Recogn Artif Intell 11:1271–1294. [Google Scholar]
  11. Collins DL, Pruessner JC (2010): Towards accurate, automatic segmentation of the hippocampus and amygdala from MRI by augmenting ANIMAL with a template library and label fusion. Neuroimage 52:1355–1566. [DOI] [PubMed] [Google Scholar]
  12. Collins DL, Neelin P, Peters TM, Evans AC (1994): Automatic 3D intersubject registration of MR volumetric data in standardized Talairach space. J Comput Assist Tomogr 18:192–205. [PubMed] [Google Scholar]
  13. Collins DL, Evans AC, Holmes C, Peters TM (1995a): Automatic 3D segmentation of neuro‐anatomical structures from MRI. Inform Process Med Imaging 3:139–152. [Google Scholar]
  14. Collins DL, Holmes CJ, Peters TM, Evans AC (1995b): Automatic 3‐D model‐based neuroanatomical segmentation. Hum Brain Mapp 3:190–208. [Google Scholar]
  15. Compston A, Coles A (2008): Multiple sclerosis. Lancet 372:1502–1517. [DOI] [PubMed] [Google Scholar]
  16. Cosgrove KP, Mazure CM, Staley JK (2007): Evolving knowledge of sex differences in brain structure, function, and chemistry. Biol Psychiatry 62:847–855. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Coupe P, Manjon JV, Fonov V, Pruessner J, Robles M, Collins DL (2010): Nonlocal patch‐based label fusion for hippocampus segmentation. Med Image Comput Comput Assist Interv 13 (Part 3):129–136. [DOI] [PubMed] [Google Scholar]
  18. Coupe P, Manjon JV, Fonov V, Pruessner J, Robles M, Collins DL (2011): Patch‐based segmentation using expert priors: Application to hippocampus and ventricle segmentation. Neuroimage 54:940–954. [DOI] [PubMed] [Google Scholar]
  19. Courchesne E, Yeung‐Courchesne R, Press GA, Hesselink JR, Jernigan TL (1988): Hypoplasia of cerebellar vermal lobules VI and VII in autism. N Engl J Med 318:1349–1354. [DOI] [PubMed] [Google Scholar]
  20. Courchesne E, Saitoh O, Yeung‐Courchesne R, Press GA, Lincoln AJ, Haas RH, Schreibman L (1994): Abnormality of cerebellar vermian lobules VI and VII in patients with infantile autism: identification of hypoplastic and hyperplastic subgroups with MR imaging. AJR Am J Roentgenol 162:123–130. [DOI] [PubMed] [Google Scholar]
  21. Crum WR, Camara O, Hill DL (2006): Generalized overlap measures for evaluation and validation in medical image analysis. IEEE Trans Med Imaging 25:1451–1461. [DOI] [PubMed] [Google Scholar]
  22. Davie CA, Barker GJ, Webb S, Tofts PS, Thompson AJ, Harding AE, McDonald WI, Miller DH (1995): Persistent functional deficit in multiple sclerosis and autosomal dominant cerebellar ataxia is associated with axon loss. Brain 118 (Part 6):1583–1592. [DOI] [PubMed] [Google Scholar]
  23. Diedrichsen J (2006): A spatially unbiased atlas template of the human cerebellum. Neuroimage 33:127–138. [DOI] [PubMed] [Google Scholar]
  24. Diedrichsen J, Balsters JH, Flavell J, Cussans E, Ramnani N (2009): A probabilistic MR atlas of the human cerebellum. Neuroimage 46:39–46. [DOI] [PubMed] [Google Scholar]
  25. Dimitrova A, Zeljko D, Schwarze F, Maschke M, Gerwig M, Frings M, Beck A, Aurich V, Forsting M, Timmann D (2006): Probabilistic 3D MRI atlas of the human cerebellar dentate/interposed nuclei. Neuroimage 30:12–25. [DOI] [PubMed] [Google Scholar]
  26. Dimitrova A, Gerwig M, Brol B, Gizewski ER, Forsting M, Beck A, Aurich V, Kolb FP, Timmann D (2008): Correlation of cerebellar volume with eyeblink conditioning in healthy subjects and in patients with cerebellar cortical degeneration. Brain Res 1198:73–84. [DOI] [PubMed] [Google Scholar]
  27. Fischl B, Salat DH, Busa E, Albert M, Dieterich M, Haselgrove C, van der Kouwe A, Killiany R, Kennedy D, Klaveness S, et al. (2002): Whole brain segmentation: automated labeling of neuroanatomical structures in the human brain. Neuron 33:341–355. [DOI] [PubMed] [Google Scholar]
  28. Fonov V, Evans AC, Botteron K, Almli CR, McKinstry RC, Collins DL, Brain Development Cooperative G (2011): Unbiased average age‐appropriate atlases for pediatric studies. Neuroimage 54:313–327. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Giedd JN, Snell JW, Lange N, Rajapakse JC, Casey BJ, Kozuch PL, Vaituzis AC, Vauss YC, Hamburger SD, Kaysen D, et al. (1996): Quantitative magnetic resonance imaging of human brain development: Ages 4–18. Cereb Cortex 6:551–560. [DOI] [PubMed] [Google Scholar]
  30. Giedd JN, Raznahan A, Mills KL, Lenroot RK (2012): Review: Magnetic resonance imaging of male/female differences in human adolescent brain anatomy. Biol Sex Differ 3:19. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Good CD, Johnsrude I, Ashburner J, Henson RN, Friston KJ, Frackowiak RS (2001): Cerebral asymmetry and the effects of sex and handedness on brain structure: A voxel‐based morphometric analysis of 465 normal adult human brains. Neuroimage 14:685–700. [DOI] [PubMed] [Google Scholar]
  32. Hu S, Collins DL (2007): Joint level‐set shape modeling and appearance modeling for brain structure segmentation. Neuroimage 36:672–683. [DOI] [PubMed] [Google Scholar]
  33. Kase CS, Norrving B, Levine SR, Babikian VL, Chodosh EH, Wolf PA, Welch KM (1993): Cerebellar infarction. Clinical and anatomic observations in 66 cases. Stroke 24:76–83. [DOI] [PubMed] [Google Scholar]
  34. Klockgether T (2008): The clinical diagnosis of autosomal dominant spinocerebellar ataxias. Cerebellum 7:101–105. [DOI] [PubMed] [Google Scholar]
  35. Kuper M, Thurling M, Maderwald S, Ladd ME, Timmann D (2012): Structural and functional magnetic resonance imaging of the human cerebellar nuclei. Cerebellum 11:314–324. [DOI] [PubMed] [Google Scholar]
  36. Landman BA, Ribbens A, Lucas B, Davatzikos C, Avants B, Ledig C, Ma D, Rueckert D, Vandermeulen D, Maes F, et al. (2012): MICCAI 2012 workshop on multi‐atlas labeling In: Landman BA, Warfield SK, editors. Nice. CreateSpace Independent Publishing Platform. pp. 63–66, 91–94. [Google Scholar]
  37. Larsell O, Jansen J (1972): The Comparative Anatomy and Histology of the Cerebellum: The Human Cerebellum, Cerebellar Connections, and Cerebellar Cortex. Minneapolis: University of Minnesota Press. [Google Scholar]
  38. Leiner HC, Leiner AL, Dow RS (1993): Cognitive and language functions of the human cerebellum. Trends Neurosci 16:444–447. [DOI] [PubMed] [Google Scholar]
  39. Levisohn L, Cronin‐Golomb A, Schmahmann JD (2000): Neuropsychological consequences of cerebellar tumour resection in children: Cerebellar cognitive affective syndrome in a paediatric population. Brain 123 (Part 5):1041–1050. [DOI] [PubMed] [Google Scholar]
  40. Loeber RT, Cintron CM, Yurgelun‐Todd DA (2001): Morphometry of individual cerebellar lobules in schizophrenia. Am J Psychiatry 158:952–954. [DOI] [PubMed] [Google Scholar]
  41. Luft AR, Skalej M, Welte D, Kolb R, Burk K, Schulz JB, Klockgether T, Voigt K (1998): A new semiautomated, three‐dimensional technique allowing precise quantification of total and regional cerebellar volume using MRI. Magn Reson Med 40:143–151. [DOI] [PubMed] [Google Scholar]
  42. Makris N, Hodge SM, Haselgrove C, Kennedy DN, Dale A, Fischl B, Rosen BR, Harris G, Caviness VS Jr, Schmahmann JD (2003): Human cerebellum: Surface‐assisted cortical parcellation and volumetry with magnetic resonance imaging. J Cogn Neurosci 15:584–599. [DOI] [PubMed] [Google Scholar]
  43. Makris N, Schlerf JE, Hodge SM, Haselgrove C, Albaugh MD, Seidman LJ, Rauch SL, Harris G, Biederman J, Caviness VS Jr., et al. (2005): MRI‐based surface‐assisted parcellation of human cerebellar cortex: An anatomically specified method with estimate of reliability. Neuroimage 25:1146–1160. [DOI] [PubMed] [Google Scholar]
  44. Manjon JV, Coupe P, Marti‐Bonmati L, Collins DL, Robles M (2010): Adaptive non‐local means denoising of MR images with spatially varying noise levels. J Magn Reson Imaging 31:192–203. [DOI] [PubMed] [Google Scholar]
  45. Manto M, Gruol DL, Schmahmann JD, Koibuchi N, Rossi F (2013): Handbook of the Cerebellum and Cerebellar Disorders. New York: Springer. [Google Scholar]
  46. Mazziotta JC, Toga AW, Evans A, Fox P, Lancaster J (1995): A probabilistic atlas of the human brain: Theory and rationale for its development. The International Consortium for Brain Mapping (ICBM). Neuroimage 2:89–101. [DOI] [PubMed] [Google Scholar]
  47. Okugawa G, Sedvall G, Nordstrom M, Andreasen N, Pierson R, Magnotta V, Agartz I (2002): Selective reduction of the posterior superior vermis in men with chronic schizophrenia. Schizophr Res 55:61–67. [DOI] [PubMed] [Google Scholar]
  48. Pierson R, Corson PW, Sears LL, Alicata D, Magnotta V, Oleary D, Andreasen NC (2002): Manual and semiautomated measurement of cerebellar subregions on MR images. Neuroimage 17:61–76. [DOI] [PubMed] [Google Scholar]
  49. Piven J, Saliba K, Bailey J, Arndt S (1997): An MRI study of autism: The cerebellum revisited. Neurology 49:546–551. [DOI] [PubMed] [Google Scholar]
  50. Raz N, Dupuis JH, Briggs SD, McGavran C, Acker JD (1998): Differential effects of age and sex on the cerebellar hemispheres and the vermis: A prospective MR study. AJNR Am J Neuroradiol 19:65–71. [PMC free article] [PubMed] [Google Scholar]
  51. Raz N, Lindenberger U, Rodrigue KM, Kennedy KM, Head D, Williamson A, Dahle C, Gerstorf D, Acker JD (2005): Regional brain changes in aging healthy adults: General trends, individual differences and modifiers. Cereb Cortex 15:1676–1689. [DOI] [PubMed] [Google Scholar]
  52. Schmahmann JD (1991): An emerging concept. The cerebellar contribution to higher function. Arch Neurol 48:1178–1187. [DOI] [PubMed] [Google Scholar]
  53. Schmahmann JD, Sherman JC (1998): The cerebellar cognitive affective syndrome. Brain 121 (Part 4):561–579. [DOI] [PubMed] [Google Scholar]
  54. Schmahmann JD, Doyon J, McDonald D, Holmes C, Lavoie K, Hurwitz AS, Kabani N, Toga A, Evans A, Petrides M (1999): Three‐dimensional MRI atlas of the human cerebellum in proportional stereotaxic space. Neuroimage 10(3, Part 1):233–260. [DOI] [PubMed] [Google Scholar]
  55. Schmahmann JD, Weilburg JB, Sherman JC (2007): The neuropsychiatry of the cerebellum—Insights from the clinic. Cerebellum 6:254–267. [DOI] [PubMed] [Google Scholar]
  56. Shrout PE, Fleiss JL (1979): Intraclass correlations—Uses in assessing rater reliability. Psychol Bull 86:420–428. [DOI] [PubMed] [Google Scholar]
  57. Sled JG, Zijdenbos AP, Evans AC (1998): A nonparametric method for automatic correction of intensity nonuniformity in MRI data. IEEE Trans Med Imaging 17:87–97. [DOI] [PubMed] [Google Scholar]
  58. Stoodley CJ, Schmahmann JD (2010): Evidence for topographic organization in the cerebellum of motor control versus cognitive and affective processing. Cortex 46:831–844. [DOI] [PMC free article] [PubMed] [Google Scholar]
  59. Tiemeier H, Lenroot RK, Greenstein DK, Tran L, Pierson R, Giedd JN (2010): Cerebellum development during childhood and adolescence: A longitudinal morphometric MRI study. Neuroimage 49:63–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Timmann D, Drepper J, Frings M, Maschke M, Richter S, Gerwig M, Kolb FP (2010): The human cerebellum contributes to motor, emotional and cognitive associative learning. A review. Cortex 46:845–857. [DOI] [PubMed] [Google Scholar]
  61. Vawter MP, Evans S, Choudary P, Tomita H, Meador‐Woodruff J, Molnar M, Li J, Lopez JF, Myers R, Cox D, et al. (2004): Gender‐specific gene expression in post‐mortem human brain: Localization to sex chromosomes. Neuropsychopharmacology 29:373–384. [DOI] [PMC free article] [PubMed] [Google Scholar]
  62. Voogd J (2003): The human cerebellum. J Chem Neuroanat 26:243–252. [DOI] [PubMed] [Google Scholar]
  63. Wang H, Yushkevich PA (2013): Groupwise segmentation with multi‐atlas joint label fusion. Med Image Comput Comput Assist Interv 16 (Part 1):711–718. [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Weier K, Beck A, Magon S, Amann M, Naegelin Y, Penner IK, Thurling M, Aurich V, Derfuss T, Radue EW, et al. (2012): Evaluation of a new approach for semi‐automatic segmentation of the cerebellum in patients with multiple sclerosis. J Neurol 259:2673–2680. [DOI] [PubMed] [Google Scholar]
  65. Yushkevich PA, Piven J, Hazlett HC, Smith RG, Ho S, Gee JC, Gerig G (2006): User‐guided 3D active contour segmentation of anatomical structures: Significantly improved efficiency and reliability. Neuroimage 31:1116–1128. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Information Table

Click here for additional data file. (154.2KB, docx)

Articles from Human Brain Mapping are provided here courtesy of Wiley

RESOURCES