Using 7 Tesla MRI to image the Globus Pallidus interna: striving towards brain atlases that account for inter-subject variability
26 EC, Research Project 1
Lindsey Crown UvA ID: 10234217 05-01-2012 to 31-07-2012
Supervisor: Max Keuken Co-assessor: Birte Forstmann
Spinoza Center, University of Amsterdam
MSc in Brain and Cognitive Sciences, University of Amsterdam
Abstract
The globus pallidus interna (GPi) is a central figure in action selection processing in the basal ganglia. It is also a target of deep brain stimulation for the treatment of motor disorders like Parkinson’s disease. It is a small structure whose size and location can greatly vary between patients. As such, its inclusion in brain atlases that are flexible and can account for this variability is of vital importance. In this study we attempt to visualize and manually segment the GPi in ultra-high resolution 7Tesla MRI in hopes of creating such a map as well as calculate volume estimates of the structure. However, the GPi was not visible using traditional T1-weighted MP2RAGE scans and a low level of inter-rater agreement limited the confidence in volumetric estimates made in T2-weigthed FLASH images. As a result, we suggest that these sequences alone are inadequate for accurately
visualizing the GPi and suggest that further research investigate T2* and SWI as a potential sequences for GPi atlasing.
Keywords: Basal ganglia
Globus pallidus interna Brain atlas
7 Tesla magnetic resonance imaging manual segmentation
Introduction
It is well understood that the anatomy of the human brain is highly complex, and though certain areas have been systematically mapped out, other, more difficult to view areas, have not. The basal ganglia (BG) are a collection of small subcortical nuclei that work together with the frontal cortex to execute goal-directed motor behaviors through cognitive, motor, and limbic circuits1. The primary nuclei of the basal ganglia are the
striatum, comprised of the caudate and putamen, the globus pallidus (GP), with both internal and external segments, the substantia nigra (SN), and the subthalamic nucleus (STN)2. The striatum, together with the STN, serve as the major input nuclei for the BG
while the globus pallidus interna (GPi) and the SN are the major output nuclei2.
It is hypothesized that within the basal ganglia there are three major information pathways that together function to coordinate action selection3. When a voluntary
the motor cortex to the GPi through the cortico-STN-pallidal ‘hyperdirect’ pathway. This signal extensively excites the GPi, resulting in the inhibition of large thalamic and
cortical areas associated with the desired motor program and other competing programs. Then, another signal is sent via a ‘direct’ pathway to the GPi, first arriving at the basal ganglia through the putamen and transmitting an inhibitory signal to a certain area of the GPi. In turn, these GPi neurons disinhibit their specific thalamic connections and release only the selected motor program. Finally, a third signal is sent which also enters the basal ganglia at the putamen but then moves through the GPe as part of a longer,
‘indirect’ pathway. After traveling the longest pathway, this signal is the last to reach the GPi. It then activates GPi neurons which in turn suppress their thalamic targets
extensively, effectively stopping the motor program3.
It is hypothesized that through these three pathways only the appropriate motor program is selected and then terminated at the correct time while other competing programs mediated by the GPi are inhibited. This harmony of coordinated action in the basal ganglia seems to be severely disrupted in Parkinson’s disease. Parkinson’s disease is characterized by the progressive loss of populations of dopaminergic neurons in the sustantia nigra pars compacta (SNc). This results in regional losses of striatal dopamine, primarily in posterior areas of the putamen4. Reduced dopamine input from the SNc to
the striatum would have the effect of facilitating neurons involved in the indirect pathway and reducing the activity of direct pathway neurons5. Increased inhibition from ‘indirect’
striatal neurons to the GPe would disinhibit the STN and cause inhibitory output neurons in the GPi and SNr to fire at above normal rates. Decreased activation in the direct
pathway would also reduce inhibitory influences on the GPi and SNr, and lead to excessive amounts of inhibitory basal ganglia output5.
It is believed that this over-excitation of the GPi and SNr and the corresponding excessive amount of inhibitory output from the basal ganglia is what disrupts normal action selection processing and results in abnormal behavior being transmitted back to the motor cortex. This results in tremor, rigidity, akinesia, and gait disturbances, the major motor symptoms of Parkinson’s disease6.
To treat these movement disorders, deep brain stimulation (DBS) has been highly successful. DBS is a surgical procedure whereby electrodes are implanted into a specific part of the brain and generate an electrical pulse that influences the firing pattern of neurons in the target area7. Currently the GPi and the STN are the two primary target
structures for this movement-related DBS treatment8. While recently the STN has been
the leading structure for electrode implantation, the GPi has been gaining popularity again9. It is suspected that GPi DBS may carry a lower risk of adverse cognitive and
behavioral symptoms while providing the same motor control advantages. Stimulation of the STN often has the propensity to spread current to other brain regions, while the GPi’s larger size, compared to the STN, might be a better safeguard against this8.
Brain Atlases and their Drawbacks
The use of brain atlases is of critical importance when preforming surgeries like DBS. However, brain atlases have been historically limited in their applicability as they were based on a limited number of brains and the large individual variance in anatomy10.
one or only a few individual post mortem specimens and were generally printed11,12.
Because of the difference between the dimensionality of a 2D book and a 3D Magnetic Resonance Image (MRI) as well as the difference between standard anatomy and
individual anatomy, most of the translation from atlas to actual brain had to happen in the mind of the surgeon13. This made application of the atlas to real patients’ images tedious
and prone to errors. Today, modern atlases have sought to overcome these potential sources of error by developing computerized systems whereby a patient’s anatomical MRI is warped onto a standard brain in stereotactic space, such as in the well-established Talairach atlas12. However there are still a number of problems to be reconciled with
these modern atlas approaches.
One great challenge to be faced when using an atlas is inter-individual anatomical variation. Numerous studies have illuminated the great variably in brains14,15,16. As Toga
et al. notes “Even without pathologies, brain structures vary between individuals not only in shape and size, but also in their orientations relative to each other.”10. This problem is
most apparent in the cortex due to the high degree of variability in intricate sulci and gyri pattern between subjects16. Previous studies on Brodmann’s areas (BA) 44 and 45
(Broca’s area) have demonstrated that it is not possible to define the location of a cytoarchitectonic precisely based on macroscopic landmarks because sulci pattern and borders of Broca’s area are highly variable between subjects17. Another study in 2005 by
Uylings et al. points out that, “the simple use of Brodmann cortical areas derived from the Talairach atlas can lead to erroneous results in the specification of pertinent BA. This in turn can further lead to wrong hypotheses on brain system(s) involved in normal functions or in specific brain disorders.” 16.
This problem is not unique to the cortex, however, and extends to subcortical structures as well. Forstmann et al.’s 2010 study on the STN found that after manually segmenting the STN, the highest overlap across participants was only 45.83% for the left STN and 50% for the right18 indicating that both the volume and the precise localization
of the STN vary to a large degree between individuals.
Another important factor to consider whilst evaluating some of the widely available atlases (19,11,20,12,21) is rather they were done in-vivo or ex-vivo. This is relevant
as post-mortem brains are subject to non-linear shrinkage and the relation of structures to each other can change dramatically when brains are removed from the skull or subjected to histological procedures22. Many widely used brain atlases are still based on individual
post-mortem data 21,23. For example, the widely-used Talairach atlas12 was originally
based on one hemisphere of a single, post-mortem brain of a 60 year-old women10.
Furthermore, this atlas is rather sparse due to a variable inter-slice distance between two and five mm. 13 The Talairach atlas, however, has since been updated over the years and
is now registered in 3D space and incorporates a range of imaging datasets, allowing it to continue to be one of the standards in brain-mapping.
The changes made to the Talairach atlas represent a dramatic change in the way brain atlases are constructed. Probabilistic, MRI-based brain maps have been rising in number and carry the advantage of in-vivo 3D axis registration, and statistical structural localization based on a large population of brains12,23-25.
Yet while probabilistic maps can be a highly powerful tool to account for
population variability, the majority of these maps were created from images acquired on MRI machines with field strengths of 1.5T and below and suffer from poor
signal-to-noise ratios (SNR). As a result they tend to have large voxel sizes and oftentimes very fuzzy borderlines between structures that makes creating a very precise map difficult26.
Even with up to 100-200 slices, with 2mm section thickness, this is still too low of a resolution to view certain small structures like those found in the basal ganglia10. In fact,
as a result of this resolution limitation, many major atlases simply do not include the GPi at all(12,23-25).
Visualizing the GPi
When attempting to visualize the GPi with MRI, signal strength, contrast, and sequence choice are very important factors to consider. As previously mentioned, due its size, similar composition, and close proximity to the GPe, the GPi can be difficult to delineate. MRI precision, measured by SNR and contrast-to-noise ratio (CNR) has been shown to increase nearly linearly with field strength1,27,2,28. Ultra-high field strength 7T
MRI increases signal-to-noise ratio (SNR) and improves image contrast, allowing for identification of structures that cannot be fully visualized at 1.5 or 3T2,29. In one
comparative study by Metcalf et al. (2010) illustrating the improved CNR in higher field strengths, 7T MRI was compared to 1.5T MRI in lesion detection in patients with multiple sclerosis. 3,30 At 7T nearly twice as many lesions were detected than at 1.5T 3,30.
Another important element to optimizing GPi visibility is selecting the right contrast and pulse sequence. In general, T1-weighted images are strong for structural contrast in anatomical images and elicit the strongest signal from white matter, a medium signal from grey matter and very little signal from fluid. T2 images are generated using spin-echo based pulse sequences and appear brightest in fluids, intermediate in grey
matter, and darkest in white matter. T2* -weighted contrasts are sensitive to the amount of deoxygenated hemoglobin present and is therefore commonly used in fMRI 4,31.
For the difficult task of parsing out structures and determining their location in the brain, either manual or automated segmentation is possible. Most mapping projects utilize automated segmentation techniques, which warp individual images onto a stereotactic, averaged brain. Various algorithms are then used to clarify tissues and labeled structure boundaries(5,32-36). This can be a complex process because many
structures that maybe anatomically distinct can have very similar tissue composition and signal properties5,33.
Manual segmentation between two inter-raters is an alternative to automated segmentation. Carmichael et al (2005) demonstrated it to be more accurate and more sensitive to individual differences in the hippocampus than automated segmentation6,37.
Specifically for deep brain grey nuclei like the caudate nucleus, thalamus, and GPi, segmentation can be especially difficult. This is likely due to their small size and the heterogeneity and inconsistency of their borders with the surrounding white matter7,34.
Because the GPi is divided only by a thin lamina from the GPe and is morphologically very similar, it is especially difficult to segment8,38. These challenges to localization and
segmentation are only compounded when attempting to factor in inter-individual
variation. When assessing the accuracy of surgical ablations for the GPi for treatment of Parkinson’s disease, a study by Reich et al. found that standard atlases being used to determine the GPi’s distance from the midline in preoperative planning largely do not take into account the individual variation in this measurement9,39.
GPi with T1-weighted MP2RAGE and T2-weighted gradient echo FLASH 3D
sequences. We will then use these high-resolution images to manually segment the GPi. By combining these previously individually successfully techniques, we will calculate volume estimates of the GPi, compare these estimates to previous study’s estimates and evaluate the success of our process and calculations.
Materials and Methods:
Participants
30 young healthy participants (16 male and 14 female) with a mean age of 24.2 and standard deviation of 2.4 years were scanned as part of a collaborative project with the Max Plank Institute for human Cognitive and Brain Sciences in Leipzig, Germany. Subjects were matched for age and gender. All participants were right-handed, as confirmed by the Edinburgh Inventory 40. All participants had normal or
corrected-to-normal vision and had no history of neurological or psychiatric disorders. The local ethics committee approved the study and all subjects gave their informed written consent before scanning. The participants received a monetary reward for their participation.
Data Acquisition
Data was acquired on a 7T Magnetom MRI system (Siemens, Erlangen, Germany) using a 24-channel head array Nova coil (NOVA Medical Inc., Wilmington, MA, USA). Whole brain and slab images covering the basal ganglia were acquired using a T1-weighted MP2RAGE scan sequence (voxel resolution 0.7mm isotropic, 240 sagittal
slices, TR=5000ms, T11/T12= 900/2750ms) with an acquisition time of 9:07 minutes. A T2-weighted multi-echo spoiled 3 dimensional (3D) gradient echo FLASH sequence (voxel size 0.5mm isotropic, 128 slices, TR=41 ms, TE1/TE2/TE3=11.22/20.39/29.57, flip angle=13°) with an acquisition time of 17:18min was also acquired.
Manual segmentation of the GPi
Manual segmentation was preformed on FSL 4.1.1 viewer
(http://www.fmrib.ox.ac.uk/fsl/) by two independent raters. Manual segmentation was preformed one subject, one hemisphere at a time using exclusively the before-mentioned spoiled 3D gradient echo FLASH sequences. This was done by loading all three echoes of the flash sequence into FSLviewer simultaneous and manually adjusting the contrast values so that GPi borders were as visible as possible. A 3D image of the structure was then acquired by going back and forth through all three echoes and tracing all visible borders of the structure. (See Figure 1) Each rater preformed this process independently and upon completion the individual masks were added and thresholded such that only voxels labeled as being part of the GPi by both raters were included in the final image. Calculations of inter-rater reliability, hemispheric differences, and gender differences were done using Microsoft Excel (2011) version 14.1.4 for Mac.
Figure 1: A sample image showing manual segmentation using FSL viewer. All three echoes are loaded into the program and a 3D image is created by tracing GPi border in each slice.
Results:
The mean volume of the inter-rater GPi mask was 402.90mm3 for the left and 407.59mm3 for the right with a standard deviation of 69.56mm3 and 67.21mm3 respectively. There was no significant difference between hemispheres in either of the raters individually or in the final inter-rater masks (p=0.46). One rater did find a gender difference in GPi volumes with males being significantly larger (p=0.05), however this effect was not observed in the final rater masks (p=0.58). The overall average inter-rater agreement was 63.55% with a standard deviation of 7.34%.
Segmentation was preformed exclusively with FLASH sequences, as visualization of the GPi with MP2RAGE was not possible. (see Figure 2) Although it was possible to see the GPi in the FLASH sequence, it still proved difficult and a significant training effect was found when comparing the first half of masks with the second half in which the first half was significantly larger than the second half (p=0.0004). This suggests that FLASH might still not be optimal for viewing the GPi. The GPi and GPe are separated by only a very thin lamina and appear nearly equal in grey values; this renders the delineation of the GPi from the GPe highly contingent upon the visibility of this lamina. This might explain some of the low inter-rater values found.
The volumes obtained were significantly different from previous GPi volume estimates. Vasques et al. reports a mean volume of 629.7mm3 on the left and 596.3mm3 on the right; this is significantly larger than our estimate (p= 7.84 x 10^-17 for the left and 8.51 x 10^-16 for the right)8,34. Lange et al. reports volumes of 520mm3 and
430mm3 for men and women respectively, thus, taking an average between these two groups because separate estimates were not given for left and right, by comparison this estimate was also significantly larger than our own (p=7.00 x 10^-11)10,41. The largest
estimate of all comes from Lenglet et al. (2012) who find a volume of 804mm3 for the left GPi and 778mm3 for the right. 42 These numbers are nearly double our estimates.
The Mai atlas, the only atlas in which specific volume estimates were found for the GPi, has the structure estimated at 238mm3 for the left and 289mm3 for the right. This also varies significantly from our own findings yet in the opposite direction (p=2.80 x 10^-13 for the left and 7.58 x 10^-11 for the right)11,12,21.
Segmentation Results
Inter-rater volumes were calculated using the terminal-based ‘fslmaths’ function by adding both rater masks and then thresholding the combined mask such that only voxels marked by both raters were included in the inter-rater mask. The inter-rater mask volume was then calculated through the ‘fslstats’ function and masks were averaged across all subjects for each hemisphere using Microsoft Excel (2011). Hemispheric differences, gender differences, and training effect were also calculated using Microsoft Excel and significance was determined for all values using two-tailed paired t-tests.
Table 1. Inter-rater values for the left and right GPi volume as well as p values for paired T-tests on hemispheric differences, gender differences, and training effect (a comparison between the first half of subjects drawn and the second).
Volume Left (mm3) Volume Right (mm3) Hemispheric difference Gender difference Training effect Rater 1 532.62 551.13 P=0.006 P=0.050 (male>female) P=0.0018 Rater 2 507.05 506.17 P=0.286 P=0.92 P=0.0022 Inter-rater 402.90 407.59 P=0.465 P=0.584 P=0.0004
Commonly Used Atlases
A search was conducted through commonly used atlases to find information related to their methods and their inclusion of the GPi. Atlases included in the study were found from their inclusion with major MR software packages such FSL
(http://www.fmrib.ox.ac.uk/fsl/) and from searching for published articles related to atlasing, segmentation, and the basal ganglia.
Table 2. An overview of the major atlases included in this study including information on the type of atlas (probabilistic or not), number of subjects with age and gender ratio if available, rather it was done in-vivo or ex-vivo, at what field strength it was conducted at, what contrasts and sequences were used, what sort of segmentation method was employed, and rather or not the study included the GP.
NG= not given N/A=not applicable Atlas name Type Subjects(in-vivo/ex-vivo) Field Strength /voxel size (mm3) Sequences Segmentatio n method GP inclusion GPi Volume ICBM Deep Nuclei 13,24 Populati on probabil istic
53 (in-vivo) 3T T1-weighted manual no N/A
Harvard-Oxford (with FSL) 12,25 14,15,43 37 (in-vivo) 1T (1x1.5) T1-weighted spoiled gradient recall (SPGR) semi-automated yes NG Julich 16,23 Probabil istic 10 (ex-vivo) 1.5T/ 1x1x1.17 T1-weighted FLASH No N/A Talairac h probabil istic 50+ (in-vivo)
10,12 16,44 MNI structura l 17,45
305 (in vivo) manual
Mai 16,21 1 (ex-vivo) Yes R: 289mm2 L: 238mm2 Allen http://ww w.brain-map.org (ex-vivo) 3T T1-weighted T2 FLAIR T1-MPRAGE Yes N/G
Individual Study Data
Another search was conducted for empirical articles relating to imaging the GP or basal ganglia, volume estimates, and optimal MR technique for deep brain structures. Studies found relevant were noted and compared and reported in table 3.
Table 3 Relevant data from various imaging studies of the GP and/or GPi and data on what structure was looked at (just the GP or the GPi specifically), the number of subjects, volume estimates, field strength, sequence, and method of segmentation used.
Study Author structure Number of Subjects Volume (left/right) mm3 Field strength Sequence (voxel size- mm3) Segmentation method manual/auto
(2007) healthy (0.9375x 0.9375 x 0.9375) workstation Reich (2000) GPi 10 healthy N/G 1.5T FSE-IR FFE Pre-operative stereotactic localization Nolte (2011) GPi 9 healthy 1 PD N/G 3T FLAIR (0.43x0.43x0.43) T1-MPRAGE (0.49x0.49x0.49) T2-SPACE (0.60x0.60x0.60) T2* FLASH2D (0.5x0.5x0.5) SWI (0.75x0.75x0.75) Osirix imagining software, Inter-rater assessment Peran (2009) GP 30 healthy 2026.4/209 5.5 3T T1-weighted T2*-weighted MDEFT (1x1x1) FIRST- for T1-weighted FSL Vasques (2008) GPi 1 patient with primary DYT1 Dystonia 629.7/596. 3 1.5T T1-weighted T2-weighted automated Lenglet (2012)
GPi 4 healthy 804/778 7T T1-weighted MPRAGE
T2-weighted 2D turbo spin echo SWI 3D GE manual Lange (1976) GPi/GPe 13 healthy GPi: 520/430 GPe: 1220/1065 Post-mortem N/A Post-mortem analysis Abosch (2010)
GPi 6 healthy N/G 7T SWI 3D GRE
T2-weighted 2D turbo
automated (Amira
spin
T1-weighted GRE
software)
Discussion
Finding volume estimates and atlases of the GPi proved to be a challenging endeavor. In most of the major atlases researched (see Table 2), the globus pallidus was not included18,24,12,19,23,11,20,45. In those where it was included, obtaining volume estimates in
all but the Mai atlas 12,21,21 was not possible. As a result, GPi volume estimates come from
single studies and the variance between these study estimates is quite large.
Although inter-rater agreement in the current study (calculated as percent overlap) was only 63.55%, it is worth nothing that estimates were very similar for each rater individually (532.62mm3 and 507.035mm3 for the left and 551.13mm3 and 506mm3 for the right) Final inter-rater volume estimates (402.90mm3 for the left and 407.59mm3 for the right) were, as noted in the results, significantly lower than previously reported estimates 22,34,21,42,23,41 but this could be due to the low level of inter-rater agreement. Yet,
also as previously noted, volume estimates were significantly higher than the Mai atlas. Vasques et al. (2008) reported a volume of 629.7mm3 on the right and 596.3mm3 on the left, however this estimate is based on a single 12 year-old female DBS patient and therefore does not incorporate any form of individual difference, nor would one use it in a healthy population. The GPi volume estimates of Lange et al. were 520mm3 in men and 430mm2 in women (individual hemispheres not given) but these were ex-vivo, based on post-mortem analysis and so also are limited in comparability. The Mai atlas was the only atlas that yielded a specific volume estimate for the GPi, which was 238mm3 for the left and 289mm3 for the right. However, these estimates are not based on the actual brain
the atlas depicts. This means that so far, to the best of our knowledge, currently there are no probabilistic atlases of the GPi freely available. The results also seem to indicate that even for single studies, there are no, or at least a very limited number of robust in-vivo healthy adult estimates of GPi volume.
The other examined atlases and studies given in Tables 2 and 3 which present volumetric data do not present volume estimates for the GPi separately, rather, these studies report estimates for the GP as a whole 12,26,46. Although the GPi and GPe are often
grouped together in imaging studies and atlases, they are thought to have distinctly different functions. The GPi in involved in all three major action selection pathways and projects to the thalamus, the GPe however receives input from the STN and Striatum and is involved primarily in the indirect pathway, projecting ultimately towards the GPi 10,38.
Because the GPi is an important target for DBS electrode implantation, this makes it all the more important to have maps which differentiate between the two structures.
It is also noteworthy that even combining the GPi and GPe together estimates vary widely between studies. Ahsan (2007) and Peran (2009) each report volume estimates of the GP but differ by nearly 700mm3 between each other
(1327mm3/1322mm3 vs. 2026.4mm3/2095.5mm3 respectively)13,26,12,46. This perhaps
reflects the challenges encountered when attempting to accurately image and measure deep brain structures like those of the basal ganglia.
All segmentation was preformed on FLASH images because visualization of the GPi with MP2RAGE was not possible. (see Figure 2)
a. b.
Fig. 2 The same subject’s GPi scanned with MP2RAGE (a) and FLASH (b) Both images are of roughly the same anatomical space, making it clear that based on visual
Previous studies have confirmed the difficulty in viewing the GPi using MPRAGE 23-25,32,46. In a study by Glenthoj (2007), basal ganglia structure volumes were calculated for
patients with schizophrenia and healthy controls. The study was preformed on a 1.5T scanner using a T1-weighted MPRAGE sequence. The authors preformed manual segmentation of the caudate, putamen, accumbens, and globus pallidus but had to discard the globus pallidus from further analysis due to low inter-rater agreement (0.52 ICC). 26,47
However, although the GPi was more visible in FLASH, as the inter-rater agreement was low and a significant training effect was present for both raters, this suggests that chosen MR sequence was still not optimal for visualizing the GPi.
Brain iron concentrations are maximal in the GP, SN, red nucleus, caudate and putamen and generally produce a reduced signal in regular T2-weighted images. 10,48 This
could possibly explain some of the difficulties encountered while trying to define strict borders around the GPi. Other imaging studies of the GP and basal ganglia report optimal image contrast using T2* FLASH sequences 12,23-25,29,32,46 particularly in
combination with a new MR sequence: susceptibility-weighted imaging (SWI). 29
SWI employs the magnetic susceptibility of differences between the tissue of interest and the surrounding tissue. At high magnetic fields (3T or higher) SWI allows for high spatial resolution and improved SNR49. SWI has recently been shown, especially
at high magnetic fields, to exhibit superior contrast within tissues compared to
conventional T1- and T2-weighted images and allows clearer delineation of structures within grey matter. 29
In a study published by Nolte (2012) visibility of the GPi was compared in T2*FLASH2D, MPRAGE, and SWI. The study concluded that T2*FLASH2D combined
with SWI provided the most reliable demarcation of the GPi and recommended this combination for future studies of the GPi 32. Absoch et al. (2010) in an imagining study
of DBS targets also reported that using SWI at allowed for easy demarcation of basal ganglia structures and that they could clearly separate the GPi from the GPe 29. This
suggests that the results of our study could have possibly been enhanced by additionally using SWI in combination with a T2*-weighted sequence and might be a suggestion for future studies.
References:
1. Haber SN. The primate basal ganglia: parallel and integrative networks. Journal of
Chemical Neuroanatomy. 2003;26(4):317–330.
2. Utter AA, Basso MA. The basal ganglia: An overview of circuits and function.
Neuroscience & Biobehavioral Reviews. 2008;32(3):333–342.
3. Nambu A, Tokuno H, Takada M. Functional significance of the cortico-subthalamo-pallidal “hyperdirect” pathway. Neurosci Res. 2002;43(2):111–117.
4. Cho Z-H, Min H-K, Oh S-H, et al. Direct visualization of deep brain stimulation targets in Parkinson disease with the use of 7-tesla magnetic resonance imaging. Journal
of Neurosurgery. 2010;113(3):639–647.
5. Redgrave P, Rodríguez M, Smith Y, et al. Goal-directed and habitual control in the basal ganglia: implications for Parkinson's disease. Nat Rev Neurosci. 2010;11(11):760– 772.
6. Lozano AMA, Dostrovsky JJ, Chen RR, Ashby PP. Deep brain stimulation for Parkinson's disease: disrupting the disruption. Lancet Neurol. 2002;1(4):225–231. 7. Lozano AM, Snyder BJ, Hamani C, Hutchison WD, Dostrovsky JO. Basal ganglia physiology and deep brain stimulation. Stanley F, Karen M, Lucien C, Stephen G R, eds.
Mov. Disord. 2010;25(S1):S71–S75.
8. Okun MS, Foote KD. Subthalamic nucleus vs globus pallidus interna deep brain stimulation, the rematch: will pallidal deep brain stimulation make a triumphant return?
9. Chen SY, Tsai ST, Hung HY, Lin SH, Pan YH, Lin SZ. Targeting the Subthalamic Nucleus for Deep Brain Stimulation—A Comparative Study Between Magnetic Resonance Images Alone and Fusion with Computed Tomographic Images. WNEU. 2011;75(1):132–137.
10. Toga AW, Thompson PM, Mori S, Amunts K, Zilles K. Towards multimodal atlases of the human brain. Nat Rev Neurosci. 2006;7(12):952–966.
11. Schaltenbrand G, Wahren W, Hassler R. Atlas for Stereotaxy of the Human Brain. Thieme; 1977.
12. Talairach J, Tournoux P. Co-planar stereotaxic atlas of the human brain: 3-dimensional proportional system: an approach to cerebral imaging. 1988.
13. Ganser KA, Dickhaus H, Metzner R, Wirtz CR. A deformable digital brain atlas system according to Talairach and Tournoux. Medical Image Analysis. 2004;8(1):3–22. 14. Mazziotta JC, Woods R, Iacoboni M, et al. The myth of the normal, average human brain—The ICBM experience: (1) Subject screening and eligibility. NeuroImage. 2009;44(3):914–922.
15. Rajkowska G, Goldman-Rakic PS. Cytoarchitectonic definition of prefrontal areas in the normal human cortex: II. Variability in locations of areas 9 and 46 and relationship to the Talairach Coordinate System. Cerebral Cortex. 1995;5(4):323–337.
16. Uylings HBM, Rajkowska G, Sanz-Arigita E, Amunts K, Zilles K. Consequences of large interindividual variability for human brain atlases: converging macroscopical imaging and microscopical neuroanatomy. Anat Embryol. 2005;210(5-6):423–431. 17. Amunts KK, Schleicher AA, Bürgel UU, Mohlberg HH, Uylings HBH, Zilles KK. Broca's region revisited: cytoarchitecture and intersubject variability. J Comp Neurol. 1999;412(2):319–341.
18. Forstmann BU, Keuken MC, Jahfari S, et al. Cortico-subthalamic white matter tract strength predicts interindividual efficacy in stopping a motor response. NeuroImage. 2010;60(1):1–6.
19. Brodmann K. Vergleichende lokalisationslehre der großhirnrinde. Leipzig; 1909. 20. Kapur T, Grimson WEL, Wells WM, Kikinis R. Segmentation of brain tissue from magnetic resonance images. Medical Image Analysis. 1996;1(2):109–127.
21. Mai JK, Paxinos G, Voss T. Atlas of the Human Brain. 2007.
22. Roland PE, Zilles K. Brain atlases--a new research tool. Trends Neurosci. 1994;17(11):458–467.
probabilistic cytoarchitectonic maps and functional imaging data. NeuroImage. 2005;25(4):1325–1335.
24. Mazziotta JC, Toga AW, Evans A, Fox P, Lancaster J. A Probabilistic Atlas of the Human Brain: Theory and Rationale for Its Development:: The International Consortium for Brain Mapping (ICBM). NeuroImage. 1995;2(2):89–101.
25. Makris N, Goldstein JM, Kennedy D, et al. Decreased volume of left and total anterior insular lobule in schizophrenia. Schizophrenia Research. 2006;83(2-3):155–171. 26. Ahsan RL, Allom R, Gousias IS, et al. Volumes, spatial extents and a probabilistic atlas of the human basal ganglia and thalamus. NeuroImage. 2007;38(2):261–270. 27. Maubon AJ, Ferru JM, Berger V, et al. Effect of Field Strength on MR Images: Comparison of the Same Subject at 0.5, 1.0, and 1.5 T1. Radiographics.
1999;19(4):1057–1067.
28. Magee T, Shapiro M, Williams D. Comparison of high-field-strength versus low-field-strength MRI of the shoulder. AJR Am J Roentgenol. 2003;181(5):1211–1215. 29. Abosch A, Yacoub E, Ugurbil K, Harel N. An Assessment of Current Brain Targets for Deep Brain Stimulation Surgery With Susceptibility-Weighted Imaging at 7 Tesla.
Neurosurgery. 2010;67(6):1745–1756.
30. Metcalf M, Xu D, Okuda DT, et al. High-Resolution Phased-Array MRI of the Human Brain at 7 Tesla: Initial Experience in Multiple Sclerosis Patients. Journal of
Neuroimaging. 2010;20(2):141–147.
31. Huettel S, Song A, McCarthy G. Functional Magnetic Resonance Imaging. {Sinauer Associates}; 2004.
32. Nölte IS, Gerigk L, Al-Zghloul M, Groden C, Kerl HU. Visualization of the internal globus pallidus: sequence and orientation for deep brain stimulation using a standard installation protocol at 3.0 Tesla. Acta Neurochir. 2011;154(3):481–494.
33. Heckemann RA, Hajnal JV, Aljabar P, Rueckert D, Hammers A. Automatic anatomical brain MRI segmentation combining label propagation and decision fusion.
NeuroImage. 2006;33(1):115–126.
34. Vasques X, Cif L, Hess O, Gavarini S, Mennessier G, Coubes P. Stereotactic model of the electrical distribution within the internal globus pallidus during deep brain stimulation. J Comput Neurosci. 2008;26(1):109–118.
35. Zhang Y, Brady M, Smith S. Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm. IEEE Trans
Med Imaging. 2001;20(1):45–57.
IEEE Trans Med Imaging. 1998;17(1):98–107.
37. Carmichael OT, Aizenstein HA, Davis SW, et al. Atlas-based hippocampus segmentation in Alzheimer's disease and mild cognitive impairment. NeuroImage. 2005;27(4):979–990.
38. Nambu A. Globus pallidus internal segment. Progress in Brain Research. 2007;160:135–150.
39. Reich CA, Hudgins PA, Sheppard SK, Starr PA, Bakay RA. A high-resolution fast spin-echo inversion-recovery sequence for preoperative localization of the internal globus pallidus. AJNR Am J Neuroradiol. 2000;21(5):928–931.
40.
41. Lange HH, Thörner GG, Hopf AA, Schröder KFK. Morphometric studies of the neuropathological changes in choreatic diseases. J Neurol Sci. 1976;28(4):401–425. 42. Lenglet C, Abosch A, Yacoub E, De Martino F, Sapiro G, Harel N. Comprehensive in vivo Mapping of the Human Basal Ganglia and Thalamic Connectome in Individuals Using 7T MRI. Bankiewicz K, ed. PLoS ONE. 2012;7(1):e29153.
43. Frazier JAJ, Chiu SS, Breeze JLJ, et al. Structural brain magnetic resonance imaging of limbic and thalamic volumes in pediatric bipolar disorder. Am J Psychiatry.
2005;162(7):1256–1265.
44. Lancaster JLJ, Woldorff MGM, Parsons LML, et al. Automated Talairach atlas labels for functional brain mapping. Hum. Brain Mapp. 2000;10(3):120–131.
45. Collins DL, Holmes CJ, Peters TM, Evans AC. Automatic 3-D model-based neuroanatomical segmentation. Hum. Brain Mapp. 2004;3(3):190–208.
46. Péran P, Cherubini A, Luccichenti G, et al. Volume and iron content in basal ganglia and thalamus. Hum. Brain Mapp. 2009;30(8):2667–2675. Available at:
http://onlinelibrary.wiley.com/doi/10.1002/hbm.20698/full.
47. Glenthoj A, Glenthoj BY, Mackeprang T, et al. Basal ganglia volumes in drug-naive first-episode schizophrenia patients before and after short-term treatment with either a typical or an atypical antipsychotic drug. Psychiatry Research: Neuroimaging.
2007;154(3):199–208.
48. Ye FQF, Allen PSP, Martin WRW. Basal ganglia iron content in Parkinson's disease measured with magnetic resonance. Mov. Disord. 1996;11(3):243–249.
49. Deistung A, Rauscher A, Sedlacik J, Stadler J, Witoszynskyj S, Reichenbach JR. Susceptibility weighted imaging at ultra high magnetic field strengths: Theoretical considerations and experimental results. Magn. Reson. Med. 2008;60(5):1155–1168.