1
MAPPING THE GLOBUS PALLIDUS INTERNAL AND EXTERNAL SEGMENT USING 7
TESLA QUANTITATIVE SUSCEPTIBILITY MAPPING
Rebecca Sier, 5893011, May 2013 Cognitive Science Center Amsterdam Supervisor: M.C. Keuken
Co-assessor: B.U. Forstmann
ABSTRACT
Due to its role in motor control, the globus pallidus internal segment (GPi) is a target of deep brain stimulation (DBS). Due to its location in the basal ganglia and its small volume it is best visualized using ultra-high resolution 7 Tesla (T) magnetic resonance imaging (MRI). Quantitative susceptibility maps (QSM) are suggested to provide best contrast for iron-rich regions like the GPi. In the present study, masks of the internal and external segment of the globus pallidus were individually segmented in QSM to generate atlas probability maps. Secondly, to assess QSM usability, inter-rater values of the GPi masks were compared to inter-rater values of GPi masks made in a FLASH sequence. Thirdly, QSM values of the two pallidal segments of left and right hemispheres were compared to assess differences in iron
distribution. Results show that QSM is superior to FLASH in visualizing the GPi, that interindividual variability exists in the GPi and GPe and that there are no significant differences in QSM values of left and right GPe and GPi.
2 INTRODUCTION
Considering increasingly specialized treatments such as deep brain stimulation (DBS) for specific brain disorders, exact knowledge of the organization of the brain is of vital importance. In the treatment of several movement diseases like Parkinson, Huntington and dystonia with DBS an electrode is placed in the globus pallidus internal segment (GPi), part of the globus pallidus (GP) that itself is one of the subcortical nuclei forming the basal ganglia (BG).
The GP plays a major role in control of motor activity through several cortico-basal ganglia pathways (Haber, 2003; Nambu, Tokuno, & Takada, 2002; Nambu, 2007; Utter & Basso, 2008). The surgical procedure of placing the DBS electrodes is precarious though, due to the location and size of the GPi. The GPi is located directly medial of the GPe and lateral to the substantia nigra (Mai, Paxinos, & Voss, 2008). The size of the GPi makes it even harder to target due to the fact that it is less than half the volume of the entire globus pallidus (GP), which is estimated to be roughly between 1300 and 2000 mm3 (Ahsan et al., 2007; Anastasi et al., 2006; Lenglet et al., 2012; Péran et al., 2009; Peterson et al., 1993). Next to that, distinguishing the GPi from the external GP segment (GPe) is complicated by their locations, seamlessly fitting one another, with only a thin sheet of white matter known as the lamina
pallidi medialis as a border in between. A third complication would be the anatomical variability
between brains (Amunts et al., 1999; Mazziotta et al., 2009), decreasing the usability of standard brain atlases (Evans, Janke, Collins, & Baillet, 2012). It is therefore desirable to image and atlas the GPi per individual patient, with highest possible detail, facilitating surgical procedures like DBS.
Keeping up with recent developments in magnetic resonance imaging (MRI), several attempts have been made to atlas the brain with increasing detail. Studies on imaging the GP show how T1-weighted sequences using 1.5 and 3 Tesla (T) magnets could not be used to discern between the GPi and GPe (Ahsan et al., 2007; Nölte, Gerigk, Al-Zghloul, Groden, & Kerl, 2012; Péran et al., 2009; Peterson et
3 al., 1993). While gray and white matter are contrasted clearly using T1, differences in longitudinal relaxation times of the gray matter that make up the GPi and GPe are less pronounced.
When delineating the individual segments of the GP, T2 and especially T2* sequences are found to give a useful contrast, enabling localization of the GPi (Abosch, Yacoub, Ugurbil, & Harel, 2010; Aquino et al., 2009; Lenglet et al., 2012; Nölte et al., 2012; Vasques et al., 2009). The improved contrasts created with T2* sequences are in line with knowledge on the chemical composition of the GP. It is broadly accepted that the BG, especially the GP, contain a relatively high iron concentration which increases with age (Aquino et al., 2009; Daugherty & Raz, 2013; Haacke et al., 2005; Hallgren & Sourander, 1958; Langkammer et al., 2012; Péran et al., 2009; Yao et al., 2012). Now, T2* sequences contain information on magnetic susceptibility of brain tissue, for it is the relaxation time due to both spin-spin relaxation (i.e. regular T2) and the acceleration of spin dephasing by static inhomogeneities in the B0 magnetic field. The paramagnetic character of iron adds to these static inhomogeneities, making it possible for T2* to contrast iron-rich with iron-poor brain regions. Using such a sequence is a first step in contrasting the BG from surrounding tissue.
The higher signal-to-noise (SNR) and contrast-to-noise (CNR) ratios obtained using a stronger magnet (Maubon et al., 1999) led researchers to use 7T magnets for higher precision in the (manual) segmentation of the two segments of the GP and other parts of the BG (Abosch et al., 2010; Deistung et al., 2012; Forstmann et al., 2012; Keuken et al., 2013; Lenglet et al., 2012).
Fairly recently, it was found that combining strong magnets with susceptibility weighted imaging (SWI) improves the CNR in GPi imaging significantly (Abosch et al., 2010; Lenglet et al., 2012; Nölte et al., 2012; Yao et al., 2012). SWI exploits the difference in magnetic susceptibilities even better than T2* by directly contrasting the phase images of neighboring brain regions (Haacke et al., 2005; Yablonskiy & Haacke, 1994).
4 A weakness of SWI and other sequences exploiting phase images is that they are influenced by non-local magnetic susceptibilities, i.e. magnetic inhomogeneities that are present in other regions than the region of interest (ROI). This particular problem and other hurdles that account for noise in the SWI method are removed by several algorithms processing the data. The resulting images are quantitative susceptibility maps (QSM), providing an even better contrast of iron-rich and iron-poor regions (Deistung et al., 2012; Langkammer et al., 2012; Schweser, Deistung, Lehr, & Reichenbach, 2011; Schweser, Deistung, Sommer, & Reichenbach, 2012)
Next to providing a useful contrast for delineating the GPi and GPe, QSM values of gray matter are a linear measure of the tissue’s iron content (Deistung et al., 2012; Langkammer et al., 2012;
Schweser et al., 2011). This feature may be useful for automatic segmentation protocols separating iron-rich from iron-poor regions. The automation of segmentation is desired, for manual segmentation is labor intensive and is influenced by observer biases. However, current automatic segmentation algorithms still result in worse masks of the GPi and GPe compared to manual segmentation (Ahsan et al., 2007). Thus, the algorithms are yet to be improved. A possible difference in QSM values might contribute to this aim.
To the best of our knowledge, no studies are known in which the differences in iron content between GPi and GPe are assessed. However, both the GPi and GPe can be segmented using SWI or QSM at 7T, since the lamina pallidi medialis is a layer of white matter, and therefore has a different magnetic susceptibility (Abosch et al., 2010). Furthermore, no reports on manual segmentation of the GPi and GPe using QSM in 7T MRI are found to having been published. Therefore the claim that QSM is more useful than regular T1 or T2 sequences in segmenting the GPi has yet to be shown.
The aim of the current study is to map the anatomical features of the GPi and the GPe using 7T MRI. Manually segmented QSM masks of the GPi and GPe will be used to analyze the volume,
5 the MNI standard stereotactic space and probability maps will be derived. Next to that, the use of QSM in segmenting the GPi will be compared to the use of standard FLASH sequences, providing T2* contrast. To this end inter-rater coefficients resulting from manual segmentation by different investigators are assessed, as well as resulting volume estimations.
The results will contribute to a new, high-resolution atlas of the brain, useful in a clinical as well as a scientific environment. Furthermore, the calculated QSM values may contribute to an algorithm enabling automatic segmentation of the GP segments.
METHODS Participants
For the acquisition of brain imaging data, 30 participants (14 females) with mean age 24.2 (SD = 2.4) were scanned. All participants signed a consent form before the scanning session. None of the participants had a history of neurological, major medical, or psychiatric disorders. The study was approved by the local ethics committee at the Max Planck Institute for Human Cognitive and Brain Sciences and subjects gave their written informed consent.
Data acquisition of ultra-high resolution anatomical images
The participants’ brains were scanned using a 7T Siemens Magnetom MRI system, using a 24-channel head array Nova coil (NOVA Medical Inc., Wilmington MA) with three sequences: MP2RAGE (Marques et al., 2010) (repetition time TR = 5000 ms; inversion times TI1, TI2 = 900/2750 ms) with 0.7 mm (240 sagittal slices) and 0.6 mm (slab of 128 slices) isotropic resolution, and a multiecho spoiled 3 dimensional (3D) gradient echo (FLASH) (Haase, Frahm, Matthaei, Hänicke, & Merboldt, 1986). The FLASH slab (0.5 mm isotropic voxels) had 128 slices (TR = 41 ms) with three echo times (TE): 11 / 20 / 30
6 ms. The MP2RAGE and FLASH slabs were oriented parallel to the AC-PC line. QSM are derived from the FLASH images using the superfast dipole inversion algorithm (Schweser et al., 2012; Wharton, Schäfer, & Bowtell, 2010).
Manual segmentation of the GPi and GPe
Manual segmentation was carried out by three independent researchers A, B and C using the FSL 5.0.2 viewer. The segmentation was done in native subject space. Both researchers A and B segmented the GPi and GPe in QSM. Researcher A also segmented the GPi in the FLASH images, as did researcher C. Both image modalities were thus segmented by two different researchers: both the GPi and GPe in QSM and the GPi in FLASH scans. A segmentation protocol specified landmarks the researchers should take into account when segmenting. The GP is situated medially from the putamen, the GPi medially from the GPe. The manual segmentation was done in several steps: first, either the FLASH or QSM image of an individual subject was loaded into the viewer. Then, the visibility of the GP was maximized by changing the contrast values in the viewer. Third, either the coronal, sagittal or transverse view was randomly picked to start delineating either the GPe or GPi. The sequence in which the right or left hemisphere and the internal or external GP segment were segmented per subject was randomized. Finally, the volumes of inter-rater masks and inter-rater reliability were calculated. For the latter, Cohen’s kappa (Cohen, 1960), DICE’s coefficient (Dice, 1945) and the intra-class correlation coefficient (ICC) (Shrout & Fleiss, 1979) were used. All further analysis regarding volume and QSM values are based on the inter-rater masks, containing only the voxels to which both raters agreed on. The inter-rater values for on the one hand the FLASH masks, and on the other hand the QSM masks, were corrected for possible volume effects and compared in order to assess if either QSM or FLASH can best be used for delineating the GPi. Average QSM numbers were calculated for the inter-rater masks of left and right GPi and GPe, giving a measure of the concentration of iron in these structures.
7
Computation of probability maps and atlasing of the GPi and GPe
The masks were aligned to the MP2RAGE whole-brain image of the same subject and then normalized to the 0.4 mm3 MNI template as provided by the CBS High-Res Brain Processing Tools
(http://www.cbs.mpg.de/institute/software/cbs-hrt/index.html). All registration steps were done using an automatic linear registration algorithm using MIPAV (www.mipav.cit.nih.gov). The calculation of Cohen’s kappa, DICE coefficients and statistics was done using R.
RESULTS
Atlasing of the GPi and GPe
Figure 1 and 2 show the probability maps derived from 30 participants for the GPe and the GPi
respectively. The highest overlap across participants for both the left and right GPe was 100% (peak MNI coordinate in millimeters for left GPe: x = -16.70, y = 0.82, z = -0.13; peak MNI coordinate in millimeters for right GPe: x = 18.90, y = 0.02, z = 1.47). For both the left and right GPi the highest overlap across participants was 96.67% (peak MNI coordinate in millimeters for left GPi: x = -19.50, y = -8.38, z = -3.33; peak MNI coordinate in millimeters for right GPi: x = 20.90, y = -8.38, z = -2.93).
The centre of gravity coordinates in millimeters for the left and right GPe were x = 19.52, y = -3.86, z = -0.39 and x = 20.03, y = -2.62, z = -0.35 repectively. For the GPi left and right centre of gravity coordinates were x = -18.10, y = -7.33, z = -3.31 and x = 18.70, y = -6.29, z = -3.18 respectively.
8
Right Hemisphere Left Hemisphere
Fig. 1. Probability maps are based on 30 individually segmented left
and right 7T MRI GPe masks. Red indicates minimal overlap of 3.33%, yellow indicates maximal overlap of 100%.
9
Right Hemisphere Left Hemisphere
Fig. 2. Probability maps are based on 30 individually segmented left and
right 7T MRI GPi masks. Red indicates minimal overlap of 3.33%, yellow indicates maximal overlap of 97%.
10
Volume calculations
Table 1 displays each individual’s left and right GPi and GPe volume according to QSM inter-rater masks, including each segment’s mean volume per hemisphere. The mean volume of all the GPe masks in QSM is 918.49 mm3 (SD = 123.38). The mean volume of all the GPi masks in QSM is 365.87 mm3 (SD = 59.85).
Subjects Left GPi (mm3) Right GPi (mm3) Left GPe (mm3) Right GPe (mm3)
pp01 269.88 290.88 804.75 908.00 pp02 316.25 285.50 799.38 763.50 pp03 289.74 353.62 801.50 790.62 pp04 352.50 366.63 912.38 903.25 pp05 330.25 306.63 908.88 899.38 pp06 314.88 257.88 1031.13 816.00 pp07 379.75 394.00 953.38 916.13 pp08 356.00 358.88 942.75 905.38 pp09 331.50 422.25 1075.88 1061.00 pp10 311.75 286.00 963.01 855.38 pp11 297.50 364.50 739.50 705.88 pp12 391.50 313.00 1060.63 1061.13 pp13 466.87 462.25 996.25 959.25 pp14 548.00 398.13 1225.76 1190.88 pp15 480.00 474.75 1073.25 1024.38 pp16 399.25 367.00 787.00 851.13 pp17 328.00 355.75 820.01 767.76 pp18 466.13 487.38 1073.76 1027.13 pp19 338.13 411.13 1013.26 934.13 pp20 410.88 403.75 1016.38 1025.26 pp21 446.00 400.50 1064.13 1030.75 pp22 333.13 353.50 746.13 796.13 pp23 325.75 360.75 803.62 785.25 pp24 359.50 421.75 943.12 875.75 pp25 354.13 344.88 887.88 915.38 pp26 391.00 375.25 814.63 783.13 pp27 339.50 363.75 1086.38 1110.38 pp28 323.87 273.25 781.37 697.25 pp29 383.13 336.38 902.88 808.01 pp30 Mean (SD) 364.13 366.63 (63.16) 363.50 365.11 (57.42) 945.13 932.47 (124.16) 967.50 904.50 (123.09) Table 1
Individual volume estimates for the left and right GPi and GPe masks for the 30 participants using ultra-high resolution 7T MRI QSM.
11 Subjects Left GPi (mm3) Right GPi (mm3)
pp01 405.13 394.38 pp02 411.63 421.50 pp03 455.75 428.87 pp04 483.88 510.25 pp05 456.50 439.50 pp06 480.00 448.50 pp07 449.38 433.75 pp08 383.63 415.13 pp09 416.00 494.88 pp10 418.88 391.63 pp11 336.88 369.38 pp12 523.63 521.00 pp13 432.87 449.62 pp14 447.38 380.50 pp15 424.88 426.88 pp16 295.50 337.00 pp17 324.88 328.13 pp18 282.25 280.75 pp19 418.75 378.88 pp20 300.38 278.63 pp21 330.63 329.38 pp22 324.38 347.38 pp23 418.00 405.62 pp24 395.75 437.50 pp25 343.75 345.75 pp26 388.00 390.88 pp27 540.38 502.13 pp28 261.50 361.62 pp29 470.25 484.50 pp30 466.13 493.87 Mean (SD) 402.90 (71.74) 407.59 (65.35) Table 2
Individual volume estimates for the left and right GPi masks for the 30 participants using ultra-high resolution 7T MRI FLASH.
Table 2 displays each individual’s left and right GPi volume according to FLASH inter-rater masks,
including each segment’s mean volume per hemisphere. The mean volume of all the GPi masks in FLASH is 405.24 mm3 (SD = 68.08).
The difference in volume of the GPi according to masks made in FLASH and QSM is significant (t(188) = 3.36, p = 0.001). Thus, the QSM scan sequence results in a smaller mask than the FLASH scan sequence.
No significant differences in volume were found between hemispheres for either the GPe and GPi QSM or GPi FLASH masks. As expected, all individuals’ GPe volumes in QSM are found to be bigger than the corresponding GPi volumes.
The volume estimates for the probability masks are displayed in Table 3.
Inter-rater reliability
Table 4 displays each individual’s inter-rater coefficients of the masks per GP segment, hemisphere and MRI sequence. Cohen’s kappa, the DICE coefficient and the ICC value were used to assess the inter-rater reliability.
The masks made in QSM had an inter-rater reliability of 0.88 (SD = 0.02) for the GPe and 0.83 (SD = 0.03) for the GPi, using Cohen’s kappa. The DICE coefficient gave an inter-rater reliability of 87.67
12 (SD = 1.32) for the GPe and 82.89 (SD = 2.52) for the GPi. The mean ICC value was 0.86 (SD = 0.05) for the GPe and 0.80 (SD = 0.07) for the GPi.
The GPi masks made in FLASH resulted in a Cohen’s kappa of 0.77 (SD = 0.05) and 77.46 (SD = 4.99) for the DICE coefficient. The mean ICC value for GPi FLASH masks was 0.81 (SD = 0.07).
The difference in inter-rater values of the GPi masks segmented in QSM and FLASH is significant according to both Cohen’s kappa (t(58) = 5.31, p < .001) and the DICE coefficient (t(58) = 5.32, p < .001). An ANCOVA for QSM versus FLASH volume estimates controlling for Cohen’s kappa inter-rater values was found to have statistically significant main effects (t = -7.25, p < .001). Thus, although volume depends on the Cohen’s kappa inter-rater value, it depends on the scan modality too, without an interaction between scan modality and inter-rater value. Therefore using either FLASH or QSM to visualize the GPi influences the estimated volume of the structure.
Sequence, GP segment, hemisphere Volume (mm3)
FLASH, GPi, left 3308.63 FLASH, GPi, right 3295.25 QSM, GPe, left 5891.00 QSM, GPe, right 6153.75 QSM, GPi, left 3156.88 QSM, GPi, right 3088.63
Table 3
Volume estimates for the left and right GPi and GPe probability maps for the 30 participants using 7T ultra-high resolution MRI FLASH and QSM.
13
QSM values
Table 5 displays all mean QSM values of the GPi and GPe inter-rater masks per hemisphere and
individual. According to (Langkammer et al., 2012) the QSM value reflects the amount of iron in ppm. No significant difference in the QSM values were found when comparing between hemispheres and GP segments. Therefore the current QSM values would not be useful in improving automatic segmentation algorithms.
GPe GPi
QSM QSM FLASH
Subjects DICE Cohen’s kappa
ICC DICE Cohen’s kappa
ICC DICE Cohen’s
kappa ICC pp01 85.62 0.86 0.82 82.46 0.82 0.80 74.54 0.75 0.69 pp02 85.68 0.86 0.80 80.83 0.81 0.75 72.61 0.73 0.76 pp03 86.22 0.86 0.83 81.15 0.81 0.87 83.40 0.83 0.79 pp04 88.59 0.89 0.85 82.67 0.83 0.83 78.87 0.79 0.87 pp05 87.69 0.88 0.86 81.89 0.82 0.83 80.85 0.81 0.87 pp06 85.27 0.85 0.81 77.89 0.78 0.68 81.33 0.81 0.85 pp07 88.15 0.88 0.84 84.36 0.84 0.83 83.14 0.83 0.80 pp08 88.60 0.89 0.85 80.78 0.81 0.84 75.18 0.75 0.71 pp09 90.20 0.90 0.89 82.38 0.82 0.78 80.27 0.80 0.88 pp10 87.20 0.87 0.86 78.43 0.78 0.70 72.18 0.72 0.71 pp11 85.80 0.86 0.84 84.89 0.85 0.84 79.72 0.80 0.88 pp12 88.09 0.88 0.88 83.23 0.83 0.82 80.03 0.80 0.73 pp13 89.20 0.89 0.92 87.07 0.87 0.89 83.45 0.83 0.86 pp14 87.44 0.87 0.93 81.90 0.82 0.80 74.24 0.74 0.85 pp15 88.66 0.89 0.90 85.88 0.86 0.85 78.18 0.78 0.93 pp16 88.12 0.88 0.87 85.91 0.86 0.86 75.93 0.76 0.84 pp17 88.37 0.88 0.90 82.93 0.83 0.82 78.47 0.78 0.89 pp18 88.65 0.89 0.87 84.74 0.85 0.79 63.34 0.63 0.82 pp19 88.55 0.89 0.86 81.72 0.82 0.73 81.29 0.81 0.81 pp20 89.60 0.90 0.91 81.58 0.82 0.75 71.47 0.71 0.83 pp21 88.25 0.88 0.87 85.89 0.86 0.86 67.84 0.68 0.75 pp22 86.36 0.86 0.86 85.52 0.86 0.78 79.26 0.79 0.83 pp23 86.07 0.86 0.80 82.11 0.82 0.76 75.36 0.75 0.84 pp24 89.24 0.89 0.87 86.14 0.86 0.85 81.66 0.82 0.83 pp25 88.42 0.88 0.86 82.36 0.82 0.77 77.28 0.77 0.81 pp26 86.21 0.86 0.78 84.01 0.84 0.78 79.69 0.80 0.80 pp27 87.27 0.87 0.82 76.60 0.77 0.73 81.18 0.81 0.80 pp28 88.73 0.89 0.86 84.80 0.85 0.82 69.86 0.70 0.66 pp29 87.56 0.88 0.82 84.79 0.85 0.79 83.97 0.84 0.76 pp30 86.37 0.86 0.80 81.62 0.82 0.77 79.22 0.79 0.82 Mean (SD) 87.67 (1.32) 0.88 (0.01) 0.86 (0.04) 82.89 (2.52) 0.83 (0.03) 0.80 (0.05) 77.46 (4.99) 0.77 (0.05) 0.81 (0.06) Table 4
Cohen’s kappa, DICE coefficients and ICC values of inter-rater GPi and GPe masks per hemisphere, per MRI sequence and per individual.
14 DISCUSSION
In the present study, ultra-high 7T data was acquired to manually segment the GPe and GPi of 30 individuals. QSM were used to segment both the GPe and GPi, a FLASH sequence was used for segmenting only the GPi. Inter-rater masks resulting from QSM provided volume estimations and probability maps in MNI-space of both GP segments. Inter-rater values of the GPi segmentation in both
Subjects Left GPe Right GPe Left GPi Right GPi pp01 0.0351775 0.0525485 0.0296025 0.0556775 pp02 0.047667 0.059451 0.040751 0.039308 pp03 0.0358315 0.0407785 0.0384125 0.0354115 pp04 0.0812065 0.0927625 0.0577005 0.0688515 pp05 0.1138455 0.1208025 0.1028905 0.1137325 pp06 0.0637275 0.0870425 0.0586775 0.0949445 pp07 0.0646875 0.0875485 0.0544675 0.0893985 pp08 0.061996 0.080801 0.072833 0.0739 pp09 0.057978 0.056338 0.043786 0.057416 pp10 0.1086725 0.1454355 0.1102755 0.1503355 pp11 0.071576 0.067825 0.072302 0.044507 pp12 0.092285 0.101124 0.067227 0.088506 pp13 0.0290795 0.0262105 0.0365535 0.0305485 pp14 -0.143959 -0.13175 -0.144047 -0.147297 pp15 0.061745 0.062532 0.05157 0.046516 pp16 0.0637155 0.0757565 0.0380125 0.0731035 pp17 0.1155815 0.1197465 0.0794915 0.0918455 pp18 0.1050015 0.1125695 0.0950625 0.1062805 pp19 0.12084 0.129542 0.117097 0.130595 pp20 0.0730125 0.0663265 0.0526515 0.0458835 pp21 0.1203105 0.1047805 0.1015525 0.1148265 pp22 0.060173 0.067896 0.069634 0.073224 pp23 0.0486625 0.0301345 0.0116475 0.0137675 pp24 0.0861715 0.1025345 0.0785115 0.0989205 pp25 0.083139 0.080821 0.068713 0.074773 pp26 0.089131 0.09586 0.075061 0.078062 pp27 0.0507665 0.0661165 0.0383615 0.0601225 pp28 0.0767415 0.0952515 0.0605105 0.0882385 pp29 0.1203115 0.1343375 0.0985545 0.1187195 pp30 0.119443 0.123155 0.114167 0.118292 Mean (SD) 0.130915 (0.01879) 0.138907 (0.019584) 0.120166 (0.017260) 0.131378 (0.023052) Table 5
15 QSM and FLASH gave an indication of the usability of these sequences to delineate the GPi. Ultimately, QSM values of both GP segments gave an indication of a difference in iron distribution.
The probability masks and individual volumes reveal interindividual variability in the precise location and size of the 30 subjects’ GPe and GPi. It is therefore necessary to localize the GP segments per individual, especially in a clinical environment, when for example preparing DBS surgery. The present probability maps provide a useful addition to the localization of GPe and GPi.
The significantly differing inter-rater values for GPi masks made in FLASH and QSM indicate that QSM are more useful in imaging the GPi. As expected, the contrast provided by differences in magnetic susceptibility helps discerning the iron-rich GP segments from surrounding tissue. The white matter sheet separating GPi from GPe is visible using ultra-high resolution QSM, due to its low magnetic susceptibility compared to that from the GPe and GPi. This particular sheet was less visible in FLASH sequences, making it harder to distinguish the two GP segments from each other.
However, some caution is advised when using QSM for exact delineation of the GPi. Table 6 shows how several MRI studies show a great variation in volume estimations of the GPi. Next to that, volume measurements from post mortem data by (Lange & Albring, 1979) are smaller than those
calculated from present inter-rater masks for the QSM data. Thus, it is possible that, like in FLASH, masks
MRI/Post-mortem (PM) Left GPi (mm3) Right GPi (mm3) Left GPe (mm3) Right GPe (mm3) Left GP (mm3) Right GP (mm3) Heckers et al. (1991) PM X X X X 1820 1800
Lange & Albring (1979) PM 238 289 560 656 798 945
Present study MRI (7T) 366.63 365.11 932.47 904.50 1299.1 1269.61
Ahsan (2007) MRI (1.5T) x x x X 1486.5 1406
Anastasi (2006) MRI (1.5T) x x x x 1180 1230
Hokama et al. (1995) MRI (1.5T) x x x x 1150 1110
Péran (2009) MRI (3T) x x x x 2026.10 2095.50
Vasques et al. (2009) MRI (1.5T) 596.3 629.7 x x x x
Table 6
16 made in QSM overestimate the volume of the GPi. On the other hand, a post mortem study by (Heckers, Heinsen, Heinsen, & Beckmann, 1991) resulted in larger GP volumes than estimated in the present study. Both post mortem studies correct for tissue shrinkage. Since both post mortem studies are outdated and controdictory, it is suggested to directly compare new post-mortem data to volume estimates in QSM data. This might give a better understanding of the usability of QSM in delineating the GPi.
The QSM values were not found to be significantly different when comparing between GPi and GPe or between hemispheres. Therefore this particular feature of QSM is not useful for improvement of automatic segmentation algorithms that are able to separate between the GPi and GPe.
QSM values could be of other use, though. As stated in the introduction QSM values are a linear measure of iron distribution. Next to that, iron is known to accumulate with age, increasing
exponentially until a plateau is reached around age 30 (Aquino et al., 2009; Daugherty & Raz, 2013; Hallgren & Sourander, 1958). Measuring iron distribution could be useful in detecting abnormal iron accumulation, which is indicative of brain diseases such as Huntington’s, Parkinson’s and Alzheimer’s (Bartzokis et al., 1999; Haacke et al., 2005; Vymazal et al., 1999; Ye, Allen, & Martin, 1996).
Furthermore, knowing that the rate of iron accumulation depends on age may be useful for future segmentation. The iron distribution and therefore QSM values in GP segments could differ significantly after all, depending on age. Future research has yet to address this point.
In conclusion, findings from the present study provide evidence for the interindividual variability in the GPe and GPi and for QSM being superior to FLASH in imaging the GPi. In addition a probability map for both GPe and GPi was provided.
17 ACKNOWLEDGEMENTS
This research was done at the Cognitive Science Center Amsterdam. Many thanks to Max C. Keuken and Birte U. Forstmann for the supervision and the opportunity to do this research.
18 REFERENCES
Abosch, A., Yacoub, E., Ugurbil, K., & Harel, N. (2010). An assessment of current brain targets for deep brain stimulation surgery with susceptibility-weighted imaging at 7 tesla. Neurosurgery, 67(6), 1745–56; discussion 1756. doi:10.1227/NEU.0b013e3181f74105
Ahsan, R. L., Allom, R., Gousias, I. S., Habib, H., Turkheimer, F. E., Free, S., Lemieux, L., et al. (2007). Volumes, spatial extents and a probabilistic atlas of the human basal ganglia and thalamus.
NeuroImage, 38(2), 261–70. doi:10.1016/j.neuroimage.2007.06.004
Amunts, K., Schleicher, a, Bürgel, U., Mohlberg, H., Uylings, H. B., & Zilles, K. (1999). Broca’s region revisited: cytoarchitecture and intersubject variability. The Journal of comparative neurology,
412(2), 319–41. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/10441759
Anastasi, G., Cutroneo, G., Tomasello, F., Lucerna, S., Vitetta, A., Bramanti, P., Di Bella, P., et al. (2006). In vivo basal ganglia volumetry through application of NURBS models to MR images. Neuroradiology,
48(5), 338–45. doi:10.1007/s00234-005-0041-4
Aquino, D., Bizzi, A., Grisoli, M., Garavaglia, B., Bruzzone, M. G., Savoiardo, M., & Chiapparini, L. (2009). Age-related Iron Deposition in the Basal Ganglia : Quantitative Methods : Results : Conclusion :
Radiology, 252(1).
Bartzokis, G., Cummings, J. L., Markham, C. H., Marmarelis, P. Z., Treciokas, L. J., Tishler, T. A., Marder, S. R., et al. (1999). q Original Contribution, 17(2), 213–222.
Cohen, J. (1960). A Coefficient of Agreement for Nominal Scales. Educational and Psychological
Measurement, 20(1), 37–46. doi:10.1177/001316446002000104
Daugherty, A., & Raz, N. (2013). Age-related differences in iron content of subcortical nuclei observed in vivo: A meta-analysis. NeuroImage, 70, 113–21. doi:10.1016/j.neuroimage.2012.12.040
Deistung, A., Schäfer, A., Schweser, F., Biedermann, U., Turner, R., & Reichenbach, J. R. (2012). Toward in vivo histology: A comparison of quantitative susceptibility mapping (QSM) with magnitude-, phase-, and R(2)(⁎)-imaging at ultra-high magnetic field strength. NeuroImage, 65, 299–314. doi:10.1016/j.neuroimage.2012.09.055
Dice, L. R. (1945). Measures of the Amount of Ecologic Association Between Species. Ecology1, 26(3), 297–302.
Evans, A. C., Janke, A. L., Collins, D. L., & Baillet, S. (2012). Brain templates and atlases. NeuroImage,
62(2), 911–22. doi:10.1016/j.neuroimage.2012.01.024
Forstmann, B. U., Keuken, M. C., Jahfari, S., Bazin, P.-L., Neumann, J., Schäfer, A., Anwander, A., et al. (2012). Cortico-subthalamic white matter tract strength predicts interindividual efficacy in stopping a motor response. NeuroImage, 60(1), 370–5. doi:10.1016/j.neuroimage.2011.12.044
19 Haacke, E. M., Cheng, N. Y. C., House, M. J., Liu, Q., Neelavalli, J., Ogg, R. J., Khan, A., et al. (2005).
Imaging iron stores in the brain using magnetic resonance imaging. Magnetic resonance imaging,
23(1), 1–25. doi:10.1016/j.mri.2004.10.001
Haase, A., Frahm, J., Matthaei, D., Hänicke, W., & Merboldt, K.-D. (1986). FLASH imaging: rapid NMR imaging using low flip-angle pulses. 1986. Journal of magnetic resonance (San Diego, Calif. : 1997),
67, 258–266. doi:10.1016/j.jmr.2011.09.021
Haber, S. N. (2003). The primate basal ganglia: parallel and integrative networks. Journal of Chemical
Neuroanatomy, 26(4), 317–330. doi:10.1016/j.jchemneu.2003.10.003
Hallgren, B., & Sourander, P. (1958). Hallgren, 1958.pdf. Journal of Neurochemistry, 3, 41–51.
Heckers, S., Heinsen, H., Heinsen, Y., & Beckmann, H. (1991). Cortex, white matter, and basal ganglia in schizophrenia: a volumetric postmortem study. Biological psychiatry, 29(6), 556–66. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/2054430
Keuken, M. C., Bazin, P.-L., Schäfer, A., Neumann, J., Turner, R., & Forstmann, B. U. (2013). Ultra-high 7T MRI of structural age-related changes of the subthalamic nucleus. The Journal of neuroscience : the
official journal of the Society for Neuroscience, 33(11), 4896–900.
doi:10.1523/JNEUROSCI.3241-12.2013
Langkammer, C., Schweser, F., Krebs, N., Deistung, A., Goessler, W., Scheurer, E., Sommer, K., et al. (2012). Quantitative susceptibility mapping (QSM) as a means to measure brain iron? A post mortem validation study. NeuroImage, 62(3), 1593–9. doi:10.1016/j.neuroimage.2012.05.049 Lenglet, C., Abosch, A., Yacoub, E., De Martino, F., Sapiro, G., & Harel, N. (2012). Comprehensive in vivo
mapping of the human basal ganglia and thalamic connectome in individuals using 7T MRI. PloS
one, 7(1), e29153. doi:10.1371/journal.pone.0029153
Mai, J. K., Paxinos, G., & Voss, T. (2008). Atlas of the Human Brain (Third., pp. 1–271). New York, London, Burlington, San Diego: Academic Press.
Marques, J. P., Kober, T., Krueger, G., Van der Zwaag, W., Van de Moortele, P.-F., & Gruetter, R. (2010). MP2RAGE, a self bias-field corrected sequence for improved segmentation and T1-mapping at high field. NeuroImage, 49(2), 1271–81. doi:10.1016/j.neuroimage.2009.10.002
Maubon, a J., Ferru, J. M., Berger, V., Soulage, M. C., DeGraef, M., Aubas, P., Coupeau, P., et al. (1999). Effect of field strength on MR images: comparison of the same subject at 0.5, 1.0, and 1.5 T.
Radiographics : a review publication of the Radiological Society of North America, Inc, 19(4), 1057–
67. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/10464808
Mazziotta, J. C., Woods, R., Iacoboni, M., Sicotte, N., Yaden, K., Tran, M., Bean, C., et al. (2009). The myth of the normal, average human brain--the ICBM experience: (1) subject screening and eligibility.
20 Nambu, A. (2007). Globus pallidus internal segment. Progress in brain research, 160(06), 135–50.
doi:10.1016/S0079-6123(06)60008-3
Nambu, A., Tokuno, H., & Takada, M. (2002). Functional significance of the cortico-subthalamo-pallidal “hyperdirect” pathway. Neuroscience research, 43(2), 111–7. Retrieved from
http://www.ncbi.nlm.nih.gov/pubmed/12067746
Nölte, I. S., Gerigk, L., Al-Zghloul, M., Groden, C., & Kerl, H. U. (2012). Visualization of the internal globus pallidus: sequence and orientation for deep brain stimulation using a standard installation protocol at 3.0 Tesla. Acta neurochirurgica, 154(3), 481–94. doi:10.1007/s00701-011-1242-8
Péran, P., Cherubini, A., Luccichenti, G., Hagberg, G., Démonet, J.-F., Rascol, O., Celsis, P., et al. (2009). Volume and iron content in basal ganglia and thalamus. Human brain mapping, 30(8), 2667–75. doi:10.1002/hbm.20698
Peterson, B. S., Riddle, M. A., Cohen, D. J., Katz, L. D., Smith, J. C., & Leckman, J. F. (1993). Original Contribution. Magnetic Resonance Imaging, 11, 493–498.
Schweser, F., Deistung, A., Lehr, B. W., & Reichenbach, J. R. (2011). Quantitative imaging of intrinsic magnetic tissue properties using MRI signal phase: an approach to in vivo brain iron metabolism?
NeuroImage, 54(4), 2789–807. doi:10.1016/j.neuroimage.2010.10.070
Schweser, F., Deistung, A., Sommer, K., & Reichenbach, J. R. (2012). Toward online reconstruction of quantitative susceptibility maps: Superfast dipole inversion. Magnetic resonance in medicine :
official journal of the Society of Magnetic Resonance in Medicine / Society of Magnetic Resonance in Medicine, 000(January). doi:10.1002/mrm.24405
Shrout, P. E., & Fleiss, J. L. (1979). Intraclass Correlations: Uses in Assessing Rater Reliability.
Psychological Bulletin, 86(2), 420–428.
Utter, A. a, & Basso, M. a. (2008). The basal ganglia: an overview of circuits and function. Neuroscience
and biobehavioral reviews, 32(3), 333–42. doi:10.1016/j.neubiorev.2006.11.003
Vasques, X., Cif, L., Hess, O., Gavarini, S., Mennessier, G., & Coubes, P. (2009). Stereotactic model of the electrical distribution within the internal globus pallidus during deep brain stimulation. Journal of
computational neuroscience, 26(1), 109–18. doi:10.1007/s10827-008-0101-y
Vymazal, J., Righini, a, Brooks, R. a, Canesi, M., Mariani, C., Leonardi, M., & Pezzoli, G. (1999). T1 and T2 in the brain of healthy subjects, patients with Parkinson disease, and patients with multiple system atrophy: relation to iron content. Radiology, 211(2), 489–95.
Wharton, S., Schäfer, A., & Bowtell, R. (2010). Susceptibility mapping in the human brain using
threshold-based k-space division. Magnetic resonance in medicine : official journal of the Society of
Magnetic Resonance in Medicine / Society of Magnetic Resonance in Medicine, 63(5), 1292–304.
21 Yablonskiy, D., & Haacke, E. M. (1994). Yablonskiy, Theory of NMR Signal Behavior.pdf. MRM, 32, 749–
763.
Yao, B., Li, T.-Q., Van Gelderen, P., Shmueli, K., De Zwart, J. a., & Duyn, J. H. (2012). Corrigendum to “Susceptibility contrast in high field MRI of human brain as a function of tissue iron content” [NeuroImage 44/4 (2009) 1259–1266]. NeuroImage, 62(3), 2173.
doi:10.1016/j.neuroimage.2012.03.055
Ye, F. Q., Allen, P. S., & Martin, W. R. (1996). Basal ganglia iron content in Parkinson’s disease measured with magnetic resonance. Movement disorders : official journal of the Movement Disorder Society,
