Structural brain data is key for the understanding of brain function and networks, i.e., connectomics. Here we present data sets available from the ‘atlasing of the basal ganglia (ATAG)’ project, which provides ultra-high resolution 7 Tesla (T) magnetic resonance imaging (MRI) scans from young, middle-aged, and elderly participants. The ATAG data set includes whole-brain and reduced field-of-view MP2RAGE and T2*-weighted scans of the subcortex and brainstem with ultra-high resolution at a sub-millimeter scale. The data can be used to develop new algorithms that help building high-resolution atlases both relevant for the basic and clinical neurosciences. Importantly, the present data repository may also be used to inform the exact positioning of electrodes used for deep-brain-stimulation in patients with Parkinson’s disease and neuropsychiatric diseases.
Machine-accessible metadata file describing the reported data (ISA-tab format)
Background & Summary
Large collaborative projects between scientific groups spread around the world are aimed to increase our understanding of the human brain. Large human connectome studies1,
Recent exciting advancements in the field of ultra-high resolution magnetic resonance imaging at 7 Tesla (or higher) allow in vivo neuroimaging of the human brain with unprecedented anatomical detail7,
For the acquisition of the structural brain scans, 30 young participants (14 females) with mean age 23.8 (s.d. 2.3), 14 middle-aged (7 females) with mean age 52.5 (s.d. 6.6), and 10 elderly (3 females) with mean age 69.6 (s.d. 4.6) were included (Table 1). All participants had normal or corrected-to-normal vision, and none of them suffered from neurological, psychiatric, or somatic diseases. All subjects were right-handed, as confirmed by the Edinburgh Inventory15. The study was approved by the local ethics committee at the University of Leipzig, Germany. All participants gave their written informed consent prior to scanning and received a monetary compensation.
The structural data were acquired using a 7 T Siemens Magnetom MRI scanner using a 24-channel head array Nova coil (NOVA Medical Inc., Wilmington MA) and consisted of three sequences: a whole-brain MP2RAGE, a MP2RAGE covering a smaller slab16,17, and a multi-echo 3D FLASH18. The whole-brain MP2RAGE had 240 sagittal slices with an acquisition time of 10:57 min (repetition time (TR)=5,000 ms; echo time (TE)=2.45 ms; inversion times TI1/TI2=900/2,750 ms; flip angle=5°/3°; bandwidth=250 Hz/Px; voxel size=(0.7 mm)3; Table 2 (available online only)). The MP2RAGE slab consisted of 128 slices with an acquisition time of 9:07 min (TR=5,000 ms; TE=3.71 ms; TI1/TI2=900/2,750 ms; flip angle=5°/3°; bandwidth=240 Hz/Px; voxel size=(0.6 mm)3; Table 3 (available online only)). The FLASH slab consisted of 128 slices with an acquisition time of 17:18 min (TR=41 ms and three different echo times (TE): 11.22/20.39/29.57 ms; flip angle=14°; bandwidth=160 Hz/Px; voxel size=(0.5 mm)3; Table 4 (available online only)). Both slab sequences consisted of axial slices tilted −23 degrees to the true axial plane in scanner coordinates. This angle in combination with the used field of view ensured that the entire Basal Ganglia were scanned. To get a better inversion of the magnetization in the lower parts of the brain (e.g., the Cerebellum), a TR-FOCI inversion pulse was implemented in the MP2RAGE sequence16.
Unless indicated otherwise, all MRI data files were converted from DICOM to NIfTI format using an in-house dicom-to-nifti converter. This linux compatible converter is available via https://github.com/isis-group/isis.
The MP2RAGE sequence results in four different volumes for each subject: INV1, INV2, UNI and T1. The INV1 volume reflects the gradient echo sequence with an inversion time of 900 ms. The INV2 volume reflects the gradient echo sequence with an inversion time of 2,750 ms. The UNI volume is the combined volume of the two inversion times. Finally, the T1 volume is a T1 estimation map derived from the two inversion times (Marques et al.17). The FLASH sequence results in two different volumes per echo time per subject resulting in nine different volumes in total. Besides the standard T2* weighted magnitude image, the phase images are also provided and can be used to calculate susceptibility weighted images as well as quantitative susceptibility maps (e.g., Deistung et al.19).
All structural scans were anonymized by zeroing out the voxels in the vicinity of the facial surface, teeth, and auricles following a similar procedure as described by Hanke et al.20 All data were reoriented to the standard MNI space using the fslreorient2std tool as implemented in fslutils 5.0.2 (Figure 1).
All data records listed in this section are available from NITRIC (Data Citation 1: NITRC http://www.nitrc.org/projects/atag_mri_scans/) or Dryad (Data Citation 2: Dryad http://doi.org/10.5061/dryad.fb41s). A README file with a detailed description of the content of all downloads is available in Dryad. Additional material and information are also provided in Data Citation 1: NITRC http://www.nitrc.org/projects/atag_mri_scans/ and Data Citation 2: Dryad http://doi.org/10.5061/dryad.fb41s.
Unless noted otherwise, all MRI data files were converted from DICOM to NIfTI format using an in-house dicom-to-nifti converter. In order to de-identify data, information on centre-specific study and participant codes have been removed using an automated procedure. All human participants were given sequential integer IDs.
In line with Gedamu et al.21, motion artifacts in the structural volumes were estimated by calculating the noise ratio between the phase encoding direction and read direction outside of the brain. Two ROIs of +/−1,225 mm2 was drawn in the sagittal plane; 5 mm lateral of the skull, and in the coronal plane; 5 mm anterior of the skull, in the magnitude image of the second inversion time of the MP2RAGE sequence and FLASH sequences. The sagittal ROI corresponds to the read direction for the MP2RAGE whole brain and phase encoding direction for the MP2RAGE and FLASH slab, whereas the coronal ROI corresponds to the phase encoding direction for the MP2RAGE whole brain and read direction for the MP2RAGE and FLASH slab. The mean signal was extracted from both ROI’s and the mean phase encoding direction signal was divided by the mean read direction signal. The closer this ratio is to 1, the less motion artifacts are present. Following Gedamu et al.21, we estimated that any ratio below 2 reflects little to no motion artifacts (see Figure 2 for an example of the data quality).
One sided t-tests were conducted to test whether any of the groups showed significant motion artifacts in any of the sequences. All ratios per sequence and age group were significantly lower than 2 (MP2RAGE whole-brain: young (t(29)=−17.93, P<0.001); middle-aged (t(13)=−5.44, P<0.001); elderly (t(8)=−7.19, P<0.001), MP2RAGE slab: young (t(29)=−35.06, P<0.001); middle-aged (t(13)=−23.43, P<0.001); elderly (t(8)=−13.33, P<0.001), FLASH echo 1: young (t(29)=−3.74, P<0.001); middle-aged (t(13)=−17.68, P<0.001); elderly (t(8)=−16.97, P<0.001), FLASH echo 2: young (t(29)=−6.88, P<0.001); middle-aged (t(13)=−14.88, P<0.001); elderly (t(8)=−6.31, P<0.001), FLASH echo 3: young (t(29)=−10.36, P<0.001); middle-aged (t(13)=−6.23, P<0.001); elderly (t(8)=−19.53, P<0.001); Table 5 (available online only)).
There was no main effect of age on motion for the MP2RAGE whole brain (F(2,50)=1.29, P=0.29) or MP2RAGE slab (F(2,50)=0.8, P=0.44). There was a main effect of age and echo time on motion for the FLASH sequence (age: F(2,147)=4.97, P=0.008, echo time: F(2,147)=10.45, P<0.001). Post-hoc testing showed that the young had significantly more motion artifacts than both the middle-aged and elderly (young versus middle-aged: t(103.18)=5.61, P<0.001, young versus elderly: t(79.03)=5.25, P<0.001) whereas the middle-aged and elderly did not differ significantly (t(65.73)=−0.59, P=1.0). Post-hoc testing showed that the first echo time had significantly less motion artifacts than both the second and third echo time (first echo versus second echo: t(90)=−3.29, P=0.003, first echo versus third echo: t(82.17)=−3.77, P=0.001) whereas the second and third echo time did not differ significantly (t(101.66)=−0.72, P=0.92). All post-hoc testing was Bonferroni corrected at an alpha of 0.05.
Signal to noise ratio
To estimate the Signal to Noise Ratio (SNR), the mean signal from an axial slice just above the corpus callosum was divided by the standard deviation of the signal in the read direction ROI both in the magnitude image of the second inversion time of the MP2RAGE sequence and FLASH sequences. To improve the estimation of noise a Rician correction was applied22. As this is still an approximation of the true SNR, the term SNRapprox. is used. For the three different sequences there was a main effect of age on SNRapprox. (MP2RAGE whole brain: F(2,50)=48.3, P<0.001; MP2RAGE slab: F(2,50)=5.94, P=0.005; FLASH: F(2,147)=6.90, P=0.001). Additionally there was a main effect of echo time on SNRapprox. (F(2,147)=11.75, P<0.001).
Post-hoc testing showed that for the MP2RAGE whole brain, the young had a significantly higher SNRapprox than both the middle-aged and elderly (young versus middle-aged: t(33.72)=8.87, P<0.001; young versus elderly: t(18.37)=8.41, P<0.001) whereas the middle-aged and elderly did not differ significantly (t(17.61)=0.46, P=1.0). A similar pattern was found for the MP2RAGE slab. The young had a significantly higher SNRapprox than both the middle-aged and elderly (young versus middle-aged: t(24.8)=2.86, P=0.017; young versus elderly: t(17.46)=2.92, P=0.019) whereas the middle-aged and elderly did not differ significantly (t(20.56)=−0.10, P=1.0). The young had a significantly higher SNRapprox in the FLASH sequence than the middle-aged (t(70.80)=3.35, P=0.003) but did not differ from the elderly (t(36.31)=0.16, P=1.0). The middle-aged and elderly did not differ in SNRapprox for the FLASH sequence (t(51.87)=−2.16, P=0.071). Post-hoc testing showed that the first echo time had significantly more SNRapprox than both the second and third echo time (first echo versus second echo: t(97.2)=4.89, P<0.001, first echo versus third echo: t(88.4)=8.05, P<0.001). The second echo time had significantly higher SNRapprox than the third echo time (t(100.91)=3.42, P=0.002). All post-hoc testing was Bonferroni corrected at an alpha of 0.05 (Table 6 (available online only)).
In addition to the SNRapprox. calculation and the noise ratio between the phase encoding direction and read direction, the scans were visually inspected by two independent researchers. The FLASH magnitude scans were checked for ghosting, wrapping, or shading artifacts. The MP2RAGE UNI scans were checked for ghosting, wrapping, shading, and the presence of ‘zebra stripe’ artifacts. Finally the MP2RAGE T1 scans were checked for ghosting, wrapping, shading, the presence of ‘zebra stripes’, and CSF clipping artifacts where ‘1’ corresponds to not present at all and ‘5’ corresponds to severely present.
Ghosting artifacts are generally caused by motion and appear as a ‘ghost’ image of the brain in phase encoding direction. Wrapping artifacts are usually caused by anatomical features protruding outside of the imaged field of view but still within the sensitive volume of the RF coil. Shading artifacts were defined as a non-homogenous intensity throughout the entire brain. Zebra stripes were defined as well defined alternating black and white stripes present in the brain. Finally, CSF clipping artifacts were defined as the voxels in the CSF that have a signal dropout and appear black (McRobbie et al., 2006).
The mean rating for each scale for each checked volume is given in Table 7 (available online only). Volumes that had a higher rating on that quality check than the rest of the age group based on the +/−1.5* interquartile range are highlighted with an asterisk.
As a result of the scan parameters of the MP2RAGE sequence, a number of participants show T1 clipping artifacts in the T1 map located in the CSF. This is indicated in Table 7 (available online only). Note that these clipping artifacts do not affect the T1 values reported in the grey and white matter tissue.
The procedures we employed in this study resulted in a dataset that is highly suitable for automated processing. Data are shared in documented standard formats, such as NIfTI or plain text files, to enable further processing in arbitrary analysis environments with no imposed dependencies on proprietary tools. All processing performed on the released data article were produced by open-source software on standard computer workstations.
Forstmann, B. U. Dryad http://doi.org/10.5061/dryad.fb41s (2014)
We thank Domenica Wilfling and Elisabeth Wladimirov for taking such good care of all our participants. This research line is financially supported by the European Research Council (BUF).
About this article
Human Brain Mapping (2019)