PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of wtpaEurope PMCEurope PMC Funders GroupSubmit a Manuscript
 
Neuroimage. Author manuscript; available in PMC 2017 March 8.
Published in final edited form as:
PMCID: PMC5341769
EMSID: EMS64917

Voxel-based morphometry at ultra-high fields. A comparison of 7T and 3T MRI data

Abstract

Recent technological progress enables MRI recordings at ultra-high fields of 7 Tesla and above leading to brain images of higher resolution and increased signal-to-noise ratio. Despite these benefits, imaging at 7T exhibits distinct challenges due to B1 field inhomogeneities, causing decreased image quality and problems in data analysis. Although several strategies have been proposed, a systematic investigation of bias-corrected 7T data for voxel-based morphometry (VBM) is still missing and it is an ongoing matter of debate if VBM at 7T can be carried out properly. Here, an optimized VBM study was conducted, evaluating the impact of field strength (3T vs 7T) and pulse sequence (MPRAGE vs MP2RAGE) on gray matter volume (GMV) estimates. More specifically, twenty-two participants were measured under the conditions 3T MPRAGE, 7T MPRAGE and 7T MP2RAGE. Due to the fact that 7T MPRAGE data exhibited strong intensity inhomogeneities, an alternative preprocessing pipeline was proposed and applied for that data. VBM analysis revealed higher GMV estimates for 7T predominantly in superior cortical areas, caudate nucleus, cingulate cortex and the hippocampus. On the other hand, 3T yielded higher estimates especially in inferior cortical areas of the brain, cerebellum, thalamus and putamen compared to 7T. Besides minor exceptions, these results were observed for 7T MPRAGE as well for the 7T MP2RAGE measurements. Results gained in the inferior parts of the brain should be taken with caution, as native GM segmentations displayed misclassifications in these regions for both 7T sequences. This was supported by the test-retest measurements showing highest variability in these inferior regions of the brain for 7T also for the advanced MP2RAGE sequence. Hence, our data support the use of 7T MRI for VBM analysis in cortical areas, but direct comparison between field strengths and sequences requires careful assessment. Similarly, analysis of inferior cortical regions, cerebellum and subcortical regions still remains challenging at 7T even if the advanced MP2RAGE sequence is used.

Keywords: VBM, Ultra-high field, 7 Tesla, MP2RAGE, MPRAGE, Test-Retest

1. Introduction

Structural magnetic resonance imaging (sMRI) has become a reliable and well-established research method for the detailed assessment of anatomical brain data in vivo. Normal brain development as well as brain abnormalities can be studied by comparing different study populations of interest (May & Gaser 2006). A regularly applied method for such investigations is voxel-based morphometry (VBM), where local volume or concentration of gray matter is measured by performing a voxel-wise comparison between or within groups (Ashburner & Friston 2000; Wright et al. 1995). Several studies have applied this technique to assess structural brain changes in terms of aging (Good et al. 2001; Draganski et al. 2011), brain pathology (Nugent et al. 2006; Teipel et al. 2005; van Tol et al. 2014), or neuroplasticity (Kraus et al. 2014; Maguire et al. 2000). Routinely, a T1-weighted, magnetization-prepared rapid gradient echo (MPRAGE) sequence (Mugler & Brookeman 1990) at a field strength of 3 Tesla (3T) is used, as this sequence achieves excellent image contrast between gray matter (GM), white matter (WM) and the cerebrospinal fluid (CSF) (van der Kouwe et al. 2008). Technological progress during the last few years now enables MRI recordings at ultra-high fields of 7 Tesla (7T) and above, leading to brain images of higher resolution and to a substantial increase in the signal-to-noise ratio (Hahn et al. 2013; Sladky et al. 2013; Bazin et al. 2013). Despite these benefits, structural imaging at 7T exhibits distinct drawbacks, such as intensity inhomogeneities, which cause severe problems in automated MRI data analysis (Belaroussi et al. 2006). This so-called bias field, generated by the inhomogeneities of the transmit B1+ and receive B1- fields at increased high static magnetic fields (B0), leads to intensity variations across the entire brain. While reception B1− inhomogeneities are easily removed, transmission B1+ field inhomogeneities are more severe, as they effectively change the contrast (Marques et al. 2010). These inhomogeneities strongly affect image quality and impede the process of image segmentation and quantitative data analysis at ultra-high fields. Several strategies have been suggested to account for this problem (Marques et al. 2010), whereby a method proposed by van de Moortele showed promising results (Van de Moortele et al. 2009). Following this approach, a separate proton density weighted 3D gradient echo (GRE) image is acquired in addition to the MPRAGE image aiming for bias field reduction. Another sophisticated way of dealing with these inhomogeneities is by using the magnetization-prepared 2 rapid acquisition gradient echo, or MP2RAGE, pulse sequence (Marques et al. 2010), which is a modified version of the conventional MPRAGE sequence adopting the approach of van de Moortele by integrating the acquisition of the 3D GRE image. The spatially uniform contrast of the MP2RAGE sequence is achieved by a rapid acquisition of the two volumes at different inversion times, where the images are afterwards combined and sources of inhomogeneities are compensated for. While the first inversion time is used to produce a T1-weighted image, the second inversion time is long and therefore produces an approximate proton density-weighted contrast. The combination of these two images delivers a synthetic image with strong contrast between the different tissue types across the entire brain. Hence, strategies for imaging at 7T exist and studies for cortical thickness measurements have already been conducted (Fujimoto et al. 2014; Lüsebrink et al. 2013). However, VBM utilizes a different approach in assessing brain anatomical changes compared to cortical thickness measurements and it still remains a matter of debate if VBM analyses can be carried out at 7T ultra-high fields properly in all brain areas. To address this issue, whole-brain VBM analysis was conducted using ultra-high field data (7T) and comparison to the 3T standard was carried out to assess reliability for each brain region. Furthermore, two different pulse sequences (MPRAGE, MP2RAGE) were tested at 7T to observe their influence on gray matter volume estimates. In addition, test-retest analysis at two time-points was conducted to support reliability of gained VBM results. Taken together, we aimed to provide a thorough analysis for which brain regions VBM yields reliable results for ultra-high fields and the respective sequence. Our investigations will be useful for further VBM studies at ultra-high fields to draw attention to brain areas, which are problematic at 7T.

2. Materials and Methods

2.1. Participants

22 healthy subjects (mean age ± SD = 26.5 ± 6.2 years, 13 females) without any neurological or psychiatric disorders were included in this study and scanned at 3T (MPRAGE) and 7T (MPRAGE and MP2RAGE). Out of these 22 participants 10 subjects (mean age ± SD = 26.36 ± 7.3, 6 females) underwent the same procedure at a second time point to assess test-retest reliability measurements (mean days between measurements ± SD = 81 days ± 49). All participants were recruited via advertisement at the Medical University of Vienna, Austria and underwent a general physical and neurological examination at the screening visit including medical history, electrocardiogram and routine laboratory tests. Inclusion criteria were age between 18 and 50, general health based on history, physical examination electrocardiogram, laboratory screening and Structured Clinical Interview (SCID I & II) for DSM IV. Exclusion criteria comprised any severe diseases, implants or metal parts, current substance abuse and pregnancy. All participants provided written informed consent after written and oral presentation of an information form and they received reimbursement after participation. The study was approved by the Ethics Committee of the Medical University of Vienna and procedures were performed according to the Declaration of Helsinki.

2.2. Data acquisition

Structural MRI was carried out at the MR Center of Excellence at the Medical University of Vienna, Austria, with a 3 Tesla whole-body scanner (Siemens Tim Trio, Erlangen, Germany) and an ultra-high field whole-body 7T MRI scanner (Siemens Magnetom) using 32-channel head coils at both scanners. Participants were measured at 3T (MPRAGE, T1; 256x240 matrix, 160 slices, voxel size 1x1x1.1 mm3, TE=4.21 ms, TR=2300 ms; TI=900 ms; α=9°; total acquisition time 7 min, 46 sec) and 7T with a MPRAGE sequence (T1; 320x310 matrix, 224 slices, voxel size 0.7x0.7x0.74 mm3, TE=3.66 ms, TR=1980 ms; TI=900 ms; α=9°; total acquisition time 8 min, 58 sec) and a MP2RAGE sequence (T1; 384x312 matrix, 192 slices, voxel size 0.68x0.68x0.74 mm3, TE=3.07 ms, TR=4096 ms; TI1=900 ms; TI2=3190 ms; α1=4°; α2=5°; total acquisition time 11 min, 20 sec) using optimized parameters for each sequence (Figure 1). MP2RAGE images were calculated from images acquired with inversion times TI1 and TI2 following the procedures of (Marques et al. 2010). Second-order shimming was performed for both MPRAGE and MP2RAGE sequences on both scanners prior to data acquisition. The scanners used in this study were calibrated by the manufacturer (SIEMENS Healthcare, Erlangen, Germany) upon installation of the gradient systems in order to account for gradient nonlinearity effects, which otherwise might introduce image distortions (Caramanos et al. 2010).

Figure 1
Overview of a structural image of the same subject obtained with (A) 3T MPRAGE, (B) 7T MPRAGE and (C) 7T MP2RAGE. 3T is considered as standard for structural imaging so far and delivers overall good image quality. 7T MPRAGE exhibit a strong bias field, ...

2.3. Voxel-Based Morphometry

VBM analysis was carried out in SPM8 (http://www.fil.ion.ucl.ac.uk/spm) using the VBM8 toolbox (http://dbm.neuro.uni-jena.de/vbm/) and MATLAB 7.12 (MathWorks, Natick, MA). After an initial bias correction, images were segmented into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF), registered to MNI space and spatially normalised via the Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra (DARTEL) algorithm (Ashburner 2007). GM segments were modulated with the option 'non linear only' accounting for the effect of spatial normalization, resulting in an analysis of relative differences in regional gray matter volume, already corrected for individual brain size. Subsequently, GM volumes were smoothed with an isotropic Gaussian kernel of 8mm (FWHM). 3T MPRAGE and 7T MP2RAGE data were processed using SPM8 and the VBM8 toolbox with default parameters. Although the 'uniform' 7T MP2RAGE image exhibits a high amount of artificial background noise, the segmented gray matter volumes delivered good results and segmentation quality was not affected. Hence, the algorithm performed robustly despite the noisy background and therefore no binary masking had to be applied.

2.4. Bias field correction for 7T MPRAGE data

For the acquired 7T MPRAGE data an adapted bias correction pipeline has been used to account for the strong intensity inhomogeneities. As no additional gradient echo image (GE) has been recorded along with the 7T MPRAGE sequence to account for the bias field (Van de Moortele et al. 2009), we used an optimized preprocessing pipeline for that data (Figure 2). First, nonparametric nonuniform intensity normalisation (N3) algorithm (Sled et al. 1998) with optimized parameters for 7T data (proto-iters=1000, distance=15, n=1) was used (Lüsebrink et al. 2013), which is implemented within the FreeSurfer software package (Dale et al. 1999; Fischl & Dale 2000), (http://surfer.nmr.mgh.harvard.edu/). This inhomogeneity correction algorithm uses an iterative approach, is independent of the pulse sequence and can be applied before standard preprocessing is conducted (Figure 3). After that step, data was processed with the VBM8 pipeline using adapted bias correction values (bias regularisation: extremely light (default: very light); bias FWHM cutoff: 30mm (default: 60mm)) to deal with the remaining intensity inhomogeneities.

Figure 2
VBM processing pipeline for all three conditions is shown. T1-weighted images are processed with the VBM8 toolbox, including bias correction, segmentation, registration to MNI space and DARTEL normalisation. This is followed by a final smoothing step. ...
Figure 3
Alternative preprocessing pipeline utilizing the nonparametric nonuniformity normalization (N3) algorithm and adapted bias correction values for 7T MPRAGE. (A) Here, a single subject in native space is shown using standard preprocessing steps. Strong ...

2.5. Statistical analysis

A repeated measures ANOVA was carried out in SPM8 with post-hoc paired t-tests between the three conditions. Here, the default gray matter mask of the VBM8 toolbox was used to exclude irrelevant voxels from the statistical analyses. All data reported were corrected for multiple comparisons using a voxel-level threshold of p<0.05 family-wise error corrected (FWE).

To test the reliability of the VBM results obtained at two different time points, two distinct metrics were used. We evaluated test-retest variability (TRV) and consistency using the intra-class correlation coefficient (ICC). While TRV quantifies the variability of a measurement, the ICC is an index for reliability. For the assessment of the different regions of interest the automated anatomical labeling (AAL) atlas (Tzourio-Mazoyer et al. 2002) was used. Originally, the AAL atlas consists of 116 labels. For purpose of clarification and better overview, we merged the left and right hemispheres as we expected no hemispheric influences on GMV values due to a second measurement. Further, all cerebellar subparts have been merged to a unified cerebellum region. Finally, 46 labels were generated and a test-retest analysis of 10 subjects were conducted for all three measurements (3T MPRAGE, 7T MPRAGE and 7T MP2RAGE) at two distinct time points (Supplement I, Table 1). To assess reliability of VBM results, test-retest variability (TRV) was evaluated by using the equation:

%TRV=|M1M2|(M1+M2)/2*100

where M1 indicates the first (test) and M2 the second (retest) measurement (Yoder et al. 2011). In addition, we calculated Pearson’s correlation coefficient. The test-retest consistency was then assessed by using the intraclass correlation coefficient (ICC). ICC values were calculated using IBM SPSS statistics 19 for Windows using a two factor mixed effects model and the type consistency (McGraw & Wong 1996; Shrout & Fleiss 1979). Values are in the range between 0 and 1, where an ICC of 0 indicates no reliability and 1 indicates perfect reliability (Weir 2005). Hence, values closer to 1 indicate that most of the variance can be explained by between-subject variation, meaning high consistency of M1 and M2.

3. Results

3.1. VBM Analysis

Repeated measures ANOVA revealed a significant difference between the conditions (3T MPRAGE, 7T MPRAGE and 7T MP2RAGE) F(2,42)=18.32 (threshold value). Subsequently conducted post-hoc paired t-tests (Comparisons have also be carried out by means of test-retest measurements, see supplementary material) yielded the following results (Figure 4):

Figure 4
Post-hoc paired t-test results of VBM analysis of conditions (A) 3T MPRAGE > 7T MPRAGE, (B) 3T MPRAGE > 7T MP2RAGE and (C) 7T MPRAGE > 7T MP2RAGE. Cross hair at x=-13, y=-22, z=9 mm in MNI space (t>=5.23; p<0.05 ...

3T MPRAGE vs 7T MPRAGE

The comparison between 3T and 7T MPRAGE data (Figure 4A) revealed higher gray matter volume estimates for 3T within the thalamus (t=33.71), putamen (t=31.65), fusiform gyrus (t=23.15), insula (t=16.87), inferior frontal gyrus (t=17.36), inferior temporal gyrus (t=16.41), middle temporal gyrus (t=15.47), the cerebellum (t=15.24) and to some extent in the superior occipital gyrus (t=8.74). 7T MPRAGE displayed higher GMV than 3T in the caudate nucleus (t=21.38), anterior and middle cingulate cortex (t=16.65), middle frontal gyrus (t=15.12), superior frontal gyrus (t=10.71), hippocampus (t=13.58) and the supplementary motor area (t=11.50).

3T MPRAGE vs 7T MP2RAGE

Similarly, the analysis of 3T and 7T MP2RAGE data (Figure 4B) showed higher GMV for 3T in the fusiform gyrus (t=13.80), cerebellum (t=13.48), thalamus (t=12.19), inferior temporal gyrus (t=10.55), middle temporal gyrus (t=10.41), temporal pole (t=8.80), middle occipital gyrus (t=8.58), inferior occipital gyrus (t=6.38), gyrus rectus (t=11.57), middle frontal gyrus (t=8.74) and the inferior frontal gyrus (t=8.50). 7T MP2RAGE yielded higher values in the hippocampus (t=11.25), amygdala (t=9.69) middle and anterior cingulate cortex (t=7.34), superior frontal gyrus (t=9.68), supplementary motor area (t=9.38), precuneus (t=6.10) and the caudate nucleus (t=6.87).

7T MPRAGE vs 7T MP2RAGE

The comparison of the two 7T sequences (Figure 4C) indicated higher GM values for MPRAGE in the caudate nucleus (t=9.21), hippocampus (t=7.09), parahippocampal gyrus (t=7.03), cerebellum (t=8.03), inferior temporal gyrus (t=7.17), temporal pole (t=6.57) and middle frontal gyrus (t=6.72). On the contrary, MP2RAGE sequence showed higher GMV in the putamen (t=25.31), thalamus (t=19.37) and posterior parts of the inferior temporal gyrus (t=7.03).

3.2. Test-Retest reliability

The test-retest variability (%TRV) values for each of the 46 ROIs are stated along with mean GMV values and standard deviation. In addition, the intraclass correlation coefficient (ICC) and the Pearson’s r were calculated (Supplement Table 1). For the purpose of clarity, we will focus on %TRV values as this metric gives us the best insight of how reliable the two measurements of a certain field strength and pulse sequence are. %TRV values varied strongly between the conditions 3T MPRAGE, 7T MPRAGE and 7T MP2RAGE and region (Figure 5). 3T MPRAGE measurements showed best overall results concerning variability across all ROIs (1.6% ± 0.8%), followed by 7T MPRAGE (4.5% ± 1.6%) and 7T MP2RAGE (5.5% ± 4.6%). However, analysis revealed that the %TRV differed strongly amongst regions.

Figure 5
Test-retest variability (%TRV) values for (A) 3T MPRAGE, (B) 7T MPRAGE and (C) 7T MP2RAGE for each of the 46 regions of interest for 10 subjects. *) value for temporal pole (middle temporal gyrus) at 7T MP2RAGE was cut as values were larger than depicted ...

3T MPRAGE

For 3T MPRAGE 78% of the 46 ROIs exhibited excellent reliability of 1-3%. Values below 1% variability were observed for calcarine fissure and surrounding cortex, hippocampus, precuneus, rolandic operculum, superior temporal gyrus and supramarginal gyrus. Above the 3% variability were the thalamus, gyrus rectus, pallidum and putamen.

7T MPRAGE

For the 7T MPRAGE condition about 70% of all ROIs were between 3-6% variability. Entire range was from 1.7% till 7.9%. Hence, the distribution was quite equal across areas. Values below 3% TRV were found for hippocampus, insula, Heschl gyrus, median cingulate and paracingulate gyri, posterior cingulate gyrus, rolandic operculum and superior temporal gyrus. Values above 7% were obtained for superior parietal gyrus, inferior occipital gyrus, cerebellum, thalamus, pallidum, temporal pole (middle temporal gyrus) and the inferior temporal gyrus.

7T MP2RAGE

7T MP2RAGE exhibited the strongest variability across the ROIs with a range from 1.5% till 23.5%. 67% of all ROIs showed variability between 2-7%. Values below 2% TRV were observed for inferior parietal gyrus, supramarginal gyrus, posterior cingulate gyrus, superior parietal gyrus, precuneus, precentral gyrus and postcentral gyrus. Highest variability (above 7% TRV) was obtained for temporal pole (middle temporal gyrus), inferior temporal gyrus, cerebellum, temporal pole (superior temporal gyrus), inferior occipital gyrus, middle temporal gyrus, fusiform gyrus and gyrus rectus.

4. Discussion

This study investigated the application of VBM for different field strengths (3T and 7T) and pulse sequences (MPRAGE and MP2RAGE). Direct comparison revealed marked differences with 7T showing higher GMV in frontal areas, but 3T exhibiting higher values especially in temporal areas. On the other hand, 3T MPRAGE showed best test-retest reproducibility, followed by 7T MPRAGE and less stable values for 7T MP2RAGE. To the best of our knowledge, no whole-brain VBM study at 7T was performed so far observing the reliability of gained results at ultra-high fields. Hence, assessment if VBM analysis can be carried out properly in all areas of the brain was highly needed.

The MPRAGE sequence, which is tailored for field strengths of 3T, was not optimal for the use in an ultra-high field setting and could not be used without any further conditioning. The problem with intensity inhomogeneities at higher fields is already well known (Belaroussi et al. 2006) and has especially severe consequence for morphometric brain analyses. With the standard processing pipeline, used at lower field strengths, strong misclassifications have been observed across the entire brain using MPRAGE at 7T, whereby, especially cortical areas showed strong signal losses (Figure 3). Hence, we argue that the application of this sequence without any further precautions and preprocessing can lead to erroneous results, as signal drop-outs were observed to exceed acceptable levels. To overcome these problems, a bias correction processing pipeline is proposed here using the N3 algorithm (Sled et al. 1998) and strongest (least regularised) bias correction values in SPM. This allowed us to account for the bias field to conduct VBM analysis in previously problematic cortical areas.

Comparison between the conditions 3T MPRAGE and 7T MPRAGE revealed higher GMV values for 7T MPRAGE especially in superior frontal parts of the cortex, caudate nucleus, cingulate cortex and the hippocampus compared 3T MPRAGE. In addition, test-retest metrics showed low variability, indicating good reliability in these brain areas. In comparison to the 3T MPRAGE sequence, which can be considered as the gold standard in MRI measurements so far, these overestimations for 7T MPRAGE are most likely due to the strong bias correction values.

The advanced MP2RAGE sequence (Marques et al. 2010; Van de Moortele et al. 2009) already accounts for the bias field and was especially developed for imaging at higher field strengths (Marques et al. 2010). Hence, no initial problems in superior cortical parts of the brain have been encountered. As for the 7T MPRAGE sequence, comparison (3T MPRAGE vs 7T MP2RAGE) showed higher GMV estimates for 7T MP2RAGE in similar areas, such as in the superior frontal parts of the cortex, caudate nucleus, cingulate cortex and the hippocampus. However, differences were not as pronounced as for the 7T MPRAGE condition. In addition, more GMV was displayed for amygdala and precuneus in comparison to 3T MPRAGE. Due to the lack of a 'ground truth' in VBM measurements, these results can only be reported in a relative way. Hence, GMV overestimations for 7T can also be interpreted as 3T underestimations. Nevertheless, results in these brain regions might be attributed to the higher resolution available at 7T, as no imaging artefacts have been observed in these areas.

These results implicate that direct comparisons with studies at 3T should be done with caution in these brain areas and that such a difference between field strengths needs to be kept in mind for instance with multicentre studies. Again, TRV analysis indicated low variability and hence good reliability in these brain areas, implying that data can still be used for longitudinal analysis or comparison across population using exclusively 7T MPRAGE or 7T MP2RAGE data for a given investigation. This might be important for clinical study designs when evaluating treatment effects or group differences.

On the contrary, GMV underestimations have been observed for both 7T conditions in comparison to 3T in inferior regions of the brain. A reduction in GMV was predominantly observed in the thalamus, inferior frontal, inferior temporal areas and the cerebellum for both sequences. 7T MPRAGE specific reductions have been observed for putamen and insula, whereas 7T MP2RAGE yielded lesser GM metrics for temporal pole and gyrus rectus. These results were also corroborated by high test-retest variability scores indicating poorer reliability in these brain regions. Hence, segmentations in inferior parts of the brain remain challenging and results should be carefully evaluated in these areas. Especially the cerebellum and the inferior temporal regions were not well segmented due to signal drop-out. This was also observed in another study using ultra-high field data, where inferior parts of the brain were excluded from the analysis (Lüsebrink et al. 2013). Further, using our proposed alternative pipeline for 7T MPRAGE data, the thalamus and putamen showed slight signal drop-outs. Again, also these areas need careful evaluation by checking possible misclassifications in the gray matter segments. Signal drop-outs for 7T MP2RAGE in comparison to 7T MPRAGE occurred not due to the bias field, but due to so-called hyperintensities represented by a bright signal across the inferior parts of the brain (Figure 6). While in sequences which employ only one TI and in which the resulting image is not calculated as a fraction of two images, hyperintensities are a result of incomplete signal inversion occurring predominantly in air-tissue border regions (Bernstein et al. 2006), the hyperintensities shown in Figure 6 are a direct consequence of the low values in the denominator of Equation 1 in (Marques et al., 2010) used for the calculation of MP2RAGE values. This loss of SNR in inferior brain regions might be caused by increased B1-transmit distortions at ultra-high fields (Adriany et al., 2005; Van de Moortele et al., 2009), which cannot be fully compensated for by the MP2RAGE sequence. Hence, problems within inferior brain regions even persist for 7T MP2RAGE.

Figure 6
Strong hyperintensities (crosshair) are visible in the region of the inferior frontal lobes (A), and inferior temporal lobes (B) using the MP2RAGE sequence at 7T. These artifacts cause severe problems during the segmentation process, leading to gray matter ...

The overall test-retest analysis revealed that reliability strongly depends on the region under investigation and on the used field strength and pulse sequence. Figure 5 gives an overview which areas delivered reliable results for a particular field strength and pulse sequence and in which region VBM results should be interpreted with caution. These findings will be useful for further VBM studies to interpret and trust gained results especially at higher field strengths (> 3 Tesla). Best overall test-retest values were observed for 3T. Surprisingly, MPRAGE revealed better test-retest scores than MP2RAGE at 7T, which can be explained by strong signal drop-outs in inferior brain regions for MP2RAGE. Especially the inferior temporal regions showed markedly high variability for 7T MP2RAGE. Hence, the hyperintensities of 7T MP2RAGE need to be overcome to enable full use of its favourable image contrast. It already has been suggested that improvements in coil design can lead to better image quality, especially in these temporal brain regions (Fujimoto et al. 2014) and that better RF-shim and parallel transmission may also potentially be a source for an improvement in image quality (Lüsebrink et al. 2013).

Another limitation for the use of ultra-high field MR data is that current computational methods are not yet designed for such high field strengths (Bazin et al. 2013). This is especially the case for VBM analyses, where high resolution data is downsampled during preprocessing within default VBM pipelines. However, cortical thickness analysis at ultra-high field of 7T has already been carried out (Lüsebrink et al. 2013) with FreeSurfer (http://surfer.nmr.mgh.harvard.edu/) using a modified pipeline for 7T data. It was shown that higher field strength, which enabled higher resolution images, led to a more accurate segmentation of the cortical mantle and therefore to lower cortical thickness values in comparison to 3T. Hence, we expect that VBM pipelines tailored for 7T with its higher resolution will make use of the full potential of higher field strengths as it will lead to more accurate results if higher resolution can be conserved during the entire processing pipeline. In addition, prospective studies are needed revealing interactions between field strengths, sequences and disease status (Marchewka et al. 2014). Again, artifacts impeding image quality have to be overcome first.

5. Conclusion

We suggest that ultra-high field data can be used for VBM analysis in most brain regions using both pulse sequences (MPRAGE, MP2RAGE), however certain issues have to be taken into consideration. Due to the strong intensity inhomogeneities when using the MPRAGE sequence at ultra-high fields, an alternative preprocessing pipeline is inevitable. The more sophisticated MP2RAGE already accounts for these inhomogeneities and can be used without further conditioning. However, like 7T MPRAGE, the 7T MP2RAGE sequence also exhibits strong artifacts in form of hyperintensities in inferior brain regions, making it challenging for both sequences in these basal areas. Furthermore, we cannot conclude that ultra-high field structural benefits are evident, still we showed that VBM is reliable and possible with both sequences at ultra-high fields in most brain regions. Computational improvements and tailored VBM pipelines for ultra-high field data will deliver the full potential of 7T data. To summarize, our data supports the use of 7T MRI for VBM analysis in cortical areas but direct comparison between field strengths and sequences requires careful assessment. Similarly, analysis of temporal lobes, cerebellum and subcortical regions still remains challenging at 7T.

Supplementary Material

Supplement, Figure 1

Difference in percent (% Δ) for the comparison of (A) 3T MPRAGE vs 7T MPRAGE, (B) 3T MPRAGE vs 7T MP2RAGE and (C) 7T MPRAGE vs 7T MP2RAGE are depicted. *) bars for thalamus were cut as values were larger than displayed scale. Higher values indicate larger differences.

Supplement, Table 1

Test-retest results of 3T MPRAGE, 7T MPRAGE and 7T MP2RAGE for each region of interest were calculated using test-retest variability (%TRV), intraclass correlation coefficient (ICC) and Pearson’s r.

Supplement, Table 2

Difference in percent (% Δ) of 3T MPRAGE vs 7T MPRAGE, 3T MPRAGE vs 7T MP2RAGE and 7T MPRAGE vs 7T MP2RAGE for each region of interest was calculated using the test-retest formula.

Acknowledgements

This research was supported by a grant of the Austrian Science Fund (FWF P23021) to R. Lanzenberger. We thank E. Akimova, M. Spies, P. Baldinger, A. Höflich, T. Vanicek, A. Kautzky, A. Komorowski, U. Moser, R. Gmeinder, O. Keskin, J. Atanelov, E.K. Tempfer-Bentz, C. Tempfer and D. Winkler for their medical, technical or administrative support. We further thank F. Lüsebrink for his remarks regarding the bias correction step at ultra-high fields.

Footnotes

Conflict of Interest Statement

The authors declare no competing financial interests in the context of this study.

References

  • Adriany G, et al. Transmit and receive transmission line arrays for 7 Tesla parallel imaging. Magn Reson Med. 2005;53:434–445. [PubMed]
  • Ashburner J. A fast diffeomorphic image registration algorithm. NeuroImage. 2007;38(1):95–113. [PubMed]
  • Ashburner J, Friston KJ. Voxel-based morphometry - The methods. NeuroImage. 2000;11(6):805–21. [PubMed]
  • Bazin P-L, et al. A computational framework for ultra-high resolution cortical segmentation at 7 Tesla. NeuroImage. 2013:1–9. [PubMed]
  • Belaroussi B, et al. Intensity non-uniformity correction in MRI: existing methods and their validation. Medical image analysis. 2006;10(2):234–46. [PubMed]
  • Bernstein M, Huston J, Ward H. Imaging artifacts at 3.0T. Journal of magnetic resonance imaging. 2006;24(4):735–46. [PubMed]
  • Caramanos Z, et al. Gradient distortions in MRI: Characterizing and correcting for their effects on SIENA-generated measures of brain volume change. NeuroImage. 2010;49(2):1601–1611. [PubMed]
  • Dale AM, Fischl B, Sereno MI. Cortical Surface-Based Analysis. NeuroImage. 1999;9:179–194. [PubMed]
  • Draganski B, et al. Regional specificity of MRI contrast parameter changes in normal ageing revealed by voxel-based quantification (VBQ) NeuroImage. 2011;55(4):1423–34. [PMC free article] [PubMed]
  • Fischl B, Dale a M. Measuring the thickness of the human cerebral cortex from magnetic resonance images. Proceedings of the National Academy of Sciences of the United States of America. 2000;97(20):11050–5. [PubMed]
  • Fujimoto K, et al. Quantitative comparison of cortical surface reconstructions from MP2RAGE and multi-echo MPRAGE data at 3 and 7 T. NeuroImage. 2014;90:60–73. [PMC free article] [PubMed]
  • Good CD, et al. A voxel-based morphometric study of ageing in 465 normal adult human brains. NeuroImage. 2001;14(1 Pt 1):21–36. [PubMed]
  • Hahn A, et al. Comparing neural response to painful electrical stimulation with functional MRI at 3 and 7T. NeuroImage. 2013;82:336–343. [PubMed]
  • Van der Kouwe AJW, et al. Brain morphometry with multiecho MPRAGE. NeuroImage. 2008;40(2):559–69. [PMC free article] [PubMed]
  • Kraus C, et al. Gray matter and intrinsic network changes in the posterior cingulate cortex after selective serotonin reuptake inhibitor intake. NeuroImage. 2014;84:236–244. [PubMed]
  • Lüsebrink F, Wollrab A, Speck O. Cortical thickness determination of the human brain using high resolution 3T and 7T MRI data. NeuroImage. 2013;70:122–31. [PubMed]
  • Maguire EA, et al. Navigation-related structural change in the hippocampi of taxi drivers. Proceedings of the National Academy of Sciences of the United States of America. 2000;97(8):4399–4403. [PubMed]
  • Marchewka A, et al. Influence of magnetic field strength and image registration strategy on voxel-based morphometry in a study of Alzheimer's disease. Human Brain Mapping. 2014;35:1865–1874. [PubMed]
  • Marques JP, et al. MP2RAGE, a self bias-field corrected sequence for improved segmentation and T1-mapping at high field. NeuroImage. 2010;49(2):1271–81. [PubMed]
  • May A, Gaser C. Magnetic resonance-based morphometry: a window into structural plasticity of the brain. Curr Opin Neurol. 2006;19:407–411. [PubMed]
  • McGraw K, Wong S. Forming Inferences About Some Intraclass Correlation Coefficients. Psychological methods. 1996;1(1):30–46.
  • Van de Moortele P-F, et al. T1 weighted brain images at 7 Tesla unbiased for Proton Density, T2* contrast and RF coil receive B1 sensitivity with simultaneous vessel visualization. NeuroImage. 2009;46(2):432–46. [PMC free article] [PubMed]
  • Mugler J, Brookeman J. Three-dimensional magnetization-prepared rapid gradient-echo imaging (3D MP RAGE) Magnetic Resonance in Medicine. 1990;15:152–157. [PubMed]
  • Nugent AC, et al. Cortical abnormalities in bipolar disorder investigated with MRI and voxel-based morphometry. NeuroImage. 2006;30(2):485–97. [PubMed]
  • Shrout P, Fleiss J. Intraclass Correlations: Uses in Assessing Rater Reliability. Psychological bulletin. 1979;86(2):420–428. [PubMed]
  • Sladky R, et al. High-resolution functional MRI of the human amygdala at 7 T. European journal of radiology. 2013;82(5):728–33. [PMC free article] [PubMed]
  • Sled JG, Zijdenbos aP, Evans aC. A nonparametric method for automatic correction of intensity nonuniformity in MRI data. IEEE Transactions on medical imaging. 1998;17(1):87–97. [PubMed]
  • Teipel SJ, et al. Measurement of basal forebrain atrophy in Alzheimer’s disease using MRI. Brain : a journal of neurology. 2005;128(Pt 11):2626–44. [PubMed]
  • Van Tol M-J, et al. Voxel-based gray and white matter morphometry correlates of hallucinations in schizophrenia: The superior temporal gyrus does not stand alone. NeuroImage Clinical. 2014;4:249–57. [PMC free article] [PubMed]
  • Tzourio-Mazoyer N, et al. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. NeuroImage. 2002;15(1):273–89. [PubMed]
  • Weir J. Quantifying test-retest reliability using the intraclass correlation coefficient and the SEM. Journal of Strength and Conditioning Research. 2005;19(1):231–240. [PubMed]
  • Wright IC, et al. A voxel-based method for the statistical analysis of gray and white matter density applied to schizophrenia. NeuroImage. 1995;2:244–252. [PubMed]
  • Yoder KK, et al. Test-retest Variability of [11 C]Raclopride Binding Potential in Nontreatment-Seeking Alcoholics. Synapse. 2011;65(7):553–561. [PMC free article] [PubMed]