Impact of image acquisition on voxel-based-morphometry investigations of age-related structural brain changes
Introduction
Over the past decade, an increasing number of studies have used voxel-based morphometry (VBM) for assessing structural brain changes (Draganski et al., 2011, Ferreira et al., 2011, Good et al., 2001, Hutton et al., 2009). VBM is a whole-brain technique to investigate so-called local tissue-density changes, typically employing three-dimensional (3D) T1-weighted magnetic resonance imaging (MRI) data sets. Recent developments led to an increased accuracy of segmentation (Ashburner and Friston, 2005) and registration (Ashburner, 2007, Ashburner and Friston, 2005, Klein et al., 2010) and, thereby, to improved statistical assessment. Besides sophisticated image processing, the quality of the input images is also relevant in VBM, for example, a sufficient signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR), and minimal image artifacts are of critical importance. Naturally, input quality does not just impact VBM but image (pre-)processing methods in general. In this context, it is noteworthy that volume-based pre-processing is also employed in surface-based approaches, for instance FreeSurfer uses ‘non-parametric non-uniform intensity normalization’, N3, (Sled et al., 1998) and ‘constrained Laplacian anatomic segmentation using proximity’, CLASP, (Kim et al., 2005) for bias correction and segmentation, respectively. However, only some studies have systematically investigated the impact of image acquisition parameters, including the employed pulse sequence, on VBM and whole-brain measures (Acosta-Cabronero et al., 2008, Bach Cuadra et al., 2013, Helms et al., 2009, Klauschen et al., 2009, Krueger et al., 2012, Pereira et al., 2008, Shuter et al., 2008, Stonnington et al., 2008, Tardif et al., 2009, Tardif et al., 2010).
To study the influence of resolution, Pereira et al. (2008) used different interpolated voxel sizes comparing patients with Alzheimer's disease and patients with semantic dementia to healthy controls. They showed that interpolation effects are highly dependent on the acquired image volume itself and have to be treated carefully. In two studies, Tardif et al., 2009, Tardif et al., 2010 investigated the impact of different acquisition protocols on VBM results. In particular, three MRI sequences—the ‘fast low-angle shot’ technique, FLASH, (Frahm et al., 1986); the ‘magnetization-prepared rapid gradient echo’ imaging, MP-RAGE, (Mugler and Brookeman, 1990, Mugler et al., 1992); and the ‘modified driven equilibrium Fourier transform’ approach, MDEFT, (Lee et al., 1995)—were compared at 3 T and 1.5 T. Each protocol yielded a distinct regional sensitivity pattern to morphometric gray-matter density (GMD) changes. Results from power analyses showed that MP-RAGE required more subjects than FLASH to detect GMD changes but offered higher CNR and improved tissue classification. The MDEFT protocol, which is currently used only in research settings, yielded the highest CNR and the smallest GMD variability. A limitation of both studies was the small number of subjects (≤ 9). Similarly, Helms et al. (2009) suggested by means of VBM that the segmentation of deep gray-matter (GM) structures is improved when using acquisition protocols that allow the computation of magnetization-transfer maps compared to optimized T1-weighted MRI protocols. In those former studies paired t-tests showed the effect of mixing two different imaging parameters into one VBM analysis. However, this analysis does not provide direct information on altered detectability of structural brain changes due to the variation of the parameters. In principle, such changes can be assessed by statistical interaction analysis of two experimental factors, where one of the factors is a well-established structural process. Note, however, that the VBM analysis is only an indirect assessment of brain structure as it interprets GMD measures, which are estimated using a-priori template and model information.
To expand the scope of previous findings, the objective of the current work was to evaluate the impact of acquisition parameters in detecting age-related structural brain changes causing atrophy and reduction in the GMD. Previous research showed GMD changes across the whole cortex (Good et al., 2001), which provides excellent conditions to study distinct sensitivity profiles of different imaging parameters employing interaction analyses. Besides interaction tests, paired t-tests were utilized to detect artificially introduced differences in GMD estimates due to variation of acquisition parameters.
Following similar motivation as discussed by Tardif et al., 2009, Tardif et al., 2010, the MP-RAGE technique was employed as one of the most popular anatomical sequences in morphometric studies. Due to ubiquitous availability, good GM/white matter (WM) contrast, and relatively short scan times, the MP-RAGE sequence has been frequently used in investigations of neurodegenerative diseases (Camicioli et al., 2009, Feldmann et al., 2008) including multicenter trials of the Alzheimer's Disease Neuroimaging Initiative (ADNI) (Jack et al., 2008, Mueller et al., 2005). Briefly, MP-RAGE consists of an initial preparation of the magnetization by an adiabatic inversion pulse, followed by a relaxation delay (inversion time, TI) and a segmented acquisition period employing a snapshot FLASH sequence (Haase, 1990). Typically, the FLASH readout proceeds through the entire partition-encoding loop during each segment (i.e., during each repetition time, TR, of the inversion pulse), and the phase-encoding gradient is incremented from segment to segment (i.e., with every TR) (Deichmann et al., 2000, Mugler and Brookeman, 1990, Mugler et al., 1992).
Additional imaging with the recently published ‘magnetization-prepared 2 rapid gradient echoes’ sequence, MP2RAGE, (Marques et al., 2010) was included for a comparison of acquisition techniques. It is a variation of the MP-RAGE sequence, in which two FLASH readouts are quasi-simultaneously acquired at different TI. The images can be combined by computing the product of the (complex) signal intensities divided by the sum of the squared intensities to obtain so-called ‘uniform images’ with T1 contrast, which are largely free of both the radiofrequency (RF) reception bias field and RF transmit-field inhomogeneity. Although the SNR might be reduced on the ‘uniform images’ due to noise propagation, contrast is improved as unwanted proton-density contrast and residual T2* contrast are removed by the image combination as well. We thus hypothesize that MP2RAGE offers potential advantages of (i) an intrinsic bias correction based on a well-defined physical concept of signal generation and (ii) improved contrast between GM, WM, and cerebro-spinal fluid (CSF) that might lead to better tissue classification and segmentation.
For further comparisons, acquisitions were performed with both a 12-channel coil and a 32-channel coil to study the influence of imaging hardware, and with two different voxel dimensions to study the effect of image resolution and/or interpolation.
Section snippets
Subjects
Thirty-six healthy Caucasian adults grouped into 12 young (6 females, mean age plus/minus one standard deviation: 22.3 ± 1.1 years), 12 middle-aged (6 females, 46.6 ± 1.4 years), and 12 elderly (6 females, 71.8 ± 1.9 years) subjects participated in the study. All participants gave written consent after being informed about the possible risks and discomforts of the experimental procedure. Subjects also completed a health history questionnaire to assess their suitability for undergoing MRI scanning.
Results
Unless explicitly otherwise mentioned, the following results were obtained with SPM12b. All statistical tests refer to data, which were processed using an interpolated isotropic voxel size of 1.5 mm (standard settings). Consistent results with minimal variation were obtained with an interpolated voxel size of 0.8 mm (data not shown).
Discussion
In the current study, aging effects served as a paradigm to investigate the impact of the MRI acquisition protocol on GMD estimates obtained with VBM (SPM12b/DARTEL). In particular, influences from different RF coils, imaging sequences, and image resolutions were statistically assessed by two-sample t-tests of the main effect and interaction of age and acquisition parameters and, additionally, by paired t-tests of different acquisition parameters.
Conclusions
Systematic evaluation of GMD changes associated with age yielded a significant impact from acquisition parameters on VBM analyses. These parameters include the pulse sequence, RF coil, and image resolution. Our findings from statistical tests and quantitative image analyses are consistent with the following hypotheses: (i) Performance of MP-RAGE and MP2RAGE with SPM12b processing is slightly different, with potential advantages in tissue classification and segmentation for MP2RAGE. (ii) The
Acknowledgments
We are grateful to Dr. Tobias Kober (Siemens Schweiz AG, Renens) for sharing the MP2RAGE sequence and insightful comments and to Dr. John Ashburner for information on the different segmentation approaches in SPM. Appreciation is extended to Sylvie Neubert for assistance during the experiments. This work was funded in part by the Helmholtz Alliance ICEMED—Imaging and Curing Environmental Metabolic Diseases, through the Initiative and Network Fund of the Helmholtz Association and by the
References (44)
- et al.
The impact of skull-stripping and radio-frequency bias correction on grey-matter segmentation for voxel-based morphometry
NeuroImage
(2008) A fast diffeomorphic image registration algorithm
NeuroImage
(2007)- et al.
Unified segmentation
NeuroImage
(2005) - et al.
Voxel-based morphometry reveals extra-nigral atrophy patterns associated with dopamine refractory cognitive and motor impairment in parkinsonism
Parkinsonism Relat. Disord.
(2009) - et al.
Optimization of 3-D MP-RAGE sequences for structural brain imaging
NeuroImage
(2000) - et al.
Regional specificity of MRI contrast parameter changes in normal ageing revealed by voxel-based quantification (VBQ)
NeuroImage
(2011) - et al.
Neurostructural predictors of Alzheimer's disease: a meta-analysis of VBM studies
Neurobiol. Aging
(2011) - et al.
A voxel-based morphometric study of ageing in 465 normal adult human brains
NeuroImage
(2001) - et al.
Nonstationary cluster-size inference with random field and permutation methods
NeuroImage
(2004) - et al.
Improved segmentation of deep brain grey matter structures using magnetization transfer (MT) parameter maps
NeuroImage
(2009)
A comparison between voxel-based cortical thickness and voxel-based morphometry in normal ageing
NeuroImage
Anisotropic partial volume CSF modeling for EEG source localization
NeuroImage
Automated 3-D extraction and evaluation of the inner and outer cortical surfaces using a Laplacian map and partial volume effect classification
NeuroImage
Evaluation of volume-based and surface-based brain image registration methods
NeuroImage
Cortical thickness determination of the human brain using high resolution 3 T and 7 T MRI data
NeuroImage
MP2RAGE, a self bias-field corrected sequence for improved segmentation and T1-mapping at high field
NeuroImage
The Alzheimer's Disease Neuroimaging Initiative
Neuroimaging Clin. N. Am.
Impact of inconsistent resolution on VBM studies
NeuroImage
Ten simple rules for reporting voxel-based morphometry studies
NeuroImage
Reproducibility of brain tissue volumes in longitudinal studies: effects of changes in signal-to-noise ratio and scanner software
NeuroImage
Interpreting scan data acquired from multiple scanners: a study with Alzheimer's disease
NeuroImage
Sensitivity of voxel-based morphometry analysis to choice of imaging protocol at 3 T
NeuroImage
Cited by (32)
Convergent imaging-transcriptomic evidence for disturbed iron homeostasis in Gilles de la Tourette syndrome
2023, Neurobiology of DiseaseRapid volumetric brain changes after acute psychosocial stress
2023, NeuroImageThe age-dependent relationship between resting heart rate variability and functional brain connectivity
2019, NeuroImageCitation Excerpt :The samples differed significantly, for example in resting HRV (young subjects), sex distribution, sample size, BP, and MRI/ECG acquisition parameters (Supplementary Table 1, Tables 4 and 5). Although we accounted for within- and between-sample variance in the second-level GLM, these differences may still have influenced our results (e.g., for structural MRI; Streitbürger et al., 2014). Further, we calculated the RMSSD using 10 s of ECG data, which has been shown to be a valid measurement (Munoz et al., 2015; Nussinovitch et al., 2011a, 2011b).
Power estimation for non-standardized multisite studies
2016, NeuroImageCitation Excerpt :Variation in scaling factors could explain why a study may estimate different effect sizes based on the pulse sequence used. For example, Streitbürger et al. (2014) found significant effects of RF head coils, pulse sequences, and resolution on VBM results. The estimation of scaling factors in our model depends on good scan–rescan reliability.
- 1
These authors contributed equally.