Issues with threshold masking in voxel-based morphometry of atrophied brains

doi:10.1016/j.neuroimage.2008.08.045

NeuroImage

Volume 44, Issue 1, 1 January 2009, Pages 99-111

https://doi.org/10.1016/j.neuroimage.2008.08.045 Get rights and content

Abstract

There is great interest in using automatic computational neuroanatomy tools to study ageing and neurodegenerative disease. Voxel-based morphometry (VBM) is one of the most widely used of such techniques. VBM performs voxel-wise statistical analysis of smoothed spatially normalised segmented Magnetic Resonance Images. There are several reasons why the analysis should include only voxels within a certain mask. We show that one of the most commonly used strategies for defining this mask runs a major risk of excluding from the analysis precisely those voxels where the subjects' brains were most vulnerable to atrophy. We investigate the issues related to mask construction, and recommend the use of alternative strategies which greatly decrease this danger of false negatives.

Introduction

In essence, voxel-based morphometry (VBM) (Ashburner and Friston, 2000) involves voxel-wise statistical analysis of data derived from structural Magnetic Resonance (MR) brain images of multiple subjects. The images analysed are obtained through tissue segmentation, spatial normalisation, and spatial smoothing. Statistical analysis employs a mass-univariate parametric or non-parametric general linear model at each voxel. More precisely, the calculations are performed at each voxel within some mask. There are several reasons why masking is necessary, mostly related to the multiple comparison problem. Family-wise error (FWE) correction using random field theory (RFT) is generally more powerful for smaller analysis regions (this is commented on further in the discussion), and perhaps more importantly, masking is necessary for successful estimation of the smoothness of the residuals (John Ashburner, personal communication), which is a key part of the RFT correction procedure (Kiebel et al., 1999). If non-parametric permutation methods (Nichols and Holmes, 2002) are employed for FWE correction, the effect of the analysis region on computational complexity may also be important (Belmonte and Yurgelun-Todd, 2001). Correction of the false-discovery rate (Genovese et al., 2002) also depends on masking, since non-brain voxels could otherwise skew the distribution of p-values on which it is based. Furthermore, masking can also partially alleviate a problem of implausible false positives occurring outside the brain due to the very low variance in voxels with consistently low smoothed tissue density — the extreme limit of the phenomenon described by Reimold et al. (2006). Finally, while not specifically considered here, multivariate machine learning, classification or decoding approaches (Lao et al., 2004, Vemuri et al., 2008, Friston et al., 2008) can also benefit from masking as an initial feature selection or dimensionality reduction step.

Having emphasised above that smaller masks generally lead to higher sensitivity and clarified interpretation, it is important to recognise the obvious risk that overly restrictive masks will lead to false negatives, as potentially interesting voxels are excluded from the statistical analysis. In this paper, we argue that there is a particular danger of false negatives arising in VBM studies of pathological brains when computing the analysis mask using a commonly employed approach with settings that appear reasonable a priori. This approach is used by the popular Statistical Parametric Mapping (SPM) software (http://www.fil.ion.ucl.ac.uk/spm/). We recommend the use of different mask-generation strategies, which we show to reduce this danger. In a three-part experiment using SPM, we (a) use simulated data to investigate properties of preprocessing relevant to masking; (b) explore the behaviour of standard and more novel methods of masking, considering variable patient group composition; and (c) test the practical importance of our recommendation on a particular example of a real VBM study. We propose two main masking options: one is a fully objective parameter-free algorithm, which we hope will find wide-spread applicability; the other allows expert knowledge to be exercised in cases where the automatic strategy is found to be unsuitable.

Section snippets

Masking strategies

The SPM software commonly used for VBM studies offers several alternatives to specify the mask for statistical analysis. If available, a precomputed mask can be explicitly requested, or the analysis mask can be automatically derived by excluding voxels in which any of the images have intensity values below a certain threshold. This threshold can be specified as an absolute value, constant for all the images, or as a relative fraction of each image's ‘global’ value. The global value can itself

Simulated images

Fig. 1 illustrates typical results for VBM preprocessing using SPM5's unified segmentation model (Ashburner and Friston, 2005). The estimated segmentation is in close agreement with the simulation's underlying model,³ but the inter-subject correspondence following spatial normalisation is only approximate. This imperfect overlap

Conclusions

With many diseases there is a spectrum of severity of focal atrophy; the most vulnerable regions might also be the most likely to have outlying subjects with particularly severe absence of tissue. The standard masking procedure in the SPM software risks missing findings in the most severely atrophied brain regions. It is important to note that the missed atrophy when using overly restrictive masks might not be readily apparent from consideration of the ‘glass-brain’ maximum intensity projection

Acknowledgments

We are grateful to John Ashburner, for helpful comments on the need for masking in residual smoothness estimation, and to Susie Henley and Jonathan Rohrer for helpful discussion and for experimenting with our new mask-creation software. We wish to thank the anonymous reviewers, who made several important suggestions. G.R.R. is funded by an EPSRC CASE Studentship, sponsored by GlaxoSmithKline. J.D.W. is supported by a Wellcome Trust Intermediate Clinical Fellowship. N.C.F. acknowledges support

References (25)

Acosta-CabroneroJ. et al.
The impact of skull-stripping and radio-frequency bias correction on grey-matter segmentation for voxel-based morphometry
Neuroimage
(2008)
AshburnerJ.
A fast diffeomorphic image registration algorithm
Neuroimage
(2007)
AshburnerJ. et al.
Voxel-based morphometry—the methods
Neuroimage
(2000)
AshburnerJ. et al.
Unified segmentation
Neuroimage
(2005)
FristonK. et al.
Bayesian decoding of brain images
Neuroimage
(2008)
GenoveseC.R. et al.
Thresholding of statistical maps in functional neuroimaging using the false discovery rate
Neuroimage
(2002)
KarasG.B. et al.
A comprehensive study of gray matter loss in patients with Alzheimer's disease using optimized voxel-based morphometry
Neuroimage
(2003)
KiebelS.J. et al.
Robust smoothness estimation in statistical parametric maps using standardized residuals from the general linear model
Neuroimage
(1999)
LaoZ. et al.
Morphological classification of brains via high-dimensional shape transformations and machine learning methods
Neuroimage
(2004)
RidgwayG.R. et al.
Ten simple rules for reporting voxel-based morphometry studies
Neuroimage
(2008)

ShenD. et al.

Very high-resolution morphometry using mass-preserving deformations and HAMMER elastic registration

Neuroimage

(2003)

SmithS.M. et al.

Advances in functional and structural MR image analysis and implementation as FSL

Neuroimage

(2004)

Cited by (256)

Gray Matter Adaptations to Chronic Pain in People with Whiplash-Associated Disorders are Partially Reversed After Treatment: A Voxel-based Morphometry Study
2024, Journal of Pain
Gray matter (GM) changes are often observed in people with chronic spinal pain, including those with chronic whiplash-associated disorders (CWAD). These GM adaptations may be reversed with treatment, at least partially. Pain neuroscience education combined with exercise (PNE+Exercise) is an effective treatment, but its neural underlying mechanisms still remain unexplored in CWAD. Here, we performed both cross-sectional and longitudinal voxel-based morphometry to 1) identify potential GM alterations in people with CWAD (n = 63) compared to age- and sex-matched pain-free controls (n = 32), and 2) determine whether these GM alterations might be reversed following PNE+Exercise (compared to conventional physiotherapy). The cross-sectional whole-brain analysis revealed that individuals with CWAD had less GM volume in the right and left dorsolateral prefrontal cortex and left inferior temporal gyrus which was, in turn, associated with higher pain vigilance. Fifty individuals with CWAD and 29 pain-free controls were retained in the longitudinal analysis. GM in the right dorsolateral prefrontal cortex increased after treatment in people with CWAD. Moreover, the longitudinal whole-brain analysis revealed that individuals with CWAD had decreases in GM volumes of the left and right central operculum and supramarginal after treatment. These changes were not specific to treatment modality and some were not observed in pain-free controls over time. Herewith, we provide the first evidence on how GM adaptations to CWAD respond to treatment.
This article presents which gray matter adaptations are present in people with chronic pain after whiplash injuries. Then, we examine the treatment effect on these alterations as well as whether other neuroplastic effects on GM following treatment occur.
The neural substrates of how model-based learning affects risk taking: Functional coupling between right cerebellum and left caudate
2023, Brain and Cognition
Higher executive control capacity allows people to appropriately evaluate risk and avoid both excessive risk aversion and excessive risk-taking. The neural mechanisms underlying this relationship between executive function and risk taking are still unknown. We used voxel-based morphometry (VBM) analysis combined with resting-state functional connectivity (rs-FC) to evaluate how one component of executive function, model-based learning, relates to risk taking. We measured individuals’ use of the model-based learning system with the two-step task, and risk taking with the Balloon Analogue Risk Task. Behavioral results indicated that risk taking was positively correlated with the model-based weighting parameter ω. The VBM results showed a positive association between model-based learning and gray matter volume in the right cerebellum (RCere) and left inferior parietal lobule (LIPL). Functional connectivity results suggested that the coupling between RCere and the left caudate (LCAU) was correlated with both model-based learning and risk taking. Mediation analysis indicated that RCere-LCAU functional connectivity completely mediated the effect of model-based learning on risk taking. These results indicate that learners who favor model-based strategies also engage in more appropriate risky behaviors through interactions between reward-based learning, error-based learning and executive control subserved by a caudate, cerebellar and parietal network.
Neuroimaging research in Williams syndrome: Beginning to bridge the gap with clinical care
2023, Neuroscience and Biobehavioral Reviews
Williams syndrome (WS) is a genetic disorder affecting multiple organ systems. Cardinal features include cardiovascular disease, distinct facies, and a unique cognitive profile characterized by intellectual disability, hypersociability, and visuospatial weaknesses. Here, we synthesize neuroimaging research in WS with a focus on how the current literature and future work may be leveraged to improve health and quality of life in WS. More than 80 neuroimaging studies in WS have been conducted, the vast majority of which have focused on identifying morphometric brain differences. Aside from decreased volume of the parieto-occipital region and increased cerebellar volume, morphometric findings have been variable across studies. fMRI studies investigating the visuospatial deficit have identified dorsal stream dysfunction and abnormal activation of the hippocampal formation. Minimal work has been done using PET or MRS. Future approaches that conduct neuroimaging in tandem with clinical phenotyping, utilize novel imaging techniques to visualize brain vasculature or provide biochemical and molecular information, and include more homogenous age groups across the lifespan, have significant potential to advance clinical care.
Behavioral and neural correlates of pareidolic illusions in dementia with Lewy bodies
2023, Parkinsonism and Related Disorders
Pareidolia, a form of visual illusions phenomenologically similar to complex visual hallucinations, is a phenomenon that is associated with visual hallucinations in dementia with Lewy bodies (DLB). This study aimed to identify commonalities and differences in behavioral and neural correlates between pareidolic illusions and visual hallucinations in DLB.
Forty-three patients with DLB underwent the scene pareidolia test, which evokes and measures pareidolic illusions, and standardized neuropsychological and behavioral assessments. Regional cerebral blood flow (rCBF) was measured by single-photon emission computed tomography. Factor analysis was performed to assess the relationships among pareidolic illusions, cognitive functions, and behavioral symptoms. Partial least squares correlation analysis was used to investigate the relationship between these symptoms and rCBF.
Factor analysis yielded three behavior factors: the first factor (hallucinations/fluctuations) consisted of pareidolic illusions, visual hallucinations, and fluctuating cognition; the second factor (general cognitive function) consisted of general cognitive function and working memory; and the third factor (visual processing) consisted of visual processing and pareidolic illusions. Partial least squares correlation analysis identified two brain-behavior correlation patterns: (1) rCBF reduction in the frontal and perisylvian/periventricular regions was associated with lower general cognitive function and lower visual processing; and (2) rCBF reduction in the bilateral occipitotemporal cortex was associated with more severe hallucinations/fluctuations and lower visual processing.
At the behavioral level, pareidolic illusions are associated with visual hallucinations, fluctuating cognition, and visual processing in DLB. At the neural level, pareidolic illusions may arise from the synergistic effects of global neuropathological changes and occipitotemporal cortical dysfunctions.
The para-hippocampal–medial frontal gyrus functional connectivity mediates the relationship between dispositional optimism and procrastination
2023, Behavioural Brain Research
Procrastination is a prevalent phenomenon throughout the world, which can lead to worse consequences across life domains, such as academic performance, mental health, and even public policy. Despite the evidence for the association between dispositional optimism and procrastination, the neural mechanisms underlying this link remain unexplored. To address this issue, we employed voxel-based morphometry (VBM) and resting-state functional connectivity (RSFC) methods to explore the underlying links between dispositional optimism and procrastination in a large sample (N = 408). The self-report results showed that dispositional optimism was negatively associated with procrastination (r = −.30, p < .001). The VBM analysis indicated that dispositional optimism was positively correlated with gray matter volumes (GMV) in the right para-hippocampal (rPHC), and negatively correlated with GMV in the left cerebellum. Moreover, the functional connectivity analysis with the rPHC as a seed region revealed that rPHC-rMFC (right medial frontal gyrus) functional connectivity was negatively associated with dispositional optimism. Furthermore, the mediation analysis showed that the rPHC-rMFC connectivity partially mediated the relationship between dispositional optimism and procrastination. These results suggested that the rPHC-rMFC connectivity engaged in less task aversiveness by episodic prospection may underlie the association between dispositional optimism and procrastination, which provides a new perspective to understand the relationship between dispositional optimism and procrastination.
Word-producing brain: Contribution of the left anterior middle temporal gyrus to word production patterns in spoken language
2023, Brain and Language
Vocabulary is based on semantic knowledge. The anterior temporal lobe (ATL) has been considered an essential region for processing semantic knowledge; nonetheless, the association between word production patterns and the structural and functional characteristics of the ATL remains unclear. To examine this, we analyzed over one million words from group conversations among community-dwelling older adults and their multimodal magnetic resonance imaging data. A quantitative index for the word production patterns, namely the exponent β of Heaps’ law, positively correlated with the left anterior middle temporal gyrus volume. Moreover, β negatively correlated with its resting-state functional connectivity with the precuneus. There was no significant correlation with the diffusion tensor imaging metrics in any fiber. These findings suggest that the vocabulary richness in spoken language depends on the brain status characterized by the semantic knowledge-related brain structure and its activation dissimilarity with the precuneus, a core region of the default mode network.

View all citing articles on Scopus

View full text