Dataset Information

Effects of spatial smoothing on functional brain networks.

ABSTRACT: Graph-theoretical methods have rapidly become a standard tool in studies of the structure and function of the human brain. Whereas the structural connectome can be fairly straightforwardly mapped onto a complex network, there are more degrees of freedom in constructing networks that represent functional connections between brain areas. For functional magnetic resonance imaging (fMRI) data, such networks are typically built by aggregating the blood-oxygen-level dependent signal time series of voxels into larger entities (such as Regions of Interest in some brain atlas) and determining their connection strengths from some measure of time-series correlations. Although it is evident that the outcome must be affected by how the voxel-level time series are treated at the preprocessing stage, there is a lack of systematic studies of the effects of preprocessing on network structure. Here, we focus on the effects of spatial smoothing, a standard preprocessing method for fMRI. We apply various levels of spatial smoothing to resting-state fMRI data and measure the changes induced in functional networks. We show that the level of spatial smoothing clearly affects the degrees and other centrality measures of functional network nodes; these changes are non-uniform, systematic, and depend on the geometry of the brain. The composition of the largest connected network component is also affected in a way that artificially increases the similarity of the networks of different subjects. Our conclusion is that wherever possible, spatial smoothing should be avoided when preprocessing fMRI data for network analysis.

SUBMITTER: Alakorkko T

PROVIDER: S-EPMC5698731 | biostudies-literature | 2017 Nov

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Effects of spatial smoothing on functional brain networks.

Alakörkkö Tuomas T Saarimäki Heini H Glerean Enrico E Saramäki Jari J Korhonen Onerva O

The European journal of neuroscience 20171020 9

Graph-theoretical methods have rapidly become a standard tool in studies of the structure and function of the human brain. Whereas the structural connectome can be fairly straightforwardly mapped onto a complex network, there are more degrees of freedom in constructing networks that represent functional connections between brain areas. For functional magnetic resonance imaging (fMRI) data, such networks are typically built by aggregating the blood-oxygen-level dependent signal time series of vox ...[more]

PMID: 28922510

Similar Datasets

Project description:Neuroimaging methods with enhanced spatial resolution such as functional magnetic resonance imaging (FMRI) suggest that the subcortical striatum plays a critical role in human reward processing. Analysis of FMRI data requires several preprocessing steps, some of which entail tradeoffs. For instance, while spatial smoothing can enhance statistical power, it may also bias localization towards regions that contain more gray than white matter. In a meta-analysis and reanalysis of an existing dataset, we sought to determine whether spatial smoothing could systematically bias the spatial localization of foci related to reward anticipation in the nucleus accumbens (NAcc). An activation likelihood estimate (ALE) meta-analysis revealed that peak ventral striatal ALE foci for studies that used smaller spatial smoothing kernels (i.e. <6mm FWHM) were more anterior than those identified for studies that used larger kernels (i.e. >7mm FWHM). Additionally, subtraction analysis of findings for studies that used smaller versus larger smoothing kernels revealed a significant cluster of differential activity in the left relatively anterior NAcc (Talairach coordinates: -10, 9, -1). A second meta-analysis revealed that larger smoothing kernels were correlated with more posterior localizations of NAcc activation foci (p<0.015), but revealed no significant associations with other potentially relevant parameters (including voxel volume, magnet strength, and publication date). Finally, repeated analysis of a representative dataset processed at different smoothing kernels (i.e., 0-12mm) also indicated that smoothing systematically yielded more posterior activation foci in the NAcc (p<0.005). Taken together, these findings indicate that spatial smoothing can systematically bias the spatial localization of striatal activity. These findings have implications both for historical interpretation of past findings related to reward processing and for the analysis of future studies.

Project description:Ultra-high Field (≥7T) functional magnetic resonance imaging (UHF-fMRI) provides opportunities to resolve fine-scale features of functional architecture such as cerebral cortical columns and layers, in vivo. While the nominal resolution of modern fMRI acquisitions may appear to be sufficient to resolve these features, several common data preprocessing steps can introduce unwanted spatial blurring, especially those that require interpolation of the data. These resolution losses can impede the detection of the fine-scale features of interest. To examine quantitatively and systematically the sources of spatial resolution losses occurring during preprocessing, we used synthetic fMRI data and real fMRI data from the human visual cortex-the spatially interdigitated human V2 "thin" and "thick" stripes. The pattern of these cortical columns lies along the cortical surface and thus can be best appreciated using surface-based fMRI analysis. We used this as a testbed for evaluating strategies that can reduce spatial blurring of fMRI data. Our results show that resolution losses can be mitigated at multiple points in preprocessing pathway. We show that unwanted blur is introduced at each step of volume transformation and surface projection, and can be ameliorated by replacing multi-step transformations with equivalent single-step transformations. Surprisingly, the simple approaches of volume upsampling and of cortical mesh refinement also helped to reduce resolution losses caused by interpolation. Volume upsampling also serves to improve motion estimation accuracy, which helps to reduce blur. Moreover, we demonstrate that the level of spatial blurring is nonuniform over the brain-knowledge which is critical for interpreting data in high-resolution fMRI studies. Importantly, our study provides recommendations for reducing unwanted blurring during preprocessing as well as methods that enable quantitative comparisons between preprocessing strategies. These findings highlight several underappreciated sources of a spatial blur. Individually, the factors that contribute to spatial blur may appear to be minor, but in combination, the cumulative effects can hinder the interpretation of fine-scale fMRI and the detectability of these fine-scale features of functional architecture.

Project description:BACKGROUND:Many methods of spatial smoothing have been developed, for both point data as well as areal data. In Bayesian spatial models, this is achieved by purposefully designed prior(s) or smoothing functions which smooth estimates towards a local or global mean. Smoothing is important for several reasons, not least of all because it increases predictive robustness and reduces uncertainty of the estimates. Despite the benefits of smoothing, this attribute is all but ignored when it comes to model selection. Traditional goodness-of-fit measures focus on model fit and model parsimony, but neglect "goodness-of-smoothing", and are therefore not necessarily good indicators of model performance. Comparing spatial models while taking into account the degree of spatial smoothing is not straightforward because smoothing and model fit can be viewed as opposing goals. Over- and under-smoothing of spatial data are genuine concerns, but have received very little attention in the literature. METHODS:This paper demonstrates the problem with spatial model selection based solely on goodness-of-fit by proposing several methods for quantifying the degree of smoothing. Several commonly used spatial models are fit to real data, and subsequently compared using the goodness-of-fit and goodness-of-smoothing statistics. RESULTS:The proposed goodness-of-smoothing statistics show substantial agreement in the task of model selection, and tend to avoid models that over- or under-smooth. Conversely, the traditional goodness-of-fit criteria often don't agree, and can lead to poor model choice. In particular, the well-known deviance information criterion tended to select under-smoothed models. CONCLUSIONS:Some of the goodness-of-smoothing methods may be improved with modifications and better guidelines for their interpretation. However, these proposed goodness-of-smoothing methods offer researchers a solution to spatial model selection which is easy to implement. Moreover, they highlight the danger in relying on goodness-of-fit measures when comparing spatial models.

Dataset Information

Effects of spatial smoothing on functional brain networks.

Publications

Effects of spatial smoothing on functional brain networks.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets