Pre-processing of the raw fMRI time-series, prior to statistical analysis, generally has two main goals: firstly, to reduce unwanted or uninteresting variability from data and; secondly, to prepare data for statistical analysis and inference given that many statistical tests applied in fMRI studies make assumptions that are met through such preprocessing steps. In practice, this involves modifying the raw data in a series of steps often including image realignment - to correct and to diagnose head motion artifact in a time series, image normalization - to transform data from different subjects into a common neuro-anatomical space, spatial smoothing - to filter (or blur) the data to reduce spatial noise and to improve its normality for statistical parametric tests, and temporal smoothing -to remove low or high frequency noise sources, such as mentioned above.
There are a growing number of ways to perform statistical analysis in BOLD fMRI experiments. This has been assisted greatly by the development of publicly available neuroimaging analysis software packages, such as Statistical Parametric Mapping (http://www.fil.ion.ucl.ac.uk/spm/), FMRIB Software Library (http://www.fmrib.ox.ac.uk/fsl/ FSL), Analysis of Functional Neuro Images (http://afni. nimh.nih.gov/). The majority of fMRI studies to date have adopted a conventional voxel-based mapping approach based on extensions of the general linear model for time-series analysis. The basic premise behind such approaches is that the observed fMRI data can be accounted for by a combination of several experimental (or model) parameters and uncorrelated (or independently distributed) noise. Given the high number of statistical tests performed (voxel by voxel) some correction factor for multiple
Magnetic Resonance Imaging (Functional). Fig. 5. Global functional connectivity of a large-scale and distributed brain network characterized from two distinct task states using independent component analysis (ICA). Top panel a: Correlated fluctuations of the BOLD signal among regions of the so-called ''default-mode network'' in a group of healthy subjects scanned at rest (NB: time-course plot is of a single subject). Bottom panel b: Correlated fluctuations of the BOLD signal among ''default mode network'' regions in the same group of subjects performing a moral dilemma task (NB: time course plot is of the group mean). The task consisted of four alternating 30 s control (C) and moral dilemma (D) condition blocks (CDCDCDCD). (Modified from Harrison et al. (2008) Proc Natl Acad Sci USA 105:9781-9789. © 2008 by the National Academy of Sciences of the USA.)
comparisons will generally be applied, leading to the generation of statistically thresholded "activation" maps related to the experiment at hand. This may be performed for the whole brain or specific regions of interest.
Other techniques, based on multivariate analysis techniques can also be used to investigate which brain areas are "activated" by a task or a stimulus in fMRI studies. These techniques, as opposed to the general linear model approach, are data driven and therefore do not require the specification of experimental models a priori. Another important distinction between this class of statistical tests and the former is that they are sensitive for testing not only "where" activation occurs in a given experimental context but also how different regions or networks of regions may interact or show interdependence in their activities over time (Fig. 5). Such relationships have been characterized as representing distinct forms of brain-functional connectivity, which has become a topic of specific interest with BOLD fMRI in recent years.
Major Strengths of the Method
• Is a safe, noninvasive, highly repeatable and widely available technique for measuring changes in brain activity in vivo.
• Has superior spatial resolution compared to other human neuroimaging techniques.
• Affords high flexibility in experimental design and data modeling.
Major Limitations of the Method
• Measures neuronal activity indirectly via changes in blood oxygenation levels.
• Has a temporal resolution in the order of seconds due to the nature of the hemodynamic response.
• Is susceptible to influences of non-neural changes in the body.
Was this article helpful?