After designing an fMRI paradigm and running the experiment and the data collection, various analysis steps must be applied on resulting data before the neuroscientists and physicians can achieve answers to the questions about activities corresponding to the experiment. The goal of computer-based analysis is to determine automatically, those parts of the brain which respond to stimuli that presented to the subjects. The fMRI analysis methods are composed of several basic stages: Pre-processing, signal detection and description and extraction of the brain connectivity.
The goal of preprocessing is to eliminate different kinds of artifacts such as motion correction. Pre-processing consist of spatial or temporal filtering of fMRI data and improving the image resolution. After preprocessing, signal detection is carried out. The purpose of signal detection is to determine which voxels are activated by the stimulation and it is commonly achieved by applying a test statistic. The output of this stage is an activation map which indicates those parts of brain which have been activated in response to the stimulus. The purpose of signal description is modeling the BOLD response shape by several parameters and relating these parameters to the description of the stimulation context. Finally, the connectivity analysis tries to estimate brain networks. Before statistical analysis and signal detection, it is necessary to improve the signal quality by preprocessing the raw data obtained from the MRI scanner, including artifact detection, baseline correction, movement correction, and image restoration. The pre-processing step applies different image and signal processing techniques to reduce the noise and the artifacts of the raw fMRI data. The pre-processing steps are applied individually in particular orders to the fMRI data.
Nearly all fMRI data are collected using two-dimensional MRI acquisition, in which the data are acquired one slice at a time, with the timing of the slice acquisition evenly spread over the repetition time (TR). In some cases, the slices are acquired in ascending or descending order. In another known method as interleaved acquisition, every other slice is acquired sequentially. These differences in the acquisition time of different voxels are problematic for the analysis of fMRI data. The goal of slice timing correction is to adjust the voxel time series so that common reference timing exists for all voxels. The time corresponding to the first slice is often chosen to be the reference.
Statistical Analysis Methods
At the pre-processing stages, the quality of the fMR images is improved. After that, statistical analysis is attempted to determine which voxels are activated by the stimulation. Most fMRI studies are established upon the correlation of hemodynamic response function with stimulation. Activation defines the local intensity changes in the images. These methods can be grouped into two broad categories: the univariate methods (hypothesis testing methods), and the multivariate methods (exploratory methods).
The univariate methods attempt to define which voxels can be characterized as activated given one signal model. This allows the parameterization of the response and then the estimation of the model parameters. The univariate methods are widely used to analyze brain images obtained from fMR imaging. In these methods, signal estimation and the presence or the absence of activation are defined by the statistical test. One of the typical methods is Statistic Parametric Mapping (SPM), which is based upon the hypothesis of linear correlation between neuro-activities and the tasks.
Multivariate methods are also applied to fMRI data analysis, which extract information from dataset, often with any prior knowledge of the experimental conditions. They use some structural properties, such as decorrelation, independence, similarity measures, that can discriminate between features of interest present in the data. Unlike the univariate methods which carry out voxel-wise statistical analysis, multivariate methods provide statistical inference about the whole brain so as to describe brain responses in terms of spatial patterns. A wide range of multivariate statistical methods is being increasingly employed to analyze the fMRI time series. fMRI data are essentially multi-variate in nature, since information about thousands of measured locations (voxels) are being impacted in each scan. The univariate or voxel based analysis approaches have been traditionally used to analyze neuroimaging data.
The vast majority of fMRI data analysis techniques employed by neuroscientists use a GLM. The GLM is one of the most common approaches in fMRI statistical analysis which is the construction of a model that describes the way in which the BOLD response depends on the stimulus. In general linear modeling first a model must be set up and then this model has to be fit to the data.
Functional magnetic resonance imaging or functional MRI (fMRI) measures brain activity by detecting changes associated with blood flow. This technique relies on the fact that cerebral blood flow and neuronal activation are coupled. When an area of the brain is in use, blood flow to that region also increases.
Preprocessing of raw fMRI data involves recognition of outlier data followed by multiple steps to correct for noise, motion, signal drifts, slice timing discrepancies, and spatial distortions.
Functional magnetic resonance imaging (fMRI) uses MR imaging to measure the tiny changes in blood flow that take place in an active part of the brain.