The advent of neuroimaging methods, in particular functional Magnetic Resonance Imaging (fMRI) opened the possibility to non-invasively map cognitive functions in the healthy human brain. The principle of fMRI is to measure task-evoked physiological changes in blood flow and local metabolism that correlate with the related neural activity. The underlying physiological and statistical procedures will be shortly summarized in the following sections.
What kind of physiological signal is measured with fMRI? Neural activity leads to an increased metabolism – the consumption of glucose and oxygen, which is needed to restore concentration gradients in the neuron that are changed following neural activity. Metabolites are supplied by the vascular system with one essential aspect being the arterial blood supply of oxygenated hemoglobin. The increase in oxygenation usually exceeds the actual demand in the respective brain region. Early research on the MRI signal demonstrated that deoxygenated hemoglobin is paramagnetic while oxygenated hemoglobin is diamagnetic. The different ratio of oxygenated and deoxygenated blood after neural activity leads to temporal local field inhomogenities that are reflected in the T2* decay time which crucially depends on the field homogenity. It has been shown that these differences in magnetic properties after neural activity can be measured with MRI (Ogawa, Lee, Kay, & Tank, 1990). This effect was named the BOLD (blood-oxygenation-level-dependent) effect which was then applied to functional measurements with fMRI (Bandettini, Wong, Hiks, Tikofsky, & Hyde, 1992; Kwong et al., 1992; Ogawa et al., 1992). Thus, the BOLD contrast describes the difference in MRI signal on T2*-weighed images as a function of the amount of deoxygenated hemoglobin.
The BOLD response to neural activity consists of a short onset delay, a rise to a peak after a few seconds, a return to baseline, and a prolonged undershoot. Usually, this takes about 5-12 seconds, excluding the undershoot (Aguirre, Zarahn, & D’Esposito, 1998; Friston, Frith, Turner, & Frackowiak, 1995a; Miezin, Maccotta, Ollinger, Petersen, & Buckner, 2000). This is also called the hemodynamic response function (HRF). Amplitude and latency of the HRF depend on the strength of the evoking stimulus on the one hand but also on the region where it is measured. In addition, high inter-individual variability in the shape of the function has been measured (Aguirre et al., 1998; Handwerker, Ollinger, & D’Esposito, 2004).
Depending on the specific research question, different experimental designs can be used to increase the signal-to-noise ratio in the task-evoked BOLD-signal. In blocked designs, the different experimental conditions are presented block-wise with high task frequency within one block. That way a very strong signal that develops over the course of the block can be measured. In event-related designs, single trials are presented within the same blocks with longer inter-stimulus intervals (ISI). In event-related designs, the HRF can be determined for every task trial individually, giving the possibility to compare trials within one task block and to eliminate error trials. However, event-related designs have usually smaller detection power, e.g. the ability to detect an activation, than block designs. Study 1 and 2 of this dissertation used pure blocked experimental designs; Study 3 was measured in a mixed block and event-related design.
As outlined above, the functional task-related measurements consist of T2* images that are acquired on a slice-by-slice basis in high temporal frequency (1.5- 2.2 seconds per brain volume in the present studies) in small 3-dimensional units (voxels) in the brain.
Before the actual statistical analysis of the task-related BOLD-signal changes in certain brain regions, a couple of preprocessing steps are performed on these time series data depending on the applied experimental design. Using the software SPM2 (http://www.fil.ion.ucl.ac.uk/spm/), in the present studies images were slice-time corrected to account for differences in acquisition time between slices (Study 3 only) and motion corrected (all studies) and unwarped (Study 3 only) to account for movement of the participants in the scanner. The images were then spatially normalized into the standard MNI atlas space using the high-resolution T1-weighted anatomical images of every participant and applying the normalization parameters to the functional images after having coregistered those to the anatomical images. Then, the functional data were smoothed with an 8-mm FWHM Gaussian kernel to suppress residual differences in functional and gyral anatomy during inter-subject averaging. In addition, a high-pass filter was applied during analysis to eliminate residual low-frequency noise. The data were statistically analysed using the general linear model (GLM) for serially auto-correlated data (Friston et al., 1995b). In the GLM the given experimental design is correlated with the brain activity in every voxel in the brain. For that purpose the predictors (regressors) are convolved with a given HRF. The obtained parameter estimates (beta values) per voxel describe the degree to which this voxel correlates with the given experimental design. That way, voxels are identified that show increased BOLD-signal changes under certain task contexts. These signal increases are then tested for statistical significance in a second-level analysis across all participants, either in comparison to a baseline condition or to another task condition. Usually two task conditions are compared to eliminate specific cognitive processes (Braver et al., 1997). The resulting statistical maps can then be overlaid onto normalised anatomical images to locate the significant activities in the brain. That way, regions are identified on a whole-brain basis that are related to the manipulated cognitive processes.
In addition to the statistical analyses on a whole-brain level, so-called regions-of-interest (ROI) analyses were applied in all three studies. The goal of an ROI-analysis is to test specific hypotheses about the pattern of task-related activity in a particular brain region. A-priori ROIs can be identified either as anatomical or/and as functional ROIs. For anatomical ROIs, usually, structural T1 images are used to draw the ROIs on specific sulci, gyri or other subcortical structures. The mean signal changes or the related parameter estimates from voxels included in these regions are then tested with respect to the formulated hypothesis.
Another powerful approach that was applied in the present studies to test the effects of the parametric manipulations is to use functional ROIs. The advantage of functional ROIs is that they can take functional subdivisions within anatomical regions into account. These ROIs are based on functional criteria such as the results of a whole-brain analysis. Some researchers use the group activity peaks from other studies as centers of ROI-masks for their analyses. Another way that takes inter-individual variabilities in functional neuroanatomy better into account, is to use the individual ROIs determined within the participants who also perform the actual experiment. This approach was applied in the present three studies. It is also called the localizer technique. Either prior to the main experiment (Saxe, Brett, & Kanwisher, 2006) or based on another functional contrast within the main experiment (Friston, Rotshtein, Geng, Sterzer, & Henson, 2006), the functional ROIs are isolated from a contrast that eliminates the respective region of interest (see for example Nieto-Castanon, Ghosh, Tourville, & Guenther, 2003).
ROI-approaches have the advantage that the number of statistical comparisons is greatly reduced compared to the whole-brain analysis, thus minimizing the need for correction of multiple comparisons. Note however, that solely relying on an ROI-approach might prevent one from discovering other regions involved in the processes of interest. Therefore a combined whole-brain and ROI-approach is often a reasonable approach.
Given that functional integration plays an important role in most cognitive processes, investigating the interaction of different brain regions during task performance is important to better understand the neural dynamics of cognitive processes. Different approaches to measure functional connectivity between brain regions have been proposed (Lee & Mechelli, 2003). PPI analysis is such a functional connectivity method which also takes the specific task context into account.
The aim of a PPI analysis is to explain neural responses in one brain region in terms of the interaction between the neural responses in another brain region and a specific psychological context (Friston et al., 1997). PPI analysis thus measures context-sensitive changes in functional connectivity between two regions. For example, PPI analysis can be used to investigate whether posterior brain region related to the perceptual processing of a task are differentially coupled with lateral prefrontal regions under specific task conditions (Friston et al., 1997). This was done in Study 3 of this dissertation.
In general, the PPI analysis consists of a design matrix with three regressors: (1) the “psychological variable” representing two task contexts to be compared, (2) the physiological variable representing the neural response in a given brain region (“seed region”) and (3) the interaction of term (1) and (2). The corresponding subject-specific contrast images of the interaction term are then entered into a random effects analysis. The whole-brain analysis identifies voxels that show increased functional coupling with the according seed region under task context 1 compared to task context 2. To test specific hypotheses about certain brain regions, this approach can be combined with the ROI approach described above.
|© Die inhaltliche Zusammenstellung und Aufmachung dieser Publikation sowie die elektronische Verarbeitung sind urheberrechtlich geschützt. Jede Verwertung, die nicht ausdrücklich vom Urheberrechtsgesetz zugelassen ist, bedarf der vorherigen Zustimmung. Das gilt insbesondere für die Vervielfältigung, die Bearbeitung und Einspeicherung und Verarbeitung in elektronische Systeme.|
|DiML DTD Version 4.0||Zertifizierter Dokumentenserver|
der Humboldt-Universität zu Berlin