Functional Magnetic Resonance Imaging - Essay Example

analysis of epoch- and event- related experimental designs in Functional magnetic resonance imaging designs Outline Introduction 2. Task design: problems and solutions 3. fMRI experimental designs in detail 4. Event-related fMRI vs. epoch-related fMRI 5. Conclusion 1. Introduction fMRI is functional neuroimaging method which allows to disclose local changes in brain activity via hemodynamic measures. In fact, fMRI is based upon the following chain of observations and concepts. It is well known that changes in behavioural and cognitive task demands lead to changes in neural activity. Such changes in neuronal activity correlate strongly with changes in blood properties; this is called hemodynamic effect. The oxygen content of the blood increases in a region of the brain with induced activity. Physiologically, oxyhemoglobin concentration increases in an activated region, whereas local deoxhemoglobin concentration decreases. fMRI measures brain activity indirectly, namely by measuring change in oxygen content; this is called the Blood Oxygenation Level Dependent (BOLD) contrast mechanism. However, fMRI has certain limitations. For instance, the fMRI signal reflects changes in oxygen content with high but insufficient spatial-time resolution. Hemodynamics in response to neuronal activity is revealed on a spatial-temporal scale far longer than the neuronal activity itself. Here, so-called temporal "blurring" of the fMRI signal is caused by both inertance and residual effects. In spite of such obstacles, changes in neural activity associated with individual trials or components of a trial in a task can be observed. Moreover, it is possible to capture brain activity associated with a single momentary cognitive act of mentally rotating a stimulus, without recourse to averaging over events (Buckner & Logan 2001, p. 31). Special fMRI experimental designs such as event-related fMRI designs are required in these cases. 2. Task design: problems and solutions There are numerous difficulties in separating the processing roles of specific brain areas. Usually such separation is provided either by well matched task comparisons or through convergence across multiple studies. However, brain activity changes can be relative changes between pairs of tasks, gradual or even nonlinear changes across a series of tasks, or correlations between different tasks. How can tasks and trials within a task be constructed to separate brain cognitive operations This is a key problem of fMRI experimental design. There are several approaches for its solving. The basic approach is to have subjects engage in a target behavioural task for a period of time and then contrast that task period with periods where subjects perform a reference task. Here, the subject might perform a target task, and the measurement obtained during the performance of that task would be contrasted with a measurement obtained when the subject performed a matched reference task. How to substantiate this approach It is obvious that brain activity will change between the two task states and therefore will correlate selectively with the manipulated task demands. When using fMRI, images are taken of the brain repeatedly and in sequence. Brain areas of activation are identified by examining which specific regions change signal intensity as the task state changes from the reference condition to the target task. Then, statistical procedures ranging from direct comparisons between task states to more sophisticated estimations of correlations among task states can be employed to identify those regions whose activity change is unlikely to occur by chance. Unfortunately, tasks designed by such approach may cause differences in the processing strategies adopted by subjects during task performance by means of the blocking of trials, which may result in differential patterns of neural activity that do not have to do with the item-specific processes elicited by the individual trials. This issue can appear in delicate forms in cognitive paradigms where subject strategies may be influenced by the predictability of events; see details in Cabeza & Kingstone (2001). Recent design approaches such as event-related fMRI can avoid the problems of analyzing and interpreting data from blocked designs by isolating individual trials of tasks. Event-related fMRI (or ER-fMRI) is a set of methods which allow "functional neuroimaging procedures to regain the experimental flexibility afforded to traditional cognitive paradigms" (Buckner & Logan 2001, p. 27), including imaging brain areas active during rapidly presented and randomly intermixed types of trials. In accordance with Henson (2004), ER-fMRI is simply the use of fMRI to detect responses to individual trials, in a manner analogous to the time-locked event-related potentials (ERPs) recorded with EEG. The neural activity associated with each trial is usually modelled as a delta function (an "event") at the trial onset. ER-fMRI procedures allow contrasting activity associated with different trial types and even separating activity associated with subcomponents of a trial. For instance, comparison of an ER-fMRI study and a blocked trial study of memory retrieval shows that modulation of the prefrontal areas in relation to retrieval success is closely tied to the use of a blocked trial procedure (Buckner & Logan 2001, p. 34). Here, ER-fMRI separation of successfully retrieved and correctly rejected items indicates equal levels of prefrontal signal change. In this case, the blocked trial design, where item events are predictable, may have encouraged subjects to adjust their strategy and influenced prefrontal participation. It is important that ER-fMRI designs can circumvent this issue by presenting trials randomly and contrasting different trial types under conditions where the specific upcoming event type cannot be predicted; e.g. see Burock et al (1998). In fact, ER-fMRI methods have extended the spectrum of experimental task designs and analytical techniques for neuroimaging studies. As follows from the example above, ER-fMRI allows design techniques to move from "blocked" tests (in which long periods of task performance are integrated) to procedures that isolate individual trial events or subcomponents of trial events. This feature provides much greater flexibility in fMRI experimental design by allowing for selective averaging of stimulus events or task conditions that may be intermixed on a trial-by-trial basis. Additionally, by focusing on responses to single events rather than to extended blocks, ER-fMRI design provides "a means of examining questions regarding the dynamics and time course of neural activity under various conditions"; see details in Cabeza & Kingstone (2001). 3. fMRI experimental designs in detail Let us consider the basis of event-related and epoch-related fMRI experimental designs in detail. fMRI technique is closely related with properties of the BOLD-contrast hemodynamic response. There are two key characteristics of the BOLD hemodynamic response most important for event-related experimental designs. Firstly, the BOLD hemodynamic response can be detected following even brief periods of stimulation. Secondly, the BOLD response can be characterized by a predictable response function. The first characteristic of the BOLD hemodynamic response is proved by numerous observations of clear increasing in the hemodynamic signal even for short periods of neuronal activity, e.g. 0.5 sec movements. These observations show that fMRI can detect hemodynamic responses to very brief neuronal events, making it possible to be used in a really event-related fashion (Rosen et al 1998). So, a number of event-related fMRI experimental designs have exploited these methods. The second characteristic of the BOLD hemodynamic response is the nature and consistency of the shape of the response to a given brief and fixed interval of neuronal activity. It is known that the hemodynamic response is delayed in onset from the time of presumed neural activity by about 3-5 sec. Also, more subtle components of the BOLD hemodynamic response may have considerably longer recovery periods. From experimental data sets, it is possible to incorporate explicit models of the hemodynamic response function into analysis of time-series data, in order to better account for the response lag and delayed offset properties (Wimmer 2003, Henson 2004). Advanced analytic methods are necessary to exploit features of ER-fMRI design schemata more effectively. Indeed, fMRI time series can be considered as the underlying neuronal activity expressed through the hemodynamic response with added noise. The noise sources include measurement noise and some physiological effects. Both fMRI signal and all noise-producing effects must be modelled and then analysed to provide (Wimmer 2003, p. 55): 1) optimum experimental design; 2) optimum filtering of the fMRI time series to obtain efficient parameter estimates; 3) robustness of the statistical inference about the parameter estimates that ensue. Several fMRI experimental designs utilize full implementation of the statistical framework of the general linear model. Such methods seem to be the most flexible because interactions of event types with time and performance variables can be simply coded. Complex analytical models explore so-called design matrices to obtain values of the response or dependent variables. Here, two main ways of experimental design are used in the analysis of fMRI time series, namely epoch- and event-related designs: "The design matrix can be configured in terms of epochs or events corresponding to an epoch or event related response" (Wimmer 2003, p. 57). Epoch-related design was the first design for fMRI time series. For this type of fMRI designs measurements of prolonged states of brain activity are necessary. Also this adoption of state-based design is called as "block" design. For the epoch-related response scans are treated as time series and a block designs used. Blocks of the stimuli or task conditions are presented typically for several seconds, alternating with periods of rest or control conditions. The epoch-related design seems to be superior in terms of theoretical results and interpretability. A newer fMRI design, the event-related design, allows to take into account certain psychological constraints or confounding factors, such as habituation or fatigue. In this design stimuli are presented as isolated brief events separated in time so that the individual response to single events can be identified. Event-related design yields higher frequency signals. In general, the argumentation is in favour of the event-related design. Some advantages of the event-related design are (Wimmer 2003, p. 57): 1) the randomized trial order; 2) the subjective classification of trials; 3) the fact that some events can only be indicated by subject in time; 4) the application to trials which cannot be blocked; 5) these models are even more accurate for blocked designs. In both these designs the explanatory variables are created by convolving a set of delta functions (or so-called "stick" functions), indicating onset times of a particular epoch/event with a small set of basis functions that model the hemodynamic responses to those epochs/events. The idea behind the convolution with the hemodynamic response function is that only experimental effects whose frequency structures survive convolution with the hemodynamic response function can be estimated with any efficiency. That is because the hemodynamic response function, a function of blood oxygenation, flow and volume, models the BOLD impulse response. This function reaches its maximum oxygenation at 3-5 seconds post stimulus and returns to baseline after 20-30 seconds. Further an initial undershoot can be observed; it differs among regions and subjects. Shorter stimulus onset asynchrony (SOA), which is the time between the onset of two consecutive appearances of the same condition/trial type, is more sensitive. For that reason, if the BOLD response is modelled the SOA can be shorter than the return time to baseline without affecting the analysis. Experimental variance should therefore be elicited with reference to the transfer function of the hemodynamic response function; see details in Wimmer (2003) and Henson (2004). For optimal fMRI experimental design it is important that the design complies with the natural constraints imposed by the hemodynamic response to ensure that experimental variance occupies these intermediate frequencies. Noise is much more prevalent at low frequencies; hence those low frequencies should be avoided by experimental design. There are many basis functions for the epoch- and the event-related fMRI designs. The simplest and mostly used basis function for the epoch-related fMRI design is a step function with value 1 in the activation periods and 0 in the rest periods. Other possible basis functions for the epoch-related fMRI design are fixed response (as a rule, half-sine) and exponential decay, applicable when response decreases over time within an epoch, and basis functions (as a rule, discrete cosine), applicable when steady state response does not occur. The most common basis function for the event-related fMRI design is the canonical hemodynamic response function, which is based on empirical data and usually consists of two gamma functions. In the application of event-related fMRI design it is important to determine the boundary conditions for effectively applying the method, e.g. how close in time separate trial events can be presented. Extremely rapid presentation rates can provide a powerful means of mapping brain function (Burock et al 1998). Here, it is useful to analyse the linearity of the BOLD hemodynamic response over sequential neuronal events. It is known that, on first approximation, changes in intensity or duration have near linear and additive effects on the BOLD response. Moreover, data from visual sensory responses suggest that the hemodynamic response of one neural event summates in a roughly linear manner on top of preceding events. Delicate departures from linear summation have been observed in nearly every study that has examined response summation. Nevertheless, the nonlinearities may be subtle enough to still be considered approximately linear; see details in Cabeza & Kingstone (2001). On the base of these facts it is possible to conclude that it may be reasonable to carry out event-related fMRI experimental designs using presentation rates that are much faster than the time course of the BOLD hemodynamic response (Buckner & Logan 2001, p. 34f). So, ER-fMRI experimental designs can be effectively applied to cognitive tasks and higher-order brain regions. Also, event-related fMRI models can be used in the context of blocked designs (Mechelli et al 2003). They can differ from epoch models with respect to the temporal form or shape of the predicted hemodynamic response. Although the two fMRI models assume the same amount of integrated synaptic activity, the event-related model expresses higher frequencies than the epoch model. Once the assumed synaptic activity is convolved with the hemodynamic response, the two fMRI models predict differential hemodynamic responses, with the epoch model reaching its peak later and returning to baseline sooner than the event-related model; see details in Mechelli et al (2003). This corresponds to differential response onsets and offsets for the event-related and the epoch fMRI model. According to Mechelli et al (2003), the advantage of the event-related fMRI model results from its early onset rather then it's late offset relative to the epoch fMRI model. 4. Event-related fMRI vs. epoch-related fMRI Preliminary analysis shows that event-related fMRI experimental designs can produce more flexible and effective solutions for functional neuroimaging than epoch-related fMRI designs. Let us summarize relative advantages of event-related fMRI. First of all, usage of event-related fMRI experimental design allows intermixing trials of different types, rather than blocking them. The counterbalancing or randomising of different trial-types ensures that the average response to a trial-type is not biased by a specific context or history of preceding trial-types. This is important because the (unbalanced) blocking of trial-types might, for example, induce differences in the cognitive "set" or strategies adopted by subjects. This means that any difference in the mean activity during different blocks might reflect such "state" effects, rather than "item" effects specific to individual trials; see details in Henson (2004). Then, event-related fMRI methods allow categorisation of trial types according to the subject's behaviour. This might include separate modelling of trials with correct and incorrect task performance, or parametric modelling of trial-by-trial reaction times. For instance, in so-called "subsequent memory" experiments subjects perform a simple "study" task on a series of items, followed by a surprise memory test. The latter allows the items in the study task to be categorised according to whether they were later remembered. Brain regions can then be isolated whose activity "predicts" subsequent memory (Henson 2001). Next advantage of event-related fMRI designs reflects the identification of events whose occurrence can only be indicated by the subject, e.g. the spontaneous transition between the perception of ambiguous visual objects (note that in this case the objective stimulation is constant). Then, event-related fMRI methods allow some experimental designs that cannot be easily blocked, e.g. when the stimulus of interest is one that deviates from the prevailing context, and therefore cannot be blocked by definition. Finally, event-related fMRI methods potentially allow more accurate models of the data. Indeed, even when trial-types are blocked, modelling the BOLD hemodynamic response to each trial within a block may capture additional variability that is not captured by a simple neuronal model. Furthermore, it is possible distinguish between state effects and item effects; see details in Henson (2004). There are also disadvantages of event-related fMRI designs associated with randomised character of these designs. Firstly, such designs are generally less efficient for detecting effects than blocked designs with short SOAs and reasonable block lengths. Secondly, such psychological manipulations as changes in selective attention or task may exert stronger effects when blocked. Also it is necessary to note that even in blocked design fMRI, an event-related analysis may provide a more accurate model of the hemodynamic responses than an epoch-related analysis. This is because "the temporal shape of the predicted response differs between the event-related and the epoch model, with the former reaching its peak sooner and returning to baseline later than the latter" (Mechelli et al 2003, p. 806). Furthermore, in a number of experiments the event-related fMRI model explains changes in activities that are not accounted for by the epoch model. In accordance with Mechelli (2003), the advantage of the event-related over epoch model is caused by its early onset rather than its late offset, relative to the epoch model. 5. Conclusion It is necessary to distinct event-related experimental designs and epoch-related designs in functional magnetic resonance imaging. In event-related fMRI designs stimuli of different types are intermixed whereas in blocked fMRI designs stimuli of the same type are presented in blocks. Effects of interest in blocked fMRI designs are usually modelled with some form of step basis function convolved with a synthetic hemodynamic response function. Here, it is assumed that synaptic activity and hemodynamics are attained within each block. On the contrary, effects of interest in event-related fMRI designs are modelled by convolving each trial onset (i.e., a "stick" function) with a synthetic hemodynamic response function; see details above. Here, the hemodynamic responses to stimulus-induced neuronal transients are modelled without assuming constant within-block activity; e.g. see Clare (1997), Buckner & Logan (2001), Gore (2003), and Mechelli et al (2003). Event-related fMRI design is characterized by numerous advantages over epoch-related fMRI design, namely: the randomized trial order; the subjective classification of trials; the fact that some events can only be indicated by subject in time; the application to trials which cannot be blocked; more accuracy even for blocked designs, etc. Epoch-related fMRI models may provide greater sensitivity than event-related fMRI models only under certain circumstances. For instance, if the hemodynamic response rises slowly or quickly returns to baseline, an epoch-related fMRI model may explain the response profile better than an event-related fMRI model; this is because epoch models reach their peak later and return to baseline earlier than event-related models (Mechelli et al 2003). However, in most cases event-related fMRI models may characterize the form of the observed hemodynamic response better than an epoch-related fMRI models, thereby maximizing fMRI sensitivity. References Buckner, RL & Logan, JM 2001, 'Functional neuroimaging methods: PET and fMRI', in R Cabeza & A Kingstone (eds), Handbook of functional neuroimaging of cognition, The MIT Press, pp. 27-48. Burock, MA, Buckner, RL, Woldorff, MG, Rosen, BR & Dale, AM 1998, 'Randomized event-related experimental designs allow for extremely rapid presentation rates using functional MRI', NeuroReport, vol. 9, pp. 3735-3739. Clare, S 1997, Functional MRI: methods and applications, PhD Thesis, University of Nottingham. Retrieved Oct 27, 2007 from http://users.fmrib.ox.ac.uk/stuart/thesis/fmri.pdf Gazzaley, AH & D'Esposito, M 2005, 'BOLD functional MRI and cognitive aging', in R Cabeza, L Nyberg & D Park (eds), Cognitive neuroscience of aging: linking cognitive and cerebral aging, Oxford University Press, pp. 107-131. Gore, JC 2003, 'Principles and practice of functional MRI of the human brain', The Journal of Clinical Investigation, vol. 112, no. 1, pp. 4-9. Henson, RN 2001, 'Event-related fMRI: introduction, statistical modelling, design optimisation and examples', Japanese Journal of Cognitive Neuroscience. Retrieved Oct 29, 2007 from http://www.mrc-cbu.cam.ac.uk/rh01/henson-jjcn.pdf Henson, RN 2004, 'Analysis of fMRI time series: linear time-invariant models, event-related fMRI and optimal experimental design', in Frackowiak, Friston, Frith, Dolan & Price (eds), Human Brain Function, pp. 793-822. Kida, I & Hyder, F 2006, 'Physiology of functional magnetic resonance imaging: energetics and function', in PV Prasad (ed), Magnetic Resonance Imaging: methods and biologic applications, Humana Press, Totowa, pp. 175-195. Mechelli, A, Henson, RN, Price, CJ & Fristona, KJ 2003, 'Comparing event-related and epoch analysis in blocked design fMRI', NeuroImage, vol. 18, pp. 806-810. Rosen, BR, Buckner, RL & Dale, AM 1998, 'Event-related functional MRI: past, present, and future', Proceedings of the National Academy of Sciences USA, vol. 95, pp. 773-780. Wimmer, K 2003, 'fMRI time series', in FMRI Time Series Analysis with the software SPM99, Proceedings of the OENB WG of the Vienna University of Technology, pp. 55-63. Read More