Fmri Experimental Design And Analysis

Because images can be collected at the level of seconds (TR for a single slice can be < 1 s, whole brain coverage < 3 s), it becomes possible to collect hundreds of images consecutively, with the primary limiting factors being subject to fatigue or movement over time, or hardware processing constraints. This allows a wide range of study designs. The prototypical fMRI experimental design involves a "boxcar" in which two behavioral tasks alternate over the course of a scanning session, and the fMRI signal between the two tasks or between a task and a resting condition is compared. In the most typical application of this block design, subjects will perform multiple trials of the stimulation (i.e., experimental) task (say for 20 seconds) and then multiple trials of the control task (say for the next 20 seconds), and these conditions will repeatedly alternate over time. The primary analysis essentially involves a subtraction in which one condition is subtracted from the other.

Event-related designs provide the primary alternative to the block design (Buckner et al., 1996; D'Esposito, Zarahn, & Aguirre, 1999). In these studies, individual trials are treated as discrete events, rather than being grouped together as a block of trial. The trials can either be performed in a temporally discrete manner, such that the hemodynamic response is allowed to return to baseline between each trial, or trials can be performed in a manner in which the hemodynamic responses temporally overlap, but are separated enough that the responses can be modeled in relation to a reference function. If responses have significant temporal overlap, as is the case with rapid event-related designs, successful estimation of the evoked hemodynamic responses rely on random presentation of stimuli (i.e., trial Type A is followed by Type B as often as B is followed by A) and highly jittered intertrial-interval durations (Buckner et al., 1996).

Block designs have an advantage over event-related designs in that they provide strong signal detection characteristics over relatively brief times (a single functional scan on the level of 4-7 minutes is often sufficient to detect a substantial BOLD change) (Liu, Frank, Wong, & Buxton, 2001). However, the interpretational power of this design is limited because it cannot disambiguate differential contributions of events occurring within a block or trial (see Figure 13.1b). As described following, event-related designs provide a far more powerful tool in separating the different components of a task.

Consider a spatial delayed response task. The task has three main epochs; a cue period where stimuli to be remembered are presented (say the location of a briefly appearing dot), an unfilled retention period where the location of the dot must be retained in memory, and finally a response period where a memory-guided response is required (say a saccade to the remembered location). In a typical block design, a control condition (not requiring maintenance but attempting to control for other sensory and motor features) is subtracted from the delayed response condition. Because the requirements of the experimental and control tasks have similar visual and motor attributes, but differ in the attribute of interest (i.e., maintenance of the location), subtracting these two blocks is reasoned to yield areas active during memory maintenance. The inferential framework of cognitive subtraction attributes differences in neural activity between the two tasks to the specific cognitive process (i.e., maintenance; Friston et al., 1996; Posner, Petersen, Fox, & Raichle, 1988). However, the assumptions required for this method may not always hold (Zarahn, Aguirre, & D'Esposito, 1999) and could produce erroneous interpretation of functional neuroimaging data. Cognitive subtraction relies on the assumption of pure insertion—that a cognitive process can be added to a preexisting set of cognitive processes without altering the other processes. If pure insertion fails as an assumption, then a difference in the BOLD signal between the two tasks might be observed, not because a specific cognitive process was engaged in one block and not the other, but because the added cognitive process and the preexisting cognitive processes interact.

Continuing with our delayed-response example, the insertion of a maintenance requirement may directly impact the other encoding and retrieval/response processes (e.g., visual encoding; why encode the cue if it will not be used to guide the response made after the delay?). The result is a failure to meet the assumption of cognitive subtraction. Thus, inferences drawn from the results of such blocked experiments may fail to specifically isolate maintenance-related activity.

Event-related designs allow researchers to statistically disambiguate the hemodynamic signals specifically related to encoding the cue stimulus and generating memory-guided responses from the maintenance-related activity present in the retention interval (Aguirre & D'Esposito, 1999). Event-related designs model each component of the trial independently (e.g., cue, delay, and response; see Figures 13.1c and 13.Id). Task designs are often complicated due to the sluggish hemodynamic response, but are feasible as long as different components of the task are temporally varied in relation to each other so that separate aspects of the task can be modeled. Such designs allow separate identification of brain regions involved in encoding spatial locations, maintaining that information across the retention interval, and making the memory-guided response. The ability to model maintenance separately from other task components thus makes it possible to avoid assumptions of pure insertion.

Was this article helpful?

0 0

Post a comment