Task analysis complements neuroimaging: an example from working memory research

Abstract

When functional neuroimaging researchers draw conclusions about the sensory, cognitive, or motor processes that are associated with changes in brain activity, they are making assumptions about the component processes involved in performing a complex behavioral task. We demonstrate the danger in making such assumptions using, as an example, the n-back task, which has been widely used in neuroimaging studies of working memory. Neuroimaging researchers have assumed that the letter n-back task only engages processes involved in the short-term maintenance and manipulation of verbal information. We report three behavioral experiments demonstrating that the letter n-back task additionally recruits spatial processes. A fourth experiment suggested that the location n-back task may recruit verbal processes in addition to spatial processes. These results call into question conclusions that have been drawn about the neural basis of working memory. More broadly, our results demonstrate that task analysis is a vital partner of neuroimaging in the cognitive neuroscience enterprise.

Introduction

Recently developed functional neuroimaging methodologies, such as functional magnetic resonance imaging (fMRI) and positron emission tomography (PET), have made important contributions to the understanding of the neural basis of sensory, cognitive, and motor behavior. The scope of these contributions is limited, however, for a variety of reasons D'Esposito et al., 1999, Friston et al., 1996, Jennings et al., 1997, Kosslyn, 1999, Newman et al., 2001, Poeppel, 1996, Uttal, 2001. One critical limitation of neuroimaging is that it requires that assumptions are made about the component processes involved in performing a behavioral task. When a neuroimaging report states that a change in activity in a particular brain region (or regions), measured while participants perform a task, is associated with a particular process, the authors have assumed that (a) the process is required for performance of the task, and (b) no other processes required for performance of the task are associated with the change in activity. Long before neuroimaging tools were developed, cognitive psychologists had developed methods for determining the component processes comprising cognitive tasks. Thus, it is critical that cognitive psychology be used in conjunction with neuroimaging to confirm which processes are involved in which tasks.

Although in some cases neuroimaging researchers directly adopt tasks for which cognitive psychology has thoroughly analyzed the component processes, this is not always the case. An example of the latter is the n-back task Cohen et al., 1994, Gevins et al., 1990, which has been widely used in neuroimaging studies of working memory despite minimal prior task analysis.¹ In the n-back task, participants are presented with a sequence of stimuli, and asked to judge whether the current stimulus matches the stimulus that preceded it by n places in the sequence. For example, in a 3-back task with the sequence of letters B, Q, D, P, Q, participants should respond positively to the second Q because it matches the letter that appeared three letters earlier. Neuroimaging researchers were attracted to the n-back task for several reasons Braver et al., 1997, Cohen et al., 1997, Jonides et al., 1997. First, the n-back task was thought to be a more representative example of a working memory task than the tasks used in earlier neuroimaging studies because it was presumed to require manipulation of information held in working memory in addition to mere maintenance of information. Second, it allowed the use of the method of parametric variation instead of the method of subtraction, thereby avoiding some of the problematic assumptions associated with the latter Friston et al., 1996, Jennings et al., 1997. Third, manipulation of the value of n allowed the systematic manipulation of working memory load, and brain regions showing changes in activity coincident with changes in load were thought to play specific processing roles. Neuroimaging researchers have rationalized their use of the n-back task, despite a lack of task analysis, by claiming that it activates the same brain areas as other working memory tasks whose component processes are well understood. There are at least two consistent exceptions to this claim: dorsolateral Brodmann area 6 (dorsal premotor cortex, or PMd) and right posterior parietal cortex (PPC), two regions that are not active during other nonspatial working memory tasks, are active during nonspatial n-back tasks D'Esposito et al., 1998, Smith and Jonides, 1999.

According to current theory, which was introduced by Baddeley and Hitch (1974) and refined by many contributors Baddeley, 1986, Baddeley and Logie, 1999, there are two main types of component systems involved in working memory tasks: the executive system and the subordinate systems. The executive controls the maintenance, manipulation, and retrieval of the contents of working memory, and is thought to be involved in all working memory tasks regardless of the domain of information that is being maintained. Different subordinate systems, under the control of the executive, are thought to be involved in the maintenance of specific domains of information (e.g., verbal, spatial, and visual form). In other words, the executive is thought to be a domain-general system, whereas the subordinate systems are thought to be domain-specific systems. Neuroimaging studies of working memory have exploited this distinction in an effort to identify the brain regions associated with these systems. The logic is as follows: if you design two working memory tasks that are identical except for the domain of the information to be maintained, then brain regions that are activated by both tasks (excepting sensory and motor activation) are associated with the executive, and brain regions that are activated by just one of the tasks are associated with one of the subordinate systems. Studies employing this logic have concluded that prefrontal cortical areas are associated with the executive, and more posterior regions, especially in parietal and frontal cortices, are associated with the different subordinate systems D'Esposito et al., 1998, Smith and Jonides, 1999.

Before the n-back task was introduced in neuroimaging studies, regions of activity in PMd and right PPC were associated with the spatial subordinate system (Jonides et al., 1993). However, studies using the n-back task have shown that PMd and right PPC are active for all domains of information, not just spatial information (e.g., D'Esposito et al., 1998, Nystrom et al., 2000, Postle et al., 2000b, Zurowski et al., 2002). Application of the aforementioned logic forces the conclusion that these PMd and right PPC activations are associated with domain-general processes rather than spatial-specific processes. An alternative possibility is that erroneous assumptions have been made about the component processes involved in the n-back task. The n-back task is similar to other working memory tasks in that it requires the maintenance of stimulus identity and serial position information. However, the n-back task is unique in that it requires the dynamic comparison of serially presented stimuli. Perhaps such comparisons are aided by the mental transformation of serially presented information into a spatial representation. Experiment 1, Experiment 2, Experiment 3 were designed to test the hypothesis that the n-back task, regardless of the domain of information to be maintained, recruits spatial processing.