Cognitive Control and Value-based Decision-Making

Cognitive Control and value-based decision-making are two critical and often opposing functions that shape human actions. While cognitive control insure behavioral flexibility in the face of changing environments by exerting fine-tuned control including inhibiting of prepotent responses. Value-based decision-making on the other hand is driven my the motive of reward maximization. 

Cognitive control comprises several executive functions including flexible task switching, response inhibition, detection of errors and response conflict, and working memory. Although there is a long research tradition in cognitive control, it is yet unclear, whether cognitive control can be thought of a unitary construct or a collection of distinct, but related functions. We tested 344 participants on several measures of cognitive control and the Iowa Gambling Task, a measure of value-based decision-making. Strikingly, we observed that several distinct executive functions fractionated into different areas of the prefrontal cortex suggesting that the brain maintains separate neural module for implementing these different cognitive functions.

We also looked for a common neural signature of all cognitive control function by extract a single common factor using factor analysis. The neural signature of this common factor was located the anterior cingulate cortex (ACC), overlapping with the neural correlates of two switching tasks (Trail Making B, Wisconsin Card Sorting Test, WCST) suggesting that this common factor reflects the critical ability to flexible sift between task and response sets. In contrast, the neural correlate of value-based decision-making was located in the ventromedial prefrontal cortex (vmPFC) consistent with many other studies that report value signals in this area. Finally, contrasting the common cognitive control factor with the scores for value-based decision-making revealed two PFC networks: lateral one (including ventro- and dorsolateral PFC, shown in red) that is mainly concerned with cognitive control, and a more medial one (including vmPFC and orbitofrontal cortex, shown in blue) that is processing the motivational value of predictive stimuli.

Model-based Lesion Mapping of Cognitive Control

Model-based fMRI (Gläscher & O'Doherty, 2010) enables powerful insights into which brain regions might implement a specific computational process derived from a cognitive model. In principle the same approach is also possible for lesion mapping by relating parameters of a computational model to lesion overlap analysis. The general idea is to estimate the model parameters either hierarchically (or individually) thus allowing for subject-by-subject variation among them. This variability in the model parameters is then analyzed using lesion overlap analysis. Through this method is thereby possible to identify critical neural structures for a neurobiological implementation of a cognitive model.

A crucial different exists between model-based fMRI and model-based lesion mapping. In model-based fMRI we can analyze a specific computational signal (e.g. expected values or prediction errors) as it unfolds over time and identify those areas that show a correlation with this signal across the entire experiment. In model-based lesion mapping we can only access the structural brain images and related those to the model parameters. These model parameters usually modulate the specific computational signals that we can measure with model-based fMRI. Thus, with model-based lesion mapping we can rather detect the modulatory influence on the cognitive operations, whereas with model-based fMRI we can identify the actually computational signal.

We used a sample of more than 350 lesion patients who had been administered the WCST, the most widely used test of cognitive control. The most important index is the number of perseverative errors, which are those errors that are committed when a patient fails to detect a switch in the correct sorting dimension. We used a recent cognitive model for the WCST (Bishara et al., 2010, see left figure) and modeled the test data from all patients using Hierarchical Bayesian estimation.

Of all the different scores derived from the WCST, perseverative errors were the only one with a substantial neural correlate in the right PFC (shown in red). In addition, punishment sensitivity was the only model parameter that also showed a significant neural correlate (shown in green), which overlapped to a large degree with the effects for perseverative errors (overlap shown in yellow). This suggests that decreased sensitivity to negative feedback in the WCST may be a cause for increase perseverative errors, which also makes intuitive sense as negative feedback is the only way that a subject is informed that a shift in the sorting dimension may have occurred.

Domain-specific and General Intelligence

The Wechsler Adult Intelligence Scale (WAIS) is the most widely used test of intellectual abilities worldwide. Besides the most general intelligence quotients for verbal and perceptual abilities (verbal IQ and performance IQ), one can also extract four domain specific indes scores that measure difference aspects of intelligence: verbal comprehension, perceptual organization, working memory, and processing speed.

We used the lesion mapping to identify those areas in the brain that are critical for these different domains of intelligence. We found the verbal comprehension and working memory shared a lesion deficit effect in the left hemisphere reaching from inferior frontal gyrus to superior parietal cortex including the white matter tract connecting these regions. However, verbal comprehension has its focus in the inferior frontal cortex (Broca's area), whereas the strongest effect for working memory was in the parietal cortex, which is part of the fronto-parietal attention network. Perceptual organization mainly depends on the brain regions in the right parietal cortex, whereas processing speed includes several distributed regions throughout the cerebral cortex in accordance with the nature of this intelligence domain. (Processing speed measures (among other things) the efficiency of cortical communication.) 

Since the early days of defining and measuring intelligence it was widely known that most tests of cognitive abilities are moderately correlated with each other, meaning that people who perform well on one test tend to be also good performers on other test of cognitive functions. This lead one of the early psychologist working on intelligence to define a general intelligence factor, g, which captures this cognitive capacity common to all cognitive functions. Using hierarchical factor analysis, we also extracted g from the WAIS test data and mapped its brain regions using lesion mapping. Interestingly, we found that g mostly depended on the main white matter tract in the left hemisphere connecting prefrontal cortex and parietal cortex. This underlines that efficient interregional communication between brain regions are the basis for good cognitive performance.

Furthermore, we also detected a region in the frontal pole that was related to general intelligence, but that could not be explained by any of the contributing, domain-specific cognitive functions. This suggests that general intelligence, is also "more than the sum of its parts" and that it is likely to involve very abstract capacities of executive functions, like allocation of processing resources to different cognitive tasks.