Functional Imaging of Neurocognition

Mark D'Esposito, M.D., Helen Wills Neuroscience Institute and Department of Psychology, University of California, Berkeley, California.

Semin Neurol. 2000;20(4) 

In This Article

Lesion Studies Versus Functional Imaging Studies

Lesion studies (i.e., studies of patients with deficits that follow specific brain damage) and functional neuroimaging studies of normal, healthy subjects provide complementary but different types of information regarding brain-behavior relationships. Amajor premise of this review is that both of these types of studies are necessary to provide an inferentially sound basis for drawing conclusions about the neural basis of cognition. Converging evidence from both types of studies is necessary because of inferential limitations of lesion and functional neuroimaging studies performed in isolation.

The nature of functional neuroimaging studies is that they support inferences about the association of a particular brain system with a cognitive process. However, functional neuroimaging is unable to prove that the observed activity is necessary for a putatively isolated cognitive process. This is because one never has perfect control over the cognitive processes in which a subject engages. Although an experiment may control the conditions or tasks to which a subject is exposed, it cannot conclusively demonstrate that a subject is differentially engaging a single, identified cognitive process. It should be noted that "a more sensitive behavioral test" cannot solve this problem, as it is always possible that the subject engages unnecessary cognitive processes that either have no overt, measurable effects or are perfectly confounded with the process of interest. As a result, observed neural activity may be the result of some confounding neural computation which is not itself necessary for the execution of the cognitive process seemingly under study. An equivalent formulation of these statements is to note that, essentially, neuroimaging is an observational,correlative method.[9] It is important to note, however, that the inferences that can be drawn from functional neuroimaging studies apply to all methods of physiological measurement, including single-unit and multiunit electrophysiology, EEG, magnetoencephalography, hemodynamic measures, and measures of glucose metabolism.

The inference of necessity cannot be made without a demonstration that the inactivation of a brain system disrupts the cognitive process in question. However, un-like the lesions produced in animal models, which can be precisely created with neurotoxins, lesions in human patients are often extensive and damaging, not only to local neurons but also to "fibers of passage."[10] It is also possible that connections from region "A" support the continued metabolic function of region "B" but that region "A" is not computationally involved in certain processes undertaken by region "B." Mechanisms that might be imagined to produce such an effect include diaschisis11 or retrograde transsynaptic degeneration. Consequently, lesions of the kind often studied in humans cannot conclusively demonstrate that the neurons within a specific area are themselves critical to the computational support of an impaired cognitive process.

An example of these issues regarding the types of inferences that can be drawn from various lesion versus electrophysiology studies is illustrated as follows. Single-unit recordings in monkeys have revealed neurons in lateral prefrontal cortex (PFC) that increase their firing during a delay between the presentation of information and its later use in behavior.[12,13] These studies have been taken as evidence that lateral PFC represents a neural correlate of working memory, the cognitive processes governing the temporary maintenance and manipulation of information.[14] The necessity of this region for working memory has been demonstrated in monkey studies that have shown that lesions of the lateral PFC impair performance on delayed-response tasks but not on tasks that require visual discrimination and saccades without the requirement of holding information on line.[15] Delay-specific neurons have also been found in the hippocampus,[16,17] a region thought to be involved in long-term memory, as opposed to working (i.e., short-term) memory. Lesions of the hippocampus, however, do not impair performance on delayed-response tasks (with short delay periods), suggesting that the hippocampus may be involved in maintaining information over short periods of time but is not necessary for this cognitive operation.[18] Observations in humans also support this notion. For example, the well-studied patient HM, with complete bilateral damage to the hippocampus and severe inability to learn new information, can nevertheless perform normally on working memory tasks.[19]

When combined, however, a stronger level of inference results from lesion and functional neuroimaging studies. One type of combination might be that (1) lesions to a cortical area impair a given cognitive process and that (2) the cognitive process, when engaged by intact subjects, evokes neural activity in the same cortical area. The inference that the neuroanatomical area is computationally necessary for the cognitive process is now rendered less vulnerable to the faults detailed before for each method in isolation, although it is still possible to conceive of failures. As a result, neuroimaging and lesion studies are complementary in that each provides inferential support that the other lacks.

Interpretation of functional neuroimaging studies can potentially lead to another type of inferential failure. Two types of inferences can potentially be applied to functional neuroimaging data (Zarahn E, Aguirre GK, submitted). A "forward inference," which is typical for most imaging experiments, derives from the assumption that if a particular brain region is activated by a cognitive process (evoked by a particular task) then the neural activity in that brain region must depend on engagement of that particular cognitive process. For example, a brain region that responds with a greater magnitude of fMRI signal to face stimuli as compared with other stimuli (cars, buildings, etc.) would be considered to be activated by face perception. Alternatively, a "reverse inference" derives from the assumption that if a particular brain region is activated then a cognitive process must have been engaged by the subject during the study. This type of inference is prevalent within discussion sections of functional neuroimaging papers. For example, activation within prefrontal cortex during a mental rotation task20 might be taken as evidence that subjects were using working memory to remember the identity of the rotated target (this assumption was derived from previous imaging studies that have shown activation of PFC during working memory tasks). It can be shown in general that reverse inferences of this type are logically incompatible with simultaneous forward inferences (Zarahn E, Aguirre GK, submitted). In our previous example, one cannot be sure that the working memory was engaged to evoke PFC activation because some other cognitive process could have activated this region.[21] In sum, this discussion of the types of inferences that can be drawn from different types of studies should help the investigator (as well as the consumer) to derive conclusions from fMRI data that do not rely on faulty assumptions.