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

Basic Principles of Functional MRI

Over the past few years, fMRI has become the predominant functional neuroimaging method for cognitive activation studies because of several clear advantages over PET. For example; fMRI does not require injection of a radioisotope into subjects and is otherwise noninvasive; it has better spatial resolution, on the order of millimeters; and it has better temporal resolution, on the order of a few seconds compared with tens of seconds. Nevertheless, there are still some advantages of PET scanning in selected circumstances. For example, imaging with fMRI is difficult within the orbitofrontal and anterior/inferior temporal lobe regions because of susceptibility artifact caused by the sinuses in these regions. Improvements in pulse sequences for acquiring fMRI data, however, should eventually eliminate these artifacts.[22,23] Electrical recordings such as the electromyelogram (EMG) or electro-oculogram (EOG) during fMRI scanning also present a technical challenge that has been tackled in some laboratories.[24] The fMRI data are exquisitely sensitive to subject motion during scanning, rendering studies of some subject populations (e.g., agitated patients) problematic, but studies can be performed successfully while the subject makes overt verbal responses.[25] Finally, as the pulsing of the fMRI gradient coils makes a considerable amount of noise, challenges exist for the study of auditory processes.[26,27]

A basic understanding of the theorized physiological mechanism of fMRI signal change provides a foundation for the design and interpretation of fMRI studies of cognition. For more detail of MRI physics, the reader is referred to MRI Physics for Radiologists.[28] In MRI, (hydrogen) nuclei in tissue are excited (i.e., put into a higher energy state) by brief exposure to radio-frequency (RF) energy. The excitation of hydrogen ions results in a change in the orientation of the magnetic moments (i.e., direction of the respective magnetic fields) of the nuclei in the tissue away from the main magnetic field of the MRI scanner. After excitation, the magnetic moments take time to reorient with the main magnetic field (i.e., return to a lower energy state). As they reorient, each nucleus precesses (a wobbling motion similar to that of a spinning top) about the main magnetic field, releasing RF energy at a frequency that depends upon its precession frequency. The precession frequency of a nucleus depends upon the local magnetic field. Deoxyhemoglobin is paramagnetic, meaning that it possesses magnetic properties. Because it is paramagnetic, increases in the concentration of deoxyhemoglobin in a brain region increase local inhomogeneity in the magnetic field, which leads to a decreased magnetic resonance signal.

The sensitivity to detect fMRI signals seems to be due to the blood flow-mediated relationship between the concentration of deoxyhemoglobin and neural activity. When a neural event occurs in a region of the brain, there is a subsequent increase in local blood flow[29] that paradoxically results in a decrease in the concentration of deoxygenated hemoglobin in the local microvasculature of the activated region.[30] This change leads to an increase in fMRI signal that has been called the blood oxygenated level dependent (BOLD) response.[31,32] Thus, fMRI measures blood flow and is an indirect measure of neural activity.

The temporal dynamics (i.e., the time scale on which changes can occur) of neural activity are quite rapid, even in association (i.e., nonprimary) cortices such as the parietal cortex. For example, neural activity in the lateral intraparietal area of monkeys increases within 100 milliseconds of the visual presentation of a saccade target.[33] In contrast, it is a well-replicated observation that the fMRI signal gradually increases to its peak magnitude within [4,5,6] seconds after an experimentally induced, putatively brief (i.e., < 1 second in duration) change in neural activity and then decays back to baseline after several more seconds.[34,35,36] This slow time course of fMRI signal change in response to such a brief increase in neural activity is informally referred to as the BOLD fMRI hemodynamic response (or simply hemo-dynamic response). An example estimated hemodynamic response (acquired from the primary sensorimotor cortex of a normal human subject during a motor response to a visual stimulus) is illustrated in Figure 1. In contrast, neural activity changes in primary motor cortex associated with simple reaction time tasks are, for the most part, concluded within a few hundred milliseconds of presentation of the response cue.[37] Thus, neural dynamics and neurally evoked hemodynamics (as measured with fMRI) are on quite different time scales.

A summary measure of the hemodynamic response function (i.e., fMRI signal change in response to a brief increase in neural activity) from primary motor cortex across several subjects. The fMRI signal displayed is derived from a motor task during which subjects made a button press every 16 seconds when a fixation cross changes to a circle. Note that the fMRI signal peaks approximately 5 seconds after the onset of the motor response (which occurred at time zero).

The sluggishness of the hemodynamic response limits the effective temporal resolution of the fMRI signal to a few seconds, as opposed to the millisecond temporal resolution of electrical recordings of neural activity (such as from single-unit studies in monkeys or event related potentials (ERP) studies in humans). A consequence of the limited temporal resolution of fMRI is that one will have reduced statistical power if attempting to distinguish sequential changes in neural activity that are occurring rapidly with respect to the hemodynamic response. That is, the ability to resolve statistically the changes in fMRI signal associated with two neural events is likely to require a separation of those events by a duration that is relatively long compared with the width of the hemodynamic response. To provide some sense of the relevant time scales, we estimate that evoked fMRI responses to discrete neural events separated by at least 4 seconds are well within the range for resolution.[38] However, several investigators have reported significant differential functional responses between two trial types with trials spaced as closely as 500 milliseconds apart,[39,40,41] provided that these stimuli are presented in a random fashion. Some tasks, such as delayed-response tasks, have individual trial events (such as the presentation of the cue, the delay period, and the response) whose order cannot be randomized. In these types of tasks, temporal resolution on these short time scales that is possible with trial events that can be randomized (less than 2 seconds) will not be possible.

Perhaps counter to intuition, the limited temporal resolution of fMRI does not necessarily mean that one cannot detect with reasonable statistical power brief changes in neural activity (as demonstrated in Fig. 1). Several investigators have reported fMRI signal changes that were correlated with events lasting brief periods of time. For example, fMRI responses have been observed in the sensorimotor cortex in association with single finger movements42 and in the visual cortex during very briefly presented (i.e., 34 msec) visual stimuli.[43] Thus, fMRI has superb temporal and spatial resolution, which makes it a powerful methodology for neuroscience research.