MEG versus BOLD MR Imaging: Functional Imaging, the Next Generation?

Blood oxygen level–dependent functional MR (BOLD fMR) imaging, long accepted as a powerful research technique in the cognitive sciences and neurosciences, recently has gained acceptance as a clinical tool. Clinical applications of BOLD imaging have focused primarily on the preoperative localization of the motor, sensory, and language centers of the brain in anticipation of tumor or vascular malformation resection, and in the functional evaluation of focal cortical dysplasias in epilepsy. Neuronal firing is not directly measured by fMR imaging. Rather, in BOLD imaging, as its name implies, neuronal activation is inferred from small, local MR signal changes (on the order of 3% at 1.5 T field strength) proportional to hemodynamically induced alterations in net deoxyhemoglobin concentration caused by task-related increases in neuronal metabolism. Thus, although capable of spatial resolution on the order of millimeters, BOLD fMR imaging is limited in its temporal resolution to the time interval required for a change in neuronal activity to produce a measurable hemodynamic response, typically approximately 2 to 5 seconds.

Magnetoencephalography (MEG), on the other hand, a somewhat newer technique, not only directly measures the magnetic field changes associated with neuronal firing, but is capable of temporal resolution on the order of milliseconds. This high temporal resolution is likely to prove especially valuable in tracking transient, complex, coordinated neuronal activation patterns involved in higher-order cognitive functions (such as visual memory formation), which are known to occur across large segments of the brain (1). In MEG, an array of hypersensitive, superconducting magnetic detectors translate minute, rapidly changing magnetic fields (on the order of one billionth the strength of the earth's magnetic field) into detectable alterations in electric current. This is accomplished, however, at the price of decreased spatial resolution compared to that of fMR imaging. The spatial resolution of MEG, for unspecified magnetic source distributions, is typically limited to several centimeters.

In this issue of the AJNR, Roberts et al (page 1377) report on the difference between the ability of BOLD fMR imaging and MEG to quantify evoked responses. Specifically, they studied five subjects, all of whom underwent a similar sensory paradigm, using both techniques. The “task” involved successive stimulation of one, two, three, and four digits of the left hand. For fMR imaging, activation was quantified in two ways: first, as the extent of cortical activation, and second, as the amount of activation (defined as the product of the number of activated pixels and the mean signal change per pixel). For MEG, activation was also quantified in two ways: first, as the magnitude of the evoked magnetic field peak, and second, as the strength of the modeled current source, Q. Using fMR imaging, a trend toward an increased number of activated pixels for an increased number of stimulated digits was not found to be statistically significant, and a very high intrasubject and intersubject variability in pixel activation was noted. Using MEG, however, the evoked field magnitude was found to vary linearly with the number of digits activated in a statistically significant way, and intrasubject and intersubject variability of activation was noted to be much less than it had been for fMR imaging. The authors concluded that, for the particular somatosensory task they studied, robust quantification of evoked responses is possible with MEG, but robust quantification of increasing cortical area of activation is not possible with fMR imaging.

In attempting to assess the clinical impact of these results, two questions present themselves. The first is a question of methodology; were alternative experimental designs possible? Did this study compare BOLD and MEG so as to optimize the possibility of quantitation for each technique, and are the results reproducible? The second is a question of relevance; are the conclusions of this study important? Does quantification of cortical activation examinations matter clinically, and what are the potential roles for MEG and fMR imaging in patient care?

With regard to the first question, alternative experimental designs might have cast a more favorable light on the potential of fMR imaging to document reproducible, quantifiable, evoked responses. The conclusion reached by Roberts et al holds strictly only for the particular paradigm that they studied. It is possible that by varying the frequency, duration, intensity, or even the type of stimulus presentation, rather than varying only the number of digits stimulated, the resulting fMR imaging signal changes might have been more robust to quantification (or the MEG changes less). The way in which quantification is defined, which was different for the two techniques studied, could also affect the results. Because robust hemodynamic alterations are detectable after neuronal stimuli lasting only a few 10s of milliseconds, a newer method of BOLD imaging, “event related” fMR imaging, has been designed to measure activation in response to single sensory or cognitive events (2). Had this study used an “event-related” rather than a “block” design for its stimulus presentation, or had the “power” of the fMR imaging response (the area under the signal intensity versus time curve for an activated pixel) been measured, rather than the extent and amount of activation (as defined by Roberts et al), the results might have been different. fMR imaging data collection using a magnetic field strength greater than 1.5 T (3 T or higher) also could have improved the signal-to-noise ratio of the BOLD effect.

With regard to the second question, accurate, reproducible quantitation of functional activation could be valuable not only to monitor serially the response of individual patients to treatment in situations such as stroke rehabilitation, but also as an objective surrogate marker of disease progression when comparing clinical trial outcomes among different subjects. For such applications, a low intrasubject and intersubject variability is required. For preoperative planning, although localization of function is crucial, quantitation of evoked responses might help determine the significance of equivocal foci of activation, which may be present in regions of peritumoral edema or mass effect (3). In the future, MEG also may prove useful in quantifying disease states such as epilepsy, memory disorders (dementias), or language disorders, for which structural derangement might not be apparent on conventional MR imaging.

An important drawback in the use of MEG to evaluate diseases such as epilepsy, however, (despite its great sensitivity for detecting abnormal “spikes” of neuronal activation) is that anatomic localization with MEG is highly model dependent. When, as in the study by Roberts et al, the approximate brain region of an evoked response is known with high certainty (the postcentral gyrus, in the case of somatosensory stimulation), the models used to infer the strength and location of the current source, Q, can be considered reliable. When the approximate location of a current source is less certain, or the sources are multiple and widely distributed over the cortical surface, anatomic localization with MEG can be considerably less reliable.

Given the preceding discussion, it is clear that important differences exist between fMR imaging and MEG in their sensitivity for detection of evoked responses, how they measure such responses, and what those responses mean. MEG and fMR imaging techniques have largely complementary strengths and weaknesses, not only with respect to their ability to quantify functional activation, but also in their spatial and temporal resolutions. Rather than emphasize the limitations of each of these techniques, however, it makes sense to attempt to exploit their strengths in order to optimize their clinical and research utility. To this end, techniques recently have been developed that combine the spatial resolution of fMR imaging with the temporal resolution of MEG to create “anatomically constrained” functional activation maps (4). These maps are calculated by using the high resolution fMR imaging and structural MR imaging datasets, obtained during activation studies, to model more precisely the highly temporally resolved MEG current sources. The results are superimposed onto an “inflated” cortical surface representation, and can be displayed as “movies” showing spreading waves of cortical activation.

Using such “anatomically constrained” maps, it is possible to obtain highly spatially and temporally resolved functional information that is unavailable from data produced by either BOLD fMR imaging or MEG alone. This technique has been used successfully to study parallel cortical activation during semantic processing of visually presented words in diverse brain regions associated with perception, semantic processing, and response choice (4). By building on the complementary strengths of fMR imaging and MEG, precise monitoring of the spatiotemporal orchestration of complex, high-order, perceptual and cognitive neuronal activations has the potential to become a clinically useful technique for the evaluation of both normal and diseased states.