Magnetoencephalography (MEG)

What is it? What does it measure?

Magnetoencephalography (MEG) is a non-invasive technique used to measure magnetic fields generated by small intracellular electrical currents in neurons of the brain. Thus MEG provides direct information about the the dynanamics of evoked and spontaneous neural activity and the location of their sources in the brain.

MEG and EEG are closely related, the latter detecting the electric potentials generated by neural currents instead of the corresponding magnetic fields. However, it turns out that the task of inferring the sites of brain activation is often more straightforward from MEG than from EEG. This is due to the electric and magnetic properties of the tissues in the cranium and also to the fact that MEG is selectively sensitive to currents flowing tangential to the scalp, corresponding to sulcal activations. On the other hand, the interpretation of EEG is often complicated by the simultaneous presense of both sulcal and gyral sources, the latter corresponding to radial currents.

How does it work? What equipment is needed?

MEG measurements are conducted externally, using an extremely sensitive device called a superconducting quantum interference device (SQUID). The SQUID is a very low noise detector of magnetic fields, which converts the magnetic flux threading a pickup coil into voltage allowing detection of weak neuromagnetic signals. Since the SQUID relies on physical phenomena found in superconducators it requires cryogenic temperatures for operation. In a modern MEG device, an array of more than 300 SQUIDS is contained in a helmet shaped liquid  helium containing vessel called a dewar, allowing simultaneous measurements at many points over the head.  The MEG system is operated in a shielded room that minimizes interference from external magnetic disturbances, including the Earth’s magnetic field, noise gerated by electrical equiment, radiofrequency signals, and low frequency magnetic fields produced by moving magnetic objects like elevators, cars, and trains.

What do the data look like?  How is information extracted?

Presentation of sensor data

Both MEG and EEG raw data are often presented as time-dependent signals, arranged in a topographical layout. These data may either represent averages over repeated sensory stimuli or motor responses or continuous raw data. The latter are routinely employed in the analysis of abnormal epileptic activity or in the characterization of ongoing rhythmic activity.

Dipole modeling

In order to estimate the locations of activated brain areas from the measured data a suitable source model is employed. Many primary sensory responses can be adequately accounted for with a dipole model which relies on the assumption that the extent of the activity is sufficiently small to appear as a point source at a typical measurement distance of at least three centimeters. Single time instant or a whole epoch of data is then employed in the estimation of the locations and timecourses of one or more dipole sources. Since MEG does not provide anatomical information, the locations of the sources are displayed in the anatomical MRI data of the subject or patient, which are coregistered with the MEG coordinate frame using fiducial marker locations and the overall shape of the scalp.

Cortically constrained distributed models

Another possibility in the analysis is to assume that the sources have a continuous distribution in the brain or on the cortical mantle, segmented from high-resolution MR images. A suitable constraint is then applied to selected the most probable current distribution of those infinitely many choices in principle available to explain the measured data. The results of such current distribution estimation procedures can be displayed as ‘movies’ which indicate the estimated strength of the activation or a related statistic as a function of time. In the visualization of the current estimates we routinely employ the inflated cortical surface representation thus making the data easily comparable to corresponding fMRI results.

Multimodal imaging

Often the nonuniqueness of the MEG (and EEG) source estimation problem can be alleviated by incorporating information from other imaging modalities as an a priori constraint. The most natural companion to MEG is EEG, which provides additional information about the radial current sources. As indicated above, we are also currently employing anatomical MRI data as a geometrical or location constraint and as a medium for visualization of the results. In addition, we have developed procedures which allow biasing the MEG source estimates towards the fMRI activation, observed in an identical or similar experimental paradigm. Presently, the combination of fMRI, MEG, and EEG data is under further investigation within our Center for Functional Imaging Technologies. It is expected that deeper understanding of the relationship of the hemodynamic and electrophysiological signals will lead to combined models which employ the fMRI, MEG, and EEG jointly to estimate the sites and dynamics of brain activity.

Temporal resolution

Milliseconds. MEG allows real-time recording of the brain activity.

Spatial resolution

From several millimeters to a couple of centimeters, depending on the experiment.

What are some features/benefits of MEG?

MEG is completely noninvasive and non-hazardous.

The data can be collected in the natural seated position allowing more life-like cognitive
experiments than fMRI.

The measurement environment is completely silent, which facilitates especially auditory studies.

MEG has an extremely high temporal resolution (milliseconds) and also provides a good spatial resolution.

Signals can be recorded over the whole cortex.

There is no need to paste electrodes on the scalp as with EEG.

What are its limitations?

A major technical problem associated with MEG is that the localization of sources of electrical activity within the brain from magnetic measurement outside the head is complicated and does not have a unique solution. This is known as the ill-posed inverse problem, and is itself the subject of research. However, as indicated above, feasible solutions can be often obtained by using relatively simple models.

Due to the increased distance to sources and the almost spherical symmetry of the head it is difficult to provide reliable information about subcortical sources of brain activity.

MEG does not provide structural/anatomical information. Therefore MEG data often must be combined with MR data into a composite image of function overlaid on anatomy to produce activation maps.

Because the neuromagnetic signals are very weak compared to the magnetic fields in a normal laboratory environment, the MEG measurements have to be taken in a magnetically shielded room with two or more layers using a sensitive SQUID magnetometer.

Research and clinical applications

Clinically, MEG is used to detect and localize epileptiform spiking activity in patients with epilepsy. It is also used to localize brain areas important for speech, which should be avoided by the surgeon in planning for removal of brain tumors.

At the Martinos Center, researchers use MEG, often in conjunction with EEG, MRI, fMRI, and optical imaging to obtain maps of brain activity in cognitive neuroscience studies carefully designed to investigate the workings of the normal and diseased brain.


by M. Hämäläinen, updated 11/2007

back to top | back to MEG Core laboratory