Imaging Brain Activity in Real Time

Research participant with EEG

BY LEONARDO FERNANDINO
For over two decades functional magnetic resonance imaging (fMRI) has been the indisputable workhorse in human brain mapping. Its ability to localize brain activity with high spatial resolution (millimeters), coupled with its non-invasiveness, make it an excellent tool for mapping behavioral and cognitive phenomena onto detailed brain anatomy. However, since fMRI relies on changes in blood flow, volume and oxygen concentration as indicators of neural activity, millisecond-scale fluctuations in neuronal activity are not reflected in the signal. The temporal resolution of fMRI (typically > 500 ms) is therefore too coarse to track neural activity in real time and discriminate rapidly succeeding neural events.

For example, some fMRI studies indicate that auditory cortical activations in response to signed language are stronger in deaf participants than in hearing individuals, raising the possibility that the auditory cortex is rewired to process visual information in deaf individuals. An alternative interpretation, however, is that visual processing of sign language occurs entirely in the visual cortex for both groups, and that activations in the auditory cortex reflect only a later, conceptual processing stage. The speed with which these processes occur makes it challenging to determine the precise sequence in which different brain regions become active using fMRI (or any other method that relies on metabolic responses, such as fNIRS or PET).

Fortunately, other non-invasive techniques can provide such precise temporal information. Electroencephalography (EEG) and magnetoencephalography (MEG) directly track the electrical activity of brain cells by measuring its effects on the electrical and magnetic fields just outside the head. They allow researchers to record brain activity at a very high temporal resolution (milliseconds), close to the actual timescale of neural computations.

EEG measures changes in electrical potential generated by the brain, which reflect the bulk electrical activity of many pyramidal neurons depolarizing in synchrony, through electrode leads placed on the scalp and connected to an amplifier. Researchers can then estimate the locations of the neural sources of these signals. The accuracy of these estimates depends, among other things, on the number of electrodes used, with higher numbers resulting in more precise estimates. Typically, between 64 and 256 electrodes are used.

---

MEG, as the name indicates, measures the magnetic counterpart of that same electrical activity. As demonstrated by Hans Christian Oersted in 1820, an electrical current always generates a magnetic field in the space around it. Thus, the electrical activity of the brain leads to changes in the magnetic field measured just outside the head. The magnetic field of the brain is extremely weak, around 10 femtoteslas (10-14 tesla). For comparison, the Earth’s magnetic field – barely strong enough to move a compass needle – is a billion times stronger. Changes in the brain’s magnetic field are measured through sensors named SQUIDs (superconducting quantum interference devices), which must operate at extremely low temperatures, close to absolute zero. A modern MEG scanner contains typically between 150 and 300 sensors, arranged as a helmet around the head and immersed in liquid helium.

Synchronized depolarization along the apical dendrites of cortical pyramidal neurons results in a net electrical current perpendicular to the cortical surface (right). The current generates electrical (yellow) and magnetic (green) fiselds that spread through the head. Electrical changes can be detected by EEG electrodes placed on the scalp (left). Image: Hari, R., & Parkkonen, L. (2015). The brain timewise: how timing shapes and supports brain function. Philosophical Transactions of the Royal Society B: Biological Sciences, 370(1668). Reproduced with permission.

Reconstructed source of neural activity from MEG at 107 ms after auditory stimulation. The array of SQUIDs positioned around the head measures the distribution of the magnetic field (left), which is used to estimate the location of the neural source (right). Image credit: MNE Software 

In both EEG and MEG, the locations of activated brain regions must be estimated from the distribution of the signal outside the head. This so-called “inverse problem” is mathematically ill-posed in the sense that there is, in principle, an infinite number of source configurations that can generate the observed signals. Therefore additional assumptions (e.g., diffuseness of sources) are necessary to estimate a unique solution. A mathematical model of the head is used to generate a “forward model” of how electrical activity at each point affects the signal distribution outside the head. EEG researchers tend to use a generic head model, while MEG researchers usually rely on a detailed model of the participant’s cortex generated from a T1-weighted MRI. 

Research participant undergoing MEG

Nevertheless, MRI-assisted EEG source localization is becoming more common. In both cases, the inverse solution is typically computed through an iterative algorithm that searches for the combination of cortical sources whose forward-modeled signal best matches the observed signal.

The main advantage of MEG over EEG comes from the fact that, while electrical signals are blurred and distorted by the skull, magnetic fields can traverse it virtually unobstructed. Thus, the MEG signal has higher spatial resolution, which allows better estimation of its neural sources.

MEG’s main disadvantage is the cost: the scanner and magnetically shielded room run in the millions of dollars, while a high-density EEG system costs under US$150,000 (although recent advances in quantum sensing might herald the coming of low cost, room temperature MEG). Even leaving these financial considerations aside, MEG still cannot completely replace EEG (nor vice versa), since the two techniques have complementary strengths. Because a magnetic field is oriented perpendicularly to its generating current, MEG is virtually blind to activations outside the sulcal walls, while EEG is maximally sensitive to activations in the gyral crests (and somewhat sensitive to activations in the sulci). Thus, simultaneous recording of EEG and MEG can add valuable information to the source estimation procedure.

Both EEG and MEG have been used clinically in the detection and localization of seizure activity in epileptic patients, which can be crucial for diagnosis and surgical planning. In research, they have been used to study a wide variety of problems, from spatial attention and synesthesia to sentence comprehension. Researchers at the University of California San Diego, for example, used them to answer the question posed earlier, of whether the auditory cortex in deaf individuals is indeed rewired for visual processing. By recording the MEG responses to signed language in deaf and hearing participants, they showed that, in both groups, activity was confined to the visual cortex in the early perceptual stage (~100 ms), and only later, during the lexico-semantic stage (~300 ms), the auditory cortex became active. This result provides compelling evidence against the auditory rewiring hypothesis.
​
Like any technique, MEG and EEG have unique strengths and limitations. The advantage afforded by high temporal resolution is partly offset by the relatively low spatial resolution of the estimated source locations. Nevertheless, researchers who manage to incorporate these approaches into their toolkit should expect a rewarding boost in their ability to probe the living human brain.

---