Hand movements can be predicted from non-invasive brain signals

'Brain machine interfaces' (BMI) are technologies that allow controlling computers or prostheses with signals recorded from the brain. Scientists hope that in the future, severely paralyzed patients will be able to use this method to control artificial limbs with their thoughts. There are two basic types of BMI-technologies: invasive and non-invasive BMIs. In invasive BMIs, neuronal activity is recorded from electrodes that have to be implanted into the brain. Non-invasive BMIs work with sensors that are externally attached to the skull. Obviously, non-invasive technologieshave the advantage to be much easier to handle  and are virtually risk-free. But  they carry the decicive disadvantage of a much lower spatial resolution. In a joint effort, scientists from Freiburg and Tübingen, however, have succeeded in non-invasively recording a movement control signal from the brain, which, up to that time, had only been achieved by invasive BMIs. The study by the scientists around Carsten Mehring (Bernstein Center for Computational Neuroscience and University Freiburg) is published in the current issue of the 'Journal of Neuroscience'.

Left: Experimental setting. The subject is moving a joystick in one of four directions, while brain activity is recorded by MEG (sensors under the white cap) and EEG (not visible in this figure).

Right: Topographical representation of the information about intended movement directions. The head is shown from above, with the nose pointing upwards. Brain areas from which the movement could be predicted with high precision are highlighted in red, areas with low information content in blue. Information about movement directions of the right hand could primarily be derived from the left motor cortex.

Invasive and non-invasive BMIs differ not only in their technical approach, but also in where the neuronal signals for movement control are recorded. In non-invasive methods, brain activity is measured through the skull, resulting in a blurred picture, like when looking through a frosted glass. Therefore, this technology exploits brain signals that are produced by large groups of neurons. Patients or experimental subjects usually have to learn in an intensive training how to voluntarily induce certain electrical voltage changes that can be used for controlling a computer cursor. Invasive technologies that use implanted electrodes, in contrast, allow recording the activity of single neurons or small neuron groups from within the motor cortex - the brain region that is of primary importance for voluntary movements.

Scientists from Freiburg and Tübingen have now for the first time succeeded in extracting specific movement control signals from the motor cortex using non-invasive methods. Using magneto-encephalography (MEG) or electro-encephalography (EEG), they could tell from brain activity alone in which direction the subject was moving its hand. In EEG, the voltage changes that are induced by the currents flowing through electrically active neurons are measured at the skull surface. MEG records magnetic signals that are induced by these currents. Compared to  previous non-invasive methods, the approach by Mehring's goup has one decicive advantage: with it, controlling a prosthesis or a cursor would be possible in a completely intuitive way, just like in natural hand movements, and would therefore require much less training than previous non-invasive approaches.

In a follow-up study, the scientist are now testing whether healthy subjects can use this approach to control a computer with non-invasive brain signals. They concede that the precision of such a system will probably not quite reach the one of invasive systems. Based on their new results, however, one could make use of the advantages of this direct and natural approach for control of prostheses and cursors without running the high risks tied to sensor implantation.

Original Publication:

Waldert S, Preissl H, Demandt E, Braun C, Birbaumer N, Aertsen A, Mehring C. (2008).

Hand movement direction decoded from MEG and EEG. J Neurosci. 28(4):1000-8.

doi:10.1523/JNEUROSCI.5171-07.2008