Eberhart E. Fetz - Recurrent brain-computer interfaces

The third of last Thursday's talks was at the UW bioengineering department. Eberhart E. Fetz of the Washington National Primate Research Center presented current work on implantable brain-computer interfaces.

The main application of a potential brain-computer interface (BCI) is to allow profoundly paralysed people to control prostheses by thought. The ideal BCI-controlled prosthesis would feel as natural as using one's own arms and legs, but the technology is a long way from that point. No matter how good the technology gets, it seems likely that any BCI would require some period of training for the user, due to the high degree of variability between brains.

Fetz described two impressive sets of experiments. The first involved giving monkeys brain implants, and then demonstrating how fast the monkey could learn to control a simple up-or-down output, provided that the monkey received direct feedback. The monkey could see a VU-type meter which was directly affected by the BCI, in various ways. When the needle reached a certain, clearly-marked target point, the monkey would get a reward (some applesauce, if I recall correctly) and the meter would reset. There were two experimental conditions: input to the BCI would either increase or decrease the needle level, so the monkey would have to either increase or suppress the activity of the monitored brain region to get rewards. In this setup, the monkeys were able to learn to control the meter within minutes, and they could adapt impressively fast to switches between the two experimental conditions.

This is encouraging, but to really be very useful a BCI would have to be able to provide many more degrees of freedom than just 'up' and 'down'. To this end, the second set of experiments involved giving human subjects a BCI (I think it was a non-invasive EEG-type device, but I'm not sure) and asking them to move a cursor on a computer display, following the movements of a technician. Fetz showed us records from experiments, in which people were able to follow the lead approximately, but accuracy was very poor. There's clearly some way to go before an implant like this could be used to interact with a computer, let alone control prosthetic limbs.

One of the main restrictions of the experiments was that subjects only had a small number of limited-time training sessions with the BCI, so Fetz mentioned ongoing work in monkeys to develop chronically implantable BCIs. If such a thing can be installed without triggering an immune response (the main current restriction on their use) then there is some hope that the effectively endless training would improve the accuracy that can be achieved. After all, a non-paralysed person's brain has effectively had a whole lifetime of training to control its body.


Trackback URL for this entry is:


Post a comment