http://teddysratlab.blogspot.com [Full link to blog for email clients.]
For the final installment in this series, we will look at brain to machine or brain to computer interfaces that do not require a direct interface with neurons. whereas interfaces that involve electrodes implanted into brain are typically termed "brain machine interface", interfaces that operate "noninvasively" are quite frequently called "brain computer interfaces". Thus when looking at the literature, it can be a bit confusing to see reference to BMI versus BCI, without realizing that they are referring to essentially the same thing.
BCI's typically involve recording electroencephalograph information from the scalp, and using special amplifiers and processors to identify specific frequency components, and use that identification to drive a computer interface. The simplest such connection, is somewhat reminiscent of the "biofeedback" devices popular in the 70s. A biofeedback monitor is simply a single channel EEG amplifier connected to a filter for the "Alpha" frequency waveform of EEG. Using a technique called fast Fourier transformation, the relative power or quantity of waveforms that fall in the alpha frequency band between eight and 12 Hz is represented with a simple analog meter. The alpha frequency is associated with meditation, relaxation, and quiet concentration. The commercially available device from NeuroSky (http://www.neurosky.com/) was originally designed to measure increases and decreases in alpha frequency, and provide a simple on-off switch for computer interface. Their current devices measure five different frequency ranges, and can produce differential outputs to a computer based on the ratio of EEG frequency. Like many brain machine interfaces, and EEG frequency-based BCI requires the user to train themselves to alter their EEG frequencies. Thus the analogy to biofeedback training, since users learn how to produce active and quiet PEG through a process very similar to meditation.
More sophisticated BCI's are being developed, that use more than just the frequency of EEG to drive a computer interface. Increasing sophistication of a device such as NeuroSky, and the BCI2000 system pioneered by Gerwin Schalk (http://www.bci2000.org/BCI2000/Home.html), have begun to take multiple channels of EEG and look for much finer detail which represents movement intention, and focused attention. Aside from the (unintentional) alliterative nature of the previous sentence, the two processes are very similar. Research in the 1980s and 90s demonstrated that prior to an actual limb movement, areas of the brain involved in planning of movement became active several seconds before the movement. By mapping this activity, researchers could correlate the neural activity with the actual muscle activation, and control a robotic arm by "thought" alone. Such a finding is very important for developing a prosthetic to replace an amputated limb, since prosthetic movement needs to be controlled by the "intention" to move. On the other hand researchers also found that by focusing one's attention on the type of movements that they wish to make, they can also correlate brain activity with movement.
Difficult as it was to perform these analyses with electrodes implanted directly into the brain, it is even more difficult to select the information-containing components from EEG recorded from outside the skull. will the more distant and electrode is located from the neural activity that is to record, the weaker the signal, and the more noise or unrelated information can be recorded at the same time. EEG is a form of recording known as "volume conduction" in which the sum of neural signals from a large volume combine onto a given electrode. The intervening skull and scalp also contribute to attenuation of the signal from specific brain areas. The best way to correct for these effects is to use multichannel EEG in which a dozen or more electrodes are placed onto the scalp, and signals can be localized by comparison between pairs of electrodes with different spatial orientation with respect to the brain area being recorded. While this enables a finer detail and recording, it still does not compare to the detail that can be obtained with implanted electrodes. Neuroscience as a field is quite familiar with this problem. It is typically referred to as "the inverse problem" in that it requires reverse engineering the inputs (neural activity in specific brain areas) from the EEG output. It is a computationally intensive effort that has not yet been solved. However in the interim, devices such as that produced by NeuroSky and analysis platforms such as BCI 2000 to allow for derivation of information from the EEG Tech and be used to drive a computer or device.
Brain computer interfaces of all types have application not only in bionics and prosthetics, but in providing communication and device control for quadriplegics and those suffering from debilitating neural diseases such as ALS. They are an important component not only of neuroscience but of rehabilitative medicine, and have the potential to teach us much more about the brain, than we have already learned in the process of developing these devices. Thus, the current state of the art in bionics, prosthetics, and brain machine interfaces is still fairly crude. This is not to say that it is not effective, or that it does not provide sufficient information to drive a prosthetic or an interface. It is simply that we are still a long way from the "Six Million Dollar Man" ideal of an interface directly from machines to the brain and from the brain to machines.
This ends our series on bionics, prosthetics, and brain machine interfaces. I will take a brief break from the Lab Rats Guide to the Brain with some current events news articles and a report from the StellarCon science-fiction convention. Please note that I'm still having to dictate blog entries, which may result in some delays in the near future.
Until next time, take care of your brain: we can't quite rebuild it, we don't yet have the technology.