Jobs
Careers at NINDS
Why Work at NINDS

Training
Training and Career
   Development
Office of Minority
   Health & Research

Student Programs
Temporary Employment

Summer Program

Contact Us
My Privacy

NINDS is part of the
National Institutes of
Health

  Print-friendly version

   Email this page

This summer, I have been working with magnetoencephalography (MEG) as a tool for brain-computer interfacing (BCI). MEG is a non-invasive method of recording the activity of moving electrical charges within the brain using sensors outside the head. These measured magnetic fields can be used to characterize the subject's mental activity with the assistance of computerized pattern recognition methods. The software takes the information from the MEG, uses signal analysis algorithms to identify relevant features of the signal, and statistically analyzes those features to classify the activity into one of several brain states. To construct a BCI, those brain states are in turn used by a subject to control the movement of external devices or images on a screen. By providing the user with sensory feedback, he/she can modulate his/her brain rhythms, e.g. to select letters for spelling a word, to control a cursor on the screen, or, in my experiment, to fly a virtual helicopter through a 2D virtual cave.

Having little or no experience with signal analysis and statistics, I viewed this summer as a chance to acquaint myself with some prominent techniques in the context of BCI. After spending my first few weeks poring over papers and acquainting myself with the recent work in the field, I got the sense that the relative success or failure of a BCI project is hugely dependent on small nuances that seem to get overlooked in the published material. With this in mind, I set out to attempt to prove that it is possible to give a subject one-dimensional control of an object on a screen with limited calibration performed before each trial. The model that I had seen used most frequently involved doing a huge block of calibration, followed by a huge block of training trials. This method does not, in my opinion, reflect the dynamic nature of the brain. As I am writing this summary, I am still not sure how well my method will work, as I am in the beginning stages of collecting subjects and running experiments. I have developed original algorithms and computer code from scratch over the past two months which I hope will result in an operational BCI that will generate robust, easily reproducible results.

Last updated November 16, 2007