BCI - Research

State of the Art in BCI research

At this time, different labs (presented in alphabetical order in the following table) are working on communication channels between the brain and the computer. The work of two leading groups (i) the University of Tübingen and (ii) the Wadsworth Center in Albany, N.Y., is discussed in detail in the following Sections. The BCI of the University of Technology Graz will be discussed separatly.
BCI-Publications gives an overview over important BCI papers.

University Department Authors Year
Healthy Subj.
Pat.
Cl.
EEG-Signal
 FB
Country
University of Illinois Beckman-Institute and Department of Psychology Farwell et. al. 1988
4
1
2
P300
 No USA
  Spencer 1998
10
4
2
P300
 No  USA
University of Michigan Biomedical Engineering Department Huggins et al. 1999
0
15
2
Oscillatory Frequ. Comp.
 No USA
University Rochester Department of Computer Science Bayliss and Ballard 1999
5
0
2
P300
 No USA
University of Technology Graz Institute of Biomedical Engineering Flotzinger et al. 1992
1
0
2
Oscillatory Frequ. Comp.
 No Austria
  Pfurtscheller et al. 1993
1
0
2
Oscillatory Frequ. Comp.
 Yes Austria
  Pregenzer et al. 1994
3
0
2
Oscillatory Frequ. Comp.
 No Austria
  Kalcher et al. 1996
4
0
3
Oscillatory Frequ. Comp.
 Yes Austria
  Pfurtscheller et al. 1996b
3
0
2
Oscillatory Frequ. Comp.
 Yes Austria
  Pregenzer et al. 1996
1
0
2
Oscillatory Frequ. Comp.
 Yes Austria
  Pfurtscheller et al. 1997a
3
0
2
Oscillatory Frequ. Comp.
 Yes Austria
  Ramoser et al. 1997
5
0
2
Oscillatory Frequ. Comp.
 Yes Austria
  Schlögl et al. 1997
3
0
2
Oscillatory Frequ. Comp.
 Yes Austria
  Lugger et al. 1998
3
0
2
Oscillatory Frequ. Comp.
 No Austria
  Pfurtscheller et al. 1998b
4
0
2
Oscillatory Frequ. Comp.
 Yes Austria
  Guger et al.  1999a
3
0
2
Oscillatory Frequ. Comp.
 Yes Austria
  Müller-Gerking et al. 1999
3
0
3
Oscillatory Frequ. Comp.
 No Germany
  Neuper et al. 1999a
4
0
2
Oscillatory Frequ. Comp.
 Yes Austria
  Obermaier et al. 1999a
3
0
2
Oscillatory Frequ. Comp.
 Yes Austria
  Obermaier et al. 1999b
3
0
2
Oscillatory Frequ. Comp.
 Yes Austria
  Pregenzer 1999
3
0
2
Oscillatory Frequ. Comp.
 Yes Austria
  Ramoser 1999
3
0
2
Oscillatory Frequ. Comp.
 No Austria
  Guger et al. 2000
3
0
2
Oscillatory Frequ. Comp.
 Yes Austria
  Guger et al. 2001
3
0
2
Oscillatory Frequ. Comp.
 Yes Austria
University of Tübingen Institute of Medical Psychology and Behavioral Neurobiology Kotchoubey et al. 1997
13
0
3
Slow wave
 Yes Germany
  Kübler et al. 1998
0
2
2
Slow wave
 Yes Germany
  Kübler et al. 1999
0
2
2
Slow wave
 Yes Germany
  Birbaumer 1999a
0
1
2
Slow wave
 Yes Germany
University degli Studi Tor Vergata   Babiloni et al. 1999
5
0
3
Oscillatory Frequ. Comp.
 No Italy
Wadsworth Center Wadsworth Center for Laboratories and Research Wolpaw et al. 1991
5
0
2
Oscillatory Frequ. Comp.
 Yes USA
  McFarland et al. 1993
4
0
4
Oscillatory Frequ. Comp.
 Yes USA
  Wolpaw et al. 1994
5
0
4
Oscillatory Frequ. Comp.
 Yes USA
  Wolpaw et al. 1997
3
1
2
Oscillatory Frequ. Comp.
 Yes USA
  McFarland et al. 1997b
3
1
2
Oscillatory Frequ. Comp.
 Yes USA
  Miner 1998
3
1
2
Oscillatory Frequ. Comp.
 Yes USA
  McFarland et al. 1998
7
3
2
Oscillatory Frequ. Comp.
 Yes USA
  Vaughan et al. 1998
4
1
2
Oscillatory Frequ. Comp.
 Yes USA
  Wolpaw et al. 1998
11
5
4
Oscillatory Frequ. Comp.
 Yes USA
Wright-Patterson Air Force Base Scientific Services and Air Force Research McMillan and Calhoun 1995
3
0
2
VEP
 Yes USA
  Calhoun et al. 1995
0
3
2
VEP
 Yes USA
Up

Wolpaw, Wadsworth Center:

One of the leading BCI research labs in the world is the Wadsworth Center located in Albany. The BCI laboratory in Albany has focused on using the mu (8-12 Hz) and the beta (13-28 Hz) rhythms in the EEG for communication [McFarland 1993, Vaughan 1996]. By using these rhythms generated on the sensorimotor cortex, subjects learned to move a cursor on a computer screen with biofeedback [Wolpaw 1991, 1994]. The subjects use spontaneous EEG activity not tied to a specific evoking stimulus. One-dimensional control was realized with electrodes above the left and right hemisphere. The vertical cursor movement was established by summing up the mu power over both hemispheres. When the sum exceeded a given threshold, the cursor moved upwards or otherwise downwards. The task was either to hit a moving target on the screen or to move the cursor into a highlighted target on the screen. In the first case, the trial ended when the moving target disappeared from the screen and in the second case, when the subject reached the target. A fast Fourier transformation (FFT) algorithm was used to calculate the power of the mu rhythm every 200 ms of EEG derivations on the left and right sensorimotor cortex. These power values were converted into horizontal or vertical cursor movements by linear equations. The coefficients of the linear equations were updated after each trial. The classification accuracy was around 70 % to 80 %. It is important to note, that the movement is paced by the subject himself - the user can decide when to move the cursor.

Today, the maximal entropy method of autoregressive spectral estimation is preferably used for on-line analysis. Trained subjects can reach an accuracy from 70 % up to 95 % and reach the target in 1-2 s [McFarland 1997a, 1997b, 1998, Wolpaw 1998].

Two dimensional control was established by controlling horizontal movement with the sum of left and right mu power and vertical movement by the difference between left and right mu. The target was presented in one corner of the screen. The accuracy was around 60 % and the subjects needed 2-4 seconds to hit the target [Wolpaw 1994]. The ultimate goal is a mouse-like cursor control that allows to operate common mouse-driven programs [Vaughan 1996].

Miner et al. [Miner 1998] demonstrated that a BCI can be used to answer simple questions. Four trained subjects (one with ALS) controlled a vertical cursor on a video screen. The targets were replaced by the words YES and NO. Then the subjects used the cursor to answer spoken questions. The answers were confirmed by response verification. 93 % to 99 % of the questions were correctly answered and 64 % to 87 % of their answers were confirmed by the response verification. The question complexity did not interfere with the accuracy. Finally, the Wadsworth BCI appears capable of operating the "Freehand" neuroprosthesis which provides hand-grasp control to people with spinal cord injuries [Kilgore 1997, Lauer 1999].


Up

Birbaumer, University of Tübingen:
Birbaumer's lab uses slow cortical potentials (SCPs) to present biofeedback to subjects. SCPs are recorded using DC-amplifiers and are therefore often referred to as DC-potentials [Kotchoubey 1997]. The subject learns to evoke this SCPs in certain inter-tone intervals by producing either (i) positive or negative SCP shifts at the vertex or (ii) SCP asymmetry between the right and the left central area.


In the first experiments, a rocket appeared on the screen whose position was determined by the SCP shift at Cz. The subjects were instructed to move the rocket and had to find their individual best strategy. The disadvantage was that the subjects could not control the brain potentials fast enough. Subjects needed at least 10 seconds to perform one action, but for real applications, a more rapid control is necessary. Therefore, the absolute negativity versus positivity measurement was replaced by the measurement of the relative negativity in two consecutive intervals. This has 2 advantages: it is not required to produce relative positivity (more difficult than negativity [Birbaumer 1981]) and a faster timing paradigm could be introduced [Kotchoubey 1997].

Birbaumer's group investigated patients who were almost totally paralyzed. The patients were trained for 4-6 weeks and learned to control their slow cortical potentials at the vertex. This shows that the control of SCPs does not require feedback loops from the periphery [Kübler 1998]. After that a Thought Translation Device (TTD) was set up in order to enable subjects to select letters on a computer screen and to write words and sentences [Birbaumer 1992, Kisil 1992, Kübler 1999]. At this time the TTD is permanently used from 3 ALS patients in Germany who were trained over a period of some months in more than 100 sessions to control the system. The patients have to produce either cortical negative or positive differences within 2 seconds in reference to a 2 second baseline interval. The difference between active and baseline interval is transformed into cursor movement on a computer monitor and allows to operate the TTD. Two of the three ALS patients achieved a classification accuracy between 70 % and 80 %. The communication rate is approximately 1 letter every 2 minutes [Birbaumer 1999a]:

Up