How many people are able to operate an EEG-based brain-computer interface (BCI)?

C. Guger, Günter Edlinger, W. Harkam, I. Niedermayer, G. Pfurtscheller

IEEE Transactions on Neural Systems and Rehabilitation Engineering · 2003 · 526 citations · 10 references

Concepts

TL;DR

A field study with 99 healthy participants at an exposition in Graz involved two 20–30‑minute BCI sessions: an initial 40‑trial no‑feedback block followed by a 40‑trial feedback block where subjects imagined right‑hand or foot movements to control a horizontal bar using bipolar electrodes over the corresponding motor areas. After two sessions, about 93 % of participants achieved classification accuracy above 60 %, with adaptive autoregressive models applied to 39 subjects and band‑power estimation to 60 subjects.

Abstract

Ninety-nine healthy people participated in a brain-computer interface (BCI) field study conducted at an exposition held in Graz, Austria. Each subject spent 20-30 min on a two-session BCI investigation. The first session consisted of 40 trials conducted without feedback. Then, a subject-specific classifier was set up to provide the subject with feedback, and the second session - 40 trials in which the subject had to control a horizontal bar on a computer screen - was conducted. Subjects were instructed to imagine a right-hand movement or a foot movement after a cue stimulus depending on the direction of an arrow. Bipolar electrodes were mounted over the right-hand representation area and over the foot representation area. Classification results achieved with 1) an adaptive autoregressive model (39 subjects) and 2) band power estimation (60 subjects) are presented. Roughly 93% of the subjects were able to achieve classification accuracy above 60% after two sessions of training.

References

10