The non-invasive Berlin Brain–Computer Interface: Fast acquisition of effective performance in untrained subjects
Introduction
The aim of Brain–Computer Interface (BCI) research is to establish a novel communication system that translates human intentions–reflected by suitable brain signals–into a control signal for an output device such as a computer application or a neuroprosthesis. According to the definition put forth at the first international meeting for BCI technology in 1999, a BCI “must not depend on the brain's normal output pathways of peripheral nerves and muscles” (Wolpaw et al., 2000).
There is a huge variety of BCI systems (see Pfurtscheller et al., 2005, Wolpaw et al., 2002, Kübler et al., 2001, Dornhege et al., 2007b, Curran and Stokes, 2003). BCI systems relying on intentional modulations of evoked potentials can typically achieve higher information transfer rates (ITRs) than systems working on unstimulated brain signals (cf. Cheng et al., 2002, Kaper and Ritter, 2004). On the other hand, with evoked potential BCIs the user is constantly confronted with stimuli, which could become exhaustive after longer usage. Furthermore, some patient groups might not be able to properly focus their gaze and thus such a system will not be a reliable means for their communication when visual evoked potentials are employed.
One of the major challenges in BCI research is the huge inter-subject variability with respect to spatial patterns and spectrotemporal characteristics of brain signals. In the operant conditioning variant of BCI, the subject has to learn the self-control of a specific EEG feature which is hard-wired in the BCI system (see e.g., Elbert et al., 1980, Rockstroh et al., 1984, Birbaumer et al., 2000). An alternative approach tries to establish BCI control in the opposite way: while using much more general features, the system automatically adapts to the specific brain signals of each user by employing advanced techniques of machine learning and signal processing (e.g., Müller et al., 2001, Haykin, 1995; and more specifically with respect to BCI: Blankertz et al., 2004, Blankertz et al., 2006c, Blankertz et al., 2006d, Müller et al., 2003, Müller et al., 2004).
The Graz BCI group introduced the common spatial pattern (CSP) algorithm (spatial filters that are optimized for the discrimination of different condition, cf. Common spatial pattern (CSP) analysis) for the use in BCI systems (Ramoser et al., 2000) and reported in (Guger et al., 2000a) results from a feedback study with a CSP-based BCI operating on a 27 channel EEG. The feedback study encompassed 6 sessions on 4 days for each of three subjects that were experienced with BCI control. Nevertheless the result for two out of three subjects was at chance level in the first feedback session and reasonable BCI control was only obtained from the 2nd feedback session on. The feedback application did not allow to explore what information transfer rates could be obtained because it relied on a synchronous design where each binary decision needed 8 s, limiting the highest possible ITR to 7.5 bits per minute (bpm) at a theoretical accuracy of 100%. In a more recent publication (Krausz et al., 2003) 4 patients with complete or partial paralysis or paresis of their lower limbs were trained to operate a variant of the Graz BCI that uses band power features of only 2 bipolar channels. As feedback application, a basket game was used in which the subject controls the horizontal position of a ball that falls downward at constant speed. The aim in this application is to hit one of two basket targets at the bottom of the screen. On the second and third day the maximum ITR of 6–16 runs of 40 trials each for the 4 subjects was between 3 and 17.2 bpm (mean 9.5 ± 5.9).
A study from the Wadsworth BCI group (McFarland et al., 2003) investigates the influence of trial duration and number of targets on the ITR in their BCI system that uses operant conditioning for letting the users learn to modulate the amplitude of sensorimotor rhythms. Eight subjects (2 patients, one spinal injury at c6 and one cerebral palsy) trained over several months to operate a BCI application similar to the basket game described above, but with vertical cursor control and a variable number of target fields. The average ITR from 8 runs of 20 to 30 trials for the 8 subjects was between 1.8 and 17 bpm (mean 8.5 ± 4.7) at the individual best number of targets. In a more recent study in cooperation with the BCI group in Tübingen (Kübler et al., 2005) a similar methodology was successfully used with 4 patients suffering from Amyotrophic Lateral Sclerosis (ALS). This was the first study demonstrating that ALS patients are capable of voluntarily modulating the amplitude of their sensorimotor rhythms to control a BCI.
Based on offline results, del Millán et al. (2002) suggest to use a local neural classifier based on quadratic discriminant analysis for the machine learning part. Using this system asynchronously in an online feedback with three classes (left/right-hand motor imagery and relax with eyes closed) three subjects were able after a few days of training to achieve an average correct recognition of about 75% whereas the wrong decision rates were below 5%. In del Millán and Mouriño (2003) it was reported that with this system a motorized wheelchair and a virtual keyboard could be controlled. In the latter case trained subjects were able to select a letter every 22 s. In a preliminary study the best subject was reported to be able to do selections every 7 s. Note that brain signals for one class were produced by closing the eyes.
Here we demonstrate how an effective and fast BCI performance can be realized even for untrained subjects by use of modern machine learning techniques (cf. Algorithms and procedures and Methodological and technical details).
Section snippets
Neurophysiology and features
According to the ‘homunculus’ model, as described by Jasper and Penfield (1949), for each part of the human body there exists a corresponding region in the primary motor and primary somatosensory area of the neocortex. The ‘mapping’ from the body part to the respective brain areas approximately preserves topography, i.e., neighboring parts of the body are represented in neighboring parts of the cortex. For example, while the feet are located close to the vertex, the left hand is represented
Results
For three subjects the combination left vs. right was found optimal, for four subjects left vs. foot and for the remaining two subjects right vs. foot (the criterium for selecting a binary pair of tasks was the discriminability of the corresponding classes of brain signals which have been acquired in the calibration measurement, see Feedback sessions and Validation). For one subject no sufficient separation was achieved (see Investigation the failure).
Discussion
The Berlin Brain–Computer Interface makes use of a machine learning approach towards BCI. Working with high dimensional, complex features obtained from 128 channel EEG allows the system a distinct flexibility for adapting to the individual characteristics of each user's brain. The result from a feedback study with 10 subjects demonstrates that the BBCI system (1) robustly transfers the discrimination of mental states from the calibration to the feedback sessions, (2) allows a very fast
Acknowledgments
We would like to express our thanks to the anonymous reviewers who gave valuable comments, criticism and suggestions for the revision of the first draft.
This work was supported in part by grants of the Bundesministerium für Bildung und Forschung (BMBF), FKZ 01IBE01A/B, by the Deutsche Forschungsgemeinschaft (DFG), FOR 375/B1, and by the IST Programme of the European Community, under the PASCAL Network of Excellence, IST-2002-506778. This publication only reflects the authors' views.
References (55)
- et al.
Learning to control brain activity: a review of the production and control of EEG components for driving brain–computer interface (BCI) systems
Brain Cogn.
(2003) - et al.
Biofeedback of slow cortical potentials. I
Electroencephalogr. Clin. Neurophysiol.
(1980) - et al.
EEG source localization: implementing the spatio-temporal decomposition approach
Electroencephalogr. Clin. Neurophysiol.
(1998) - et al.
Brain–computer interface (BCI) operation: optimizing information transfer rates
Biol. Psychol.
(2003) - et al.
Event-related dynamics of cortical rhythms: frequency-specific features and functional correlates
Int. J. Psychophysiol.
(2001) - et al.
Imagery of motor actions: differential effects of kinesthetic and visual-motor mode of imagery in single-trial EEG
Brain Res. Cogn. Brain Res.
(2005) - et al.
Event-related EEG/MEG synchronization and desynchronization: basic principles
Clin. Neurophysiol.
(1999) - et al.
Visually guided motor imagery activates sensorimotor areas in humans
Neurosci. Lett.
(1999) - et al.
Mu rhythm (de)synchronization and EEG single-trial classification of different motor imagery tasks
NeuroImage
(2006) - et al.
EEG-based communication: analysis of concurrent EMG activity
Electroencephalogr. Clin. Neurophysiol.
(1998)
Brain–computer interfaces for communication and control
Clin. Neurophysiol.
Über das Elektroenkephalogramm des Menschen
Arch. Psychiatr. Nervenkrankh.
The though translation device (TTD) for completely paralyzed patients
IEEE Trans. Rehabil. Eng.
Boosting bit rates and error detection for the classification of fast-paced motor commands based on single-trial EEG analysis
IEEE Trans. Neural Syst. Rehabil. Eng.
The BCI competition 2003: progress and perspectives in detection and discrimination of EEG single trials
IEEE Trans. Biomed. Eng.
The Berlin Brain–Computer Interface: EEG-based communication without subject training
IEEE Trans. Neural Syst. Rehabil. Eng.
The Berlin Brain–Computer Interface presents the novel mental typewriter Hex-o-Spell
The Berlin Brain–Computer Interface: machine learning based detection of user specific brain states
J. Univer. Comput. Sci.
The BCI competition III: validating alternative approaches to actual BCI problems
IEEE Trans. Neural Syst. Rehabil. Eng.
Design and implementation of a brain–computer interface with high transfer rates
IEEE Trans. Biomed. Eng.
Organization of thalamic and cortical alpha rhythm: spectra and coherences
Electroencephalogr. Clin. Neurophysiol.
Asynchronous bci and local neural classifiers: an overview of the adaptive brain interface project
IEEE Trans. Neural Syst. Rehabil. Eng.
A local neural classifier for the recognition of EEG patterns associated to mental tasks
IEEE Trans. Neural Netw.
Boosting bit rates in non-invasive EEG single-trial classifications by feature combination and multi-class paradigms
IEEE Trans. Biomed. Eng.
Increase information transfer rates in BCI by CSP extension to multi-class
Combined optimization of spatial and temporal filters for improving brain–computer interfacing
IEEE Trans. Biomed. Eng.
Improving human performance in a real operating environment through real-time mental workload detection
Cited by (762)
Improvement of motor imagery electroencephalogram decoding by iterative weighted Sparse-Group Lasso
2024, Expert Systems with ApplicationsManifold embedded instance selection to suppress negative transfer in motor imagery-based brain–computer interface
2024, Biomedical Signal Processing and ControlSpectral representation of EEG data using learned graphs with application to motor imagery decoding
2024, Biomedical Signal Processing and ControlHilbert–Huang Transformation-based subject-specific time–frequency-space pattern optimization for motor imagery electroencephalogram classification
2023, Measurement: Journal of the International Measurement ConfederationQuantification of event related brain patterns for the motor imagery tasks using inter-trial variance technique
2023, Engineering Applications of Artificial IntelligenceMBGA-Net: A multi-branch graph adaptive network for individualized motor imagery EEG classification
2023, Computer Methods and Programs in Biomedicine