Martin Bogdan

Support vector channel selection in BCI

Designing a brain computer interface (BCI) system one can choose from a variety of features that may be useful for classifying brain activity during a mental task. For the special case of classifying electroencephalogram (EEG) signals we propose the usage of the state of the art feature selection algorithms Recursive Feature Elimination and Zero-Norm Optimization which are based on the training of support vector machines (SVM) . These algorithms can provide more accurate solutions than standard filter methods for feature selection . We adapt the methods for the purpose of selecting EEG channels. For a motor imagery paradigm we show that the number of used channels can be reduced significantly without increasing the classification error. The resulting best channels agree well with the expected underlying cortical activity patterns during the mental tasks. Furthermore we show how time dependent task specific information can be visualized.

Neural mechanisms of brain-computer interface control

Brain-computer interfaces (BCIs) enable people with paralysis to communicate with their environment. Motor imagery can be used to generate distinct patterns of cortical activation in the electroencephalogram (EEG) and thus control a BCI. To elucidate the cortical correlates of BCI control, users of a sensory motor rhythm (SMR)-BCI were classified according to their BCI control performance. In a second session these participants performed a motor imagery, motor observation and motor execution task in a functional magnetic resonance imaging (fMRI) scanner. Group difference analysis between high and low aptitude BCI users revealed significantly higher activation of the supplementary motor areas (SMA) for the motor imagery and the motor observation tasks in high aptitude users. Low aptitude users showed no activation when observing movement. The number of activated voxels during motor observation was significantly correlated with accuracy in the EEG-BCI task (r=0.53). Furthermore, the number of activated voxels in the right middle frontal gyrus, an area responsible for processing of movement observation, correlated (r=0.72) with BCI-performance. This strong correlation highlights the importance of these areas for task monitoring and working memory as task goals have to be activated throughout the BCI session. The ability to regulate behavior and the brain through learning mechanisms involving imagery such as required to control a BCI constitutes the consequence of ideo-motor co-activation of motor brain systems during observation of movements. The results demonstrate that acquisition of a sensorimotor program reflected in SMR-BCI-control is tightly related to the recall of such sensorimotor programs during observation of movements and unrelated to the actual execution of these movement sequences.

Robust EEG channel selection across subjects for brain-computer interfaces

Most EEG-based brain-computer interface (BCI) paradigms come along with specific electrode positions, for example, for a visual-based BCI, electrode positions close to the primary visual cortex are used. For new BCI paradigms it is usually not known where task relevant activity can be measured from the scalp. For individual subjects, Lal et al. in 2004 showed that recording positions can be found without the use of prior knowledge about the paradigm used. However it remains unclear to what extent their method of recursive channel elimination (RCE) can be generalized across subjects. In this paper we transfer channel rankings from a group of subjects to a new subject. For motor imagery tasks the results are promising, although cross-subject channel selection does not quite achieve the performance of channel selection on data of single subjects. Although the RCE method was not provided with prior knowledge about the mental task, channels that are well known to be important (from a physiological point of view) were consistently selected whereas task-irrelevant channels were reliably disregarded.

EEG alpha spindle measures as indicators of driver fatigue under real traffic conditions

OBJECTIVE The purpose of this study is to show the effectiveness of EEG alpha spindles, defined by short narrowband bursts in the alpha band, as an objective measure for assessing driver fatigue under real driving conditions. METHODS An algorithm for the identification of alpha spindles is described. The performance of the algorithm is tested based on simulated data. The method is applied to real data recorded under real traffic conditions and compared with the performance of traditional EEG fatigue measures, i.e. alpha-band power. As a highly valid fatigue reference, the last 20 min of driving from participants who aborted the drive due to heavy fatigue were used in contrast to the initial 20 min of driving. RESULTS Statistical analysis revealed significant increases from the first to the last driving section of several alpha spindle parameters and among all traditional EEG frequency bands, only of alpha-band power; with larger effect sizes for the alpha spindle based measures. An increased level of fatigue over the same time periods for drop-outs, as compared to participants who did not abort the drive, was observed only by means of alpha spindle parameters. CONCLUSIONS EEG alpha spindle parameters increase both fatigue detection sensitivity and specificity as compared to EEG alpha-band power. SIGNIFICANCE It is demonstrated that alpha spindles are superior to EEG band power measures for assessing driver fatigue under real traffic conditions.

Coupling brain-machine interfaces with cortical stimulation for brain-state dependent stimulation: enhancing motor cortex excitability for neurorehabilitation.

Motor recovery after stroke is an unsolved challenge despite intensive rehabilitation training programs. Brain stimulation techniques have been explored in addition to traditional rehabilitation training to increase the excitability of the stimulated motor cortex. This modulation of cortical excitability augments the response to afferent input during motor exercises, thereby enhancing skilled motor learning by long-term potentiation-like plasticity. Recent approaches examined brain stimulation applied concurrently with voluntary movements to induce more specific use-dependent neural plasticity during motor training for neurorehabilitation. Unfortunately, such approaches are not applicable for the many severely affected stroke patients lacking residual hand function. These patients require novel activity-dependent stimulation paradigms based on intrinsic brain activity. Here, we report on such brain state-dependent stimulation (BSDS) combined with haptic feedback provided by a robotic hand orthosis. Transcranial magnetic stimulation (TMS) of the motor cortex and haptic feedback to the hand were controlled by sensorimotor desynchronization during motor-imagery and applied within a brain-machine interface (BMI) environment in one healthy subject and one patient with severe hand paresis in the chronic phase after stroke. BSDS significantly increased the excitability of the stimulated motor cortex in both healthy and post-stroke conditions, an effect not observed in non-BSDS protocols. This feasibility study suggests that closing the loop between intrinsic brain state, cortical stimulation and haptic feedback provides a novel neurorehabilitation strategy for stroke patients lacking residual hand function, a proposal that warrants further investigation in a larger cohort of stroke patients.

Online adaptation of a c-VEP Brain-computer Interface(BCI) based on error-related potentials and unsupervised learning.

The goal of a Brain-Computer Interface (BCI) is to control a computer by pure brain activity. Recently, BCIs based on code-modulated visual evoked potentials (c-VEPs) have shown great potential to establish high-performance communication. In this paper we present a c-VEP BCI that uses online adaptation of the classifier to reduce calibration time and increase performance. We compare two different approaches for online adaptation of the system: an unsupervised method and a method that uses the detection of error-related potentials. Both approaches were tested in an online study, in which an average accuracy of 96% was achieved with adaptation based on error-related potentials. This accuracy corresponds to an average information transfer rate of 144 bit/min, which is the highest bitrate reported so far for a non-invasive BCI. In a free-spelling mode, the subjects were able to write with an average of 21.3 error-free letters per minute, which shows the feasibility of the BCI system in a normal-use scenario. In addition we show that a calibration of the BCI system solely based on the detection of error-related potentials is possible, without knowing the true class labels.

Automatic Cluster Detection in Kohonen's SOM

Kohonens self-organizing map (SOM) is a popular neural network architecture for solving problems in the field of explorative data analysis, clustering, and data visualization. One of the major drawbacks of the SOM algorithm is the difficulty for nonexpert users to interpret the information contained in a trained SOM. In this paper, this problem is addressed by introducing an enhanced version of the Clusot algorithm. This algorithm consists of two main steps: 1) the computation of the Clusot surface utilizing the information contained in a trained SOM and 2) the automatic detection of clusters in this surface. In the Clusot surface, clusters present in the underlying SOM are indicated by the local maxima of the surface. For SOMs with 2-D topology, the Clusot surface can, therefore, be considered as a convenient visualization technique. Yet, the presented approach is not restricted to a certain type of 2-D SOM topology and it is also applicable for SOMs having an n-dimensional grid topology.

Online artifact removal for brain-computer interfaces using support vector machines and blind source separation

We propose a combination of blind source separation (BSS) and independent component analysis (ICA) (signal decomposition into artifacts and nonartifacts) with support vector machines (SVMs) (automatic classification) that are designed for online usage. In order to select a suitable BSS/ICA method, three ICA algorithms (JADE, Infomax, and FastICA) and one BSS algorithm (AMUSE) are evaluated to determine their ability to isolate electromyographic (EMG) and electrooculographic (EOG) artifacts into individual components. An implementation of the selected BSS/ICA method with SVMs trained to classify EMG and EOG artifacts, which enables the usage of the method as a filter in measurements with online feedback, is described. This filter is evaluated on three BCI datasets as a proof-of-concept of the method.

Nessi: an EEG-controlled web browser for severely paralyzed patients

We have previously demonstrated that an EEG-controlled web browser based on self-regulation of slow cortical potentials (SCPs) enables severely paralyzed patients to browse the internet independently of any voluntary muscle control. However, this system had several shortcomings, among them that patients could only browse within a limited number of web pages and had to select links from an alphabetical list, causing problems if the link names were identical or if they were unknown to the user (as in graphical links). Here we describe a new EEG-controlled web browser, called Nessi, which overcomes these shortcomings. In Nessi, the open source browser, Mozilla, was extended by graphical in-place markers, whereby different brain responses correspond to different frame colors placed around selectable items, enabling the user to select any link on a web page. Besides links, other interactive elements are accessible to the user, such as e-mail and virtual keyboards, opening up a wide range of hypertext-based applications.

Automated EEG feature selection for brain computer interfaces

A brain computer interface (BCI) utilizes signals derived from electroencephalography (EEG) to establish a connection between a persons state of mind and a computer based signal processing system that interprets the EEG signals. The choice of suitable features of the available EEG signals is crucial for good BCI communication. The optimal set of features is strongly dependent on the subjects and on the used experimental paradigm. Based upon EEG data of an existing BCI system, we present a wrapper method for the automated selection of features. The proposed method combines a genetic algorithm (GA) for the selection of feature with a support vector machine (SVM) for their evaluation. Applying this GA-SVM method to data of several subjects and two different experimental paradigms, we show that our approach leads to enhanced or even optimal classification accuracy.