Bayesian brain-computer interfacing interpretation of patient intentions from single-trial EEG

Research: Brain-computer interfacing (BCI) allows people to communicate their intent to the outside world without having to rely on any aspect of conventional motor function. It can be used to create a new communication channel for the severely handicapped, or to augment existing means of control for healthy individuals. Most BCI systems are based on measuring electroencephalographic (EEG) activity from the scalp. With signal processing techniques, this activity is then on-line and continually analyzed to infer the users intention. In order to realize successful applications of BCI systems, it is essential to improve their performance (typically measured in the number of error-free bits that can be communicated per minute). In this project, we focus on the enhancement of the signal processing techniques, in particular to face the following challenges. Low signal-to-noise ratio, EEG signals are the result of many sources of activity inside the brain. Nevertheless, most current BCI approaches treat the signals \\as is", without reference to the way they are generated. Explicit modeling of these sources will lead to a much better representation of the background noise and, after subtraction of that noise, to a much better signal-to-noise ratio. Highly non-stationary, Users of a BCI system considerably adapt during an experiment, if only in their attempt to maximize correct feedback. This adaptivity of the user requires the use of signal processing techniques that can track non-stationarities. Also this aspect is ignored in most BCI systems. Our approach will be based on an integrated probabilistic Bayesian framework. That is, we will use probability models to describe the relationships from brain activity to sensor signals over time. Prior knowledge, in particular about the localization of sources inside the brain, is implemented probabilistically as well. Algorithms for classification and tracking follow from the use of Bayes rule, which is known to be the optimal procedure from a decision-theoretic point of view. For example, for optimal classification, Bayes rule inverts the probability of sensor signals given brain activity to the probability of brain activity (representing a persons intentions) given sensor signals. Bayesian updating of the models parameters enables the system to track non-stationarities. Utilization: We expect our methods to considerably improve the performance and hence applicability of brain-computer interfacing in general. Within this project we plan to demonstrate the merits of our approach on two applications in more detail: FES-induced cycling and EEG monitoring of the depth of anesthesia. FES-induced cycling, Patients with paraplegia regain the possibility to cycle by applying Functional Electrical Stimulation (FES) to their lower extremities. A cycling paraplegic strongly depends on the use of his/her hands and arms for critical tasks like steering and propulsion during cycling and body support during standing-up or getting-down. Subtle control actions (e.g., modifying the stimulation for brake and accelerate commands) can then be issued by mind control using a BCI system, which is much more natural and robust. EEG monitoring, In this project we will focus on monitoring the depth of anaesthesia. Patients can be anesthetized to the extent where they are aware of undergoing surgery but are unable to communicate such. Better EEG monitoring methods will help prevent such traumatic experiences. The first application stresses the active use of brain-computer interfaces. It will be pursued in close collaboration with the users BerkelBike, SMK Research, and BraInquiry. Other applications in the medical domain include BCI systems that help locked-in patients to regain their ability to communicate or to control a device. In a non-medical setting, our research can improve imaginary control in, among others, natural interfaces to computers (e.g., for computer games) and mobile telephones, and virtual and augmented reality systems. The monitoring application corresponds to a more passive use of brain-computer interfacing. The results of our research will have direct impact on on-line monitoring in general. An example of this is the on-line monitoring of heart activity for the early detection of cardiac arrhythmias and strokes. Alkmaar Medical Centre (Anesthesiology) and Vitraton (pacemakers) have expressed their keen interest in our research proposal and willingness to cooperate.