ICBME2008 Abstract— in this study we have employed most effective feature-based and model-based methods to extract motor activity changes characterized by EDR/ERS patterns in Electroencephalogram signal. We exploited for the first time Kat'z fractal dimension to identify motor related changes with application in asynchronous BCI systems. Methods applied in this study include Fractal Dimension, Band Power, m-spacing estimate of Entropy, and Quadratic Model-based detector according to a previous work in [2]. Although we assess our methods on an idle versus execution of foot movement data set, our long term goal is to employ them on imagery movement data as well. Evaluations of our early stage experiments reveals true positive rates of 76.7%-96.7% with their false positive rates of 6.45%-3.23%, respectively. Keywords- Asynchronous BCI, Asynchronous Brain-Controlled Switch, ERD/ERS Patterns, Fractal Dimension I. INTRODUCTION RAIN-COMPUTER INTERFACE (BCI) is a system in which user's intentions are conducted towards an external device or neural prosthesis or even is used to control Functional Electrical Stimulation (FES), not requiring any physical execution. BCI systems, based on their mode of operation, are divided into two major classes, i.e. synchronous and asynchronous. Synchronous BCIs operate in a system controlled manner, where system orders the user when to start imagining executing a task. Signal processing in synchronous systems is limited within previously defined time windows, in which user is allowed to operate. On the contrary, there are asynchronous systems which allow the user to produce motor related patterns whenever he/she wishes to. Here the neurophysiological signal e.g. EEG should be continuously monitored to be able to detect "event" related patterns from "idle" thinking. In these systems, the challenging issue is to distinguish the occurrence of motor related changes from spontaneous EEG, accurately enough to have a reliable brain-controlled switch. Furthermore, the design of an asynchronous BCI has been carried out so far in two major ways. One way has Manuscript received September 19, 2007. E. B. Sadeghian is a M.Sc student with the Department of Biomedical Engineering, Amir Kabir University of Technology (Tehran Polytechnic), Tehran, Iran ( e-mail: e_bsadeghian@ aut.ac.ir ). M. H. Moradi is associated professor with the Department of Biomedical Engineering, Amir Kabir University of Technology, Tehran, Iran (e-mail: mhmoradi@ aut.ac.ir). been the incorporation of event detection task into classification of motor activity; by thresholding the classifier's scores [1]. Hereby functions of synchronous and asynchronous systems are gathered in one system. The veritable point is that in these systems the errors related to detection of every class of motor activities are incorporated in system's total performance. Take the example of detection of right hand versus left hand movement; here the imperfection due to detection of right hand movement occurrence versus idling accumulates with imperfection due to detection of left hand activity versus idling. Therefore system's total performance would suffer from both simultaneously. The second way of designing an asynchronous BCI is to detect the occurrence of motor activity prior to motor classification [3]. This is the area that has not received enough attention from BCI community in contrast to motor classification which has been already well studied in literature [4], [5]. Accordingly, there are feature and model- based methods capable of detecting motor related patterns from spontaneous EEG signals. Feature-based methods employ features capable of extracting oscillatory or low frequency changes like ERD/ERS patterns or ERPs respectively. LF-ASD [3] applies a wavelet based detector to extract energy changes in 1-4Hz frequency band. But in data with no significant ERPs, as here is the case, one needs to rely on ERD/ERS extracting features as Band Power [6], Entropy, etc. In model-based methods, we fit models describing "motor related event" and "idle/rest" parts of a signal. Thereafter act of detection changes to the question of "which model fits better?" In this paper we employ Band Power, m-spacing estimate of Entropy, Fractal Dimension from feature-based methods and "Quadratic detector" from model-based ones, to detect spectral changes of ERD/ERS patterns caused by execution of foot movement. The feature or detector signal is then fed to a threshold detector (THD) to discriminate between event and idle parts of the signal. II. DATASET This data has been provided by the Laboratory of Brain Computer Interfaces (BCI-Lab), Graz University of Technology (Prof. Gert Pfurtscheller, Reinhold Scherer) [7]. Recording consists of 2 subject's EEG of 3 runs with 30 trials each. At t=0s a cross “+” was presented; then at Detection of ERD/ERS in single-channel EEG with Application in Asynchronous BCI E. B. Sadeghian, M. H. Moradi B