Evaluation of Ssvep as Passive Feedback for Improving the Performance of Brain Machine Interfaces

Research in brain-computer interfaces have focused primarily on motor imagery tasks such as those involving movement of a cursor or other objects on a computer screen. In such applications, it is important to detect when the user is interested in moving an object and when the user is not active in this task. This paper evaluates the steady state visual evoked potential (SSVEP) as a feedback mechanism to confirm the mental state of the user during motor imagery. These potentials are evoked when a subject looks at a flashing objects of interest. Four different experiments are conducted in this paper. Subjects are asked to imagine the movement of flashing object in a given direction. If the subject is involved in this task, the SSVEP signal will be detectable in the visual cortex and therefore the motor imagery task is confirmed. During the experiment, EEG signal is recorded at 4 locations near visual cortex. Using a weighting scheme, the best combination of the recorded signal is selected to evaluate the presence of flashing frequency. The experimental result shows that the SSVEP can be detected even in complex motor imagery of flickering objects. The detection rate of 85% is achieved while the refreshing time for SSVEP feedback is set to 0.5 seconds. INTRODUCTION Brain-computer interfaces (BCIs) have gained greater attention in the past several years because they provide the possibility to directly create a communication channel between the human brain and the computer by translating human intentions into control signals for the computer. In non-invasive BCIs, electroencephalography (EEG) is commonly used due to its excellent time resolution, ease of data acquisition, portability and lower system cost. Other brain monitoring mechanisms such as positron emission tomography (PET) and magnetic resonance imagery (MRI) find only restricted application in brain-computer interfaces although they are widely used in medical and research settings [1]. Brain-computer interfaces have been primarily used to provide assistance to severely disabled people [1,2]. However, BCIs can also be used as an alternative interface for regular human computer interaction. New developments in BCIs have already enabled users to navigate in virtual scenes, manipulate virtual objects or play games just by means of their cerebral activity [3]. BCIs are also being considered as effective tools in future computer aided design systems by engaging visual imagery in design process [4] or helping the selection of geometries in virtual environments[5]. Motor imagery i.e. the imagery involved in motion has been the primary focus of most BCIs [2, 3]. Here signals are obtained during imagined sensorimotor rhythms (SMR). Typically, the SMRs are detected based on features of the μ and β rhythms (8–12 and 18–26Hz) [1]. Changes in the amplitudes of these frequency bands are referred to as event-related desynchronization (ERD) (i.e. decrease) and event-related synchronization (ERS) (i.e. increase). The rhythms decrease or desynchronize with movement or its preparation, and increase or synchronize after movement and with relaxation [6]. BCIs based on sensorimotor rhythms (SMR) are the basic elements for movement control in virtual environments. It has been shown that using SMR based BCI, it is possible to control the 2D motion of a cursor [7], [8]. The main advantage of motor imagery classification is that it requires no external stimuli and the ongoing EED is used to classify the mental task. However, its implementation in continuous human computer interaction is subjected to false _______________________________________________ * These authors have equal contributions to this manuscript. 2 Copyright © 2012 by ASME detection of movement because certain brain activities involved even in an idle state can mimic motor imagery. Moreover, while BCI research efforts have succeeded in providing communication for some users, it has often been report that (very roughly) 20% of subjects do not exhibit BCI performance adequate for effective control [9]. Therefore in order to use BCI as an alternative in human computer interactions, it is critical to increase its robustness. To overcome these problems a hybrid BCI approach has been proposed. The main idea of hybrid BCI is to use a stimulus based response such as P300 or visual evoked potential with sensorimotor rhythms to increase the robustness of the BCI [8], [10]. Similarly, to reduce the false alarm in classification of ongoing EEG signals, Pfurtscheller et al [11] have proposed a brain switch by combining visual evoked potential and event related synchronization (ERS)-based BCI. Although, they could achieve high robustness by combining SSVEP and ERS, their hybrid BCI is only designed to detect one mental task and may not have the same performance in complex situations. In this paper, we have conducted a series of experiments to verify the robustness of hybrid BCIs in multi-mental task situations. To this end, steady state visual evoked potential (SSVEP) is used in combination with different motor imagery. The idea is that in human computer interaction, the motor imagery usually occurs when the subject is gazing on a virtual object to move or rotate. Therefore by flashing the object of interest on the screen, it may be possible to get passive confirmation about correct detection of user’s intent. This passive confirmation (feedback) is achieved through SSVEP detection. The organization of this paper is as follow: The background behind visual evoked potential and SSVEP is provided in next section. It is followed by detailed experimental methods and signal processing algorithms for detecting SSVEP. Finally, the experimental results and conclusions are presented in the final sections. STEADY STATE VISUAL EVOKED POTENTIAL Among different brain signals that have been employed for EEG based BCIs, VEP (Visual Evoked Potentials) based system has been studied since 1970s [12]. It is commonly accepted as a method that provides high information transfer rate and needs less user training. VEP is the response of human brain to the visual stimulus. It is categorized into transient VEP (TVEP) and steady state VEP (SSVEP) which correspond to visual stimulus with low and high frequency, respectively. TVEP arises when the stimulation frequency is less than 2 Hz, while SSVEP appears when the repetition rate of the stimulus is higher than 6 Hz [13]. It is well agreed that SSVEP has a wider area of application than TVEP because in most cases, the human’s brain is considered in steady state of excitability in which the responses that elicited by the high frequency visual stimulus will overlap each other. Since the characteristics caused by two kinds of stimulus are different, researchers usually use temporal methods for TVEP analysis and frequency analysis for SSVEP case [14]. A Steady-State Visual Evoked Potential (SSVEP) is a resonance phenomenon arising mainly in the visual cortex when a person is focusing his/her visual attention on a light source flickering with a frequency above 6 Hz [13]. The SSVEP can be elicited up to at least 90 Hz [15] and could be classified into three ranges: low (up to 12 Hz), medium (12-30) and high frequency (> 30 Hz). In general, the SSVEP in low frequency range has larger amplitude responses than in the medium range. Thus, the lower frequencies are easier to detect. The high-frequency SSVEP ranges have the advantage of a minimum visual fatigue caused by flickering, making the SSVEP-based BCI a more comfortable and stable system [16]. At a same time these frequencies experience the weakest SSVEP which make the SSVEP detection a more difficult task and requires computationally expensive algorithm. MATERIALS AND METHODS EEG signals were recorded using the Emotiv neuroheadset at 4 channels on the scalp. The names of channels that are used for this study are based on the international 10-20 system are: P7, O1, O2, and P8. Signals were recorded at sampling rate of 2048 Hz, and sent to the computer wirelessly after being downsampled to 128 Hz. The selected visual stimulation was a three-dimensional cube that flashed on the screen. The background is black while the cube had white surfaces and blue edges. All surfaces of the cube were flashing at a frequency of 13 Hz. Four different experiments were conducted in this study. Each experiment consists of two parts: part one is conducted in only one trial during which the cube is not flashing. In this part of the experiment, subject is asked to gaze at the object and conduct an imagery movement in a given direction. This experiment only contains motor imagery and therefore can be used as the control. Part two contains 5 trials in which the subject is performing the same task as part one but with presence of the flashing stimulus. In each trial for all experiments, subject is asked to sit on a comfortable position and look at the screen. EEG data for each trial is recorded once the cube appears on the screen until it disappears. The motor imagery task that subject is instructed to perform each experiment are shown in the Figure 1. The arrows describe how the cube is rotating or translating and its direction. The numbers denote different states during the experiment process. All the arrows and numbers in the figures are only for the convenience of description and do not exist in the experiment. In the first experiment, the cube only rotates along the horizontal axis which passes its geometric center. In the second experiment, the cube moves from the position of initial state 1 to that of final state 2, it is also rotating along the vertical axis