Home
Search
Collections
Journals
About
Contact us
My IOPscience
P300 speller BCI with a mobile EEG system: comparison to a traditional amplifier
This content has been downloaded from IOPscience. Please scroll down to see the full text. 2014 J. Neural Eng. 11 036008 (http://iopscience.iop.org/1741-2552/11/3/036008) View the table of contents for this issue, or go to the journal homepage for more
Download details: IP Address: 134.193.117.53 This content was downloaded on 26/05/2014 at 02:20
Please note that terms and conditions apply.
Journal of Neural Engineering J. Neural Eng. 11 (2014) 036008 (8pp)
doi:10.1088/1741-2560/11/3/036008
P300 speller BCI with a mobile EEG system: comparison to a traditional amplifier Maarten De Vos 1,2,3 , Markus Kroesen 4 , Reiner Emkes 4 and Stefan Debener 2,3,4 1
Methods in Neurocognitive Psychology, Department of Psychology, University of Oldenburg, Oldenburg, Germany 2 Research Center Neurosensory Science, University of Oldenburg, Oldenburg, Germany 3 Cluster of Excellence Hearing4all, University of Oldenburg, Oldenburg, Germany 4 Neuropsychology Lab, Department of Psychology, University of Oldenburg, Oldenburg, Germany E-mail: [email protected] Received 15 January 2014, revised 14 March 2014 Accepted for publication 14 March 2014 Published 24 April 2014 Abstract
Objective. In a previous study, we presented a low-cost, small and wireless EEG system enabling the recording of single-trial P300 amplitudes in a truly mobile, outdoor walking condition (Debener et al (2012 Psychophysiology 49 1449–53)). Small and wireless mobile EEG systems have substantial practical advantages as they allow for brain activity recordings in natural environments, but these systems may compromise the EEG signal quality. In this study, we aim to evaluate the EEG signal quality that can be obtained with the mobile system. Approach. We compared our mobile 14-channel EEG system with a state-of-the-art wired laboratory EEG system in a popular brain–computer interface (BCI) application. N = 13 individuals repeatedly performed a 6 × 6 matrix P300 spelling task. Between conditions, only the amplifier was changed, while electrode placement and electrode preparation, recording conditions, experimental stimulation and signal processing were identical. Main results. Analysis of training and testing accuracies and information transfer rate (ITR) revealed that the wireless mobile EEG amplifier performed as good as the wired laboratory EEG system. A very high correlation for testing ITR between both amplifiers was evident (r = 0.92). Moreover the P300 topographies and amplitudes were very similar for both devices, as reflected by high degrees of association (r > = 0.77). Significance. We conclude that efficient P300 spelling with a small, lightweight and quick to set up mobile EEG amplifier is possible. This technology facilitates the transfer of BCI applications from the laboratory to natural daily life environments, one of the key challenges in current BCI research. Keywords: P300, mobile EEG, brain–computer interface, wireless amplifier, comparison (Some figures may appear in colour only in the online journal)
been made, resulting in more powerful and robust BCI systems for communication. However, the development of simple, reliable and affordable EEG recording technology suitable for daily life applications received much less attention. Indeed, one of the key BCI challenges seems to bring the technology into real, daily-life applications [5, 6]. Aiming toward this goal, this study compared the performance of a small, wireless and quick to set up EEG system [7] with a state-of-the-art but
1. Introduction The fascinating brain–computer interface (BCI) technology has been developed to provide a means of communication [1, 26, 27] and for cognitive monitoring [28]. However, various key problems hamper the practical implementation of BCIs. Over the past few years, major progress in the field of signal processing (e.g. [2]), dry electrode technology [29, 30] and paradigm development (e.g. [3]) and software usability [4] has 1741-2560/14/036008+08$33.00
1
© 2014 IOP Publishing Ltd
Printed in the UK
M De Vos et al
J. Neural Eng. 11 (2014) 036008
Figure 1. Experimental setup. With the small and wireless mobile EEG amplifier, signals are wirelessly transmitted (left), while for the traditional wired laboratory EEG amplifier (right), a flat-ribbon cable is needed to connect the same cap to the recording amplifier. In this setup, electrode positions, electrode impedances and all parameters potentially influencing the experiment were kept the same for the two recording conditions.
amplifier (Brainamp, Brainproducts GmbH, Herrsching, Germany), using a repeated measurements’, single session design. We chose as paradigm the classical visual ERP speller [11], as it is among the most frequently used BCI paradigms and has been originally developed as a means of communication for locked-in patients (e.g. [12]). In this paradigm, a matrix of characters (typically 6 × 6) is shown to the individual, and each row and column highlighted (flashed) through enhancing the contrast. When the user focuses on only one letter during a whole sequence of highlighting all rows and columns in a random fashion, the paradigm corresponds to a visual oddball task. The ERP speller task is known to elicit a robust P300 brain response in most individuals. Typically, this highlighting of all rows and columns is repeated multiple times and the corresponding trials are averaged, which improves the ERP signal-to-noise ratio and facilitates decoding of the attended letter from the brain activity, but reduces the information transfer rates (ITRs). Using the identical electrodes cap and a software environment that could handle both amplifier systems [13], it was possible to isolate the influence of the EEG amplifier on P300 matrix spelling performance. This was achieved by using the identical electrode cap and electrode preparation with both amplifiers (see figure 1), and identical stimulation and signal processing. Hence, any difference in ERP spelling performance could be unambiguously attributed to the different EEG amplifiers used.
bulky laboratory EEG system in an event-related potentials (ERP) BCI speller task. Inspired by the marketing of wireless consumer EEG electronics, our group recently modified a wireless 14-channel EEG system from the company Emotiv (www.emotiv.com) and merged it with a research quality EEG electrode cap (www.easycap.de). As illustrated in figure 1, this resulted in a fully head mounted small, wireless and lightweight EEG with full flexibility regarding electrodes positioning. Our wireless, gel electrodes EEG system was found to deliver good-quality single-trial P300 signals recorded during a 2-class oddball task while subjects were freely walking outdoors [7]. For BCI applications, the system has interesting features, because it is cost-efficient, requires very few minutes of set-up time and is so small (5 × 5 × 2 cm) and lightweight (55 grams) that its presence is hardly noticed by the user. In a follow-up study, the wireless EEG system was validated with a 3-class auditory oddball paradigm recorded during outdoor seated and walking conditions [8]. It was found that in 19 out of 20 individuals, single-trial P300 amplitudes could be used to decode rare target versus non-target events. These first studies complement and extend a previous report on P300 acquisition during indoor walking [9]. While they clearly demonstrate the potential of the wireless EEG system to be used in daily life environments, they provide no information regarding how well, or how poorly, consumer EEG hardware performs when rigorously compared to state-of-the-art laboratory EEG devices. It is conceivable that low-cost consumer electronics is just not up to par with well-established high-quality laboratory EEG devices. In light of the fact that by far most BCI work seems to be done in experimental, laboratory environments, it appears important to determine the quality of different EEG systems in general, and in particular focus on those that could help in progressing from laboratory settings to uncontrolled daily-life environments. The significance of this research is also mentioned in a recent BCI competition report, which identified the evaluation of mobile, wireless EEG systems as an important goal for the near future [10]. Here, we compared our mobile EEG as introduced previously [7] to a well-established, wired laboratory EEG
2. Material and methods 2.1. Participants
Fifteen healthy individuals free of past or present neurological or psychiatric conditions participated in the study. Data from two individuals had to be discarded, one due to a mistake by the experimenter and one due the subject not following task instructions. The remaining sample consisted of 13 individuals (9 female, age 23 to 45 years). All participants gave written informed consent prior to participation. Several participants had performed the ERP speller paradigm already before, but 2
M De Vos et al
J. Neural Eng. 11 (2014) 036008
no one had substantial experience in BCI use. The study was approved by the University of Oldenburg ethics committee.
Stimulation parameter settings revealing robust spelling performance were used, as the goal was not to achieve optimal ITR. Stimulus duration was 125 ms and interstimulus interval was respectively 62.5 ms for the wireless EEG and 60 ms for the wired system. Note that these values represent approximations that cannot be exactly obtained, due to the screen refresh rate of 60 Hz. The difference was inevitable because of the difference in sampling rates between the two devices. However, given the small difference it seems unlikely that this contributed a noticeable bias to the results. Between letters a 2000 ms pause was implemented. Four different German sentences of 19 letters each were generated and used in balanced order for training and testing: HEINZ_MALT_53_PUDEL, ASTRID_KLAUT_13_PKW, ANDREA_HOLT_12_WOKS and GERD_NUTZT_25_KEILE. They were designed aiming for similar spread over rows and columns of the 6 × 6 letter matrix. The training block was done without feedback, and the online testing block with feedback. The analyzed EEG epochs started at 0 ms and lasted 800 ms. A decimation frequency of 20 Hz was used. Step-wise linear discriminant analysis was used for classifier training [14], using default BCI2000 parameter settings (e.g. all channels were used without a spatial filter, p-value for including features in the step-wise procedure was 0.1 and for excluding 0.15).
2.2. Data acquisition
EEG data were recorded from 14 sintered Ag/AgCl electrodes (at the 10–20 sites FPz, F3, Fz, F4, C3, Cz, C4, TP9, TP10, P3, Pz, P4, O1 and O2), which were inserted into an elastic cap manufactured by Easycap (www.easycap.de). The reference and ground electrodes for the wired EEG system were placed at FCz and AFz, and for the mobile EEG the same positions were used for common mode suppression and driven right leg. The mobile EEG amplifier weighted 55 g (size 5 × 5 × 2 cm3) and was tightly attached to the cap, approximately between electrodes O1 and O2. The wired 32-channel EEG amplifier was positioned behind the subject’s head and connected with a cable of approximately 108 cm length. The wired EEG system was of size 18.7 × 15 × 11.2 cm3 (including a power supply box) and weighted 2960 g. For this system, a fiber-optic cable was required to transmit the data from the amplifier to an additional hardware box, which was connected to a recording PC. Figure 1 visualizes the setup for both conditions. The cable between the cap and the lab amplifier was the only difference besides the amplifier itself. The sampling rate for the mobile EEG system was 128 Hz and could not be modified (0.16–45 Hz band-pass). The sampling rate for the wired EEG was set to the closest possible rate (200 Hz; 0.1–200 Hz bandpass). Electrode impedances were checked before the experiment and were lowered below 20 k prior to recording onset. Pilot recordings revealed that the electrode impedance remained stable over the total recording duration (approximately 35–45 min). We therefore refrained from checking electrode impedances separately for each amplifier. This also ensured a very short pause needed to switch from one recording system to another (
Search
Collections
Journals
About
Contact us
My IOPscience
P300 speller BCI with a mobile EEG system: comparison to a traditional amplifier
This content has been downloaded from IOPscience. Please scroll down to see the full text. 2014 J. Neural Eng. 11 036008 (http://iopscience.iop.org/1741-2552/11/3/036008) View the table of contents for this issue, or go to the journal homepage for more
Download details: IP Address: 134.193.117.53 This content was downloaded on 26/05/2014 at 02:20
Please note that terms and conditions apply.
Journal of Neural Engineering J. Neural Eng. 11 (2014) 036008 (8pp)
doi:10.1088/1741-2560/11/3/036008
P300 speller BCI with a mobile EEG system: comparison to a traditional amplifier Maarten De Vos 1,2,3 , Markus Kroesen 4 , Reiner Emkes 4 and Stefan Debener 2,3,4 1
Methods in Neurocognitive Psychology, Department of Psychology, University of Oldenburg, Oldenburg, Germany 2 Research Center Neurosensory Science, University of Oldenburg, Oldenburg, Germany 3 Cluster of Excellence Hearing4all, University of Oldenburg, Oldenburg, Germany 4 Neuropsychology Lab, Department of Psychology, University of Oldenburg, Oldenburg, Germany E-mail: [email protected] Received 15 January 2014, revised 14 March 2014 Accepted for publication 14 March 2014 Published 24 April 2014 Abstract
Objective. In a previous study, we presented a low-cost, small and wireless EEG system enabling the recording of single-trial P300 amplitudes in a truly mobile, outdoor walking condition (Debener et al (2012 Psychophysiology 49 1449–53)). Small and wireless mobile EEG systems have substantial practical advantages as they allow for brain activity recordings in natural environments, but these systems may compromise the EEG signal quality. In this study, we aim to evaluate the EEG signal quality that can be obtained with the mobile system. Approach. We compared our mobile 14-channel EEG system with a state-of-the-art wired laboratory EEG system in a popular brain–computer interface (BCI) application. N = 13 individuals repeatedly performed a 6 × 6 matrix P300 spelling task. Between conditions, only the amplifier was changed, while electrode placement and electrode preparation, recording conditions, experimental stimulation and signal processing were identical. Main results. Analysis of training and testing accuracies and information transfer rate (ITR) revealed that the wireless mobile EEG amplifier performed as good as the wired laboratory EEG system. A very high correlation for testing ITR between both amplifiers was evident (r = 0.92). Moreover the P300 topographies and amplitudes were very similar for both devices, as reflected by high degrees of association (r > = 0.77). Significance. We conclude that efficient P300 spelling with a small, lightweight and quick to set up mobile EEG amplifier is possible. This technology facilitates the transfer of BCI applications from the laboratory to natural daily life environments, one of the key challenges in current BCI research. Keywords: P300, mobile EEG, brain–computer interface, wireless amplifier, comparison (Some figures may appear in colour only in the online journal)
been made, resulting in more powerful and robust BCI systems for communication. However, the development of simple, reliable and affordable EEG recording technology suitable for daily life applications received much less attention. Indeed, one of the key BCI challenges seems to bring the technology into real, daily-life applications [5, 6]. Aiming toward this goal, this study compared the performance of a small, wireless and quick to set up EEG system [7] with a state-of-the-art but
1. Introduction The fascinating brain–computer interface (BCI) technology has been developed to provide a means of communication [1, 26, 27] and for cognitive monitoring [28]. However, various key problems hamper the practical implementation of BCIs. Over the past few years, major progress in the field of signal processing (e.g. [2]), dry electrode technology [29, 30] and paradigm development (e.g. [3]) and software usability [4] has 1741-2560/14/036008+08$33.00
1
© 2014 IOP Publishing Ltd
Printed in the UK
M De Vos et al
J. Neural Eng. 11 (2014) 036008
Figure 1. Experimental setup. With the small and wireless mobile EEG amplifier, signals are wirelessly transmitted (left), while for the traditional wired laboratory EEG amplifier (right), a flat-ribbon cable is needed to connect the same cap to the recording amplifier. In this setup, electrode positions, electrode impedances and all parameters potentially influencing the experiment were kept the same for the two recording conditions.
amplifier (Brainamp, Brainproducts GmbH, Herrsching, Germany), using a repeated measurements’, single session design. We chose as paradigm the classical visual ERP speller [11], as it is among the most frequently used BCI paradigms and has been originally developed as a means of communication for locked-in patients (e.g. [12]). In this paradigm, a matrix of characters (typically 6 × 6) is shown to the individual, and each row and column highlighted (flashed) through enhancing the contrast. When the user focuses on only one letter during a whole sequence of highlighting all rows and columns in a random fashion, the paradigm corresponds to a visual oddball task. The ERP speller task is known to elicit a robust P300 brain response in most individuals. Typically, this highlighting of all rows and columns is repeated multiple times and the corresponding trials are averaged, which improves the ERP signal-to-noise ratio and facilitates decoding of the attended letter from the brain activity, but reduces the information transfer rates (ITRs). Using the identical electrodes cap and a software environment that could handle both amplifier systems [13], it was possible to isolate the influence of the EEG amplifier on P300 matrix spelling performance. This was achieved by using the identical electrode cap and electrode preparation with both amplifiers (see figure 1), and identical stimulation and signal processing. Hence, any difference in ERP spelling performance could be unambiguously attributed to the different EEG amplifiers used.
bulky laboratory EEG system in an event-related potentials (ERP) BCI speller task. Inspired by the marketing of wireless consumer EEG electronics, our group recently modified a wireless 14-channel EEG system from the company Emotiv (www.emotiv.com) and merged it with a research quality EEG electrode cap (www.easycap.de). As illustrated in figure 1, this resulted in a fully head mounted small, wireless and lightweight EEG with full flexibility regarding electrodes positioning. Our wireless, gel electrodes EEG system was found to deliver good-quality single-trial P300 signals recorded during a 2-class oddball task while subjects were freely walking outdoors [7]. For BCI applications, the system has interesting features, because it is cost-efficient, requires very few minutes of set-up time and is so small (5 × 5 × 2 cm) and lightweight (55 grams) that its presence is hardly noticed by the user. In a follow-up study, the wireless EEG system was validated with a 3-class auditory oddball paradigm recorded during outdoor seated and walking conditions [8]. It was found that in 19 out of 20 individuals, single-trial P300 amplitudes could be used to decode rare target versus non-target events. These first studies complement and extend a previous report on P300 acquisition during indoor walking [9]. While they clearly demonstrate the potential of the wireless EEG system to be used in daily life environments, they provide no information regarding how well, or how poorly, consumer EEG hardware performs when rigorously compared to state-of-the-art laboratory EEG devices. It is conceivable that low-cost consumer electronics is just not up to par with well-established high-quality laboratory EEG devices. In light of the fact that by far most BCI work seems to be done in experimental, laboratory environments, it appears important to determine the quality of different EEG systems in general, and in particular focus on those that could help in progressing from laboratory settings to uncontrolled daily-life environments. The significance of this research is also mentioned in a recent BCI competition report, which identified the evaluation of mobile, wireless EEG systems as an important goal for the near future [10]. Here, we compared our mobile EEG as introduced previously [7] to a well-established, wired laboratory EEG
2. Material and methods 2.1. Participants
Fifteen healthy individuals free of past or present neurological or psychiatric conditions participated in the study. Data from two individuals had to be discarded, one due to a mistake by the experimenter and one due the subject not following task instructions. The remaining sample consisted of 13 individuals (9 female, age 23 to 45 years). All participants gave written informed consent prior to participation. Several participants had performed the ERP speller paradigm already before, but 2
M De Vos et al
J. Neural Eng. 11 (2014) 036008
no one had substantial experience in BCI use. The study was approved by the University of Oldenburg ethics committee.
Stimulation parameter settings revealing robust spelling performance were used, as the goal was not to achieve optimal ITR. Stimulus duration was 125 ms and interstimulus interval was respectively 62.5 ms for the wireless EEG and 60 ms for the wired system. Note that these values represent approximations that cannot be exactly obtained, due to the screen refresh rate of 60 Hz. The difference was inevitable because of the difference in sampling rates between the two devices. However, given the small difference it seems unlikely that this contributed a noticeable bias to the results. Between letters a 2000 ms pause was implemented. Four different German sentences of 19 letters each were generated and used in balanced order for training and testing: HEINZ_MALT_53_PUDEL, ASTRID_KLAUT_13_PKW, ANDREA_HOLT_12_WOKS and GERD_NUTZT_25_KEILE. They were designed aiming for similar spread over rows and columns of the 6 × 6 letter matrix. The training block was done without feedback, and the online testing block with feedback. The analyzed EEG epochs started at 0 ms and lasted 800 ms. A decimation frequency of 20 Hz was used. Step-wise linear discriminant analysis was used for classifier training [14], using default BCI2000 parameter settings (e.g. all channels were used without a spatial filter, p-value for including features in the step-wise procedure was 0.1 and for excluding 0.15).
2.2. Data acquisition
EEG data were recorded from 14 sintered Ag/AgCl electrodes (at the 10–20 sites FPz, F3, Fz, F4, C3, Cz, C4, TP9, TP10, P3, Pz, P4, O1 and O2), which were inserted into an elastic cap manufactured by Easycap (www.easycap.de). The reference and ground electrodes for the wired EEG system were placed at FCz and AFz, and for the mobile EEG the same positions were used for common mode suppression and driven right leg. The mobile EEG amplifier weighted 55 g (size 5 × 5 × 2 cm3) and was tightly attached to the cap, approximately between electrodes O1 and O2. The wired 32-channel EEG amplifier was positioned behind the subject’s head and connected with a cable of approximately 108 cm length. The wired EEG system was of size 18.7 × 15 × 11.2 cm3 (including a power supply box) and weighted 2960 g. For this system, a fiber-optic cable was required to transmit the data from the amplifier to an additional hardware box, which was connected to a recording PC. Figure 1 visualizes the setup for both conditions. The cable between the cap and the lab amplifier was the only difference besides the amplifier itself. The sampling rate for the mobile EEG system was 128 Hz and could not be modified (0.16–45 Hz band-pass). The sampling rate for the wired EEG was set to the closest possible rate (200 Hz; 0.1–200 Hz bandpass). Electrode impedances were checked before the experiment and were lowered below 20 k prior to recording onset. Pilot recordings revealed that the electrode impedance remained stable over the total recording duration (approximately 35–45 min). We therefore refrained from checking electrode impedances separately for each amplifier. This also ensured a very short pause needed to switch from one recording system to another (
