|
IMPORTANT
This work is outdated and the provided code only works with the first generation MUSE pictured on the left. It is identifiable by by the MUSE logo being written in white. This MUSE is no longer available and is no longer supported. |
Our laboratory has been getting a lot of interest about our work with MUSE - and a lot of it has to do with whether or not MUSE is truly capable of ERP research. The short version is - YES! We have our first methods paper under review at Psychophysiology and soon as it is published it will be available here. With that said, we are happy to share the longer version below with some sample data.
The excerpt below is taken from our submitted paper - below this except, the evidence.
"In recent years there has been an almost explosive growth of low-cost (i.e., less than $500 USD) electroencephalographic (EEG) recording systems. While most of the systems on the market offer software developer kits allowing scientists to access the raw data for research purposes – little to no work has been done to validate the effectiveness of these systems for research, and specifically for research using event-related potential (ERP) designs. Indeed, the ERP methodology is difficult to implement in low-cost non-standard research grade equipment for several reasons. First and foremost, there is the issue of data quality and whether or not low-cost systems can deliver sampling rates (i.e., >= 250 Hz) and data quality (i.e., noise free, a small number of artifacts) conducive for traditional ERP analyses. Two other key issues that are a cause for concern relate to the issue of experimental timing: first, how one can “mark” the data for subsequent ERP analysis, and second, the issue of non-standard electrode locations for analysis – ERP components are typically associated with analysis of specific electrode locations.
Data quality is of paramount importance for the results of ERP experiments to be interpreted meaningfully. Indeed, while this issue is obvious to any scientist using EEG or ERP designs, these concerns were specifically highlighted in the seminal Picton et al. (2000; see also Luck, 2014) paper in which issues such as electrode type and quality (e.g., Coles et al., 1986; Kutas, 1997), the minimum number of electrodes necessary for meaningful interpretation (e.g., Srinivasan et al.,1998), and the capabilities of the amplifier (e.g., Cadwell & Villarreal, 1999) were stated to specifically impact data quality and thus the ability to draw meaningful conclusions from EEG/ERP data. Specific amplifier characteristic such as the number of bits available for the converter (8 minimum), the gain of the amplifier, and the common-mode rejection ratio were all stated in the Picton paper (Picton et al., 2000) to have specific minimum values necessary to achieve sufficient data quality. Thus, the concern here with low-cost systems is simply whether or not the hardware meets these proposed standards as the supposition is that if they do not these systems will not be able to provide EEG data of sufficient quality for meaningful interpretation. While we agree that all of the aforementioned concerns are valid, a more meaningful test of data quality is rather straightforward – collect data from a low-cost system and directly determine whether or not a low-cost EEG system can provide data that reliably results in visible and statistically quantifiable ERP components.
Other issues that occur when using low-cost EEG systems relate to the issue of event timing and the use of non-standard electrode locations. Typically, in an ERP paradigm an event marker is sent from the stimulus computer to the recording computer via a parallel or TTL cable to “mark” the data. Importantly, marking the data in this manner affords the ability to precisely extract epochs of data centered on the onset of events of interest and thus allows the researcher to create event-related average waveforms for subsequent analysis (Coles et al., 1986; Luck, 2014). Extending from this, the researcher typically seeks to analyze specific electrode channels for specific ERP components – channels where the ERP component is maximal and has been reported before (Rugg & Coles, 1995; Luck, 2014). Here, we hypothesized that ERPs could be recorded with a MUSE system in a dramatically more efficient way at a fraction of the cost making trading off with these issues a worthwhile goal. To accomplish this, first, with regard to experimental timing we developed a protocol in which we recorded data streaming directly on to a stimulus computer immediately following the presentation of experimental events of interest for 1000 ms thus negating the need to mark the continuous EEG data. Second, after an a priori analysis of existing data, we decided to simply analyze ERP components at non-standard locations. For example, while the P300 ERP component is typically maximal at posterior locations on the midline we did not have an electrode(s) at this location and as such we were forced to analyze the P300 component at electrode locations that we did have (here, electrodes TP9 and TP10). While obviously our solutions to these issues are not ideal, the purpose of the present research was to demonstrate that a low-cost EEG system could be used to conduct ERP research and thus we “worked with what we had”."
The excerpt below is taken from our submitted paper - below this except, the evidence.
"In recent years there has been an almost explosive growth of low-cost (i.e., less than $500 USD) electroencephalographic (EEG) recording systems. While most of the systems on the market offer software developer kits allowing scientists to access the raw data for research purposes – little to no work has been done to validate the effectiveness of these systems for research, and specifically for research using event-related potential (ERP) designs. Indeed, the ERP methodology is difficult to implement in low-cost non-standard research grade equipment for several reasons. First and foremost, there is the issue of data quality and whether or not low-cost systems can deliver sampling rates (i.e., >= 250 Hz) and data quality (i.e., noise free, a small number of artifacts) conducive for traditional ERP analyses. Two other key issues that are a cause for concern relate to the issue of experimental timing: first, how one can “mark” the data for subsequent ERP analysis, and second, the issue of non-standard electrode locations for analysis – ERP components are typically associated with analysis of specific electrode locations.
Data quality is of paramount importance for the results of ERP experiments to be interpreted meaningfully. Indeed, while this issue is obvious to any scientist using EEG or ERP designs, these concerns were specifically highlighted in the seminal Picton et al. (2000; see also Luck, 2014) paper in which issues such as electrode type and quality (e.g., Coles et al., 1986; Kutas, 1997), the minimum number of electrodes necessary for meaningful interpretation (e.g., Srinivasan et al.,1998), and the capabilities of the amplifier (e.g., Cadwell & Villarreal, 1999) were stated to specifically impact data quality and thus the ability to draw meaningful conclusions from EEG/ERP data. Specific amplifier characteristic such as the number of bits available for the converter (8 minimum), the gain of the amplifier, and the common-mode rejection ratio were all stated in the Picton paper (Picton et al., 2000) to have specific minimum values necessary to achieve sufficient data quality. Thus, the concern here with low-cost systems is simply whether or not the hardware meets these proposed standards as the supposition is that if they do not these systems will not be able to provide EEG data of sufficient quality for meaningful interpretation. While we agree that all of the aforementioned concerns are valid, a more meaningful test of data quality is rather straightforward – collect data from a low-cost system and directly determine whether or not a low-cost EEG system can provide data that reliably results in visible and statistically quantifiable ERP components.
Other issues that occur when using low-cost EEG systems relate to the issue of event timing and the use of non-standard electrode locations. Typically, in an ERP paradigm an event marker is sent from the stimulus computer to the recording computer via a parallel or TTL cable to “mark” the data. Importantly, marking the data in this manner affords the ability to precisely extract epochs of data centered on the onset of events of interest and thus allows the researcher to create event-related average waveforms for subsequent analysis (Coles et al., 1986; Luck, 2014). Extending from this, the researcher typically seeks to analyze specific electrode channels for specific ERP components – channels where the ERP component is maximal and has been reported before (Rugg & Coles, 1995; Luck, 2014). Here, we hypothesized that ERPs could be recorded with a MUSE system in a dramatically more efficient way at a fraction of the cost making trading off with these issues a worthwhile goal. To accomplish this, first, with regard to experimental timing we developed a protocol in which we recorded data streaming directly on to a stimulus computer immediately following the presentation of experimental events of interest for 1000 ms thus negating the need to mark the continuous EEG data. Second, after an a priori analysis of existing data, we decided to simply analyze ERP components at non-standard locations. For example, while the P300 ERP component is typically maximal at posterior locations on the midline we did not have an electrode(s) at this location and as such we were forced to analyze the P300 component at electrode locations that we did have (here, electrodes TP9 and TP10). While obviously our solutions to these issues are not ideal, the purpose of the present research was to demonstrate that a low-cost EEG system could be used to conduct ERP research and thus we “worked with what we had”."
Sample Data
Presented here is data from a standard visual oddball paradigm. At the top, you see data collected from a Brain Vision ActiChamp system with a standard analysis.
Here, you see data from the same system during the same task, but using what we call a "reduced" analysis to parallel what we have to do with the MUSE data.
And here you see the MUSE ERP data. While it is not identical to the data from the big array system - the commonality of the pattern between the MUSE ERP data and the big system data is impossible to miss.
In other words... IT WORKS!
Here, you see data from the same system during the same task, but using what we call a "reduced" analysis to parallel what we have to do with the MUSE data.
And here you see the MUSE ERP data. While it is not identical to the data from the big array system - the commonality of the pattern between the MUSE ERP data and the big system data is impossible to miss.
In other words... IT WORKS!