Usability of four commercially-oriented EEG systems

Twelve of sixteen participants provided overall preference rating data for all four of the systems, ranking their most-preferred system as a one and their least-preferred system as a four. Table 3 shows rank descriptive scores for each system, omitting any participants that did not complete reports for all cases.

The subjective ratings, based on an ordinal scale, were analyzed using a non-parametric Friedman's test with Wilcoxon sign-rank tests for a priori pair-wise comparisons between BioSemi and each of the other systems (table 4). A statistically significant difference in ratings was found among the systems overall (χ²(3) = 16.5, p = 001; Kendall's W = 0.458). Follow up comparisons reveal that the B-Alert X10 was preferred significantly more often than the ActiveTwo system (table 4; z = −2.44, p < 05). The rating comparisons between the other systems suggest that subjects generally preferred the ActiveTwo more often than the EPOC and HMS, although these differences fell short of the criterion for significance. Across the group, all of the systems received rankings ranging from being the most-preferred (1) and least-preferred (4), except for the B-Alert X10 which was ranked by all participants in the top two slots, reflecting the overall comfort of the system.

Additionally, thirteen of the sixteen participants provided comfort ratings for all four systems on a scale from 1 to 7, with 1 labeled as 'very comfortable' and 7 labeled as 'very uncomfortable' table 5 shows the descriptive comfort rank scores for each system. Similar to the ranking data shown in table 3, the variability between the minimum and maximum ratings was smallest for the B-Alert X10 system (1–4) reflecting the overall comfort of the system, although the HMS also showed a slightly tighter range (1–6) than the ActiveTwo and EPOC systems (1–7).

A Friedman's test indicated overall significant differences in comfort ratings between systems, (χ²(3) = 20.52, p < 001; Kendall's W = 0.526). Follow up comparisons showed comfort ratings between the B-Alert X10 and the ActiveTwo were significantly different indicating more comfort for the X10 system (z = −2.35, p < 05), while the HMS was rated as significantly less comfortable than ActiveTwo (z = 1.97, p < 05) and there was no significant difference with the EPOC system (table 6). For comparisons of both preference and comfort, including partial data did not substantially alter the results.

Overall, concerns about system comfort identified by participants can be grouped into three broad categories: discomfort due to pressure on specific points of the scalp, discomfort from overall weight or constriction of the unit as a whole, and difficulty of general application and clean up.

When a head unit does not fit correctly, pressure may not be distributed across the area as designed, resulting in discomfort for the subject. For the laboratory-grade ActiveTwo system, discomfort typically occurs with a cap that is too small. Over an extended period of wear, this constant pressure may result in subject discomfort. However, this is a rare problem because researchers can select an appropriately sized cap for the user from a wide range available from the manufacturer. For the EPOC and HMS systems, however, the electrode array is held in place by spring-loaded pressure. On the EPOC, these are plastic pre-bent arms, while the HMS headset uses spring-loaded metal arms. In both cases, subjects reported feeling 'hot spots' that developed over the hour long experimental session due to the pressure of the electrode pad. Generally, these types of complaints did not appear immediately, but rather after several minutes or even an hour of wearing time. Also, problems of this nature were mostly common with individuals with larger heads, due to the increased pressure created by the spring action (especially the Emotiv system—see also section 4.3). We did not receive any similar complaints about the ABM B-Alert X10, likely because it uses relatively little pressure at the electrode site to hold it in place.

The ActiveTwo system components worn on the head are not very heavy (170 g), but both the B-Alert X10 and EPOC are notably lighter (110 and 125 g, respectively). The HMS, in contrast, headset is considerably heavier (510 g) and bulkier than the other units. The heavy HMS head unit resulted in some reports of neck pain and headaches by participants towards the end of the experimental session. This lack of comfort may arise because the weight is focused on fewer locations on the scalp rather than being distributed across the scalp in a diffuse number of points, as with the conventional caps used in the ActiveTwo. Multiple participants noted that the HMS was comparatively heavier, and one specifically complained about discomfort due to the system weight. The EPOC system also elicited similar problems stemming from uneven weight distribution, as it has fewer points of contact with the scalp, while no such complaint arose with the B-Alert X10 head unit. For future systems, determination of a comfortable, maximum weight for an EEG head unit will vary across participants, but designs that minimize pressure points are likely to be more comfortable overall.

Finally, we were surprised at participant feedback regarding system application and clean-up. While participants clearly would prefer minimal gel in their hair, their overall concerns were biased less towards this factor than towards the general comfort of use and ease of application; this is reflected in the ratings of overall preference described above (tables 3 and 4). For example, many noted how simple and non-invasive the saline-based EPOC system was; however, participants who had a lack of comfort from EPOC pressure points found this to be a greater problem than the gel used by the B-Alert X10 or ActiveTwo systems. Likewise, although the HMS system required virtually no clean-up and only minimal preparation from subjects, the difficulties encountered in twisting the electrodes to achieve a good signal (see section 4.4) and pressure from the electrode pins outweighed this advantage. This is probably why participants had an overall preference for the B-Alert X10 system. Even though it required some gel, the cleanup was minimal (especially when compared to ActiveTwo), and it was lightweight with virtually no problems of comfort, even for participants with larger head sizes who were included in this study.

A major challenge in comparing data across different EEG systems is the inconsistency of electrode locations. Although each system states the 10–20 locations for the electrodes (the industry standard scheme based on an inter-electrode percentage relative to the total scalp distance; see table 1), the accuracy of these intended locations during an EEG recording session may vary based on how the EEG systems fit across different head sizes and shapes. In the conventional method utilized by the laboratory-grade ActiveTwo system, the caps stretch between electrode locations to distribute them approximately equivalently with the percentage-based inter-electrode distances from subject-to-subject, and a specific cap size is chosen so that the range is correct. However, this requires purchasing and maintaining multiple caps. In contrast to this approach, the EPOC and HMS systems target a one-size-fits-all approach by utilizing semi-rigid components that allow the electrodes some flexibility in fitting different head sizes; however, the inter-electrode distance (for some electrodes) is fixed, limiting the total compensation possible for inter-electrode distances. The B-Alert X10, alternatively, comes in a range of sizes of electrode strips with a fixed distance between electrodes and attachment points at the end of the strips to an elastic headband.

In all of these systems, the manufacturer provides instructions for placement of the system on the wearer's head based on aligning one or two electrodes to anatomical landmarks (ActiveTwo: Cz; B-Alert X10: Cz; EPOC: T7/T8; HMS: Cz). The user must then accept a certain error in placement of the rest of the electrodes. For the ActiveTwo, the most common solution to this problem is to purchase a variety of cap sizes so that the error is minimized for each subject. For the EPOC, the fixed curvilinear distance between T7/T8 and the other electrode sites cannot be adjusted. Similarly, the HMS electrodes radiate out from CZ, which results in varying 10–20 location errors, highlighting the same fundamental issue for EEG systems with fixed inter-electrode distances. The B-Alert X10 allows setting the position of a single electrode site, but the inter-electrode distances are fixed along the plastic strip. The strips are pulled tight and fastened with Velcro to the headband in order to maintain good connectivity, which allows for some judgment in electrode placement when the system is put on the wearer. However, it should be noted that the plastic strip yielded a unique issue where it buckled in the back, with the plastic strip having to fold and bunch in order for the connector to align properly, resulting in some connectivity problems with the electrodes near the back of the head.

Another critical variation among the systems is the scalp areas covered by the electrodes (figure 2). Most notable in the coverage differences, the EPOC system lacks recording locations along the midline. When selecting a system for real-world applications, EEG experiments should consider which scalp locations produce the strongest signal of interest and take care to select an EEG system that will provide consistent and sufficient coverage of these scalp areas across participants.

Zoom In Zoom Out Reset image size

To determine the extent of the variability in electrode placement between systems, we digitized electrode placement of each system on ten different individuals with an ELPOS system (Zebris; Isny, Germany). This generates polar coordinates (azimuth, latitude, and radius) for each electrode position relative to anatomical points. These coordinates are compared to the actual 10–20 coordinates projected onto a sphere using the radius measured at that location, yielding differences in cord length between points (error in mm). Error reported here thus represents the accuracy and inter-individual variability in electrode placement for each system relative to the manufacturer's stated position. For the purpose of this study, we used the width of the plastic electrode snap for the ActiveTwo system. Based on indentations and residue left on subjects' skin after wearing the ActiveTwo, it is rare that gel extends beyond this area.

We found that the average error in the placement of the center of the electrode was comparable for all four systems (figure 3). However, if the size of the sensor is taken into consideration (see the horizontal bars in figure 3), the HMS system is the most accurate due to the large diameter of the sensors. It should be noted that for all systems, including the ActiveTwo, the size of the scalp area covered by the electrode compensates for the imprecision in the 10–20 location.

Zoom In Zoom Out Reset image size

The three wireless systems showed imprecision at specific regions of the scalp based on the form factor of the head unit. Figure 4 plots this error for each electrode in relation to the laboratory-grade ActiveTwo system. We found that the B-Alert X10 repeatedly showed errors and inter-individual variability at position POz. This is likely a result of how the cap is bent in order to plug into the electronics unit (see section 4.3). Otherwise, the variability of the B-Alert X10 is similar to variability in the ActiveTwo system. The EPOC shows the highest amount of error at O1 and O2 at locations at the back of the head over occipital cortex and F3 and F4 at locations at front of the head over frontal cortex. These positions are most susceptible to location errors based on the fixed length of the semi-rigid arms and variability in head size that changes the distance between these arms when expanded to fit the head circumference. The HMS shows the highest amount of position error for the P3 and P4 electrodes positioned over the parietal cortex. However, since the HMS has relatively large electrodes, collecting potential across a broad surface area, the position offset is likely least relevant for this system.

Zoom In Zoom Out Reset image size

Across the four systems, three approaches were implemented to facilitate conductivity between the electrodes and the scalp. The first was electrically conductive gel, which is the traditional approach for laboratory EEG systems used by the ActiveTwo and B-Alert X10 systems. This approach allows for electrical contact through hair, but increases set up time to gel each electrode (64 for ActiveTwo and 9 for X10) and requires clean up time for the participant at the end of the EEG recording session. The application of the gel is accomplished more quickly using the ABM design because there are only nine channels, and each electrode has the gel applied prior to application to the participant's head. Thus, the experimenter only needs to push down on the electrode once it has been placed on the participant's head to release the gel. The BioSemi system requires that a syringe of gel be used to fill each electrode separately after the EEG cap is positioned on the participant's scalp, which is time consuming (taking two experimenters working concurrently around 20 m to gel 64 electrodes).

The second approach utilized by the EPOC system was saline-soaked foam pads, which is also capable of establishing a connection through most types of hair. The pads are soaked in saline before they are inserted in the headset, so there is nothing to apply after the headset is positioned on the participant's scalp. This approach is also advantageous to the participant because the only residue left in their hair is saline. Since saline is roughly similar to sweat, it dries quickly after completion of the session and is not bothersome to the user.

The third approach was 'dry' electrodes on the HMS system, which required no gel or additional fluid. The dry electrodes are designed to work through hair by inserting a tool into a notch on the top of the sensor and twisting the sensor back and forth to separate hair, enabling direct contact between the sensor and the scalp. The pins are spring-loaded on concentric rings so that they roughly conform to the scalp. Thus, the array of electrode pins used by the HMS system is capable of contacting the scalp through hair. However, the HMS electrode can become tangled in long, thick hair resulting in an uncomfortable experience for the wearer and/or insufficient connectivity. We were unable to establish sufficient connectivity with the HMS system in one participant with thick, curly hair, despite success with the other three systems. As a result, although the HMS is more convenient for the participant because no liquid residue remains in their hair, two major trade-offs for the system are not comfortable to the participant from the twisting of the sensor in their hair and additional cleaning by the experimenter to remove tangled hair from the unit.

In addition to the initial application of the system, the total time to prepare the participant was also affected by the effort necessary to obtain acceptable electrode impedance, which indicates the quality of connection with the scalp. For example, with the HMS unit, even though the initial time to setup the headset on the participant was very quick, it often took considerable effort to make strong, reliable contact between the electrodes and the scalp at all positions simultaneously, especially at the Cz location. Additionally, this system required a 'settling time' of several minutes for the impedance to come within an acceptable range after application. As a result, the total time for preparation often matched or exceeded what was typical for the gel-based systems. In contrast, the EPOC and B-Alert X10 systems only required a small amount of manual manipulation of the electrodes to work the conductive liquid into contact with the scalp because the application of saline or conductive gel could be completed prior to the participant's arrival for the study. Thus, based on the current designs, the less dense, wet electrode systems were generally faster to apply than the dry electrode system. The BioSemi ActiveTwo still required the most time, simply due to the substantially larger number of channels to set up.

Another component of the total setup time relates to the ease of reading the current impedance values of each electrode. The interface for each of the three wireless systems was easier to use than the 'ActiveView' software for the laboratory-grade comparison system. In ActiveView, the voltage offset (impedance is not reported, but offset from the common mode follower circuit is used as a surrogate) is shown in a bar plot for all 64 channels on the same screen, so it is often time consuming to decipher which bar on the screen corresponds to which location on the scalp. In contrast, the other three systems report impedance in a graphical depiction that plots each electrode on the scalp, which greatly eases identification. It should be noted, however, that while both the HMS and EPOC software report in real-time, the current software for B-Alert X10 requires a manual button click to begin calculation to compute the impedance for the entire set of channels (taking takes several seconds for each electrode). As a result, the total time to read impedance on all channels for the B-Alert X10 system can be extremely long, and the duration of this process is extended every time a channel is manipulated and must be re-checked. Altogether, it was universally agreed among the experimenters collecting data that while the EPOC took the least amount of time and effort for total setup, and the ActiveTwo cap consistently took the longest, it was unpredictable whether the B-Alert X10 or HMS might be equally time consuming as the substantially higher-density ActiveTwo system.

Across all three approaches in the newer systems, we were pleasantly surprised by the stability of the connection between the electrodes and the scalp. Conductive gel (used by both ActiveTwo and B-Alert X-10) dries out a small amount during the 60 min experimental session which can result in an itchy sensation on the scalp for the participant, but it remained conductive throughout the experiment. The saline-soaked felt pads used in the EPOC remained damp enough to be conductive throughout the experiment, although this may be contingent on the relative humidity of the area where they are being used. The dry electrode sensors on the HMS system also maintained their conductivity as long as the headset position was not adjusted by a head scratch or other manipulation at the scalp.

As neuroscience moves from the laboratory into more complex settings, environments that cause a participant to sweat should increase the durability of scalp-electrode connections for every connection type. However, for any fluid conductivity-based system, there is a risk of bridging the gaps between electrode sites in high-density electrode systems. Interestingly, sweat may enhance the signal quality in the pin-array style of dry electrodes by bridging the small inter-pin distances. Excessive sweating, however, may result in an electrical bridge between pin arrays identical to the bridging between electrodes in other systems. Future EEG systems designs will need to be implemented and tested in whole-body-movement paradigms and various temperature environments to examine these real-world recording considerations.