{"id":574,"date":"2010-08-01T15:25:51","date_gmt":"2010-08-01T15:25:51","guid":{"rendered":"http:\/\/www.physiologicalcomputing.net\/wordpress\/?p=574"},"modified":"2021-12-22T20:21:48","modified_gmt":"2021-12-22T20:21:48","slug":"functional-vocabulary-an-issue-for-emotiv-and-brain-computer-interfaces","status":"publish","type":"post","link":"http:\/\/www.physiologicalcomputing.net\/?p=574","title":{"rendered":"Functional vocabulary: an issue for Emotiv and Brain-Computer Interfaces"},"content":{"rendered":"

The Emotiv<\/a> system is a EEG headset designed for the development of\u00a0 brain-computer interfaces.\u00a0 It uses 12 dry electrodes (i.e. no gel necessary), communicates wirelessly with a PC and comes with a range of development software to create applications and interfaces.\u00a0 If you watch this 10min video<\/a> from TEDGlobal, you get a good overview of how the system works.<\/p>\n

First of all, I haven’t had any hands-on experience with the Emotiv headset and these observations are based upon what I’ve seen and read online.\u00a0 But the talk at TED prompted a number of technical questions that I’ve been unable to satisfy in absence of working directly with the system.
\n
\nAs I said, the Emotiv system runs on 12 EEG channels.\u00a0 According to the International 10-20 system, the channels are from frontal (AF3, AF4, F3, F4, F7, F8) and\u00a0 fronto-central (FC5, FC6) areas with less coverage of\u00a0 occipital (O1, O2), parietal (P8) and temporal sites (T7, T8).\u00a0 In essence, we’re talking about an EEG system with the majority of coverage over the frontal cortex.\u00a0 Looking at these details<\/a> of the researcher package, the system uses a real-time Fast Fourier Transform analysis so EEG data can be analysed in any of the classic EEG bands (delta, theta, alpha, beta) or customised bands.<\/p>\n

Looking at the TED video, a new user on the Emotiv can train the system by performing a number of mental actions on a 3D cube displayed on the screen.\u00a0 In the demo, the user is asked to ‘push’ the cube away and also attempts to make the cube disappear.\u00a0 The demo looks completely unrehearsed and what was really impressive was that a single trial of 7 seconds or so was necessary\u00a0 in order to capture an EEG template that could subsequently be associated with an action.<\/p>\n

<\/p>\n

This begs a question, which is admittedly dry and boring, but fundamental to this type of system.\u00a0 If I bought myself an Emotiv system (and who knows, I may do at some point in the future), how many unique commands does the system need to reliably identify for me to adopt a Brain-Computer Interface (BCI) as an alternative to or supplement for my conventional input device?<\/p>\n

This is what I’ve called the functional vocabulary of the BCI.\u00a0 In the case of the Emotiv system, distinct patterns of EEG activity are matched to specific commands\/verbs.\u00a0 If the system is trying to recognise the difference between a resting or neutral state and say a ‘push’ command, this is a two-category classification problem that any BCI should be able to do.\u00a0 As we add 3 more commands to the vocabulary of the system, such as pull the cube, move the cube left and move the cube right, we now have a four-category classification problem, which requires a greater\u00a0 level of sensitivity from the system.\u00a0 Logic also dictates that as we add more commands to the vocabulary, we increase the probability of false positives, i.e. when the system mistakes a ‘pull’ command from the user’s attempt to ‘move right’.\u00a0 So, the functional vocabulary of this particular BCI design is defined by the number of unique commands that can be reliably detected by the system (based on the training procedure).\u00a0 The ‘reliably’ part of that definition is important because the system has to successfully classify different patterns of EEG data consistently over time.<\/p>\n

There are a couple of usability issues surrounding the functional vocabulary of the system.\u00a0 First of all, how many commands need to be successfully implemented by the BCI in order for it to function as a working interface?\u00a0 This will vary depending on the requirements of the interface and whether the BCI is the sole means of input control – or whether it’s paired with a conventional input device, such as keyboard\/mouse or a gamepad.\u00a0 Based on what I’ve said above, it would be very sensible for the interface design to minimise the number of BCI-enabled commands to minimise the possibility of false positives.<\/p>\n

The second question surrounds the relationship between the recognition accuracy of the BCI and user perceptions of acceptability.\u00a0 For most input devices, we are accustomed to 100% accuracy – this may be less true for motion control input systems, such as the Wii or Kinect – but still, the ‘feel’ of those interfaces is pretty close to what used to be called in user interface circles WYSIWYG<\/a>. In the case of BCI, user acceptability will be strongly determined by the capability of the system to deliver WYTIWYG (What You Think Is What You Get).<\/p>\n

There is a secondary issue here that is perhaps only of interest to researchers\/system developers and that is the relationship between the number of psychophysiological channels and the functional vocabulary of the system.\u00a0 One would think that more channels = greater potential for discrimination\/classification with respect to the psychophysiological data, which translates into a higher number of items into the functional vocabulary.\u00a0 The problem with this presumed link between number of channels and the functional vocabulary is simple: the surface of the brain functions as a conductor and there is a lot of crosstalk between different EEG channels, especially for those sites that are located close together.\u00a0 Perhaps this is the crux of the problem, the propensity of\u00a0 psychophysiological channels to be correlated works against the optimisation of a BCI system, which seeks to maximise the number of working commands that it can offer to the user.<\/p>\n

I should also add a caveat; some categories of BCI do not work in this way.\u00a0 Those BCI systems that are based on a generic measure of cognitive (e.g. P300-based systems) or visual attention (e.g. those systems based on SSVEP) are concerned with identifying the item that elicits the greatest ‘interest’ from an array.\u00a0 This is a completely different dynamic as commands or alphanumeric alternatives are presented in a visuo-spatial array – and the array can present different items at different times, hence extending the vocabulary of the system.<\/p>\n

Obviously the user interfaces issues described above are not the whole story when it comes to BCI design for healthy users. A system like the Emotiv is a futuristic and sexy way to communicate with a computer, especially for the gaming market.\u00a0 It provides an illusion of telepathic powers, something that appeals to anyone brought up on comics and sci-fi movies.\u00a0 Fun is a strong motivation to purchase a BCI and train yourself how to use it, but it’s hard to have fun if communication between your brain and computer is repeatedly lost in translation.<\/p>\n","protected":false},"excerpt":{"rendered":"

The Emotiv system is a EEG headset designed for the development of\u00a0 brain-computer interfaces.\u00a0 It uses 12 dry electrodes (i.e. no gel necessary), communicates wirelessly with a PC and comes with a range of development software to create applications and interfaces.\u00a0 If you watch this 10min video from TEDGlobal, you get a good overview of […]<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"spay_email":""},"categories":[5,7],"tags":[18,28,37,61],"jetpack_featured_media_url":"","jetpack_shortlink":"https:\/\/wp.me\/pY315-9g","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"http:\/\/www.physiologicalcomputing.net\/index.php?rest_route=\/wp\/v2\/posts\/574"}],"collection":[{"href":"http:\/\/www.physiologicalcomputing.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/www.physiologicalcomputing.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/www.physiologicalcomputing.net\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"http:\/\/www.physiologicalcomputing.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=574"}],"version-history":[{"count":1,"href":"http:\/\/www.physiologicalcomputing.net\/index.php?rest_route=\/wp\/v2\/posts\/574\/revisions"}],"predecessor-version":[{"id":4727,"href":"http:\/\/www.physiologicalcomputing.net\/index.php?rest_route=\/wp\/v2\/posts\/574\/revisions\/4727"}],"wp:attachment":[{"href":"http:\/\/www.physiologicalcomputing.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=574"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/www.physiologicalcomputing.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=574"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/www.physiologicalcomputing.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=574"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}