Automated EEG Analysis Is Good For: Pharmaceutical Development, Advertising, Marketing, Interface Design, Brain Dysfunction Classification, Videogame Design

What is an Automated Method and Why is it Useful?

An automated methodology is one in which a complex multi-step procedure is made easy to use.  For example, in a methodology that reveals brain function related to human behaviour or changes to brain function that result from some stimulus, there are usually multiple steps in processing the raw brain activity data collected.  At minimum, there are 3 main steps: (1) the set-up of the experiment, (2) the mathematical analysis of the raw data collected to obtain interpretable results, (3) the interpretation of the results of the experiment derived from the analysis.  By automating the mathematical analysis of the raw data (step 2), the experimentor is freed to maximize their attention on the things that matter to them: the set-up of the experiment and the interpretation of the results.  A good automated methodology will provide easily interpreted results and make all of the complex analysis transparent to the user.  Using the automated method should be as simple as pressing a button.

Automated brain activity analysis methods have the potential to make analysis of brain function ubiquitous in non-traditional fields.  For example, the fields of neuropsychological evaluation, pharmaceutical development, marketing research, and video game development are currently adopting brain scan -type procedures to understand constraints on human behavior and how we can modify human behavior via various interventions and stimuli.  Analysis of brain function in these non-traditional fields has the potential to facilitate (1) optimization of stimuli to maximize its impact in our memories, (2) make it easy to find things on a computer display, (3) our understanding of how cognitive brain function is affected by pharmaceutical intervention, (4) classification and diagnosis of various brain dysfunctions, (5) understanding cognitive strategies employed to succeed in playing video games.   

The most common non-invasive method of obtaining information about brain function is by recording and analyzing the time-varying electrical activity measured from the scalp caused by activity of the brain.  This time-varying activity, called the electroencephalogram (EEG) has been used for many years to give us hints as to what might be happening in the brain and how activities of the brain relate to stimuli presented in experiments. 

The analysis of the EEG waveform, using standard non-automated methods, only provides hints as to what is going on in the brain  The EEG waveform, in its basic form, does not tell us exactly what the brain is doing; it only tell us that something about brain function is different given two different stimuli presented.  In other words, standard methods examine the EEG as a phenomenological process and this process changes when the activities inside the brain change.  Hence, to derive information about the brain when the EEG is considered a phenomenology, one must design an incremental set of experiments that figuratively ‘probe’ the brain to reveal what is going on inside.  This incremental experimental design encompasses an iterative procedure of hypothesis testing and experimental manipulation. 

In a hypothesis methodology, scientists researching brain function attempt to predict how the EEG will change based on manipulations in the stimuli received by study participants.  These predictions are based on their current understanding a theoretical model of how the brain functions. Because such research is driven by a theoretical model and building onto this theoretical model is iterative, standard EEG analysis methods require the contexts in which brain function is examined not be altered too dramatically.  Changes to the contexts (or paradigms) in which brain function is examined must be modified incrementally and these contexts must be kept very simple.  Moreover, these contexts must elicit brain activity that has as few active brain areas as possible. In other words, standard EEG processing methods, which are based upon a hypothesis testing strategy, are effective for examining simple brain function in well-understood behavioral paradigms.

By combining aspects of the standard methodology and some of the latest analysis methods into a larger methodology for examining brain function, a system-level analysis of brain function that identifies what is happening in many areas of the brain and how the activity of these areas relate to each other is available. This ‘system-level’ analysis provides monitoring of sensory-level brain function in conjunction with higher-level cognitive processing in the brain.  Moreover, a marriage of the latest analysis methods with standard methods allows for the analysis of brain activities of individual persons or comparisons of individual persons to a group.

Automation of a methodology that encapsulates the latest data analysis methods allows for the algorithms by non-specialists who are primarily concerned with the data processing result – what is happening in the brain, and not the details of how the result was determined. Automation provides ‘simplicity of use’ while at the same time supporting the creation of detailed ‘big-picture’ system-level descriptions of brain function.  The characteristic of encapsulation has additional advantages.  For example, encapsulation limits the user from modifying the original data or excluding segments of data based on visual inspection; this has the advantage of maintaining objectivity in the analysis of the data.    

An automated and encapsulated processing also allows for processing data obtained in a variety of contexts and behavioral paradigms.  This provides the freedom to apply brain analysis methods in novel circumstances in fields such as in: marketing research that examines aspects of human-product interaction, neuropsychological evaluations that employ a variety of behavioral tests, video game development that explores how games can be modified to require specific types of cognitive processing.  (Video games can be designed to be ‘good for you’ by exercising particular aspects of brain function.  This genre of games is referred to as ‘serious games’ and is a growing industry.) 

While automation of brain activity analysis clearly has some advantages, there are noteworthy disadvantages.  Generally, automated methods provide go-no-go results and the person making decision based on those results doesn’t know how close some results are to the threshold of this go-no-go case.  This is an issue of ‘fixed-thresholds’.  For example, in application, it is often a question of whether or not results derived ‘automatically’ can be trusted.  An EEG technician I once interviewed stated that she would never use an automated procedure for detecting epileptic activity.  She said if the automated process is wrong, then her patient will erroneously have their driver’s license revoked.  Another issue is that often automated processes do not provide an indication of the reliability of a result.  Since the user is not looking at the data themselves, there is no ‘feel’ for consistency in the data.  This is an important point because EEG typically contains many artifacts that have very high power and can have a significant impact on the overall variance of the data and in the outcome of automated processing.  This is an issue of ‘unknown result reliability’.

One way to address the notable disadvantages of automation is by providing ‘supervised automation’.  That is, the user of the automated procedure should be provided with a display of continuous values for each step of the automated process.  This information does not provide any sort of intervention in the automated process on the part of the user, but it does provide information that the user can use to help them chose to act on the result of the automated process.  For example, if the algorithm is designed to detect epileptic activity or classify brain activity in some way, a qualitative indicator such as a plot can be used to compare actual processing results against a priori determine decision points.  Similarly, the reliability of results can be determined and depicted in a way that the user can decide what results they will accept and reject.  This empowers the user so that they are more likely to be comfortable using the automated method.

In general, automated methods that encapsulate advanced data analysis methods offer the possibility of learning much more from the data than standard methods allow because the detail contained in the raw data can be revealed.  However, for the traditional investigator of brain function, automated methods might be a bridge too far from the data.  If this is the case, then I suggest using automated methods in conjunction with standard manual methods.  In this case, automated methods can also be used to provide important detail in a qualitative description of what might be going on the brain in a way that supports findings obtained by standard methods.  The qualitative description that automatic methods can provide can be to identify possible avenues of future research using the standard methods.

We’ve developed an automated methodology called MOST-EEG (Multiple Origin Spatio-Temporal –EEG) for analyzing brain function which is suitable for the applications described above. In this example, we applied MOST-EEG to create a data-driven model of brain function related to two different cognitive strategies in game playing.  In this case, we used the UnReal game engine to provide a 1^st-person perspective video game experience much like Wolfenstein 3D, Descent, Quake, or Half-Life. However, in this case, our custom levels were designed to isolate specific aspects of brain function.  To avoid any potential ethical issues, the levels are devoid of monsters and weapons.

Links:  Example automated brain function analysis results showing right-hemisphere brain activity related to play 3D videogames. More information about the algorithm used to calculate these results called MOST-EEG .  Research and technology development organization which investigates how brain function changes during drug treatment therapies called Applied Brain and Vision Sciences Inc.  Applied Brain and Vision Sciences Inc., employs the MOST-EEG algorithm to identify what areas of the brain are affected by pharmacetuical treatments and how the coordination of these areas changes with such treatments.