Cs.appstate.edu
Efficacy Evaluation of the Emotiv EPOC Neuroheadset & Associated SoftwareEric Jackson1 and Jonathan Tate21Macon Early College, Franklin, NC2South Iredell High School, Statesville, NCResearch Experience for Teachers, Department of Computer Science Appalachian State University, July 2015Abstract: The Emotiv EPOC neuroheadset is a non-medical grade electroencephalogram (EEG) made available by the company Emotiv based in San Francisco, California. It takes part of its name, EPOC from the technique in EEG analysis of a time-related event as displayed in EEG data. In this study, the EPOC and software was evaluated for efficacy in four separate trials: 1st as a non-medical EEG displaying brainwave activity over time, 2nd utilizing the Cognitive Suite to control a computer game with only brainwave activity, 3rd utilizing the Affective Suite to display a change in basic emotional states over time, and 4th utilizing the Expressive Suite where a virtual robot mimics the headset wearer’s facial expressions. The results were varied as each trial proved to have differing difficulties of use and meaning of data. It was determined that the headset and software works well as a non-medical EEG, not well at all controlling software with brainwave activity, inconclusive at modelling basic emotions, and does not work well mimicking facial expressions.Key words: Emotiv, EPOC, neuroheadset, brainwave, brain-computer interface. Introduction1.1 Device InformationThe Emotiv EPOC is a consumer grade, high resolution, multichannel electroencephalography (EEG) system. This device is designed to allow everyday people the opportunity to measure, observe, and monitor the brain. It can be thought of as a bridge to consumers from commercial EEG units operated and interpreted by trained professionals: the commercial units can cost tens of thousands of dollars. Emotiv touts the use of the unit as a Brain-Computer-Interface (BCI) and a therapeutic device. These non-traditional functionalities are enabled via proprietary mathematical algorithms and corresponding application suites which encourage the user to explore and train the software. The application suites include the Affectiv (emotion recognition), Expressiv (facial recognition), and Cognitiv (brain state awareness). At the heart of the unit is the non-invasive EEG headset which is designed to provide EEG signal to the software via an encrypted wireless connection. Emotiv EPOC uses sequential sampling method, single ADC (analog to digital converter), at a rate of 128 SPS (2048 Hz internal). Sequential sampling is a method that is used to reduce drift and jitter errors which is introduced by timing errors in the sampling processes. [1]Emotiv EPOC operates at a resolution of 14 bits per channel with a frequency response between 0.16 - 43 Hz. The Emotiv EEG headset uses a wet electrode system whose conductivity is enhanced with a saline solution. [1]1.2 EEG OverviewElectrical activity in the brain is seen in an EEG as patterns of waves which indicate electrical impulses related to the communication between neurons. In general, EEG signals have a broad spectral content similar to pink noise, which can reveal oscillatory activity in specific frequency bands. Only large populations of active neurons can generate electrical activity which is recordable by an EEG. This activity is then characterized using time-frequency analysis [3] via mathematical algorithms [4].EEGs have been used historically since 1935 in attempts to effectively diagnose or monitor certain health conditions which have affected normal neural activity in the brain. For example, when seizure activity is present EEG waves will appear as rapidly spiking waves on the EEG data capture. Patients with lesions of the brain, which can result from tumors or stroke, may have unusually slow EEG waves depending on the size and the location of the lesion.5996180Figure 1: Evidence of epilepsy [2]00Figure 1: Evidence of epilepsy [2]An EEG can also be used to diagnose other disorders that influence brain activity, such as Alzheimer's, certain psychoses, and sleep disorders such as narcolepsy. EEGs may also be used to evaluate trauma or the extent of brain damage in comatose patients. During surgery, monitoring cerebral blood flow using an EEG may be indicated. [5]When employing a traditional EEG system, any data collected during electromusculeographic and electrooculographic events (EMG and EOG respectively), are either completely discarded or are filtered. This data is considered noise or “dirty signal”. In contrast, Emotiv uses the distribution of sensors around the face to triangulate muscle sources and to build classification systems to identify specific facial expressions in the Expressiv portion of their software.One of the more novel aspects of the Emotiv unit is its wireless connectivity/portability. This allows the user a great degree of freedom when wearing the unit. Traditional EEGs are administered in quiet locations, with the patient being still, lying down or in a reclining position while “attached” to several devices to facilitate data capture and analysis. Even though the Emotiv device capitalizes on EMG and EOG events, it is recommended to limit movement, but the wired connection to the computer is not present.1894205103505Figure 2: Screenshot of sensor location00Figure 2: Screenshot of sensor locationIn both traditional EEG configurations and the Emotiv, electrodes are placed in locations and assigned names which are specified by the International 10-20 system. This method was developed to ensure standardized reproducibility so that a subject's studies could be compared over time and subjects could be compared to each other. The "10" and "20" refer to the fact that the actual distances between adjacent electrodes are either 10% or 20% of the total front–back or right–left distance of the skull. This system is based on the relationship between the location of an electrode and the underlying area of cerebral cortex. Each site has a letter to identify the lobe and a number to identify the hemisphere location. The letters F, T, C, P and O stand for frontal, temporal, central, parietal, and occipital lobes, respectively.The EPOC has CMS/DRL references in the P3/P4 locations to offer optimal positioning for accurate spatial and spectral resolution as you can see in figure 2. [1]15868656985Figure 3: Conventional EEG setup [7]00Figure 3: Conventional EEG setup [7]1.3 Brain-Computer-Interfacecenter1461110Figure 4: Detecting electrical signals [10]0Figure 4: Detecting electrical signals [10]A BCI is a system that receives central nervous system (CNS) information in the form of electrical signal. This signal is then analyzed in such a way as to create identifiable signatures which can be used to enact triggers for the initiation of some action – typically an action that the user is unable to perform due to a physical limitation. The EPOC device and software is designed to be used to increase user awareness of mental states and to “use” this awareness to initiate the triggers in the Emotiv BCI system. The Emotiv SDK system is enabled to allow users to sandbox their own adaptions of the interface, with limitations on access to the proprietary algorhithms. There are 3 distinct types of BCI which are characterized at a high level based on the method used to acquire the EEG recording – Non-invasive systems use the application of sensors on the scalp to capture electrical signal. Non-invasive systems have the limitations of poor spatial resolution (where is the signal originating?) due to low signal to noise ratio (SNR) and interference from unwanted EMG and EOG signals. Semi-invasive procedures involve the placement of electrodes on the surface of the brain which provide better spatial discovery and higher signal to noise ratios, but it does require surgery. And finally, the invasive method involves the placement of micro-electrodes into the brain cortex.[6] These recordings, of course, have a very high spatial accuracy and very high signal to noise ratio, but the tendency of the brain is to create scar tissue around the foreign body. This reduces long term efficacy.The Emotiv system is of course non-invasive. It suffers from all the limitations of low SNR and unwanted signals. Due to its consumer grade nature, easily replaceable parts and supplies are required to keep the unit operating up to its full potential. There is a constant requirement for saline solution to keep the electrodes wet, the 14 electrodes need to be installed and removed for each use, the unit needs to be charged before each use, and after consistent use the electrode pads which hold the saline solution will need to be replaced. 1.4 The Software Suites15506191509623Figure 5: Cognitive Suite00Figure 5: Cognitive SuiteThe Cognitiv Suite is purported to use only brainwave input, while filtering out muscle activity from both the face and the scalp. However, by their own admission, even after stringently stating that there is NO muscle activity in the Cognitiv data, they include some muscle data to encourage novice users. All actions in this suite must be trained against an established neutral baseline which is also trained. These actions are not externalized, but are rather attempts to control a model in the Emotiv Control Panel environment. Actions include “ideas” such as push, pull, move to each side, rotate, and drop. The actions cannot be correlated to a thought of the action being performed, but are interpolated associations to an electrical state in the brain. This would be considered an entire BCI system of electrical data, triggering, and action, for example. There are significant mathematical manipulations which are applied to the signal to initiate an action against the model. 15309853002280Figure 6: Expressive suite 00Figure 6: Expressive suite The Expressiv Suite presents the user with the same type of training environment as the Affectiv suite for training facial expressions such as raised and furrowed brows, blinks, winks, and eye movements. In general, the recognition of facial movements is simply pattern matching between specific sensors. The wave forms are characterized then compared to other sensor outputs. These comparisons could be across many channels and there may be multiple types of fitting and matching algorithms employed to build a recognition profile.center1002665Figure 7: Affective Suite00Figure 7: Affective SuiteThe Affective Suite attempts to report or create a visualization of emotional states based on universal brainwave characteristics. However, the confounding issue is still how to handle artifacts, and there is no standardized EEG data available that is in the proper context of emotional determination. The Affectiv Suite, in contrast to the Emotiv and Cognitiv Suites, does not have a training program. It is important though, that each user creates and references only their user profile. Emotiv states that this is what allows their software to increase detection accuracy as the user engages the software – this is obviously a training function in an attempt to create some type of characterization. Emotions such as excitement, engagement, frustration and relaxed are the focus of the Affectiv Suite. Once again, Emotiv claims that it disregards any motion artifacts. And again it applies proprietary mathematical algorithms, this time in the form of fractal signal analysis. In the suite, there is a Status that notifies the user if there is too much noise in the signal. If such a condition arises, the data capture simply stops. In other words, don’t become too animated when experiencing excitement because you will inhibit the software’s ability to “understand” your emotional state.center1213866Figure 8: Screen shot of Test Bench EEG stream00Figure 8: Screen shot of Test Bench EEG streamTest Bench allows viewing and saving of raw EEG streams from compatible headsets in the form of proprietary XML file type. This data can be exported to Excel as a csv file. In theory, a research can study the raw EEG data being captured by the device. However, by Emotiv’s own admission – “Testbench does not contain any of our detections, there is absolutely no way you can view the Testbench data and make any conclusions about the EEG data or detections (Geoff – Emotive Customer Service). What is to be clearly noted here is Geoff’s reference to our detections. In other words, there is no signal processing occurring in Test Bench and a user cannot make any conclusions about what the signal means. However, a user can notice that “some” event has possibly occurred. Testbench has 4 available displays. We only referenced 3 – EEG, Gyro, and Data Packets. The EEG displays the raw EEG data from the headset. It has various selection tools that can be used to alter the display. The Gyro view will show x and y axis motion data, and the Data Packet will show connectivity quality. It should be noted, that the unit did appear have excellent connectivity. In the Testbench, there is the ability to create markers in the files by so that data analysis can be simplified. We attempted to create a virtual COM port as instructed in the user’s manual, but there were issues with the software compatibility.[1],[9]2.0 MethodGlobal Procedure: The following procedure was conducted before each of the below trials.The USB transceiver dongle was inserted into a USB port. The felt tabs connected to the electrodes were moistened with saline solution until saturated and then installed onto the headset arms. The headset was powered on and placed on the subject’s head by sliding it from the top of the head down where the arm with the rubber insert is just behind the ear lobe. The software was initiated to check for good connection on each sensor and the sensors were gently pressed/adjusted until a green light appears for each sensor in the EPOC Control Panel Headset Setup screen or any screen in Test Bench.2.1 EEG EvaluationProcedure - baseline: The Testbench software was initialized and the EEG tab selected with a 200?V channel spacing. The subject was instructed to remain as motionless as possible minimizing any facial movement with eyes open to establish a baseline output. This was done over a period of three to five minutes to ensure consistency in the readout, and to ensure subject was motionless and comfortable. There was also an attempt to clear the subject’s mind of any complex thought or emotion aiding in the neutral baseline state. The EEG data is displayed in ?V per time and the largest time scale on the axis is approximately 10 seconds. 87630840105Figure 9: Baseline – eyes open00Figure 9: Baseline – eyes openData: The raw data can be exported into Microsoft Excel as a csv file and graphed as a line connected scatter plot. This was found to have little use as it is produces a graph identical to the EEG readout in the software. Since the purpose of investigation was only to evaluate the effectiveness of the equipment and software, producing graphs other than the graphs from the software was not necessary. All data in the EEG evaluation section will be displayed as cropped screenshots of Testbench software. The graph for the baseline is shown below in figure 9Result: A clear baseline.Procedure – eyes open then closed: The subject was instructed to remain as motionless as possible minimizing any facial movement with eyes open to establish an output that matches the baseline. When this was achieved, the subject was instructed to gently close the eyes.Data: 200?V channel spacing. There was a clear change in activity, specifically an increase in average amplitude and an increase in frequency. Below, the eyes closed region is to the right side of the dashed line. 0-3810O1O2Figure 10: Eyes open then closedO1O2Figure 10: Eyes open then closedResult: an increase in amplitude and frequency is shown, especially in the O1 and O2 sensors. These are sensing the occipital lobes of the brain. The occipital lobe is responsible for sight, image recognition and perception. When the eyes close the activity should increase in an attempt to recognize the new image of the back of the eyes.Procedure – flashing lights eyes closed: The subject was instructed to remain as motionless as possible minimizing any facial movement with eyes closed. When this was achieved, a keychain flashlight was flashed on and off approximately 3 times per second for 5 seconds at a distance of 10cm and switching between the left and right eye. Data: The channel spacing was decreased to 100?V for increased sensitivity. 50470248082Figure 11: eyes closed flashing lightsFigure 11: eyes closed flashing lightsResult: To the left of the dashed line represents no lights flashing and to the right represents the flashing light pattern. There seems to be no real discernable difference between the two sides. Procedure EEG and Affective Suite Evaluation – breathing deeply: The subject was instructed to remain as motionless as possible minimizing any facial movement with eyes closed. When this was achieved, the subject was to breathe deeply for 3 minutes remaining as motionless as possible. -381001095375Figure 12: breathing start00Figure 12: breathing startData: The Control Panel software was added to the Testbench screen shot in an attempt to display any emotional state change. The red line represents engagement/boredom, blue represents frustration, green represents meditation and black represents instantaneous excitement. The vertical axis can represent a qualitative increase or decrease calculated by EEG frequency data most commonly associated with the above listed emotional states. The higher the red line the more engaged someone is and the lower the red line displays an increase in boredom.center0Figure 13: breathing end00Figure 13: breathing endResult: Figure 12 represents an eyes closed baseline EEG similar to the right side of figure 10; however, there is an obvious change in figure 13 displaying a decrease in amplitude and a small increase in frequency. The control panel Affective Suite shows no average change in red line engagement/boredom and green line meditation, a significant increase in instantaneous excitement, and an initial increase then decreasing in blue line frustration. The subject was asked if there were any outstanding emotions that were experienced during the test. The subject experienced only concentration and mild frustration during breathing, but a significant feeling of relaxation at the end. This could be exemplified by the increasing/decreasing blue line frustration. However, there seems to be no concrete indicator that the Affective Suite accurately displayed the emotional state of the subject. 2.3 Cognitive Suite Evaluation14382751262380Figure 14: cognitive suite training00Figure 14: cognitive suite trainingProcedure: The goal of the Cognitive Suite is to train the software to recognize brainwave patterns to be able to control software with only brainwave commands. When the software recognizes a specific pattern that has been trained as “Push” then the software can be programmed to activate a keyboard or mouse action thereby controlling a game or other software. The Control Panel software was initiated and the Cognitive Suite tab was selected. Under the training tab “Neutral” was selected and the subject remained motionless and attempted to clear the mind of all thoughts to obtain a neutral brainwave pattern. The action “Push” was trained and for 8 seconds the subject attempted to establish a brainwave pattern the software would recognize and interpret as a pushing command. The more training the user practices the higher percentage of ability the software scores the user. Results: After 30 minutes of training a score of 10% was reached; however, shortly after the score dropped to 0%. An additional 15 minutes of training wouldn’t increase the percentage further than 1%. Another adult male and a teenage female trained with “Push” and scored a 100% and 99% respectively, but were unable to use that training to consistently push the little box in figure 14. All other attempts to train in another action i.e. pull, left, right, etc. were met with very slow progress and/or termination of training due to time constraints. Using the neuroheadset to control simple actions with the associated software proved difficult to inconclusive, as training was difficult and the use of the training inconsistent. 2.4 Expressive Suite EvaluationProcedure: The goal of the Expressive Suite is to train the software to recognize facial/scalp movement (EMG) and eye movement (EOG) patterns to be able to control software. The software contains a virtual robot shown in figure 15 that will mimic your facial movements in real time. When the software recognizes a specific pattern that has been trained as “Push” then the software can be programmed to activate a keyboard or mouse action thereby controlling a game or other software.center1371600Figure 15: expressive suite training00Figure 15: expressive suite trainingThe Control Panel software was initiated and the Expressive Suite tab was selected. Under the training tab “Neutral” was selected and the subject remained motionless and attempted keep the eyes and face from moving, and any muscle from the neck up as relaxed as possible. The action “Raise Eyebrows” was trained and for 8 seconds the subject’s eyebrows were raised so the pattern could be established and the robot could mimic the wearer of the headset. Care was taken that the subject’s eyes did not blink during the training. If the robot did not mimic the facial movement, the wearer could click the Sensitivity tab and adjust the sensitivity until the facial movements were repeated. The possible facial actions are: blink, right wink, left wink, look right/left, raise brow, furrow brow, smile, clench, right smirk, left smirk and laugh. Data/results: There is no data to be displayed other than qualitative success or failure. All facial actions were trained and the only facial actions that were repeated by the robot was the raise and furrow brow.3.0 ConclusionThe Emotiv EPOC is a high quality consumer EEG device that attempts to function well beyond the inherent limits of non-invasive EEG. These limitations are low signal to noise ratio and very poor spatial recognition. The development of semi and fully invasive EEG techniques is to overcome these issues. However, there is always a trade-off as discussed in the BCI section. Per Emotiv’s SDK user manual significant and proprietary mathematical methods are applied to either remove or leverage EOG or EMG signal. This discretionary use is applied to the various software suites in order to meet Emotiv’s predefined objectives. In other words, there is a disconnect between what information an EEG can supply, and what Emotiv is attempting to do with that information. Our research indicates that no matter how clever these methods are, the results produced are inconsistent and of low quality.Software installation and use was both cumbersome and confusing. Of note was the failed attempt to create data markers in the Testbench application. This is one of the more important requirements for robust data exploration. Markers are analogous to time stamps and are very helpful when attempting to analyze EEG data sets which are very large. Creating a virtual COM (communication) port failed due to software incompatibility. There was no work around available for this problem. There various software applications were not integrated which increased the learning curve for using the device properly. In addition, several features of the software failed continuously. Specifically the training of various actions in both the Cognitiv and Effectiv Suites was very inconsistent from session to session and between test subjects. Many times the training of an action would be proceeding as expected, then the progress would be lost and could not be recovered. In the Effectiv Suite it was not possible to reliably train more than 3 expressions out of 11.In conclusion, the device appears to function well as an electroencephalogram if the subject remains still. Curiously enough, the design of the device and the placement of the battery compartment prevents the user from using the device in a traditional clinical setting – the user can’t recline or lay down; they must remain in a seated position and as motionless as possible for most uses. This would make accurate comparisons between this device and a commercial EEG using a traditional setting. Further study using a medical-grade EEG in comparison with the EPOC would shed more light on the quality of the EPOC’s data stream and would perhaps allow the EPOC to be standardized.Using the Cognitive Suite - the device is ineffective and inconsistent for one of the main design parameters – using the device to perform complex video game control using only brain waves. Using the neuroheadset to control simple actions with the associated software proved impossible in the research time frame of 3 weeks. Training of more than 1 out of 12 available actions was unsuccessful in the Cognitive Suite. In the Emotiv Suite, the research showed no correlation between perceived mental state and what the application indicated. There was no correlation between a task being performed and what the program indicated. Acknowledgements This work as funded by the US National Science Foundation (NSF) Research Experience for Teachers (RET) in the Department of Computer Science at Appalachian State University in Boone, North Carolina. Special thanks to Dr. R. Mitchell Parry and Dr. Rahman Tashakkori for their help and guidance.References[1] Troy Giunipero (n.d.) Emotiv Software Development Kit User Manual for Release 2.0.0.20. [Online]. Available at: (Accessed: 07/2015).[2] Joseph Sirven, MD | Robert S. Fisher, MD, PhD (09/2013) Juvenile Myoclonic Epilepsy, Available at: (Accessed: 7/2015).[3] L. Cohen (1995) Time–Frequency Analysis, 1st edn., New York: Prentice Hall.[4] Towle, Vernon L.; Bola?os, José; Suarez, Diane; Tan, Kim; Grzeszczuk, Robert; Levin, David N.; Cakmur, Raif; Frank, Samuel A.; Spire, Jean-Paul (March 6 2003) The spatial location of EEG electrodes: locating the best-fitting sphere relative to cortical anatomy, Available at: (Accessed: 7/2015).[5] John Hopkins Medicine () Electroencephalogram (EEG), Available at: (Accessed: 7/2015).[6] Panoulas, K.J., Hadjileontiadis, L.J.,Panas, S.M. (2010) 'Brain-Computer Interface (BCI): Types, Processing Perspectives and Applications', in Tsihrintzis, G.A., Jain, L.C., (ed.) Multimedia Services in Intelligent Environments SIST. Berlin Heidelberg: Springer-Verlag , pp. 299.[7] Teplan, M (2002) 'Fundamentals of EEG Measurement', Measurement Science Review, 2, pp. Section 2.[8] Dimigen, O. (2013) Toolbox for Simultaneous Eye tracking & EEG: Tutorial, Available at: (Accessed: 7/2015).[9] Roschler (2010) The Emotiv EPOC: Meeting the Future Head On, Available at: (Accessed: 7/2015).[10]Po Lei Lee, Yung Yang Lin (2002) Basic Principles of Electroencephalography & Magnetoencephalography, Available at: (Accessed: 7/2015). ................
................
In order to avoid copyright disputes, this page is only a partial summary.
To fulfill the demand for quickly locating and searching documents.
It is intelligent file search solution for home and business.