SREE NARAYANA GURUKULAM COLLEGE OF ENGINEERING

KOLENCHERY

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

Seminar Report

BRAIN COMPUTER INTERFACE

Submitted by:

JYOTHY S MENON

Reg.No. 56430

2005-2006

SREE NARAYANA GURUKULAM COLLEGE OF ENGINEERING

KOLENCHERY

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

CERTIFICATE
This is to certifi that the Seminar Report entitled "BRAIN

COMPUTER INTERFACE" was presented by JYOTHY S MENON

(Reg. No. 56430) offinal year Computer Science and Engineering, Sree

Narayana Gurukulam College of Engineering , Kolenchery in partial

fu@lment of the requirement for the award of Degree in Computer Science and Engineering of Mahatma Gandhi University during the Academic year 2005-2006.

Prof. Dr. Janahan La1 Head of Department Computer Science & Engg. SNGCE, Kolenchery.

P.S. Smijesh Staff in charge Computer Science & Engg. SNGCE, Kolenchery.

I express my sincere thanks to Dr. (Prof.) P.S Janahanlal Head of the Department for providing me the guidance and facilities for the seminar.
I extend my sincere gratitude to Lecturer P.S Smijesh for his co-

operation for presenting the seminar.

I also extend my sincere thanks to all other faculty members of

Computer Science and Engineering Department and my friends for their support and encouragement.

JYOTHY S MENON.

Brain Computer Interface

CONTENTS

SLNO

TOPIC

PAGE NO

1. 2.

Abstract.........................................................................................2 Introduction...................................................................................3 2.1. What is a Brain-Computer Interface (BCI)?.......................... 3 2.2. How does the brain work?.................................................... 6 2.3. How can we make it possible?............................................. 7

3.

BCI Architecture............................................................................9 3.1. EEG Acquisition Unit..........................................................0 I 3.2. EEG Signal Processing Unit.............................................. 15 3.2.1. Preprocessing Unit................................................. 16 3.2.2. Feature extraction & classification unit................... 16 3.3. Output Unit......................................................................... 17

4.

BCI Functioning...:........................................................................ 18 BCI Implementation Example....................................................... 25 Factors affecting BCI.................................................................... 27 Key issues in the development & application of BCI systems......28 Practical BCls............................................................................... 29 Conclusion................................................................................... -30 References.................................................................................. 31

5.

6.
7. 8. 9. 10.

Department of Computer Science & Engineering

SNGCE. Kadayiruppu

Brain Computer Interface

1. Abstract
A brain-computer interface (BCI) is literally a direct technological interface between a brain and a computer not requiring any motor output from the user. That is, neural impulses in the brain are intercepted and used to control an electronic device. The brain's electrical output is translated by a computer into physical outputs,.such as moving a cursor on a computer screen. To make the computer understand what the brain intends to communicate necessitates monitoring the brain activity. Among the possible brain monitoring methods, the scalp recorded electroencephalogram (EEG) constitutes an adequate alternative because of its good time resolution, relative simplicity and noninvasiveness. The EEG signals are analyzed and mapped into actions inside the computer rendered environment. A BCI allows a person to communicate with or control the external world without using the brain's normal output pathways of peripheral nerves and muscles. Messages.and commands are expressed not by muscle contractions, but rather by electrophysiological signals from the brain. BCls provide an alternative communication and control option for the severely disabled.There has been great success in using cochlear implants in humans as a treatment for non-congenital deafness. There is also promising research in vision science indicating retinal implants may some day prove to be similarly successful in treating non-congenital blindness.

Department of Computer Science & Engineering

SNGCE; Kadayiruppu

Brain Computer Interface

2. Introduction

2.1. What is a Brain-Computer Interface (BCI)?

Since the creation of computer rendered environments, the computers power has been increasing to an amazing speed. Each year, the possibilities offered by a computer are growing extraordinarily. However, the communications with a machine have unfortunately not significantly evolved. We still have to content ourselves of an old keyboard, a mouse and a screen display. It is undeniable that it would be a wonderful evolution to improve our interactions with the machines. Following this idea, eminent computer professionals were concerned about developing a new way of communication using the brain. Based on measures of the electrical activity of the brain, the goal was to discover the user's thoughts in order to execute his orders. To accomplish this prowess, the natural properties of the brain were used: to identify a particular mental activity of the subject, we observe the variation of his mental activities in frequency and in time using an appropriate material. A mathematical algorithm will next analyze the collected data and guess about the subject thoughts.

This technology offers creating a completely new way of communication offering lots of possibilities. For example, for persons with movements' disabilities, a BCI could help them to command their electric wheelchair without needing somebody else help, giving them more independence. As another example, we can image substitute some damaged nerves by computers intercepting the brain messages and redirecting them to the muscle. For people having to deal with epilepsy crisis, it would be interesting to improve their knowledge of the different processes that stimulate a crisis and try to avoid them, controlling their own brain as another muscle.
Department of Computer Science & Engineering SNGCE, Kadayiruppu

Brain Computer Interface

The idea of using our brains to directly control a machine isn't particularly new. As far back as 1967, Edmond Dewan described experiments using subjects wired to an electroencephalograph (EEG), which records and graphs the electrical activity of the brain. With practice, the subjects were able to reduce the amplitude of their brain's alpha rhythms, to transmit Morse code to a teleprinter. Research into the Brain Computer Interface, or BCI, began in earnest in the early 70's, when the United States Department of Defense saw the promise of fighter pilots using their minds to directly control their planes. Given the technology of the time, there was limited success, and the program was cancelled. But the groundwork was laid for a field of research now growing rapidly. A major motivation has been to help patients suffering from conditions such as cerebral palsy, or spinal injuries, which inhibit physical control, but which leave intellectual faculties intact. Over the last decade, great advances have been made. Automatic systems capable of understanding different facets of human communication will be at the heart of human-computer interfaces (HCI) in the near future. An HCI which is built on the guiding principle: "think and make it happen without any physical effort" is called a brain-computer interface (BCI). Indeed, the "think" part of this principle involves the human brain, "make it happen" implies that an executor is needed (here: a computer) and "without any physical effort" means that a direct interface between the brain and the computer is required. To make the computer understand what the brain intends to communicate necessitates monitoring the brain activity. Among the possible brain monitoring methods, the scalp recorded electroencephalogram (EEG) constitutes an

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

adequate alternative because of its good time resolution, relative simplicity and noninvasiveness when compared to other methods such as: functional magnetic resonance imaging, positron emission tomography, magnetoencephalography and electrocorticogram. Furthermore, there is clear evidence that observable changes in EEG result from performing given mental activities.

A successful BCI must be bidirectional. Getting information into the brain

is relatively easy: we can use the normal sensory channels, such as sight or hearing. But getting the information out of the brain by studying its electrical signature is a much harder problem. The sheer complexity of the brain's measurable activity produces EEG traces which present a formidable problem of interpretation. However, by focusing on very specific areas of brain activity, such as motor function, it is possible to analyze EEG data using filters, Fourier transforms, and neural networks, to extract some useful signal from the noise. Electrophysiological signals reflecting brain activity are acquired from electrodes on the scalp or within the head and processed to produce measurements of specific signal features, such as amplitudes of evoked potentials or EEG rhythms or firing rates of single neurons that reflect the user's intentions. These features are translated into commands that operate a device, such as a word processor or a neuroprosthesis. Successful operation depends on the interaction of two adaptive controllers, the user and the system. The user must develop and maintain good correlation between his or her intentions and the signal features selected by the BCI; and the BCI must select features that the user can control and must translate those features into device commands correctly and efficiently. Brain-computer interface (BCI) technology is a potentially powerful new communication and control option for those with severe motor disabilities. The pace and volume of BCI research have grown very rapidly over the past decade.
Department of Computer Science & Engineering SNGCE, Kadayiruppu

Brain Computer Interface

The success of this exciting work depends on close and productive interaction of scientists, engineers, and clinicians from many different disciplines and requires recognition and attention to a number of crucial issues. This meeting is designed to foster such interdisciplinary interactions and address these crucial issues, and thereby promote the development of BCI systems of practical value to people with disabilities.

2.2 How does the brain work?

Our brain is mostly composed of neurons interconnected to each others forming an enormous network (Figure 1). They communicate together through their axons using small electric impulses (pV).

During a particular mental activity (MA), we can observe electric potential variations of the brain active regions. Neurons are generating small electrical variation that summed over a region give a potential variation in the space. These variations can be decomposed in series of electrical maps. This means that a mental activity can be seen as a sequential continuation of brain electrical states.

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

Based on recordings of the brain electrical activity, we would like to reflect as well as possible the dynamic of the functional states of the brain and identify the states representing a mental task as well has possible in order to be able to recognize later on the same mental activity.

2.3 How can we make it possible?

In order to measure the electrical activity of the brain in a non-invasive way, nowadays, the best choice is to employ electroencephalogram (EEG) material. EEG measurements have proved their utility in the medical world and are not expensive. The 10120 international system is used for positioning the electrodes on the scalp. This norm defines 16 electrodes positions best positioned in order to represent as well as possible the electrical activity of the brain (Figure 2).

Why not using another more sophisticated way to measure the brain electrical activity? We could for example use the Magnetic Image Resonance
Department of Computer Science & Engineering SNGCE, Kadayiruppu

Brain Computer Interface

method (IRM)? Let's remember our first objective. All we would like to do is to identify some pattern in the mental activity of the subject in order to give him an immediate feed back. The IRM are useful to describe the brain form and to study the blood fluctuations but they don't measure the electrical activity of the patient. It is easy to imagine that there is a correlation between the electrical activity and the blood fluctuations in the brain, but the latency (variation in time) is fundamentally different! Another more interesting idea is to use microchips introduced directly to the contact of the brain. It poses some new problems like difficulties of avoiding infections in the human body, or transmitting the data measured at the contact of the brain to the computer (through wireless communication?). Moreover, this kind of solutions will just measure the brain activity in a very close region of the brain. Studying the ecological states of a few trees randomly selected in a forest, is not enough to conclude about the ecological state of the entire forest. So, it would be necessary to increase the number of introduced microchips augmenting at the same time the discomfort of the user. Moreover, integrating chips in the human corps is not yet very well defined in the medicine world and neither in the ethic of the humanity. In conclusion, the EEG systems based on 10120 international positioning, is efficient enough for our objective: detecting some special mental activities.

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

4
-3.3001 0.5917 -6.0412 -6.1430 -1.1795 -4.1720 -1.0386 electrodes -8.7089 -1.0109 0.4586 -2.0777 -2,1974 1.3035 -1.5028

n u m b at samples t

-1.2996
-2.2864

Fig: Matrix of data To record the electrical activity, we sample the mental activity using 16 electrodes and store them in a matrix (16xNumber of measurements, Figure
4). Each column contains the 16 values collected in the electrodes at a certain

time t. For the analysis of the values, Matlab will be employed, having an important strength in mathematical analysis.

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

10

3. BCI Architecture

To make the computer understand what the brain intends to communicate necessitates monitoring the brain activity. Among the'possible brain monitoring methods, the scalp recorded electroencephalogram (EEG) constitutes an adequate alternative because of its good time resolution, relative simplicity and noninvasiveness when compared to other methods such as: functional magnetic resonance imaging, positron emission tomography, magnetoencephalography and electrocorticogram. Furthermore, there is clear evidence that observable changes in EEG result from performing given mental activities. In the following, an EEG based BCI will be simply called a BCI. Current BCls use the following EEG signals:
Event Related Potentials (ERPs)

ERPs are transient signals which are characterized by a voltage deviation in the EEG and are caused by external stimuli or cognitive processes triggered by external events. When the user pays attention to a particular stimulus, presented by the BCI an ERP that is time locked with that stimulus appears in her EEG. The changes induced by the ERP in the EEG can be detected by the BCI. Therefore, by focusing her attention to the adequate stimuli, the user can command the BCI. The advantage of an ERP based BCI resides in the fact that little training is necessary for a new user to gain control of the BCI. Nonetheless, the communication is slow since the user must wait for the relevant stimulus presentation.

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

11

Steady State Visual Evoked Responses (SSVERs)

SSVERs are elicited by a visual stimulus that is modulated at a fixed frequency. SSVERs are characterized by an increase in the EEG activity at the stimulus frequency. Through feedback, users learn to voluntarily control their SSVER amplitude, whose variations can be detected by the BCI.
Slow Cortical Potential Shifts (SCPSs)

SCPSs are shifts of cortical voltage, lasting from a few hundred milliseconds up to several seconds. Users can learn to produce slow cortical amplitude shifts in an electrically positive or negative direction for binary control. This skill can be acquired if the users are provided with a feedback on the course of their SCPS production and if they are positively reinforced for correct responses.
Oscillatory sensorimotor activity

The 8-12 Hz and 18-26Hz activities recorded over the motor cortex exhibit noticeable changes during movement, preparation for movement and imagined movement. Indeed, such activities decrease in the hemisphere that is opposite (contralateral) to the movement and increase in the other hemisphere (ipsilateral). The frequency ranges and the magnitude of the changes are user dependent; if trained, a BCI can detect these changes and react according to a previously established protocol.
Spontaneous EEG signals

These signals are recorded during the performance of mental activities other than imagined motor tasks and are not elicited by external stimuli (e.g. mental counting, mental rotation of an object, etc.). The BCI can function with spontaneous signals if the patterns characterizing the corresponding mental activities are learned by the BCI in a training phase.

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

12

A BCI can be defined as a communication system in which the messages or commands that the user sends to the external world do not pass through the brain's normal output pathways of peripheral nerves and muscles. The user communicates with the BCI by performing mental activities (MAS) that the BCI is able to recognize because of previous training. An unknown MA triggers a neutral answer from the BCI; the nature of this answer depends on the context in which the BCI is used. By convention we denote by MAo an unknown MA. BCI consist of three units, namely EEG acquisition, processing and output.

Digitized signals
I

Acquisition

EEGSrgnai Processing
extraction

n
output
User
Feedback

Fig: BCI architecture

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Intetface

13

While the user is performing the MAS that control the BCI, hislher EEG signals are acquired, digitized and sent to a computer where the signals are processed and translated into commands which generate actions in the output unit. These actions that can be noticed by the user constitute a feedback that the user exploits so as to modulate her mental activity to make the BCI accomplish her intents. The correspondence between the patterns present in the EEG, resulting from a mental activity, and the actions generated by the output unit is determined using machine learning techniques. Therefore a training phase, in which the user is asked to perform the MASthat will be used to command the BCI, is carried out. The training phase is composed of several sessions and results in models associated with each MA that serve as reference for their respective MA. The models are built through several training sessions in order to take into account different external (environmental) and internal (user related) conditions which can induce variations in the EEG. The MA models resulting from the training phase are validated in a training with feedback (TRWF) phase in which the user is asked to perform the MAS and a feedback, telling her if the MA she just performed was successfully recognized (positive feedback) or not (negative feedback) is provided. This implies simultaneous learning of both, the user (who adapts her EEG depending on the feedback) and the BCI (who updates its MA models). When the TRWF phase is completed the user can start controlling the BCI
by performing the MAS for which the BCI was trained (application phase). The

time schedule of the application phase is depicted in the following figure.

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

14

EEG tfWS

Fig: Application phase: Time scheduling

The amount of EEG that is analyzed by the BCI to generate an action, the EEG trial duration and the interval between two consecutive actions, the BCI period depend on the type of MAS and on the user. These parameters are determined during the training phase, in which some typical values for these parameters are tested and those values that provide the best results in terms of recognition error are finally selected. The above mentioned process, namely training, training with feedback and application is carried out several times throughout the use of the BCI.

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

15

3.1 EEG Acquisition Unit

EEG signals are measured at the scalp by affixing an array of electrodes positioned according to the 10-20 international system and with reference to digitally linked ears (DLE). DLE referenced voltages are obtained by using the average of voltages at both ear lobes as reference. The ear lobes are selected because they constitute an almost quiet reference, As a matter of fact, they present small influences due to temporal activity.

Fig: International 10-20 system of electrodes-placement. The electrode labels correspond to their position with respect to the brain zones, i.e. Fronto-polar (Fp), Frontal (F), Central (C), Temporal (T), Parietal (P) and Occipital (0). Odd indexes are located in the left hemisphere and even ones in the right hemisphere .If Ve is the voltage at electrode e, VA1 and VA2 the voltages at left and right ear lobes respectively and if we arbitrarily choose VA1 as the physical reference, then the DLE referenced voltage of electrode e is

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

16

3.2 EEG Siqnal Processing Unit

The purpose of signal processing is twofold the MAS(feature extraction). 3.2.1 Preprocessing Unit The preprocessing aims at identifying those EEG trials contaminated by noise. External electromagnetic perturbations, muscular and ocular movements are the main sources of noise. External perturbations are essentially present at high frequencies (above 40 Hz). Since the interesting EEG frequencies are located below this limit, each EEG trial is filtered between 1 and
40 Hz. Muscular movements induce relatively high frequency (above 25 Hz), high

- firstly, maximization of the

signal-to-noise ratio (preprocessing) and secondly, effective characterization of

amplitude components (muscle artifacts) in electrodes located in the motor area

[4]. We set a power threshold (determined during the training phase) on the

frequency band above 25 Hz, in order to detect such artifacts. Ocular artifacts are identified by tracking the abrupt power changes in the signals recorded at prefrontal electrodes.

3.2.2 Feature extraction and classification unit In the context of BCI, the features used for EEG trial classification mainly result from the time, frequency, and time-frequency analysis. Most of the research groups work in the frequency domain and extract the information characterizing mental activities from the nonparametric and parametric spectral representations of EEG. Also, the joint spectral properties of the EEG components are analyzed in for detecting particular emotional states. The

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

17

relationship between the time evolutions of the signals coming from different electrodes serves as an indication of motor activities in. Useful information can also be extracted from particular brain configurations that can be interpreted in terms of brain states. The feature extraction and classification unit determines the commands where Ne is the sent by the user to the BCI. An EEG trial X belongs to2 R~~~~~~~~ number of electrodes and Ttrial is the number of samples in the EEG trial. The rows of X X (1)... X (Ne) contain the signals recorded at different electrodes. The goal of feature extraction is to map the EEG trial space, RNe xTtrial into a feature
' space, R ~(Df is the dimensionality of the feature space) which is suitable for the

discrimination of EEG trials resulting from the performance of different MAS. The mapping function is usually determined using the physiological knowledge of the MAS that are used to control the BCI. Common mappings result from timespace, frequency (parametric and nonparametric) and timefrequency analysis of EEG trials.

3.3 Output Unit The commands from the feature extraction unit are sent to the output subsystem. It then generates a "system answer" that constitutes a feedback to the user who can modulate his mental activities so as to produce those EEG patterns that make the BCI accomplish his intents.

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

18

4.

BCI Functioning
The BCI period is the average time between two consecutive answers and

the EEG trial duration is the duration of EEG that the BCI needs to analyze in order to generate an answer.

Fig: BCI Scheduling

We call "neutral state" when nothing happens (the BCI provides a neutral answer), the "active state" when the BCI executes something, the "neutral EEG
-

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

19

set" as composed of those EEG trials that elicit the neutral answer and the "active EEG setnthe complement of the neutral EEG set. The ideal BCI is a two-state machine whose state changes occur at a rate defined by the BCI period and are determined by a Boolean variable 61 (activation) which becomes true when the BCI detects an element of the active EEG set and false otherwise.

51=o

Neutral

Fig: Ideal BCI

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

20

We propose a less ideal BCI by introducing a transition state so that the BCI cannot switch from the neutral to the active state immediately. The BCI remains in the transition state as long as a second Boolean variable B2 (confirmation) is false.

Fig: Less ideal BCI

We consider a BCI that is commanded by NM MAS, each of them associated with an action. The set of actions that the BCI is able to perform is 'noted as {Ao, All ..., ANMd where A. is the neutral action that corresponds to an unknown MA.

Department of Computer Science & Engineering SNGCE, Kadayimppu

Brain Computer Interface

21

MA generator

*xlm)l

b"'

Fig: BCI Model BCI functioning is modeled as depicted in the figure. Each BCI period the MA generator (which is composed of the user's brain, the acquisition device and the feature extractor) produces a vectorex (m) drawn from a probability density function (PDF) p(x (m)) which depends on the action at time (m - I), A(m conditions) and internal (user's state of mind). The vector x(m) is then processed by the recognition algorithms whose result is the vector L(m) containing NMA scores that measure the likelihood that x(m) corresponds to each MA. The taken action can be noticed by the MA generator who adapts its behavior consequently (feedback). Additionally, the current action A(m) depends on L(m) and the last T actions A(m-1), ...,A(m-T), this dependency is presented in section 4.2. In order to produce the desired actions, the MA generator should adapt its PDF through reinforcement learning. While external factors can be controlled, internal ones are harder to control even by the user. It becomes then necessary

- 1)

and on extra-system factors (Ext) that can be external (noise and environmental

Departmentof Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

22

to introduce a mechanism that allows the recognition algorithms to adapt to the user's current PDF; this mechanism is presented in Section 4.1.

4.1 Dynamic characterization of the PDF associated with an MA

The parameters of recognition algorithms need to be flexible in order to dynamically characterize the PDF associated with each MA under different extrasystem conditions. The characterization process is carried out in several training sessions and it is continuously updated under the assumption that the extrasystem conditions remain constant during each session.
Feature vectar belonging to the MAq class

MAq Cue

,--n\H kmv

0
MA generator

Fig: Training without feedback (open loop) The first characterization of MAq that is denoted by OMAq (0) is obtained after training without feedback session (open loop) in which the user is asked, by a visual cue, to perform the MAq while the produced feature vectors are stored to compute OMAq (0). Subsequent characterizations, OMAq (k) for k > 0, are obtained after training with feedback sessions (closed loop), in which the user is asked to perform MAq and the system uses the last characterization to tell the user if the current MA was successfully identified.

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

23

Feature vector Wonging

*ns(

k)

dwq!ILLl j

Fig: Training with feedback (closed loop) An effective way to integrate the current characterization with the precedent one consists in considering OMAq as a set composed of those vectors that define the boundaries of the PDFq.
4.2 Action generation

The action generator is a state machine whose (NMA + 1) states correspond to the NMA actions and the neutral one. In the ideal scheme the action at time m, depends only on L (m) .Since the recognition of MAS is not perfect; the action generator is not allowed to abruptly change its state. Instead, a dynamic transition mechanism is introduced that depends on the recognition error. As the user performance increases, i.e. the recognition error rate decreases, the mechanism converges to the ideal scheme. Consider that at time m the BCI is executing the action Aql and, according to the ideal scheme criteria A (m + I) A@. The transition towards As = is allowed if, under the ideal scheme criteria, A (m+l) = A (m+2) = ... = A (m+~q2ql) A@ ,otherwise the BCI keeps executing Aql. =

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

24

The value of T is determined such that

"l &

log E

1%

~q2ql

where pq2ql is the probability that a vector x, generated during the performance of MAql was recognized as belonging to the MAQ class and is a constant that characterizes the degree of confidence on the transition from A to A@,e.g. if ,, 99% confidence is required then, the value of is set to 0.01.

One can easily verify that

-q2ql
t

~n4l-Q

1 1

i.e. the smaller the probability

of confusion between MA@and MAq1the shorter the transition duration. Therefore, as the user performance improves, the transition scheme evolves towards the ideal scheme.

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

25

5. BCI Implementation Example

Three types of MAS were used: imagination of left (MAI) or right (MA2) index finger movements and a baseline (MA3) where the subject can imagine anything except MA1 or MA2. The goal was to allow the user to control the movement, to the left or to the right, of a cursor on a computer screen. Our BCI implementation can be explained in terms of three states: the
neutral state in which the BCI whether recognizes MA3 or cannot recognize any

known MA, the active state in which the BCI recognizes MA1 or MA2 and answers with an action and a transition state between the above mentioned states. State changes occur at a rate defined by the BCI period, and are determined by the activation of two Boolean variables: detection (BI) and confirmation (82). 61 is true when MA1 or MA2 are recognized and B2 is true when B1 is true and if the M previously recognized MAS are equal to the currently recognized MA.

Fig. BCI States

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

26

The parameter M (latency time) depends on the rate of false positive recognitions, although it would not be larger than two seconds. The set of BCI parameters are summarized in the following table. BCI period EEG trial duration Classification parameters Latency time

The optimal values for these parameters are determined during several training sessions and continuously updated.

- -

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

27

6. Factors affecting BCI

The achievement of a successful BCI system depends on design factors as well as on the user who should acquire a new skill consisting in controlling her EEG in order to interact with the system. Therefore, human factors such as fatigue, stress or boredom are of great influence. Effective BCls adapt to each user on three levels.
-

First, when a new user accesses the BCI, it adapts to that user's signal characteristics.

Second, periodic adjustments are necessary to reduce the impact of EEG variations that results from different user's mind states.

Third, the adaptive capacity of the brain is engaged in the sense that, through feedback the brain activity will modify so as to produce those EEG patterns that best control the BCI.

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

28

7. Key issues in the development & application of BCI

Recognition that BCI research and development is an interdisciplinary endeavor, involving neuroscience, psychology, engineering, mathematics, computer science, and clinical rehabilitation. ldentification of those signal features that users are best able to control. The extent to which feature control can be independent of neuromuscular control. Development of behavioral methods for helping users to gain and maintain feature control. Development of signal acquisition and processing methods for deriving the features. Selection of the best algorithms for translating the features into device commands. Identification and elimination of artifacts. Adoption of standard methods for evaluating short-term and long-term BCI performance. Selection of appropriate applications and appropriate matching of applications and users. Department of Computer Science & Engineering SNGCE, Kadayiruppu

Brain Computer Interface

29

8. Practical BCls
.There has been great success in using cochlear implants (Cochlear implants are hearing devices that can help people with certain kinds of hearing impairment or who are entirely deaf.)in humans as a treatment for non-congenital deafness. There is also promising research in vision science indicating retinal implants may some day prove to be similarly successful in treating noncongenital blindness. Simple brain-computer interfaces already exist in the form of

neuroprosthetics, with a great deal of neuroscience, robotics, and computer science research currently dedicated to furthering these technologies. Recent achievements demonstrate that it is currently possible to implement brain-computer interfaces that allow neuronal clusters to directly control computers. Current BCls are mainly designed for people with severe motor disabilities in order to provide them with new ways of communication which do not depend on the brain's normal output channels of peripheral nerves and muscles. Present-day BCls provide maximum information transfer rates up to 25bitslmin. Furthermore, BCls can be used as a complement to other HCI devices to enrich the interaction between humans and computers.
Department of Computer Science & Engineering SNGCE, Kadayiruppu

Brain Computer Interface

30

9. Conclusion

Over the past decade, productive BCI research programs have arisen. Facilitated and encouraged by new understanding of brain function, by the advent of powerful low-cost computer equipment, and by growing recognition of the needs and potentials of people with disabilities, these programs concentrate on developing new communication & control technology for those with severe neuro - muscular disorders. The immediate goal is to provide these users, who may be completely paralyzed or "locked in," with basic communication capabilities so that they can express their wishes to caregivers, operate simple word processing programs etc. Future progress depends on attention to the key issues listed previously. With adequate recognition and effective engagement of these issues, BCI systems could provide an important new communication and control option for those with motor disabilities. They might also give to those without disabilities, a supplementary control channel or a control channel useful in special circumstances.

Department of Computer Science & Engineering

SNGCE, Kadayiruppu

Brain Computer Interface

31

9. References
http://www.cs.man.ac.uk/-toby/writing/PCW/bci.html http://en.wikipedia.orglwikilBrain-computer interface
http://ltsl pcl9.epfl.ch/repositorv/Garcia2004 1464.pdf http://ltsl pcl9.epfl.ch/re~ositorv/Garcia2003 438.pdf http://ltsl pcl9.epfl.ch/repositon//Garcia2003 436.pdf Many more on the web ................................................

Department of Computer Science &'ngineering

SNGCE, Kadayinrppu