The Viability of Brain-Computer Interface Controlled Prosthetics

moanafternoonInternet και Εφαρμογές Web

11 Δεκ 2013 (πριν από 3 χρόνια και 6 μήνες)

163 εμφανίσεις

Session B10

3024



University of Pittsburgh Swanson School of Engineering
1


Mar. 7, 2013


The Viability of
Brain
-
Computer Interface Controlled
Prosthetics


Alexander Burval (
akb53@pitt.edu
, Bon 6:00), Zachary Wool (
zrw4@pitt.edu
, Budny 4:00)




Abstract
-

This paper will discuss and report the scientific
advances that led to Brain
-
Computer Interface (BCI)
technology and its social impacts. BCI technology allows for
neurological signals to be translated into computer signals.
We will focus on the use of it in the translation of brain
signals into mechanical motion, specifically, the control of a
robotic pros
thetic arm. Patients visualize moving their arm
or hand, then these thought patterns are interpreted by the
system and then returned as mechanical comma
nds which
move the prosthetic [1
]. The end goal of this research is to
develop a BCI controlled prosthet
ic for public use.


We will report on the basic history of BCI and its
components as well as the issues that are keeping it from
flourishing. Currently, there are drawbacks

as well as
ethical issues surrounding
BCI which prevent practical
everyday use. As

the goal is to have this technology
commonly available, these issues are the forefront of BCI
research

and a large part of our discussion
.



Key Words: Brain
-
Computer Interface, Prosthetics,
Robotics, Rehabilitation,
Bioengineering, Neurological
Signals


A
N
I
NTRODUCTION TO
B
RAI
N
-
C
OMPUTER
I
NTERFACE

IN
P
ROSTHETICS


A Brain
-
Computer Interface (BCI)

is a system developed to
take

input from a user’s brain and output
commands that
control

a device.

The point of BCI technology is to allow
device control via thin
king. This involves the translation of
human thought patterns into computerized code. Then that
code is

used as output commands for whatever the users
want their device to do.

This paper will discuss the
applications
BCIs

have within the motor control of
p
rosthetics.


BCIs started out
with
a
simple muscle
-
based control

design and
then
moved on to pure thought
-
based control of
devices.
Recent
breakthroughs
within human trials have
encouraged the scientific community. Different designs and
component
combinations

of BCI systems have been

proposed
to try and reach the end goal of BCI prosthetics:
giving independence to those hampered
by physical defects
or injury. Although BCIs have been around for a
several
decades
,

there is limited research in the are
as of real world
application.


BCI shows
promise in a lab setting;
however aspects of
the technology inhibit the capabilities of the overall system.

The need
for

skilled technician
s

and

lack of real wo
rld trials
are

a couple of the

issues that must be answered
before
taking BCIs into the everyday world.


With human based clinical trials becoming the majority
of BCI testing, certain ethics must
also
be discussed so that
no harm comes to the volunteers. Issues of ethics also arise
wit
hin the potential of the everyday person having a BCI
within their own control.


Despite the need for improvements, BCI technology
offers the best chance of giving independence to those
who
lack

the ability to
control

their limbs.



D
EVELOPMENT OF
BCI

T
ECHNOLOGY


BCIs began with visual evoked potentials, that is, simple
sight
-
based recognition and thought
.
These early designs

focused more on muscle co
ntrol than pure brain signals [2
].
While this achievement proved the theory behind BCI, it did
not
add to
the overall
BCI research

because it was not a
brain
-
based control system
.


An e
lectroencephalograph

(EEG) based device was soon
used

by
Larry

Farwell and Emmanuel Donchin

to allow
users to type out words
. Focused attention on a specific
location
allowed the user to
pick a letter out of a matrix [2
].
This attention
-
based, rather than muscle
-
based, system
represented a significant advance as the system moved away
from muscle dependence. Users soon learned to use
sensorimot
or rhythms, activity record
ed from scalp locations
associated with the motor cortex,

to move a c
omputer cursor
across a

screen. Rapid switching of the user’s intended target
required a bidirectional thought process that had yet to be
included in tests before. These studies were repl
icated and
extended to multidimensional

control by subseq
uent
researchers [2
].


More recently,
there has been
investigation into single
-
neuron activity detection from implants within a subcranial
cortex. These studies have been, for the most part,
performe
d

on non
-
human primates; however
there have been
trials o
n human users.
Certain important issues regarding
long term implants in humans have yet to be resolved
be this
research
[2
].

These issues will be discussed within
Limitations of BCI Technology
.


Adaptive or “zero
-
training” systems have been proposed
by Klaus
-
Robert Muller and Benjamin Blankertz
that would
allow both the machine and the user to adapt to perform the
overall goal or task.
This plan would be
easier on a user new
to BCIs
;

however, unle
ss this

design can prove to be
more
efficient and beneficial there
is no cause
for such a design
[2
].

Alexander Burval

Zachary Wool



University of Pittsburgh Swanson School of Engineering
2


March 7, 2013



BCIs have

a significant history
and each new design is
working to overcome the limitations of previous designs.

Improvements may have been slow from our
point of view,
but there is hope of much to come as research continues.


E
XPERIMENTAL
P
ATIENT
D
ATA


Initially, the majority of BCI experimental trials were
performed on non
-
human primates. Only within the last
twenty years have trials been predominantly hu
man based.
The primates showed much more success in their trials than
the first human trials. Jennifer Collinger was one of the first
patients within human trials to perform more successfully
than her primate counter parts.



Collinger
, see figure 1,

perfo
rmed a range of activities
with a BCI implant
that she would in her day
-
to
-
day life.

As
early as the second day, she was able to manipulate the
prosthetic within a three dimensional workspace. Her
progress surprised the scientists when she completed tasks
in
days though the
y thought it would take weeks [3]
.


FIGURE
-
1


Jennifer Collinger operating a BCI controlled arm

[5]


For someone using a BCI, performing everyday tasks
requires learning new skills in order to operate the system.
Cathy Hutchinson, a stroke victim, learned to operate a BCI
controlled prosthetic arm. She performed many everyday
tasks, including bringing a b
ottle
to her mouth to drink
coffee [4]
. Ms. Scheuermann, who suffers from a
degeneration of the connections between her brain and her
muscles,
was implanted with an electrode grid within the
part of her brain controlled her right arm and hand
movement. She

was able to feed herself chocolate within a
year of BCI experience. Her progress was a great leap
forward in BCI human trials
[5]
.


Each of these patients’ stories shows how BCIs are
allowing other
-
dependent users to perform many tasks once
considered imp
ossible on their own. Their successes give
rejuvenation to the studies and hope for future research.


C
OMPONENTS OF A

BCI


Similarities between different BCI designs allow it to be
broken down into four main
components
: Signal
Acquisition, Feature Extrac
tio
n, Feature Translation, and
Device

Output.


Signal Acquisition


Signal Acquisition is the measurement of the user’s brain
sign
als using a particular sensor
[6]
. There are both invasive
and noninvasive sensors that are in use. N
oninv
asive sensors
are
primarily used with

e
lectroencephalography

(EEG)
.
Invasive sensors must be implanted within
or on the surface
of t
he brain its
elf to record

signals
.

Each will be

discussed
in
following sections.


Feature Extraction


Features, being signals produced by the user that the BCI
can use to interpret intent, must
be extracted from the entire
signal spectrum gathered through signal acquisition.
The
extraction process becomes difficult when features cannot be
distinguished fro
m other signals within the spectrum.
The
signals must be analyzed so that these features can be
interpreted effectively.

The

original

“fundame
ntal problem” in real time BCI
was that there wa
s not enough data received to do proper
data analysis. This proble
m can be circumvented given
enough computing power via nested cross validation. All
data must be
processed through this validation sequence so
that there is

limited biased data
[7]
.

Biased data comes from
conceptually wrong procedure
s (these skewed concept
s are
based off of the assumption that some adaptive signals are
universally constant)
which lead to severe underestimation
of the generalization error,

an estimation based on the data
.




There are

two

main

steps to the extraction of features
:
filtering
and

classifying. Filtering is done in either spectral
or spatial filtering. Spectral
filtering
limits the frequencies of
the brain signals that are processed based on the desired
response

[7]
.

The frequencies extracted via spectral filtering

of invasive BC
Is are dominated by oscillati
ons at
frequencies in the alpha

and beta bands. The amplitude of
such oscillations diminishes when subjects imagine
movement. The opposite effect occurs for frequencies in the
gamma range. Based on these fluctuations scientists

can
determine which frequencies to omit from data processing.
Issues arise in attempting to eliminate

false positive signals.
When a

task is being focused on, real positives can be
distinguished from false positives, however, “in the absence
of a control
task the brain can spontaneously
generate
signals…which mimic those of volitionally produced
signals, yielding false alarms”

[8]
.

Alexander Burval

Zachary Wool



University of Pittsburgh Swanson School of Engineering
3


March 7, 2013


Spatial filtering focuses data collection based on the
probable location in the brain of where the desired response
should com
e from. The goal of classifying data is to find an
equation that provides a good estimate of the probable
distribution of data based on the type of signal

[7]
.

The
problem with this is that probability distributions are not
given, thus they must be estimat
ed via one of many methods
(see Dornhege, 2007).


Fe
ature Translation


Feature translation must

be dynamic in order to
accommodate and adapt to spontaneous or learned changes
in the signal features. This ensures that the user’s possible
range of feature values covers the
full range of device
controls
[6]
.

Extracted frequencies are sent through a co
mputer
system that translates them into device output. This process,
if done signal by signal, would be almost impossible. Thus,
BCI scientists create a library of position
s and movements
for the device
.
This database takes recorded signals from a
base sub
ject who performs specific movement based on
time. These signals are taken from the firing rate of a
specific population of neurons in the primary motor cortex.
This activity of the cells is related in vector form to the
kinematics representing the limb

ti
me based movements
.
[7]
.


According to Dornhege

(et al.)
,

“To model what aspects
of movement
s

are represented (encoded) by the brain, we
adopt a probabilistic approach and learn a generative model
of neural activity”.
The goal is to find a function that
relates
the prosthetic kinematics

at a specific time with
corresponding neural firing rates.

There are four components
to the generative model
.

First is the type of data, that is,
spikes, rates and local field potentials. Second, important
variable behavio
rs like joint
-
angles, torques, muscle
activation, hand direction and action. Third is the functional
relationship between the behavior and neural activity, for
example linear, exponential, etc. Finally, the model of the
“noise” within the signals (“noise”

can arise from any
number of inadequacies of the model)

[7]
.


Device Output





Once the signals are translated
,

they are then
used as direct
commands with
i
n the device intended for use
[6]
. Initially
BCI’s were used for

directing computer cursors
or spelling
words, but now
devices
such
as robotic prosthetics and
virtua
l wheelchairs have been added [2
].


While BCIs have been limited to purely health
-
based
need, once commercialized, they have the potential to be
used outside of medical need. That is
, a BCI could be used
to control various objects within the home, such as television
or kitch
en tools. Also, computer access

could be modified to
be purely controlled by a person’s brain. Already BCI
-
based
control of in home devices has been
applied

with g
reat
success as it gave the patients independence without their
caregivers
[9]
.
At some point, this technology will be able to
affect the lives of people without handicaps as an everyday
device.
Commercial applications such as these wi
ll promote

investment
s in the research from non
-
medical based
institutions.


FIGURE 2



A basic flow chart of BCI components and their path of
operation

[2
]


Flow of Inter
-
System Communication


While each

component discussed above is
an essential part
of the BCI system, the pathway between
all four components
is what allows the entire syst
em to function. Figure 2
(above)

shows the pathway of this process of turning brain
signals into output commands.


First signals are taken from the
brain with a
signal
scanner

(such as an EEG) that reads the signals and then
stores them inside the computer.
The signals are then
separated to determine the user’s intent from the range of
signals acquired.
Once the
se

signals are acquired
, features
are ex
tracted and
then translated using dynamic programs
which

turn these features into commands that the output
device can understand. The device output could be anything
from a computer cursor to a robotic prosthetic.
This process
is repeated
continuously

as the user
goes about
his
tasks.


Through the use of this process
, the user can

send
commands
,

through the system,
to a device
,

such as a robotic
arm
,

by thinking about what

they wish to happen.


N
ON
IN
VASIVE
V
ERSUS

I
NVASIVE
D
ESIGNS


Noninvasive Design



As its name su
ggests, noninvasive BCIs minim
ize the direct
contact of the BCI recording device to the brain.
This
increases the ease of implementing the device. While this
Alexander Burval

Zachary Wool



University of Pittsburgh Swanson School of Engineering
4


March 7, 2013


decreases the accuracy of
gathering signals, it
prevents much
of
t
he patient risk fr
om surgery
[6]
.


One of the most common noninvasive BCI
designs
is the
scalp
-
recorded EEG. An EEG sums the action potentials of
large groups of neurons
which fire

at close to the same time
[10]
.
Placement on the scalp, although beneficial to the
patient,
cau
ses
certain

issues. The main issue

is the
contamination of recorded signals by
non
-
brain activity [2
].
That is, the user’s intent may be mixed with other signals
within the brain which then cannot be
separately

read
by the
EEG.



EEG has

had
success in
controlling two or three
dimensional motion, but some specific
limitations have
prevented the

use
of EEG as the main BCI design
. Current
EEG
-
based
electrodes

requ
ire a certain level of skill
to
position as well as

perio
dic maintenance to ensure
su
fficient

contact to the skin
.
More stable and convenient electrodes
are being developed as an answer to these limitations [2].


EEG is not the only noninvasive design that was
successful. E
lectrooculography

(EOG) measures the resting
potential of the retina
to control devices
[11]

rather than
picking up brainwaves from the scalp
.

A hybrid of both EEG
and EOG will be discussed in the next section.


Hybrid Proposal



A hybrid EEG

EOG system was proposed that combined
attributes of both EEG and EOG into a single

system. The
patients within the trials used their eyes to control several
BCI operated tools (com
puter cursor, robot arm, etc.). A
negative EOG sensor was place on the left side of the left
eye and a positive is place on the right side of the right eye.
These two sensors allowed a detection of right and left
commands. Positive and negative EEG electrodes were place
on the scalp to detect alpha and beta wave activity. “Forward
movement [was] detected when the beta activity [was]
greater than the alpha acti
vity i.e. the subjects [needed] to
think about something.” This proposed hybrid system can
lead to approximately 96% classification accuracy, which is
higher than the approximately 86% accuracy of recently
developed BCI a
lgorithms with synchronized
results
.
Additionally, it only employs two bioamplifiers and five
electrodes, combined with a short training time of around 20
minutes, which allows users with little experience to use the
system. Due to these advantages, this system appears more
practical in eve
ryday life than current EEG or EOG systems
[11]
.


Invasive

Design


BCIs started out as simple EEG
-
based designs, and despite
limitations having arisen since then, their use has resulted in
many successes. However, rather than working to fix these
limitatio
ns, some scientists have moved on to invasive
designs.

Invasive
designs
require
more direct contact to the brain
,

as these
designs try to maximize

prox
imity to the neurons
.

Generally, a grid of electrodes is surgically implanted either
on or within the br
ain.
With surgery
come

many health risk
s

from infection and unpredicted reactions to the implant.


One of the main reasons behind the development of
invasive BCIs

wa
s the belief that
only invasive BCIs would

be capable of real
-
time
,

multidimensional control of a
robotic arm or
other
neuro
prosthetic

[2
]. Although, t
his
belief has been disproved by non
-
invasive BCI experimental
results
, invasive designs have their own advantages. Direct
implantatio
ns may be cumbersome now, but they

all
ow for a
more permanent connection to the BCI s
ystem. Additionally,
the proximit
y to the brain itself helps to limit false alarms
that would otherwise hinder the output of the BCI

better than
a non
-
invasive system
.



Invasive designs

have
issues besides ri
sks from surgery.
As the brain is a soft and pliable organ, it moves and
accelerates within the brain cavity. When stiff, non
-
moving
electrodes are anchored to the brain, there is the potential for
harm to come to both the electrodes and the brain itself.
To
counteract this issue, electrodes are being developed that sit
on top of the brain and can move
along with the brain’s

motion. In addition, c
ontinued presence of
electrodes
within
the brain cortex increases the possibility of

cell death
within
the surro
unding tissue
[12]
.
This provides real challenges
for long term use of these designs.


Like
their
non
-
invasi
ve counterparts
, invasive designs
have their drawbacks that prevent
their application in real
life. Both
have proven very successful in their
performances
withi
n trials. Each design has
its

own advantages

and
disadvantages
.




M
OVEMENT
C
ONTROL


The output of BCI algorithms can be applied to many
different kinds of devices. In th
e area of prosthetics
there are
two ways BCI output
s motion control
:


kinematic control
and goal selection.


Kinematic control

requires an individual command for
every degree of freedom a prosthetic has
, that is, every axis
of rotation it has
. The BCI must
return these commands from
the user as

movements
in real time.
Currently, robotic arms
have seven or more degrees of freedom (compared to a
human arm having upwards of twenty). As robotics
technology continues to progress
,

robotic
prosthetics

will
increase in

degrees of freedom. Since kinematic control
requires an ind
ividual command for every degree of
freedom,

kinem
atic control
becomes

increasingly

demanding
upon the BCI system
. However, this kind

of
movement
command allows for
more flexibility in a wider variety of
circumstances.


While kinematic control requires a
command for every
individual m
ovement, goal selection
only outputs the
final
location or placement from the user. H
ardware and software
Alexander Burval

Zachary Wool



University of Pittsburgh Swanson School of Engineering
5


March 7, 2013


handle

achieving that outcome

in the most efficient way
possible
. This control method is much less demanding of the
system
.

This is
because it requires fewer input commands
from the user in order to control the movement which
simplifies the computing process.
Thus, g
oal selection

wo
uld
be able to convey

c
ommands, such as “making coffee

, that
are simpler from the user’s

side.

On the other hand, goal
-
based commands
require

a constant update
to the system of
its

surroundings in order to properly perform task
s

which
will increase the necessary computing power
.
In order to
update the system, pattern recognition software is u
sed.


Traditional pattern recognition claims “that all available
information is in the training set.” Current BCIs use the
process called
biomimetic pattern recognition (
BPR
)

which
leans away from this idea. BPR claims that there can be a
prior knowledge i
n the relationship between signals through
the Principle of Homology
-
Continuity. This principle, in its
most basic form, claims, that if two things share a specific
trait, then what
separates

them from each other must be
changed gradually. Through this the
y can be considered part
of one similar large group. Under this basis, BPR relates
similar physical structures into samples of the same class
creating separate
environmental structures

[13]
.


Control requirements for devices vary greatly between
simple
forward
-
backward motion and motion in three
-
dimensional space. Both kinematic and goal selection are
then

used in a variety of different

devices for different
output
. A combination of both kinematic and goal selection
commanding is
also
feasible

[2
]
.


L
IM
ITATIONS OF
BCI

T
ECHNOLOGY


Lack of Specific Research


Much of BCI research has been done on non
-
human
primates and only recently have human trials been the
majority of clinical
trials.
More research in the are
a of
human trails would
m
ove BCI

closer to

an everyday reality.


A significant portion of BCI research has been done to
perfect the performance of translation software.
Original
testing was done in

real time. However
,

much of the
following studies have use
d

pre
-
recorded signals to test their
progr
ams. These offline tests, while beneficial to the
translation program
,
do

not help reach the end goal of BCI
prosthetics. That is, BCIs must run in real
-
time to provide
r
eal time feedback to the user [2
].


Technician Necessity


Brain
-
computer interfaces have yet to be able to
automatically read a

user’
s mind
.
Instead, a process of
twenty to thirty minutes must take place where the user
“tunes” his mind with the BCI system.
This “tuning” syncs
the user
with the

BCI
system so that
the machine can
more
easily
interpret

the user’s brain signals in real time.
The

syncing

process also allows the BCI to separate

the user’s
intent from false positive readings.
This process requires a
skilled technician, who is also needed onsite during th
e
testing to maintain the system

[8]
.


This dependence on a technician prevents a BCI
-
abled
patient to be independent

in their everyday life. This human
assistant is impractical
in

real life application and thus the
translation algorithms must be improved
so that they
function either without the tuning or only an initial proc
ess
with their permanent user [3]
.


Error Analysis


BCIs have a poor track record for reliability and
performance
. While translation algorithms a
re improving,
issues have surfaced

in
deriving the user’s intent from other
random “noise” in the brain.

Non
-
invasive BCIs have a more
difficult time
than invasive sensors due to

their lack of
proximity

to the brain.
[10]
.
Signals become mixed as pure
thoughts (both conscious and unconscious)
are mixed with
intended motor signals.


Controlled
Environment


Most research into BCI has been done inside closed
laboratory environments. Usually the patient is comfortable
and at ease during their trials.
This does not depict real
world situations where

everyday events can be
unpredictable. Thus,

research into the emotional and
stressful situations
needs to be done in order

to predict the
difference in BCI

computing in such situations
[14]
.


F
UTURE
BCI


Due to limitations of technology, BCI
currently
ca
nnot reach
its
end goal;

as
the efficiency of systems improve

and o
ur
understanding of the brain

increases
, BCIs will
continue to
improve.

Already, scientists have started to propose
solutions to these limitations. The EEG/EOG hybrid,
discussed above, helps to improve both the training time and
the classification accuracy of the user’s intentions. Error
analysis is being improved to try and s
ynchronize the BCI
with the user’s intentions
[11]
. Others have proposed ideas
such as wireless communication and the eventual perfection
of the BCI’s translational algorithms.


Although much of the discussion above
has been related
to how BCI applies to
paraplegics

(
and others who suffe
r
from lack of muscular control),

the technology has

other
commercial

(discussed above) and medical applications
.
As
patients used
BCI
s

in

their trials
,

it was discovered that the
ir

neural
-
muscular connections
gre
w in
strength over time [3].
This “mental” physical therapy would aid people that suffer
from

non
-
permanent

loss of muscular control
, such as stroke
victims,

to quicken their recovery.
Using a BCI to improve
recovery will allow the patient to maintain mental
Alexander Burval

Zachary Wool



University of Pittsburgh Swanson School of Engineering
6


March 7, 2013


co
nnections to their muscles while their muscles

are
unable
to be used. This will speed up the rehabilitation process
drastically.


Presently, t
hese limitations of BCI prevent its ev
eryday
use;

however,
they do not
halt

BCI researchers’

strives

to
overcome

such obstacles.
Human trials have demonstrated
that BCIs are worth pursuing and scientists continue to strive
to make what is now an initial hope into an eventual reality.


E
THICAL
I
SSUES

OF
BCI



Health Concerns


BCI
technology has potential benefits,

but as of now it
s full
effects on the subject
have not yet been fully explored
[14]
.
Con
c
erns arise

in the areas of personal identity, mental
workload, and surgery risk within research.

Thus, if such
risks outweigh the overall benefits of the project, the patient
testing of BCI systems must
be
discontinue
d until

those risks
can be reduced or avoided.



Modifying the human brain with artificial components
has the potential to
alter the per
sonality
as well
[15]
. While
these personality changes may not be significant to everyday
life, they do change the patient into something they were not
before.
If BCI

technology’s

intent is to improve physical
capabilities
,

it is important to weigh the pos
sible alterations
within the mind as well.


As BCI requires additional

focus

from

the user

(compared to an average person)
, their mental workload
from performing tasks
must be taken into account
.
All trials
have occurred

within emotionally neutral, stress
-
free
situations, which are not comparative to actual experiences

[15]
.
Daily exposure to stressful or mentally demanding
labors could affect the user’s mental health.
If the mental
health of a patient is in severe dange
r, then the use of BCI
should be ceased. That is, if their daily lives are impacted
with
excessive stress or

other harmful emotions from
extended BCI use,

the benefits of the system do

not
outwe
igh these

negative

psychological effects
.


In the case of inv
asive designs,
s
urgery always has risks
due to human error or unpredictable circumstances. A test
subject must consider such risks before being given the
implant needed for experimental testing.
Thus it is not
personally beneficial to people who do not suf
fer from
paralysis or other muscle inhibitions to go through
experimentation.
S
ince the s
urgical implant of the
electrode
is
invasive
, a solution is to use patients already equipped
with

an implant for research purposes
.
However,
the reason
these patients
have an implant is due to their own medical
issues (often times epilepsy). Su
ch conditions would not
improve from this

BCI research and thus neither would the
patient.

The question arises as to whether to
put those
patients through such

tasks when they hav
e n
othing to gain
from the overall

research.


The risk of surgery in long term patients would not be as
drastic if the implant lasted a lifetime. However, as it stands,
the chip has yet to be tested over a lifetime’s use.

A
s of now
,

signals have been obtai
ned after

a thousand days of testing
since implantation

[6]
.

While this length of time is nowhere
near ideal, it provides hope to the reality of a lifetime BCI
implant

for future models
.



BCI research focuses upon benefiting those suffering
from
physical disabilities. In order to improve the
technology, experiments must be performed on humans

directly
. As such, personal health risks
,

as stated above
,

must
be taken into consideration before such experiments are
performed.


Responsibility

of BCI Owners


A tool
can be manufactured for a task by a company, but it is
put to use by an individual. If harm or damage comes to
someone or something through
the use of that

tool
, it must
be asked

as to whether the person is at fault th
rough use or
th
e company from

a malfunction

in the design
.

Since motor
control within a BCI is controlled through electrical signals
generated by a computer, it may be difficult for the real
agent behind such actions to be determined

[15]
.


BCI actions are controlled

through com
puter decoded
algorithms
from the
user’s brain signals. A

possible
program
fault

could cause a physic
al action that was
not intended by
the user
. If this were to happen
,

the manufacturer would be
considered at fault

because the issue would lie
within the
program i
tself
.

The real question arises as to whether the
prosthetic is
considered

a part of the person or a simple tool
used by the person. If
the
prosthetic
is a part of the user, then
the user would be
responsible

for it just

as
they are
responsible for their own limbs

[15]
.


It would be unrealistic to require a BCI to work
efficie
ntly without a single error;

even the most
sophisticated device might malfunction
due to an

unpredictable cause. As such, precautions must be taken to
prevent h
arm
caused by

such errors.

If no errors within the
device can be found, then the user who uses the tool must
then be at fault for their personal actions.


Sportsmanship


As BCI prosthet
ics continue to progress

in their real world
application, in theory they would give more independence to
those
who

have lost

limb
s. Specifically, people
who gain this
new replacement limb may wish to compete in ph
ysical
competitions
. The issue arises as to wheth
er or not it is fa
ir
for a person aided by an artificial limb
to compete against
an
able
-
bodied athlete
.


In a study

done by International Association of Athletic
Federations,

the
Cheetah
TM
,
a
non
-
BCI
re
placement leg for
sprinting, it was found that it

cost the sprinter

wi
th the
prosthetic

less energy to run the same length as an able
-
bodied sprinter
[16]
. Fully trained athletes may be at the
Alexander Burval

Zachary Wool



University of Pittsburgh Swanson School of Engineering
7


March 7, 2013


peak of human training, but BCI
-
aided athletes have the
potential to
be
just as good with less effort because of their
prosthetics. W
hile, neuroprosthetics allow an impaired
patient to do as much as their non
-
impaired counterpart, as
technology progresses, some day they might be able to do
more.


Performance aiding drugs and suits are already being
outlawed b
y professional comp
et
it
ions.

Soon those
professionals must also decide how to approach the idea
of
the formerly handicapped out
-
performing their able bodied
opponent
s
.


T
HE
O
VERALL
V
IABILITY OF
BCI

T
ECHNOLOGY


It is important to remember that
BCI technology, while
showin
g notable improvements within
lab setting
s
,

still has a
long way to go before becoming a real world prosthetic
device.
Over the past few decades, BCIs have developed
from a simple experiment
in
to a feasible solution to

solving

t
he complete dependence
of th
ose suffering from paraplegia
,

or other issues
,

on others
.

While BCI does have hope in the future, many realistic
issues must be solved before real world applications can be
reached.
Translation
al algorithms m
ust be perfected to
ensure that mechanical
output is in line with the user’s
intention.

BCI systems

can be unreliable because of this. In
addition, the training time that is needed to sync the user
with the BCI requires a skilled operator who cannot be there
in everyday situations
.

Additiona
lly, e
thical concerns must be addressed during
this technological pursuit.
The r
isks of volunteers within the
trials must be conside
red as well as their personal benefit
from the trials. If

BCI prosthetic
s

were to become a reality,
the idea of responsibility mus
t be discussed. That is, who is
responsible for actions don
e with

a prosthetic
,
the

user or the
manufacturer? As the line between BCI prosthetics and
human limbs diminishes,
th
ese questions must be answered
before such situations cause harm to people.

Alth
ough, BCIs have much to gain from the expansion of
their technology, their goal to give independence to the
handicapped is a worthwhile one.
Regardless of the ethical
and design concerns, this is a topic of research that should
continue to be pursued.

Onc
e complete,
BCI technology

will

gi
ve freedom to many who would never have
independent mobility otherwise.



R
EFERENCES


[1
]
Business Insights: Global, Biotech Week
. (May 30, 2012). “Paralyzed
individuals use thought
-
controlled robotic arm to reach and grasp.”
http://bi.galegroup.com/global/article/G
ALE|A291015822/35500084df8cc9
e8aab5388aea518b10?u=upitt_main

[2] D. J. McFarland, J. R. Wolpaw. (2010). “Brain
-
Computer Interfaces for
the Operation of Robotic and Prosthetic Devices.”
Science Direct.

(Online
Article).
http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=4640662&tag=1

[3]
G. Courtine, S. Micera, J. DiGiovanna, J. del R Millan. (December 17,
2012). “Brain
-
machine interface: c
loser to therapeutic reality?”
The Lancet
.
(Online Article).
http://www.thelancet.com/journals/lancet/article/PIIS0140
-
6736%2812%2962164
-
3/fulltext

[4]
R
. T. Gonzalez. (May 2012). “Watch This Paralyzed Woman Control a
Robotic Arm Using Only Her Mind.”
Io9.com.

(Online Article).
http://io9.com/5910859/watch
-
this
-
paralyzed
-
woman
-
control
-
a
-
robotic
-
arm
-
using
-
only
-
her
-
mind

[5]
UPMC/University of Pittsburgh Schools of Health Sciences. (December
16, 2012). “Woman with Quadriplegia Feeds Herself Chocolate Using
Mind
-
Controlled Robot Arm

in Pitt/UPMC Study.” (Online Article).
http://upmc.com/media/NewsReleases/2012/Pages/bci
-
press
-
release
-
chocolate.aspx

[6]

D. J. Krudienki, J. J.Shih, and J. R.
Wolpaw. (Mar. 2012).

“Brain
-
Computer Interfaces in Medicine.” Mayo Clinic Proceedings. 87.3.
p268. (Online Report).

http:
//go.galegroup.com/ps/i.do?action=interpret&id=GALE|A284565321&
v=2.1&u=upitt_main&it=r&p=AONE&sw=w&authCount=1

[7]

Dornhege, Millán,Hinterberger, et. Al. (2007). “Toward Brain
-
Computer Interfacing.” (Print Book). pp 146
-
277

[8]

A. Torres Valderrama, P. Pac
lik, et al. (2012). “Error probability of
intracranial brain computer interfaces under non
-
task elicited brain states.”
Science Direct
. (Online Article).

http://www.sciencedirect.com/science/article/pii/S138824571200404X


[9] J. N. Mak, J. R. Wolpaw. (2009). “Clinical Applications of Brain
-
Computer Interfaces: Current State and Future Prospects.” National Institute
of Health. (Online Article).
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2862632/

[10]

I. Daly, S.J. Nasuto, K. Warwick. (May 30, 2001). “Brain computer
interface control via functional connectivity dynamics.”
Science Direct.
(On
line Article).

http://www.sciencedirect.com/science/article/pii/S0031320311002032

[11]

Y. Punsawad. Y. Wongsawat, and M. Parnichkun. (2010). “Hybrid
EEG
-
EOG Brain
-
Computer
Interface System for Practical Machine
Control.” 32
nd

Annual International Conference of the IEEE EMBS. Buenos
Aires, Argentina. (Online Report).
http://ieeexplore.ieee.org/sta
mp/stamp.jsp?tp=&arnumber=5626745

[12]

D. Moran. (2010). Evolution of brain
-
computer interface action
potentials, local field potentials and electrocorticograms.”
Science

Direct.
(Online Article).

http://www.sciencedirect.com/science/article/pii/S0959438810001789

[13]

Y. Wu, Y. Ge. (2012). “A novel method for motor imagery EEG
adaptive classification based biomimetic pattern recognition.” (Online
Article).
http://www.sciencedirect.com/science/article/pii/S0925231212007266

[14]

E. A. Felton, J. C. Williams, G.C., Vanderheiden, et. al. (2012).
“Mental Workload During

Brain
-
Computer Interface Training.” Department
of Neurology, School of Medicine, Johns Hopkins University, Baltimore,
MD,

USA. Department of Biomedical Engineering, University of
Wisconsin
-
Madison, Madison, WI, USA. (Online Report).
http://www.tandfonline.com/doi/pdf/10.1080/00140139.2012.662526

[15]

J. Clausen. (Nov. 2008). “Moving Minds: Ethical Aspects of Neural
Motor Prostheses.” Department of Medical Ethics, University of
Tubingen,
Tubingen, Germany. (Online Article).
http://www.ncbi.nlm.nih.gov/pubmed/19072905

[16]

P. J. Zettler
. (2009). “Is it Cheating

to Use Cheetahs: The Implications
of Technologically Innovative Prostheses for Sports Values and Rules.”
Boston University International Law Journal. (Print Article). Issue 1, page
367
-
410.


A
DDITIONAL
S
OURCES



F. J. Freyer. (May 17, 2012). “Paralyzed
pair
prove power of thought.”
Providence Journal.

(Print Article).

J. L.

Contreras. (August 2012). “Ethi
cal Considerations Behind Brain
-
Computer Interface Research.” Oandp.com . (Online Article).
http://www.oandp.com/articles/2012
-
08_12.asp

Alexander Burval

Zachary Wool



University of Pittsburgh Swanson School of Engineering
8


March 7, 2013



A
CKNOWLEDGEMENTS


Dedicated to the scientists who pursue
research that benefits
the health and progress of mankind.

Dedicated to Derrick Kreider for ke
eping us motivated and
on track.