IpsiHand: An EEG-based Brain Computer Interface for Rehabilitation and Restoration of Hand Control following Stroke and Traumatic Brain Injury Using Ipsilateral Cortical Physiology

yakzephyrΤεχνίτη Νοημοσύνη και Ρομποτική

24 Νοε 2013 (πριν από 3 χρόνια και 4 μήνες)

65 εμφανίσεις

The

undergraduate

authors

would

like

to

thank

D
.

Bundy

for

providing

technical

support

and

manuscript

comments,

Dr
.

E
.

Leuthardt

for

advice

throughout

the

work,

and

Dr
.

A
.

Nehorai

for

his

support

as

well
.

This

project

supported

in

part

by

The

National

Collegiate

Inventors

and

Innovators

Alliance,

The

Washington

University

School

of

Engineering,

and

Emotiv

Systems
.

Combining

the

discovery

of

signals

in

contralesional

hemisphere,

electronics,

and

advances

in

rehabilitation,

IpsiHand

offers

a

new

rehabilitation

option

for

stroke

and

TBI

survivors
.

IpsiHand

was

able

to

process

EEG

signals

for

real
-
time

hand

control

with

accuracy

consistent

with

previous

studies

[
8
]

and

theoretical

ROC

analysis
.

Combining

BCIs

and

orthotic

devices

induces

neural

plasticity

and

improves

motor

function

[
8
]
.

Furthermore,

IpsiHand

therapy

is

unhampered

by

the

severity

of

neural

pathway

injury

by

circumventing

the

injury
.

Compared

to

other

devices,

IpsiHand

facilitates

plasticity

most

directly,

is

cheaper

and

more

portable
.


Future

Directions
:


We

plan

improvements

in

portability

and

signal

processing
.


Portability

Goals
:

Currently
,

a

laptop

processes

the

EEG

signals

used

for

orthosis

movement
.

We

plan

to

miniaturize

the

processing

to

a

micro
-
computer

for

portability,

allowing

patients

to

even

use

the

device

as

a

replacement

for

normal

hand

function

in

daily

life
.

Signal

Processing

Goals
:


Expand

the

system’s

ability

to

adapt

to

spatially

non
-
stationary

EEG

signals
.

Possibilities

include

adaptive

feature

selection

and

adaptive

spatial

reference

selection
.


Remove

muscular

and

other

artifacts

with

spatial

and

temporal

filtering
.

This

is

essential

for

device

performance

outside

of

a

research

setting
.

Artifact

sources

include

eye

blinks,

EMG,

ECG,

and

breathing
.


Build

real
-
time

feature

selection

to

eliminate

need

for

screening

procedure

and

improve

user

friendliness
.


Improve

control

signal

normalization

and

adaptation

through

least

mean

squares

algorithm

and

linear

regression

techniques
.

An

Emotiv

EPOC
TM

EEG

headset

records

EEG

signals

from

the

scalp

with

14

channels
.

The

headset

aligns,

bandpass

filters,

and

digitizes

the

signal

at

128

Hz

and

transmits

wirelessly

to

a

laptop
.

EEG

signals

tend

to

be

small

and

spatially

diffuse

compared

to

invasively

recorded

signals,

so

maximizing

the

signal
-
to
-
noise

ratio

is

imperative

to

device

performance
.

We

used

a

large

bi
-
polar

reference

spatial

filter

to

attenuate

noise

from

wide

areas

of

the

scalp

and

detect

signals

specific

to

a

particular

brain

area
.

This

also

makes

IpsiHand

resilient

to

electrode

placement

variations

[
6
]
.



Signal

processing

was

carried

out

in

the

BCI
2000

framework,

a

development

platform

that

allows

for

rapid

recording,

filtering,

and

feature

selection

of

brain

signals

[
2
]
.

Initial

screenings

determine

the

EEG

features

that

our

algorithm

will

use

to

contrast

movement

from

rest
.

During

screening,

the

user

alternates

between

periods

of

attempted

hand

movement

and

periods

of

rest
.

With

screening

data,

we

identified

the

specific

electrode

channels

and

frequency

bins

with

consistent

changes

in

power

spectrum

between

hand

movement

and

rest

conditions
.

The

power

of

the

selected

channel

is

normalized

to

0

mean

and

unit

variance

using

a

buffer

of

previous

trial

data

and

sent

to

LabView

software

for

conversion

into

control

signal
.




In

LabView,

the

signal

is

compared

to

a

user

defined

threshold

and

mapped

to

an

actuator

position

command

according

to

𝑥
𝑖
=
𝑥
𝑖

1
+
𝑔
𝑠


where

x
i

is

the

commanded

position

of

the

i
th

program

iteration,

g

is

a

user

defined

gain,

and

s’

is

the

thresholded

signal
.

IpsiHand: An EEG
-
based Brain Computer Interface for Rehabilitation and Restoration of Hand
Control following Stroke and Traumatic Brain Injury Using Ipsilateral Cortical Physiology

Sam
Fok
, Raphael Schwartz, Mark
Wronkiewicz
, Charles Holmes, Jessica Zhang, Thane Somers, David Bundy, Dr. Eric
Leuthardt

Washington University in St. Louis

Abstract

Methods

Results

Discussion

Stroke

and

traumatic

brain

injury

(TBI)

cause

long
-
term,

unilateral

loss

of

motor

control

due

to

brain

damage

on

the

opposing

(
contralateral
)

side

of

the

body
.

Conventional

therapies

are

ineffective

at

restoring

function

in

about

half

affected
.

Brain

computer

interfaces

(BCIs)

show

promise

for

rehabilitation

but

remain

primarily

restricted

to

the

research

stage
.

Furthermore,

traditional

BCIs

cannot

work

if

areas

such

as

M
1

are

damaged
.

We

present

a

novel

BCI,

IpsiHand,

which

circumvents

signal

sourcing

issues

in

an

injured

brain

as

well

as

risk

associated

with

invasive

recordings
.

IpsiHand

uses

electroencephalography

(EEG)

to

record

novel

motor

intent

signals

and

control

a

powered

hand

orthosis,

which

allows

the

undamaged

hemisphere

to

control

both

hands
.

Through

sensory

and

proprioceptive

feedback

and

neural

plasticity,

IpsiHand

can

strengthen

ipsilateral

neural

pathways
.

Introduction

References

Stroke

and

TBI

combined

are

the

leading

cause

of

disability

in

the

US,

with

around

a

million

cases

annually
.

Half

report

trouble

with

hand

movement,

and

conventional

physical

therapy

produces

little

improvement

after

3

months

post

injury

[
1
]
.

Therapies

requiring

the

patient

to

actively

control

their

impaired

limb

are

most

likely

to

induce

reorganization

of

neural

pathways

and

improving

control

but

require

intensive

interaction

between

the

patient

and

practitioner

[
2
]
.


BCIs

promise

new,

more

effective

motor

therapies
.

They

are

traditionally

applied

in

cases

where

central

nervous

signals

are

cut

off

from

their

destination

by

injury
.

Electrical

signals

are

recorded

from

the

brain

to

circumvent

injuries

and

control

devices

to

actuate

a

target

limb,

which

recouples

intent

to

move

and

movement

[
3
]
.



Despite

promise,

conventional

BCIs

cannot

be

applied

to

cases

of

brain

injury

where

damaged

primary

motor

cortex

contralateral

to

the

affected

limb

produces

no

signals
.

However,

recent

study

found

distinct

cortical

physiology

associated

with

ipsilateral,

contralesional

hand

and

limb

movement

in

regions

distinct

and

separable

from

the

primary

motor

cortex

[
4
]
.

These

signals

exist

in

cortex

anterior

to

ipsilateral

primary

motor

cortex

at

frequencies

below

40
Hz

[
5
]
.

We

used

a

non
-
invasive

EEG

consumer

headset

to

record

from

cortex

and

control

an

orthosis

that

opens

and

closes

a

subject’s

hand
.

As

the

least

invasive

recording

technique,

EEG

is

most

practical

for

immediate

application

in

the

clinic
.


IpsiHand

demonstrates

the

synthesis

of

neurophysiology,

consumer

electronics,

and

signal

processing

to

develop

new

devices

for

more

effective

therapies
.

Implementation

of

this

design

constitutes

a

fundamentally

new

approach

to

restoring

function

in

stroke

and

TBI

survivors
.

Synchronous

neuronal

firing

over

large

areas

of

the

cortex

are

recorded

from

the

scalp

and

processed

onboard

a

laptop
.

The

signal

is

filtered

to

generate

a

control

signal
,

which

is

then

sent

to

a

linear

actuator

fitted

to

an

orthosis

controlling

the

patient’s

finger

closure
.

IpsiHand

was

tested

with

three

healthy

subjects

to

verify

the

ability

to

use

non
-
conventional

signals

from

cortex

on

one

side

of

the

brain

to

control

a

hand

on

the

same

side

of

the

body
.

We

found

that
:


1
.

Hand

movement

correlates

with


ipsilateral

signals
.


2
.

IpsiHand

can

use

EEG

signals

to

move

the

hand
.

[
1
]

HS
-
Jorgensen,

et_al,
-
"Outcome

and

time

course

of

recovery

in

stroke
:

Part

II
:

Timecourse

of

recovery
.

The

Copenhagen

Stroke

Study
,"
Archives

of

Physical

Medicine

and

Rehabilitation
,

1995
.

[
2
]

G

Schalk
,

D

McFarland,

T

Hinterberger
,

N

Bribaumer
,

and

J

Wolpaw
,

"
BCI
2000
:

a

general

purpose

brain

computer

interface

(BCI)

system
,“
IEEE

Transactions

on

Biomedical

Engineering
,

pp
.

1034
-
1043
,

2004
.

[
3
]

D

Broetz

and

et

al,

"Combination

of

brain
-
computer

interface

training

and

goal
-
directed

physical

therpy

in

chronic

stroke
:

A

case

report,"

J

Neurorehab

and

Neural

Repair
,

pp
.

674
-
679
,

2010
.

[
4
]

Kimberly

J

Wineski

et

al
.
,

"Unique

cortical

physiology

associated

with

ipsilateral

hand

movements

and

neuroprosthetic

implications,“

Stroke
,

pp
.

3351
-
3359
,

2009
.

[
5
]

KJ_Wineski

et
.

al
.
,

"Unique

cortical

physiology

associated

with

ipsilateral

hand

movements

and

neuroprosthetic

implications,“

Stroke,

2009
.

[
6
]

DJ_McFarland
,

et
.

al,

"Spatial

filter

selection

for

EEG
-
based

communication,

“Electroencephalography

and

Clinical

Neurophysiology
,

vol
.

103
,

1997
.

[
7
]

Ernst

Niedermeyer

and

Fernando

Lopes

da

Silva,

Electroencephalography
:

basic

principles,

clinical

applications,

and

related

fields
.:

Lippincott

Williams

&

Wilkins,

2004
.

[
8
]

DJ_McFarland
,

et
.

al,

"Electroencephalographic

(EEG)

control

of

three
-
dimensional

movement,"

J
.

Neural

Eng
.
,

vol
.

7
,

2010
.

Acknowledgements

Signal Processing and Control

Signal Acquisition

Mechanical
Orthosis

A

Becker

Oregon

Talon
TM

prefabricated

orthosis,

designed

to

couple

wrist

motion

to

hand

closure,

was

fitted

with

a

powered

linear

actuator

(Firgelli

Miniature

Linear

Motion

Series

L
16
),

controlled

through

our

LabView

algorithm
.


The

orthosis

size

is

adjustable,

which

is

key

in

a

clinical

setting

with

a

variety

of

patients
.

To

mechanically

prevent

hyperextension

or

hyperflexion

of

the

hand,

the

range

of

actuator

motion

is

mapped

only

onto

the

natural

range

of

finger

joint

rotation
.

Left
:

Correlation

colormap

between

left

hand

movement

and

rest

per

electrode

and

per

frequency

bin
.

Electrodes

named

according

to

the

10
-
20

EEG

system(
Right
)

[
7
]
.

Bins

with

high

correlation

are

good

candidates

for

control

signals
.

Clusters

(dotted

red

circles)

of

high

correlation

in

ipsilateral

cortex

noted

around

the

12
Hz

bins

in

F
3

through

P
7

and

also

in

channel

F
3

around

the

22
Hz

bin
.

Window Length



2.6 Seconds



2.0 Seconds



1.0 Seconds



0.5 Seconds


-

-

-

Random Guess

Window Length (Seconds
)

Accuracy of Classification

2.60

96.2%

2.0

92.3%

1.0

86.5%

0.5

80.8%

Classification

accuracy

using

best

threshold

as

determined

by

ROC

analysis

per

window

length
.

In

actual

trials,

through

10

sets

of

trials

with

non
-
impaired

individuals

we

achieved

an

81
.
3
%

success

rate

for

the

1
D

cursor

task

and

orthosis

using

a

0
.
5
s

window
.

Note

that

this

is

slightly

higher

than

predicted

by

ROC

analysis
.


Top
:

Left
:

Spatial

map

of

L
.
hand

vs
.

rest

correlations

per

channel

at

12
Hz
.

Note

correlations

present

across

frontal

cortex
.


Top
:

Right
:

Spatial

map

of

L
.
hand

vs
.

rest

correlations

per

channel

at

22
Hz
.

Note

correlation

only

present

unilaterally

in

channel

F
3
.


Bottom
:

Left
:

Raw

spectrum

of

channel

F
3

shows

difference

in

power

around

12

and

22
Hz

bins

during

L
.
hand

and

rest

conditions
.

Bottom
:

Right
:

ROC

curves

shows

classification

performance

using

varying

window

lengths

and

thresholds
.

Note

that

longer

window

have

higher

performance

but

also

increase

the

system

latency
.

The

features

identified

during

screening

were

used

to

modulate

orthosis

closure

and

1
D

cursor

movement
.

The

subject

was

tasked

with

moving

the

cursor

to

a

target

randomly

located

at

either

the

left

or

right

side

of

the

screen

(below)
.


Screening Results

Online Performance Results