Ch.1 Introduction to Brain-Computer Interfacing

unknownlippsAI and Robotics

Oct 16, 2013 (3 years and 7 months ago)

114 views

Ch.1

Introduction to Brain
-
Computer
Interfacing

Overview


Fairytales: translating thoughts into actions without acting
physically.


Recent BCI technologies made it possible!


Aims of BIC


Restore sensory functions


Restore motor functions


Application areas:
neuroprostheses
, computer games, etc.


Technologies for monitoring brain activity:
electroencephalography (EEG), invasive electrodes,
magnetoencephalography

(MEG), positron emission tomography
(PET), functional magnetic resonance imaging (
fMRI
), optical
imaging (
fNIRS
)


Portable and practical: EEG,
fNIRS
, invasive electrodes


Technological bottleneck: sensors


Advanced signal processing and machine learning play a key role.


Other issues: robustness, online adaptation to
nonstationarities
,
sensor fusion, classifier and filter parameter tuning

Hans Berger early 1900s

Electromagnetic Fields of Neural Activity


=65m

at f = 100Hz

in head tissues

Sensors: average distance to neural generators = 0.15m




Quasistatic

assumption

:

Neglect propagation of EM waves




Hans Berger (1929)



Reasonably low
-
cost



Widely used in clinical practice + Neuropsychology research units


EEG

MicroMed

Electrical

Geodesics

NeuroScan


EEG

Magnetoencephalography

(MEG)


SQUID (Superconducting
QUantum Interference Device)


Records magnetic fields
produced by electrical
activity in the brain.


Study neuronal activity by
means of magnetic fields.


Temporal resolution in msec.


Spatial resolution in mm,
better than that of EEG.




SQUID
-

James Zimmerman (1968), David Cohen (1972)



Cost = MRI



About 50 centers worldwide


still counting . . .


MEG

PET



Positron Emission Tomography (PET) is a technique for
measuring the concentrations of positron
-
emitting
radioisotopes within the tissue of living subjects. These
measurements are made outside of the living subjects.
PET can be broken down into several steps:


label a selected compound with a positron
-

emitting
radionuclide


administer this compound to the subject of study


image the distribution of the positron activity as a
function of time by emission tomography


The main positron
-

emitting radionuclides used in PET
include Carbon
-
11, Nitrogen
-
13, Oxygen
-
15, and
Fluorine
-
18, with half
-
lives of 20 min, 10 min, 2 min, and
110 min respectively.





PET System

PET Images

fMRI

BOLD &
fMR

Images

fNIRS

NIRS Signals & Images

Approaches to BCI Control


Two separate approaches, but mostly
mixed of these two.

1)
Learning to voluntarily regulate brain
activity by means of
neurofeedback

and
operant learning principles.

2)
Machine learning procedures that enable
the interference of the statistical signature
of specific brain states or intentions within
a calibration session.



The Biofeedback Approach



Voluntary control of the brain response


Biofeedback is a procedure to acquire voluntary
control over the autonomous parameter of the
brain.


Subjects receive visual, auditory, or tactile
information about their cardiovascular activities
(heart rate, blood pressure), temperature, skin
conductance, muscular activity, electrical brain
activities, blood oxygenation responses.


Subjects are asked to either increase or
decrease the activity of interest.


By means of the feedback signal, participants
receive continuous information about the
alteration of the activity.


The Machine Learning Approach



Detection of the relevant brain signal


Training is moved from subjects to learning
algorithm


Decoding algorithms are individually adapted to the
users that perform the task.


Learning algorithms require examples from which
they can infer the underlying statistical structure of
the respective brain state.


Subjects are first required to repeatedly procedure a
certain brain state during a calibration session.


Machine learning algorithms extract spatiotemporal
blueprints of these brain activities which are used in
subsequent feedback session.


Challenge is the trial
-
to
-
trial variability. Advanced
machine learning techniques are essential.



Integration of the Two Approaches



In practice, BCIs will neither rely solely on
feedback learning of uses nor only on
machine learning approaches.


Co
-
adaptation of the learning use and
algorithm is inevitable.


BCI illiterates: typically about 20% of the
users are unable to successfully classify the
brain activation patters.


Further research work is needed.

Clinical Target Groups


Individuals in need of a BCI for motor
control and communication


Examples


Amyotrophic lateral sclerosis


Cervical spinal cord injury


Brain stem stroke

BCI for Healthy Subjects


Recent interest as a HCI technology


Addition to keyboard, computer mouse, speech or
gesture recognition devices


Ch. 23, 24, 25 for first examples


Brain signals read in real
-
time on a single trial basis
could provide direct access to human brain states
which can be used to adapt HMI.


Monitoring tasks such as alertness monitoring,
cognitive workload, alertness, task involvement,
emotion or concentration.


Current bottlenecks: sensor prices, error rate, price
of EEG. Need fashionable, cheap, contactless EEG
caps.



Brain Pong (
Dornhege

et al, 2006)


Recording Methods, Paradigms,
and Systems for BCI


Current BCIs differ in how the neural
activity of the brain is recorded, how
subjects are trained, how the signals are
translated into device commends, which

application is provided to the user.


Noninvasive Recording Methods for BCI


Recorded from the Scalp (EEG)


Magnetic Activity of the Brain (MEG)


Blood Oxygen Level Dependent (BOLD,
fMRI
)


Blood Flow (NIRS)


EEG
-
BCIs


Slow Cortical Potential BCI (SCP
-
BCI)


Sensorimotor

Rhythm BCI (SMR
-
BCI)


P300 BCI (P300
-
BCI)


Steady
-
State Visual Evoked Potential BCI
(SSVEP
-
BCI)

Generic Noninvasive BCI Setup

Slow Cortical Potentials (SCPs)


SCP
-
BCI


SCP
-
BCI requires users to achieve
voluntary regulation of brain activity.


The traditional S1
-
S2 paradigm


Detection of contingent negative variation
(CNV): a negative SCP shift seen after a
warning stimulus (S1) two to ten seconds
before an imperative stimulus (S2) that
requires users to perform a task
.