Emtiyaz (Emt) CS, UBC - Emtiyaz Khan's Homepage · PDF fileBrain-Computer Interface Overview,...

Post on 04-Feb-2018

219 views 2 download

Transcript of Emtiyaz (Emt) CS, UBC - Emtiyaz Khan's Homepage · PDF fileBrain-Computer Interface Overview,...

Brain-Computer InterfaceOverview, methods and opportunitiesEmtiyaz (Emt)CS, UBC

Overview• BCI : application, structure, challenges, opportunities• Two examples of BCI : P300 and ERD/ERS• ERD/ERS BCI – Methods• ERD/ERS BCI – Our Work

Please don’t sleep, we will watch videoshttp://www.youtube.com/watch?v=NIG47YgndP8

http://www.youtube.com/watch?v=qCSSBEXBCbY

Why a Brain-Computer Interface?1.Use the capabilities of remaining pathways2.Detour around the points of damage (FES)3.Or if you have enough time to kill, use BCI

Military Applications Other Fancy Use

Locked-in SyndromeAmyotrophic Lateral SclerosisMultiple Sclerosis

Image Bayliss’ Thesis, and various webpages

Why a Brain-Computer Interface?

A General Brain-Computer Interface (BCI)

Image Pfurtscheller 2006

Evoked Potentials

Operant Conditions

Implanted Methods

BCI: Challenges and Opportunities

Image Pfurtscheller 2006

Evoked Potentials

Operant Conditions

Implanted Methods

ChallengesHigh Signal VariabilityUnderlying Physics unknownLinear or nonlinear Model?Man-Machine learning dilemma

One obvious opportunityOptimization of Electrode position

P300

ERD/ERS

P300 BCI

Images from http://ida.first.fraunhofer.de/projects/bci/competition_ii/albany_desc/albany_desc_ii.html

AverageBCI competetion II DatasetTask relevant stimuli

Non-task relevant response

Sequential data, Discrete o/p, Correlation based models

ERD/ERS BCIAmplitude attenuation/ enhancementin the specific frequency bands

Photographs from Pfurscheller 2002

Sequential data, Continuos o/p,

Strong Spectral charasteristics

C3 C4

ERD/ERS – Methods

Inter-trial Variance (IV) Method

Photographs from Pfurtscheller 1998

Only offline

Sensitive to frequency

band selection

RLS ApproachAdaptive Autoregressive (AAR) Model

Solved with Recursive least square (RLS) algorithms and features classified with Linear Discriminant Analysis (LDA)

observation noise

Photographs from Pfurtscheller 2000

Our WorkDone at the Indian Institute of Science, Bangalore, Indiain 2002-04

PublishedM. E. Khan and D. N. Dutt, "An Expectation-Maximization Algorithm Based Kalman Smoother Approach for Event-Related Desynchronization(ERD) Estimation from EEG", Vol. 54, No. 7, July 2007, IEEE Transactions on Biomedical Engineering

Time-varying AR Model

RLS algorithm

Kalman filters

Use EM to get the parameters{A, Q, R, x0, S0)

EMKS

RLS

KS

Effect of EM Learning on Tracking

Average

Single TrialSignal

Optimization

Improved AR coefficientsTracking, and variance isreduced

Spectrum EstimationOptimization

Signal

Single Trial

Average

Improved Frequency TrackingVariance is reduced

Motor Imagery Data: Spectrum estimatesSignal

Single Trial Spectrum Estimate

Right Hand

Left Hand

Motor Imagery Data : ERD estimation

Final Comments• BCI datasets provide opportunity to test new

algorithms• Every two year there is a BCI competition. Some free

datasets can be found there (just google BCI competition)

• Do it for fun and blame it on humanity!!