recent advances in eegrecent advances in eeg th l da
TRANSCRIPT
Recent Advances in EEGRecent Advances in EEG
T h l d A li tiTechnology and Applications
Tzyy-Ping Jung
Swartz Center for Computational Neuroscience and Center for Advanced Neurological Engineering
University of California San DiegoUniversity of California San Diegoand
Department of Computer ScienceNational Chiao-Tung University, Hsinchu, Taiwan
1 In spite of e iting ad ances in ne oscience
Serious Problems with My Work
1. In spite of exiting advances in neuroscience,
mathematical modeling and neurotechnology, the
current and dominant emphasis of this field failed to
fully capitalize the potential of neuroscience and
neurotechnology to address questions, needs andneurotechnology to address questions, needs and
applications outside of research settings.
2. My work has done very little, if any, to help needed
peoplepeople.
Current Limitations
• The lack of portable and user-acceptable systems p p yfor monitoring brain and body dynamics;
• The lack of mathematical mining and modeling methods to find statistical relationships betweenmethods to find statistical relationships between moment-to-moment variations in environmental, beha io al and b ain d namic eco dingsbehavioral, and brain dynamic recordings;
• How can we leverage scientific principles and• How can we leverage scientific principles and advanced technologies?
Neuroimaging modality for Neuroergonomics
PET MEGfMRIEEG
• In all modalities but EEG, the sensors are heavy.
• EEG is the only modality that does not require the y y qhead/body to be fixed.
• EEG might enable the monitoring of the brain functions g gof unconstrained participants performing normal tasks in the workplace and home. However, …..
EEG Electrodes
In common applications, EEG signals are measure by an electrode with electrolyte gel placed directly on the skin.
The coupling between skin and electrode can be described as a layered conductive and capacitive structure, with series combinations of parallel RC elements
Equivalent circuit elements.
EEG Electrodes
Comparison of electrical coupling of the skin-electrode interface between electrodes
Typically, one of the RC sections dominates and the electrical coupling may be simply represented as a single element with conductance gc in parallel with
it C Y (j ) j C
From Chi, Jung & Cauwenberghs, in press.
capacitance Cc , Yc(jω) = gc + jωCc.
EEG Electrodes
The conventional notion that low resistance (high conductance) is essential for good electrode performance could be misleading in certain cases.g p g
Source input-referred noise
Source input-referred noisepower density:
Vs, rms can be reduced to zero in two limits: either infinite coupling ,conductance (low-resistance contact sensing), or infinite coupling impedance (capacitive noncontact sensing). This presents a rather interesting dichotomy—either of the two extreme cases of zero resistance g yand infinite resistance of skin-electrode contact are actually optimalfor low-noise signal reception
Practical Design Considerations
•To abrade the skin to obtain a low contact resistance (5–10k ).( )
•To employ an amplifier with very high input impedance such that that the skin-electrode impedance becomes negligible.that the skin electrode impedance becomes negligible.
Comparison for EEG Electrodes
Standard wet electrodes : low skin impedance, and buffer the electrode against mechanical motion. But, they may be messy, time-consuming, irritating and the signal quality degrades over time.
Rigid metal electrodes: subject to motion artifacts
Dry foam electrode (Gruetzmann et al., 2007): comfortable and stable with increased resistance to motion artifact.
MEMS sensors: low skin impedance. But, they may be irritating and difficult to penetrate the hairs.
Microprobe electrodes: sensitive to motion artifacts.p
Non-contact sensors: sensitive to motion artifacts, poor settling times. Friction between the electrode and insulation g(e.g. hair) can cause large voltage excursion at the sensitive input.
Comparison between EEG Signals collected by Wet and dry Electrodes
Dry MEMS Sensor
Correlations between EEG acquired by a conventional and dry MEMS electrodes
1,000 seconds (4,000 pt.)
Novel Dry Sensors for Hairy Sites
No preparation, no gel, through the hair
Fast and low-cost fabrication (US$50 less than US$0.5)
Flexible substrate + Once-forming a set of dry electrode Flexible substrate + Once forming a set of dry electrode fingers -> easily contact with scalp
Flexible Substrate (Silicone)
Once-formingMulti-fingers
Scalp
Comparison EEG signals between using Wet and Novel Dry Sensors in Occipital Area
50
100
ro V
olt)
0.5
1
Correlation
0
mpl
itude
(mic
r
0
Cor
rela
tion
0 0 5 1 1 5 2 2 5-100
-50E
EG
Am
0 0 5 1 1 5 2 2 5
-1
-0.5
C
WetDry
40 Correlations between EEG acquired by a conventional and dry electrodes
0 0.5 1 1.5 2 2.50 0.5 1 1.5 2 2.5Time(sec)
20
30
Th l ti f t t l 96%
10
0 The correlation of total 96% data are greater than 0.97. (non-filtering raw data)
0.94 0.95 0.96 0.97 0.98 0.99 10
Correlation Coefficient
Novel Dry Foam-based EEG Electrode
Decreased Impedance Low Movement Artifacts In forehead and hairy site, the correlation with wet
electrode all are over 90% similarity. L t 0 5 US/ it d t Low cost <0.5 US/unit and easy to wear MRI Compatible
Polymer Foam + Conductive Fabric
Mobile & Wireless EEGLaboratory EEG
NCTU’ MW EEGNCTU’s MW-EEG
D MEMS Bi A & ADCDry MEMS EEG Sensor
Bio-Amp & ADC
μcontrollerμcontrollerand Bluetooth
Specification of MINDO
EEG Headband
F Di ib d Ci iFeatures Distributed Circuits
MiniaturizatioDAQ: 20 x 18 (2 or 4 pieces)
Miniaturization
Size (mm) MCU: 40 x 25
Weight < 100 gS li R 512HSampling Rate 512Hz
Bandwidth Filter to 1 - 50 Hz
Gain 6000 timesOutput current
(working) 31.58 mA(working)Battery Life
(3.7V, 1,100mA) 33 hours
Current Limitations
• The lack of portable and user-acceptable systems p p yfor monitoring brain and body dynamics;
• The lack of mathematical mining and modeling methods to find statistical relationships betweenmethods to find statistical relationships between moment-to-moment variations in environmental, beha io al and b ain d namic eco dingsbehavioral, and brain dynamic recordings;
• How can we leverage scientific principles and• How can we leverage scientific principles and advanced technologies?
A Typical Event-related Potential ExperimentExperiment
Difficulties in Observing Distributed EEG dynamics
Scalp EEG data
CortexLocal
Synchrony
Local SkinDomains of
synchronySynchrony
y y
Scalp EEG signals appear to be noisy because they each sum anoisy because they each sum a mixture of signals generated in
many brain areas.y
Scott Makeig / UCSD 05/08
From Jung et al., 2000.
Modeling and Mining Distributed EEG Dynamics
Cocktail Party
Independent component analysis (ICA) pioneered for EEG by Makeig, Bell, Jung &
Blind source separation by ICA p y g, , g
Sejnowski (1996), separates high-density EEG signals into their source contributions.
y
ICA separates brain and non-brain signals contributing to high-density EEG recordings This allows direct
Cortical source projections
EEG recordings. This allows direct analysis of distinct source activities in response to stimulus presentation and subject’s behavior.subject s behavior.
Cortical source localization
Subject BEM head model
Advanced forward head models allow advanced inverse models based on cortical topology to identify the sources of the scalp
localization
S. Makeig, 2007
p gy y pEEG recordings and to study their interrelationships.
Miniaturization of EEG Acquisition and On-Line processing a d O e p ocess g
I t t ti A lifi & Si D lt ICAInstrumentation Amplifier &Band-Pass Filter
Sigma-DeltaADC
ICAsignal process
DSP
Application
On-Chip Signal Processing
A VLSI Implementation of a 4-Channel Independent Component Analysis
ProcessorProcessor
Department of Electronics EngineeringNational Chiao Tung University, Taiwan
Prof. Wai-Chi Fang Teamwfang@mail nctu edu [email protected]
System-on-Chip Design and System Integration Laboratory
24
System Architecture of Mobile & Wireless BCI
Dry MEMS Sensors
EEG DAQ, Amp, and ADC
Wireless telemetry y
Wireless receiver
Wearable on-line signal processing
Visualization & feedback receiver signal processing & feedback
Work-in-Progress
VLSI implementation of bio-amp, ADC and wireless moduleVLSI implementation of bio amp, ADC and wireless module 16-channel High-end EEG measurement systemDSP PCB -> On-chip Signal Processing (SoC)p g g ( )
RAM_BRAM_ASDMA
RAM_BRAM_ASDMA
Core_A Core_BCore_A Core_B
Front-end chip: Bioamp+ filters + ADC Back-end Chip: Dual Core DSP
Current Limitations
• The lack of portable and user-acceptable systems p p yfor monitoring brain and body dynamics;
• The lack of mathematical mining and modeling methods to find statistical relationships betweenmethods to find statistical relationships between moment-to-moment variations in environmental, beha io al and b ain d namic eco dingsbehavioral, and brain dynamic recordings;
• How can we leverage scientific principles and• How can we leverage scientific principles and advanced technologies?
What is a BCI?A brain–computer interface (BCI), sometimes called a
What is a BCI?b a co pute te ace ( C ), so et es ca ed a
direct neural interface or a brain–machine interface, is a direct communication pathway between a brain and an
l d i BCI f i d i iexternal device. BCIs are often aimed at assisting, augmenting or repairing human cognitive or sensory-motor functionsfunctions.
Wolpaw et al. 2002.
BCIs for communication and device control
• ECoG -based BCIs.
ECoG control of vertical cursor movement using specific motor imagery to
th dmove the cursor up and rest to move it down. Modified from LeuthardtModified from LeuthardtEC et al, 2004.
BCIs for communication and device control
• EEG-based BCIs(A) P300 event-related
i l Cpotential BCI.
(B) Sensorimotor(B) Sensorimotor rhythm BCI.
(C) S d i l(C) Steady-state Visual Evoked Potential
P300 BCIsP300 BCIs
• Farwell and Donchin 1988
• P300 Speller
Mu BCIsMu BCIs
SSVEP BCIs
Photic driving
tStimulus >6Hz
t tSteady-state VEP
SSVEP BCIs
Passive or Affective BCIsPassive or Affective BCIs
(A) Cognitive state monitoring and management
(B) Emotional state estimation(B) Emotional state estimation
(C) Attention, intention
(D) Clinical applications, e.g. seizure detection and/or
prediction neurorehablitationprediction, neurorehablitation
(E) etc.
Testing the Mobile & Wireless BCI in Testing the Mobile & Wireless BCI in a VRa VR--based Dynamic Driving Simulatorbased Dynamic Driving Simulatora VRa VR based Dynamic Driving Simulatorbased Dynamic Driving Simulator
Implemented of Cognitive-state Monitoring on the DSP ModuleMonitoring on the DSP Module
4-channel EEG
Down-sampling
ICA(multiplying W to the data)
ICA Training
Copy the
Removing artifactual components
pynew W
when ICA convergescomponents
FFT
Auditory feedback to the subject FFT
Drowsiness
39
Drowsiness Detection
EEG-based Cognitive-Status MonitoringD El t dWet Electrodes Dry Electrodes
DSP and Display module
2 5 x 1 5 in2.5 x 1.5 in
Lin at. al., Proc. of the IEEE, July 2008.
Acknowledgement