Slide 1

28/04/2008 Agenda Dr. G. Otte

Principles of EEG/ERP

05/05/2008
EBM & Statistical Evidence in
clinical studies (stats without
maths)
 A psychiatric (?) case history...
• A young woman, 45 years is admitted because of
depression. Her parents tell us that she is a math
teacher who acts a bit strange since 6 months. Her
fiancee has recently broken the engagement and now
she often absent at work, refuses to eat and “since
then” hardly ever speaks (according to parents).
• In the clinic her behaviour looks like depression at
times even bizar shidzofrenic (social regression,
looks sad, depressed, mutistic, avoids friends,
anxiety..).
• She develops nervous tics and is very anxious and
easily “jumpy” aka startled on sudden noises
2
 Triphasic pseudoperiodic
complexes
Creutzfelt – Jacob (a human prion disease (BSE ?)

I am a mean
mad Cow

3
How to explore the brain ?

of this patient ? 4
 De brain is firmly packed for
transportation by Genetics, Inc

5
Glass Skull would be “nice”

6
7
 Not really
“patient friendly”

8
The Hart and the Brain

• The hart is a pump

• You can “hear” it (use your stethoscope !!)

• The brain is no pump: contains water (in the

ventricles) but no “valves” or moving parts
• Use stethoscope : hear only...silence....
9
Electrical Activity

10
 CATON

1875 University
of Liverpool
11
First Picture of human EEG
Blood vessel

Pravdisch-Neminsky EEG-trace

12
E.E.G
Hans
Berger

...may I succeed and obtain my goal

after more then 20 years work, the
creation of a mirror of the brain....

The electroencefalogram

Dr. Hans Berger

Jena University
16/11/1924 13
Hans Berger: een Psychiater !
 E.E.G
Electro- (electrical)
Encefalo-(brain)
Gram. (write down) 15
 EEG: technique
• Passive registration on the scalp
surface of electrical voltage
differences in the resting brain
• Atraumatic , easily tolerated by the
patients and easy to reproduce.
• Functional type of investigation
complementary to other imaging
techniques like CT, MRI, PET-scan.

16
 Data from
Neuroanatomy

17
 Data from
Neuroanatomy

18
BRAIN CORTEX

19
20
 Data from
Neuroanatomy

21
 Data from
Neuroanatomy

22
23
The EEG measures electrical
cortical synaptic field activity

• One mini colums neocortex (3 mm high and

radius 0.3 mm) contains +/- 100 pyramidal
cells and +/- 1 million synapses.
• One macro column contains 1000 mini
colums (= 100.000 Pyramidal Cells)
• We need at least 600 macrocolumns
(synchroneusly active) to get a measurable
dipool (6 cm2)
24
 Read about it

• Info from Paul

Nunez
(sec edition)
• The “ bible” on
Neuro physics of
EEG

25
 All of All of signal
biophysiology analysis

26
 Lorento di No
• Open and closed cortical fields
• fi Hippocampus is a closed field

27
 MEG and EEG do NOT measure the
same

• EEG synaptic fields

• MEG magnetical field generated by
axonal current flow.

28
INTRACELLULAR: MILLIVOLT ORDER

29
INTRACORTICAL

30
SUBDURAL

31
ATTENUATION !

32
 ATTENUATION !
• In amplitude: factor 1000 (we need
amplifiers)
• In frequency: high frequencies (cellular
action potentials ie “spikes” ) disappear and
only the low frequencies persist as EEG
waves : “low pass filter = high cut”.

33
Relationship EEG vs generators :
ExcitatoryPSP/InhibitoryPSP:
simultaneous registration

34
Analog system

35
 Digital system

Visualisation Lo pass
filters Amplifier Head box
and registration

optoisolation 36
37
Brain oscillations

38
 Brain oscillations
Alfa 8-13 50-100 µV Occipital Rest
Hz Eyes
closed
Beta 20-30 5-20 µV Frontal Mental
activity
Theta 4-7 Hz 30-70 µV Temporal Young

Delta 0-4 Hz 30-80 µV Diffuse Deep sleep

Mu 9-10 30-60 µV Parietal Motor

Hz activity
39
There are other frequencyband subdivisons possible….

40
41
 Roberto Pasqual Marqui

Analysis of topological identical frequencybands with

sLoreta method
42
1994

43
 Ant
Post

44
EEG registration

45
Electrical eye on the brain

46
 While Medical Imaging has a fantastic
spatial resolution

EEG/ERP outperforms medical imaging in the time resolution

47
BRAIN CORTEX
EEG measures the extracellular field potentials generated
at the apical dendritic synapses of large numbers
synchronous activated pyramidal cortical neurons:
synaptic activity = neurotransmitter driven !

48
 TIME WINDOW: 10-800 msec

• Field of Interest for Neuropsychophysiology

• Study of integrated temporal dynamics of emotion,
cognition, consciousness
• In vivo evaluation of neurotransmitter (dys)
functions by their synaptic activity and postsynaptic
extracellular potentials (P50, LDAEP, P300, CNV).
• Objective evaluation of dynamical elements of
psychopathology
• Opening doorway to endophenotyping maior
psychiatric disorders

49
 TIME WINDOW: 10-800 msec

• Field of Interest for Neuropsychophysiology

• Study of integrated temporal dynamics of emotion,
cognition, consciousness
• In vivo evaluation of neurotransmitter (dys)
functions by their synaptic activity and postsynaptic
extracellular potentials (P50, LDAEP, P300, CNV).
• Objective evaluation of dynamical elements of
psychopathology
• Opening doorway to endophenotyping maior
psychiatric disorders

50
 TIME WINDOW: 10-800 msec

• Field of Interest for Neuropsychophysiology

• Study of integrated temporal dynamics of
emotion, cognition, consciousness...
• In vivo evaluation of neurotransmitter (dys)
functions by their synaptic activity and postsynaptic
extracellular potentials (P50, LDAEP, P300, CNV).
• Objective evaluation of dynamical elements of
psychopathology
• Opening doorway to endophenotyping maior
psychiatric disorders

51
 TIME WINDOW: 10-800 msec

• Field of Interest for Neuropsychophysiology

• Study of integrated temporal dynamics of emotion,
cognition, consciousness
• In vivo evaluation of neurotransmitter (dys)
functions by their synaptic activity and
postsynaptic extracellular potentials (P50,
LDAEP, P300, CNV).
• Objective evaluation of dynamical elements of
psychopathology
• Opening doorway to endophenotyping maior psychiatric
disorders

52
 TIME WINDOW: 10-800 msec

• Field of Interest for Neuropsychophysiology

• Study of integrated temporal dynamics of emotion,
cognition, consciousness
• In vivo evaluation of neurotransmitter (dys)
functions by their synaptic activity and postsynaptic
extracellular potentials (P50, LDAEP, P300, CNV).
• Objective evaluation of dynamical elements of
psychopathology
• Opening doorway to endophenotyping maior
psychiatric disorders

53
 TIME WINDOW: 10-800 msec

• Field of Interest for Neuropsychophysiology

• Study of integrated temporal dynamics of emotion,
cognition, consciousness
• In vivo evaluation of neurotransmitter (dys)
functions by their synaptic activity and postsynaptic
extracellular potentials (P50, LDAEP, P300, CNV).
• Objective evaluation of dynamical elements of
psychopathology
• Opening doorway to endophenotyping maior
psychiatric disorders

54
Electrocorticography

55
 Electro-Cortico-Graphy
• Direct Registration on the cortex
(superficial or deep).
• The filters : (meninges-skull-skin) are
partially eliminated: signals with much
better spatial resolution !
• Electro-registration is largely connected to
electro-stimulation (cfr tests by Penfield,
Jose Delgado et al)

56
 57
Prof. Dr. Jose Delgado: the first electrical Toreador
 EEG Techniques

Filter !

58
59
60
61
62
 X ray

Dense electrode
sheets / maps

63
64
 Electrodes are inserted
intracortically

10 X 1O array

65
No needles in
My brain: I am
not a fakir !!

66
67
68
69
70
71
“Borg” - aspect

72
Famous Bionics

73
We can go on……and on…

74
Evolution is towards “small is
beautifull”
Miniaturisation
Telemetrics

75
 76
Telemetric solution
Magnetoencefalography: another way to bypass
the annoying filter effects of bone-scalp as
magnetic fields pass them unattenuated. But:

• Not really portable

• Extremely expensive
• Very delicate : needs expensive shielding
environment
•EEG and MEG: not the same current source

77
 “Classical EEG”
• Registration by surface scalp electrodes
• Non traumatic
• Painless - Cheap - Easy to repeat
• No blood loss
• But:
– Less spatial resolution
– More low pass filtered signals
78
 H. Jasper
10-20 System Montreal, 1950

79
Fp: prefrontal
F: frontal
C: central
P: parietal
T: temporal
O: occipital Uneven index: Left
Even Index: Right 80
How to understand 10-20 ?
Sagittal plane
• One defines 3 planes
• Sagittal: line Nasion-Inion
with 3 point marks:
– Central mark (halfway): Cz
– 10% of distance above inition
– 10% of distance above nasion

81
 10-20 II: frontal plane
• The frontal meridian
– From left tragus
through the Cz to right
tragus: (cartilageous
part of the ear)
– On this line two marks:
• 10% of the distance
above left tragus
• 10 % of the distance
above right targus
82
 10-20 III: Horizontal plane
• The horizontal
meridian /line
– From supra
nasion via super
tragus to super
inion
– On both (left and
right) sides.

83
Registration

84
Technique Fixing the electrodes
The EEG technician is
responsible for optimal
contact between skin and
electrodes in order to get the
resistance as low as possible
(Impedance less then 5-10K
Ohm: the lower the better)
Desquamation of death skin
cells and appliance of
electrolytic gelly is needed
for a better conduction and
lower resistance
(impedance=res to AC) 85
 A rubber helmet (fixed
below the chin) serves as a
fixing frame for the
skinelectrodes
(painless bloodless
technique)

86
Variant Systems

Rubber helmet

Electrocap
87
The electrodes are connected to a
preamplification stage

88
Number of channels~spatial
resolution

89
 Routine : 8 of 16 channals
• Actually already 32
• Research: 64-128-256
• Each channel: separate ampli
• Cost of system~nr channels
• Fixing 256 electrodes in a standardised way
takes a lot of work and time.
• Some electrocaps can simplify that daunting
task. (Geodesics, Inc)
90
Geodesics Electrocap
92
 EEG accessoires

Electrodes
Electrocaps and gel (Ag/AgCl)

Cable to ampli
93
 Supplies

• Electrode caps
• Electrodes
• Gels et al
94
 Most used caps
• Easy Caps
• Electro caps
• Neuroscan Quick cap

• Waveguard (ANT)
• Hydrocel nets (EGI)

95
 Easycap
• http://www.easycap.de/easycap/index.htm

Exists for 21, 36 en 78 Ch EEG

96
Electro-cap

http://www.electro-cap.com/

97
Neuroscan Quick Cap
proprietary

http://www.neuroscan.com/landing.cfm 98
EGI: Hydrocel sensor nets

http://www.egi.com/

99
 Waveguard by ANT

http://www.ant-neuro.com/
64, 128, 256 Ch
100
A simple and handy system

101
 Extension for high density EEG

• 20 20 (usefull is 17 X 17: 289 locations 5 % systeem

• Website Robert Oostenveld
• Artikel Oostenveld

http://oase.uci.kun.nl/~roberto/index.php/electrode/ 102
 Extension of 10-20 system
Bvb het 5% system

http://oase.uci.kun.nl/~roberto/index.php/electrode/

103
 Extension of 10-20 system

Interesting literature freely available here

http://www.ant-neuro.com/products/caps/waveguard/layouts
104
105
ANT-Mousemat
106
 ANT adaptor

EGI sensor net

107
 Our system: private
Flexible Rubber Combicaps
system
• No high density (21 electrodes)
• Non shielded
• Very practical
• Adapts easily to different skull “models”
volumes and forms
• Quick application technique (developed by
my technician Mr Rik Broothaers)
Broothaers
108
109
 What electrodes ?
• CAVE!!!! Not all electrodes can do everything
(like ERP)
• Prefer Ag/AgCl electroden : always safe !

• Gold and Platinum (they look shiny and

expensive) exert a capacity effect so high pass
(low cut) filter that will attenuate the low
frequencies we need in most ERP studies (fi
CNV) 110
Ag/AgCl electrodes

Silver and gold 111

Electrodes to headbox

Headbox 112
Headbox to Amplifier
Older EEG machines

113
Control buttons per channel

114
Other control buttons: measuring impedances

115
Impedances

116
Older multichannel EEG machines
were rather impressive

117
 ModernEEGand headbox

Fiberoptic
isolation:
patient safety

118
 Even high density ampli’s are
compact

21 ch

128 ch (cascaded)

64 Ch 119
Modern EEG labs

F. Ferrerlaan 88a Gent (neurolab P.C dr guislain)

Preparing sleep EEG 120

 What
electrode gel ?

• Best a good electrolyt (most manufacturers

have their own brand)
• Do not use “ betonniet”: very patient
unfriendly if after each EEG patients look
like this…
121
First results of our new EEG gel….
122
 Dirty hair ?

Camel Hair
Grease 123
A more drastical solution…
124
 Degreasing gels (hydro) are available

Advice to patients : wash hair. Do not use gels!! (Prepare a

small patient brochure)
125
Before starting a registration session
• Signal “garbage” in = analysis garbage “out”.
• Low impedances are best (resistance to AC:)
(<5kOhm).
• Slight skin abraision removes dead skin cells and
eases application of electrolyte gel and lowers
impedance
• Measure impedance before starting registration
but also during the exam at regular time intervals.

126
 From channels to derivations to
montages
• A channel

• A derivation

• A montage

127
Two channels

128
A montage

129
 Montages
• Bipolar: from active electrode to active
electrode
• Referential: from active electrode to a so
called “””neutral””” referential electrode
• Other:
– Laplacian derivation
– Averaged reference

130
From Synaptic field to Signal

Basic Physiological Principles

131
We ‘see’ (measure) electrical fields
by the electrical “eye” of our
electrodes

Positive en negative charges ‘cancel’ (at rest) 132

 What happens in depolarisation?

Depolarisation
in an axon

133
 We see (measure) electrical fields
with our electrode (eyes)

Depolarisation
in one axon

134
135
136
137
138
Axonal depolarisation sinusoidal

139
If the electrode is away : it ‘sees’ both the
depolarisation and the repolarisation front

140
EEG < synaptical cortical fields

141
Corticale Dipool lagen

142
143
144
145
 Differential amplification : Difference between inputs

Monopolar (referential) Bipolar : less magnitude

146
147
 Clinical applications

The differential ampli gives, per channel the difference

between input 1 and input2. in (A): bipolare montage; in
(B) monopolar of referential. 148
 Convention: scope
A positive potential is downward deviation on
the scope !!!

149
 2
1

Input of differential ampli

• Output = (30) – (+70):= - 40 so upward
(negative) even if both inputs are positive )
• remember: If input 2 is relative more
positive then input 1 then deflection is
UPWARD.

150
 Convention ampli
Output= input 1 – input 2
+

Input 2

Input 1

If input 2 is more positieve then input 1

Then UPWARD deflection 151
 Sequential from nasion to inion

2 more negative
then 1
2 >pos then 1
2 > pos then 1
2 is EQ to 1

152
 Situation B

2 > neg dan 1

2 EQ 1

2 >pos 1
2 (slightly) > 1

153
 Situation C

2 EQ 1
2 > pos 1

2 (slightly) > pos 1

2 EQ 1

154
 Next situation

2 > pos 1
2 EQ 1

2 EQ 1
2 >neg 1

155
 Referential montage

2 > pos 1
2 >> pos 1

2 > pos 1
2 EQ 1

Here is input 1 always same: reference electrode 156

 Referentiale montage

2 > pos 1
2 >> pos 1

2>1
2 (slightly) > pos dan
1

157
 Referentiele montage

2 > + then 1
2 > + then 1

2 > + then 1
2 > + then 1

Contamination of reference 158

 Referential

2 more pos then 1

2 more pos then 1

2 more pos then 1

2 EQ 1

159
 Ideal ‘neutral’ reference ?
• An amplifier measures potential differences
• The ideal “neutral” reference does not
exist !!
• For high density EEG one can use a
mathematical mean of all individual
channels as averaged reference
• In practise most used are : biauriculair or
bimastoid.
160
(EKG artefact on the referential bimastoid:
assymetrical contamination in all channels)

161
 Output= input 1 – input 2

Input 2

Input 1

If inout 2 is more pos thenh input 1: deflection upward

162
 163
Impedance measurement
EEG registration

164
“Classical” EEG

Registration on paper

165
 Modern EEG machine
Paper registration

Transit from analog to

digital
166
digital
Modern EEG
Stroboscoop

“head”box

Screen

Keybord

Mouse (not alive)

167
 Reminder: before starting a
recording session (and during)
• Signal “garbage” in = analysis garbage “out”.
• Prepare a very good skin-electrode contact in
order to get a very low impedance (resistance to
AC: alternate current) (<5kOhm).
• Slight skin abrasion and application of electrolyte
gel to increase current.
• Impedance values at each electrode must be
monitored before and during registration.

168
Impedances

169
 Start a registration

Patient. Register 8 chan. Neurologist:

EEG interpretation.

170
 Amplification Technique
2001

1982: Prof. Dement

171
1990
172
173
Interpretation of the waveforms

174
Normal EEG

175
EEG Oscillations (rythmes)

176
 “Brainwaves”
• Alfa:
Alfa
– 8-13 Hz 30-80 µV (rest activity : eyes closed)
– occipitally localised
– “Disappearing” (desynchronisation) at eyes
opening: “visual halt reaction”.
– waxing en waining aspect
– Activated by stroboscopic stimulation (light
flashes at increasing frequency) fotostroboscopic
alfa drive - following”.
– Beta:
Beta
– 14-30 Hz—5-20 µV
– frontally.
177
 Slower waves
• Theta rythmes
– 4-7 Hz
– Temporal field (young people)
– Pathological if abundant or focal.
• Delta rythmes
– 0.5-3 Hz
– Always pathological except in young
children or during deep sleep stage III-IV.

178
179
 Brain oscillations
Alfa 8-13 50-100 µV Occipital Rest
Hz Eyes
closed
Beta 20-30 5-20 µV Frontal Mental
activity
Theta 4-7 Hz 30-70 µV Temporal Young

Delta 0-4 Hz 30-80 µV Diffuse Deep sleep

Mu 9-10 30-60 µV Parietal Motor

Hz activity
180
 Frequency
Timedomain domain

Alfa

Beta

181
 Frequency
Timedomain domain

Theta

Delta

182
 Frequency
Timedomain domain

Mu

183
 EEG: Science of artefacts
• Technical:
Technical
– Power supply noise (50 Hz)
– Electrodeartefacts (bad contacts)
– High impedance
– Other electrical devices (vb environment IC, respiration devices,
electrical matras, pacemaker, DBS, kine, ultra short waves, GSM….)
• Biological:
Biological
– Blink-artefacts: very important !!!!
– Eye movements (corneo-retinal dipole)
– Muscle-artefacts: high frequency (occipital, temporal,frontal, occipital)
– Transpiration-artefacts.
– ECG artefacts
– Ballisto-pletysmographic artefact
– Respiration, movement,…. 184
 EEG specialist
• A pattern recogniser !
• Trained by many years of experience in recognising
typical disease “patterns” in several instances of normal vs
pathological EEG (sleep: somnogram, wake, baby’s ,
siblings and young children , coma patients: alfacoma,
burst-suppression, epilepsia, artefacts…)
• Humans are good (“wet”) biological neural network
pattern recognition “devices” but have problems
evaluating small frequency variations and powerspectra
changes.
• Important: anesthesia, coma: power in slow frequencies
determines the evolution stages.

185
A neurological vision on the
EEG
Triphasic
complexes of
Creutzfeld Jacob

186
Normal EEG

187
Blink artefacts

188
 Normal

50 Hz

Blinks

Muscular

189
190
191
 Visual halt reaction
Eyes closed

Eyes open 192

Epileptic Attack

193
Epileptic Attack

194
Epileptic Spikes

195
Epileptic Attack ?

196
Blinkartefacts ?

197
Spike wave epileptic complexes 198
Epileptic Discharges ?

199
Epilepsia ?

200
Electrode artefact

201
Frontal Blinkartefacts

202
Eyeblink artefacts ?

203
FIRDA in metabolic encefalopathy

204
Burst suppression coma ?

205
ILS: Intermittent light stimulation

206
Lateralisation: Pathology !!!

207
Power supply artefact

208
Pletysmo artefact

209
Plethysmo artefact ?

210
Spike wave complexes

211
Trifasic waves

212
Rolandic Spikes

213
 Clinico-pathological cases
• A young man with criminal behaviour
– Frontal delta waves on EEG: frontal meningeoma
• “Hysterical” Arab lady (mutistic):
– General slow waves : postpartum trombosis of cerebral
sinus (sinus longitudinalis superior.)
• “A man acting strangely”:
– Epileptic focus left temporal lobe:
psychomotoric epilepsia malignant glioma.
• Hysterical behaviour in 3 men:
– TBC meningitis, carcinomateuse meningitis and a patient
with encafalitis.
214
 More ‘Iconography’

http://www.psychophysiology.blogspot.com/ 215
 Digital EEG: a new horizon
• Frequency analysis
• Brain mapping
• Evoked potentials
• Monitoring: Holter 24 h
• SleepEEG
• Video EEG
• Cortical EEG
216
Digitalisation

217
Computer and Multimedia
Clinico-Electrical correllations
Video

218
Video-EEG

219
Holter: registration >24h

II

220
 EEG telemetrics

Patients are allowed to move freely and more naturally

during registration.
221
 Frequency analysis
• The Fourier transform was a breakthrough.
• Dedicated DSP processors (digital filters)
and fast postprocessing software allowed to
evaluate the signals in real time.
• Frequency shifts are important in coma and
anesthesia (brainmonitor), monitoring, vb
CSA:
CSA Compressed Spectral Array.

222
 Digitalisation of the
signal

• Opens the world of DSP:

Digital Signal Processing

223
 A “typical” EEG analysis

Enormous line noise

contamination
On “analog” EEG
machines with in-pen
systems such would
result in messy stuff.
But digital EEG:
FILTERS !!

224
Notch filter removes 50 hz artefacts

In combination with high pass filters we also eliminate

the slow psychogalvanic oscillations: stabilises the
baseline ans allows “reading” of EEG.
225
FFT: Fast Fourier Transform

226
227
228
CSA
229
 Digital Computer Power

Calculate the powerspectra and “map” topologically. 230

Glioblastoma Multiforme

CT scan Brainmap

231
Activity in the delta band
Power Spectra Brain mapping

232
modern EEG....
233
A Remote view at the data over the internet

234
EEG mapping

235
 Fourier Analysis
Analyses the signal according to its
.frequency components

Transform the signal : x(t) from the time domain to

the frequency domain. 236
Fourier transform en inverse

237
Signal : 4 frequency components

x(t)=cos(2*pi*10*t)+cos(2*pi*25*t)+cos(2*pi*50*t)+cos(2*pi*100*t)
238
The FFT-transform

239
? What do we notice about that signal

It is generated by the Fourierformula

The FFT shows what frequencies are present in
the signalNOT when . they are appearing
The signal at hand is a perfectly “stationary”
signal: the composing frequencies are present
at each moment in time: itsstatistical
.proporties are time-invariant

240
 ? What about this signal

Evidently non-stationary because first low frequency

component and later higher frequency components.
241
Non stationary: 100 Hz—50 Hz—25 Hz—10 Hz

242
243
 Same FFT: but
essentially two
different signals

244
 !! Important
FT: only “valid” with stationary signals (signals
whose statistical proporties are time invariant.
(unless one is only concerned with presence of
(frequencies and not moment of appearance

? Is the EEG a stationary signal

!! NO
245
 Resolution problem
It is impossible to know what frequency
component is present at what precise
instance or point in time (compare to
(Heisenbergs uncertainty principle
We only measure the frequency components
within a certain time interval

246
Fourier transform and its inverse

X(t): is multiplied by an exponential term and integrated

over “all time”
247
 How to solve the stationarity
? problem
Chop the non-stationary signal in short
.“piecewise” stationary fragments
The divison depends on the length of the
.“chosen window “w
Multiplicate w(t) with x(t) for each t=t+1 en
.calculate the FT of this product

248
 S hortT imeF ourierAnalysis
Technique developed by Denis Gabor :STFT

The STFT “maps” a signal in a two-dimensional

function of time and frequency. 249
The Resolutionparadox: Heisenberg

Thesmaller the window the less frequency

components we can “trap” so the less
.frequency resolution we can obtain
Alonger window (called the “support”)
increases the freq.resolution but lowers the
(timeresolution.(limit is the classical FT

250
Problemparadox of STFT
:Narrow support

.bad frequency resolution

Large Support
,good frequency resolution
.bad time resolution
251
252
(Different support (windows lengths

w(t)=exp(-a*(t^2)/2); 253
 STFT

254
Good timeresolution but limited frequencyresolution
 Larger Window

Better frequency-
resolution but overlap
in timedomein

255
256
MRA:M ultiR esolutionAnalysis
Use a window with variable length:wavelet
.transform
The basic function is called the “mother”
.wavelet
Thee transform indroduces two new measures:
.scale andtranslation

257
 (Scale (s
“Compare to “scale
.on a map
Large scale: less resolution but good
.“overview”; small scale: better resolution
.Is the reciproke of frequency
Scale reflects a dilatation function: s>1:
.dilatation en s<1 is compression
258
 Translation
The way by which the supportwindow is slided over
. the signal
Comparable to STFT
Gives time information

259
 CWT
• Memo: The CWT is the integral over all time
of the signal x(t) multiplied by a translated
and scaled version of a certain wavelet
function..
step proces 5 •

Step 1:
Select a wavelet and compare it
to a section at the start of the
signal.
260
 CWT
Stap 2:
Calculate the transform, C, a measure of
similarity between the waveform and the
selected section of the signal x(t). A large C
means better similarity.

Note: result depends on the

selected wavelet
261
Relation between scale and frequency of a cos 262
signal
 CWT
Step 3: “translate” slide the wavelet to the
right and repeat step 1-2 untill end-of-
.signal

263
 CWT
Step 4: scale (dilate wavelet and repeat step 1-3

264
 :CWT
.Step 5: repeat for all values of s in the scale

265
 Algorithme
;S=1
Start
LOOP 1: s end of scale
LOOP 2: Tau 0 to end of signal
Calculate C: the WaveletTransform
Tau=tau+1
End LOOP2
(S=S+1 (dilatation off the motherwavelet
End LOOP 1
266
267
268
269
270
271
272
 Neurophysiological signals
High frequencies: mostly short time present
((burst
Low frequencies: mostly for a longer
.timespan present
Wavelet-analysis allows to capture both types

273
CWT vs DWT (discrete wavelet
(transform
With DWT one uses digital filter banks and
half pass filterbanks and subsampling:
.called subbandcoding
Better and more efficient calculations

274
Wavelet Analyse Vs. Fourier Analysis
:Fourier analysis •
Splits the signal in its sinusoidal frequency
.components

:Wavelet analysis •
Splits the signals intranslated andscaled
versions of the original (mother .) wavelet

275
 Modern Digital EEG
Opens a new window on brain
function
276
1965
Evoked Potentials

Latency <Duration>

Stimulus Response 277

 E
Evoked Potentials
are small

E
EP/ERP
G 278
We have to ‘extract’ them from the
EEG

Averagers (software)

279
Extraction averaging

280
A/ Stimulus Locked Event

281
B/ Independent from (random)
background

282
 Principle of the averager
• The algebraic summation of digitized epoch’s
(epoch: piece of EEG signal ) will out-average
the ‘random’ background (noise) (EEG) to
‘zero’ while the stimulus timelocked response
EP will be enhanced.
• Signal / noise ratio will increase

283
 Averaging

1 segment

EEG epoch

2 align

ENSEMBLE AVERAGING 284

Background is “averaged out”
Results in improved S/N ratio

285
Segments: 150 msec before stim/
800 post stim

epoch epoch epoch epoch epoch epoch

286
 After “ensemble” averaging

Evoked Potentials are substracted

from background 287
N1

N1
P2

P2
N1

P2
288
Evaluation to reference database (Duffy BEAM system)

289
290
 Labeling
peaks

N1 N400

- N2

P1 P2

+
P3a P600
291
P3 P3b
 We are in a generous mood !!!

To psychiatry !

To the neurologist

292
BERA: Brain Stem Auditory Evoked
Potential
293
294
Neurological
VEP

Stim: Flash or pattern

Registration: O1 - O2

295
VEP example

296
SSEP: somatosensory evoked potential
297
Electrodiagnosis
Registration of electrical potential
differences by means of superficial
(skin) electrodes at the scalp
Spontaneous activity: EEG.
– Evoked: EP
• Obligatory:
Obligatory neurological.
• Event related:
related psychofysiological.
Event Related Potentials
P300
CNV
MMN
N400
P50
P300
This is an auditory discrimination task.
– The patient must differentiate between 2
different tones.
• a frequent low-tone stimulus.
• a rare high freq stim. (randomly presented in
the testset +/- 15 tot 20 % of stimuli).
– React to “rare” stimulus by pressing a
contact.
 P300
Active (event Eyes open: fixation
Patient driven) attent
Auditive Frequent Rare
Stimulus 70 dB Low tone 80% High tone:20%
800 Hz --40 ms 1470 Hz --40 msec
ISI: 1 sec Random 30-50 epochs
Tijd distributio artefact free
n (averaging)
Auditive RT Rare
Taak
discrimination stimulus
Settings TC: 1 sec 33 Hz Fz Cz Pz
Registration technique
F
z

Cz

Pz

Rare
Frequent N1
N1 N2
SNA
P1 P3a P3b
P2 P2
 Dubble Averaging

Frequente stimulus
P3a P3b
Rare stimulus
 P-300 Complex
Neuro- Psychofysiological
N1

N2

SW

P1
P2

P3a P3b

P3
 Amplitudo

L
N1
A
P2
T
E N2
N
P3a
C
P3b
Y
SW
 Latencies

N1
N2 SNA

P2
P3a P3b

0 100 200 300 msec

Duration of the P300 complex

400 msec
 Normale waarden
Opgemaakt via ROC curven
Leeftijd
P3a P3b
Tot 34 j 6-26 250-315 7-25 290-390
µV msec µV msec

35-62 j 4-18 230-335 4-16 290-400

µV msec µV msec
 “Principles” of averaging
NEUROLOGY PSY

1. Time invariance
?
Stimulus and response are “time locked”
(fixed interval)
2. Amplitude Invariance
?
Each identical stimulus results in an
identical response (EP)
3. Noise Invariance
EP is independent on/of background ?
EEG activity (background activity is
considered random)
309
 The border neuro/psy = 10-100
msec
• Short latency EP’s are determined by the
Physical stimulus Characteristics but not
by the reaction of the patient (fi BERA,
Pattern VEP, SSEP..)
• Long latency potentials are more
determined by task related reaction of the
patient: Event Related Potentials (ERP)

310
 The border neuro/psy = 10-100
msec
• Short latency EP’s are determined by the
Physical stimulus Characteristics but not
by the reaction of the patient (fi BERA,
Pattern VEP, SSEP..)
• Long latency potentials are more
determined by task related reaction of the
patient: Event Related Potentials (ERP)

311
 The border neuro/psy = 10-100
msec
• Short latency EP’s are determined by the
Physical stimulus Characteristics but not
by the reaction of the patient (fi BERA,
Pattern VEP, SSEP..)
• Long latency potentials are more
determined by task related reaction of the
patient: Event Related Potentials (ERP)

312
 Task related ERP

ERP ‘s are multicomponent evoked responses and

attention, emotion (motivation) will tend to influence.
This can lead to violations of the first law of
neurological averaging: increased jitter between
stimulus and respons

PS: don’t forget: caused by increased time variance on stimulus side : complex
stimuli embedded in multimedia video or audio files typical in psychiatric and 313
neurocognitive test paradigms )
 Illustration
Identical EP but increased time
viariability : ‘jitter’

Average: “false” low

amplitudo

Stimulus

Average 314
 Consequence ? Neurological EP

• In ERP: beware of to much trust in averager output.

• We will need to focus more on individual responses
instead of only looking at global and “grand averages”
• New analysis paradigms are needed !

315
 Task related ERP

Violate the second law of neurological averaging:

identical stimuli stop generating responses with
identical amplitudo due to the dynamical
phenomenon of “habituation” 316
 Consequence ?
S/N versus dynamics

• Beware of “neurologically” averaging many

hundreds (or thousands ) of stimuli (trying to
improve the S/N ratio in order to create “nice”
stable EP’s: one could “scotomise” the fast
habituating responses in ERP paradigm’s and
loose the dynamics.
dynamics
• Go for machines that allow sequential averagers
317
 Auditory Evoked Potential: odd
ball paradigm
Fz Habituation is often
‘scotomised’ by
neurological averaging

N1

P2

Fast habituating phenomena are lost

318
 Hansen’s Axioma

THERE IS NO
SUBSTITUE FOR
CLEAN
CLEAN DATA
DATA
Jon Hansen (experimenter and technical manager at Steve Hillyards lab
319
UCSD)
 What about the third “law”
• Is the background independent of the
EP ?

320
 ERP can result from

• Evoked: stimulus evoked time locked

amplitudoincrease (EP).
• Induced: stimulus or task evoked partial or
full synchronisation of phase in a
background frequency band
• Combined

321
 Induced ERP’s ?
• Without fase synchronisation “invisible”
events to linear averagers”.

322
 Stimulus influencing background
EEG

323
NIHIL ???
 Testsituation: repetitive stimuli

Noise of noise + ??

Clearcut induced oscillation can NOT be visualised by ordinairy ensemble

averaging.

Output averager:nihil 324

 New tools are needed
• ERS/ ERD (Gert Pfurtscheller)
• EEG patterns that reflect true cortical activation/
focal ERD/ surround ERS fi in executive motor
imagery (basics of Brain Computer interface).
• PS fMRI BOLD signals reflect increased
metabolism but only TMS and ERS/ERD can
determine true cortical activation or inhibition

Another reason
why we need
EEG/ERP 325
 Peaks versus components
• The ERP peaks are
linear combinations
of the generator
activity at several
neuronal
assemblies
• “Peaks are not
important,
components are”
(Steve Luck)
326
 Peaks are a fuzzy projection of the
generating components

Sometimes it is easy….

327
 R. Ramirez, 2005
Sometimes more difficult: a plane ? A cross???
3 components A-B-C

329
In this experiment the activity in component C is increased

330
Same experiment adding extra attention by the patient: is the augmentation of
the peak a proof of further increased activity of component C ?

331
NO !!

332
333
 Moving from
Neurologic EP’s to
“psychiatric” ERP’s
• “Crossing the border”
border has been difficult and
disappointing for many !!
• Due to increased technical and biological variance in the
results of ERP they experienced a loss of “stability” ->
sensitivity and specificity that we have been so used to in
neurology (EP, spike wave...)
• As neurologists with our limited toolset we failed to
capture (or appreciate) the dynamics of those signals
and “turned” away from ERP to more gratifying medical
imaging techniques (PET, MRI, fMRI) and their - inviting-
good spatial resolution.
334
 Now we learned that
• The transition is possible and necessary but we
need to use (and get used to) our new tools
(ICA, Loreta,seq ave) in order to fully capture the
richness of EEG temporal dynamics in this first
second time frame.
• We need technical training and adhere to sound
principles of signal analysis and standardised
techniques in order to lower variance and
increase diagnostic and dynamic yield.
• We need a clear view and understanding of
clinical significance fi “trait versus state” markers
in ERP use.
335
Trait vs State

• Low voltage P300

– State marker in
depression
– Trait marker in
shidzophrenia
– Vulnerability marker in
first degree relatives of
alcoholics.
• We need good clinical
referral information !!!

336
337
LORETA analyse:
analyse inverse problem solution

338
 ERP image
• Each epoch is replaced by a line with each
pixel in a color matching the amplitudo on
that point.
• Each line is drawn sequentially stacked on
top of the previous
• One can spot trends as we can now “sort”
the trial according to whatever criterium
necessary (so not only the stimulus but
also patient reaction fi reaction time)
339
 ERP-Image Plotting
• Display single
trials as color-
coded horizontal
lines (e.g., red is
+µV, blue is -µV,
green is 0).

• Sort all trials

according to
some variable of
interest (here,
subject RT).

• Smooth
vertically.

340
2006 Jung et al., Human Brain Mapping, 200
ERSP (t.f) plots of ICA components

341
 Component analysis
N1 MMN
Fz
Fz

N1 P2

MMN

Cz Cz

P2
N1
MMN
Pz
Pz

P2
Component ERP
Slide 343

Auditory attended odd ball …

N2b or MMN ?

N1
MMN

N2b
Fz

P2

P3

Activation of N2-P3a components

Component ERP
 P3b
P3a

Constant N2a (MMN)

Habituation: N2b

344
 Technical progress
 Allows high density EEG analysis increasing
also the spatial granularity of source
localisation
Slide 346

ICA and LORETA: source

Localisation
 Technical progress
 Allows complementary integration between
neurophysiology and medical imaging by new filter
paradigms (fi based on ICA) filtering out
stimulation artefacts in real time.
 Technical progress
Allows integration of ERP and TMS opening new
perspectives in functional brain diagnosis and
therapeutics: age of functional neuronavigation.

Visor Courtesy by ANT

This is still EEG Jim, but not as we knew it

349
The Future ?

350
Integration with structural and functional imaging

351
 Modern developments
• Integration of visual and functional data
• EEG mappings superimposed on CT or
MRI images
• Morphology + Functional 3 D maps
• Time dynamics are added
• Psychological tasks can functionally be
traced in the brain
352
353
Bergers: Mirror of the brain ?

354
75 Y. Hemiparesis Ri post stroke

355
43j. Therapy-resistent epilepsia.

356
 The “ final question” (??):
Is Hans Berger’s dream realised
“EEG as a “Mirror of the brain”?
• Yes !? (No ??) we came a long way since
Berger thanks to FFT and CWT and even if
we have not captured the “soul” we have
increased our window span on the brain.
• Integration of EEG and fMRI data have
performed even better
357
 A first real mirror on the brain

Subject looks at 1750 natural

images
A voxel specific receptive
field theory was built taking
into account location,
intensity but also orientation
and spatial frequency
Inverse decoder ! information.

From brain pattern to Identification of 120 new

“meaning” images was 92% accurate 358
Seeing Your dreams ?

359
How do we proceed ?

360
 Neurofeedback
• Biofeedback via neuronal signals
• Already in use by “operant conditioning”
mechanismes in psychiatry and behaviour
psychology.
• The patient is asked by imaging certain
experiences and sensations or certain mental
activities to try to alter or influence certain signal
characteristics of his/hers EEG patterns (fi power
in occipital alpha bands with relaxation)
• Use to alleviate tension, stresscontrol, anxiety,
somatisation disorders.. etc..
361
 Neuronal Biofeedback

By Telemetry

Interaction with a VR
environment 362
Use in in psychotherapy

363
 B.C.I
Brain Computer Interface

364
 BCI: the final frontier
• Use the EEG as a control signal
• Direct interface from “thought” to
“effector” (robot, cursor, an object) without
interface of peripheral nerve or muscle
• Application field: tetraplegic patients,,
locked-in patients (conscious but cannot
move a single muscle)

365
 Extraction off significant “signal
characteristics (“features”)

Feed as input parameters to a

pattern recognition/classifyer system
366
367
368
 Challenge
• Non Invasive
• Non dependent on other bioelectrical
signals (eye movements, muscle,
peripheral nerve impulses).
• No external stimuli: only the “thinking”
about an action causes the action to
happen.
• Asynchroon: the system detects the
mental activity and reacts upon.
369
Recognition of stimulusfrequency: photic drive
370
 EEG as control signal
• One uses the EEG signal to acquire control
on an appropriate effector
• Two ways:
• Timedomain: amplitudo of a certain waveform vb
P300
• Frequency domain: extraction of certain
characteristics from defined frequency bands.

371
Use off P300 372
 Frequency Domain
• One a patient “thinks” about a movement
(imagines the movement without performing it
physically) this “mental activity” causes a
desynchronisation of the mu rythme over the
motor cortex of the contralateral hemisphere .
This is an internal event related
desynchronisation ERD
• One can capture decreasing (ERD) or increasing
(ERS) power at certain time of this movement
related event in certain frequency bands.
373
 Problems
• ERS/ERD are low amplitude signal events (2-10
µV) masked by abundant back ground activity in
the other frequency bands of the ongoing EEG
signal (typically at 50-100 µV.)
• Averaging of those postevents ERD is not a good
solution as these events are not phase correllated.
• What frequencies are best fitted to recognise arm
of leg movements ?

374
 Problems
• The ERD is often different in amplitudo and
location from patient to patient : so not
necessarily largest at the parieto-central
electrodes C4 C5.
• One can increase the spatial resolution by
increasing the number of electrodes around
that area on the scalp: higher electrode-
density.
375
Electrode-density

Motor Cortex

376
Registration

Movement:
captured by
EMG 377
ERD

378
Mu desynchronisation

379
380
381
Even play chess

382
 ERD of Mu ritme
• Maximally about 70-80 % accuracy
• It is important to look at all frequency
components from: 0-40 Hz
• Contains
• Mu ritme : 10 Hz
• Alfa: 8-13Hz
• Beta : 13-20 Hz
• Gamma: 38-42 Hz

383
The event: segmented

384
 What happens ?
• At rest: 10 Hz rythme parietal (µ rythme)
• Before and during the movement: ERD:
desynchronisation of the mu rythme and lowering of its
amplitudo. The “ power” goes down..
• Just before the effective movement a peak of ERS with
rising amplitudo (and power) in the gammaband (40
Hz).
• After the movement : power rises in 20 Hz beta band.
• Everything is embedded in ongoing beta (14-20 Hz) and
alfa (8-10 Hz) activity.
385
386
Gamma peak ERS

387
 Peak in the
ERS: peak in betaband
the gamma postmovement
band
premovement

388
 Time
Frequency map
Gamma-rythme:
ERS: Pre-movement

Beta-rythme:
ERD: during movement
ERS post 1 movement

Mu-Ritme:
ERS pre en post2 movement
en ERD during and post post1 389
Selection of epoch’s

390
 Kohonen oa…
• One constructs a “classifyer” fi a SVM or
Kohonen neural network that is trained at feature
vectors derived from the frequency patterns that
are evoked by certain mental activities.
• This neural network will the during the application
phase drive a cursor dependent on the
classification it makes from a certain state of brain
activity.

391
LVQ

392
393
BCI is a fascinating challenge

All interested and capable to

participate in this adventure are
welcome to join our group here at this
location (contact Prof Boullart)
394
But first for some well earned
sleep….

395