ARO Mid-Winter Meeting, 2001 1 Analysis of Evoked ...cansl.isr.umd.edu/simonlab/pubs/ARO01.pdfMURI #...

ARO Mid-Winter Meeting, 2001 1

Institute for Systems ResearchUniversity of Maryland

Center for Auditory and Acoustic Research

Analysis of Evoked Potentials on the Surface of Primary Auditory Cortex Combining Ripples and ICA

Institute for Systems Research & Department of Electical and Computer Engineering

University of MarylandSupported by:

MURI # N00014-97-1-05001 from the office of Naval Research# NIDCD T32 DC00046-01 from the National Institute on Deafness and Other Communication Disorders

# NSFD CD8803012 from the National Science Foundation American Society of Electrical Engineers

This poster is available at <http://www.isr.umd.edu/CAAR/pubs.html>

Nikolaos Kanlis Didier DepireuxJonathan Simon Shihab Shamma




Synopsis

EEG Recording made with 24 channels in 1mm on AI surface

Physiological Sources of Neural Activity seperated among channels using ICA (Independent Component Analysis).

Pure tone stimuli: Separated sources reveal tonotopy

Moving Ripple Stimuli: Separated sources reveal traveling waves with:

- Direction given by tonotopy

- Phase-locking to ripple speed

2




Recording DeviceThin-film micro-electrode

array, developed by Anthony Owens and Shihab Shammarecording Evoked Potentials (EPs).

24 gold electrodes (40x40µm ) sandwiched between two layers of biocompatible polyimide.

Rests directly on cortex surface.

Flexible enough to conform to the shape of the cortex

Simultaneous recording, independent of the state of the animal and the level of anesthesia.

2

1 2 3 54

6 7 98

11

10

12 13 14

16

15

17 18

21

19

22 23

20

24 25

Ventral

Dorsal

Caudal

Rostral

AI

1mm




24 Channels

EKG Measurement Gain-Normalization

3 Amplification Stages

Optically-Coupled Isolation Main Stage Amplifier.

60Hz Suppression by Aluminum Shielding & Averaging

Experimental Setup




StimuliData were collected from a total of 7 anesthetized domestic Ferrets (Mustela Putorius), in response to monaurally presented acoustic stimuli.

Tone stimuli were 200 ms in duration, with 10 ms rise/fall time, at various frequencies ranging from 2 to 11 kHz with intensity ranging from 65 to 85 dB.

Moving Ripples were 1700ms in duration, with 8ms rise/fall time, repeating every 4.2 secs,with intensity ranging from 65 to 85 dB also. Ripples were chosen because they are dynamic and broadbandas most natural sounds, with the ability to match the different cell Receptive Fields (elaborate description is given on Slide 13.)




Goal: to decompose the recorded EPs in the responses of separate / independent populations of neurons by using some criterion ofseparability that can be related to the neurophysiology.

Evoked PotentialsUnlike unit activity, Evoked Potentials (EPs) consist primarily of

synaptic activity. Peaks of the EPs appearing later than 10ms post-stimulus, are thought to reflect auditory processing by different neural centers/structures (neocortex and possibly subcortical structures).

Ele

ctri

cal A

ctiv

ity (m

Vol

t)

Electrode #8

200 400

-0.5

0

0.5

1

1.5

200 400

-0.5

0

0.5

1

1.5

Electrode #17




Information OverlapThe simultaneous multi-electrode, recording-from-the-surface

approach, records overlapping signals from closely placed electrode sites.

Each electrode receives signals not only from neurons directly underneath, but is a weighted sum (linear mixture) of the activity of various separate/independent distributed neurophysiologicalnetworks.

We approximate the mixing process as simultaneous (un-filtered, un-delayed).




Linear ICAIndependent Component Analysis (ICA), both

(1) unmixes the separate sources’ activity, and(2) reduces the information overlap between electrodes

• Model:Instantaneous Linear Mixing

X(t) = A*S(t)A*W = P*D*I

P: Permutation MatrixD: Diagonal Scaling MatrixI: Identity Matrix

• Method:Estimate weights so to maximize outputentropy H(y) => minimize mutual information I(y)

• Goal:Learned weights approximate inverse of mixing matrix




ICA Example on Speech

0 0.5 1 1.5 2

-5

0

5

0 0.5 1 1.5 2

-5

0

5

0 0.5 1 1.5 2

-5

0

5

Even then, if she took one step forward he could catch her (Female)

She had your dark suit in greasy wash water all year (Male)

She had your dark suit in greasy wash water all year (Female)

0 0.5 1 1.5 2

-1-0.5

0

0.51

0 0.5 1 1.5 2

-1

0

1

0 0.5 1 1.5 2

-1

0

1

0 0.5 1 1.5 2

-5

0

5

0 0.5 1 1.5 2

-5

0

5

0 0.5 1 1.5 2

-5

0

5

ORIGINAL SOURCES

RECOVERED SOURCES

Linear

ICA

MIXED SOURCES

Linear Mixing Process

1.0

1.01.0

1.4 0.61.2

1.60.81.8




Pure Tone Data I

experiment 184s – recording 1tone1to4k5s

1000 2000-10123 # 1

1000 2000-10123 # 2

1000 2000-10123 # 3

1000 2000-10123 # 4

1000 2000-10123 # 5

1000 2000-10123 # 6

1000 2000-10123 # 7

1000 2000-10123 # 8

1000 2000-10123 # 9

1000 2000-10123 # 10

1000 2000-10123 # 11

1000 2000-10123 # 12

1000 2000-10123 # 13

1000 2000-10123 # 14

1000 2000-10123 # 15

1000 2000-10123 # 16

1000 2000-10123 # 17

1000 2000-10123 # 19

1000 2000-10123 # 18

Single Tones 1.0KHz – 1.4KHz – 2.0KHz – 2.8KHz – 4.0KHz Concatenated




Pure Tone Analysis19 Independent Components with Contribution Percentage Scores

1000 2000-505

1015 # 1

21.8895

1000 2000-505

1015 # 2

20.2003

1000 2000-505

1015 # 3

11.3979

1000 2000-505

1015 # 4

10.8803

1000 2000-505

1015 # 5

6.5295

1000 2000-505

1015 # 6

4.9273

1000 2000-505

1015 # 7

3.8556

1000 2000-505

1015 # 8

3.753

1000 2000-505

1015 # 9

3.6906

1000 2000-505

1015 # 10

3.1043

1000 2000-505

1015 # 11

2.5716

1000 2000-505

1015 # 12

2.1589

1000 2000-505

1015 # 13

1.9297

1000 2000-505

1015 # 14

1.4423

1000 2000-505

1015 # 15

0.86589

1000 2000-505

1015 # 16

0.32142

1000 2000-505

1015 # 17

0.20025

1000 2000-505

1015 # 18

0.14242

1000 2000-505

1015 # 19

0.1392

4 2 65

73 1




Spatial Maps I

#1 min = 5 max = 61

#2 min = 2 max = 64

#3 min = 5 max = 34

#5 min = 4 max = 35

#8 min = 3 max = 29

#7 min = -2 max = 30

1

7 5

4

32

6

1.0KHz 1.4KHz

2.0KHz

2.8KHz4.0KHz

Tonotopic Organization revealed/verified by surface

methods.




Stimuli - Moving Ripples

20

where log ( ) FxF

=

0F

( )( , ) 1 sin 2S x t A w t xπ= + ∆ ⋅ ⋅ ⋅ + Ω⋅ + Φ

= 4Hz and = 0.4/0.8/1.6 cyc/octw ± ΩParameters:




Moving Ripple DataMoving Ripple: w = 4Hz, Ω = 0.4 cycles/octave

2000 4000-0.5

0.51

0

# 1

2000 4000-0.5

00.5

1 # 2

2000 4000-0.5

00.5

1 # 3

2000 4000-0.5

00.5

1 # 4

2000 4000-0.5

00.5

1 # 5

2000 4000-0.5

00.5

1 # 6

2000 4000-0.5

00.5

1 # 7

2000 4000-0.5

00.5

1 # 8

2000 4000-0.5

00.5

1 # 9

2000 4000-0.5

00.5

1 # 10

2000 4000-0.5

00.5

1 # 11

2000 4000-0.5

00.5

1 # 12

2000 4000-0.5

00.5

1 # 13

2000 4000-0.5

00.5

1 # 14

2000 4000-0.5

00.5

1 # 15

2000 4000-0.5

00.5

1 # 16

2000 4000-0.5

00.5

1 # 17

2000 4000-0.5

00.5

1 # 18

experiment 184s – recording mr.04.w4




Moving Ripple Analysis

2000 4000-10

0

10

20# 1

19.1757

2000 4000-10

0

10

20# 2

12.6825

2000 4000-10

0

10

20# 3

10.751

2000 4000-10

0

10

20# 4

10.3894

2000 4000-10

0

10

20# 5

9.3473

2000 4000-10

0

10

20# 6

6.63

2000 4000-10

0

10

20# 7

6.5014

2000 4000-10

0

10

20# 8

6.2708

2000 4000-10

0

10

20# 9

5.4353

2000 4000-10

0

10

20# 10

4.2562

2000 4000-10

0

10

20# 11

2.1397

2000 4000-10

0

10

20# 12

1.549

2000 4000-10

0

10

20# 13

1.3834

2000 4000-10

0

10

20# 14

1.2461

2000 4000-10

0

10

20# 15

1.224

2000 4000-10

0

10

20# 16

0.53606

6 2

3

5

1

4




Spectral Analysis

0 50

10

20

30phase = 26 conf = 55%

0 50

10

20

30phase = -8 conf = 68%

0 50

20406080

100120140 phase = -158 conf = 69%

0 50

10

20

30phase = 72 conf = 81%

0 50

50

100

150phase = 53 conf = 93%

0 50

5

10

15

20 phase = -46 conf = 56%

0 50

20406080

100120 phase = -67 conf = 86%

0 50

50

100

150phase = -103 conf = 70%

0 50

10

20

30

phase = 14 conf = 58%

0 50

20

40

60

phase = -166 conf = 85%

0 50

510152025

phase = 123 conf = 69%

0 50

20

40

60

phase = -56 conf = 64%

0 50

2468

101214 phase = -16 conf = 45%

0 50

20406080

100120140 phase = 10 conf = 82%

0 50

5

10

15phase = -121 conf = 35%

0 50

1020304050 phase = 59 conf = 83%

power in 4Hz harmonic setConfidence = , Criterion for RED: conf > 80%total power




Spatial Maps II

#4 min = -3 max = 47

#5 min = - 39 max = 2#7 min = 2 max = 43

#10 min = - 29 max = 6

#14 min = -10 max = 22

#16 min = -4 max = 15

Traveling wavesuccession




Reconstructed Responses

-0.05

0

0.05 # 1

-0.05

0

0.05 # 2

-0.05

0

0.05 # 3

-0.05

0

0.05 # 4

-0.05

0

0.05 # 5

-0.05

0

0.05 # 6

-0.05

0

0.05 # 7

-0.05

0

0.05 # 8

-0.05

0

0.05 # 9

-0.05

0

0.05 # 10

-0.05

0

0.05 # 11

-0.05

0

0.05 # 12

-0.05

0

0.05 # 13

-0.05

0

0.05 # 14

-0.05

0

0.05 # 15

-0.05

0

0.05 # 16

-0.05

0

0.05 # 17

-0.05

0

0.05 # 18

-0.5

0

0.5

1

50 1000 1850

50 1000 1850 50 1000 185050 1000 185050 1000 1850

50 1000 185050 1000 185050 1000 1850

50 1000 185050 1000 185050 1000 185050 1000 1850

50 1000 185050 1000 185050 1000 185050 1000 1850

50 1000 1850

50 1000 185050 1000 1850

Reference4Hz

Reconstructed Responses

smoothed up to the 5th 4Hz harmonic.




Moving Wave Movie




DiscussionTonotopic organization revealed from the surface of the cortex, as

documented to be the case (Shamma et. al. 1993) with single unit recordings, from sub-cortical layers.

ICA identified components of physiological importance, unlike the orthogonal componets extracted by PCA used in previous studies (Barth & Di, 1990).

ICA performed on moving-ripple responses, revealed components with phase-locking to the ripple speed.

Response reconstruction using phase-locked components is done in a coherent way, so that traveling wave on AI was revealed.

ICA in this case, can considered as another way of “filtering”,helping remove noisy sources




Future ConsiderationsSystematic recordings of moving ripples, so that surface

SpectroTemporal Responsive Fields (STRF) can be constructed as for single-neuron STRFs.

Application of convolutive-ICA that compensates for delays & filtering in the mixing process




Selected ReferencesMicro-Electrode ArrayOwens L.A., Denison J.T., Versnel H., Rebbert M., Peckerar M. & Shamma A.S., “Multi-

electrode array for measuring evoked potentials from surface of ferret primary auditory cortex”, J. of Neuroscience Methods, vol.58, pp.209-220, 1995.Independent Component Analysis and Applications

Bell A.J. and Sejnowski T.J., “An information maximization approach to blind separationand blind deconvolution”, Neural Computation, vol.7(6), pp.1129-1159, 1996.

Kari Torkkola, “Blind Separation of Convoluted Sources Based on Information Maximization”, Proc. of IEEE Workshop on Neural Networks for Signal Processing NNSP-96, Kyoto,Japan.

T.-P. Jung, S. Makeig, T.-W. Lee, M. McKeown, G. Brown, A. J. Bell, T. J. Sejnowski. “Independent Component Analysis of Biomedical Signals”, International Workshop ICA'00, Helsinki, pp.633-644, June 2000.Auditory Cortex and Rippled Spectra Stimuli

Daniel S. Barth and Di Shi, “Three dimensional analysis of auditory-evoked potentials in rat neocortex”, J. Neurophysiology, vol.64, pp.1527-1536, 1990.

Shamma A.S., Fleshman, J.W., Wiser P.R. & Versnel H., “Organization of response areas in ferret primary auditory cortex”, J. Neurophys., vol.69(2), pp.367-383, 1993.

Shamma A.S., Versnel H., Kowalski N., “Ripple Analysis in Ferret Primary Auditory Cortex. I. Response Characteristics of Single Units to Sinusoidally Rippled Spectra”, Auditory Neuroscience, vol.1, pp.233-254, 1995.

Depireux A.D., Simon Z.J and Shamma A.S., “Measuring the dynamics of neural responsesin primary auditory cortex”, Comment in Theoretical Biology, 1998

ARO Mid-Winter Meeting, 2001 1 Analysis of Evoked ...cansl.isr.umd.edu/simonlab/pubs/ARO01.pdfMURI #...

Documents

Transcript of ARO Mid-Winter Meeting, 2001 1 Analysis of Evoked ...cansl.isr.umd.edu/simonlab/pubs/ARO01.pdfMURI #...