ANALYSIS OF AUDITORY EVOKED POTENTIAL by VIKASH DAGA, B.E. A THESIS IN ELECTRICAL ENGINEERING Submitted to the Graduate Faculty of Texas Tech University in Partial Fulfillment of the Requirements for the Degree of IVIASTER OF SCIENCE IN ELECTRICAL ENGINEERING Approved December, 2002
52
Embed
ANALYSIS OF AUDITORY EVOKED POTENTIAL A THESIS IN ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
ANALYSIS OF AUDITORY EVOKED POTENTIAL
by
VIKASH DAGA, B.E.
A THESIS
IN
ELECTRICAL ENGINEERING
Submitted to the Graduate Faculty of Texas Tech University in
Partial Fulfillment of the Requirements for
the Degree of
IVIASTER OF SCIENCE
IN
ELECTRICAL ENGINEERING
Approved
December, 2002
^
z^
ACKNOWLEDGEMENTS
v-'L ^h laccn tharkfiil to my advisor Dr. Mary Baker for her help, support and guidance
throughout the project. I am grateful to Dr. Dwayne Paschall for providing the data set and
guiding me and agreeing to be in my graduate committee to evaluate my work. I am also
grateful to Dr. Thomas Trost and for being in my graduate committee and evaluating my
work.
I am thankful to all my friends who provide me support through out my work and
study and especially thankful to Sri Raja for guiding me in some bad times. I am thankful
to my parents and my family for their support without which I would not have been where I
am right now. It is because of their love, guidance and encouragement that I was able to
make it so far.
11
TABLE OF CONTENTS
ACKNOWLEDGEMENTS ii
LIST OF TABLES v
LIST OF FIGURES vi
CHAPTER
1. INTRODUCTION 1
1.1 Aim of the thesis 1
1.2 Introduction to EEG and ABR 1
1.3 Earlier work and thesis outline 2
2. AUDITORY BRAINSTEM RESPONSE 3
2.1 Electroencephalography 3
2.2 Auditory evoked potential 3
2.3 Auditory brainstem response 5
2.4 Instrumentation 6
2.5 Interpreting ABR waveforms 9
2.6 Absolute latency 11
2.7 Interwave latency intervals 12
2.8 Latency curve 13
2.9 Clinical application 14
3. WAVELET TRANSFORMATION 16
3.1 Mathematical transformation 16
iii
3.2 Fourier transform 16
3.3 Short-time Fourier transform 17
3.4 Wavelet transform 17
3.5 Scale 18
3.6 Time and frequency resolution 20
3.7 Discrete wavelet transform 23
4. PROCEDURE 27
4.1 Introduction of data set 27
4.2 Data collection 27
4.3 Preprocessing 29
4.4 Slope method 30
4.5 Wavelet transform 32
4.6 Spectrogram approach 35
5. RESULT AND CONCLUSION 36
5.1 Slope method 36
5.2 Wavelet analysis 36
5.3 Conclusion 41
5.4 Future work 41
REFERENCES 43
IV
LIST OF TABLES
2.1 Summary of auditory evoked potentials 4
2.2 Normal values for adult females 10
5.1 Average and standard deviation of wave V latency 41
LIST OF FIGURES
2.1 Neural generators of the ABR in humans 6
2.2 Electrode placement for ABR recording 7
2.3 Block diagram of ABR recording 10
2.4 Normal ABR waveform of an adult male 12
2.5 Latency curve 13
2.6 Latency curve comparison 15
3.1 A cosine wave for s =1 19
3.2 A cosine wave for s > 1 19
3.3 A cosine wave for s < 1 20
3.4 Time and frequency diagram for wavelet transform 21
3.5 Time and frequency diagram of STFT 22
3.6 Time and frequency diagram of Fourier transform 22
3.7 Different levels of wavelet decomposition of signal 26
4.1 10-20 Intemational system for electrode placement 28
4.2 Test details 29
4.3 A typical waveform 30
4.4 Slope method 32
4.5 Daubechie wavelets 33
VI
4.6 Reconstmcted waveform 34
4.7 Spectrogram 35
5.1 Reconstmcted waveform at different wavelet coefficients 37
5.2 Latency curve comparison 38
5.3 Wave V latency for 80 dB 38
5.4 Wave V latency for 60 dB 39
5.5 Wave V latency for 55 dB 39
5.6 Wave V latency for 40 dB 40
5.7 Wave V latency for 20 dB 40
Vll
CHAPTER 1
E^JTRODUCTION
1.1 Aim of the thesis
Auditory brainstem response is the response of the auditory nervous system to an
acoustic stimulus. It is characterized by a series of peaks. The time of occmrence of these
peaks has clinical importance in assessing hearing disorders of patients and identifying
the region of disorder. The proper identification of the occmrence of these peaks is very
difficult due to noise in the signal as well as background EEG signals. The aim of the
thesis is to identify one of the several peaks (Wave V) with the help of various signal
processing techniques.
1.2 Introduction to EEG and ABR
Electroencephalography measures electrical signals from brain, recorded with the
help of electrodes placed on the scalp. EEG recordings are used to diagnosis various
neurological disorders. The EEG signal can be analyzed visually by neuroscientists to
yield some information, but computerized methods have been developed to analyze the
signals to get a better understanding of the human brain.
Auditory brainstem response occurs in the first 10-15 ms of the auditory evoked
potential. Auditory brainstem response is used in the following:
1. estimating the level in different patients.
2. screening newboms who are at risk for hearing loss.
3. evaluate patients with suspected retrocochlear pathology.
The Auditory brainstem response or ABR as it is commonly known, is identified
by the presence of various peaks which are produced when the stimulus is encoded by
neurons along the ascending auditory pathway. The various peaks in the ABR are named
as wave I, II, III and so on in order of their occurtence. The most important peak out of
these peaks is wave V peak which is produced by lateral lemniscus [1]. These peaks can
be identified in normal cases but in abnormal cases these peaks are hidden by the
presence of noise. The diagnosis is done by looking at the time of occiurence of the peak
which is known as the latency of the peak.
1.3 Earlier work and thesis outline
Automatic recognition of wave V in the auditory brainstem response has been the
subject of research for the past few years, and a lot of work has been done in the field.
Various approaches like partem recognition and neural network and signal processing
techniques have been used for automatic detection of wave V peaks. Habrakan [2] used
neural networks to extract features to identify peak V in brainstem auditory evoked
potenfial. Wilson [3] used discrete wavelet analysis for the peak identification.
In this thesis. Chapter 2 describes about the evoked potential and gives a brief
description of auditory brainstem response. Chapter 3 describes the basics of wavelet
transformation and the reason for it being used in the thesis. Chapter 4 gives details about
the procedure followed, from data collection to processing. Chapter 5 summarizes the
result and conclusion as well as the future scope of the work.
CHAPTER 2
AUDITORY BRAINSTEM RESPONSE
2.1 Electroencephalography
Neurons exchange information in the form of electric signals. The recording of
this electrical activity of the neurons is called electroencephalography, or EEG. An
Austrian psychiatrist Hans Bergerin was the first to measure these signals in the late
1920s.
EEG signals are used for clinical and research purposes. The most common
method of recording an EEG signal is by placing surface electrodes on the scalp. The
electrical response to specific stimuli is known as the evoked potential. The evoked
potential can be a visual evoked potential caused by a flash of light or an auditory evoked
potential caused by an audio click.
2.2 Auditory evoked potential
Electrical potentials recorded due to an acoustic stimulus is known as an auditory
evoked potential. The auditory stimulus can be a click, tone burst, white noise, and
others. Latency is the time interval between the presenting of the stimuli and the
occurtence of the cortesponding response. Auditory evoked potentials are usually divided
into three time epochs each with different latencies. These are [4]:
1. fast response - latency of 0-10 ms,
2. middle response - latency of 10-50 ms and
3. slow response - latency of 50-500 ms.
Table 2.1 gives a list of all the auditory evoked potentials that occur in the first
500 ms after the onset of the stimulus along with their latencies, neural generators and the
places where the potentials are recorded [1].
Table 2.1 Summary of auditory evoked potentials [1]
Response Electrocochleography
Auditory Brainstem Response
Frequency Following Response
SNio Response
Middle Latency Response (MLR)
40 Hz Response
Late Potentials (Ni-P2 P300, CNV)
Latency 0.2-4.0 ms
1.5-10.0 ms
6-25 ms
8-12 ms
10-80 ms
Every 25 ms
80-500 ms
Recording Site Middle ear at the pronmontory
Vertex to earlobe or mastoid
Vertex to earlobe or mastoid
Vertex to earlobe or mastoid Vertex to earlobe or mastoid
Vertex to earlobe or mastoid Vertex to earlobe or mastoid
The ABR is a sequence of electrical potentials that are generated in the brainstem
and central auditory pathway in response to a stimulus in the ear. ABR occurs as the fast
response of the auditory evoked potenfial [4]. ABR reflects the transmission of the
stimuli through the brain stem auditory pathways. A normal waveform is usually
characterized by a series of 5 to 7 peaks that occur within 12 ms after the onset of the
stimulus. These peaks are labeled in roman numerals from wave I to wave VII. These
peaks are the response of the various neurons present in the brainstem auditory pathway
of the stimuli. The various neurons responsible for the peaks are these [1]:
a. Vlllth nerve for the wave I,
b. Lateral nemuscus for wave II,
c. Oliver bundle for wave III,
d. Lateral lemniscus for wave IV and
e. Inferior colliculus for wave V.
The most prominent peaks are the wave I, III, V. The latencies of these peaks for
normal persons is generally [1]:
a. 1.6 ms for wave I,
b. 3.7 ms for wave III and
c. 5.4 ms for wave V.
Figure 2.1 shows the ascending auditory pathway and the neural generators
responsible for the various peaks. The figure is that of cochlear nerve fibers and the
neurons responsible for the generafion of peaks are shown [4].
DCNi
A A A / T r i I
"""rt-
1 ' -"^ i ^.
MG
Figure 2.1 Neural generators of the ABR in humans (reproduced from "The Auditory Brainstem Response" edited by John T. Jacobson, P.27)
2.4 Instrumentation
An instrument that captures and stores auditory evoked potential has to perform
four major functions. These include generafing an acousfic stimulus, amplification of the
electric potential and averaging and storing the signal. These are explained in detail
below.
2.4.1 Generation of an acoustic stimulus and recording
The stimulus most often used in ABR recording is an acoustic stimulus, which is
a click with time duration of 100 microseconds. The intensity of the stimulus is measured
in decibels and ranges from 80 dB to 20 dB.
The ABR is recorded by artaching electrodes to the surface of the patient's scalp.
The most common method of clinical ABR recording uses four electrodes, one at the
vertex (referted as Gi), two electrodes at the inner surface of both the ear lobes or over
the mastoids (G2), and the ground electrode at the forehead. The electrode artangement is
shown in Figure 2.2. In the Intemational 10-20 system, which is most commonly used,
the Gi electrode position is called Cz, while the G2 positions are called Ai and A2,
respectively, for left and right ears.
Figure 2.2 Electrode placement for ABR recording (reproduced from "Auditory Evoked Potentials" by Lind J. Hood and Charles I. Berlin, P. 12)
Electrodes are placed after cleaning the skin and then placing conductive
electrode paste or gel on the skin and electrode. The electrode is securely attached with
tape. The impedance of the electrodes is measured to determine whether it is within the
acceptable limits.
2.4.2 Amplification
The amplifier used in measuring auditory evoked potential is a differential
amplifier. It amplifies only the difference voltage of the electrodes and cancels the
common voltage. This helps in reducing the noise from, for example, the building power
lines, which is common to both the electrodes. Evoked potentials have their amplitudes
in the order of 0.1-0.5 microvolt, so the gain of the amplifier is on the order of 100,000.
This increases the amplitude to a range where it can be better identified and analyzed [4].
2.4.3 Signal averaging
Before averaging the signal is passed through an analog-to-digital converter. A
signal averager is then used to increase the signal-to-noise ratio (SNR). The signal is
assumed to be a sum of ABR and noise. Let x(j) be the EEG signal, a(j) is the ABR signal
and n(j) is the noise where/ which varies from 1 to k (total number of sample points), is
the number of sampled points. The EEG signal can be written mathematically as
x(j) = a(j) + n(j) 2.1
and represents one recorded waveform. If w records are collected a matrix of ^ columns
and m rows can be formed. Each sampled point can be represented as:
8
MW = a(ij) + ri(ij) 2.2
where / goes from 1 to m. Since the ABR signal is identical to each of these records, die
equation can be rewritten as:
x(ij) = aO) + n(ijf 2.3
The noise is assumed to be random, so it will differ in each record and hence the index /
cannot be removed from the noise part. The SNR can be taken as the ratio of the RMS
energy of ABR signal and the RMS energy of noise. The RMS of ABR signal is aQ'), for
a particular value of time the ABR signal is constant. The noise is assumed to be of
Gaussian distribution with mean value zero and variance a^. The RMS value of n(ij)
equals its standard deviation a. So for a particular value of time the SNR of the record is
a(j)/ a. When the average of all the records is taken the variance of the noise, which is m
independent samples from the same distribution is a /m, and hence its RMS value is
a/m^''^. So the over all SNR of the ABR signal is m^''^ a(j)l a which is m^'^ times of the
SNR of a single record. Finally, the resultant waveform is stored in a digitized format in a
computer. The various fiinctions performed by the instmment are shown in Figure 2.3.
2.5 Interpreting ABR waveforms
Some well-defined peaks characterize an ABR waveform. Parameters associated
with these peaks which help in diagnosis of hearing disorders are (i) latency (in ms), (ii)
latency difference between primary peaks (interpeak latency), (iii) peak amplitude
(microvolt), (iv) I-V amplitude ratio, and (v) waveform morphology. The values of these
parameters depend on the age as well as the sex of the subjects. Table 2.2 shows values
of these parameters for aduh females [1].
Stimulus! Section ;
Earphones i i
Calibrated Attenuator
i I
Stimulus Generator
SUBJECT
1
i 1
» C\Bf,i tieCuuuc:]
' r
Amplifiers and filters
A/D Converter
1 r
Averager
!Recording and i Amplification 1 Section
1 Averaging • ^prt inn
CVA converter Computer
Plotter
Storing Section
Figure 2.3 Block diagram of ABR recording
Table 2.2 Normal values for adult females [1]
Parameter Presence of waveform components Expected latency at 75dB Wave I Wave III WaveV Interwave latency intervals Wave I-III Wave III-V Wave I-V Wave V latency-intensity function between 50 and 70 dB Wave V/I amplitude ratio
F-^:.^Jr:^--^<^::»^^^-^ T ^ 5 > i M r v ^ ^ ^^---^^K^V
'
IE r I 1 ' • " ' • * — • " • "
—•—wavelet identified 1 1 1
—•—clinically identified
1 3 5 7 9 11 13 15 17 Record
Figure 5.5 Wave V latency for 55 dB
12
0
Intensity 40 dB
-•—wavelet identified
••—clinically identified
" T ^' I r
1 2 3 4 5 6 7 8 9 1011 1213
Record
Figure 5.6 Wave V latency for 40dB
40
Intensity 20 dB
-•—wavelet identified
•—clinically identified
1 2 3 4 5 6 7 8 9 10
Record
Figure 5.7 Wave V latency for 20dB
Table 5.1 compares the average values and standard deviation of wave V latency
between clinically identified peaks and wavelet identified peaks.
Table 5.1 Average and standard deviation of wave V latency
Intensity 80 60 55 40 20
Clinically Identified Average Lat.
7.29 7.46 7.02 7.47 7.10
Standard dev. 0.97 0.70 0.55 0.81 0.70
Wavelet Resulted Average Lat.
7.36 7.92 6.92 7.32 6.91
Standard dev. 0.76 0.87 0.50 0.91 0.79
The other features, like the slope of the reconstructed waveform, the interpeak
latencies and the amplitude ratio yield no results. The spectrogram approach had shoyvn
some results but the values obtained fi-om the spectrogram could not be quantified and
hence it was a qualitative result and not quantitative.
41
5.3 Conclusion
The reconstructed wave of wavelet decomposition level 3 using Daubechie 5
mother wavelet can be used as feature to extract the wave V peak from the waveform.
Wilson [3] has shoym that the wavelet used does not significantiy effect peak
identification.
5.4 Future work
Along with one more features the automated identification accuracy of wave V
can be improved. The other wavelets can potentially be used to extract features for
identification. The spectrogram approach could be used to classify the cases into normal
and abnormal.
42
REFERENCES
1. Linda J. Hood, Charles I. Berlin, Auditory Evoked Potentials, Austm, Pro-ed, 1986.
2. Habraken JBA, van Gils MJ, Cluitmans PJM. Identification of Peak V in Brainstem Auditory Evoked Potentials with Neural Networks. Computers in Biology and Medicine, 23 (5), pp. 369-380, 1993.
3. Wayne J. Wilson, Mark Winter, Gill Kert and Farzin Aghdasi, "Signal Processing of the Auditory Brainstem Response: Investigating into the use of Discrete Wavelet Analysis", Procs. South African IEEE Symposium on Communications and Signal Processing - COMSIG'98, University of Cape Toym, Rondebosch, September 7-8 1998, pp 17-22.
4. John T. Jacobson, The Auditory Brainstem Response, College-Hill Press, Inc, San Diego, CA, 1985.
6. Ernest J. Moore, Bases of Auditory Brain-Stem Evoked Responses, Grune & Stratton, New York, 1983.
7- Barbara Burke Hubbard, The world According to Wavelets, A K Peters Wellesley, Wellesley, Ma, 1998.
8. Agostino Abbate, Casimer M. DeCusatis, Pankaj K. Das, Wavelets and Subbands, Birkhauser, Boston, MA, 2002.
9. J. Robert Boston, "Automated Interpretation of Brainstem Auditory Evoked Potentials: A Prototype System", IEEE Transactions on Biomedical Engineering, vol. 36, pp. 528-532, May 1989
10. Wayne J. Wilson, Mark Winter, Cartnel Nohr and Farzin Aghdasi, "Signal Processing of the Auditory Brainstem Response: Clinical effects of variations in Fast Fourier Transform Analysis", Procs. South African IEEE Symposium on Communications and Signal Processing - COMSIG'98, University of Cape Town, Rondebosch, September 7-8 1998, pp 23-28
11. M.J. van Gils. Peak Identification in Auditory Evoked Potentials using Artificial Neural Networks, Ph.D. Thesis, Eindhoven University of Technology, The Netherlands, 1995.
12. Wilson, W.J., and Aghdasi, F., "Discrete Wavelet Analysis of tiie Auditory Brainstem Response: Effects of Stimulus Intensity and Subject Age Gender and Test Ear", Proceedings of the IEEE AFRICON'99 conference, Vol. 1, pp. 291-296, September 1999.
13. Wilson, W.J., and Aghdasi, F., "Discrete Wavelet Analysis of the Auditory Brainstem Response: Effects of Stimulus Intensity and Subject Age Gender and Test Ear", Proceedings of the IEEE AFRICON'99 conference. Vol. 1, pp. 291-296, September 1999.
14. Hanrahan, H. E. (1990). Extraction of features in auditory brainstem response (ABR) signals. COMSIG 90, Proceedings of the third South African Conference on Communications and Signal Processing, IEEE catalog number 90TH0314-5/90, 61-66.