Professional Documents
Culture Documents
Nonlinear Adaptive Filtering of Stimulus Artifact
Nonlinear Adaptive Filtering of Stimulus Artifact
Abstract—Noninvasive measurements of somatosensory evoked distortion of the SEP, [4]. Fourth, the mechanisms by which
potentials have both clinical and research applications. The elec- the electrical stimulus is coupled into the recording are varied,
trical artifact which results from the stimulus is an interference complex and distributed, making minimization of the SA at the
which can distort the evoked signal, and introduce errors in
response onset timing estimation. Given that this interference is source problematic [5].
synchronous with the evoked signal, it cannot be reduced by the Given that the conventional methods are not effective for SA
conventional technique of ensemble averaging. The technique of reduction in SEP measurements, alternative approaches have to
adaptive noise cancelling has potential in this regard however, and be considered. A technique which has potential in this regard
has been used effectively in other similar problems. An adaptive is adaptive noise cancelling (ANC). In this approach an esti-
noise cancelling filter which uses a neural network as the adaptive
element is investigated in this application. The filter is imple- mate of the SA is obtained via an adaptive filter and the esti-
mented and performance determined in the cancelling of artifact mate subtracted from the composite SA plus SEP signal. ANC
for in vivo measurements on the median nerve. A technique of filters have been used in other biological signal acquisition ap-
segmented neural network training is proposed in which the plications. Ye and Choy [6] used such a filter for the reduction of
network is trained on that segment of the record time window respiratory artifact in rheopneumography measurements. Sada-
which does not contain the evoked signal. The neural network is
found to generalize well from this training to include the segment sivan and Narayana [7] developed an ANC filter for the reduc-
of the window containing the evoked signal. Both quantitative tion of electrooculogram in electroencephalography measure-
and qualitative measures show that significant stimulus artifact ments. Thakor and Zhu [8] have used an ANC filter for the re-
reduction is achieved. moval of the QRS complex in ECG in order to improve detec-
Index Terms—Adaptive, artifact, cancelling, evoked, filter, re- tion of the -wave. Suzuki et al. [9] developed a real-time ANC
sponse, stimulus. filter for suppression of ambient noise in lung sound measure-
ments.
I. INTRODUCTION In the application of noise cancelling to SA reduction,
McGill et al. [4] estimated the artifact waveform from sub-
and
III. METHODOLOGY
A block diagram of the measurement system for acquiring
Fig. 2. Feedforward neural network structure.
SEP from the median nerve is shown in Fig. 3. Median nerve
sensory fibers are stimulated at the finger and SEP acquired at
the wrist using surface Ag/AgCl electrodes. The resulting SEP
the first application to SA reduction, Grieve et al. [14] inves- and SA are amplified over the bandwidth 10 Hz to 10 kHz,
tigated the pi–sigma NN configuration with some success but sampled at a rate of 25.6 kHz, and the digital data stored for
concluded that a feedforward NN configuration would be more subsequent off-line processing. Following each stimulus, a data
appropriate. The feedforward multilayer perceptron (MLP) NN window of 10 ms or 256 samples is recorded. The Bruel and
will be considered in this paper for the ANC filter and SA can- Kjaer 2032 signal analyzer is used in an on-line ensemble aver-
cellation. aging mode in order to confirm the presence of SEP during the
The general configuration for the ANC filter is given in Fig. 1. recording session. In order to ensure no SEP crosstalk between
The primary channel input is the addition of evoked re- the primary and reference inputs to the ANC filter, a necessary
sponse , artifact , and uncorrelated noise . The reference condition for an ANC filter, a double-stimulus single-channel
channel input is the addition of artifact , correlated with technique is used. In this technique a series of alternating sub
through the system , and uncorrelated noise . The adap- and supra threshold stimuli are applied to the finger. The re-
tive filter estimates from and is adapted by the error sponses to the subthreshold stimuli are used as the reference
through an appropriate training algorithm. Following conver- data, and those to the suprathreshold as the primary data. Thus
gence of , the ANC filter output is the minimum mean no channel crosstalk is possible. This technique also has the ad-
square error estimate of . vantage of requiring only a single data acquisition channel and
Different structures can be used for the adaptive filter, associated hardware.
, and perhaps the most common is the linear FIR filter. Two Grass S11 stimulators with isolators are used to generate
However the structure of interest here is the NN adaptive the series of stimuli. Fig. 4 shows a typical series of stimuli and
filter because of its ability to generalize from training to test acquired data for this technique.
data, and to model nonlinear transfer functions of arbitrary
order. The general structure of a feedforward MLP neural
IV. EXPERIMENT 1
network is given in Fig. 2. For the implementation in this
paper, samples, , from form the In order to optimize the NN parameters and evaluate the NN
-dimensional input feature vector for the input layer. The ANC performance, the first experiment acquired data without
input layer is fully connected to the hidden layer through the the presence of SEP. In this way the cancellation performance
weights , where is the number of can be examined over the full SA duration, including the time
hidden nodes. The hidden layer is connected to the output layer interval in which the SA would otherwise overlap the SEP. If the
through the weights , and hyperbolic SEP were present and unknown it would be impossible to pre-
tangents are used for activation functions. The bias inputs and cisely evaluate the filter’s performance in cancelling the SA. At
GRIEVE et al.: NONLINEAR ADAPTIVE FILTERING OF STIMULUS ARTIFACT 391
Fig. 4. (a) Alternating sub and supra electrical stimuli and (b) resulting SA and SEP.
the same time the network parameters can be studied and op- training, the record time windows were divided into two seg-
timized—in particular network size and training data window ments, and , such that segment contains no SEP and ,
will be investigated. The methodology of Section III was used to which is the remainder, contains the SEP. This training method
collect 300 primary and reference records. However, for this ex- is referred to as segmented-training. Fig. 5 shows an example of
periment both reference and primary stimuli were subthreshold, the segments and . Due to the finite conduction velocity of
thus ensuring no SEP in either channel. the nerve fiber, there is always a time delay between the stimulus
Because the SA and SEP are correlated, it is essential to application and the onset of the SEP. This interval determines
ensure that during NN training with SEP present, the primary the magnitude of the time segment . The value of for prac-
channel training data not contain SEP. Otherwise the NN filter tical application can thus be based on the minimum expected
will adapt to cancel not only SA but also the SEP. Thus for nerve conduction delay between stimulus and SEP arrival at the
392 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 47, NO. 3, MARCH 2000
Fig. 6. Network size test on subthreshold stimulus data showing performance as a function of the number of input and hidden nodes.
REFERENCES
model nonlinearities of arbitrary order. A paired -test shows a
significant difference in their performances at the 5% signifi- [1] C. D. McGillem, J. I. Aunon, and D. G. Childers, “Signal processing in
evoked potential research: Applications of filtering and pattern recogni-
cance level. tion,” CRC Crit. Rev. Bioeng., vol. 9, pp. 225–265, Oct. 1981.
Fig. 8 shows the results of the SA cancellation for the NN and [2] S. B. Harrison and D. F. Lovely, “Identification of noise sources in sur-
NRLS FIR adaptive filters for three subjects which are typical of face recordings of spinal somatosensory evoked potentials,” Med. Biol.
Eng. Comput., vol. 33, pp. 299–305, 1995.
the six subjects. These figures allow a qualitative assessment of [3] D. Regan, Human Brain Electrophysiology. New York: Elsevier Sci-
the cancellation in the segment where the SA and ence, 1989.
SEP overlap, as well as in the segment . The relative [4] K. C. McGill, K. L. Cummins, B. B. Dorfman, L. J. Berlizot, K. Luetke-
meyer, D. G. Nishimura, and B. Widrow, “ On the nature and elimination
abilities of the NN and NRLS adaptive filters to generalize from of stimulus artifact in nerve signals evoked and recorded using surface
to are particularly evident in Fig. 8(b). electrodes,” IEEE Trans. Biomed. Eng., vol. BME-29, pp. 129–137, Feb.
1982.
[5] R. N. Scott, L. McLean, and P. A. Parker, “Stimulus artifact in so-
VI. CONCLUSIONS AND DISCUSSION matosensory evoked potential measurement,” Med. Biol. Eng. Comput.,
vol. 35, pp. 1–5, 1997.
The overlap of SA with SEP can cause significant signal dis- [6] J. Ye and T. T. Choy, “On-line respiratory artifact removal via adap-
tortion and onset timing errors. An ANC filter with a NN as tive FIR filters in rheopneumographic measurements,” Med. Biol. Eng.
the adaptive element was shown to provide substantial SA re- Comput., vol. 32, pp. 620–624, 1994.
[7] P. K. Sadasivan and D. N. Dutt, “A nonlinear estimation model for adap-
duction. An MLP with 12 input and six hidden nodes, and seg- tive minimization of EOG artefacts from EEG signals,” Int. J. Biomed.
mented training achieved a SA reduction in the segment of Comp., vol. 36, pp. 199– 207, 1994.
GRIEVE et al.: NONLINEAR ADAPTIVE FILTERING OF STIMULUS ARTIFACT 395
[8] N. V. Thakor and Y. S. Shu, “ Applications of adaptive filtering to ECG Philip A. Parker (S’70–M’73–SM’86) received the
analysis: Noise cancellation and arrhythmia detection,” IEEE Trans B.Sc. degree in electrical engineering from the Uni-
Biomed. Eng., vol. 38, pp. 785–793, Aug. 1991. versity of New Brunswick (UNB), Fredericton, NB,
[9] A. Suzuki, C. Sumi, K. Nakayama, and M. Mori, “Real-time adaptive Canada, in 1964, the M.Sc. degree from the Univer-
cancelling of ambient noise in lung sound measurements,” Med. Biol. sity of St. Andrews, St. Andrews, U.K., in 1966, and
Eng. Comput., vol. 33, pp. 704–708, 1995. the Ph.D. degree from the UNB in 1975.
[10] V. Parsa, P. A. Parker, and R. N. Scott, “Adaptive stimulus artifact re- In 1996, he joined the National Research Council
duction in noncortical somatosensory evoked potential studies,” IEEE of Canada as a Communications Officer and the fol-
Trans. Biomed. Eng., vol. 45, pp. 165–179, Feb. 1998. lowing year he joined the Institute of Biomedical En-
[11] , “Convergence characteristics of two algorithms in nonlinear stim- gineering, UNB, as a Research Associate. In 1976, he
ulus artifact cancellation for electrically evoked potential enhancement,” was appointed to the Department of Electrical Engi-
Med. Biol. Eng. Comput., vol. 36, pp. 1–13, 1998. neering, UNB, and currently holds the rank of Professor in that department. He
[12] W. G. Stevens, “The current-voltage relationship in human skin,” Med. is also a Research Consultant to the Institute of Biomedical Engineering, UNB.
Electron. Biol. Eng., vol. 1, pp. 389–399, 1963. His research interests are primarily in the area of biological signal processing.
[13] M. Yuwaraj, A. Gluzmann, P. Madsen, and H. Kunov, “Adaptive
stimulus artifact cancellation in otoacoustic emission testing,” in Proc.
20th Canadian Medical and Biological Engineering Conf., 1994, pp.
106–107. Bernard Hudgins (M’97) received the Ph.D. degree
[14] R. Grieve, P. A. Parker, and B. Hudgins, “Adaptive stimulus artifact can- from the University of New Brunswick (UNB), Fred-
cellation in biological signals using neural networks,” presented at the ericton, NB, Canada, in 1991.
17th IEEE EMBES Conf., Montreal, P.Q., Canada, Sept. 1995. He is currently a Senior Research Associate with
[15] S. Haykin, Neural Networks: A Comprehensive Foundation. New the Institute of Biomedical Engineering at UNB
York: Macmillan College, 1994. His primary research interests are in the area of
myoelectric signal processing for the control of
artificial limbs.
Dr. Hudgins was the recipient of a Whitaker
Foundation Investigator Award and recently spent
two years on a NATO workgroup assessing alterna-
tive technologies for cockpit applications. He has been the Region 7 (Canada)
representative on the IEEE EMBS Advisory Committee for the past two years.