|
STIMULATION
ARTIFACT REMOVAL ALGORITHM FOR REAL-TIME SURFACE EMG APPLICATIONS |
|
T. Keller\
and M. R. Popovic** |
|
*ParaCare - Institute for Rehabilitation and
Research, |
SUMMARY
|
A direct and intuitive method to control a neuroprosthesis for grasping
is to use surface EMG (SEMG) activity of muscles that subjects can voluntary
control, e.g. in the case of C5 or C6 SCI subjects the deltoid muscles. The
measured voluntary SEMG activity in such applications is contaminated with
stimulation artifacts (SA) that are much higher in amplitude compared to raw
SEMG signals. Hence, to be able to use SEMG signals for control purposes one
has to remove the SA from the measured SEMG signal. In closed-loop applications
the SA can produce a positive feedback, which further stresses the importance
of removing the SA from the measured signal in close-loop SEMG control
applications.
Well-established SA removal techniques are artifact blanking and
filtering methods. Real-time SA blanking methods, either hardware built
sample-hold circuits or software blanking routines in digital processed SEMG
signals loose all EMG information during the blanking period. Especially with
current controlled stimulators, which have a very high output impedance, the
electrode-tissue impedance can cause a SA of several milliseconds. Most of the
SEMG SA filtering techniques are not viable in case of current stimulators
using surface stimulation electrodes, since the long lasting SA tail overlaps
in frequency and time domain with the voluntary SEMG activity.
A new method that encounters the randomness and stationarity of
voluntarily generated EMG is presented. An ensemble averaged SA with
exponential forgetting was subtracted from the recorded SEMG and an almost
artifact free SEMG signal was obtained. Measurements with multi-channel
stimulation patterns showed fast convergence of the algorithm. The algorithm
was significantly less sensitive to changes of the stimulation pulse amplitude
than to changes of the stimulation pulse width. The method can be implemented
in real-time applications and requires a low computational power.
STATE OF THE ART
SEMG signals that are recorded during surface functional electrical
stimulation (
Hardware /2, 3/ and software /1,
4/ artifact blanking or sample-and-hold blanking methods
blanked or sampled-and-held the SEMG during the SA while loosing all signal
information during that time.
SA filtering methods /5-8/ reduced the SA using linear, non-linear, or/and
adaptive filtering, gain switching, slew rate limiting, or constant
current/voltage switching techniques. Because the SEMG signal and the SA
overlapped in time and frequency domain, all applied filters influenced the
quality of the SEMG signal.
Software artifact subtraction methods /9-11/ subtracted a more or less pure SA from the mixed
SEMG. The presented methods differed in the way the pure SA was obtained. For
the control of neuroprostheses the proposed SA subtraction algorithms cannot be
used, because the produced SAs changed with the action (e.g. grasping or
releasing) over time and differed from a priori extracted SAs.
To overcome the above problems an enhanced ensemble averaged SA
subtraction method with real-time capabilities was developed.
MATERIAL AND METHODS
Algorithm
The SA was extracted from the first 125 samples (12.5 ms) post
stimuli of the recorded SEMG signal that lasted 500 samples (50 ms)
between two artifacts. A moving ensemble averaging algorithm with exponential
forgetting was used to extract the SA and the direct muscle responses. The
algorithm was deliberately kept very simple by applying a first order infinite
impulse response (IIR) filter for the exponential forgetting. For each sample n the following recursive filter output
was calculated:
, where p is the
weight that controls the forgetting. 
is the nth sample of the SEMG curve 
measured at time t and
is the extracted SA.
Small p values stand for fast
forgetting. The moving ensemble averaged SA (Y(t)) then was subtracted from the SEMG (X(t)). The algorithm did not process the SEMG from samples 126 to
499 post stimuli since it was always SA free.
Experiment
A COMPEX MOTION constant current stimulator provided a three channel
stimulation sequence that alternating opened and closed the subjects' hand.
COMPEX (5050MED) self-adhesive electrodes were used to stimulate the finger
extensors (channel 1) during hand opening, the finger flexors (channel 2), and
the thenar muscle (channel 3) during hand closing.
The stimulation frequency was 20 Hz.
Two COMPEX biofeedback sensors (gain: 1400, bandwidth: 100-4000 Hz)
were taped on the skin surface: one between the finger extensor stimulation
electrodes over the M. extensor carpi radialis, and one on the M. pars clavicularis
of the contralateral deltoid muscle. The sampling frequency was 10 kHz.
|
Figure 1: shows the stimulation and recording electrode locations. |
Table 1:
show the performed stimulation sequences with different transition times. |
Stimulation sequences similar to the one used by our neuroprosthesis for
grasping were applied to produce the time variant SAs. In each sequence the
hand was closed for 2 s and then opened for 2 s.
A trial consisted of eight concatenated stimulation sequences that
represented a typical grasping task with different transition times (see Table 1). When during the transitions the pulse width was
changed (between 0 and 250 ms) it is marked with (PW) and when the pulse amplitude was changed (between 0
and 12 mA / 8 mA)
it is marked with (AMP). Two such trials were conduced, one without and one
with voluntary muscle contraction.
Signal Processing
The raw SEMG recording between two stimuli was divided into two parts
(see Figure
2):
A the SA, 12.5 ms long
B the remaining SA free part
curve that was not processed
Figure 2: The data processing steps are shown for
the SA removal method and the result is compared to normal voluntary
contraction for SEMG signals recorded over the wrist extensor muscles.
The recorded signals were processed as follows (see Figure 2):
1. Part A (first
125 samples) (curve „) were cut from the
raw SEMG signal for each stimuli
2. The moving ensemble average algorithm provided
curve … that was
3. subtracted from „ and resulted in curve †.
4. The result from step 3 was concatenated with
the SA free part B (curve
‡) and the first 3 ms after stimulus containing residual SAs were
blanked (darker shaded in Figure
2).
RESULTS
During the constant stimulation phases of 2 s, the SAs were almost
completely eliminated from the recorded SEMG signals for both electrode
locations. In the processed wrist extensor SEMG signal only a few residual SA
spikes during stimuli were left (see Figure 2, curve ‚). The rest of the curve was SA free. The SA
recorded on the deltoid muscle occurred only during the stimuli. No SA tail was
produced. It has to be mentioned that in Figures 3 and 4 only the first
12.5 ms post stimuli are concatenated. The SA free part B is not shown.
|
Figure 3: The SA that was recorded between the stimulation electrodes on the M.
ext. carpi radialis
changed strongly during transitions for changing PWs. |
Figure 4: Even very low stimulation amplitudes produced similar SA between the
stimulation electrodes for constant (250 µs) PWs |
The transitions from hand opening to hand closing or vice versa were
more problematic. If the pulse widths were rapidly changed and shorter than 100
ms, then the SA changed dramatically from pulse to pulse (see Figure 3) and caused errors in the estimation of the SA.
Trials with different forgetting weights p
could not reduce this effect. Optimal results were obtained with a forgetting
weight p = 1.
If the pulse amplitude
was changed the SA remained almost the same, even for very low amplitudes. Here
also a forgetting weight p = 1 was
optimal.
DISCUSSION
A novel SA removal method for real-time applications was presented. The
algorithm subtracted a moving ensemble averaged SA with exponential forgetting
from the SA contaminated SEMG of a voluntary activated muscle. The algorithm
was capable of eliminating SA tails in presence of voluntary SEMG activity,
even if the SA shapes were changing due to changing stimuli. The stimulation
spikes could not be eliminated. We suggest blanking the signal during that
saturated period (see dark shaded region in Figure
2). For fast transitions with pulse amplitude modulated
stimulation pattern (PW 250 ms) it could be shown that the SA removal
performance remained good. Fast changing stimulation pulse widths during transitions
produced SA tails that could not be removed, because the SA changed strongly
from pulse to pulse.
REFERENCES
/1/
T. Keller, A. Curt, M. R. Popovic, V. Dietz, and A.
Signer, Grasping in High Lesioned Tetraplegic
Subjects Using the EMG Controlled Neuroprosthesis, J. NeuroRehab., vol. 10, pp. 251-255,
1998.
/2/ J. A. Freeman, An
electronic stimulus artifact suppressor, EEG
and Clinic. Neurophysiol., vol. 31, pp. 170-2,
1971.
/3/ J. Minzly,
J. Mizrahi, N. Hakim, and
A. Liberson, Stimulus artefact
suppressor for EMG recording during FES by a constant-current stimulator, Med. Biol. Eng. Comput,
vol. 31, pp. 72-5, 1993.
/4/ T. Handa,
H. Takahashi, C. Saito, Y. Handa, M. Ichie, J. Kameyama, and N. Hoshimiya, Development of an FES system controlled by EMG
signals, Proc. of Int. Conf. IEEE EMBS, 1990.
/5/ M. Solomonow,
R. Baratta, T. Miwa, H. Shoji, and R. D. Ambrosia, A technique for recording the
EMG of electrically stimulated skeletal muscle, Orthopedics, vol. 8, pp. 492-5, 1985.
/6/ C. M. Epstein, A simple
artifact-rejection preamplifier for clinical neurophysiology,
Am. J. EEG Techn.,
vol. 35, pp. 64-71, 1995.
/7/ F. Del Pozo and J. M. R. Delgado, Hybrid stimulator for chronic
experiments, IEEE Trans. Biomed. Eng , vol. BME-25, pp. 92-4, 1978.
/8/ V. Parsa,
P. Parker, and R. Scott, Convergence characteristics of two algorithms in
non-linear stimulus artefact cancellation for
electrically evoked potential enhancement, Med.
Biol. Eng. Comput., vol. 36, pp. 202-14, 1998.
/9/ K. C. McGill, K. L.
Cummins, L. J. Dorfman, and B. B. Berlizot,
On the nature and elimination of stimulus artifact in nerve signals evoked and
recorded using surface electrodes, IEEE
Trans. Biomed. Eng., vol. 29, pp. 129-37, 1982.
/10/ T. Blogg
and W. D. Reid, A digital technique for stimulus artifact reduction, EEG and Clinic. Neurophysiol.,
vol. 76, pp. 557-61, 1990.
/11/ T. Wichmann,
A digital averaging method for removal of stimulus artifacts in neurophysiologic experiments, J. Neurosci. Method., vol. 98, pp. 57-62,
2000.
ACKNOWLEDGEMENTS
This project was supported by grants from the Federal Commission for
Technology and Innovation, Switzerland - Project No. 4891.1 and the Swiss
National Science Foundation - Project No. 5002-057811
AUTHORS ADDRESS
|
Dipl. Ing. Thierry
Keller |
e-mail: kellert@balgrist.unizh.ch home page:
http://www.aut.ee.ethz.ch/~fes/ |