PDA

View Full Version : Summary: Removal of heartbeat signals from EMG recordings



Zvika Ben-haim
03-05-1998, 07:59 AM
First of all, we would like to sincerely thank the many people who
responded to our question. The responses were serious and well-focused,
and have helped us a great deal in attempting to solve our problem.
Attached is a summary of the responses we received for our query.

Original Query
> I am interested to know what is the preferred method to remove heart
> beat signals from EMG recordings of trunk muscles. with minimum loss
or
> distortion of the EMG signal itself. A summary of the responses I will

> submit to the list members.
(Original question posted by Dr. Ruth Dickstein)

Summary of responses on removal of ECG artifacts from EMG signals
Highpass Filtering: The most common reply was to use some form of
highpass filtering, with suggested cutoffs ranging between 1 and 40 Hz.
Many suggested performing an FFT and zeroing the lower frequencies, then
performing an inverse FFT, in order to obtain a sharper cutoff than by
using ordinary analog filters. An article describing the trade-off in
choice of cutoff frequency is: Redfern et al, "High-pass filtering to
remove electrocardiographic interference from torso EMG recordings,"
Clin. Biomech. 1993; 8; 44-48.

Gating: A simple and usually satisfactory method is known as gating. The
idea is that values above a trigger value are assumed to be heartbeat
signals, and are treated as missing values. This has the advantage of
fast and automatic data processing. The gating should be performed
before integration or any other method of smoothing to avoid data loss.

Subtraction: Another common reply suggested subtracting the data of a
'clean' heartbeat from all occurances of heartbeats in the data. Better
results can apparently be achieved by selecting as a 'clean' heartbeat
an average 'template' of many heartbeats recorded with no muscle
activity.

Bartolo et al (see "Analysis of diaphragm EMG signals: comparison of
gating vs. subtraction for removal of ECG contamination", J. Appl.
Physiol. 80(6); 1898-1902) compared the gating and subtraction methods
using several common numerical processing techniques, such as mean
power, zero crossing frequency and so on. They found no significant
difference in results between signals processed using these two methods,
and concluded that the added computational power required to use the
more complex method of subtraction does not yield significantly better
results.

A combination of the subtraction and filtering techniques was also
suggested. In this method the subtraction would be performed in the
frequency domain (on an FFT signal). In other words, the FFT of a clean
heartbeat would be subtrated from the FFT of the data, and the result
converted back using an inverse FFT. As far as we know this has not been
attempted yet with EMG signals.

Wavelets: Another suggestion was to use wavelet smoothing to remove
heartbeat signal frequencies without distorting EMG data at the same
frequency. For a technical introduction to wavelet theory, see Chui
(1994): "Wavelets: Theory, Algorithms and Applications", San Diego:
Academic Press.

Adaptive Noise Cancellation: Using adaptive noise cancellation with an
extra channel in which only the ECG is collected. For details on this
method see Medical Engineering and Physics, 1997, Vol 19 No 3 pp
279-285, "A comparison of ECG cancellation techniques applied to the
surface recording of somatosensory evoked potentials", J. S. Black and
D. F. Lovely.

Use of histograms: Finding a histogram of the EMG amplitudes, and
removing (or keeping as missing values) all amplitudes which clearly do
not belong to the EMG.

Improved electrode placement: Positioning the ground electrode in such a
manner that the heartbeat reaches the ground and active electrodes
simultaneously and does not get recorded.


Thanks again to all who responded to our question.

Sincerely,

Zvika Ben-Haim
zvikabh@aluf.technion.ac.il




First of all, we would like to sincerely thank the many people who responded
to our question. The responses were serious and well-focused, and have
helped us a great deal in attempting to solve our problem. Attached is
a summary of the responses we received for our query.

Original Query
> I am interested to know what is the preferred method to remove heart
> beat signals from EMG recordings of trunk muscles. with minimum loss
or
> distortion of the EMG signal itself. A summary of the responses I
will
> submit to the list members.
(Original question posted by Dr. Ruth Dickstein)

Summary of responses on removal of ECG artifacts from EMG signals
Highpass Filtering: The most common reply was to use some form
of highpass filtering, with suggested cutoffs ranging between 1 and 40
Hz. Many suggested performing an FFT and zeroing the lower frequencies,
then performing an inverse FFT, in order to obtain a sharper cutoff than
by using ordinary analog filters. An article describing the trade-off in
choice of cutoff frequency is: Redfern et al, "High-pass filtering
to remove electrocardiographic interference from torso EMG recordings,"
Clin. Biomech. 1993; 8; 44-48.

Gating: A simple and usually satisfactory method is known as
gating. The idea is that values above a trigger value are assumed to be
heartbeat signals, and are treated as missing values. This has the advantage
of fast and automatic data processing. The gating should be performed before
integration or any other method of smoothing to avoid data loss.

Subtraction: Another common reply suggested subtracting the data
of a 'clean' heartbeat from all occurances of heartbeats in the data. Better
results can apparently be achieved by selecting as a 'clean' heartbeat
an average 'template' of many heartbeats recorded with no muscle activity.

Bartolo et al (see "Analysis of diaphragm EMG signals: comparison
of gating vs. subtraction for removal of ECG contamination", J. Appl. Physiol.
80(6); 1898-1902) compared the gating and subtraction methods using several
common numerical processing techniques, such as mean power, zero crossing
frequency and so on. They found no significant difference in results between
signals processed using these two methods, and concluded that the added
computational power required to use the more complex method of subtraction
does not yield significantly better results.

A combination of the subtraction and filtering techniques was also suggested.
In this method the subtraction would be performed in the frequency domain
(on an FFT signal). In other words, the FFT of a clean heartbeat would
be subtrated from the FFT of the data, and the result converted back using
an inverse FFT. As far as we know this has not been attempted yet with
EMG signals.

Wavelets: Another suggestion was to use wavelet smoothing to
remove heartbeat signal frequencies without distorting EMG data at the
same frequency. For a technical introduction to wavelet theory, see Chui
(1994): "Wavelets: Theory, Algorithms and Applications", San Diego:
Academic Press.

Adaptive Noise Cancellation: Using adaptive noise cancellation
with an extra channel in which only the ECG is collected. For details on
this method see Medical Engineering and Physics, 1997, Vol 19 No
3 pp 279-285, "A comparison of ECG cancellation techniques applied to the
surface recording of somatosensory evoked potentials", J. S. Black and
D. F. Lovely.

Use of histograms: Finding a histogram of the EMG amplitudes,
and removing (or keeping as missing values) all amplitudes which clearly
do not belong to the EMG.

Improved electrode placement: Positioning the ground electrode
in such a manner that the heartbeat reaches the ground and active electrodes
simultaneously and does not get recorded.
 

Thanks again to all who responded to our question.

Sincerely,
 
Zvika Ben-Haim
zvikabh@aluf.technion.ac.il