Hi,
What you are trying to do should be easy. We do it all the time as an
"add-on" benefit for my company's other medical applications - all
using labview. I don't want to send the actual code since it is proprietary,
but I think I can help with advice. You might have considered some of
these things already, but I will list the primary considerations.
Hopefully that will help you solve your problem. Once you have done the peak
finding correctly the rest is easy, but I will review that at the end at well.
To do an excellent heart rate monitor you only need 3 vis - "collector"
"filter" and "peak finder".
Peak finding: with the ECG there are two problems with setting the
peak finding parameters - wandering baseline and different shapes of the
individual waveforms for different individuals (and to a much lesser extent,
the same individual).
Wandering baseline: the ECG tends to have a wandering
baseline, even ECG with a good analog high pass. The "high end" recommended
high pass cut off for medical diagnostic purposes is about 0.05 Hz, and
typically this will still allow for enough baseline wander to mess with your
peak finding occasionally. So digitally filter your signal beforehand
with a high pass filter with a much high frequency - like 1 Hz (use the
pre-fabed "filter.vi" in the signal analysis pallet, or any of the
many lower level filters vis).
Different Shapes: The other issue is
that different people will have different shaped PQRST complexes. Since you
want to set the peak threshold below the R-peak but above the other features
that might trigger the peak finder - mainly the P as T waves. The thing to
remember is that if you are going for a heart rate monitor ONLY you are allowed
to filter the hell out of the signal since you don't care about the exact shape
of the PQRST. In fact you really are
looking for a filter that removes as much of the lower frequency P and T waves
leaving mostly R peaks to do the peak finding. You might try razing the high
pass cut off until you are satisfied that both the drift and the shapes are
best for peak finding.Yry it. It goes without saying that depending on your high frequency
noise you may ALSO want to low pass filter your signal before peak finding, though this may not be as crucial.. If you have prominent
60 Hz interference you may want to low pass as a frequency below 60 Hz.
Also remember to set the number of peak samples in the peak finder to be
enough to prevent multiple peak detections - something like 1/4 or thereabouts
of the width of the R-wave (in samples).
After filtering the signal you then need to set the threshold. Since everybody has different R-wave peak magnitudes,
you can't unusually just fix the value. The way to do this is to fix some
percentage of the MAXIMUM value in a give window of say 15 second or more. If you are doing you filtering correctly then
this will give you the approximate (max) value of the (filtered) R-wave. Once
you have your max value you then set the threshold at a certain fraction (you
decide) of this value. Try 0.6 for
example. Then you will always "cut" only
the R-waves, and only once each if you set your peak samples like I said. There
are alway going to be "mutant" ECGs that are "pathelogical"
- such as a lead I that has a prominent NEGATIVE R-wave "peaks". You
can decide how to deal with these case on your own. The resulting output of the
peak finder will then work for just about any person, For about %95 of ECGs what
I described will work fine, and I believe once you do it once correctly, dealing
with the rare cases will be easy for you to pick up without handholding
Heart Rate calculation (easy part once you have the peak locations). The size of the location vector (output of
the peak finder) is the number of peaks. I think there is also an output that
has just this number too. The net sample separation (call it delta S) is the sample
value of the last peak minus the sample value of the first peak. The corresponding
time separation is delta S x delta t (where delta t is the sampling time) or
delta S/(sampling rate). The heart rate is the number of beats per minute. Use the time calculated above, and the number
of detected beats minus 1 as the number of beats in that time interval and make
sure you convert to beats per minute dimensions. All of the above should be done on a rolling
window of many seconds since you usually want an average heart rate over
several second - you can use the "collect" vi to do this. If you don't
care about beat to beat fluctuations you can just calculate the heart rate from
each peak separation time (be sure you know your sampling rate, of course). The way it would be done on a hospital monitor
is the first way - using a longish moving window, say 15-30 seconds, It should be no problem for a modern PC to
keep up. Alternatively you could do the calculation on consecutive 15-30 second
intervals instead of a rolling window.
So to do this right you only need the collect vi, the filter vi and the peak
finder vi. I hope this helped in some way.
Marko
CardioPrint Biometrics
