Gas station without pumps

2022 April 16

ECG: 2-electrode vs. 3-electrode

Filed under: Circuits course — gasstationwithoutpumps @ 12:23
Tags: , , , ,

In Lower PVC frequency, I said “I did not do direct comparisons of the 2-electrode and 3-electrode configurations—I’ll have to try that sometime soon.” So I did that earlier this week, recording resting ECGs first with a 3-electrode configuration (with the bias electrode on my sternum, halfway between the LA and RA electrodes) and then with a 2-electrode configuration (with the bias wire clipped to the RA electrode).  The 60 Hz noise was slightly higher with the 2-electrode configuration, but after filtering and signal averaging the two recordings were almost identical:

resting-2022-Apr-13

The waveforms after signal averaging were remarkably similar. The PVC burden was also similar (20.1% for the 3-electrode recording and 20.5% for the 2-electrode recording).

bpm-resting-2022-Apr-13

The pulse rate from looking at time between spikes worked well for the resting recordings, but the autocorrelation method failed completely, so I did not plot it. The rapid fluctuation in heart rate within a narrow range is real, not an artifact of the algorithm—the heart beats are not perfectly periodic, and the PVCs may be making them even less periodic. The 2-electrode recording probably started a little after 400 seconds—PteroDAQ only time stamps when the file was saved, not what the t=0s time was. I should probably fix PteroDAQ to change that, recording both.

exercise-2022-Apr-13

I tried recording a session on the exercise bike also. The PVCs are mainly during resting at the beginning of the session and at the end of a recovery at the end of the session—the PVC burden was only 1.4%.

bpm-exercise-2022-Apr-13

For the exercise recording, the noise really disrupted the spike-based pulse detection, but did not interfere as much with the autocorrelation-based pulse detection. My peak pulse rate was about 151.5 bpm, by the autocorrelation measure. I’m not sure whether the sudden changes in pulse rate at 100s (when I started pedaling) and around 556s (about 130s into the recovery time) are real or not—the noise in the recording makes it a little difficult to determine the “correct” pulse rate.

The noise during exercise was not 60Hz noise and seemed to vary with whether I was inhaling or exhaling, so I think that it was probably caused by EMG signals from the pectoral muscles or perhaps the diaphragm. The spike detector was clearly missing a lot of the spikes, but making it more sensitive would probably result in false triggering on the EMG noise. I’m wondering whether putting the electrodes on my back, over the scapulae, would reduce the EMG noise, but placing those electrodes and clipping to them would be difficult without an assistant.

The autocorrelation-based pulse detection seems more reliable when exercising, as my pulse is more periodic and has few PVCs, and the autocorrelation method is less susceptible to aperiodic noise.  The spike-based pulse detection seems more reliable when resting, when the pulse is not as periodic and PVCs disrupt the pattern.

I’m also wondering whether a more strenuous exercise session would raise my pulse rate, or whether I’m getting close to my maximum heart rate.  The standard formula for maximum heart rate by age suggests that this may be close to my maximum, but the exercise does not seem all that strenuous, and a couple of years ago I could routinely push to 170 bpm (though perhaps on a device that was an unreliable reporter—it was built into a treadmill at the gym).  So sometime in the next few weeks I’ll try using a higher power output and seeing where my heartbeat tops out.  I’ll probably need to increase the cadence, rather than the resistance, as I’ve been using about 70rpm and 28Nm to get about 205W.  Raising that to 80rpm or even 90rpm is probably easier than increasing the torque.

2022 March 29

Better heartbeat detection

Filed under: Circuits course — gasstationwithoutpumps @ 14:37
Tags: , , , , ,

In Lower PVC frequency, I promised “I’ll report on the algorithms when I get something a little better than I have now.”

I’ve played with three algorithms this week:

  • Doing spike detection, then measuring the time for 2n periods by looking at the time between the spikes n before andafter the current spike.  This gives a fine resolution both in time and frequency, and provides smoothing because adjacent measurements overlap by 2(n-1) periods. It is, however, very susceptible to errors due to miscalling the spikes.  Missed spikes result in too low a frequency, and extraneous spikes result in too high a frequency. I could increase n to reduce the effect of miscounts, but at some loss of time resolution when pulse rate changed.
  • Taking FFT of a block of samples (say 4096, which is about 17 seconds at 240 Hz) and looking for a high energy frequency.  Temporal resolution is poor (even with 50% overlapping blocks we get one measurement every 8.5s) and frequency resolution also poor (bins are about 3.5 bpm wide). I tried improving the frequency resolution by looking at the phase change for the peak between adjacent windows, but that didn’t solve the main problem, which was that choosing the right peak in the spectrum was often difficult.  The simple algorithms I tried for choosing the peak often failed. I eventually gave up on this technique.
  • Taking the autocorrelation of a block of samples (using rfft and irfft) and looking for a peak. The time for that peak is the period, which can be inverted to get the frequency.  This method provides the same coarse time resolution as the FFT method (same size blocks), but has much better frequency resolution, as even the fastest reasonable pulse rate (240bpm or 4Hz) has 60 samples at a sampling rate of 240Hz. I tried accentuating peaks “of the right width” in the autocorrelation by doing some filtering of the autocorrelation, and I tried looking for harmonic errors (where 150bpm might result in a larger peak in the autocorrelation at 75bpm, 50bpm, or 37.5bpm). Even with all the tweaks I could think of, I still had a number of way-off estimates, though median filtering removed most of the anomalies.  Of course, median-of-5 filter makes the time resolution even worse, as median could have come from any of 5 windows (with 50% overlap, that means a time range of 12288 samples or 51.2 seconds!).

I did most of my algorithm testing on one data set (the exercise set from 23 March 2022), and the algorithm is almost certainly overtrained on that data.

bpm-2022-Mar-23

Here are both algorithms applied to the two data sets from 23 March. On the exercise set, the autocorrelation method did an excellent job (except right at the end of the run), but the 12-period measure clearly shows missing and extra peaks. On the resting set, the 12-period measurements were very good, but the autocorrelation ones failed at one point, even with median-of-five filtering. The autocorrelation measurements were also consistently somewhat low.

bpm-resting-2022-Mar-23

To try to figure out why the autocorrelation estimates of the pulse rate were low, I tried superimposing the filtered ECG signal on the plot. The PVCs are visible as large downward spikes. Having one or more PVCs in the window seems to make the autocorrelation estimates somewhat too low. I still have no explanation for why the autocorrelation measure fails so badly around 50 seconds.

Although the autocorrelation measure makes a nice smooth plot on the exercise data set, I sacrificed a lot of temporal resolution to get that. I think that I would do better to make a more robust spike detector to improve the period-based measurements.

%d bloggers like this: