Accessing IBI intervals about heartrate_analysis_python

[paulvangentcom]

Hi Francois,

Thanks for the kind words. Can you share the data behind the two plots you place with me? I am working on an extension to the enhance_peaks() function that will normalise locally rather than globally. This would give me a chance to test it.

[paulvangentcom]

Additionally, take a look at the heartrate arduino repo here, especially the AVR simple logger with scaling: https://github.com/paulvangentcom/heartrate_analysis_Arduino/blob/master/implementations/AVR/SimpleLogger_Scaling_AVR/SimpleLogger_Scaling_AVR_USB/SimpleLogger_Scaling_AVR_USB.ino

Docs here: https://embedded-heart-rate-analysis-toolkit.readthedocs.io/en/latest/implementations.html#simple-logger

This is an implementation for a PPG sensor like the Pulse Sensor you use, plus Arduino. It auto-scales the data as it is recording. This might help you achieve what you need without using the enhance_peaks() in heartpy.

[paulvangentcom]

Regarding the RR-intervals. I assume your sample rate is 50Hz. The module finds the nearest highest point as a peak location. This means the resolution is tied to the sampling rate for now. I'm working on a solver to interpolate each peak and find the maximum that way, which could be more accurate but I don't have an ETA for that.

For now I recommend upping your sample rate to 1000Hz for 1ms R-R accuracy. The linked Arduino sketch can reach those sampling speeds easily.

-Paul

[FrancoisCosta]

Hi Paul, thanks for the reply!

I've attached the csv file with the Pulse Sensor Data I've been using. It was sampled at 50Hz over a period of approximately 15mins.
Let me know if you would like longer datasets, I should have several hours in the next couple of weeks (I aim to collect data to analyse sleep).

PulseSensorData_vanGent.txt

The first graph is believe represents the [:15000] samples whereas the second graph represents [30000:]

Regarding the RR intervals - ah I see! I'll give that Arduino sketch a go tomorrow morning and let you know how it goes (I don't have access to the sensor today unfortunately).

Please let me know if/when the local enhance_peaks() function and the IBI interpolated solver get implemented :) would be of great help for my project.

[paulvangentcom]

I'll push the peak position interpolation step up my list then and hope to get to it this week. I'll keep you updated.

-Paul

[FrancoisCosta]

Perfect! Thanks :)

[FrancoisCosta]

Hi Paul, any update on this?

Thanks,
FB

[paulvangentcom]

Hi Francois,

Not yet I'm afraid. I need to find the time to fit heartpy work in between my phd work so it doesn't always work out...

I hope this weekend I can have a look.

-Paul

[paulvangentcom]

update, I've finished the interpolation method, which upsamples and re-estimates peak position. On the first peak in your signal it looks like this:

Or, for the first 10 seconds of your signal you can go from 20ms accuracy (50Hz):
array([ 280., 1160., 2080., 2980., 3860., 4740., 5620., 6480., 7300., 8120., 8920., 9700.])

to peaks estimated to 1ms accuracy (1000Hz)
array([ 276., 1166., 2079., 2984., 3865., 4737., 5622., 6477., 7295., 8111., 8914., 9692.])
(or of course higher if you want, but I don't see why)

You can specify the precision you want and the method does the rest. I need to test the implementation and update the docs. I'll probably push the update to github today or tomorrow. Just wanted to keep you informed.

[FrancoisCosta]

Excellent work! Thanks Paul!
I look forward to giving it a go :)
Do you foresee any negative impact on the accuracy of the peak location when upsampling and re-estimating?

[paulvangentcom]

Please pull the latest version, I've implemented the above described interpolation to the process() and process_segmentwise() functions. Usage is simple:

import heartpy as hp

data = hp.get_data('data.csv')

#the 'high_precision' flag sets the mode as active
#the 'high_precision_fs' variable sets the target sample rate to upsample to
#default is 1000.0 Hz, meaning 1-ms peak accuracy
wd, m = hp.process(data, sample_rate = 100.0, high_precision = True, high_precision_fs = 1000.0)

print(wd['RR_list'])

Gives

[1019. 982. 968. 998. 1051. 1092. 982. 909. 892. 952. 1085. 1157. 1130. 1023. 1041. 1067. 1046. 944. 963. 1035. 1094. 1012. 980.]

Good luck!

• Paul