Closed arashtavakoli closed 3 years ago
Hi Arash,
It seems there is substantial noise still in the signal. Can you share a CSV file with me?
Paul
Hi Paul,
Thank you so much for your reply. Here is a csv uncleaned file. This file is resampled at 100hz. I am also happy to meet you online if you have any suggestions. Thank you so much for your help.
https://drive.google.com/file/d/1imQk9Rc4MyoAFeODQaCTBHWyWaJ3Nz8E/view?usp=sharing
Best, Arash
Hi Arash,
I had a quick look, find a quick notebook with some analysis here: Arash.zip
The signal is really quite noisy, and segments will not give a good solution.
Paul
Hi Paul,
I took my time to really understand what you did in the notebook. Thank you soo much. I am just wondering how do you find the values for the bandpass filter? basically, this number can change depending on how much movement there is I guess. Also, do you have any general suggestions in this regard? I understand that you use the Poincare plot but I was not sure if there is anyway to actually automate this process? Also as you mentioned it is way too noisy. Do you think such signal is of any use or I should just completely throw it away?
Sorry for too many questions. I would be very happy to also talk to you on phone.
Cheers, Arash
On Tue, Jun 29, 2021 at 3:56 AM Paul van Gent @.***> wrote:
Hi Arash,
I had a quick look, find a quick notebook with some analysis here: Arash.zip https://github.com/paulvangentcom/heartrate_analysis_python/files/6731968/Arash.zip
The signal is really quite noisy, and segments will not give a good solution.
Paul
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/paulvangentcom/heartrate_analysis_python/issues/70#issuecomment-870410635, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJKRJTEDEEQUTSVEJY4RXUTTVGDCDANCNFSM44D7KPOQ .
-- Arash Tavakoli, E.I.T. Graduate Research Assistant Department of Civil and Environmental Engineering University of Virginia
Hi Arash,
The bandpass values I pick based on expectation of the bpm range. 0.7 - 2.0 Hz in this case would translate to 42 - 180 bpm. You can expand the range if you expect higher values. Lower I wouldn't quickly expect.
The noisiness is in the original signal, I did include the poincare plot but that was for illustrative purposes mainly : ). If possible I would definitely consider getting another way of collecting the signal. pulsesensor.com has quite good ones you can put on fingertips and earlobes. Quite a lot of peaks are discarded by heartpy, most likely because the underlying signal is too influenced by the present noise, which shifts the peaks to unexpected and unlikely positions.
You can use this type of signal quite well for BPM only (it's resistant to noise) but not really for the HRV metrics (very sensitive to noise)
Cheers Paul
Hi,
Thank you so much for this awesome package. I have been using it and unfortunately, it often misses lots of the peaks in my own PPG data. I have used the code below on my data and I have also attached an example of how it labels almost all peaks as rejected. Any suggestion would be much appreciated. My data is already at around 90-100 hz so I scaled it, and then resampled it at 90hz.
df_final["PPG1_clean"] = hp.filter_signal(df_final["scaled_ppg"], [0.7, 2.5], sample_rate=90, order=3, filtertype='bandpass') for s in [[0, 10000], [20000, 30000], [30000, 40000], [40000, 50000]]: wd, m = hp.process(df_final["PPG1_clean"].values[s[0]:s[1]], sample_rate = 90, high_precision=True, clean_rr=True) hp.plotter(wd, m, title = 'zoomed in section', figsize=(12,6)) peaks = nk.ppg_findpeaks(df_final["PPG1_clean"].values[s[0]:s[1]],show=True) hp.plot_poincare(wd, m) plt.show()
hrv_indices = nk.hrv(peaks["PPG_Peaks"], sampling_rate=240, show=False)