Hi. Thanks for the code.
I was benchmarking the performance of different detection algorithms on MIT-BIH database and I noticed that Pan-Tompkins and Two Moving Average execution was rather slow.

Both algorithms depend on the MWA function to calculate the moving window average. I benchmarked the performance of two different moving average functions with the following results:

Start benchmark!!!
Execution time in seconds using original moving average function:
3.5515489461900143

Execution time in seconds using moving average with convolution:
0.018413547830014067

Execution time in seconds using moving average with cumulative sum:
0.005457540920051542

Best,
Ioannis

Python >= 3.4 ships pathlib with its standard library. Furthermore, the pathlib package on PyPI is not maintained anymore. Instead, they recommend to use the pathlib2 package. Therefore, the entry in setup.py should be modified as follows (https://stackoverflow.com/a/32643122/1112283):

'pathlib2;python_version<"3.4"'

The package import should then also be changed to something along the following lines:

try:
    import pathlib
except ImportError:
    import pathlib2 as pathlib

I have an ecg recording which gives the error:

File "C:\toolkits\Anaconda3\lib\site-packages\ecgdetectors.py", line 310, in engzee_detector MM.append(newM5) UnboundLocalError: local variable 'newM5' referenced before assignment

It seems like the newM5 variable is defined in one clause and used in another:

           if i < 5*self.fs:
                M = 0.6*np.max(low_pass[:i+1])
                MM.append(M)
                if len(MM)>5:
                    MM.pop(0)

            elif QRS and i < QRS[-1]+ms200:

                newM5 = 0.6*np.max(low_pass[QRS[-1]:i])

                if newM5>1.5*MM[-1]:
                    newM5 = 1.1*MM[-1]

            elif QRS and i == QRS[-1]+ms200:
                MM.append(newM5)
                if len(MM)>5:
                    MM.pop(0)    
                M = np.mean(MM)  

The package biosppy is never used in this project. Therefore, it should be removed from ecgdetectors.py and setup.py.

Furthermore, hrv.py contains unused imports of random and subprocess, these should also be removed.

"Stationary Wavelet Transform detector" works really great but I think there is some auto thresholding problem with it. When some disturbances occur in the ECG waveform the remaining normal R waves can not be detected for a while.

If this problem can be solved I think this is the best way to detect R waves.

For example in the attached pictures you can easily see what I mean.

(orange lines: swt_detector, green lines: pan_tompkins_detector, red lines: two_average_detector, blue lines: biosppy's default detector)

In ecgdetectors.py, function hamilton_detector, line 140, the loop counter i is incremented (i += 1).

This has no effect, because it is the last loop statement and i is reassigned by the for loop.

Try this in python to see:

for i in range(10):
    print(i)
    i += 1
    print(i)

I found this R-peak detector in MATLAB that looks interesting, and maybe I am mistaken, but I don't find its equivalent in Python: http://www.librow.com/articles/article-13
Would it be worth adding?

Hi!
First of all, I'm really appreciated with this library. This saves my life!
I'm researching about detecting PVC by using R-signal.
I found my ECG data type is quite different with the one in example folder in this repository.
In example, there are thousands of real values,
In my data, there are integers values, which some of them are consecutive.
To use this library, How should I convert the data to?

Hello,

For almost eight months I have been using py-ecg-detectors library specifically Engzee (Engelse and Zeelenberg) detector for detecting R-peaks in EKG signals. But from last week (4th Nov 2021), I observed that this detector is throwing an error r_peaks as [], an empty vector for different lengths of EKG data recorded.

Other detectors such as Pan-Tompkins, SWT are able to detect R-peaks but at wrong places.

Looking forward for your help in resolving the issue with Engzee detector as this was detecting very accurately the R-peaks.

I am using Windows 10 system to run the code and Pycharm python IDE.

Thanks,
Ravi

Hi Thanks for your code.
I tried to use use usage_example.py but it seems installing it via pip does not download the example data or even create any directory related to the package! Is it fine? or I should download manually the data set.

the error is like this:

C:\ProgramData\Miniconda3\lib\site-packages

NameError Traceback (most recent call last)
in
8 current_dir= ecgdetectors .file
9 print(path)
---> 10 current_dir = pathlib.Path(file).resolve()
11 print(current_dir)
12 example_dir = current_dir.parent/'example_data'/'ECG.tsv'

NameError: name 'file' is not defined

Best Rouhi

Getting the error that I am comparing an int and a list in hrv.py on line 127.
This is due to the fact that we need to make that list an np.array()

I am worried about how to detect mouse ECG, QRS can be detected well, but P, T wave is not a good way.

When I do pip install py-ecg-detectors I get the following error:

ERROR: Could not install packages due to an EnvironmentError: [Errno 13] Permission denied: '/usr/local/templates'
Consider using the `--user` option or check the permissions.

I think this is caused by the data_files list in setup.py. I think you should use package_data instead of data_files to include this folder in the site-packages package structure (https://docs.python.org/3.8/distutils/setupscript.html#installing-package-data).

I was reading the source code and noticed that in the SWT detector, a bandpass filter is used after the doubling, instead of a moving window average, which is what they described in their paper. Is there a particular reason for this?

Hi,

It seems that the f.read() in setup.py for the long_description has a problem with encoding.

File "C:\Users\Koha\AppData\Local\Temp\pip-install-_i_lu5wz\py-ecg-detectors_cab3225f72c3476ea1207fffa3e5d3fe\setup.py", line 6, in
long_description = f.read()
^^^^^^^^
File "C:\Python311\Lib\encodings\cp1252.py", line 23, in decode
return codecs.charmap_decode(input,self.errors,decoding_table)[0]

So I have to walk around by removing the long_description in the setup file

My python version is 3.9

Hi, thank you for your library, in first. I really appreciate your work.
I am a researcher and I am doing some tests on a Pam Tompkins derived algorithm I developed and I am using your Pam Tompkins algorithm to do some tests on false positive and false negative to validate the effectiveness of my algorithm.
Unfortunately I found an error calling the function "panPeakDetect" itself, because I do my own filtering method.
This happened while I was running my testing on the sample 104 of the MIT Arrhythmia DB.

The error code is:
ecgdetectors.py", line 590, in panPeakDetect
missed_peak = missed_section_peaks2[np.argmax(detection[missed_section_peaks2])]
TypeError: list indices must be integers or slices, not list

The problem is that you are passing a list (missed_section_peaks2) to find the signal index of what is in missed_section_peaks2.

My solution is to iterate over that list as follow returning a list containing only the signal to pass to np.argmax:

if len(missed_section_peaks2)>0:     
                                arg_max = []
                                for m in missed_section_peaks2:
                                    arg_max.append(detection[m])
                                missed_peak = missed_section_peaks2[np.argmax(arg_max)]
                                missed_peaks.append(missed_peak)
                                signal_peaks.append(signal_peaks[-1])
                                signal_peaks[-2] = missed_peak 

This way missed_section_peaks2 will contain the max argument in that section.
Not sure if it is correct, but it works now.

Again thank you for your precious work.

Kindly,
Antonio Augello

Here the check should probably be lenght(s_pks)

Good work, and thanks for putting it together in a package!

I run it on a few test cases, however, all of the detectors fail and put the peaks at entirely wrong positions except for Engzee, do you know why this is the case? Do you recommend doing some filtering before or similar?

Hi,

Thanks for making this package! I'm currently using it for various tasks and it saves me a ton of time.

I would like to contribute and add a wqrs detector (W Zong’, GB Moody), is this repo still being maintained ? is it worth me adding a PR ?

Thanks 😄