Pan-Tompkins QRS Detector — Python

A Python port of the Pan-Tompkins real-time QRS detection algorithm for ECG signals.

Original MATLAB implementation by Hooman Sedghamiz (Feb 2018), MSc Biomedical Engineering, Linköping University.
Python port by Dr. Hatem Zehir.
Python conversion retains the original BSD 3-Clause license — see LICENSE.

Algorithm Overview

The Pan-Tompkins algorithm is a classic, widely used method for detecting QRS complexes (R-peaks) in ECG signals in real time. It processes the ECG through a signal-processing pipeline:

1. Bandpass filter (5–15 Hz) — removes baseline wander and high-frequency noise
2. Derivative filter — emphasises the steep slopes of QRS complexes
3. Squaring — makes all values positive and amplifies large slopes
4. Moving-window integration (~150 ms window) — produces a smooth envelope
5. Adaptive thresholding & decision logic — distinguishes true QRS complexes from T-waves and noise using two running thresholds updated after each beat

References

• Pan, J. & Tompkins, W. J., "A Real-Time QRS Detection Algorithm", IEEE Trans. Biomed. Eng., BME-32(3), March 1985.
• Sedghamiz, H., "Matlab Implementation of Pan Tompkins ECG QRS detector", 2014. ResearchGate

Installation

pip install -r requirements.txt

Dependencies:

Usage

import numpy as np
from pan_tompkins import pan_tompkins

# Load your ECG signal and sampling frequency
# ecg : 1-D numpy array of the raw ECG
# fs  : sampling frequency in Hz (e.g. 200, 360, 500)

qrs_amp, qrs_idx, delay = pan_tompkins(ecg, fs, gr=True)

print(f"Detected {len(qrs_idx)} QRS complexes")
print(f"R-peak sample indices: {qrs_idx}")

Parameters

Returns

Quick Demo

Running the script directly generates a synthetic ECG (~70 bpm, 10 s at 200 Hz) and detects QRS complexes:

python pan_tompkins.py

Expected output:

Detected 12 QRS complexes
R-peak indices: [ 100  270  440  610  780  950 1119 1289 1459 1629 1799 1969]
Delay: 15.0 samples

Notes

• For fs = 200 Hz, separate low-pass (12 Hz) and high-pass (5 Hz) Butterworth filters are applied sequentially.
• For all other sampling rates, a single bandpass Butterworth filter (5–15 Hz) is used.
• The derivative filter kernel is interpolated to match the sampling rate when fs ≠ 200.
• filtfilt (zero-phase filtering) is used throughout, so delay reports the moving-average window contribution only.

License

BSD 3-Clause License. Copyright (c) 2018, Hooman Sedghamiz. See LICENSE for full details.

Copyright (c) 2018 Hooman Sedghamiz
Copyright (c) 2026 Dr. Hatem Zehir

Citation

If you use this implementation in your research, please cite:

@article{PanTompkins1985,
  author = {Pan, J. and Tompkins, W. J.},
  title = {A Real-Time QRS Detection Algorithm},
  journal = {IEEE Transactions on Biomedical Engineering},
  year = {1985},
  volume = {BME-32},
  number = {3},
  pages = {230--236},
  doi = {10.1109/TBME.1985.325532}
}

@software{ZehirPythonPort2026,
  author = {Zehir, Hatem},
  title = {Pan-Tompkins QRS Detector — Python},
  year = {2026},
  url = {https://github.com/Hatem-Zehir/pan-tompkins-qrs-detector}
}

🤝 Contributing

Contributions, bug reports, and feature requests are welcome.

Please open an issue or submit a pull request.