Complete Pan Tompkins Implementation ECG QRS detector

Detects QRS complex in an ECG signal based on Pan Tompkins algorithm

Complete Implementation of Pan Tompkins;
If you found this script useful please cite the following references;
%% References :
%[1] Sedghamiz. H, "Matlab Implementation of Pan Tompkins ECG QRS detector.", March 2014. https://www.researchgate.net/publication/313673153_Matlab_Implementation_of_Pan_Tompkins_ECG_QRS_detect
AND
%[2] PAN.J, TOMPKINS. W.J,"A Real-Time QRS Detection Algorithm" IEEE
%TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. BME-32, NO. 3, MARCH 1985.
%% Author : Hooman Sedghamiz
% Linkoping university
% email : [e-mail]
% Copyright march 2014
-----------------
%% Method :
%% PreProcessing
% 1) bandpass filter(5-15 Hz)
% 2) derivating filter to high light the QRS complex.
% 3) Signal is squared.
% 4)Signal is averaged of noise (0.150 seconds length).
% 5) depending on the sampling frequency of your signal the filtering
% options are changed to best match the characteristics of your ecg signal
%% Decision Rule
% At this point in the algorithm, the preceding stages have produced a roughly pulse-shaped
% waveform at the output of the MWI . The determination as to whether this pulse
% corresponds to a QRS complex (as opposed to a high-sloped T-wave or a noise artefact) is
% performed with an adaptive thresholding operation and other decision
% rules outlined below;
% a) FIDUCIAL MARK - The waveform is first processed to produce a set of weighted unit
% samples at the location of the MWI maxima. This is done in order to localize the QRS
% complex to a single instant of time. The w[k] weighting is the maxima value.
% b) THRESHOLDING - When analyzing the amplitude of the MWI output, the algorithm uses
% two threshold values (THR_SIG and THR_NOISE, appropriately initialized during a brief
% 2 second training phase) that continuously adapt to changing ECG signal quality. The
% first pass through y[n] uses these thresholds to classify the each non-zero sample
% (CURRENTPEAK) as either signal or noise:
% If CURRENTPEAK > THR_SIG, that location is identified as a “QRS complex
% candidate” and the signal level (SIG_LEV) is updated:
% SIG _ LEV = 0.125 ×CURRENTPEAK + 0.875× SIG _ LEV
% If THR_NOISE < CURRENTPEAK < THR_SIG, then that location is identified as a
% “noise peak” and the noise level (NOISE_LEV) is updated:
% NOISE _ LEV = 0.125×CURRENTPEAK + 0.875× NOISE _ LEV
% Based on new estimates of the signal and noise levels (SIG_LEV and NOISE_LEV,
% respectively) at that point in the ECG, the thresholds are adjusted as follows:
% THR _ SIG = NOISE _ LEV + 0.25 × (SIG _ LEV ? NOISE _ LEV )
% THR _ NOISE = 0.5× (THR _ SIG)
% These adjustments lower the threshold gradually in signal segments that are deemed to
% be of poorer quality.
% c) SEARCHBACK FOR MISSED QRS COMPLEXES - In the thresholding step above, if
% CURRENTPEAK < THR_SIG, the peak is deemed not to have resulted from a QRS
% complex. If however, an unreasonably long period has expired without an abovethreshold
% peak, the algorithm will assume a QRS has been missed and perform a
% searchback. This limits the number of false negatives. The minimum time used to trigger
% a searchback is 1.66 times the current R peak to R peak time period (called the RR
% interval). This value has a physiological origin - the time value between adjacent
% heartbeats cannot change more quickly than this. The missed QRS complex is assumed
% to occur at the location of the highest peak in the interval that lies between THR_SIG and
% THR_NOISE. In this algorithm, two average RR intervals are stored,the first RR interval is
% calculated as an average of the last eight QRS locations in order to adapt to changing heart
% rate and the second RR interval mean is the mean
% of the most regular RR intervals . The threshold is lowered if the heart rate is not regular
% to improve detection.
% d) ELIMINATION OF MULTIPLE DETECTIONS WITHIN REFRACTORY PERIOD - It is
% impossible for a legitimate QRS complex to occur if it lies within 200ms after a previously
% detected one. This constraint is a physiological one – due to the refractory period during
% which ventricular depolarization cannot occur despite a stimulus[1]. As QRS complex
% candidates are generated, the algorithm eliminates such physically impossible events,
% thereby reducing false positives.
% e) T WAVE DISCRIMINATION - Finally, if a QRS candidate occurs after the 200ms
% refractory period but within 360ms of the previous QRS, the algorithm determines
% whether this is a genuine QRS complex of the next heartbeat or an abnormally prominent
% T wave. This decision is based on the mean slope of the waveform at that position. A slope of
% less than one half that of the previous QRS complex is consistent with the slower
% changing behaviour of a T wave – otherwise, it becomes a QRS detection.
% Extra concept : beside the points mentioned in the paper, this code also
% checks if the occured peak which is less than 360 msec latency has also a
% latency less than 0,5*mean_RR if yes this is counted as noise