BriefThe purpose of this project is to develop an algorithm that will allowaccurate detection/characterisation of ECG features. The aim is to develop an algorithmthat will identify QRS complexes by providing the sample numbers of the r-wavepeaks for an ECG being evaluated. The algorithm will also be required to labelQRS complexes as normal or abnormal. Performance should be reported using anumber of metrics. The metrics that are most widely used in such reporting are sensitivity and positive predictive value.
All algorithms must be developed so thatall results are completely reproducible by the marker. ECG BackgroundThe electrocardiogram (ECG) reflects of the heart’s electrical activityas measured at the body surface. Changes in the ECG are studied to detect cardiacabnormality. Computerised interpretation of the ECG is challenging and has beenresearched for many decades. One of the main challenges in this area is thehuge variability that can be seen within subjects and between subjects and betweencardiac abnormalities.ECG Features (EE)Whenthe hearts electrical activity passes through the heart and reaches the surfaceof your skin this is then read and displayed as an electrocardiogram.
The ECGis recorded by attaching electrodes to the skin in the relevant positions. Thevoltage that is measured between the electrodes varies and this is how the waveis plotted. The wave is split up into three parts.Firstly the P Wave, this is linked with the atrial depolarisation of theheart, typically less than 120ms. The main part of the ECG is the QRS complex,consisting of 3 deflections.
First deflection being the Q wave, secondly R waveand finally this S wave. These waves are linked with the ventriculardepolarisation. Lastly the T wave is a is the slow downslope of the ECG,representing ventricular repolarisation. The average resting heart beats 60-100 perminute, roughly 100,000 times each day. It is common for your heart to go outof rhythm and can sometimes be felt as a flutter in the chest, this irregularor abnormal heartbeat is called an arrhythmia.
An arrhythmia can produce anuneven heartbeat, increase or decrease the speed of a beat thus altering thenature of an ECG trace. However, anomalies within an ECG can often be theresult of an underlying heart condition or disease. The length, amplitude and morphology of aQRS complex is useful in diagnosing cardiac arrhythmias, myocardial infarction,ventricular hypertrophy and other various heart conditions. A prolonged QRSduration length indicates conditions such as bundle branch block orhyperkalemia and an increased QRS amplitude can indicate cardiac hypertrophy.Practitioners can derive heart conditions and syndromes through the individualanalysis of Q, R and S parts of the wave. An abnormality within the Q wavetypically indicates the presence of an infarction and a weak R wave progressionis commonly accredited to conditions such as anterior myocardial infarction andWolff-Parkinson-White syndrome.
However, this can also be the result of a poorECG recording procedure. QRSDetection Algorithm Research (RB)Pan& Tompkins (RB)Having known the desirable passband tomaximize the QRS spectral energy is approximately 5-15 Hz, in order toattenuate out noise Pan and Tompkins passed the signal through a digitalbandpass filter composed of high-pass and lowpass filters in order to reducethe influence of muscle noise, 60 Hz interference, baseline wander, and T-waveinterference. To implement this, they cascaded the low-pass and high-passfilters to achieve a 3 dB passband from about 5-12 Hz. The next step in theiralgorithm was to differentiate the signal to gather Information about theslope.
They then squared the signal. This makes all data points positive andintensifies the slope of the frequency response curve of the derivative whilstrestricting false positives caused by T waves with higher than usual spectralenergies. The next stage was to implement moving window integration. The movingwindow integrator produced a signal that included information about the slopeand the width of the QRS complex in addition to the slope of the R wave. Panand Tompkins then divided their algorithm into three stages, learning phase 1,learning phase 2, and Peak detection.
They determined that Learning phase 1 detection thresholds based uponsignal and noise peaks and that Learning phase 2 determines RR-interval averageand RR-interval limit values. The subsequent peak detection phase does therecognition process and produces a pulse for each QRS complex. 1