The most well-known area of biomedical signal processing is ECG analysis. An ECG record consists of heart beats and different noises. The heart beat waves, namely P, Q, R, S and T waves can show very different shapes depending on patients, ecg leads, etc. In addition, the ECG signal is disturbed with low and high frequency noises, making QRS recognising very difficult.
The most important function in ECG record evaluation is a good QRS detector. The reliability of every other algorithm depends on it. This algorithm was developed to improve QRS detection in holter and stress test systems. It is based on a unique – signal/noise ratio – solution with high noise rejection rate and precise QRS recognising. Both time and frequency domain parameters were used to help the decision. The algorithm uses wavelet transformation for frequency decomposition in order to determine the spectral characteristics of the different QRS forms and artefacts. Some of the frequency components are used to filter out baseline wandering and other noises.
As the QRS detector is the most important part of ECG signal processing, it is necessary to achive very good sensitivity and positive predictivity values.
Result on AHA database:
Sens = 99.6%
PPred = 99.9%
In an ECG record, besides normal QRS beats often occur so-called ventricular beats, which can cause serial problems, that sometimes are life threatening, so it is very important to discover them. Many problems make it hard to provide a reliable ventricular beat detector, among others baseline artefacts, burst noises, presence of branch blocks, very different shapes, etc.
Ventricular beat detection is very important part of arrhytmia analysis of ECG signals. This algorithm was developed to make VE beat detection more precise, using time and frequency domain parameters for clustering QRS beats. Multidominant QRS templates help the decision and reduce false detection. It was tested on different databases, among others on AHA (133 records: 10xx – 70xx) achiving good result:
Sens = 94.2%
PPred = 98.8%
Atrial fibrillation is an irregular and often rapid heart rate that commonly causes poor blood flow to the body. Atrial fibrillation is kept to be the major stroke risk factor.
The main task of an atrial fibrillation detection algorithm is to recognise such parts of the ecg signal which have irregular RR intervals with no presence of P wave. Algorithms which have no P wave detection often give false atrial fibrillation indication.
Detection of P wave may be the hardest problem in ECG signal analysis, because this wave has generally very low amplitude, often disturbed by noises and different heart disorders. Present atrial fibrillation detection method uses statistical analysis of RR intervals, wavelet and fast Fourier transformation for spectral analysis.
This algorithm was developed to obtain very reliable AF detection in ECG holter system using RRI/AA, PWA/FSA methods. It was tested on different databases, among them on MitBih AF database achieving good result:
Sens = 93.47%
Spec = 99,44%
Spec = 99,44%