Session P7C.6

ECG Synthesis Using a Gaussian Combination Model (GCM)

S Parvaneh*, M Pashena

Islamic Azad University
Tehran, Iran

In this paper a method for electrocardiogram (ECG) synthesis is presented. The synthesizer utilizes morphological properties of ECG around extermum points.
For implementing our algorithm, we first selected a single period of ECG and assembled a database for normal ECG and some arrhythmic ECG. In the next step we determined the number of Gaussians in our morphologic model. This quantity could be determined manually or based on number of local extermums. The second method of Gaussian number determination yields superior results compared to the first method.
For ECG synthesis, we fit a Gaussian around each extermum and minimized local error that is defined as local difference of real ECG and our model. Range of Gaussian fitting has been determined using two methods:
1- Zero crossing method: two zero crosses around each extermum with local maxima or minima defined as Gaussian fitting range.
2- Minimum bank method: we first gathered a bank of extermum extracted from signal and determine proper range for fitting as a map that includes each extermum.
The fitting should be done for each Gaussian in GCM method in order to synthesize the signal.
With GCM method we have represented one cycle of ECG with a combination of Gaussians (each Gaussian has a mean and a variance). In order to provide realistic ECG with true timing, we present each synthesized ECG cycle based on heart rate variability data.
Using fitting method and similarity of morphologic shape of ECG with our Gaussian Combination Model provides reasonable results with negligible error (based on the method and nature of signal) and using heart rate variability information enriched the timing properties of our model.
The zero crossing method is faster than the minimum bank method but has less accuracy. Modeling time to produce each cycle of ECG is in order of seconds and the number of Gaussians will affect this time.
In this paper attempts have been made to model the time domain ECG signal directly and based on morphological features. This synthesis method has been used to generate a database of normal and abnormal ECG that can be used for teaching purposes and evaluation of ECG signal processing algorithms. Each Signal in the database is represented with mean and variance of Gaussians that generate that signal with minimum error. As a result, we have a collection of Gaussian parameters (equal to the number of Gaussians that have been used in GCM model).
The above GCM method can be used in ECG compression, ECG evaluation with artificial neural networks, etc.

(Abstract Control Number: 155)