Conclusions

A new dynamical model has been introduced which is capable of replicating many of the important features of the human ECG. Moreover, many of the morphological changes observed in the human ECG manifest as a consequence of the geometrical structure of the model. Model parameters may be chosen to generate different morphologies for the PQRST-complex. The power spectrum of the RR-intervals can be selected a priori and used to drive the ECG generator. This allows the operator to prescribe specific characteristics of the heart rate dynamics such as the mean and standard deviation of the heart rate and spectral properties such as the LF/HF ratio. In addition the average morphology can be controlled by specifying the positions of the P,Q,R,S and T events and the magnitude of their effect on the ECG. Having access to a realistic ECG provides a benchmark for testing numerous biomedical signal processing techniques. In order to establish the operational properties of these techniques in a clinical setting, it is important to know how they perform for different noise levels and sampling frequencies. A number of applications and simple extensions of the model are possible: (i) By fitting (see [17]) the model to the morphology of a particular subject's ECG and the power spectrum of their RR-intervals, a database of realistic ECGs could be created. This database could be employed for statistical hypothesis testing. Furthermore, it may be possible derive a corrected QT-interval which is independent of the heart rate. (ii) The synthetic ECG could be used to assess the effectiveness of different techniques for noise and artefact removal. These could be evaluated by adding noise and/or artefact onto the synthetic signal and then comparing the original with the processed signal. (iii) Abnormal morphological changes with time could be introduced by using a parameter to control the position of any of the P,Q,R,S or T events. This extension would be particularly useful for testing techniques which aim to detect ST depression or elevation by decreasing or increasing the $z$-position of the T wave over time. Similarly QT prolongation could be replicated by moving the T point away from the Q point in the $(x,y)$ plane (increasing $\theta_T - \theta_Q$). (iv) The model could be used to produce multi-lead ECG signals by introducing a measurement function which maps from the $(x,y,z)$ model space to the ECG signal: $s = h(x,y,z)$. Different lead configurations and modulations due to respiration and movement of the cardiac axis could be modelled using time-dependent functions for $h$. (v) Abnormal beats, such as ectopics, can be simulated by modifying the position of the R-peak for one cycle of the dynamics. The new model presented here reflects a data-driven approach to modelling the electrical activity of the heart. Key physiological features have been incorporated using motion of a trajectory throughout a three-dimensional state space. The quasi-periodicity of the cardiac cycle is represented by attraction towards a limit cycle. The model produces QT-intervals and R-peak height variation (RSA) which vary linearly with the RR-intervals as has been found in real ECGs [6,4]. It is hoped that this model will provide a valuable tool for testing biomedical signal processing algorithms applied to ECG signals with different sampling frequencies and levels of noise and/or movement artefact.