Impedance cardiography is a noninvasive technique for estimation of stroke volume (SV), based on monitoring the variation in the thoracic impedance during the cardiac cycle. The current SV calculation methods use parameters obtained by ensemble averaging of the waveform along with equations based on simplified models of the thoracic impedance and aortic blood flow profile. They often result in inconsistent estimates when compared with the reference techniques. An investigation is carried out for beat-by-beat monitoring of SV using an artificial neural network with a set of input parameters as used in the different SV equations.
A three-layer feed-forward neural network is used and the impedance cardiogram parameters are obtained using an algorithm for beat-by-beat automatic detection of the characteristic points. The training and testing are carried out using the SV values obtained from Doppler echocardiography as a reference technique after alignment of the signals from the two techniques. Results from the data from six subjects with recordings under rest and post-exercise conditions show the neural network based estimation to be more effective than the estimations based on SV equations.