Design of an optimized continuous mini-bolus thermodilution cardiac output monitor using artificial neural networks and genetic algorithms

The ability to estimate cardiac output by thermodilution, as initially described by Fegler (1954), was an important step in hemodynamic monitoring. However, the usefulness of this procedure has been hampered by the difficulty in filtering the thermal noise from the thermodilution signal in the pulmonary artery. As a result, current procedures are limited to intermittent measurements with large-bolus injections that produce an acceptable signal-to-noise ratio (SNR). This paper presents one approach to solving this problem using the nonlinear mapping ability of artificial neural networks (ANN). It is shown that the cardiac output estimated by the ANN significantly improves the classical method of computing cardiac output with small-injectates using the Stewart-Hamilton equation and are within clinically acceptable limits in comparison to the “gold standard”