Application of pattern recognition and image classification techniques to determine continuous cardiac output from the arterial pressure waveform.

The shape of the arterial pressure waveform is a non-linear function of stroke volume, heart rate and many other cardiovascular parameters. Previous attempts have been made to exploit this relationship and derive cardiac output (CO) from the arterial pressure waveform. These classical "pulse-contour" methods utilized simplifying linear assumptions, as a result they failed to adequately estimate CO over a sufficiently wide range of hemodynamic conditions. We have applied pattern recognition and image processing techniques to the problem of deriving CO from the arterial pressure waveform, thereby eliminating the need for simplifying assumptions. Computer simulations were used to develop the basic pattern recognition algorithms and compare their performance with that of published classical "pulse-contour" methods. Animal models were subsequently used to demonstrate proof of the concept. For over 200,000 individual heart beats, covering a wide range of hemodynamic conditions, the mean error, in calculated CO compared to ultrasonic flow probe determined CO, was 2.8% with a standard deviation of 9.8%.