Beat-to-beat cardiac output inference using heart sounds

Cardiac output (CO) change is the primary compensatory mechanism that responds to oxygenation demand. Its continuous monitoring has great potential for the diagnosis and management of cardiovascular diseases, both in hospital as well as in ambulatory settings. However, CO measurements are currently limited to hospital settings only. In this paper, we present an extension of the model proposed by Finkelstein for beat-to-beat CO assessment. We use a nonlinear model consisting of a two-layer feed-forward artificial neural network. In addition to demographic (body surface area and age) and physiological parameters (HR), surrogates of contractility, afterload and mean arterial pressure based on systolic time intervals (STIs), estimated from echocardiography and heart sounds are used as inputs to our models. The results showed that the proposed models - with echocardiography as reference - produce better estimations of stroke volume/CO than the Finkelstein model (12.83±10.66 ml vs 7.23±6.6 ml), as well as higher correlation (0.46 vs 0.82).