Accurate automatic detection of end-diastole from left ventricular pressure using peak curvature

Precise identification of end-diastole (ED), corresponding to the end of diastole and start of systole, is crucial for accurate assessment of cardiac function. The aims of this study were to develop a new algorithm based on peak curvature (kappa(p)) for detecting ED as a "corner" in left ventricular pressure (LVP) signals, and to compare this approach with "gold-standard" ED obtained by manual annotation (ED(man)) and ED calculated with previously described algorithms that use an LVP first-derivative threshold (dP/dt(0) or dP/dt(100)), the peak LVP second-derivative (d(2)P/dt(2)(p)) or ECG R-wave peak (ECG(R)). Using customized software, all algorithms were applied to data derived from 213 large animal studies spanning a wide range of animal ages (fetus to adult), heart rates, inotropic states, and loading conditions. Differences between ED(man) and each algorithm were then compared after defining an acceptance region for the ED detection based on ED(man) interobserver variability. ED detected with kappa(p) was the most accurate (p < 0.001) and least variable (p < 0.001), with 97% of measurements within the acceptance region and difference from ED(man) of (1.5 +/- 4.2) ms. By contrast, ED was often detected early with dP/dt(0) and dP/dt(100) , and late with d(2)P/dt(2)(p) and ECG(R). These results indicate that the peak curvature algorithm using LVP provides accurate and reliable detection of ED.