Minimally invasive monitor of cardiac output based on the machine-learning analysis of the pulse contour of the peripheral arterial pressure.
Journal:
Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference
PMID:
40040149
Abstract
In the hemodynamic management of anesthetized patients during surgical operation, minimally invasive and accurate cardiac output (CO) monitoring is strongly required. We have developed a CO monitor based on the machine-learning analysis of the pulse contour of the peripheral arterial pressure (AP) and validated it experimentally in 24 dogs. In the development, we aimed to make the monitor reliable in tracking relative CO normalized by the subject's baseline CO (%CO), and accurate in diagnosing whether %CO is maintained at not less than 100%, that is, CO is maintained at the baseline pre-operation level. This monitor uses the systolic pressure area (SYA) measured in the peripheral AP contour as a surrogate of cardiac stroke volume, and estimates %CO (%COes) by modeling the relation between %CO, product of SYA and heart rate, and mean AP with use of support vector regressor (SVR). In a train/validation group (n=12), SVR was trained and validated via 3-fold cross-validation. %COes estimated by SVR strongly correlated with %CO with a coefficient of determination (R) of 0.84, and identified %CO ≥ 100% accurately with 95% specificity at the cutoff %COes of 112%. When the SVR and the optimal cutoff %COes thus established were prospectively applied to a test group (n=12) to validate their generalizability, %COes estimated by the SVR tightly correlated with %CO (R = 0.76), and the positive predictive value of the cutoff %COes was 94% in the test group. Further development of this CO monitor is strongly warranted for its reliable clinical application to perioperative hemodynamic management.