Machine learning-enabled estimation of cardiac output from peripheral waveforms is independent of blood pressure measurement location in an in silico population

Monitoring of cardiac output (CO) is a mainstay of hemodynamic management in the acutely or critically ill patient. Invasive determination of CO using thermodilution, albeit regarded as the gold standard, is cumbersome and bears risks inherent to catheterization. In the pursuit of noninvasive methods, prediction of CO from arterial pressure waveforms (even uncalibrated) through AI has garnered increased attention. In the current study we investigate the effect of peripheral blood pressure measurements’ (i) location, (ii) calibration and (iii) injected additive noise on the performance of AI-based CO estimation. A large previously generated virtual cohort of n = 3818 subjects with varied hemodynamic profiles served as data bank for arterial pulse waves and reference CO values. Two-layered convolutional neural networks (CNN) yielded CO estimates based on entire pressure traces from the radial, superficial temporal and common carotid arteries. The predictive ability of the CNN models was comparable across arterial locations (r ≥ 0.97, nRMSE ≤ 4.4%) and only slightly compromised following pressure signal normalization from 0 to 1 (r ≥ 0.93, nRMSE ≤ 7.8%). Following noise injection, model performance was preserved (r ≥ 0.94, nRMSE ≤ 6.6%) yet became more sensitive to loss of calibration (r ≥ 0.78, nRMSE ≤ 12.2%). Our study highlights the interchangeability of potential blood pressure measurement locations in the quest for pressure-based CO estimation.