Machine Learning Algorithm to Estimate Cardiac Output Based On Less-Invasive Arterial Blood Pressure Measurements.

Area of Science:

• Cardiovascular Physiology
• Biomedical Engineering
• Machine Learning in Medicine

Background:

• Cardiac output (CO) is a critical hemodynamic parameter representing the heart's pumping volume per minute.
• Estimating CO non-invasively using arterial blood pressure (ABP) waveforms is desirable but lacks a defined relationship.
• Existing methods for CO estimation from ABP are limited, necessitating advanced analytical techniques.

Purpose of the Study:

• To elucidate the complex, previously unknown relationship between cardiac output (CO) and arterial blood pressure (ABP) waveforms.
• To develop and validate a machine learning-based approach for estimating CO from ABP.
• To determine the optimal number of cardiac cycles for robust feature extraction in CO estimation.

Main Methods:

• Utilized machine learning and advanced feature engineering to analyze the CO-ABP relationship.
• Applied the sparse identification of non-linear dynamics (SINDy) algorithm for feature discovery from ABP.
• Investigated the impact of varying cardiac cycle numbers on feature extraction performance.

Main Results:

• Achieved clinically acceptable performance in estimating CO from ABP, validated by radial limits of agreement (RLOA) and radial bias (RBias).
• The proposed machine learning model demonstrated reliable CO estimation on an independent external dataset.
• Discovered potential similarities between the derived dynamic models and the Navier-Stokes equations.

Conclusions:

• The study successfully established a data-driven relationship between CO and ABP using machine learning.
• The developed method offers a promising, less-invasive approach for hemodynamic monitoring and CO assessment.
• Further research into the connection with fluid dynamics principles like Navier-Stokes equations may offer deeper physiological insights.