Machine Learning-Based Cardiac Output Estimation Using Photoplethysmography in Off-Pump Coronary Artery Bypass...

Area of Science:

• Cardiovascular Physiology
• Biomedical Engineering
• Machine Learning in Medicine

Background:

• Hemodynamic monitoring is vital for critically ill patients and major surgeries.
• Cardiac output (CO) is a key indicator of hemodynamic status.
• Invasive CO measurement methods necessitate non-invasive alternatives.

Purpose of the Study:

• To develop a non-invasive algorithm for classifying cardiac index (CI).
• Utilize finger photoplethysmography (PPG) and machine learning for CI classification.
• Categorize CI into low and non-low groups.

Main Methods:

• Collected PPG and thermodilution CO data from patients undergoing off-pump coronary artery bypass graft surgery.
• Employed the Relief algorithm for feature selection.
• Trained and evaluated a CatBoost machine learning model.

Main Results:

• Achieved 89.42% accuracy in the validation phase and 87.57% in the testing phase.
• Demonstrated balanced performance across low and non-low CI categories.
• Showcased robust classification capabilities for hemodynamic status.

Conclusions:

• Machine learning combined with non-invasive PPG shows potential for accurate CO classification.
• The proposed method may improve patient safety and comfort in critical care.
• Further validation in diverse populations and scenarios is warranted.