Testing an inverse modeling approach with gradient boosting regression for stroke volume estimation using patient...

Area of Science:

• Cardiovascular Physiology
• Biomedical Engineering
• Machine Learning in Medicine

Background:

• Stroke volume (SV) is crucial for assessing cardiovascular function, heart performance, and blood flow adequacy.
• Accurate SV measurement is vital for managing heart failure, surgical patients, and critical care.
• Traditional invasive methods (e.g., thermodilution) and imprecise non-invasive techniques limit continuous monitoring.

Purpose of the Study:

• To develop a novel, non-invasive method for accurate stroke volume estimation.
• To integrate a validated 1-D systemic circulation model with machine learning for SV prediction.
• To replace traditional optimization processes with a faster, regression-based approach.

Main Methods:

• Developed a gradient boosting regression model for SV estimation.
• Utilized an in silico-generated dataset mimicking 1-D systemic circulation dynamics.
• Validated the method against the gold standard thermodilution technique in 24 patients.

Main Results:

• The ML-based method showed satisfactory agreement with thermodilution (MAE: 16 mL, nRMSE: 21%, Bias: -9.2 mL, LoA: [-47, 28] mL).
• A significant correlation was found (r=0.7, p<0.05).
• Predicted SV (68±23 mL) was slightly underestimated compared to reference SV (77±26 mL).

Conclusions:

• The novel ML-enhanced 1-D model provides accurate and rapid SV estimation.
• Reduced computational time makes the method suitable for real-time clinical applications.
• This non-invasive approach offers a promising alternative for continuous hemodynamic monitoring.