A machine-learning based analysis for the recognition of progressive central hypovolemia

Area of Science:

• Physiological monitoring
• Machine learning in healthcare
• Surgical patient safety

Background:

• Traditional monitoring (heart rate, blood pressure) has limitations in predicting central hypovolemia.
• Decreased central blood volume (CBV) can lead to cerebral hypoperfusion during surgery.
• Early detection of CBV changes is crucial for preventing surgical complications.

Purpose of the Study:

• To develop an advanced monitoring model for early detection of decreased CBV.
• To identify physiological predictors of central hypovolemia.
• To create a system that indicates the need for intervention before pre-syncope.

Main Methods:

• Twenty-eight healthy subjects underwent induced central hypovolemia using lower body negative pressure.
• Physiological parameters were measured and processed using a 30-second moving window.
• A neural network model was trained to predict time to pre-syncope.

Main Results:

• An optimal model with 10 hidden neurons and 80 iterations was identified.
• The model achieved an average prediction slope of -0.64.
• It successfully predicted the need for intervention over 200 seconds before pre-syncope in 75% of subjects.

Conclusions:

• A neural network model can detect decreases in CBV independently of heart rate and blood pressure changes.
• This model facilitates early intervention, mitigating the risk of symptomatic cerebral hypoperfusion.
• The findings support enhanced patient monitoring during surgical procedures.