Towards a Lightweight Classifier to Detect Hypovolemic Shock.

Area of Science:

• Biomedical Engineering
• Physiology
• Machine Learning in Healthcare

Background:

• Early detection of hypovolemic shock is critical for timely clinical intervention.
• The compensatory reserve metric (CRM) quantifies the body's ability to manage blood loss.

Purpose of the Study:

• To develop a lightweight, three-class predictor for CRM.
• To assess the feasibility of real-time hypovolemia monitoring using a wearable device.

Main Methods:

• Utilized a linear support-vector machine (SVM) classifier for prediction.
• Employed a moving mean filter for smoothing predictions.
• Focused on a three-class classification: good, fair, and poor compensatory reserve.

Main Results:

• Developed a predictor for CRM with good, fair, and poor classifications.
• Demonstrated a feasible model for real-time hypovolemia monitoring.
• The model requires minimal storage (408 bytes) and computational resources.

Conclusions:

• The developed SVM classifier provides a feasible approach for real-time hypovolemia monitoring.
• The lightweight nature of the model supports implementation on wearable devices.
• Early detection of declining compensatory reserve can facilitate timely medical intervention.