Enhancing Trauma Care: Machine Learning-Based Photoplethysmography Analysis for Estimating Blood Volume During...

Area of Science:

• Biomedical Engineering
• Machine Learning in Medicine
• Trauma Care Technology

Background:

• Hemorrhage is a leading cause of preventable death in trauma.
• Compensatory mechanisms can mask hemorrhagic shock, delaying critical interventions.
• Accurate estimation of blood loss is crucial for effective trauma management.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting Percent Estimated Blood Loss (PEBL).
• To utilize non-invasive photoplethysmography (PPG) waveforms for PEBL estimation.
• To offer field-deployable solutions for rapid hemorrhage detection in pre-hospital and emergency medicine.

Main Methods:

• Collected PPG waveform data during a hemorrhage and resuscitation swine study.
• Tuned and optimized various ML model architectures and prediction window lengths.
• Evaluated different data normalization approaches for model training.

Main Results:

• ML models successfully predicted PEBL in swine subjects.
• Achieved coefficient of determination values exceeding 0.8 for PEBL prediction.
• Demonstrated accurate derivation of PEBL from non-invasive PPG signals.

Conclusions:

• PEBL can be accurately estimated using non-invasive PPG waveforms and ML.
• This approach offers a promising tool for improving trauma care and casualty triage.
• Non-invasive, field-deployable solutions can enhance decision-making in emergency medicine.