A Machine Learning Approach for Pulse Detection using Cerebral Oximetry Signals during Out-of-Hospital Cardiac...

Area of Science:

• Cardiology
• Biomedical Engineering
• Critical Care Medicine

Background:

• Out-of-hospital cardiac arrest (OHCA) is a major global cause of mortality.
• Accurate discrimination between pulseless electrical activity (PEA) and pulse-generating rhythm (PR) is crucial for detecting return of spontaneous circulation (ROSC) and improving patient survival.
• Existing ECG-based classifiers struggle with PEA/PR discrimination due to signal similarities.

Purpose of the Study:

• To develop and evaluate a novel random forest (RF) classifier for improved PEA/PR discrimination.
• To assess the added value of combining cerebral oximetry signals with ECG-derived features.
• To enhance the efficiency and interpretability of ROSC detection during OHCA.

Main Methods:

• A novel RF classifier was developed using concurrent ECG and high-temporal-resolution cerebral oximetry signals.
• A dataset of 165 5-second segments (90 PR, 75 PEA) was used, with 5-fold cross-validation for feature selection and model training.
• Feature selection involved evaluating 350 discrimination features, reducing the set to enhance computational efficiency.

Main Results:

• The combined classifier, using only 11 features, achieved a mean balanced accuracy of 85.3% (±2.8%).
• Mean sensitivity and specificity were 82.4% (±4.8%) and 88.3% (±5.1%), respectively.
• The combined model demonstrated comparable performance to ECG-only models but with significantly fewer features.

Conclusions:

• Integrating high-resolution cerebral oximetry signals significantly enhances PEA/PR discrimination for ROSC detection.
• The novel RF classifier offers a computationally efficient and interpretable approach compared to traditional ECG-only methods.
• This approach holds promise for improving outcomes in patients experiencing OHCA.