Automatic detection of CO₂ rebreathing during BiPAP ventilation

Area of Science:

• Critical Care Medicine
• Respiratory Physiology
• Biomedical Engineering

Background:

• Carbon dioxide rebreathing (CO₂ rebreathing) impacts respiratory drive and work of breathing during BiPAP ventilation.
• Current methods for detecting CO₂ rebreathing are often expensive and not routinely used.

Purpose of the Study:

• To develop a real-time method for detecting CO₂ rebreathing during BiPAP ventilation.
• To identify respiratory cycle types based on CO₂ movement without direct CO₂ concentration measurement.

Main Methods:

• Observational study in an ICU involving 18 patients on BiPAP ventilation.
• Analysis of airflow, pressure, and CO₂ movement across 4747 respiratory cycles.
• Application of statistical tests (ANOVA, t-tests, CDA) and a multilayer perceptron (MLP) artificial neural network.

Main Results:

• Three distinct respiratory cycle types were identified based on CO₂ movement; type III cycles induce rebreathing.
• Canonical Discriminant Analysis correctly classified 93.9% of cycles, with 97.9% accuracy for type III.
• The MLP network achieved 98.8% accuracy in automatically classifying respiratory cycle types.

Conclusions:

• Respiratory cycle types can be distinguished by CO₂ movement during BiPAP ventilation.
• Artificial neural networks offer a highly accurate method for real-time detection of CO₂ rebreathing-inducing cycles (type III).