Breathing Cycle-Aware Segmentation for Patient-Ventilator Asynchrony Detection.

Area of Science:

• Biomedical Engineering
• Critical Care Medicine
• Data Science

Background:

• Patient-ventilator asynchrony (PVA) affects 25% of ICU patients, increasing risks of lung and diaphragm injury.
• Accurate segmentation of breathing cycles is crucial for detecting PVA, but current methods are limited.

Purpose of the Study:

• To develop an unsupervised, breathing cycle-aware segmentation method for reliable PVA detection.
• To address limitations of manual annotation, fixed-length windows, rule-based, and supervised deep learning segmentation methods.

Main Methods:

• Proposed an unsupervised method integrating frequency-adaptive clustering, periodicity hints validation, and dynamic segmentation.
• Leveraged the quasi-periodic nature of ventilation waveforms for breathing cycle boundary identification.

Main Results:

• The proposed method outperformed baseline approaches on five out of six evaluation metrics on a real-world dataset.
• Accurate breathing cycle segmentation using the proposed method enhanced PVA detection performance with state-of-the-art deep learning models.

Conclusions:

• The unsupervised breathing cycle-aware segmentation method offers an effective solution for PVA detection.
• This method has the potential to be integrated into ventilation systems to support clinical decision-making and improve patient care.