Research on Classification of Patient-ventilator Asynchrony Using Permutation Disalignment Index

Abstract

The study of patient-ventilator asynchrony (PVA) is of great significance to improve the respiratory condition of critically ill patients and improve their clinical comfort. In the intensive care unit (ICU), the automatic alarm system for respiratory asynchrony usually relies on simple thresholds, resulting in frequent false positives, which will cause great interference to medical staff. In current clinical applications, PVA is still detected by visually observing the pressure, flow and volume curves, which is very time-consuming. Therefore, we aim to develop a classification model based on the permutation disalignment index (PDI) to check PVA events and facilitate the diagnosis of medical staff. We extracted the PDI of the respiratory signal, used the decision tree algorithm and the random forest algorithm for classification and evaluation. Results showed that the accuracy of classification using the PDI feature and the decision tree algorithm reached 0.964, the Recall score reached 0.966, the F1 score was 0.960. The accuracy of classification using the PDI feature and the random forest algorithm reached 0.966, the Recall score reached 0.953, and the F1 score was 0.966. It indicates that PDI is a promising feature to detect PVA.