EVALUATING MACHINE LEARNING ALGORITHMS FOR THE DETECTION OF PATIENT VENTILATOR ASYNCHRONY IN HYPOXEMIC RESPIRATORY FAILURE PATIENTS: A SYSTEMATIC REVIEW.

Publication Details

Publication Date

10/09/2026

Volume / Issue

Vol 1, Issue 4 (2026)

Article No.

021

Journal

British Journal of Contemporary Research

DOI

https://doi.org/10.67693/BJCR-HSCY2K98

Received

18 Aug 2026

Views

5

Downloads

0

Abstract

Background: Patient–ventilator asynchrony (PVA) remains a major challenge during invasive mechanical ventilation, resulting from inadequate coordination between ventilator-delivered breaths and the patient’s intrinsic respiratory effort. PVA is clinically significant due to its high prevalence and its association with ventilator-induced lung injury (VILI), prolonged mechanical ventilation, and increased ICU stay, particularly in patients with hypoxemic respiratory failure. The most frequently reported PVA subtypes include reverse triggering, double triggering, and ineffective inspiratory effort. Artificial intelligence (AI), particularly neural network-based machine learning models, has emerged as a promising approach for automated waveform analysis and real-time detection of PVA. Methods: A systematic review was conducted using structured database searches of Pub Med, Web of Science, and Embase. Ten original studies (7 prospective and 3 retrospective) published between 2018 and 2025 were included, focusing on neural network-based automated detection of PVA using ventilator waveform data. Model performance, detection of PVA sub-types, and diagnostic accuracy were evaluated. Results: Among the included studies,convolutional neural networks (CNN) were utilized in 4studies, 3 studies utilized recurrent neural network architectures(RNN), Logistic regression and decision tree were applied in 1 and 2 studies respectively with some studies employing hybrid deep learning frameworks. Sample sizes ranged from 750 patient recordings and 40,807,279 breath counts. Neural network models demonstrated high diagnostic performance, with sensitivity >0.90 and specificity ranging from >0.90, F1 scores>0.90. Accuracy (AUC) values greater >0.90. Reverse triggering and double triggering demonstrated the highest detection accuracy (>92% in several studies), whereas ineffective inspiratory effort detection accuracy ranged between 54% and 90%. Conclusion: Neural network-driven detection systems show strong potential for accurate real-time identification of PVA and may enhance ventilator management and patient outcomes.

License

This article is published under the Creative Commons Attribution 4.0 International License . Free to read, share, and adapt with attribution.