DEEP LEARNING FRAMEWORK FOR AUTOMATED PATIENT–VENTILATOR ASYNCHRONY DETECTION AND CLASSIFICATION
International Journal of Drug Delivery Technology · 08.06.2026 · 16 (50s)
DOI: 10.25258/ijddt.16.50s.40 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Drug Delivery Technology, online erschienen am 08.06.2026. Verfasst von V. Menaka und 6 weiteren Autor:innen.
Abstract (Originalsprache)
Patient–ventilator asynchrony has remained as a frequent and a clinically relevant complication in a mechanically ventilated patients, which has contributed to an increased respiratory effort, prolonged ventilation, and unfavorable outcomes. The conventional identification of asynchrony has relied on the manual interpretation of the airway pressure and flow waveforms, which has required a substantial clinical expertise and it has suffered from the observer variability. Recent advances in the machine learning have enabled the data-driven analysis of the physiological signals, which offers an opportunity to improve the scalability. Manual waveform inspection has lacked the consistency, and it has limited a continuous monitoring in an intensive care environments. Earlier machine learning (ML) approaches have used a handcrafted features and classical classifiers have achieved moderate accuracy but it have struggled to generalize across the diverse waveform patterns. There it has remained as a need for a robust framework that has automatically learned the discriminative features directly from the ventilator signals while it maintains the clinical feasibility. This study has proposed a unified deep learning framework that has employed a four convolutional neural network models for an automated detection of the patient–ventilator asynchrony. Airway pressure and flow signals have been collected, normalized, and segmented into the fixed-length epochs. Two one-dimensional CNN models have processed the raw waveform inputs, while two two-dimensional CNN models have analyzed the time– frequency spectrogram representations, which includes a transfer learning–based ResNet-50 architecture. Model training has used a supervised learning with the standardized evaluation metrics. The proposed CNN-based framework has achieved a superior performance across all the evaluation metrics. Over various iterations, the framework has attained a maximum accuracy of 96%, precision of 95%, recall of 96%, F1-score of 95%, and specificity of 97%, which consistently has surpassed the Random Forest, Support Vector Machine, and Shallow 1D CNN methods.
Zitieren
APA
Menaka, V., Mehata, K., Narayanan, N. L., Manivannan, K., Jaishankar, P., Rajakumar, P., & Manonmani, V. (2026). DEEP LEARNING FRAMEWORK FOR AUTOMATED PATIENT–VENTILATOR ASYNCHRONY DETECTION AND CLASSIFICATION. International Journal of Drug Delivery Technology, 16(50s). https://doi.org/10.25258/ijddt.16.50s.40
Vancouver
Menaka V, Mehata K, Narayanan NL, Manivannan K, Jaishankar P, Rajakumar P, et al. DEEP LEARNING FRAMEWORK FOR AUTOMATED PATIENT–VENTILATOR ASYNCHRONY DETECTION AND CLASSIFICATION. International Journal of Drug Delivery Technology. 2026;16(50s). doi:10.25258/ijddt.16.50s.40
BibTeX
@article{Menaka2026,
author = {Menaka, V. and Mehata, K.M and Narayanan, N. Lakshmi and Manivannan, K. and Jaishankar, P. and Rajakumar, P.S and Manonmani, V.},
title = {DEEP LEARNING FRAMEWORK FOR AUTOMATED PATIENT–VENTILATOR ASYNCHRONY DETECTION AND CLASSIFICATION},
journal = {International Journal of Drug Delivery Technology},
year = {2026},
volume = {16},
number = {50s},
doi = {10.25258/ijddt.16.50s.40}
}
EUVIMED