Respiratory failure & ventilation
Development of a deep-learning model for classification of intensivists' extubation decisions using ventilator graphic monitor images: a prospective observational study
Authors' conclusion
This study developed a deep-learning model capable of classifying intensivists' extubation decisions from ventilator graphic monitor images with good classification performance (AUC 0.900).
Study context
- Population
- Adults aged ≥ 20 years on mechanical ventilation undergoing SAT and spontaneous breathing trial (SBT) with a Puritan Bennett™ 980 ventilator, ICU of Yamagata University Hospital, August 2023–July 2025160 extubated, 31 non-extubated
- Participants
- 191 adults
- Exposure
- Four CNN architectures (DenseNet121, EfficientNet-B0, MobileNetV3-Large, ResNet18) classifying intensivists' extubation decisions from ventilator graphic monitor images
- Outcome
- Classification of intensivists' extubation decisions from ventilator graphic monitor images at SBT outcome determination
Methodological detailssingle-centre, deep-learning model development, threefold cross-validation, independent test set, Four study periods: threefold cross-validation on first three, fourth used as independent test set
- Single-centre
- Deep-learning model development
- Threefold cross-validation
- Independent test set
- Four study periods: threefold cross-validation on first three, fourth used as independent test set
Main finding
As reported in the abstract
Classification performance
| Group | AUC | Sensitivity | Specificity | Accuracy |
|---|---|---|---|---|
| MobileNetV3-Large | 0.900 (95% CI 0.807-0.993) | 0.900 | 0.727 | 0.759 |
MobileNetV3-Large
AUC0.900 (95% CI 0.807-0.993)
Sensitivity0.900
Specificity0.727
Accuracy0.759
- AUC range across all four architectures
- 0.827-0.900
Observational data: associations, not effects.
Further findings
Grad-CAM visualisation
- Model focused on entire waveform (pressure, flow, volume) in extubated group, primarily on flow waveform in non-extubated group
Original publication
Development of a deep-learning model for classification of intensivists' extubation decisions using ventilator graphic monitor images: a prospective observational study. J Intensive Care. Published 25 May 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →