Image augmentation and automated measurement of endotracheal-tube-to-carina distance on chest radiographs in intensive care unit using a deep learning model with external validation
Authors' conclusion
The CarinaNet model is an efficient and generalizable deep learning algorithm for the automated assessment of ETT placement on ICU chest radiographs.
Study context
- Population
- Intensive care unit (ICU) chest radiographs; external test set from the MIMIC-CXR databaseimages, external test set
- Participants
- 200 participants
- Comparison
- CarinaNet deep learning model vs human interpretation of endotracheal tube (ETT) position on ICU chest radiographs
- Outcome
- Accuracy (mean absolute error) of predicted ETT tip position, carina position and ETT-carina distance
Methodological detailsdeep learning model, transfer learning, external validation, uncertainty quantification, image segmentation
- Deep learning model
- Transfer learning
- External validation
- Uncertainty quantification
- Image segmentation
Performance
As reported in the abstract
MAE of the model on the external test set (200 MIMIC-CXR images)
- MAE, ETT tip position prediction
- 0.51 cm
- MAE, carina position prediction
- 0.61 cm
- MAE, ETT-carina distance prediction
- 0.89 cm
- Interrater agreement, ETT tip position
- 0.18 cm
- Interrater agreement, carina position
- 0.58 cm
- Interrater agreement, ETT-carina distance
- 0.60 cm
Further findings
Interrater agreement
- ETT tip position (cm)
- 0.18
- Carina position (cm)
- 0.58
- ETT-carina distance (cm)
- 0.60
Assessment of ETT placement with model support
- Improved by complementing human interpretation of chest radiographs with CarinaNet
Original publication
Image augmentation and automated measurement of endotracheal-tube-to-carina distance on chest radiographs in intensive care unit using a deep learning model with external validation. Crit Care. Published 25 January 2023. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →