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

Abstract

Background: Chest radiographs are routinely performed in intensive care unit (ICU) to confirm the correct position of an endotracheal tube (ETT) relative to the carina. However, their interpretation is often challenging and requires substantial time and expertise. The aim of this study was to propose an externally validated deep learning model with uncertainty quantification and image segmentation for the automated assessment of ETT placement on ICU chest radiographs.

Methods: The CarinaNet model was constructed by applying transfer learning to the RetinaNet model using an internal dataset of ICU chest radiographs. The accuracy of the model in predicting the position of the ETT tip and carina was externally validated using a dataset of 200 images extracted from the MIMIC-CXR database. Uncertainty quantification was performed using the level of confidence in the ETT-carina distance prediction. Segmentation of the ETT was carried out using edge detection and pixel clustering.

Results: The interrater agreement was 0.18 cm for the ETT tip position, 0.58 cm for the carina position, and 0.60 cm for the ETT-carina distance. The mean absolute error of the model on the external test set was 0.51 cm for the ETT tip position prediction, 0.61 cm for the carina position prediction, and 0.89 cm for the ETT-carina distance prediction. The assessment of ETT placement was improved by complementing the human interpretation of chest radiographs with the CarinaNet model.

Conclusions: The CarinaNet model is an efficient and generalizable deep learning algorithm for the automated assessment of ETT placement on ICU chest radiographs. Uncertainty quantification can bring the attention of intensivists to chest radiographs that require an experienced human interpretation. Image segmentation provides intensivists with chest radiographs that are quickly interpretable and allows them to immediately assess the validity of model predictions. The CarinaNet model is ready to be evaluated in clinical studies.

Keywords: Chest radiograph; Deep learning; Image processing; Intensive care unit; RetinaNet.

Fig. 1

Architecture of the RetinaNet model

Fig. 2

Bounding boxes around the carina and the endotracheal tube tip on a chest radiograph

Fig. 3

Illustration of the cluster thinning algorithm

Fig. 4

Number of chest radiographs per classification of the endotracheal tube tip position as a function of the annotated endotracheal tube–carina distance

Fig. 5

Exponential fit of the absolute error for the endotracheal tube–carina distance prediction as a function of the level of confidence in this prediction. The predictions were grouped into 8 bins based on the level of confidence and the mean absolute error was computed for each bin

Fig. 6

Confusion matrices for the classification of the endotracheal tube tip position by a clinician, the CarinaNet model, and a coupled reading

Fig. 7

Edge detection on the region of interest

Fig. 8

Clustering and cluster thinning on the binary edge map

Fig. 9

Example outputs of the CarinaNet model. Top-left and top-right chest radiograph were, respectively, classified as having acceptable quality and a hardly visible carina. The endotracheal tube tip and carina were accurately detected and the model uncertainty was inferior to 1 cm, indicating that the model result could be trusted. The segmentation was successful for both these radiographs. Bottom-left radiograph had a very unusual angle, this disrupted the segmentation step. However the endotracheal tube was entirely segmented and the endotracheal tube tip and carina were accurately detected. Bottom-right radiograph featured no endotracheal tube. The CarinaNet model properly detected the carina and output an endotracheal tube tip position but with an uncertainty of 3.3 cm indicating that the result could not be trusted. Note that the segmentation step also failed