Improving difficult direct laryngoscopy prediction using deep learning and minimal image analysis: a single-center prospective study

Jong-Ho Kim,Hee-Sun Jung,So-Eun Lee,Jong-Uk Hou,Young-Suk Kwon

DOI: https://doi.org/10.1038/s41598-024-65060-x

IF: 4.6

2024-06-22

Scientific Reports

Abstract:Accurate prediction of difficult direct laryngoscopy (DDL) is essential to ensure optimal airway management and patient safety. The present study proposed an AI model that would accurately predict DDL using a small number of bedside pictures of the patient's face and neck taken simply with a smartphone. In this prospective single-center study, adult patients scheduled for endotracheal intubation under general anesthesia were included. Patient pictures were obtained in frontal, lateral, frontal-neck extension, and open mouth views. DDL prediction was performed using a deep learning model based on the EfficientNet-B5 architecture, incorporating picture view information through multitask learning. We collected 18,163 pictures from 3053 patients. After under-sampling to achieve a 1:1 image ratio of DDL to non-DDL, the model was trained and validated with a dataset of 6616 pictures from 1283 patients. The deep learning model achieved a receiver operating characteristic area under the curve of 0.81–0.88 and an F1-score of 0.72–0.81 for DDL prediction. Including picture view information improved the model's performance. Gradient-weighted class activation mapping revealed that neck and chin characteristics in frontal and lateral views are important factors in DDL prediction. The deep learning model we developed effectively predicts DDL and requires only a small set of patient pictures taken with a smartphone. The method is practical and easy to implement.

multidisciplinary sciences

What problem does this paper attempt to address?

This paper attempts to address the problem of accurately predicting Difficult Direct Laryngoscopy (DDL). Specifically, the authors propose an artificial intelligence model based on deep learning that uses a small number of bedside photos of the patient's face and neck to predict DDL. This approach aims to improve the safety and efficiency of airway management and reduce medical risks caused by unforeseen DDL.

### Background and Problem
1. **Importance of Airway Management**: Airway management is crucial in emergency care, anesthesia, and intensive care to ensure that patients can maintain natural ventilation when intubation and other procedures are needed.
2. **Necessity of Predicting Difficult Airway**: Accurately predicting a difficult airway is essential for planning ahead and taking targeted measures to reduce the risk of airway obstruction or failure.
3. **Limitations of Existing Methods**: Common bedside assessment tests (such as the Cormack-Lehane classification) are widely used but have low sensitivity, making it difficult to reliably identify all potential difficult airways.

### Solution
1. **Data Collection**: Researchers collected 18,163 photos from 3,053 patients, including front, side, anterior neck extension, and mouth-opening views.
2. **Model Construction**: A deep learning model using the EfficientNet-B5 architecture was constructed, incorporating multi-task learning techniques to include image perspective information in the model training.
3. **Performance Evaluation**: Through cross-validation, the model performed well in predicting DDL, with an area under the ROC curve (AUC) of 0.81–0.88 and an F1 score of 0.72–0.81.

### Main Contributions
1. **Improved Prediction Accuracy**: The model significantly improved the accuracy of DDL prediction by incorporating image perspective information.
2. **Simplicity and Ease of Use**: The method is practical and easy to implement, requiring only a few simple photos taken with a smartphone for prediction.
3. **Clinical Application Prospects**: The model can be widely applied in various clinical settings, helping to improve the safety and efficiency of airway management.

### Conclusion
This study successfully developed a deep learning-based model that can effectively predict Difficult Direct Laryngoscopy using only a small number of bedside photos of the patient. This method not only improves prediction accuracy but is also simple and easy to use, with the potential for widespread application in clinical practice.