Automated Evaluation of Upper Airway Obstruction Based on Deep Learning

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Upper airway obstruction is a critical condition requiring accurate diagnosis.
• Lateral cephalograms are commonly used in assessing airway morphology.
• Current screening methods can be time-consuming and subjective.

Purpose of the Study:

• To develop a deep learning-based screening tool for evaluating upper airway obstruction.
• To automate the analysis of lateral cephalograms for improved efficiency and accuracy.

Main Methods:

• A novel convolutional neural network (CNN) model utilizing a ResNet backbone was developed.
• The model was trained and tested on a dataset of 1219 lateral cephalogram X-ray images.
• Performance was compared against the VGG16 model.

Main Results:

• The developed CNN model achieved high performance metrics: sensitivity 0.86, specificity 0.89, positive predictive value (PPV) 0.90, negative predictive value (NPV) 0.85, and F1-score 0.88.
• Heat maps provided insights into the features learned by the deep learning model.
• The model outperformed VGG16 in evaluating upper airway obstruction.

Conclusions:

• Deep learning effectively extracts relevant features from cephalograms for automated upper airway obstruction evaluation.
• The established CNN model can reduce dentists' screening workload and enhance diagnostic accuracy.
• This AI-driven approach offers a practical solution for improving upper airway obstruction assessment.