Machine learning for image-based detection of patients with obstructive sleep apnea: an exploratory study

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Sleep Medicine

Background:

• Patients with severe obstructive sleep apnea (OSA) often present with a crowded oropharynx on lateral cephalometric radiographs compared to non-OSA individuals.
• Lateral cephalometric radiography is a standard imaging technique used in diagnosing craniofacial abnormalities and sleep-related breathing disorders.

Purpose of the Study:

• To investigate the efficacy of machine learning (ML), a subset of artificial intelligence (AI), in identifying individuals with severe OSA using 2-dimensional (2D) lateral cephalometric radiographs.
• To compare the diagnostic performance of a deep convolutional neural network (CNN) with traditional manual cephalometric analysis for OSA detection.

Main Methods:

• A deep convolutional neural network (CNN) was developed and trained on 1258 lateral cephalometric radiographs and tested on 131 radiographs from individuals diagnosed with severe OSA or non-OSA.
• Three image datasets were created: full image, main region (facial profile, upper airway, craniofacial tissues), and head only. A radiologist performed manual cephalometric analysis for comparison.

Main Results:

• The CNN achieved high diagnostic accuracy, with the "main region" dataset yielding the highest area under the receiver-operating characteristic curve (AUC) of 0.92.
• Sensitivity and specificity for the "main region" were 0.88 and 0.75, respectively. The full image dataset also showed strong performance (AUC 0.89).
• Manual cephalometric analysis had a lower AUC of 0.75, indicating the superior performance of the AI model in this study.

Conclusions:

• A deep convolutional neural network can accurately detect individuals with severe obstructive sleep apnea (OSA) from lateral cephalometric radiographs.
• The findings support the potential of AI-powered image analysis for improving the efficiency and accuracy of OSA screening and triage.
• Further research is warranted to explore the clinical application of AI in diagnosing and managing OSA.