Predicting obstructive sleep apnea severity from craniofacial images using ensemble machine learning models.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Sleep Medicine

Background:

• Obstructive sleep apnea (OSA) affects a significant portion of the global population, leading to various health issues.
• Current diagnosis relies on polysomnography (PSG), which is costly and inaccessible for many.
• There is a need for alternative, cost-effective OSA diagnostic methods.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting OSA severity using 2D craniofacial images.
• To compare the model's performance against existing craniofacial analysis methods and the STOP-BANG questionnaire.

Main Methods:

• A machine learning model was trained to predict OSA severity from 2D frontal view craniofacial images.
• The model was validated using a cross-validation study on 280 patients.
• Performance was assessed using the Area Under the Curve (AUC) metric.

Main Results:

• The proposed machine learning model achieved an average AUC of 0.780.
• This significantly outperformed a previous craniofacial analysis model (AUC 0.638) and the STOP-BANG questionnaire (AUC 0.52) on the same dataset.

Conclusions:

• Deep learning models utilizing craniofacial images show significant potential for OSA diagnosis.
• This approach could substantially reduce the cost and improve accessibility of OSA diagnosis.
• Facial imaging offers a promising avenue for developing more accessible sleep apnea screening tools.