Artificial intelligence in imaging for obstructive sleep apnea: A comprehensive review

Obstructive Sleep Apnea (OSA) is a prevalent disorder, yet its primary diagnostic tool, polysomnography (PSG), is costly and complex. Artificial intelligence (AI) in medical imaging offers a promising approach by directly analyzing the anatomical underpinnings of OSA. However, a systematic comparison to guide the use of various imaging modalities is lacking. This review evaluates AI applications across different imaging techniques for OSA, compares their strengths and limitations, and proposes a framework for future applications.

This article provides a comprehensive narrative review with a structured search strategy, summarizing and discussing key original research on the application of AI in medical imaging for OSA. We synthesize findings from six databases, including PubMed, Web of Science, Scopus, IEEE Xplore, Embase, and ScienceDirect, focusing on studies published from database inception to June 2025. We included studies that applied AI to analyze medical images for OSA diagnosis, severity assessment, or phenotyping, including studies related to OSA, such as airway segmentation. From 1963 initial records, 54 studies were included in this review.

The most frequently used modalities were computed tomography (CT), including cone-beam CT (CBCT), and magnetic resonance imaging (MRI). AI was primarily utilized for upper airway segmentation, disease prediction, severity assessment, and phenotyping, significantly enhancing analytical efficiency and objectivity compared to manual methods.

AI demonstrates significant promise in the diagnosis and management of OSA. This review provides a comprehensive comparison of AI applications in OSA imaging and proposes a novel five layers decision-making framework to guide modality selection across different clinical scenarios, from population screening to surgical planning and mechanistic research. Future clinical application hinges on developing large-scale standardized datasets, enhancing model explainability, and conducting robust clinical validation.