Multimodal Artificial Intelligence for Sleep: Integrating Multichannel Signals for Comprehensive Assessment

Sleep medicine generates rich, multimodal physiologic data that has traditionally been reduced to summary metrics through labor-intensive manual interpretation. Artificial intelligence (AI), particularly deep learning, is transforming the field by enabling scalable, reproducible, and integrated analysis of polysomnographic signals. This review examines 5 interrelated frontiers of multimodal AI in sleep medicine: foundation models pre-trained on diverse biosignal datasets that may unify analysis across devices and populations; automated sleep staging systems that match individual expert reliability; AI-driven respiratory analysis that moves beyond the Apnea-Hypopnea Index toward continuous, endotype-level characterization of sleep-disordered breathing; AI-derived sleep biomarkers, including electroencephalography microstructure, brain age.