Abstract

Obstructive sleep apnea (OSA) is largely underdiagnosed due to the reference test, overnight polysomnography (PSG) requiring a sleep laboratory and trained staff. Because loud, habitual snoring is among the first outward signs of OSA, a number of recent studies ask whether deep learning can turn snoring audio and low-cost physiological signals into an early warning outside the clinic. This paper reviews seven recent studies and several supporting works for that question, including CNN, attention-based CNN–BiLSTM, ResNet with audio spectrogram transformers, and end-to-end PSG models.. We compare them by method, reported findings and stated limitations, outline the common processing pipeline and discuss privacy and ethical concerns. Deep models show promise for screening, but practical use is limited by small datasets, class imbalance, heavy computation, weak explainability and little validation in home settings. In summary, we summarize the design requirements for a lightweight, explainable apnea risk system deployable in the home.

Keywords

How to Cite This Article

Rights