🤖 AI Summary
Non-invasive, acoustic-based localization of multi-level upper airway collapse in obstructive sleep apnea (OSA) remains clinically challenging.
Method: We formulate the first snore-driven multi-label classification task, leveraging anatomical (VOTE: velum, oropharynx, tongue base, larynx) and functional (RP/RG) annotations from drug-induced sleep endoscopy (DISE). Our model fuses fine-tuned ResNet-50 and Audio Spectrogram Transformer (AST), processing 0.5-second snore spectrograms to enable whole-night dynamic obstruction profiling.
Contribution/Results: AST achieves F1 scores of 0.71 (velum), 0.80 (oropharynx), and 0.86 (retropalatal/retroglossal), with corresponding AUCs of 0.89, 0.94, and 0.97—significantly outperforming baselines. This work is the first to demonstrate high-accuracy, snore-only localization of multi-level airway collapse, establishing a novel paradigm for OSA phenotyping and personalized therapeutic intervention.
📝 Abstract
This study investigated the use of deep learning to identify multi-level upper airway collapses in obstructive sleep apnea (OSA) patients based on snoring sounds. We fi-ne-tuned ResNet-50 and Audio Spectrogram Transformer (AST) models using snoring recordings from 37 subjects undergoing drug-induced sleep endoscopy (DISE) between 2020 and 2021. Snoring sounds were labeled according to the VOTE (Velum, Orophar-ynx, Tongue Base, Epiglottis) classification, resulting in 259 V, 403 O, 77 T, 13 E, 1016 VO, 46 VT, 140 OT, 39 OE, 30 VOT, and 3150 non-snoring (N) 0.5-second clips. The models were trained for two multi-label classification tasks: identifying obstructions at V, O, T, and E levels, and identifying retropalatal (RP) and retroglossal (RG) obstruc-tions. Results showed AST slightly outperformed ResNet-50, demonstrating good abil-ity to identify V (F1-score: 0.71, MCC: 0.61, AUC: 0.89), O (F1-score: 0.80, MCC: 0.72, AUC: 0.94), and RP obstructions (F1-score: 0.86, MCC: 0.77, AUC: 0.97). However, both models struggled with T, E, and RG classifications due to limited data. Retrospective analysis of a full-night recording showed the potential to profile airway obstruction dynamics. We expect this information, combined with polysomnography and other clinical parameters, can aid clinical triage and treatment planning for OSA patients.