Artificial Intelligence in Snoring Sound Analysis: OSA Detection and Obstruction Site Classification, a Systematic...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Sleep Medicine

Background:

• Obstructive sleep apnea (OSA) is a prevalent sleep disorder.
• Accurate detection and classification of OSA are crucial for effective management.
• Snoring sounds offer a potential noninvasive biomarker for OSA diagnosis.

Purpose of the Study:

• To systematically review the application of machine learning (ML) and artificial intelligence (AI) in analyzing snoring sounds.
• To evaluate the efficacy of ML/AI models in detecting OSA.
• To assess the capability of these models in classifying upper airway obstruction sites using the Velum, Oropharynx, Tongue, and Epiglottis (VOTE) system.

Main Methods:

• A comprehensive literature search was conducted across major databases (PubMed, Scopus, etc.) from 2011 to November 2024.
• Studies were screened and reviewed following PRISMA 2020 guidelines, with 42 studies included.
• Methodological quality was assessed using PROBAST, focusing on ML models, datasets, feature extraction, and performance metrics.

Main Results:

• ML models, including SVMs and CNNs, achieved high accuracy (up to 98.6%) for OSA detection and snore classification.
• Key acoustic features like MFCCs and wavelet transforms were identified.
• VOTE classification performance varied, with targeted features improving accuracy to over 95% in some studies, though challenges like data imbalance and noise persist.

Conclusions:

• AI-powered snore analysis presents a viable noninvasive approach for OSA screening and anatomical site classification.
• Further research should emphasize multimodal data integration and real-world validation for clinical translation.
• Enhancing model generalizability is essential for widespread adoption in clinical practice.