Abstract:Objectives: Snoring is a common symptom of obstructive sleep apnea (OSA) which is considered to be potential predictors of the obstruction site. Successful treatment of OSA depend on the determination the types of obstruction site. This study aimed to develop a machine learning-based model to detect obstruction site using snoring sound.Methods: Patients with OSA underwent drug-induced sleep endoscopy (DISE) and the snoring sounds were recorded simultaneously. We extracted acoustic features based on Mel-frequency cepstral coefficients (MFCC). A k-nearest neighbors (KNN) was used for snore classification.Results: Total 42 patients with OSA were enrolled. The accuracy of model was 85.55 %, F1 score was 85.04. With combined age, gender and Body Mass Index (BMI), the accuracy of model was 87.98 %, and F1 score was 87.96. The model exhibited accuracies of 83 %, 93 % and 92 %; an AUC of 85.88, 89.22 and 88.17 in detecting ret-ropalatal, retrolingual and multilevel obstructions.Conclusion: Our results suggest that combing snoring sound with age, gender and BMI, the machine learning based model can help automatically assess obstruction site. The model may have potential utility as a clinical tool to help for clinical decision-making.

Automatic Classification of the Obstruction Site in Obstructive Sleep Apnea Based on Snoring Sounds.