PROJECT TITLE :

Acoustic Screening for Obstructive Sleep Apnea in Home Environments Based on Deep Neural

ABSTRACT:

Obstructive sleep apnea, also known as OSA, is a condition that is both persistent and common, and its associated comorbidities are well-known. However, due to a lack of access to polysomnography (PSG), the diagnostic method considered to be the gold standard for obstructive sleep apnea (OSA), many severe cases continue to go undiagnosed. We need accurate home-based screening methods for obstructive sleep apnea (OSA) that can be used on high-risk patients at a low cost to identify patients who need a polysomnogram (PSG) to evaluate their condition in its entirety. A variety of screening techniques for obstructive sleep apnea (OSA) have been researched in the past, and some of them analyze speech or breathing sounds. Nevertheless, these procedures have limitations that prevent them from being used in domestic settings (e.g., they require specialised equipment, are not robust to background noise, are obtrusive or depend on tightly controlled conditions). The authors of this study propose a novel method for screening for obstructive sleep apnea (OSA) that involves analyzing the sounds of sleep breathing that are recorded on a smartphone at home. Audio recordings made over the course of an entire night are broken up into segments, and a deep neural network analyzes each segment to determine whether or not it contains obstructive sleep apnea (OSA). The apnea-hypopnea index is then used to screen for the condition, and it is estimated from the segments that are predicted to contain evidence of obstructive sleep apnea (OSA). The proposed system was developed and evaluated with the help of audio recordings that were made during home sleep apnea testing from 103 participants over the course of one or two nights. The acoustics-based system achieved a sensitivity of 0.79 and a specificity of 0.80 when used for screening for moderate OSA. When screening for severe OSA, the sensitivity was 0.78 and the specificity was 0.93. Both of these values were very good. The system is suitable for application on smartphones designed for the general public.