Sleep Apnea Detection using Mel-spectrograms Snoring and Convolutional Neural Networks

Obstructive sleep apnea (OSA) is a chronic disease characterized by intermittent hypoxemia during sleep related to snoring. It affects the quality of life and increases the risk of severe health conditions, including cardiovascular diseases. The gold standard for diagnosing OSA is polysomnography (PSG), which requires an overnight hospital stay while physically connected to 10-15 measurement channels. PSG is costly, inconvenient, and requires the involvement of a sleep technologist. Such as, over 80% of affected individuals remain undiagnosed. Therefore, cost-effective and non-invasive screening methods for OSA play a fundamental role in improving people's file quality. Approaches based on deep learning techniques have achieved evaluable results. However, such results are not reproducible due to the lack of code and dataset, making it difficult to evaluate the impact of these methods on first-level diagnosis scenarios.In this work, we face apnea detection as a classification image problem. The introduced method exploits the Mel-spectrograms of snoring and VGG19, an architecture based on Convolutional Neural Networks (CNN), to detect apnea. We test our approach on a public dataset that stores data related to polysomnography with simultaneous audio recordings for sleep apnea studies. On this dataset, our methods archive 95, 4% of accuracy. The analysis of the performance values shows that our method reaches competitive results.