Attention Fusion Network for Fine-Grained Sleep Apnea Detection Using Respiratory Signals

Increasing life pressures have led to the manifestation of various sleep-related symptoms, among which sleep apnea is one of the most prevalent. Recent researchers have developed numerous methods to assist in the diagnosis of the Apnea-Hypopnea Index in clinical settings, such as morphology, machine learning and deep learning. However, these methods do not possess the capability for precise event localization and do not offer comprehensive performance. Therefore, this paper proposes a fine-grained sleep apnea detection neural network (FG-AFSAN) that is based on respiratory signals, which can localize each event accurately, and incorporate an attention fusion mechanism to assess the severity of sleep apnea syndrome. We evaluated our model using both public and clinic datasets, and the results demonstrate that our model achieves a comparable mean Average Precision of 79.69% in event localization, an accuracy of 71% in AHI predictions, which highlights its potential in future clinical applications for sleep apnea screening.