GRU-TSMixers: Sleep Apnea and Hypopnea Detection Based on Multi Scale MLP-Mixers

Zufang Huang, Kejing He

Abstract: Sleep apnea is one of the most common sleep breathing disorders and can lead to other serious diseases. The automated detection of sleep apnea and hypopnea events is crucial for preventing the of potential health complications associated with these disorders. This paper presents a novel deep learning model, the GRU-TSMixers, designed for the efficient and accurate identification of these events using a single channel of raw respiratory signals. Our approach leverages Gated Recurrent Units (GRU) to capture the temporal dynamics of the respiratory signal, while employing TSMixers to intricately extract features without any feature engineering. The model’s effectiveness is demonstrated through rigorous evaluation on two large-scale datasets, the Sleep-Heart-Health-Study (SHHS) and the Multi-Ethnic Study of Atherosclerosis (MESA), where it consistently outperforms traditional CNN and LSTM models as well as other state-of-the-art approaches. Our findings suggest that GRU-TSMixers not only sets a new benchmark for sleep event detection but also paves the way for advancements in non-intrusive diagnostic tools in sleep medicine.

External IDs:dblp:conf/ijcnn/HuangH24