70 / 2023-08-30 23:38:01

MuFF-E: Sleep Spindle Detection using Multi-Feature Fusion and Ensemble

Sleep Spindle Detection, U-Net, Multi-Feature Fusion, Model Ensemble.

Special Sessions > SS08 - IoT, AI Methods, and Applications in Health Monitoring and Diagnosis of Structure, Power, Medicine, etc

Sleep spindles are a distinct electroencephalographic (EEG) pattern observed during the non-rapid eye movement (NREM) sleep stage. Alterations in spindle properties may suggest changes in memory solidification or neurodegenerative diseases, so the detection of spindles is important in clinical research. Manual detection is too time-consuming and prone to intra- and inter-expert variability. Considering the distinctive frequency characteristics of spindles and the influence of gender and age on spindles, we propose MuFF-E, a U-Net architecture neural network that integrates multi-features from the time domain, time-frequency domain, and metadata for spindle detection. Additionally, we introduce a model ensemble approach applied to MuFF models, termed MuFF-E, that imitates the formation of group consensus in the MODA dataset. Using the high-quality spindle dataset MODA, MuFF-E achieves an F1 score of 0.84 at an overlapping threshold of 0.2 and an IoU score of 0.67, numerically surpassing the state-of-the-art approaches. Besides, MuFF-E still performs well on larger overlapping thresholds, reaching an F1 score of 0.82 at an overlapping threshold of 0.4.