|TITLE||Sleep Stage Scoring based on an Interpretable Machine Learning Algorithm|
This presentation discusses an unsupervised approach for sleep dynamics exploration and automatic annotation by combining modern signal processing tools and probability models. In our algorithm, we applied a nonlinear-type time frequency analysis tool to extract the frequency-domain features from a pair of brain waves and use the multiview diffusion maps to fuse them. Based on the fused feature sequence and the expert-determined sleep stages, we construct a predict model by the kernel support vector machine. The prediction performance is validated on two databases. The first one is the publicly available benchmark database PhysioNet. The second database comes from six sleep centers in Taiwan. Apart from the purpose of autoscoring, our algorithm has also been applied to assess the consistency of experts’ scoring rules. We will also report this part of results. This talk includes the joint works [1-3] with Yu-Lun Lo, John Malik, Yuan-Chung Sheu, and Hau-Tieng Wu.
 G. R. Liu, Y. L. Lo, Y. C. Sheu, and H. T. Wu Explore intrinsic geometry of sleep dynamics and predict sleep stage by unsupervised learning techniques. (To be appeared in the Springer book Harmonic Analysis and Applications)
 G. R. Liu, Y. L. Lo, J. Malik, Y. C. Sheu, and H. T. Wu Diffuse to fuse EEG spectra -- intrinsic geometry of sleep dynamics for classification. Biomedical Signal Processing and Control55 (2020)
 Gi-Ren Liu, Caroline Lustenberger, Yu-Lun Lo, Wen-Te Liu, Yuan-Chung Sheu and Hau-Tieng Wu Save Muscle Information – Unfiltered EEG Signal Helps Distinguish Sleep Stages Sensors 20 (2020)