Neural Networks for Pattern Recognition
Neural Networks for Pattern Recognition
Hi-index | 0.00 |
Recent evidence supports the positive effects of external intervention during specific sleep stages (e.g. enhanced memory consolidation and depression relief). To enable timely intervention, online automated sleep staging is required and preferably with short latency. In this paper, we propose an approach to achieve this based on the analysis of spectral features of a single electroencephalogram (EEG) channel and the use of Gaussian Mixture Models. We compare among several choices for the EEG signal location, the type of spectral features, and the duration of the signal segment (epoch) that is required to automatically identify the sleep stage. The performance metric used for comparison purposes is the kappa statistic, which measures the agreement between the automatic and manual sleep staging. The performance is higher when central EEG locations (C3, C4), longer epochs, and the power in five frequency bands are used. However, good results (kappa=0.6) can also be obtained for an epoch duration of 12 seconds.