Integrated system for analysis and automatic classification of sleep EEG

In this paper we describe a system for automatic all night sleep analysis, based on neural networks. In what concerns sleep staging, the system consists of a three-step analysis. The first step is the recognition of elementary patterns in EEG (delta, alpha, spindle and K complex waves), EOG and EMG, and spectral analysis for background activity. The second step is the determination of sleep stages based on these parameters. Automatic sleep scoring was performed using a multilayer feedforward network. The last step, is the supervision of the automatic decision using ambiguity rejection and coherence analysis. Finally, this work presents an automated system for micro arousals detection based on k-means method. The system was validated with a data set including 20 hours of signals. Overall agreement between the computer and human judges indicates considerable reliability of the system