The effect of using Elman-type feedback SOM for sleep stage diagnosis

In psychiatry, the sleep stage is one of the most important evidence for diagnosing mental disease. However, when doctor diagnose the sleep stage, much labor and skill are required, and a quantitative and objective method is required for more accurate diagnosis. For these reasons, an automatic diagnosis system must be developed. In this paper, we propose an automatic sleep stage diagnosis method by using Self-Organizing Maps (SOM). Neighborhood learning of SOM makes input data which has similar feature output closely. This function is effective to understandable classifying of complex input data automatically. We didn´t only applied SOM to EEG of normal subjects but also applied to EEG of subjects suffer from disease. The spectrum of characteristic waves in EEG of disease subjects is often different from it of normal subjects. So, it is difficult to classify EEG state of disease subjects with the rule for normal subjects. On the other hand, SOM classifies the EEG state with considering the features which data include. So, even the EEG of disease subjects is able to be classified automatically. In our experiment, first, the features included in EEG were extracted and learned by the Elman-type feedback SOM on competitive layer. The spectrum data were inputted to the Elman-type feedback SOM and data were classified on competitive layer. Next, the data were diagnosed by doctor and the sleep stages were labeled. The data of stage wake were inputted to the learned Elman-type feedback SOM, and the neuron which fires mostly was decided. This neuron is called wake winner neuron (WWN). Finally, data for testing were inputted to the learned Elman-type feedback SOM and corresponding sleep stage was diagnosed by the distance from WWN to Best Matching Unit. Experimental results were compared with the results of normal SOM and R&K method, and it was revealed that the Elman-type feedback SOM achieve the highest correct rate of sleep stage diagnosis.