A CNN model with statistical correction rules for automatic sleep stage scoring

In this study, a convolutional neural network (CNN) model with statistical correction rules is developed for the automatic sleep stage scoring. Sleep is a dynamic process which consisted of several sleep stages from light sleep to deep sleep. The convolutional neural network is designed by 6 layers, using two convolution kernels of different sizes to extract the time domain and frequency domain features separately. The statistical correction rules are extracted regarding to the dynamic transition between different sleep stages. The developed method was tested on the large amount of sleep data for the inspection of consisted sleep stages during ones overnight sleep. The prediction results by CNN model were corrected by the statistical correction rules. Totally, the sleep recording of 20 subjects were evaluated. The obtained results showed that the combination of CNN and correction rules achieved rather good and reasonable performance for sleep stage scoring.