The authors of this paper propose an interesting method for automatic sleep stage classification based on the time-frequency image (TFI) of electroencephalogram (EEG) signals. The method is useful for the diagnosis of sleep disorders and the selection of treatment options.
The main drawback of this study is that it involves only eight subjects. This is not sufficient for making statistical inferences; thus, it should be treated as a pilot study. Although statistical validation will be needed, the TFI obtained from the time-frequency representation (TFR) does seem to be a useful method for abstracting sleep stages from nonstationary EEG signals. The lowest accuracy in the confusion matrix (table 2) is 75.48 percent, but that value must be statistically adjusted for only eight observations, which converts the outcome to a statistical accuracy of less than 70 percent. The communicated findings should be validated on an adequate sample. Approximately 400 observations are needed in each category to achieve 95 percent statistical accuracy.
This paper presents a specialized and experimental study, so the target audience is sleep disorder researchers.