Abstract— This study aimed to develop an automatic
algorithm to detect the activation phases (A phases) of the
Cyclic Alternating Pattern. The sleep EEG microstructure of 4
adult, healthy subjects was scored by a sleep medicine expert.
Features were calculated from each of the six EEG bands (low
delta, high delta, theta, alpha, sigma and beta), and three
additional characteristics were computed: the Hjorth activity in
the low delta and high delta bands, and the differential
variance of the raw EEG signal. The correlation between
couples of features was analyzed to find redundancies for the
automatic analysis. The features were used to train an Artificial
Neural Network to automatically find the A phases of CAP.
The data were divided into training, validation and testing set,
and the visual scoring provided by the clinician was used as the
desired output. The statistics on the second by second
classification show an average sensitivity equal to 76%,
specificity equal to 83% and accuracy equal to 82%. The
results obtained are encouraging, since an automatic
classification of the A phases could benefit the practice in
clinics, preventing the physician from the time-consuming
activity of visually scoring the sleep microstructure over the
whole eight-hour sleep recordings. Moreover, it would provide
an objective criterion capable of overcoming the problems of
inter-scorer variability.