Automatic detection of cyclic alternating pattern

The cyclic alternating pattern is a microstructure phasic event, present in the non-rapid eye movement sleep, which has been associated with multiple pathologies, and is a marker of sleep instability that is detected using the electroencephalogram. However, this technique produces a large quantity of information during a full night test, making the task of manually scoring all the cyclic alternating pattern cycles unpractical, with a high probability of miss classification. Therefore, the aim of this work is to develop and test multiple algorithms capable of automatically detecting the cyclic alternating pattern. The employed method first analyses the electroencephalogram signal to extract features that are used as inputs to a classifier that detects the activation (A phase) and quiescent (B phase) phases of this pattern. The output of the classifier was then applied to a finite state machine implementing the cyclic alternating pattern classification. A systematic review was performed to determine the features and classifiers that could be more relevant. Nine classifiers were tested using features selected by a sequential feature selection algorithm and features produced by principal component analysis. The best performance was achieved using a feed-forward neural network, producing, respectively, an average accuracy, sensitivity, specificity and area under the curve of 79, 76, 80% and 0.77 in the A and B phases classification. The cyclic alternating pattern detection accuracy, using the finite state machine, was of 79%.