Objectives: The analysis of cyclic alternating pattern (CAP) provides impor
tant microstructural information on arousal instability and on EEG synchron
y modulation in the sleep process. This work presents a methodology for aut
omatic classification of the micro-organization of human sleep EEG, using t
he CAP paradigm.
Methods: The classification system is composed of 3 parts: feature extracti
on, detection and classification. The feature extraction part is an EEG gen
eration model-based maximum likelihood estimator. The detector part for the
CAP phases A and B is done by a variable length template matched filter, w
hile the classification criteria part is implemented on a state machine rul
ed-based decision system.
Results and conclusions: The preliminary results of the automatic classifie
r on a group of 4 middle-aged adults are presented. The high agreement betw
een the detector and visual scoring is very promising in the achievement of
a fully automated scoring system, although a more exhaustive evaluation pr
ogram is needed. (C) 1999 Elsevier Science Ireland Ltd. All rights reserved
.