Automatic spindle detector for infant data

Ana Jeroncic

University of Split - Medical School, Departement of Neuroscience, Split, Croatia

Sleep spindles are transient EEG events which have been used as a tool in automatic sleep stage scoring. In addition, spindles have been used as potential marker of sleep-related memory consolidation and of different neuropatological conditions. The development of the robust and precise automatic spindle detector is here reported. Algorithm is developed on the training dataset of visually detected spindles from all-night EEG records of 2 and 9 month old infants (N=10). One hour of continuous EEG record, C3A2-derivation, was used in each case. Since descriptor values for the detection process were calculated from the statistical properties of the EEG record tested, we have applied detector to different EEG derivation and on the adult data. It was shown to be robust to the source of data. When rigorously tested on continuous EEG records, with either very low or high frequency of visually detected spindles, detector has shown on average 89% of sensitivity and 99% of specificity, with positive prediction value of 60%. Although primarily developed for the infant data, a wider specter of application and the precision makes the detector a promising general tool for spindle detection.