P. Betterton, M. D'Alessandro, G. Vachtsevanos, and B. Litt (USA)
application-biomedicine, signal processing, sleep staging, intracranial EEG, epileptic seizures, consciousness
Automated sleep staging plays an important role in evaluation and treatment of neurological disorders impacted by state of awareness. Recent work in seizure prediction demonstrates the need for on-line calculation of state of awareness from the intracranial electroencephalogram (IEEG) during measurement of seizure precursors. The aim of this work is to develop such a method for use as input to seizure prediction algorithms. To establish a reference point, data segments from a single patient were examined, and it was found that the power spectrum of the delta band (0 4 Hz) in the left temporal depth electrode contact closest to the occipital lobe was qualitatively "flatter" during wakefulness than during sleep. An algorithm was developed to quantify this observation and automatically classify signals as "awake" or "asleep," for comparison to behavioral classification of sleep state from video recordings. This algorithm was tested on 753 hours of data from a total of five patients, with average accuracy of 72.1%. One-minute segments were used; the first 1/3 of each patient's data was used for training, and the second 2/3 was used to assess accuracy. Subjective classification of state of consciousness, misalignment of visual classification and IEEG data, and attachment of disjoint data segments may have degraded accuracy. However, further refinement of these methods, calibrated to classical scoring of sleep stage on scalp EEG, could provide sufficient accuracy for use in real-time seizure prediction and intervention devices.
Important Links:
Go Back
