Related Experiment Videos
Seizure detection using a self-organizing neural network: validation and comparison with other detection strategies
1Department of Neurology, University of California, Davis, 95616, USA. ajgabor@ucdavis.edu
Objective:
A previously described seizure detection algorithm (CNET) (Gabor, A.J., Leach, R.R. and Dowla, F.U. Automated seizure detection using a self-organizing neural network. Electroenceph. clin. Neurophysiol., 1996, 99: 257-266) was validated with 200 records from 65 patients (4553.8 h of recording) containing 181 seizures.
Design And Methods:
Performance of the algorithm was manifest by its sensitivity ((seizures detected/total seizures) x 100) and selectivity (false-positive errors/Hr-FPH). Comparisons with the Monitor detection algorithm (Version 8.0c, Stellate Systems) and audio-transformation (Oxford Medilog) were performed.
Results:
CNET detected 92.8% of the seizures and had a mean FPH of 1.35 +/- 1.35. Monitor detected 74.4% of the seizures and had a mean FPH of 3.02 +/- 2.78. Audio-transformation detected all but 3 (98.3%) of the seizures. Selectivity for this detection strategy was not defined.
Conclusions:
This study not only validates the CNET algorithm, but also the notion that seizures have frequency-amplitude features that are localized in signal space and can be selectively identified as being distinct from other types of EEG patterns. The ear is a specialized frequency-amplitude detector and when the signal is transformed into audio frequency range (audio-transformation), seizures can be detected with better sensitivity as compared to the other strategies examined.