Sleep disorder classification and epileptic seizure prediction by neural networks and particle filters

Obstructive sleep apnea (OSA) and narcolepsy are two common sleep disorders. Based on data of EEC and pupil size, classification between subjects with sleep disorders of OSA or narcolepsy is discussed. Modified adaptive resonance theory 2 (ABT2) neural networks in series are used to group different projects with or without sleep disorders. Specifically, the system is apply to classify two groups, one group including subjects with OSA and healthy controls and the other group having subjects with narcolepsy and healthy controls. This system can effectively classify different subjects in each group.; Particle filters are widely used in the tracking problem, for example, a ship is running in the sea and only measurement is angles from a fixed location. Two problems in traditional particle filters are sampling degeneracy and variance increasing with time. The backpropagation neural networks are combined with the particle filters to improve its performance in both issues. The modified particle filters can track a running ship accurately with the small number of particles from our simulation results.; Epilepsy is a kind of brain disorder and its prediction is meaningful. Normally EEG can be used to record brain activities including epilepsy. Wavelet transform is applied to epileptic EEG data and energy of frequency band of 5--12Hz can be obtained. The procedure of epileptic occurrence is able to be described by two equations with random variables. The hidden variable in the process has ability to predict seizures based on its feature. The method of particle filters combined with neural networks is applied to figure out the hidden variable and further to predict seizure onsets. Six patients' data are used to test the performance of this algorithm and about 88% of them can be identified in advance from 6 minutes to 85 minutes.