Poster | 6th Internet World Congress for Biomedical Sciences |
Jose Ramon Sepulveda Sanchis(1), Juan Guerrero(2), EMILIO SORIA OLIVAS(3), Antonio Benetó(4), Enriqueta Gómez(5), Antonio J. Serrano López(6)
(1)universidad de valencia - valencia. Spain
(2)Departamento de Electrónica. Universidad de Valencia - Burjassot. Spain
(3)DPTO INGENIERÍA ELECTRONICA. FACULTAD DE FISICAS - BURJASSOT/VALENCIA. Spain
(4)(5)Hospital la Fé - Valencia. Spain
(6)Dpto. Electrónica. Universidad de Valencia - Burjasot. Spain
[Neuroscience] |
[Health Informatics] |
In the processing of biosignals most of the problems arise due to incomplete data or a misunderstanding of data, yielding unreliable results and a subjective evaluation. In the type of data that we are concerned with (EEG), the electrical currents generated in the cortex are too small and may be modelled only poorly. Furthermore, the large size of scalp electrodes and the effects of muscle and instrument noise make EEG analysis a complex task.
The aim of this work is to get an automatic method for the detection of arousal by using EEG. Arousal is a phenomenon that normally lasts for a few seconds without causing awakenings or changes in the sleep stage. Arousals can be identified visually in the polysomnographic recordings, and are characterized by abrupt changes in the electroencephalographic(EEG) frequency, suggestive of an awake state and /or brief increases in the electromyographic (EMG) amplitude (1). Usually, the detection of such phenomena is performed manually by an expert; this method is cumbersome and highly time consuming, which suggests automatization.
The important fact is that the arousals result in fragmented, rather than shortened, sleep. As with shortened sleep, it has been demonstrated that sleep fragmentation can lead to increased sleepiness during the day. Equally important, this tool for the detection of arousals suggests applications to neurophysiological studies regarding the normal level of arousal in a healthy patient, as well as concerning the co-occurrence of arousal patterns with sleep disorders or other neurophisiological pathologies (e.g. apnoea, narcolepsy,catalepsy,…).
A method for the automatic detection of arousals during sleep, is described using neural networks. Due to the non linearity of the phenomenon, neural networks would be a suitable solution.A multilayer perceptron with the backpropagation algorithm is used. For the training, preprocessed data is used to train the neural network, with arousal and nonarousal cases.
The data used were supplied from the Hospital La Fe of Valencia Database, recorded during a night (7 hours ) of sleep investigation, during which standard polysomnographic signals were recorded continuously. The standand Rechschaffen and Kales bipolar submental EMG and referential electrooculograms(EOGs) are used, as is the referential EEG.
Mean frequency, Mobility, and Zero Cross are used as potential indicators of arousal (3), each of them in three channels. And also the dispersion in the EOG channel. These characteristics are used as inputs for the neural network.
- Mean Frequency.
This index is the first moment ( effectively, the "center of gravity")of the frequency spectrum, and it gives an estimate of the central or mean EEG frequency.
- Mobility.
Giving a measure of the standard deviation of the slope with reference to the standard deviation of the amplitude. It is expressed as a ratio per time unit, this index is an estimate of EEG frequency.
- Zero Crossing.
This index indicates the number of times the EEG waveform crossed the zero potential level per unit time. This index is related to mean EEG frequency.
Data are preprocessed subtracting the mean value and dividing by the standard deviation. In order to give a relation within the arousal every epoch is subtracted from the epoch 3 seconds before, it due to the minimum duration of arousal (1).
A set of recordings with 362 scored arousals from a 2500 epoch recording is used to train the neural network.
The sensitivity of the automatic system is 84.2% and the selectivity 66.4%.
Due to the non-linearity of the problem linear dicriminants were of no use and artificial neural networks with the apropiate pre and post processing are well suited for the detection of arousal during sleep in EEG.
[Neuroscience] |
[Health Informatics] |