ON THE USE OF NEURAL-NETWORK TECHNIQUES TO ANALYZE SLEEP EEG DATA - FIRST COMMUNICATION - APPLICATION OF EVOLUTIONARY AND GENETIC ALGORITHMS TO REDUCE THE FEATURE SPACE AND TO DEVELOP CLASSIFICATION RULES
R. Baumgartschmitt et al., ON THE USE OF NEURAL-NETWORK TECHNIQUES TO ANALYZE SLEEP EEG DATA - FIRST COMMUNICATION - APPLICATION OF EVOLUTIONARY AND GENETIC ALGORITHMS TO REDUCE THE FEATURE SPACE AND TO DEVELOP CLASSIFICATION RULES, Neuropsychobiology, 36(4), 1997, pp. 194-210
To automate sleep stage scoring, the system sleep analysis system to c
hallenge innovative artificial networks (SASCIA) has been developed an
d implemented. The aims of our investigation were twofold: In addition
to automatic sleep stage scoring the hypothesis was tested that the i
nformation of only 1 EEG channel (C4-A2) should be sufficient to autom
atically generate sleep profiles which are comparable with profiles ma
de by sleep experts on the basis of at least 3-channel EEG (C4-A2), EO
G and EMG, as EOG and EMG are seen as epiphenomena during sleep and th
e full information about the sleep stage should - according to our hyp
othesis - be available in the EEG, The main components of the SASCIA s
leep analysis system are designed to meet the requirements of flexible
adaptation to the interindividual differences of the sleep EEG. The c
ore of the SASCIA sleep analysis system consists of neural networks. S
upervised learning was implemented and the experts' scorings were incl
uded into the learning set and test set, The feature selections out of
a large number(118) are performed by genetic algorithms and the topol
ogies of the networks are optimized by evolutionary algorithms, Differ
ent mathematical procedures were used to evaluate and optimize the eff
iciency of the system. The profiles generated by SASCIA are in reasona
ble agreement with the sleep stages scored by experts according to RKR
, The development of the system is communicated in three parts: the fi
rst communication deals with the application of the neural network tec
hniques using evolutionary and genetic algorithms and with the selecti
on of feature space. The second communication shows the training of th
ese evolutionary optimized network techniques with multiple subjects a
nd the application of context rules! while the third communication sho
ws an improvement in the robustness by the simultaneous application of
9 different networks obtained from 9 subject types which were used in
combination with context rules.