Detection of rapid eye movement with rapidly adapting neuronal fuzzy systems in imprecise REM syntax

Abstract

Both living beings and artificial neuronal networks are capable of 'learning' and behavioural adaptation. But also the fuzzy program designed to detect rapid eye movements (REM) during sleep and described here, can be provided with a self-learning option that provides important information about REM sleep. The algorithm computes REM on the basis of horizontal and vertical EOG. EEG, EMG and actiography signals are employed to optimize the method and eliminate artefacts. In a second step, the fuzzy system learns to detect REM with the aid of a sample data set and a minimal set of syntax rules. From sample data and the actions and reactions of visual scorers, the program extracts additional rules and information, which are then used to build a complete fuzzy structure. Thereafter, the REM detection program optimizes the fuzzy logic structure, independently of visual monitoring, on its own. A direct comparison of the results of the algorithm in a 10-night analysis with those of two experienced visual scorers revealed a better than 95% agreement. Re-analysis with the algorithm showed a 100% concurrence. Complete visual measurement of the eye movements occurring in a single night requires several hours; this compares with only 15 minutes required by the algorithm.