Joint time and time-frequency optimal detection of K-complexes in sleep EEG.

Automated detection of waveforms such as delta and K-complex in the EEG is an important component of sleep stage monitoring. The K-complex is a key feature that contributes to sleep stages assessment. However, its automated detection is still difficult due to the stochastic nature of the EEG. In this paper, we propose a detection structure which can be interpreted as joint linear filtering operations in time and time-frequency domains. We also introduce a method of obtaining the optimum detector from training data, and we show that the resulting receiver offers better performances than the one obtained via the Fisher criterion maximization. The efficiency of this approach for K-complexes detector design is explored. It results from this study that the obtained receiver is potentially the best one which can be found in the literature. Finally, it is emphasized that this methodology can be advantageously used to solve many other detection problems.

Sleep stage scoring using the neural network model: comparison between visual and automatic analysis in normal subjects and patients.

In this paper, we compare and analyze the results from automatic analysis and visual scoring of nocturnal sleep recordings. The validation is based on a sleep recording set of 60 subjects (33 males and 27 females), consisting of three groups: 20 normal controls subjects, 20 depressed patients and 20 insomniac patients treated with a benzodiazepine. The inter-expert variability estimated from these 60 recordings (61,949 epochs) indicated an average agreement rate of 87.5% between two experts on the basis of 30-second epochs. The automatic scoring system, compared in the same way with one expert, achieved an average agreement rate of 82.3%, without expert supervision. By adding expert supervision for ambiguous and unknown epochs, detected by computation of an uncertainty index and unknown rejection, the automatic/expert agreement grew from 82.3% to 90%, with supervision over only 20% of the night. Bearing in mind the composition and the size of the test sample, the automated sleep staging system achieved a satisfactory performance level and may be considered a useful alternative to visual sleep stage scoring for large-scale investigations of human sleep.

Statistical Decision Tree: a tool for studying pharmaco-EEG effects of CNS-active drugs.

Quantitative pharmaco-EEG has become a useful technique for determining pharmacodynamic parameters after CNS-active drug administration. Nevertheless, one of the most important problems faced by practitioners of pharmaco-EEG is the difficulty in evaluating drug-specific effects. In this article, a methodology for comparing two time sequences of pharmacodynamic measurements, the Statistical Decision Tree (SDT), is proposed. This methodology, based on one- and multi-dimensional Wilcoxon signed-rank tests on EEG variables, takes into account vigilance fluctuations and placebo effects in order to pick out effects specifically due to the drug.

Neural network model: application to automatic analysis of human sleep.

We describe an approach to automatic all-night sleep analysis based on neural network models and simulated on a digital computer. First, automatic sleep stage scoring was performed using a multilayer feedforward network. Second, supervision of the automatic decision was achieved using ambiguity rejection and artifact rejection. Then, numerical analysis of sleep was carried out using all-night spectral analysis for the background activity of the EEG and sleep pattern detectors for the transient activity. Computerized analysis of sleep recordings may be considered as an essential tool to describe the sleep process and to reflect the dynamical organization of human sleep.