Feature-based detection of the K-complex wave in the human electroencephalogram using neural networks

Abstract

The main difficulties in reliable automated detection of the K-complex wave in EEG are its close similarity to other waves and the lack of specific characterization criteria. We present a feature-based detection approach using neural networks that provides good agreement with visual K-complex recognition: a sensitivity of 90% is obtained with about 8% false positives. The respective contribution of the features and that of the neural network is demonstrated by comparing the results to those obtained with i) raw EEG data presented to neural networks, and ii) features presented to Fisher's linear discriminant.

Automatic K-complexes detection in sleep EEG recordings using likelihood thresholds

“…For the majority of them, they were extracted from significant points of the possible K-complex. These significant points are similar to those of Bankman et al [12] and are illustrated in Fig. 3: -val_min and t_min (X) correspond to the minimal value of the pseudo K-complex.…”

“…However he achieved poor performances with detection rate ranging from 42% to 67%. Then, Bankman et al [12] proposed to use features-based neural networks detection, where 14 features were taken on significant points of the possible K-complex. Much better performance where obtained with a sensitivity of 90% for about 8% of FP.…”

“…Algorithms for classification problem [7], [8], [11] and [12] use EEG segment of fixed length containing K-complexes or not. The corresponding false positive rate is calculated as follow: …”

Joint Time and Time-Frequency Optimal Detection of K-Complexes in Sleep EEG

“…[6] In the expression introduced above, the bidimensional function R x (n, m) denotes the discrete time instantaneous autocorrelation function of the signal x(k):…”

Massively Parallel Classification of EEG Signals Using Min-Max Modular Neural Networks

“…Artificial neural networks have been applied to neurophysilogical data analysis such as classification of EEG signals [1,2,10]. However, training large networks on large sets of high-dimensional EEG data is a hard problem because no efficient algorithm is available for training large networks and a very long training time is required to achieve satisfactory learning accuracy [10].…”