Automated detection of sleep EEG slow waves based on matching pursuit using a restricted dictionary

In this paper, an original method to detect sleep slow waves (SSW) in electroencephalogram (EEG) recordings is proposed. This method takes advantage of a Matching Pursuit algorithm using a dictionary reduced to Gabor functions reproducing the main targeted waveform characteristics. By describing the EEG signals in terms of SSW properties, the corresponding algorithm is able to identify waveforms based on the largest matching coefficients. The implemented algorithm was tested on a database of whole night sleep EEG recordings collected in 9 young healthy subjects where SSW have been visually scored by an expert. Besides being fully automated and much faster than visual scoring analysis, the results obtained to the proposed method were in excellent agreement with the expert with 98% of correct detections and a 77% concordance in event time position and duration. These results were superior from those of the classical method both in terms of sensibility and precision.