Combining time frequency representation and parametric analysis for the enhancement of transients in sleep EEG signal

The study of the electroencephalographic (EEG) signal contributes to sleep analysis. In the microstructure of the sleep EEG signal, transient patterns are characterized by their frequency content and their time duration. The Time-Frequency Representations (TFR) take into account these time frequency characteristics but the lower energy transient signals are masked by higher energy ones. In order to overcome this problem, we introduced a method to decompose signals into a summation of oscillatory components with time varying frequency, amplitude and phase characteristics, based on the Tufts-Kumaresan algorithm. The resulting parameters, i.e. amplitude and frequency, are then used to train joint linear filtering operations of the TFR in the time frequency domain. The aim of this work is to improve the classical TFR analysis for detecting frequency transients over short time duration, to reduce the amount of useful information to few parameters that help medical doctors to analyze the microstructure of sleep by correlating information estimated from different signals.