Time-varying statistical complexity measures with application to EEG analysis and segmentation

The recently proposed instantaneous statistical dimension is compared to new conditional Renyi entropies. The motivation for introducing these time-varying complexity measures is the analysis of electroencephalograms for which nonstationarity is an inherent property. Experimental data from babies are analyzed using the proposed complexity measures. The instantaneous statistical dimension computation is based on an adaptive autocorrelation eigenspectrum computation known as APEX together with a model selection rule. The conditional Renyi entropies are based on time-frequency representation of the signal. It is shown that: 1) the three time-varying complexity measures account for a component counting property, 2) the instantaneous statistical dimension is the most robust to Gaussian white noise.