INTRINSIC MODE DECOMPOSITION OF PHYSIOLOGICAL SIGNALS FOR FEATURE EXTRACTION

摘要

This paper presents a study for extracting features from physiological signals using intrinsic mode decomposition. The complex, nonlinear and non-stationary biomedical signals are first decomposed into intrinsic mode functions (IMF). Next each IMF is subjected to morphological signal processing (MSP) for extracting features, namely, pattern spectrum entropy (PSEn), that characterize the shape-size complexity of the component signals. These along with other features like energy (E) and sample entropy (SampEn) are extracted from the individual IMF as well as the cumulative sums of IMF for characterizing the signals. The procedure is illustrated using heart sound signals digitally recorded during cardiac auscultation representing different cardiac conditions. The study examines the effectiveness of IMF based features in the assessment of cardiac state.