Shannon Entropy based on the S-Transform Spectrogram applied on the classification of heart sounds

The aim of this study is to present a complexity measure (Normalized Shannon Entropy) based on the S-Transform Spectrogram (ST-Spectrogram) plane, which can be considered as a variant of the Cohen's class. The ST-Spectrogram verifies the non-negativity condition which makes the application of the famous Shannon Entropy measure possible. A concrete application presented in this paper consists to detect pathologic heart sounds with systolic murmurs. A systolic period which contains murmurs is usually more complex than normal sound. The complexity measure applied on the ST-Spectrogram is used as feature to classify normal and pathologic heart sounds. A comparison with the classical spectrogram (STFT-Spectrogram) is performed by calculating the different Receiver Operating Curves (ROC) and the robustness against additive Gaussian noise is discussed.