Low complexity algorithm for heart sound segmentation using the variance fractal dimension

This paper presents an algorithm for S1 and S2 heart sound segmentation using variance fractal dimension. Heart sound is assumed as a non-stationary signal embedding two main sounds S1 arid S2, murmurs and eventually unusual ambient sound. The variance fractal dimension is applied to adaptively identify the boundaries of sound lobes. S1 components are detected using QRS synchronization while for S2 components a non-supervised classification approach is applied, based on temporal features of the lobes. This allows a 2-lead ECG signal to be used for the task. Some preliminary results are presented using recorded heart sounds taken a few days after valve replacement.