Heart murmur recognition and segmentation by complexity signatures.

Heart sound analysis has been a topic of investigation for several years. Since heart sounds directly encode the mechanical activity of the heart, they enable the assessment and follow-up of several types of heart disorders in pre-symptomatic states. Murmurs are the most common abnormality signature in many heart disorders. This paper introduces an algorithm for heart murmur identification. In the presence of murmurs, heart sounds exhibit chaotic behavior. In the proposed method this is assessed based upon the nonlinear dynamics of the signal. In order to segment murmurs from other heart sound components, the signal is transformed into a phase space that is later reconstructed using the embedded matrix. Based on the phase space, the complexity and the strength of the signal are computed. These features are the basis for sound component boundary location. The method has been tested with a database of heart sounds that include diverse heart lesions and heart murmurs. The algorithms achieved 91.09% sensitivity and 95.25% specificity.