Low complexity tracking for long term monitoring of heart sounds

This paper describes a novel framework for tracking heart sounds. For a given heart sound component (e.g. S1) candidate features (e.g. peaks) are mapped to states and assigned appropriate local scores. A low complexity dynamic programming based algorithm is developed to select the largest most consistent feature from the candidate features. This algorithm is also capable of detecting multiple simultaneous features for improved noise immunity. The proposed algorithm has linear complexity with time and cubic complexity with the number of features to be tracked. Data from six swine balloon-occlusion studies are used for preliminary testing of the algorithm. The S1 amplitude and heart sound ejection time (HSET) are found to be statistically significantly correlated with changes in systolic performance due to ischemia induction.