Comparison of sample entropy and AR-models for heart sound-based detection of coronary artery disease

The first reported observations of rare diastolic murmurs in patients with coronary artery disease (CAD) date back to the late sixties. Subsequently several studies have the examined signal processing methods for identification of the weak murmurs. One such method is autoregressive (AR) models. A recent study showed that CAD changes the entropy of the diastolic sound. The aim of the current study is to analyze the relationship between features from an AR-model and features describing signal entropy. Sample entropy and the poles of AR models were calculated from diastolic intervals in heart sound recordings randomly selected from a database of stethoscope recordings of good quality. In total 100 recordings were analyzed (50 patients with two recordings from each). The recordings were band pass filtered with a 8 order Chebyshev filter with pass band edge frequency at 50 Hz and 500 Hz. The result shows that both measures equally separates the CAD patients from non-CAD patients, but the measures are strongly correlated.