Complexity and Similarity Analysis of Heart Sound for Cardiovascular Disease Detection

Heart auscultation signal has served as an important primary symptom to identify pathological condition of cardiovascular system for a long time. However, clinical capabilities to properly analyze Heart Sound (HS) have regressed recently. Thus, a new approach for automated Cardiovascular Disease (CVD) diagnosis based on complexity and similarity analysis of HS is presented. The relevant technologies namely, Musical Instrument Digital Interface (MIDI) coding and N-gram encoding are utilized for feature extraction and pattern encoding of HS. Lempel and Ziv (LZ) complexity and Super-symmetric Comparison Distance (SCD) similarity measure are used to measure the complexity and similarity between two HSs individually. To identify specific problems and important attributes required for HS analysis using the proposed approach, various simulated HS signals are generated and critically tested. The effect of change in amplitude, heart rate and length of HS record are identified on the diagnosis results and their thresholds or tolerances are predefined. Through such explored results, HS records of patients with different physiological characteristics which are practically incomparable are made comparable. The final testing results of such a novel approach with an average diagnostic accuracy above 70% in identifying some CVDs through comparing similarities of pathological conditions of HS with ones in benchmark database (DB) proves its feasibility and availability.