Heart Condition Monitoring Using Phonocardiogram and Murmurs Analysis by CNN

Heart failure is the leading cause of mortality worldwide. Cardiac auscultation is one of the oldest diagnostic methods to explore the condition of the heart. Mechanical activities of heart initiated by biopotentials generated by the Sino-Atrium (SA) node of the heart causes non-stationery and non-deterministic signals commonly known as Heart Sound Signal (HSS) and the graphical representation of HSS is termed as Phonocardiogram (PCG) while recorded by an electronic stethoscope. When assessing a variety of cardiac anomalies, like valvular heart disease, irregular heartbeat, cardiac decompensation, and anatomical flaws in heart, PCG can offer extremely useful diagnostic data. The first and second (S1, S2) heart sounds, which are often the two loudest heart sounds, make up PCG. The PCG signal can contain murmurs, the third and fourth cardiac sounds (S3), and other unusual noises including clicks, snaps, and rubs in both pathological and non-pathological circumstances. Analysis of PCG thus becomes very effective in diagnosing the cardiovascular disorder, if any, in a non-invasive manner. The basic aim of the thesis is to develop an Automatic Computerized Auscultation (ACA) for effective and efficient status monitoring of the condition of the heart. ACA is a non-invasive, cost effective, fast and accurate method that may assist the physicians in early detection of cardiac illness truncating the death rate due to cardio vascular diseases by providing diagnostic and prognostic information. The proposed CNN method deals with classification of heart murmurs as innocent (normal) and pathological (abnormal) following the same stages of classification of heart sound. The best combination for murmur classification provided a classification AUC of 9345551.