Smart cardiac health management in IoT through heart sound signal analytics and robust noise filtering

Heart sound or Phonocardiogram (PCG) is one of the fundamental markers of cardiac health. With the promise of Internet of Things (IoT) and advent of wearable-captured PCG signal, automated analysis of PCG signal plays vital role in remote and mobile cardiac health management. One of the challenging problems of completely automated analytics or clinical prediction method is the frequent presence of corruption in PCG signals from multiple noise sources like motion artifacts, ambient noise and majority of the automated computational methods fail to ensure sufficient clinical utility due to their inability to eliminate the corruption in PCG signals. In this paper, we propose a personal cardiac management application and ecosystem that helps patient's on-demand cardiac health assessment. It applies novel noise filtering method on PCG signals through robust feature space optimization feature selection with audio signal processing primitives to ensure effective clinical analytics and identifies cardiac abnormality condition. The proposed scheme is a precise blend of signal processing, information theoretic and machine learning techniques. We depict more than 85% accuracy and high specificity of identifying noisy PCG signals while experimenting over annotated PCG datasets from large publicly available MIT-Physionet database. We further show that robust noise filtering of PCG signals has the capability to significantly improve the clinical utility of detecting cardiac abnormal condition by more than 40% over state-of-the-art solution.