Automated Classification of Heart Disease using Deep Learning

Abstract

Heart murmurs are irregular heartbeat patterns that may be a sign of a serious cardiac problem. These conditions can only be diagnosed by qualified professionals using a stethoscope. Given that the patient-to-doctor ratio is low in developing countries, there is a requirement for an automated system that can classify heart sounds and analyses the phonocardiogram (PCG) recording in real-time. A critical step in the diagnosis of cardiovascular disorders (CVDs) is the computerized classification of cardiac sounds. Particularly when applied to heart sound spectrograms, Deep learning techniques have been quite successful in automating the detection of CVDs. Such a system might be created using a variety of available techniques, transfer learning is one of such utilities. A modern machine learning technique that has gained popularity due to its quick training time and improved accuracy. The lack of sufficient data, effective models and ineffective training pose certain limitations. This paper aims at developing a lightweight, fast and reliable alternative for heart sound classification. The data is cleaned, processed, and transformed into an image using spectrogram signal representation. Based on the obtained experimental outcomes of this research paper, a transfer learning pipeline could make heart sound classification and CVD detection easier while requiring less training time and resources.