Deep Learning Based Phonocardiogram Signals Analysis for Cardiovascular Abnormalities Detection

Abstract

Cardiovascular diseases are among the vital causes of mortality worldwide which need early detection with the use of auscultation examination. Heart diseases could be diagnosed in a convenient way of heartbeat sound analysis. Manual auscultation is time-consuming and problematic to differentiate heart sounds related to different kinds of heart abnormalities. Also, as it requires an expert in the field, it becomes costly and quite prone to human error. Due to all these issues, there is a high demand for an automatic diagnostic system, an alternative way to human examination. This research focuses on developing Artificial Intelligence (AI) based system of the latest computational algorithms for detecting heart abnormalities from heart sounds. Heart sounds are classified as to be normal or abnormal from Phonocardiogram (PCG) signals. One of the recent techniques introduced in deep learning algorithms for audio classification is 1-Dimensional Convolutional Neural Network (1D-CNN). This research work includes a 1D-CNN as a classification algorithm. A widely used publicly available dataset of heart sounds from PhysioNet/CinC (2016) challenge is utilized. The method acquires accuracy, sensitivity, specificity, F1 score, and precision of 95.45%, 97.44%, 93.6%, 95.45%, and 95.54% respectively. The proposed approach uses a less-complicated customized 1D-CNN algorithm, outshining most of the previous competitive methods by securing high performance that makes it appropriate for diagnosing heart diseases from PCG data.