Multi-class Heart Sounds Classification Using 2D-Convolutional Neural Network
Megha Banerjee, Sudhan Majhi
Abstract
Heart disease is a major concern. To prevent this, it is important to detect cardiovascular diseases at the early stage. Early discovery of heart infections and constant treatment can lessen the death rate. However, the accurate and effective detection method of heart diseases is necessary to uncover this deadly threat at a very early stage, even without the presence of a medical professional. This paper studies the use of 2D-convolutional neural network to classify heart sounds into normal and abnormal categories. The paper reports a classification of five designated categories of heart sounds such as artifact, extra heart sound, extra systole, murmur, and normal. For the betterment of the accuracy, we have reduced the number of convolutional neural network layers with a softmax layer at the top. Each convolutional layer is followed by a max pooling and a dropout layer which finally leads to a global average pooling layer. The proposed method achieves an accuracy of 83%.