Heart Sounds Classification Using Hybrid CNN Architecture

Abstract

In this paper, we propose a hybrid model for diagnosing heart conditions by analyzing heart sounds and signals. The Hybrid CNN (Convolutional Neural Network) model is trained to classify distinguishable pathological heart sounds into three classes; normal, murmur, and extrasystole. Scalogram images of heart sounds were obtained by applying wavelet transform to heart sound signals. Images are inputs for Resnet50 and Resnet101 CNN models. The feature vectors of these architectures in the fc1000 layer are combined. Relief feature selection algorithm was applied to the obtained feature vector, and then the classification was performed with the support vector machine algorithm. Training the proposed model resulted in accuracy of 92.75%, thus, making it the best performing model in comparison to other models in this paper.