Learning Image-based Representations for Heart Sound Classification

Abstract

Machine learning based heart sound classification represents an efficient technology that can help reduce the burden of manual auscultation through the automatic detection of abnormal heart sounds. In this regard, we investigate the efficacy of using the pretrained Convolutional Neural Networks (CNNs) from large-scale image data for the classification of Phonocardiogram (PCG) signals by learning deep PCG representations. First, the PCG files are segmented into chunks of equal length. Then, we extract a scalogram image from each chunk using a wavelet transformation. Next, the scalogram images are fed into either a pre-trained CNN, or the same network fine-tuned on heart sound data. Deep representations are then extracted from a fully connected layer of each network and classification is achieved by a static classifier. Alternatively, the scalogram images are fed into an end-to-end CNN formed by adapting a pre-trained network via transfer learning. Key results indicate that our deep PCG representations extracted from a fine-tuned CNN perform the strongest, 56.2 % mean accuracy, on our heart sound classification task. When compared to a baseline accuracy of 46.9 %, gained using conventional audio processing features and a support vector machine, this is a significant relative improvement of 19.8 % (p < .001 by one-tailed z-test).

Deep-Net: A Lightweight CNN-Based Speech Emotion Recognition System Using Deep Frequency Features

“…In addition, Reference [ 21 ] similarly used CNN to learn the salient features and recognized the emotions in spectrograms. Today, deep learning approaches have become a recent trend to extract the CNN features from the speech spectrum and the spectrograms [ 22 ]. Moreover, the researchers take benefits from the deep spectrum and the spectrograms by employing transfer learning techniques to trained end-to-end SER models utilizing Alex Net [ 23 ] and VGG [ 24 ].…”

Audio for Audio is Better? An Investigation on Transfer Learning Models for Heart Sound Classification

“…Moreover, to make the current machine learning-based approaches feasible in clinical practice, a large number of expert annotations are needed, which is another difficult issue for almost all biomedical engineering fields. Motivated by the success of transfer learning (TL) in computer vision [5], natural language processing [6], and speech recognition [7], TLbased methods are now proving another paradigm for extracting higher representations from heart sound without any human expert domain knowledge [8]. Nonetheless, most existing TL-based models are pre-trained on images, such as ImageNet [9] rather than on audio data.…”

Gated recurrent unit-based heart sound analysis for heart failure screening

“…HS signals are closely related to cardiovascular diseases and have been widely studied, while objects of these researches were different. For example, the identification and classification of HS components [27,28], classification of normal and other abnormal HS [29][30][31], differentiating the murmurs between physiological and pathological [32,33]. However, the previously published papers about classification of HFrEF, HFpEF and normal were few and incomplete.…”