ResHNet: Spectrograms Based Efficient Heart Sounds Classification Using Stacked Residual Networks

ResHNet: Spectrograms Based Efficient Heart Sounds Classification Using Stacked Residual Networks
Title:
ResHNet: Spectrograms Based Efficient Heart Sounds Classification Using Stacked Residual Networks
Other Titles:
2019 IEEE EMBS International Conference on Biomedical & Health Informatics (BHI)
Publication Date:
12 September 2019
Citation:
A. Balamurugan, S. G. Teo, J. Yang, Z. Peng, Y. Xulei and Z. Zeng, "ResHNet: Spectrograms Based Efficient Heart Sounds Classification Using Stacked Residual Networks," 2019 IEEE EMBS International Conference on Biomedical & Health Informatics (BHI), Chicago, IL, USA, 2019, pp. 1-4. doi: 10.1109/BHI.2019.8834578
Abstract:
Cardiovascular disease (CVD) is one of the major contributors of global mortality rate as it accounts foralmost 31% of the worldwide deaths. As per World HealthOrganization (WHO), CVD continues to be the number onecause of death in the world. In some parts of the world,access to expert doctors and diagnosis are difficult. Thus anefficient and quick diagnosis of heart disease method is needed especially for low-income and middle-income countries whereMagnetic Resonance Imaging (MRI) and Ultrasound becomesa constraint in terms of the resources to save human life. Withthe tremendous technology advancement in the medical field,deep learning has gained more attention to automate most of the initial diagnosis of diseases. This fosters continuous researchin adopting deep learning methods for automatic classificationof heart sounds to identify any abnormalities. In this work, weaim to investigate the efficiency of introducing residual modulesin heart sounds classification using a deep neural network.This approach involves the following steps: (i) Generation ofSpectrograms for every 1D audio signal using Spectrogramgenerator module (ii) Training of residual network basedclassifier for identifying normal and abnormal heart soundsbased on the Spectrograms. The standard dataset given for 2016PhysioNet/CinC Challenge has been used here for validating ourresidual network. This method achieved around 97% accuracy on the independent hidden test set performing best withoutincorporating any segmentation or additional MFCC featuresof the audio signals and just the learned features from theimage-based representations. Various baseline results of otherdeep learning based approaches have also been considered forevaluating the robustness of this framework.
License type:
PublisherCopyrights
Funding Info:
Description:
(C) 2019 IEEE.
ISSN:
2641-3604
Files uploaded:
File Size Format Action
There are no attached files.