AMNet: Introducing an Adaptive Mel-Spectrogram End-to-End Neural Network for Heart Sound Classification

Yang Tan; Zhihua Wang; Kun Qian; Zhihao Bao; Zheyu Cao; Bin Hu; Yoshiharu Yamamoto; Bjorn W. Schuller

doi:10.1109/Healthcom56612.2023.10472362

AMNet: Introducing an Adaptive Mel-Spectrogram End-to-End Neural Network for Heart Sound Classification

Yang Tan, Zhihua Wang, Kun Qian^*, Zhihao Bao, Zheyu Cao, Bin Hu^*, Yoshiharu Yamamoto, Bjorn W. Schuller

^*Corresponding author for this work

School of Medical and Technology

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

The cardiovascular diseases (CVDs) cause tremendous deaths yearly. The Mel-spectrogram is widely used as a tool to analyse the heart sound, which facilitate a cheap and efficient diagnosis of CVDs. Nevertheless, the amplitude and frequency responses of the Mel filter banks remain constant, limiting its function to frequency selection. We propose an adaptive Melspectrogram end-to-end neural network (AMNet) for a better characterisation and classification of heart sound in the work. The core of the adaptive Mel-spectrograms (AMel) lies in an adaptive Mel filter banks whose frequency characteristics remain the same as the original Mel-spectrogram (OMel) and amplitude is learnt by the backropagation algorithm. The AMNet learns the raw audio representation directly and outputs the classification results. It reaches 43.5% Unweighted Average Recall (UAR) and surpasses the model with the OMel and the baseline by 6% UAR. It is demonstrated that the AMel characterises the heart sound more effectively.

Original language	English
Title of host publication	2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	90-94
Number of pages	5
ISBN (Electronic)	9798350302301
DOIs	https://doi.org/10.1109/Healthcom56612.2023.10472362
Publication status	Published - 2023
Event	2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023 - Chongqing, China Duration: 15 Dec 2023 → 17 Dec 2023

Publication series

Name	2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023

Conference

Conference	2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023
Country/Territory	China
City	Chongqing
Period	15/12/23 → 17/12/23

Keywords

Adaptive Melspectrogram
Computer Audition
End-to-End
Heart Sounds
mHealth

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1109/Healthcom56612.2023.10472362

Cite this

Tan, Y., Wang, Z., Qian, K., Bao, Z., Cao, Z., Hu, B., Yamamoto, Y., & Schuller, B. W. (2023). AMNet: Introducing an Adaptive Mel-Spectrogram End-to-End Neural Network for Heart Sound Classification. In 2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023 (pp. 90-94). (2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/Healthcom56612.2023.10472362

Tan, Yang ; Wang, Zhihua ; Qian, Kun et al. / AMNet : Introducing an Adaptive Mel-Spectrogram End-to-End Neural Network for Heart Sound Classification. 2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023. Institute of Electrical and Electronics Engineers Inc., 2023. pp. 90-94 (2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023).

@inproceedings{5a9101ec6999422ba9e410fe9dc80dd0,

title = "AMNet: Introducing an Adaptive Mel-Spectrogram End-to-End Neural Network for Heart Sound Classification",

abstract = "The cardiovascular diseases (CVDs) cause tremendous deaths yearly. The Mel-spectrogram is widely used as a tool to analyse the heart sound, which facilitate a cheap and efficient diagnosis of CVDs. Nevertheless, the amplitude and frequency responses of the Mel filter banks remain constant, limiting its function to frequency selection. We propose an adaptive Melspectrogram end-to-end neural network (AMNet) for a better characterisation and classification of heart sound in the work. The core of the adaptive Mel-spectrograms (AMel) lies in an adaptive Mel filter banks whose frequency characteristics remain the same as the original Mel-spectrogram (OMel) and amplitude is learnt by the backropagation algorithm. The AMNet learns the raw audio representation directly and outputs the classification results. It reaches 43.5% Unweighted Average Recall (UAR) and surpasses the model with the OMel and the baseline by 6% UAR. It is demonstrated that the AMel characterises the heart sound more effectively.",

keywords = "Adaptive Melspectrogram, Computer Audition, End-to-End, Heart Sounds, mHealth",

author = "Yang Tan and Zhihua Wang and Kun Qian and Zhihao Bao and Zheyu Cao and Bin Hu and Yoshiharu Yamamoto and Schuller, {Bjorn W.}",

note = "Publisher Copyright: {\textcopyright} 2023 IEEE.; 2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023 ; Conference date: 15-12-2023 Through 17-12-2023",

year = "2023",

doi = "10.1109/Healthcom56612.2023.10472362",

language = "English",

series = "2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "90--94",

booktitle = "2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023",

address = "United States",

}

Tan, Y, Wang, Z, Qian, K, Bao, Z, Cao, Z, Hu, B, Yamamoto, Y & Schuller, BW 2023, AMNet: Introducing an Adaptive Mel-Spectrogram End-to-End Neural Network for Heart Sound Classification. in 2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023. 2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023, Institute of Electrical and Electronics Engineers Inc., pp. 90-94, 2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023, Chongqing, China, 15/12/23. https://doi.org/10.1109/Healthcom56612.2023.10472362

AMNet: Introducing an Adaptive Mel-Spectrogram End-to-End Neural Network for Heart Sound Classification. / Tan, Yang; Wang, Zhihua; Qian, Kun et al.
2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023. Institute of Electrical and Electronics Engineers Inc., 2023. p. 90-94 (2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - AMNet

T2 - 2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023

AU - Tan, Yang

AU - Wang, Zhihua

AU - Qian, Kun

AU - Bao, Zhihao

AU - Cao, Zheyu

AU - Hu, Bin

AU - Yamamoto, Yoshiharu

AU - Schuller, Bjorn W.

PY - 2023

Y1 - 2023

N2 - The cardiovascular diseases (CVDs) cause tremendous deaths yearly. The Mel-spectrogram is widely used as a tool to analyse the heart sound, which facilitate a cheap and efficient diagnosis of CVDs. Nevertheless, the amplitude and frequency responses of the Mel filter banks remain constant, limiting its function to frequency selection. We propose an adaptive Melspectrogram end-to-end neural network (AMNet) for a better characterisation and classification of heart sound in the work. The core of the adaptive Mel-spectrograms (AMel) lies in an adaptive Mel filter banks whose frequency characteristics remain the same as the original Mel-spectrogram (OMel) and amplitude is learnt by the backropagation algorithm. The AMNet learns the raw audio representation directly and outputs the classification results. It reaches 43.5% Unweighted Average Recall (UAR) and surpasses the model with the OMel and the baseline by 6% UAR. It is demonstrated that the AMel characterises the heart sound more effectively.

AB - The cardiovascular diseases (CVDs) cause tremendous deaths yearly. The Mel-spectrogram is widely used as a tool to analyse the heart sound, which facilitate a cheap and efficient diagnosis of CVDs. Nevertheless, the amplitude and frequency responses of the Mel filter banks remain constant, limiting its function to frequency selection. We propose an adaptive Melspectrogram end-to-end neural network (AMNet) for a better characterisation and classification of heart sound in the work. The core of the adaptive Mel-spectrograms (AMel) lies in an adaptive Mel filter banks whose frequency characteristics remain the same as the original Mel-spectrogram (OMel) and amplitude is learnt by the backropagation algorithm. The AMNet learns the raw audio representation directly and outputs the classification results. It reaches 43.5% Unweighted Average Recall (UAR) and surpasses the model with the OMel and the baseline by 6% UAR. It is demonstrated that the AMel characterises the heart sound more effectively.

KW - Adaptive Melspectrogram

KW - Computer Audition

KW - End-to-End

KW - Heart Sounds

KW - mHealth

UR - http://www.scopus.com/inward/record.url?scp=85190373200&partnerID=8YFLogxK

U2 - 10.1109/Healthcom56612.2023.10472362

DO - 10.1109/Healthcom56612.2023.10472362

M3 - Conference contribution

AN - SCOPUS:85190373200

T3 - 2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023

SP - 90

EP - 94

BT - 2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023

PB - Institute of Electrical and Electronics Engineers Inc.

Y2 - 15 December 2023 through 17 December 2023

ER -

Tan Y, Wang Z, Qian K, Bao Z, Cao Z, Hu B et al. AMNet: Introducing an Adaptive Mel-Spectrogram End-to-End Neural Network for Heart Sound Classification. In 2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023. Institute of Electrical and Electronics Engineers Inc. 2023. p. 90-94. (2023 IEEE International Conference on E-Health Networking, Application and Services, Healthcom 2023). doi: 10.1109/Healthcom56612.2023.10472362

AMNet: Introducing an Adaptive Mel-Spectrogram End-to-End Neural Network for Heart Sound Classification

Abstract

Publication series

Conference

Keywords

UN SDGs

Access to Document

Other files and links

Fingerprint

Cite this