面向可穿戴式的基于LSTM神經(jīng)網(wǎng)絡(luò)的智能心音異常診斷芯片

周維新; 高肇崗; 肖宛昂

doi:10.11999/JEIT230934

面向可穿戴式的基于LSTM神經(jīng)網(wǎng)絡(luò)的智能心音異常診斷芯片

doi: 10.11999/JEIT230934

周維新^{1, 2},
高肇崗³,
肖宛昂^{1, 2, ,}

1.
中國科學(xué)院半導(dǎo)體研究所北京 100083
2.
中國科學(xué)院大學(xué)集成電路學(xué)院北京 100049
3.
中國農(nóng)業(yè)大學(xué) 北京 100083

基金項目: 中國科學(xué)院先導(dǎo)科技專項培育項目(XDPB22)

詳細(xì)信息

作者簡介:
周維新：男，博士生，研究方向為智能體音診斷算法研究、可穿戴式智能體音芯片設(shè)計

高肇崗：男，碩士生，研究方向為深度學(xué)習(xí)算法研究、數(shù)字電路設(shè)計

肖宛昂：男，研究員，研究方向為智能聲音信號處理芯片、無線通信基帶芯片以及機器學(xué)習(xí)的FPGA加速

通訊作者:
肖宛昂　waxiao@semi.ac.cn

中圖分類號: TN492;TP183
計量
- 文章訪問數(shù): 480
- HTML全文瀏覽量: 334
- PDF下載量: 91
- 被引次數(shù): 0
出版歷程
- 收稿日期: 2023-08-28
- 修回日期: 2024-01-17
- 網(wǎng)絡(luò)出版日期: 2024-01-23
- 刊出日期: 2024-02-29

Intelligent Heart Sound Abnormal Diagnosis Chip Based on LSTM for Wearable Applications

ZHOU Weixin^{1, 2},
GAO Zhaogang³,
XIAO Wan'ang^{1, 2
, ,}

1.
Institute of Semiconductors, Chinese Academy of Sciences, Beijing 100083, China
2.
School of Integrated Circuits, University of Chinese Academy of Sciences, Beijing 100049, China
3.
China Agricultural University, Beijing 100083, China

Funds: The Key Research Program of the Chinese Academy of Sciences (XDPB22)

摘要

摘要: 心血管疾病是造成全球死亡人數(shù)最多的疾病之一，因此對心血管疾病的預(yù)防與提前診斷至關(guān)重要。人工聽診技術(shù)與計算機心音診斷技術(shù)無法滿足對心音長時間聽診的需求，因而可穿戴式聽診設(shè)備越來越受到關(guān)注，但是其具有高精度與低功耗的要求。該文設(shè)計了低功耗的面向可穿戴式的基于長短期記憶網(wǎng)絡(luò)(Long Short-Term Memory, LSTM)的智能心音異常診斷芯片，提出了包括預(yù)處理、特征提取以及異常診斷的心音異常診斷系統(tǒng)，并搭建了基于聽診器的心音采集FPGA系統(tǒng)，采用了數(shù)據(jù)增強的方法解決數(shù)據(jù)集的不平衡問題。基于預(yù)訓(xùn)練模型設(shè)計了智能心音異常診斷芯片，在SMIC180 nm工藝下完成了版圖設(shè)計和MPW流片。后仿真結(jié)果表明，智能心音異常診斷芯片的診斷準(zhǔn)確率為98.6%，功耗為762 μW，面積為3.06 mm × 2.45 mm，滿足可穿戴式智能心音異常診斷設(shè)備的高性能與低功耗的需求。
- 可穿戴式 /
- 心音 /
- 異常診斷 /
- 長短期記憶網(wǎng)絡(luò) /
- 低功耗
Abstract: The gravity of cardiovascular disease hazards necessitates the utmost importance of preventive measures and early diagnosis for such ailments. Conventional manual auscultation techniques and computer-based diagnostic methods prove inadequate in meeting the demands of auscultation. Consequently, wearable devices attract increasing attention, but they are required to obtain both a high accuracy and low-power consumption. An intelligent heart sound abnormal diagnostic chip based on LSTM for wearable applications is presented. The abnormal heart sound diagnostic system is developed, including preprocessing, feature extraction, and abnormal diagnosis. Furthermore, an FPGA-based system for heart sounds acquisition is constructed. The challenge of imbalanced datasets is addressed through the implementation of data augmentation techniques. By utilizing pre-trained model as a foundation, the intelligent heart sound abnormal diagnostic chip is developed, and the layout and MPW are finished under SMIC 180nm. The post-simulation results demonstrate that the chip achieves a diagnostic accuracy of 98.6%, a power consumption of 762 μW, and an area of 3.06 mm$ \times $2.45 mm, meeting the high-performance and low-power consumption prerequisites of wearable devices.
- Wearable applications /
- Heart sound /
- Abnormal diagnosis /
- LSTM /
- Low-power

HTML全文

圖 1 基于聽診器的心音采集FPGA系統(tǒng)

下載: 全尺寸圖片幻燈片

圖 2 采用數(shù)據(jù)增強技術(shù)后的心音

下載: 全尺寸圖片幻燈片

圖 3 心音異常診斷系統(tǒng)

下載: 全尺寸圖片幻燈片

圖 4 心肺音分離效果圖

下載: 全尺寸圖片幻燈片

圖 5 心音異常診斷模型訓(xùn)練過程

下載: 全尺寸圖片幻燈片

圖 6 心音異常診斷芯片架構(gòu)

下載: 全尺寸圖片幻燈片

圖 7 診斷模型的電路設(shè)計

下載: 全尺寸圖片幻燈片

圖 8 FPGA原型驗證方案

下載: 全尺寸圖片幻燈片

圖 9 心音異常診斷芯片的版圖

下載: 全尺寸圖片幻燈片

表 1 心音診斷模型訓(xùn)練參數(shù)

訓(xùn)練參數(shù)	值
框架	Pytorch
GPU型號	TITAN5 12 GB
EPOCH	300
BATCH SIZE	32
學(xué)習(xí)率	0.0001
優(yōu)化器	Adam

下載: 導(dǎo)出CSV

表 2 不同方法的性能對比

方法	準(zhǔn)確率(%)	功耗	面積(mm²)
YASEEN等人^[19]	97.9	/	/
CHOSH等人^[20]	98.3	/	/
ALKHODARI等人^[21]	99.3	/	/
KAO等人^[22]	/	12.6 mW @ 40 nm (前仿)	/
CHEN等人^[23]	/	6.6 mW @ 65 nm(實測)	10.15
心音異常診斷芯片	98.6	762 μW @ 180 nm (后仿)	7.50

下載: 導(dǎo)出CSV

參考文獻(xiàn)(23)

[1]	WHO. The top 10 causes of death[EB/OL]. https://www.who.int/en/news-room/fact-sheets/detail/the-top-10-causes-of-death, 2020.
[2]	WANG Fei, SYEDA-MAHMOOD T, and BEYMER D. Finding disease similarity by combining ECG with heart auscultation sound[C]. 2007 Computers in Cardiology, Durham, USA, 2007: 261–264. doi: 10.1109/CIC.2007.4745471.
[3]	DOMINGUEZ-MORALES J P, JIMENEZ-FERNANDEZ A F, DOMINGUEZ-MORALES M J, et al. Deep neural networks for the recognition and classification of heart murmurs using neuromorphic auditory sensors[J]. IEEE Transactions on Biomedical Circuits and Systems, 2018, 12(1): 24–34. doi: 10.1109/TBCAS.2017.2751545.
[4]	EJAZ K, NORDEHN G, ALBA-FLORES R, et al. A heart murmur detection system using spectrograms and artificial neural networks[C]. Proceedings of the Second IASTED International Conference on Circuits, Signals, and Systems, Clearwater Beach, USA, 2004: 374–379.
[5]	MILANI M G M, ABAS P E, DE SILVA L C, et al. Abnormal heart sound classification using phonocardiography signals[J]. Smart Health, 2021, 21: 100194. doi: 10.1016/j.smhl.2021.100194.
[6]	MILANI M G M, ABAS P E, and DE SILVA L C. A critical review of heart sound signal segmentation algorithms[J]. Smart Health, 2022, 24: 100283. doi: 10.1016/j.smhl.2022.100283.
[7]	XU Weize, YU Kai, YE Jingjing, et al. Automatic pediatric congenital heart disease classification based on heart sound signal[J]. Artificial Intelligence in Medicine, 2022, 126: 102257. doi: 10.1016/j.artmed.2022.102257.
[8]	ZHENG Yineng, GUO Xingming, WANG Yingying, et al. A multi-scale and multi-domain heart sound feature-based machine learning model for ACC/AHA heart failure stage classification[J]. Physiological Measurement, 2022, 43(6): 065002. doi: 10.1088/1361-6579/ac6d40.
[9]	RENNA F, OLIVEIRA J, and COIMBRA M T. Deep convolutional neural networks for heart sound segmentation[J]. IEEE Journal of Biomedical and Health Informatics, 2019, 23(6): 2435–2445. doi: 10.1109/JBHI.2019.2894222.
[10]	TARIQ Z, SHAH S K, and LEE Y. Feature-based fusion using CNN for lung and heart sound classification[J]. Sensors, 2022, 22(4): 1521. doi: 10.3390/s22041521.
[11]	DENG Muqing, MENG Tingting, CAO Jiuwen, et al. Heart sound classification based on improved MFCC features and convolutional recurrent neural networks[J]. Neural Networks, 2020, 130: 22–32. doi: 10.1016/j.neunet.2020.06.015.
[12]	CHEN Wei, SUN Qiang, CHEN Xiaomin, et al. Deep learning methods for heart sounds classification: A systematic review[J]. Entropy, 2021, 23(6): 667. doi: 10.3390/e23060667.
[13]	SHI K, SCHELLENBERGER S, WEBER L, et al. Segmentation of radar-recorded heart sound signals using bidirectional LSTM networks[C]. 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), Berlin, Germany, 2019: 6677–6680. doi: 10.1109/EMBC.2019.8857863.
[14]	LEE S H, KIM Y S, and YEO W H. Advances in microsensors and wearable bioelectronics for digital stethoscopes in health monitoring and disease diagnosis[J]. Advanced Healthcare Materials, 2021, 10(22): 2101400. doi: 10.1002/adhm.202101400.
[15]	WEI Min, SUN Kexue, WANG Chenxi, et al. A design and implementation for heart sound detection instrument based on FPGA[C]. 2015 3rd International Conference on Machinery, Materials and Information Technology Applications, Qingdao, China, 2015: 278–282. doi: 10.2991/icmmita-15.2015.55.
[16]	JUSAK J, PUSPASARI I, and KUSUMAWATI W I. A semi-automatic heart sounds identification model and its implementation in internet of things devices[J]. Advances in Electrical and Computer Engineering, 2021, 21(1): 45–56. doi: 10.4316/AECE.2021.01005.
[17]	LI Tao, YIN Yibo, MA Kainan, et al. Lightweight end-to-end neural network model for automatic heart sound classification[J]. Information, 2021, 12(2): 54. doi: 10.3390/info12020054.
[18]	WANG Jing, CHEN Ping, ZHANG Cheng, et al. Corona virus disease 2019 respiratory cycle detection based on convolutional neural network[C]. 2021 IEEE Biomedical Circuits and Systems Conference (BioCAS), Berlin, Germany, 2021: 1–4. doi: 10.1109/BioCAS49922.2021.9644970.
[19]	YASEEN, SON G Y, and KWON S. Classification of heart sound signal using multiple features[J]. Applied Sciences, 2018, 8(12): 2344. doi: 10.3390/app8122344.
[20]	GHOSH S K, PONNALAGU R N, TRIPATHY R K, et al. Automated detection of heart valve diseases using chirplet transform and multiclass composite classifier with PCG signals[J]. Computers in Biology and Medicine, 2020, 118: 103632. doi: 10.1016/j.compbiomed.2020.103632.
[21]	ALKHODARI M and FRAIWAN L. Convolutional and recurrent neural networks for the detection of valvular heart diseases in phonocardiogram recordings[J]. Computer Methods and Programs in Biomedicine, 2021, 200: 105940. doi: 10.1016/j.cmpb.2021.105940.
[22]	KAO Chaoyang, KUO H C, CHEN Jianwen, et al. RNNAccel: A fusion recurrent neural network accelerator for edge intelligence[EB/OL]. https://arxiv.org/abs/2010.13311, 2020.
[23]	CHEN Chixiao, DING Hongwei, PENG Huwan, et al. OCEAN: An on-chip incremental-learning enhanced processor with gated recurrent neural network accelerators[C]. 43rd IEEE European Solid State Circuits Conference, Leuven, Belgium, 2017: 259–262. doi: 10.1109/ESSCIRC.2017.8094575.