采用自注意力抗干扰网络的视频房颤检测

doi:10.11996/JG.j.2095-302X.2023020313

图学学报 ›› 2023, Vol. 44 ›› Issue (2): 313-323.DOI: 10.11996/JG.j.2095-302X.2023020313

• 图像处理与计算机视觉 • 上一篇下一篇

采用自注意力抗干扰网络的视频房颤检测

陈静¹^,²(), 杨学志²^,³(), 陈鲸¹, 刘雪南¹^,²

1.合肥工业大学计算机与信息学院，安徽合肥 230009
2.合肥工业大学工业安全与应急技术安徽省重点实验室，安徽合肥 230009
3.合肥工业大学软件学院，安徽合肥 230009

收稿日期:2022-06-23 接受日期:2022-08-22 出版日期:2023-04-30 发布日期:2023-05-01
通讯作者: 杨学志(1970-)，男，教授，博士。主要研究方向为数字信号处理、计算机视觉。E-mail：xzyang@hfut.edu.cn
作者简介:陈静(1997-)，女，硕士研究生。主要研究方向为视频信号处理、视频医学。E-mail：chenj_hfut@163.com
基金资助:
安徽省科技重大专项项目(JZ2019AKKZ0469);安徽高校协同创新项目(PA2019AGXC0114)

Video atrial fibrillation detection using self-attentional anti-interference network

CHEN Jing¹^,²(), YANG Xue-zhi²^,³(), CHEN Jing¹, LIU Xue-nan¹^,²

1. School of Computer Science and Information Engineering, Hefei University of Technology, Hefei Anhui 230009, China
2. Anhui Key Laboratory of Industrial Safety and Emergency Technology, Hefei Anhui 230009, China
3. School of Software, Hefei University of Technology, Hefei Anhui 230009, China

Received:2022-06-23 Accepted:2022-08-22 Online:2023-04-30 Published:2023-05-01
Contact: YANG Xue-zhi (1970-), professor, Ph.D. His main research interests cover digital signal processing, computer vision. E-mail：xzyang@hfut.edu.cn
About author:CHEN Jing (1997-)，master student. Her main research interests cover video signal processing, video medicine. E-mail：chenj_hfut@163.com
Supported by:
Anhui Province Science and Technology Major Special Project(JZ2019AKKZ0469);Anhui University Collaborative Innovation Project(PA2019AGXC0114)

摘要/Abstract

摘要：

房颤的早期发现与诊断是降低房颤以及并发症风险的关键。视频光电体积描记术(VPPG)技术为房颤筛查提供了新途径，但易受到现实场景中运动干扰。现有VPPG房颤检测方法存在运动干扰时会造成脉冲信号失真，从而发生误判。为解决以上问题，提出一种抗干扰视频房颤检测模型。该模型使用注意力编码器网络，从包含运动干扰的脉冲信号中提取鲁棒的脉冲潜在特征，径向基分类网络根据潜在特征实现房颤检测。注意力编码器将复杂脉冲信号映射到高维子空间，重点关注有效信息，提取稳健潜在特征。径向基分类网络在房颤标签监督下提高房颤识别能力，输出可靠结果。在200名测试者参与的自建数据集上进行实验，结果表明该模型在各类场景中均表现优异。在静态场景中，检测精度较最优对比算法提高了8.1%，敏感性提高了7.5%。在动态场景中，对比算法准确度均大幅下降，所提模型精度相比提升了16.5%，特异性提升了18.3%。模型具有良好的抗运动干扰能力，可有效地消除运动干扰影响，提高现实场景中视频房颤检测精度。

关键词: 视频光电体积描记术, 房颤检测, 抗运动干扰, 注意力编码器, 潜在特征

Abstract:

Early detection and diagnosis of atrial fibrillation is the key to reducing the risk of atrial fibrillation and complications. Although the video photoplethysmography (VPPG) technology provides a new approach to atrial fibrillation screening, it is susceptible to motion interference in real-world scenarios. When the existing VPPG atrial fibrillation detection method encounters motion interference, the pulse signal will be distorted, resulting in misjudgment. To solve the above problems, an anti-interference video atrial fibrillation detection model was proposed. The model employed an attention encoder network to extract robust spiking latent features from spiking signals containing motion disturbances. A radial basis classification network then performed atrial fibrillation detection based on these latent features. The attention encoder mapped complex impulse signals into high-dimensional subspaces, focusing on effective information and extracting robust latent features. Furthermore, Radial Basis Classification Network enhanced atrial fibrillation recognition ability under the supervision of atrial fibrillation labels and output reliable results. Experiments were carried out on a self-built dataset with 200 testers, and the results show that the proposed model performed well in various scenarios. In static scenes, the detection accuracy was 8.1% higher than the optimal comparison algorithm, and the sensitivity was 7.5% higher. In dynamic scenes, where the accuracy of the comparison algorithms was greatly reduced, the accuracy of the proposed model was improved by 16.5%, and the specificity was improved by 18.3%. The model demonstrated good anti-motion interference ability, effectively eliminating the influence of motion interference and improving the detection accuracy of video atrial fibrillation in real scenes.

Key words: video photoplethysmography, atrial fibrillation detection, anti-motion interference, attention encoder, potential characteristics

中图分类号:

TP391

陈静, 杨学志, 陈鲸, 刘雪南. 采用自注意力抗干扰网络的视频房颤检测[J]. 图学学报, 2023, 44(2): 313-323.

CHEN Jing, YANG Xue-zhi, CHEN Jing, LIU Xue-nan. Video atrial fibrillation detection using self-attentional anti-interference network[J]. Journal of Graphics, 2023, 44(2): 313-323.

图/表 15

图1 时频图特征(224×224像素)((a)正常测试者；(b)房颤患者)

Fig. 1 Time-frequency diagram features (224×224 px) ((a) Normal tester; (b) AF patients)

图2 抗干扰房颤分类网络结构

Fig. 2 Anti-interference atrial fibrillation classification network structure

图3 注意力去噪编码器结构

Fig. 3 Attention denoising encoder structure

图4 多头注意力模型

Fig. 4 Multihead attention model

图5 径向基房颤分类网络结构

Fig. 5 Radial basis function atrial fibrillation classification network structure

表1 数据集特征表

Table 1 Dataset feature Table

特征	房颤患者(100例)	正常测试者(100例)
年龄(岁)	62.76±10.29	63.32±10.99
性别(男性∶女性)	59∶41	44∶56
心率(次/分钟)	72.14±11.06	78.35±13.43
刘海(%)	34.0	27.0
胡子(%)	5	9
眼镜(%)	9	7

表2 静态场景对比实验结果

Table 2 Experimental results of static scene comparison

方法	S_E	S_P	P_PV	N_PV	A_cc
Similarity	0.807	0.808	0.819	0.822	0.864
LSTM	0.812	0.819	0.834	0.831	0.875
Region	0.827	0.832	0.853	0.849	0.891
本文	0.889	0.894	0.901	0.918	0.934

表3 动态场景实验结果

Table 3 Dynamic scene experimental results

方法	IPI误差指标			房颤分类指标
方法	SD_E	M_E	M_AE	S_E	S_P	P_PV	N_PV	A_cc
Similarity	199.1	96.2	188.3	0.712	0.657	0.735	0.724	0.702
LSTM	175.6	70.5	162.1	0.731	0.697	0.754	0.763	0.721
Region	169.1	68.4	150.3	0.782	0.726	0.794	0.773	0.752
本文	142.1	51.6	115.8	0.846	0.859	0.869	0.851	0.876

图6 各对比方法脉冲信号图((a)原始信号；(b)同步PPG信号；(c) Similarity方法；(d) LSTM方法；(e) Region方法；(f)本文方法)

Fig. 6 Pulse signal diagram for each comparison method ((a) Original signal; (b) PPG signal; (c) Similarity method; (d) LSTM method; (e) Region method; (f) Ours)

图7 子空间数实验结果图

Fig. 7 Plot of experimental results for subspace numbers

表4 消融实验结果

Table 4 Ablation experimental results

方法		IPI误差指标			房颤分类指标
方法		SD_E	M_E	M_AE	S_E	S_P	P_PV	N_PV	A_cc
全连接网络+RBF-AF		196.2	95.8	183.8	0.710	0.717	0.732	0.701	0.701
去噪编码器+RBF-AF		170.7	72.4	165.1	0.821	0.726	0.729	0.794	0.759
卷积编码器+RBF-AF		171.5	72.9	163.3	0.811	0.708	0.711	0.782	0.763
本文	注意力编码器+CNN	160.8	65.4	155.1	0.837	0.752	0.720	0.802	0.794
	注意力编码器+ Loss 1	166.8	69.5	160.1	0.823	0.748	0.713	0.794	0.812
	注意力编码器+RBF-AF	142.1	51.6	115.8	0.846	0.859	0.869	0.851	0.876

图8 各对比方法重构信号和IPI序列((a)重构信号；(b) IPI序列)

Fig. 8 Each comparison method reconstructed signal and IPI sequences ((a) Reconstructed signal; (b) IPI sequences)

图9 交叉数据集实验结果

Fig. 9 Cross dataset experimental results

图10 特征可视化((a)原始信号特征；(b)注意力编码器网络特征；(c)径向基网络特征)

Fig. 10 Feature visualization ((a) Original signal feature;(b) Attention encoder network feature; (c) Radial basis network feature)

图11 时长实验结果

Fig. 11 Duration results

参考文献 29

[1]	CHUNG M K, ECKHARDT L L, CHEN L Y, et al. Lifestyle and risk factor modification for reduction of atrial fibrillation: a scientific statement from the American heart association[J]. Circulation, 2020, 141(16): e750-e772.
[2]	HINDRICKS G, POTPARA T, DAGRES N, et al. 2020 ESC guidelines for the diagnosis and management of atrial fibrillation developed in collaboration with the European Association for Cardio-Thoracic Surgery (EACTS): the task force for the diagnosis and management of atrial fibrillation of the European Society of Cardiology (ESC) developed with the special contribution of the European Heart Rhythm Association (EHRA) of the ESC[J]. European Heart Journal, 2020, 42(5): 373-498. DOI URL
[3]	GUO Y T, WANG H, ZHANG H, et al. Mobile photoplethysmographic technology to detect atrial fibrillation[J]. Journal of the American College of Cardiology, 2019, 74(19): 2365-2375. DOI PMID
[4]	SUN Y, THAKOR N. Photoplethysmography revisited: from contact to noncontact, from point to imaging[J]. IEEE Transactions on Bio-Medical Engineering, 2016, 63(3): 463-477. DOI PMID
[5]	周双, 杨学志, 金兢, 等. 采用自适应信号恢复算法的非接触式心率检测[J]. 中国图象图形学报, 2019, 24(10): 1670-1682.
	ZHOU S, YANG X Z, JIN J, et al. Non-contact heart rate detection using self-adaptive signal recovery algorithm[J]. Journal of Image and Graphics, 2019, 24(10): 1670-1682. (in Chinese)
[6]	VERKRUYSSE W, SVAASAND L O, NELSON J S. Remote plethysmographic imaging using ambient light[J]. Optics Express, 2008, 16(26): 21434-21445. DOI PMID
[7]	FAVILLA R, ZUCCALÀ V C, COPPINI G. Heart rate and heart rate variability from single-channel video and ICA integration of multiple signals[J]. IEEE Journal of Biomedical and Health Informatics, 2019, 23(6): 2398-2408. DOI PMID
[8]	WANG W J, STUIJK S, DE HAAN G. Exploiting spatial redundancy of image sensor for motion robust rPPG[J]. IEEE Transactions on Biomedical Engineering, 2015, 62(2): 415-425. DOI URL
[9]	POH M Z, MCDUFF D J, PICARD R W. Advancements in noncontact, multiparameter physiological measurements using a webcam[J]. IEEE Transactions on Biomedical Engineering, 2011, 58(1): 7-11. DOI URL
[10]	LIU X N, YANG X Z, JIN J, et al. Self-adaptive signal separation for non-contact heart rate estimation from facial video in realistic environments[J]. Physiological Measurement, 2018, 39(6): 06NT01.
[11]	GHODRATIGOHAR M, GHANADIAN H, AL OSMAN H. A remote respiration rate measurement method for non-stationary subjects using CEEMDAN and machine learning[J]. IEEE Sensors Journal, 2020, 20(3): 1400-1410. DOI URL
[12]	LIU X N, YANG X Z, WANG D L, et al. Detecting pulse rates from facial videos recorded in unstable lighting conditions: an adaptive spatiotemporal homomorphic filtering algorithm[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 1-15.
[13]	COUDERC J P, KYAL S, MESTHA L K, et al. Detection of atrial fibrillation using contactless facial video monitoring[J]. Heart Rhythm, 2015, 12(1): 195-201. DOI URL
[14]	CORINO V, IOZZIA L, MARIANI A, et al. Identification of atrial fibrillation episodes using a camera as contactless sensor[C]// 2017 Computing in Cardiology Conference. New York: IEEE Press, 2017: 1-4.
[15]	CHENG P, CHEN Z C, LI Q Z, et al. Atrial fibrillation identification with PPG signals using a combination of time-frequency analysis and deep learning[J]. IEEE Access, 8: 172692-172706. DOI URL
[16]	郭一楠, 邵慧杰, 巩敦卫, 等. 基于希尔伯特黄变换和深度卷积神经网络的房颤检测[J]. 电子与信息学报, 2022, 44(1): 99-106.
	GUO Y N, SHAO H J, GONG D W, et al. Atrial fibrillation detection based on Hilbert-Huang transform and deep convolutional neural network[J]. Journal of Electronics ＆ Information Technology, 2022, 44(1): 99-106. (in Chinese)
[17]	KWON S, HONG J, CHOI E K, et al. Deep learning approaches to detect atrial fibrillation using photoplethysmographic signals: algorithms development study[J]. JMIR MHealth and UHealth, 2019, 7(6): e12770. DOI URL
[18]	YAN B P, LAI W H S, CHAN C K Y, et al. Contact-free screening of atrial fibrillation by a smartphone using facial pulsatile photoplethysmographic signals[J]. Journal of the American Heart Association, 2018, 7(8): e008585. DOI URL
[19]	SHI J G, ALIKHANI I, LI X B, et al. Atrial fibrillation detection from face videos by fusing subtle variations[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2020, 30(8): 2781-2795. DOI URL
[20]	LI X B, CHEN J, ZHAO G Y, et al. Remote heart rate measurement from face videos under realistic situations[C]// 2014 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press: 4264-4271.
[21]	ASTHANA A, ZAFEIRIOU S, CHENG S Y, et al. Robust discriminative response map fitting with constrained local models[C]// 2013 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press: 3444-3451.
[22]	KUMAR M, VEERARAGHAVAN A, SABHARWAL A. DistancePPG: robust non-contact vital signs monitoring using a camera[J]. Biomedical Optics Express, 2015, 6(5): 1565-1588. DOI PMID
[23]	DE HAAN G, JEANNE V. Robust pulse rate from chrominance-based rPPG[J]. IEEE Transactions on Biomedical Engineering, 2013, 60(10): 2878-2886. DOI URL
[24]	WANG C, ZHU M X, WANG X, et al. Time-frequency analysis of electroencephalogram signals in a cognitive decision-making task[C]// The 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society. New York: IEEE Press: 3469-3472.
[25]	KUMAR A, TOMAR H, MEHLA V K, et al. Stationary wavelet transform based ECG signal denoising method[J]. ISA Transactions, 2021, 114: 251-262. DOI PMID
[26]	COHEN M X. A better way to define and describe Morlet wavelets for time-frequency analysis[J]. NeuroImage, 2019, 199: 81-86. DOI PMID
[27]	HUMPHREYS G W, SUI J. Attentional control and the self: the self-attention network (SAN)[J]. Cognitive Neuroscience, 2016, 7(1-4): 5-17. DOI PMID
[28]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all You need[C]// The 31st International Conference on Neural Information Processing Systems. New York: ACM, 2017: 6000-6010.
[29]	VAN DER MAATEN L, HINTON G. Visualizing data using t-SNE[J]. Journal of Machine Learning Research, 2008, 9: 2579-2625.

采用自注意力抗干扰网络的视频房颤检测

Video atrial fibrillation detection using self-attentional anti-interference network

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 15

参考文献 29

相关文章 15

编辑推荐

Metrics

本文评价

[1]	胡欣, 周运强, 肖剑, 杨杰. 基于改进YOLOv5的螺纹钢表面缺陷检测[J]. 图学学报, 2023, 44(3): 427-437.
[2]	李刚, 张运涛, 汪文凯, 张东阳. 采用DETR与先验知识融合的输电线路螺栓缺陷检测方法[J]. 图学学报, 2023, 44(3): 438-447.
[3]	毛爱坤, 刘昕明, 陈文壮, 宋绍楼. 改进YOLOv5算法的变电站仪表目标检测方法[J]. 图学学报, 2023, 44(3): 448-455.
[4]	郝鹏飞, 刘立群, 顾任远. YOLO-RD-Apple果园异源图像遮挡果实检测模型[J]. 图学学报, 2023, 44(3): 456-464.
[5]	罗文宇, 傅明月. 基于YoloX-ECA模型的非法野泳野钓现场监测技术[J]. 图学学报, 2023, 44(3): 465-472.
[6]	李雨, 闫甜甜, 周东生, 魏小鹏. 基于注意力机制与深度多尺度特征融合的自然场景文本检测[J]. 图学学报, 2023, 44(3): 473-481.
[7]	肖天行, 吴静静. 基于残差和特征分块注意力的激光打码字符分割[J]. 图学学报, 2023, 44(3): 482-491.
[8]	王佳婧, 王晨, 朱媛媛, 王笑梅. 基于民国纸币的图元素匹配检索[J]. 图学学报, 2023, 44(3): 492-501.
[9]	孙龙飞, 刘慧, 杨奉常, 李攀. 面向医学图像层间插值的循环生成网络研究[J]. 图学学报, 2023, 44(3): 502-512.
[10]	刘冰, 叶成绪. 面向不平衡数据的肺部疾病细粒度分类模型[J]. 图学学报, 2023, 44(3): 513-520.
[11]	史彩娟, 石泽, 闫巾玮, 毕阳阳. 基于双语义双向对齐VAE的广义零样本学习[J]. 图学学报, 2023, 44(3): 521-530.
[12]	吴文欢, 张淏坤. 融合空间十字注意力与通道注意力的语义分割网络[J]. 图学学报, 2023, 44(3): 531-539.
[13]	葛海明, 张维, 王小龙, 朱晶晶, 贾非, 薛亚东. 基于SfM的城市电缆隧道三维重建方法优化研究[J]. 图学学报, 2023, 44(3): 540-550.
[14]	严圆, 高欣健, 高隽, 王昕, 程前. 基于非局部信息的大气偏振模式生成方法[J]. 图学学报, 2023, 44(3): 551-559.
[15]	赵玉琨, 任爽, 张鑫云. 结合对抗样本检测和重构的三维点云防御框架[J]. 图学学报, 2023, 44(3): 560-569.