没有合适的资源?快使用搜索试试~ 我知道了~
资源推荐
资源详情
资源评论
摘 要
I
摘 要
睡眠呼吸暂停综合征(Sleep Apnea Syndrome,SAS)是一种发病率极高却不易被察
觉的睡眠疾病。它严重影响着人们的睡眠质量,更是导致高血压、中风、冠心病等危险
性疾病的主要因素之一,严重时会造成夜间猝死。临床上通常使用多导睡眠仪诊断 SAS,
但其存在费时费力、检测费用昂贵、检测过程不舒适等缺点,导致大多数患者不能得到
及时的诊断和治疗。当睡眠过程中发生呼吸暂停时患者心率变慢,恢复正常呼吸时心率
在短暂增加后逐渐恢复正常,心电信号包含了可以表征 SAS 的主要特征,所以本文基于
心电信号开发便捷、有效的 SAS 自动检测算法。
目前基于心电信号的 SAS 检测算法大多集中于手动提取传统统计学特征或神经网
络提取深层特征。前者高度依赖医学知识和波形提取准确性且提取的特征不全面,后者
特征可解释性较差,指向性不强,导致模型鲁棒性较差。针对上述问题,本文开展了如
下研究内容:
(1)提出了基于单导联心电信号统计学特征的 SAS 检测算法。本文在单导联心电
信号上提取心率变异性(Heart Rate Variability,HRV)的时域、频域、非线性特征,共
17 个特征参数对 SAS 进行检测。时域特征从心率变化的角度对 SAS 进行判断;频域特
征从频谱角度分析了 SAS 对交感神经和副交感神经活性的影响;非线性特征从 HRV 信
号的复杂度和长短期变化的角度对 SAS 进行判断。在 Apnea-ECG 数据库上通过 5 折交
叉验证,该算法对一分钟长度的 SAS 段检测的平均准确率、敏感度、特异性分别达到
95.79%、95.91%、95.68%。
(2)提出了基于判别典型相关分析(Discriminant Canonical Correlation Analysis,
DCCA)融合 HRV 统计学特征和心电信号波形深层特征的 SAS 自动检测算法。基于医
学知识提取的统计学特征是针对 HRV 信号提取的多域特征,卷积神经网络提取的深层
特征是针对心电信号波形提取的特征。本文采用 DCCA 将两类特征进行融合,不仅弥补
了两类特征在特征提取上的不足,而且充分考虑了特征类内和类间的关系,最大化同类
特征相关性的同时减弱不同类特征间的相关性,去除了冗余特征。在 Apnea-ECG 数据
河北大学硕士学位论文
Ⅱ
库上通过 5 折交叉验证,该算法对一分钟长度的 SAS 段检测的平均准确率、敏感度和
特异性分别达到 97.33%、97.73%和 96.94%。且使用该算法对 Apnea-ECG 数据库的每条
心电记录进行 SAS 检测,其准确率、敏感度、特异性均达到 100%,实现了 SAS 的自动
检测,为临床便携设备的实现提供了有力的理论依据。
关键词 睡眠呼吸暂停综合征 心电信号 统计学特征 深层特征 判别典型相关分
析
Abstract
III
Abstract
Sleep Apnea Syndrome (SAS) is a sleep disorder that is highly prevalent but not easily
recognized. It seriously affects people's sleep quality, and is one of the main factors leading to
hypertension, stroke, coronary heart disease and other dangerous diseases, especially serious
will cause sudden death at night. Polysomnography is usually used clinically to diagnose SAS,
but it has the disadvantages of being time-consuming, expensive and uncomfortable, resulting
in most patients not getting timely diagnosis and treatment. When apnea occurs during sleep,
the patient's heart rate slows down and gradually returns to normal after a brief increase when
normal breathing resumes. The ECG signal contains the main features that can characterize
SAS, so this thesis develops a convenient and effective automatic SAS detection algorithm
based on the ECG signal.
Most of the current algorithms for SAS detection based on ECG signals focus on manual
extraction of traditional statistical features or neural network extraction of deep features. The
former highly dependent on medical knowledge and waveform extraction accuracy and
incomplete extracted features, while the latter features are less interpretable and less directional,
leading to poor model robustness. To address the above issues, the following research elements
were carried out in this thesis.
(1) A SAS detection algorithm based on statistical features of single-lead ECG signals is
proposed. In this thesis, the time domain features, frequency domain features, and nonlinear
features of Heart Rate Variability (HRV) are extracted from single-lead ECG signals, and a total
of 17 feature parameters are used to detect SAS. Time domain features judged SAS from the
perspective of heart rate variation; frequency domain features analyzed the effect of SAS on
sympathetic and parasympathetic activity from the spectral perspective; the nonlinear features
judged SAS from the perspective of HRV signal complexity and long-term and short-term
variation. The average accuracy, sensitivity, and specificity of the algorithm for one-minute
河北大学硕士学位论文
IV
length SAS segment detection reached 95.79%, 95.91%, and 95.68%, respectively, by 5-fold
cross-validation on the Apnea-ECG database.
(2) An automated SAS detection algorithm based on Discriminant Canonical Correlation
Analysis (DCCA) fusing HRV statistical features and deep features of ECG signal waveforms
is proposed. The statistical features extracted based on medical knowledge are multi-domain
features extracted for HRV signals, and the deep features extracted by convolutional neural
networks are features extracted for ECG signal waveforms. In this thesis, we use DCCA to fuse
two classes of features, which not only makes up for the shortage of two classes of features in
feature extraction, but also fully considers the intra-class and inter-class relationships of
features, maximizes the correlation of similar features while weakening the correlation between
different classes of features, and removes redundant features. The algorithm achieved an
average accuracy, sensitivity and specificity of 97.33%, 97.73% and 96.94%, respectively, for
one-minute length SAS segment detection by 5-fold cross-validation on the Apnea-ECG
database. The accuracy, sensitivity, and specificity of SAS detection for each ECG record in the
Apnea-ECG database using this algorithm reached 100%, enabling the automatic detection of
SAS and providing a favorable theoretical basis for the implementation of clinical portable
devices.
Keywords Sleep apnea syndrome ECG signal Statistical features Deep features
Discriminant canonical correlation analysis
目 录
V
目 录
第一章 绪论 ............................................................................................................................ 1
1.1 课题研究背景 ........................................................................................................... 1
1.2 研究目的与意义 ....................................................................................................... 2
1.3 国内外研究现状 ....................................................................................................... 3
1.4 本文主要研究内容 ................................................................................................... 7
1.5 本文章节安排 ........................................................................................................... 8
第二章 睡眠呼吸暂停综合征与心电信号的基本知识 ...................................................... 10
2.1 引言 ......................................................................................................................... 10
2.2 睡眠呼吸暂停综合征的发病机理及类型 ............................................................. 10
2.3 睡眠呼吸暂停综合征的表现形式 ......................................................................... 11
2.3.1 正常心电图表现形式 .................................................................................. 11
2.3.2 睡眠呼吸暂停综合征心电图表现形式 ...................................................... 12
2.4 睡眠呼吸暂停综合征的心电数据库介绍及预处理 ............................................. 13
2.5 本章小结 ................................................................................................................. 15
第三章 基于 HRV 统计学特征的睡眠呼吸暂停综合征检测算法 ................................... 16
3.1 引言 ......................................................................................................................... 16
3.2 基于 HRV 特征提取的睡眠呼吸暂停综合征检测算法 ...................................... 16
3.2.1 R 波检测 ...................................................................................................... 16
3.2.2 HRV 时域特征提取 .................................................................................... 17
3.2.3 HRV 频域特征提取 .................................................................................... 19
3.2.4 HRV 非线性特征提取 ................................................................................ 20
3.2.5 支持向量机分类器 ...................................................................................... 24
3.3 算法验证和结果分析 ............................................................................................. 25
3.3.1 性能评价指标 .............................................................................................. 25
剩余54页未读,继续阅读
资源评论
- tangxiaolumama2024-02-25内容与描述一致,超赞的资源,值得借鉴的内容很多,支持!
2201_75761617
- 粉丝: 20
- 资源: 7339
下载权益
C知道特权
VIP文章
课程特权
开通VIP
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功