WO2021184410A2 - 一种非接触式的心血管健康评估方法 - Google Patents

一种非接触式的心血管健康评估方法 Download PDF

Info

Publication number
WO2021184410A2
WO2021184410A2 PCT/CN2020/081533 CN2020081533W WO2021184410A2 WO 2021184410 A2 WO2021184410 A2 WO 2021184410A2 CN 2020081533 W CN2020081533 W CN 2020081533W WO 2021184410 A2 WO2021184410 A2 WO 2021184410A2
Authority
WO
WIPO (PCT)
Prior art keywords
beat
data
bcg
ecg
assessment method
Prior art date
Application number
PCT/CN2020/081533
Other languages
English (en)
French (fr)
Inventor
何光强
白忠瑞
赵荣建
方震
Original Assignee
南京润楠医疗电子研究院有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 南京润楠医疗电子研究院有限公司 filed Critical 南京润楠医疗电子研究院有限公司
Publication of WO2021184410A2 publication Critical patent/WO2021184410A2/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
    • A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
    • A61B5/1102—Ballistocardiography
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
    • A61B5/024—Detecting, measuring or recording pulse rate or heart rate
    • A61B5/02405—Determining heart rate variability
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems

Definitions

  • the invention relates to the medical field, in particular to a non-contact cardiovascular health assessment method.
  • Cardiovascular diseases have severely affected people's lives and health. As of 2018, there are 290 million patients with cardiovascular diseases in China, and the prevalence rate is on the rise. Therefore, daily monitoring, evaluation and disease diagnosis of cardiovascular health are of great significance.
  • ECG Electrocardiogram
  • impedance electrocardiogram impedance electrocardiogram
  • echocardiogram echocardiogram
  • phonocardiogram phonocardiogram
  • Ballistocardiogram is a non-invasive, non-contact monitoring method of cardiovascular function. It is mainly caused by the change of the body's gravity caused by the blood circulation process. Compared with other cardiovascular detection technologies, it has the advantages of non-invasive, no direct contact and convenient detection, and is particularly suitable for long-term monitoring.
  • Heart Rate Variability arises from the regulation of the heart's sinus node by the autonomic nervous system, which reflects the degree of sinus arrhythmia in the heart itself and the balance of the interaction between neurohumoral factors and the sinus node.
  • the present invention provides a non-contact cardiovascular health assessment method for non-contact auxiliary diagnosis of the user's heart condition.
  • a non-contact cardiovascular health assessment method of the present invention includes the following steps: S101: data acquisition and preprocessing; S102: extracting HRV features; S103: designing a transfer learning classifier to test the data.
  • the S101 includes the following steps: S1011: collect user's BCG data, download ECG data from the database; S1012: unmix the BCG data, and remove noise from the ECG; S1013: calculate the beat cycle of the BCG data , Calculate the beat-by-beat heart cycle of ECG data; S1014: Remove the abnormal value of the beat-by-beat heart cycle of the BCG data and the beat-by-beat heart cycle of the ECG data.
  • the S102 includes the following steps: S1021: Extract HRV time domain features from the BCG data beat-by-beat heart cycle and ECG data beat-by-beat heart cycle; S1022: Extract the beat-by-beat heart cycle from the BCG data and the beat-by-beat heart cycle from the ECG data HRV frequency domain features; S1023: extract HRV nonlinear features from BCG data beat-by-beat cardiac cycle and ECG data beat-by-beat cardiac cycle; S1024: normalize and filter the time-domain, frequency-domain and nonlinear features of HRV.
  • the S103 includes the following steps: S1031: setting the parameters of the classifier; S1032: dividing the time domain feature, frequency domain feature, and nonlinear feature of BCG into training set and test set; dividing the time domain feature of ECG, Frequency domain features and nonlinear features are passed into the transfer learning model; S1033: Use the test set of BCG time domain features, frequency domain features and nonlinear features to test; adjust the classifier parameters twice.
  • an IIR filter is used in S1012 for signal unmixing and noise removal.
  • the template matching method is used to calculate the BCG signal beat-by-beat heart rate
  • the PT algorithm is used to calculate the ECG beat-by-beat heart rate.
  • the type of the classifier is a decision tree or SVM, and the number of iterations N of the classifier is 50.
  • the training set and the test set are distributed at a ratio of 4:1 in a random manner.
  • a non-contact cardiovascular health assessment method provided by the present invention utilizes the homology between BCG and ECG, and uses a transfer learning method to extract useful information from massive ECG databases for classifying BCG data Discrimination can be used for daily monitoring, evaluation and tracking of the user's heart health status, and it provides a feasible solution for BCG-based heart rate variability analysis for daily health evaluation.
  • FIG. 1 is a flowchart of the present invention
  • FIG. 2 is an effect diagram of collecting a BCG signal diagram and using a template matching algorithm to mark characteristic points according to an embodiment of the present invention.
  • FIG. 3 is an ECG signal diagram of a database provided by an embodiment of the present invention and an effect diagram of marking characteristic points using a PT algorithm.
  • Fig. 4 is a block diagram of Tradaboost migration learning provided by an embodiment of the present invention.
  • This embodiment provides a non-contact cardiovascular health assessment method, as shown in FIG. 1, including the following steps: S101: data acquisition and preprocessing; S102: extracting HRV features; S103: designing a transfer learning classifier for data carry out testing.
  • step S101 the specific operation is step S1011: collecting user BCG data.
  • the existing methods such as optical fiber, piezoelectric ceramics, piezoelectric film or video, millimeter wave radar, etc. are used to obtain BCG signals of patients with cardiovascular diseases and samples of healthy people in resting state for more than 20 minutes.
  • a piezoelectric ceramic sensor placed under the mattress is specifically used to collect 50 BCG sample data of cardiovascular disease patients and healthy people, each sample is 30 minutes, and the piezoelectric ceramic sensor model is not limited.
  • download the ECG data from the database specifically download the original ECG data with sample tags (that is, with tags for cardiovascular disease patients and healthy people) from a public data set with certain authority.
  • the downloaded disease samples and healthy samples should have the same amount of data as possible, the number of sample cases should be more than 50 cases, and the total data time should be greater than 50 hours.
  • five data sets including NSR1, NSR2, INCART, SHAREE, and CHF in the Physionet database are selected to download data of a total of 116 hours of data of 50 samples of cardiovascular disease patients and healthy people.
  • step S1012 signal unmixing of the collected BCG data, and noise removal of the downloaded ECG.
  • the BCG signal uses an IIR filter for signal unmixing, which filters out breathing components and noise components in the BCG signal by using an IIR filter with a suitable passband frequency, leaving a purer heartbeat vibration in the BCG Ingredients.
  • IIR filters include wavelet decomposition, empirical mode decomposition and so on.
  • a 6th-order Butterworth bandpass filter with a passband frequency of 8-24 Hz is selected to filter the BCG.
  • the ECG signal uses an IIR filter to remove noise, which is to use an IIR filter with a suitable frequency to remove noise and baseline drift in the ECG signal to obtain a higher quality ECG signal.
  • IIR filters include moving average filtering, wavelet decomposition, and so on.
  • a 0.2Hz-45Hz 6th-order Butterworth filter is selected to filter the ECG to remove noise.
  • step S1013 the template matching method is used to calculate the beat-by-beat heart cycle of the BCG signal, and the PT algorithm is used to calculate the beat-by-beat heart cycle of the ECG.
  • the template matching method in the prior art is used to calculate the beat-by-beat heart cycle of the BCG signal. Specifically, the template matching method is used to find the characteristic points of the BCG signal, and then the heartbeat cycle is calculated in combination with the sampling frequency.
  • FIG. 2 shows the effect of using template matching to mark the J wave position of the BCG in this embodiment, that is, the BCG signal beats the heartbeat cycle.
  • Template matching is an idea of using a template to compare and calculate the similarity with the object to be processed. It is often used for image similarity detection in image processing, it is often used for image similarity comparison, and it is also used in the field of signal processing to detect signal waveform similarity.
  • the initial BCG signal is used to generate a template, and then the template is used to compare with the template generated later (the template generated by the BCG signal collected later). The process of this comparison is template matching.
  • the above J wave position is the position of the marked point on the horizontal axis of time in FIG. 2.
  • Using the PT algorithm to calculate the ECG beat-by-beat heart cycle is through the steps of band-pass filtering, differentiation, square, sliding window integration of the PT algorithm, and more accurately marks the R wave position of each beat.
  • Figure 3 shows the effect of using the PT algorithm to mark the R wave position of the ECG in this embodiment, and the interval between adjacent R waves is the ECG beat-by-beat cardiac cycle.
  • the PT algorithm is a particularly commonly used signal processing algorithm for marking R points in ECG processing.
  • the PT algorithm is disclosed in the wearable ECG signal-based nervousness discrimination method and system thereof in Patent Publication No. CN109770920A.
  • the above-mentioned R wave position is the position of the marked point on the horizontal axis of time in FIG. 3.
  • Step S1014 Remove abnormal values in the beat-by-beat cardiac cycle of BCG and ECG, the removal method is 3 ⁇ principle, box diagram analysis, etc., to remove abnormal value points of the cardiac cycle caused by detection errors or occurrence of premature beats. So that the heart rate variability can be calculated later.
  • This embodiment adopts the 3 ⁇ principle: in the normal distribution, ⁇ represents the standard deviation, and ⁇ represents the mean value.
  • the probability that the value is distributed in ( ⁇ - ⁇ , ⁇ + ⁇ ) is 0.6826
  • the probability that the value is distributed in ( ⁇ -2 ⁇ , ⁇ +2 ⁇ ) is 0.9544
  • the probability that the value is distributed in ( ⁇ -3 ⁇ , ⁇ +3 ⁇ ) It can be considered that the value of the beat-by-beat cardiac cycle of BCG and ECG is almost all concentrated in the interval of ( ⁇ -3 ⁇ , ⁇ +3 ⁇ )], and the possibility of exceeding this range is only less than 0.3%. Out of this range is an outlier.
  • Step S102 includes the following steps:
  • Step S1021 Extracting the HRV time-domain features from the beat-by-beat heart cycles of BCG and ECG refers to extracting the HRV time-domain features from the BCG and ECG beat-by-beat heart cycles calculated in step S101.
  • the methods of extracting HRV time-domain features include average, standard deviation, root mean square deviation, peak value, median, and standard deviation of difference.
  • the method of extracting the HRV frequency domain features may be to find the power of each frequency segment of the beat-by-beat cardiac cycle sequence.
  • step S1023 the method of extracting the non-linear characteristics of the HRV is to find the axis length of the Poincaré diagram, the sympathetic index and various entropy values.
  • Normalization is a way to simplify calculations, that is, a dimensional expression is transformed into a non-dimensional expression and becomes a scalar.
  • Step S103 includes the following steps:
  • Step S1031 Setting the parameters of the transfer learning classifier refers to setting some hyperparameters of the transfer learning classifier.
  • the types of transfer learning models include feature mapping transfer learning models, sample weight transfer learning models, and so on. This embodiment adopts Tradaboost, a classifier for sample weight migration:
  • the hyperparameters are respectively set as follows: the type of base classifier is decision tree or SVM, and the number of iterations N is 50 (the value can be adjusted according to the situation).
  • Step S1032 The HRV features of BCG are randomly divided into training set (source domain training data) and test set at 4:1, and the HRV feature training set of BCG is passed into the migration learning classifier as the target domain feature of the migration learning classifier.
  • the HRV feature (auxiliary domain training data) of ECG is transferred to the migration learning model as the auxiliary domain feature of the migration learning classifier.
  • the advantage of using the HRV feature calculated in the ECG as the auxiliary domain feature is that learning from ECG data with a large amount of data helps to improve the classification accuracy and generalization ability of the target domain data, so as to improve the performance of the classifier.
  • Step S1033 Train the transfer learning classifier and use the HRV test set of the BCG for testing, and perform secondary adjustment on the parameters of the transfer learning classifier to generate the final classifier.
  • the overall step flow chart of the present invention is shown in FIG. 1, and the transfer learning flow chart corresponding to the step flow chart of the present invention is shown in FIG. 4.
  • the Tradaboost classifier used in this embodiment is modified from the Adaboost algorithm, a Boosting ensemble learning classification model.
  • the sample data is reduced to adjust the weight using the following formula:
  • ⁇ t ⁇ t /(1- ⁇ t )
  • ⁇ t and ⁇ are the weight adjustment coefficients used for the target domain sample and the auxiliary domain sample
  • ⁇ t is the error rate of the classifier in the target domain in the previous iteration
  • n is the number of target domain samples
  • N is the total iteration frequency.
  • I the weight of the i-th sample in the t-th iteration.
  • h t (x i ) and c(x i ) are the estimated value of the i-th sample by the classifier and its actual value, and the value is 1 or 0.
  • BCG samples to be classified For new BCG samples to be detected (ie, BCG samples to be classified), preprocessing and feature extraction are performed, and a classifier is used to classify them.
  • the classifier trained using migration learning can be used to classify the HRV calculated from BCG data.
  • the steps for calculating HRV from BCG are shown in S101 and S102.
  • the HRV features calculated from the BCG data to be classified are input into the classifier model, and the classifier model outputs the discriminant results: cardiovascular disease samples, healthy samples, and estimates of credibility.
  • the non-contact cardiovascular health assessment method of the present invention extracts HRV features from the existing BCG and ECG signals respectively, and then builds a model through the transfer learning classifier, and the BCG and ECG signals of the patient to be tested are input into the transfer learning classifier , After the migration learning classifier produces health discrimination results, the detection is convenient and the result accuracy is high.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Surgery (AREA)
  • Public Health (AREA)
  • Signal Processing (AREA)
  • Biophysics (AREA)
  • Pathology (AREA)
  • Physiology (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Molecular Biology (AREA)
  • Veterinary Medicine (AREA)
  • Animal Behavior & Ethology (AREA)
  • General Health & Medical Sciences (AREA)
  • Cardiology (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Psychiatry (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Fuzzy Systems (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Dentistry (AREA)
  • Power Engineering (AREA)
  • Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Description

一种非接触式的心血管健康评估方法 技术领域
本发明涉及医学领域，具体涉及一种非接触式的心血管健康评估方法。
背景技术
心血管疾病已经严重影响到人们的生命健康，截至到2018年，中国有2.9亿心血管疾病患者，且患病率处于持续上升阶段。因此心血管健康的日常监测、评价和疾病诊断有十分重要的意义。
目前常规的心血管健康状况监测方式包括心电图(Electrocardiograph,即ECG)，心阻抗图、超声心动图、心音图等，都需要使用电极等与皮肤接触来进行监测，并且对监测环境、条件以及操作人员都有一定的要求。这些限制都会给使用者造成身体上的不适和生活上的不便，不适用于长期的日常监测。
心冲击信号(Ballistocardiogram，即BCG)是一种无创、无接触式的心血管功能监测手段。它主要是由血液循环过程中造成的人体重力的变化引起的。与其他心血管检测技术相比，它具有无创、无直接接触和检测方便等优势，特别适合进行长期监测。
使用BCG信号计算心率变异性(Heart Rate Variability，即HRV)进而去评估用户健康状况是一种可行的方案。心率变异性产生于自主神经系统对心脏窦房结的调节，反映了心脏本身窦性心律不齐的程度以及神经体液因素与窦房结之间相互作用的平衡关系。
想要正确地使用通过BCG所计算的HRV进行心脏健康状况的判别，需要大量的数据支持。但目前BCG信号缺少较为完善的大型数据库且采集带有用户心脏健康状况标签的BCG数据较为繁琐。
目前网络上存在较多ECG的带有标签的高质量的数据库，而ECG与BCG具有同源性，因此两者所计算的HRV具有很大的相关性，但同时由于BCG的产生滞后于ECG，两者计算所得HRV又有一定的差别，故ECG不能直接用于BCG的分析。
综上，需要一种可靠的适用于BCG信号的用户心脏健康状况辅助诊断方法，用于非接触式对用户尤其是心血管疾病高风险人群的心脏健康状况日常监测。
发明内容
为了克服现有技术中的不足，本发明提供一种非接触式的心血管健康评估方法，用于非接触式地对用户心脏状况健康与否的辅助诊断。
为了实现上述目的，本发明的一种非接触式的心血管健康评估方法，包括以下步骤：S101：数据获取及预处理；S102：提取HRV特征；S103：设计迁移学习分类器对数据进行测试。
进一步的，所述S101中包括以下步骤：S1011：采集用户的BCG数据，从数据库下载ECG数据；S1012：将BCG数据进行信号解混，将ECG进行噪音去除；S1013：计算BCG数据逐拍心动周期，计算ECG数据逐拍心动周期；S1014：去除BCG数据逐拍心动周期和ECG数据逐拍心动周期的异常值。
进一步的，所述S102中包括以下步骤：S1021：从BCG数据逐拍心动周期、ECG数据逐拍心动周期提取HRV时域特征；S1022：从BCG数据逐拍心动周期、ECG数据逐拍心动周期提取HRV频域特征；S1023：从BCG数据逐拍心动周期、ECG数据逐拍心动周期提取HRV非线性特征；S1024：对HRV的时域特征、频域特征和非线性特征进行归一化和筛选。
进一步的，所述S103中包括以下步骤：S1031：设置分类器的参数；S1032：将BCG的时域特征、频域特征和非线性特征分为训练集和测试集；将ECG的时域特征、频域特征和非线性特征传入迁移学习模型；S1033：使用BCG的时域特征、频域特征和非线性特征的测试集进行测试；对分类器参数二次调参。
进一步的，在S1012中使用IIR滤波器进行信号解混和噪音去除。
进一步的，在S1013中，使用模板匹配法计算BCG信号逐拍心率，使用PT算法计算ECG逐拍心率。
进一步的，在S1031中，分类器的种类为决策树或SVM，所述分类器的迭代次数N为50。
进一步的，在S1032中，采用随机的方式将训练集和测试集按4:1的比例分配。
有益效果：本发明提供的一种非接触式的心血管健康评估方法，利用BCG与ECG的同源性，使用迁移学习的方法提取海量的ECG数据库中的有用信息，用于对BCG数据进行分类判别，可用于用户对心脏健康状况的日常监测、评估和追踪，且为基于BCG的心率变异性分析用于日常健康评估提供可行方案。
附图说明
下面结合附图对本发明作进一步描写和阐述。
图1是本发明的流程图；
图2是本发明实施例提供的采集BCG信号图与使用模板匹配算法标记特征点的效果图。
图3为本发明实施例提供的数据库ECG信号图与使用PT算法标记特征点的效果图。
图4是本发明实施例提供的Tradaboost迁移学习的框图。
具体实施方式
下面将结合附图、通过对本发明的优选实施方式的描述，更加清楚、完整地阐述本发明的技术方案。
本实施例提供了一种非接触式的心血管健康评估方法，如图1所示，包括以下步骤：S101：数据获取及预处理；S102：提取HRV特征；S103：设计迁移学习分类器对数据进行测试。
步骤S101中，具体的操作为步骤S1011：采集用户BCG数据。具体是通过光纤、压电陶瓷、压电薄膜或者视频、毫米波雷达等现有方式获取心血管疾病患者样本和健康人样本静息状态下20分钟以上的BCG信号。
本实施例具体采用置于床垫下方的压电陶瓷传感器采集心血管疾病患者和健康人BCG样本数据各50例，每例样本30分钟，压电陶瓷传感器的型号无限定。
同时，从数据库中下载ECG数据，具体是从具有一定权威性的公共数据集中下载带有样本标签(即带有心血管疾病患者和健康人标签)的原始ECG数据。所下载患病样本和健康样本尽量同等数据量，样本例数都应在50例以上，数据总时长需大于50小时。本实施例选择了Physionet数据库中的NSR1、NSR2、INCART、SHAREE、CHF等5个数据集中下载了心血管类疾病患者和健康人群样本各50例共116小时时长的数据。
步骤S1012中：将采集的BCG数据进行信号解混，将下载的ECG进行噪音去除。
具体的，其中BCG信号使用IIR滤波器进行信号解混，是通过使用合适通带频率的IIR滤波器将BCG信号中的呼吸成分和噪声成分等滤除，留下较为纯净的BCG中的心跳震动的成分。除了IIR滤波器，其他可选现有的方法还有小波分解、经验模式分解等。本实施例中，选用通带频率为8Hz-24Hz的6阶巴特沃斯带通滤波器对BCG进行滤波。
其中ECG信号使用IIR滤波器进行噪声去除，是使用合适频率的IIR滤波器将ECG 信号中的噪声和基线漂移等去除，得到质量更高的ECG信号。除了IIR滤波器，其他现有可选方法还有滑动平均滤波、小波分解等。本实施例选择0.2Hz-45Hz的6阶巴特沃斯滤波器对ECG进行滤波去除噪音。
步骤S1013,使用模板匹配法计算BCG信号逐拍心动周期，使用PT算法计算ECG逐拍心动周期。
使用现有技术的模板匹配法计算BCG信号逐拍心动周期，具体是使用模板匹配的方式寻找BCG信号特征点，进而结合采样频率计算心动周期。图2展示了本实施例使用模板匹配标记BCG的J波位置的效果，即BCG信号逐拍心动周期。
模板匹配是一种使用模板去跟待处理的对象进行对比计算相似度的思想。常用于图像处理中进行图像相似度的检测，常用于图像相似度的对比，也用于信号处理的领域，进行信号波形相似度的检测。本专利中，先利用初始的BCG信号产生模板，再使用该模板与在后生成的模板(后来收集的BCG信号生成的模板)进行比较，该比较的过程就是模板匹配。
上述J波位置为图2中标记点在时间横轴上点的位置。
使用PT算法计算ECG逐拍心动周期，是通过PT算法的带通滤波、微分、平方、滑窗积分等步骤，较为精准地标记出每拍心跳的R波位置。图3展示了本实施例中使用PT算法标记ECG的R波位置的效果，相邻R波的间隔即为ECG逐拍心动周期。
PT算法是心电处理中特别常用的标记R点的信号处理算法，在专利公开号CN109770920A的基于穿戴式心电信号的紧张情绪判别方法及其系统公开了该PT算法。
上述R波位置为图3中标记点在时间横轴上的位置。
步骤S1014:去除BCG、ECG的逐拍心动周期中的异常值，去除的方法为3δ原则、箱型图分析等，去除因检测错误或发生早搏等引起的异常的心动周期数值点。以便随后计算心率变异性。本实施例采用3δ原则：在正态分布中σ代表标准差,μ代表均值。x＝μ即为图像的对称轴3σ原则。数值分布在(μ-σ,μ+σ)中的概率为0.6826数值分布在(μ-2σ,μ+2σ)中的概率为0.9544，数值分布在(μ-3σ,μ+3σ)中的概率为0.9974，可以认为，BCG、ECG的逐拍心动周期的取值几乎全部集中在(μ-3σ,μ+3σ)]区间内，超出这个范围的可能性仅占不到0.3％。超出该范围的即为异常值。
步骤S102包括以下步骤：
步骤S1021:从BCG、ECG逐拍心动周期中提取HRV时域特征，是指从经过步骤S101计算得到得BCG、ECG逐拍心动周期中提取HRV时域特征。提取HRV时域特征的方式包 括平均值、标准差、均方根差、峰值、中位数、差分标准差等。
步骤S1022，提取HRV频域特征的方式可以为求解逐拍心动周期序列各频率段的功率等。
步骤S1023，提取HRV非线性特征的方式为求庞加莱图轴长、交感神经指数及各种熵值。
步骤S1024，对各HRV特征进行归一化和筛选，是对HRV进行特征归一化以便分类器使用，然后使用卡方检验、随机森林等方式对各特征的重要性进行排序，根据效果，保留重要程度较高的若干维特征。本实施例经过特征筛选，选择了各类HRV特征共30维。归一化就是把数据变成(0，1)或者(1,1)之间的小数，例如对于一个采样频率为500hz的系统，400hz的归一化频率就为400/500＝0.8。主要是为了数据处理方便提出来的，把数据映射到0～1范围之内处理，更加便捷快速。把有量纲表达式变成无量纲表达式，便于不同单位或量级的指标能够进行比较和加权。归一化是一种简化计算的方式，即将有量纲的表达式，经过变换，化为无量纲的表达式，成为纯量。
步骤S103包括以下步骤：
步骤S1031:设置迁移学习分类器的参数，是指设置一些迁移学习分类器的超参数。迁移学习模型种类为特征映射迁移学习模型、样本权重迁移学习模型等。本实施例采用了一种样本权重迁移的分类器Tradaboost：
在本步骤超参数分别设置如下：基分类器种类为决策树或者SVM、迭代次数N为50(数值可根据情况可调)。
步骤S1032:将BCG的HRV特征随机按4:1分为训练集(源域训练数据)和测试集，将BCG的HRV特征训练集作为迁移学习分类器的目标域特征传入迁移学习分类器，将ECG的HRV特征(辅助域训练数据)作为迁移学习分类器的辅助域特征传入迁移学习模型。将ECG中计算得到的HRV特征作为辅助域特征的好处在于：从数据量较大的ECG数据中学习有助于提高目标域数据分类准确性和泛化能力的信息，以提高分类器性能。
步骤S1033，训练迁移学习分类器并使用BCG的HRV测试集进行测试，对迁移学习分类器参数进行二次调参，生成最终分类器。本发明的整体步骤流程图如图1所示，本发明与步骤流程图对应的迁移学习流程图如图4所示。
本实施例使用的Tradaboost分类器是从一种Boosting集成学习分类模型Adaboost算法修改而来，当辅助域样本分类错误时，降低此样本数据采用以下公式调整权重：
β t＝∈ t/(1-∈ t)
Figure PCTCN2020081533-appb-000001
Figure PCTCN2020081533-appb-000002
其中，β t与β分别为目标域样本和辅助域样本所用的权重调整系数，∈ t是上一轮迭代中分类器在目标域上的错误率，n为目标域样本数目，N为总迭代次数。
Figure PCTCN2020081533-appb-000003
是第t轮迭代中第i个样本的权重。h t(x i)、c(x i)分别为分类器对第i个样本的估计值和其实际值，取值为1或0。
对于新的、需检测的BCG样本(即待分类的BCG样本)进行预处理和特征提取并使用分类器对其进行分类。使用迁移学习训练好的分类器可用于对从BCG数据计算的HRV的分类，从BCG中计算HRV的步骤如S101、S102所示。将从待分类的BCG数据中计算得的HRV特征输入分类器模型，分类器模型输出判别结果：心血管疾病样本、健康样本，并给出可信度估计。
本发明的非接触式的心血管健康评估方法，对已有的BCG、ECG信号分别提取HRV特征，再通过迁移学习分类器建立模型，待测病人的BCG、ECG信号输入到迁移学习分类器中，经过迁移学习分类器产生健康判别结果，检测方便、结果准确度较高。
上述具体实施方式仅仅对本发明的优选实施方式进行描述，而并非对本发明的保护范围进行限定。在不脱离本发明设计构思和精神范畴的前提下，本领域的普通技术人员根据本发明所提供的文字描述、附图对本发明的技术方案所作出的各种变形、替代和改进，均应属于本发明的保护范畴。本发明的保护范围由权利要求确定。

Claims (9)

  1. 一种非接触式的心血管健康评估方法，其特征在于，包括以下步骤：
    S101：数据获取及预处理；
    S102：提取HRV特征；
    S103：设计迁移学习分类器对数据进行测试。
  2. 根据权利要求1所述的一种非接触式的心血管健康评估方法，其特征在于，所述S101中包括以下步骤：
    S1011：采集用户的BCG数据，从数据库下载ECG数据；
    S1012：将BCG数据进行信号解混，将ECG进行噪音去除；
    S1013：计算BCG数据逐拍心动周期，计算ECG数据逐拍心动周期；
    S1014：去除BCG数据逐拍心动周期和ECG数据逐拍心动周期的异常值。
  3. 根据权利要求1所述的一种非接触式的心血管健康评估方法，其特征在于，所述S102中包括以下步骤：
    S1021：从BCG数据逐拍心动周期、ECG数据逐拍心动周期提取HRV时域特征；
    S1022：从BCG数据逐拍心动周期、ECG数据逐拍心动周期提取HRV频域特征；
    S1023：从BCG数据逐拍心动周期、ECG数据逐拍心动周期提取HRV非线性特征；
    S1024：对HRV的时域特征、频域特征和非线性特征进行归一化和筛选。
  4. 根据权利要求3所述的一种非接触式的心血管健康评估方法，其特征在于，所述S103中包括以下步骤：
    S1031：设置分类器的参数；
    S1032：将BCG的时域特征、频域特征和非线性特征分为训练集和测试集；将ECG的时域特征、频域特征和非线性特征传入迁移学习模型；
    S1033：使用BCG的时域特征、频域特征和非线性特征的测试集进行测试；对分类器参数二次调参。
  5. 根据权利要求2所述的一种非接触式的心血管健康评估方法，其特征在于，在S1012中使用IIR滤波器进行信号解混和噪音去除。
  6. 根据权利要求2所述的一种非接触式的心血管健康评估方法，其特征在于，在S1013中，使用模板匹配法计算BCG信号逐拍心率，使用PT算法计算ECG逐拍心率。
  7. 根据权利要求4所述的一种非接触式的心血管健康评估方法，其特征在于，在S1031中，分类器的种类为决策树或者SVM。
  8. 根据权利要求4所述的一种非接触式的心血管健康评估方法，所述迁移学习模型为特征映射迁移学习模型、样本权重迁移学习模型。
  9. 根据权利要求4所述的一种非接触式的心血管健康评估方法，其特征在于，在S1032中，采用随机的方式将训练集和测试集按4:1的比例分配。
PCT/CN2020/081533 2020-03-18 2020-03-27 一种非接触式的心血管健康评估方法 WO2021184410A2 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010191926.9 2020-03-18
CN202010191926.9A CN111329455A (zh) 2020-03-18 2020-03-18 一种非接触式的心血管健康评估方法

Publications (1)

Publication Number Publication Date
WO2021184410A2 true WO2021184410A2 (zh) 2021-09-23

Family

ID=71174685

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/081533 WO2021184410A2 (zh) 2020-03-18 2020-03-27 一种非接触式的心血管健康评估方法

Country Status (2)

Country Link
CN (2) CN114010171B (zh)
WO (1) WO2021184410A2 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115429284A (zh) * 2022-09-16 2022-12-06 山东科技大学 心电信号分类方法、系统、计算机设备以及可读存储介质
CN116211315A (zh) * 2023-04-10 2023-06-06 济南大学 一种单导联心电信号辅助诊断方法及诊断终端

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112336310B (zh) * 2020-11-04 2024-03-08 吾征智能技术(北京)有限公司 一种基于fcbf和svm融合的心脏疾病诊断系统
CN113420624B (zh) * 2021-06-11 2022-04-26 华中师范大学 一种非接触式疲劳检测方法及系统
CN113413163B (zh) * 2021-08-24 2021-11-19 山东大学 一种混合深度学习和低差异度森林的心音诊断系统
CN114052692B (zh) * 2021-10-26 2024-01-16 珠海脉动时代健康科技有限公司 基于毫米波雷达的心率分析方法及设备
CN114995772B (zh) * 2022-08-08 2022-10-21 南京三百云信息科技有限公司 客户数据迁移存储方法及装置

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109567780B (zh) * 2018-12-07 2021-01-15 中国科学院电子学研究所 逐拍心率计算方法、装置、电子设备及存储介质
CN110148466B (zh) * 2019-05-15 2022-11-25 东北大学 一种基于迁移学习的心冲击信号房颤计算机辅助诊断方法

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115429284A (zh) * 2022-09-16 2022-12-06 山东科技大学 心电信号分类方法、系统、计算机设备以及可读存储介质
CN115429284B (zh) * 2022-09-16 2024-05-03 山东科技大学 心电信号分类方法、系统、计算机设备以及可读存储介质
CN116211315A (zh) * 2023-04-10 2023-06-06 济南大学 一种单导联心电信号辅助诊断方法及诊断终端
CN116211315B (zh) * 2023-04-10 2023-08-04 济南大学 一种单导联心电信号辅助诊断方法及诊断终端

Also Published As

Publication number Publication date
CN111329455A (zh) 2020-06-26
CN114010171A (zh) 2022-02-08
CN114010171B (zh) 2023-10-20

Similar Documents

Publication Publication Date Title
WO2021184410A2 (zh) 一种非接触式的心血管健康评估方法
Malik et al. Sleep-wake classification via quantifying heart rate variability by convolutional neural network
Oweis et al. QRS detection and heart rate variability analysis: A survey
Anusha et al. Electrodermal activity based pre-surgery stress detection using a wrist wearable
Sansone et al. Electrocardiogram pattern recognition and analysis based on artificial neural networks and support vector machines: a review
Belgacem et al. ECG based human authentication using wavelets and random forests
Saini et al. Classification of heart diseases from ECG signals using wavelet transform and kNN classifier
Mahajan et al. Cardiac rhythm classification from a short single lead ECG recording via random forest
CN105147248A (zh) 基于生理信息的抑郁症评估系统及其评估方法
CN107742534A (zh) 患者存活性预测系统
Henzel et al. Atrial fibrillation episodes detection based on classification of heart rate derived features
Xie et al. Feature enrichment based convolutional neural network for heartbeat classification from electrocardiogram
Akhter et al. Heart-based biometrics and possible use of heart rate variability in biometric recognition systems
Murugappan et al. Time domain features based sudden cardiac arrest prediction using machine learning algorithms
Yu et al. Automatic detection of atrial fibrillation from ballistocardiogram (BCG) using wavelet features and machine learning
Xie et al. Discrete Biorthogonal Wavelet Transform Based Convolutional Neural Network for Atrial Fibrillation Diagnosis from Electrocardiogram.
Rajput et al. Automated detection of hypertension using wavelet transform and nonlinear techniques with ballistocardiogram signals
Pirbhulal et al. Analysis of efficient biometric index using heart rate variability for remote monitoring of obstructive sleep apnea
Amiruddin et al. Performance evaluation of human identification based on ECG signal
Ye et al. Automatic sleep and wake classifier with heart rate and pulse oximetry: derived dynamic time warping features and logistic model
Wu et al. Stress detection using wearable devices based on transfer learning
Kandukuri et al. Constant Q-transform-based deep learning architecture for detection of obstructive sleep apnea
Jose Review on obstructive sleep apnea detection using ecg signals
Belgacem et al. ECG based human identification using random forests
Sundaram et al. Single lead ECG discrimination between normal sinus rhythm and sleep apnea with intrinsic mode function complexity index using empirical mode decomposition

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20925226

Country of ref document: EP

Kind code of ref document: A2

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20925226

Country of ref document: EP

Kind code of ref document: A2

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 03.07.2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20925226

Country of ref document: EP

Kind code of ref document: A2