技术领域technical field
本发明涉及一种便携式心电在线智能监测诊断系统设计方法，属于智能医疗技术领域。The invention relates to a design method of a portable electrocardiogram online intelligent monitoring and diagnosis system, and belongs to the technical field of intelligent medical treatment.
背景技术Background technique
心血管病是影响健康的重要杀手，2017年6月《中国心血管病报告2016》发布。报告指 出：目前，心血管病死亡占城乡居民总死亡原因的首位，且今后10年心血管病患病人数仍将 快速增长。此外，根据世界卫生组织的报告，到2030年大约要有2330万人死于心血管病。 面对这种趋势，对心血管病的早期诊断与预防显得尤为重要。Cardiovascular disease is an important killer affecting health. In June 2017, the "China Cardiovascular Disease Report 2016" was released. The report pointed out that: at present, cardiovascular disease death accounts for the first cause of death among urban and rural residents, and the number of cardiovascular disease patients will continue to grow rapidly in the next 10 years. In addition, according to the World Health Organization, approximately 23.3 million people will die of cardiovascular disease by 2030. Faced with this trend, the early diagnosis and prevention of cardiovascular disease is particularly important.
传统的诊断方法是患者到医院，医师对其进行心电图检查，进而给出诊断结果，任务繁 重且需要医师有丰富的临床经验和专业知识；而稀缺的医疗资源，很难满足数量庞大患者群 体的要求。为了提高就诊效率，方便性和快捷性，出现了自动化诊断技术，辅助医师进行诊 断。The traditional diagnosis method is that the patient goes to the hospital, the doctor performs an electrocardiogram examination on it, and then gives the diagnosis result. The task is heavy and requires the doctor to have rich clinical experience and professional knowledge; and the scarce medical resources, it is difficult to meet the needs of a large number of patients. Require. In order to improve the efficiency, convenience and speed of medical treatment, automatic diagnosis technology has emerged to assist doctors in diagnosis.
心率变异是指心动间期之间的时间变异数，其研究对象是心动间期而不是心率。人的心 率不是一成不变的，两次心搏之间存在着微小的时间差异，计算心动间期的差异，即可了解 心率变异性(Heart rate variability,HRV)。Heart rate variability refers to the temporal variability between cardiac intervals, which is studied on cardiac intervals rather than heart rate. Human heart rate is not static, there is a small time difference between two heartbeats, and the heart rate variability (HRV) can be understood by calculating the difference in the cardiac interval.
心率变异性可以评估心脏交感神经与副交感神经对心血管活动的影响，蕴含着心血管方 面的大量信息。临床研究表明，心率变异性的降低是心肌梗死、高血压、心绞痛等心血管疾 病发病的症状。因此，心率变异性的研究，在评价心血管系统功能、预测心血管疾病的发作， 以及为心血管疾病的早期诊断具有重要的意义。Heart rate variability can evaluate the effects of cardiac sympathetic and parasympathetic nerves on cardiovascular activity, and contains a lot of cardiovascular information. Clinical studies have shown that reduced heart rate variability is a symptom of cardiovascular disease such as myocardial infarction, hypertension, and angina pectoris. Therefore, the study of heart rate variability is of great significance in evaluating the function of the cardiovascular system, predicting the onset of cardiovascular disease, and early diagnosis of cardiovascular disease.
Poincare散点图是心率变异性一种重要的研究方法：通过使用连续的心搏间期在直角坐 标系中绘制图形，反映相邻心搏间期的变化，能显示心搏间期的特征；Poincare散点图有多 种形态，包括彗星状、扇形等，不同的形状反映不同的心脏状态。Poincare scatter plot is an important research method of heart rate variability: by using continuous heart beat intervals to draw a graph in a rectangular coordinate system, it can reflect the changes of adjacent heart beat intervals, and can display the characteristics of heart beat intervals; The Poincare scatter chart has various shapes, including comet shape, fan shape, etc. Different shapes reflect different heart states.
虽然Poincare散点图是一种有效的心率变异性分析方法，但是并不能体现其随时间变化 的趋势，对于某些心血管疾病、身体健康状况等不能很好的体现其心率变异性性质。于是， 一些学者提出了改进策略，即一阶差分散点图，通过相邻心搏间期的差值来绘制散点图。然 而，这种方法又丢失了原有的心搏间期绝对值信息；因此，又有学者将二者结合起来，提出 了一种RdR散点图，以此来同时反映心搏间期及其变化。Although the Poincare scatter plot is an effective HRV analysis method, it cannot reflect its trend over time, and cannot reflect the HRV nature of certain cardiovascular diseases and physical health conditions. Therefore, some scholars have proposed an improved strategy, that is, the first-order difference scatter plot, which draws a scatter plot through the difference between adjacent heartbeat intervals. However, this method loses the original absolute value of the heartbeat interval; therefore, some scholars combine the two and propose an RdR scatter plot to reflect the heartbeat interval and its value at the same time. Variety.
目前，对于不同心血管疾病的心率变异性分析很多；但是，并没有根据散点图来自动识 别、区分不同的心血管疾病，实现心电自动化诊断。At present, there are many heart rate variability analyses for different cardiovascular diseases; however, there is no automatic identification and differentiation of different cardiovascular diseases based on scatter plots to realize automatic ECG diagnosis.
发明内容SUMMARY OF THE INVENTION
针对上述现有技术存在的问题，本发明提供一种便携式心电在线智能监测诊断系统设计 方法，通过集成稀疏核主分量分析方法能够自动识别分类散点图，对心率变异性作分析，为 实现自动化诊断、缓解紧缺的医疗资源、减少医疗资源的浪费、提高就诊效率提供基础。In view of the problems existing in the above-mentioned prior art, the present invention provides a design method of a portable ECG online intelligent monitoring and diagnosis system, which can automatically identify the classification scattergram by integrating the sparse core principal component analysis method, and analyze the heart rate variability, in order to realize the It provides the basis for automatic diagnosis, alleviating the shortage of medical resources, reducing the waste of medical resources, and improving the efficiency of medical treatment.
为了实现上述目的，本发明提供一种便携式心电在线智能监测诊断系统设计方法，具体 步骤为：In order to achieve the above object, the present invention provides a kind of portable ECG online intelligent monitoring and diagnosis system design method, and the concrete steps are:
步骤1)采集心电信号，对心电信号进行滤波去噪处理，提取R波峰值位置；Step 1) collect the ECG signal, perform filtering and denoising processing on the ECG signal, and extract the peak position of the R wave;
步骤2)使用心搏间期绘制心电RdR散点图；Step 2) use the heartbeat interval to draw an ECG RdR scatter diagram;
步骤3)对心电RdR散点图进行缩放，缩放到同一规格，转成灰度图，并对图像数据进 行归一化处理，以减少计算量；Step 3) scaling the ECG RdR scatter diagram, scaling to the same specification, converting into grayscale, and normalizing the image data to reduce the amount of calculation;
步骤4)对获得的散点图样本进行标记；Step 4) marking the obtained scatter plot samples;
步骤5)对样本进行采样，随机抽取85％的数据作为训练样本；Step 5) Sampling the samples, randomly extracting 85% of the data as training samples;
步骤6)设置参数，参数包括近似基的误差参数、高斯核函数参数以及控制限的值；Step 6) setting parameters, the parameters include the error parameter of the approximate basis, the Gaussian kernel function parameter and the value of the control limit;
步骤7)分别对每类样本求解近似基、特征值、特征向量；Step 7) solve approximate basis, eigenvalue, eigenvector for each type of sample respectively;
步骤8)分别计算每个测试样本的SPE，其值与某类样本的SPE差值最小者为测试样本的 预测类别，与实际类别比较，计算准确率，若满足要求则步骤9)，否则返回步骤6)重新设 置参数训练；Step 8) Calculate the SPE of each test sample respectively, the smallest difference between its value and the SPE of a certain type of sample is the predicted category of the test sample, compare with the actual category, calculate the accuracy, if it meets the requirements, step 9), otherwise return Step 6) reset parameter training;
步骤9)获得分类模型参数，并将获得的分类模型用于心电诊断系统当中。Step 9) Obtain classification model parameters, and use the obtained classification model in the electrocardiographic diagnosis system.
进一步，采用集成稀疏核主分量分析方法用于对所述心电RdR散点图的分类识别。RdR 散点图是一种心率变异性分析方法，可以体现其随时间变化的趋势。Further, an integrated sparse kernel principal component analysis method is used to classify and identify the ECG RdR scattergram. The RdR scatter plot is a method of analyzing heart rate variability that can show its trend over time.
进一步，将集成稀疏核主分量分析方法用于对所述心电RdR散点图的分类识别，先建立 集成稀疏核主分量分析模型，然后通过平方预测误差法选择控制指标。Further, the integrated sparse kernel principal component analysis method is used for the classification and identification of the described ECG RdR scattergram, first establish the integrated sparse kernel principal component analysis model, and then select the control index by the squared prediction error method.
集成稀疏核主分量分析是一种无监督机器学习算法，集成稀疏核主分量分析的基本方法 如下：The integrated sparse kernel principal component analysis is an unsupervised machine learning algorithm. The basic method of the integrated sparse kernel principal component analysis is as follows:
主分量分析是一种典型的无监督算法，常用于解决原始空间的线性问题，而为了在特征 空间中用线性方法解决原始空间的非线性问题，B.Scholkopf等人提出了核主分量分析 (Kernel Principal Component Analysis,KPCA)。定义从原始空间Rn到特征空间F的非线性 映射:假如给定的样本X＝{x1,…,xN},xi∈Rn，则通过映射可以获得一组向量假设该组向量满足则特征空间中的相关阵为Principal component analysis is a typical unsupervised algorithm, which is often used to solve the linear problem of the original space. In order to solve the nonlinear problem of the original space with a linear method in the feature space, B. Scholkopf et al. proposed the kernel principal component analysis ( Kernel Principal Component Analysis, KPCA). Define a nonlinear mapping from the original space Rn to the feature space F: If a given sample X={x1 ,...,xN },xi ∈Rn , then by Mapping can get a set of vectors Suppose the set of vectors satisfies Then the correlation matrix in the feature space is
如果该组向量则可令可知满足条件， 代替式中的则KPCA问题可以转换为求特征空间中相关阵的特征值λ即特征向量If the set of vectors can make know Satisfy the condition, replace the Then the KPCA problem can be transformed into finding the correlation matrix in the feature space The eigenvalue λ of is the eigenvector
其中，是样本的线性组合，令ɑ＝[ɑ1,…,ɑN]T，则当不 能显式获得的时候，引入核函数，设首先需要计算：in, is a linear combination of samples, let ɑ＝[ɑ1 ,…,ɑN ]T , then when When it cannot be obtained explicitly, introduce a kernel function, set First you need to calculate:
K＝ΦTΦ；K = ΦT Φ;
其中，矩阵K是NxN的矩阵，也称核矩阵。则问题转换为：Among them, the matrix K is an NxN matrix, also called a kernel matrix. Then the problem translates to:
Kɑ＝Nλɑ；Kɑ=Nλɑ;
其中，ɑ＝[ɑ1,…,ɑN]。当中心化的过程可以直接在K上运算：Among them, ɑ=[ɑ1 ,…,ɑN ]. when The centralization process can operate directly on K:
其中，满足是一个NxN的1矩阵。假设得到的 特征值λ1≥λ2≥…λn及其对应的特征向量ɑ1,ɑ2,…,ɑN，在特征空间中的第k个特征向量ɑi,k表示的第k个特征向量的第i个值，由归一化得变量x在归一化之后的第k个特征向量方向的投影为第k个主分量，公式为：in, Satisfy is an NxN matrix of 1s. suppose to get The eigenvalues λ1≥λ2≥…λn and their corresponding eigenvectors ɑ1 ,ɑ2,…,ɑN , the kth eigenvector in the feature space ɑi,k means The i-th value of the k-th eigenvector of , given by normalized the k-th eigenvector of the variable x after normalization The projection of the direction is the k-th principal component, and the formula is:
本发明选择一种常用的核函数，径向基核函数来进行计算。当选择径向基核函数时，会 有明显的过学习问题，主要是由于径向基核函数对应的特征空间是无线维的，通过该方法得 出的主分量的次数与给定样本的维数无关，而与样本数量有关。为解决这个问题，引入了稀 疏化方法，即为稀疏核主分量分析(Sparse Kernel Principal ComponentAnalysis,SKPCA)。The present invention selects a commonly used kernel function, the radial basis kernel function, for calculation. When the radial basis kernel function is selected, there will be obvious over-learning problems, mainly because the feature space corresponding to the radial basis kernel function is wireless dimensional, and the number of principal components obtained by this method is related to the dimension of the given sample is not related to the number of samples, but to the number of samples. To solve this problem, a sparse method, namely Sparse Kernel Principal Component Analysis (SKPCA), is introduced.
在核主分量分析中，特征向量能用样本表示为也就 是说特征向量是全部样本的线性组合，这为特征向量的稀疏化提供了一种思路。通过近似 基求解算法，来求的近似基，从而得到的稀疏化方法。In Kernel Principal Component Analysis, the eigenvectors can be used as samples Expressed as That is, the feature vector is a linear combination of all samples, which provides an idea for the sparseness of feature vectors. Through the approximate basis solving algorithm, to find approximation basis, so that we get sparse method.
近似基求解的具体步骤是：The specific steps of approximate basis solution are:
A.建立集合为近似最大无关组，XA＝φ；A. Create a collection is an approximate maximum independent group, XA = φ;
B.对于k＝2,...,N的求极小值value；B. For k=2,...,N Find the minimum value;
C.如果得到极小值value≤ε，则把对应的元素加到XA当中，否则为Xl。ε是线性相关截尾 误差，在有限的样本中求无限维空间的基几乎没有稀疏性，因此选择近似计算。C. If the minimum value value≤ε is obtained, add the corresponding element to XA , otherwise it is Xl . ε is the linear correlation truncation error. There is almost no sparsity in finding the basis of infinite dimensional space in limited samples, so approximate calculation is chosen.
D.返回步骤B，直至完成所有的计算，计算结束。D. Return to step B until all calculations are completed, and the calculation ends.
其中，步骤B中的求解过程如下：Among them, the solution process in step B is as follows:
由Lagrange条件得-2K0+2Kλ＝0，其中K0＝(k(x1,xk),…,k(xl,xk))T，K是方阵， Kij＝k(xi,xj)。当核函数是高斯径向基核函数时，矩阵K正定，得λmin＝K-1K0即为f(λ)的极小 值点。Condition by Lagrange Get -2K0 +2Kλ=0, where K0 =(k(x1 ,xk ),...,k(xl ,xk ))T , K is a square matrix, Kij =k(xi ,xj ). When the kernel function is a Gaussian radial basis kernel function, the matrix K is positive definite, and λmin =K-1 K0 is the minimum value point of f(λ).
假设求得的一组近似基为用表示近似基构成的 基向量，特征向量可以表示为问题转成：Assuming that a set of approximate bases obtained is use Represents the basis vector formed by the approximate basis, and the eigenvector can be expressed as The problem turns into:
两边同乘记得Multiply both sides remember have to
可知是特征值与特征向量的问题，即令则It can be seen that it is a problem of eigenvalues and eigenvectors, that is, make but
推导得：Deduced:
其中，K(m,:)表示核矩阵K的第m行，K(:,n)表示第n列，1N＝[1,...,1]表示1行N列的行 向量。问题也就转换成了(KI)-1Ksα＝λα，为典型的特征值与特征向量问题。Among them, K(m,:) represents the mth row of the kernel matrix K, K(:,n) represents the nth column, and 1N =[1,...,1] represents a row vector of 1 row and N columns. The problem is transformed into (KI )-1 Ks α=λα, which is a typical eigenvalue and eigenvector problem.
虽然稀疏化可以控制学习机的学习能力，防止过学习，但是稀疏化方法仍然可能会丢失 数据集的某些重要特性，即欠学习问题。因此，有必要引入集成方法，称为集成稀疏核主分 量分析(Integration Sparse Kernel Principal Component Analysis,ISKPCA)。Although sparsification can control the learning ability of the learning machine and prevent over-learning, the sparsification method may still lose some important characteristics of the dataset, that is, the under-learning problem. Therefore, it is necessary to introduce an integration method, called Integrated Sparse Kernel Principal Component Analysis (ISKPCA).
由于核主分量分析是一种无监督学习，不能从外部获得指导，即不能对某次学习结果进 行奖赏或惩罚，所以，本文选择简单平均的方法。SKPCA的集成方法具体步骤流程如下：Since kernel principal component analysis is a kind of unsupervised learning, it cannot obtain guidance from the outside, that is, it cannot reward or punish a certain learning result, so this paper chooses the simple average method. The specific steps of the integration method of SKPCA are as follows:
A.设置重复的次数Re；A. Set the number of repetitions Re;
B.计算核矩阵K＝ΦTΦ；B. Calculate the kernel matrix K=ΦT Φ;
C.计算样本集的Re组近似基ΦI1,…,ΦIRe(在求解第k组近似基ΦIk的时候，前一组近似基ΦIk-1剩余的向量为第k组的优先向量)；C. Computational sample set The approximate basis of Re group ΦI1 ,...,ΦIRe (when solving the k-th group of approximate basis ΦIk , the remaining vector of the previous group of approximate basis ΦIk-1 is the priority vector of the k-th group);
D.根据上述的SKPCA方法，对各个组的近似基ΦIk求特征值λk，特征向量ɑk，共获得Re 组特征值λ1,…,λRe和特征向量ɑ1,…,ɑRe；D. According to the above SKPCA method, calculate the eigenvalue λk and the eigenvector ɑk for the approximate basis ΦIk of each group, and obtain the eigenvalues λ1 ,…,λRe and the eigenvectors ɑ1 ,…,ɑ of the Re group.Re ;
E.对前n个特征值λ1≥…≥λn和对应的特征向量，分别求集成特征值集 成特征向量其中λik是第k组近似基ΦIk所张成的子空间为解空间的求得的第i 个特征值，是对应的特征向量；E. For the first n eigenvalues λ1 ≥...≥λn and the corresponding eigenvectors, find the integrated eigenvalues respectively integrated feature vector where λik is the i-th eigenvalue obtained when the subspace spanned by the k-th approximate basis ΦIk is the solution space, is the corresponding eigenvector;
F.对集成特征值和集成特征向量进行归一化，即得集成稀疏核主分量分析的特征值 和特征向量。F. On the integrated eigenvalues and the integrated feature vector Normalization is performed to obtain the eigenvalues and eigenvectors of the integrated sparse kernel principal component analysis.
将集成稀疏核主分量分析方法用于心电RdR散点图的分类识别，需要先建立集成稀疏核 主分量分析模型，然后选择控制指标。本发明选择一种常用的控制指标，即平方预测误差 (Squared Prediction Error,SPE)。To use the integrated sparse kernel principal component analysis method for the classification and identification of the ECG RdR scattergram, it is necessary to first establish the integrated sparse kernel principal component analysis model, and then select the control indicators. The present invention selects a commonly used control index, that is, the squared prediction error (Squared Prediction Error, SPE).
平方预测误差的具体方法：The specific method of squared prediction error:
对于第i个样本Xi，假设经过SKPCA求得其前n个主分量为t1,…,tn，对应的特征值为 λ1,…,λn。用表示特征空间N个主成分的重构向量同理则 样本X的SPE定义为：For the i-th sample Xi , it is assumed that the first n principal components are obtained by SKPCA as t1 ,...,tn , and the corresponding eigenvalues are λ1 ,...,λn . use Reconstruction vector representing the N principal components of the feature space Similarly Then the SPE of sample X is defined as:
其中，in,
与现有技术相比，本发明的有益效果如下：Compared with the prior art, the beneficial effects of the present invention are as follows:
本发明利用集成稀疏核主分量分析方法，通过计算样本数据与使用核主分量分析映射数 据之间的差值来研究数据之间的最大相关性，并以此来判断心电数据类别，研究自动识别分 类RdR散点图。使用本发明方法来对心率变异性作分析，可以为实现自动化诊断、缓解紧缺 的医疗资源、减少医疗资源的浪费、提高就诊效率提供一些基础。The invention uses the integrated sparse kernel principal component analysis method to study the maximum correlation between the data by calculating the difference between the sample data and the mapping data using the kernel principal component analysis, and judges the ECG data category based on this. Identify categorical RdR scatterplots. Using the method of the present invention to analyze the heart rate variability can provide some basis for realizing automatic diagnosis, alleviating the shortage of medical resources, reducing the waste of medical resources, and improving the efficiency of medical treatment.
附图说明Description of drawings
图1为本发明具体方法流程图；Fig. 1 is the specific method flow chart of the present invention;
图2为系统结构框图；Figure 2 is a block diagram of the system structure;
图3为RdR散点图。Figure 3 is an RdR scatter plot.
图4为随机选择测试样本与各类的SPE差值；Figure 4 shows the SPE difference between randomly selected test samples and various types;
图5为用测试数据测试分类器性能的部分测试结果图。Figure 5 is a partial test result graph of testing classifier performance with test data.
具体实施方式Detailed ways
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描 述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明 中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例， 都属于本发明保护的范围。The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.
如图1所示，本发明提供的一种便携式心电在线智能监测诊断系统设计方法，具体步骤 为：As shown in Figure 1, a kind of portable ECG online intelligent monitoring and diagnosis system design method provided by the invention, concrete steps are:
步骤1)采集心电信号，对心电信号进行滤波去噪处理，提取R波峰值位置；Step 1) collect the ECG signal, perform filtering and denoising processing on the ECG signal, and extract the peak position of the R wave;
步骤2)使用心搏间期绘制心电RdR散点图；Step 2) use the heartbeat interval to draw an ECG RdR scatter diagram;
步骤3)对心电RdR散点图进行缩放，缩放到同一规格，转成灰度图，并对图像数据进 行归一化处理，以减少计算量；Step 3) scaling the ECG RdR scatter diagram, scaling to the same specification, converting into grayscale, and normalizing the image data to reduce the amount of calculation;
步骤4)对获得的散点图样本进行标记；Step 4) marking the obtained scatter plot samples;
步骤5)对样本进行采样，随机抽取85％的数据作为训练样本；Step 5) Sampling the samples, randomly extracting 85% of the data as training samples;
步骤6)设置参数，参数包括近似基的误差参数、高斯核函数参数以及控制限的值；Step 6) setting parameters, the parameters include the error parameter of the approximate basis, the Gaussian kernel function parameter and the value of the control limit;
步骤7)分别对每类样本求解近似基、特征值、特征向量；Step 7) solve approximate basis, eigenvalue, eigenvector for each type of sample respectively;
步骤8)分别计算每个测试样本的SPE，其值与某类样本的SPE差值最小者为测试样本的 预测类别，与实际类别比较，计算准确率，若满足要求则步骤9)，否则返回步骤6)重新设 置参数训练；Step 8) Calculate the SPE of each test sample respectively, the smallest difference between its value and the SPE of a certain type of sample is the predicted category of the test sample, compare with the actual category, calculate the accuracy, if it meets the requirements, step 9), otherwise return Step 6) reset parameter training;
步骤9)获得分类模型参数，并将获得的分类模型用于心电诊断系统当中。Step 9) Obtain classification model parameters, and use the obtained classification model in the electrocardiographic diagnosis system.
进一步，采用集成稀疏核主分量分析方法用于对所述心电RdR散点图的分类识别。RdR 散点图是一种心率变异性分析方法，可以体现其随时间变化的趋势。Further, an integrated sparse kernel principal component analysis method is used to classify and identify the ECG RdR scattergram. The RdR scatter plot is a method of analyzing heart rate variability that can show its trend over time.
进一步，将集成稀疏核主分量分析方法用于对所述心电RdR散点图的分类识别，先建立 集成稀疏核主分量分析模型，然后通过平方预测误差法选择控制指标。Further, the integrated sparse kernel principal component analysis method is used for the classification and identification of the described ECG RdR scattergram, first establish the integrated sparse kernel principal component analysis model, and then select the control index by the squared prediction error method.
集成稀疏核主分量分析是一种无监督机器学习算法，集成稀疏核主分量分析的基本方法 如下：The integrated sparse kernel principal component analysis is an unsupervised machine learning algorithm. The basic method of the integrated sparse kernel principal component analysis is as follows:
主分量分析是一种典型的无监督算法，常用于解决原始空间的线性问题，而为了在特征 空间中用线性方法解决原始空间的非线性问题，B.Scholkopf等人提出了核主分量分析 (Kernel Principal Component Analysis,KPCA)。定义从原始空间Rn到特征空间F的非线性 映射:假如给定的样本X＝{x1,…,xN},xi∈Rn，则通过映射可以获得一组向量假设该组向量满足则特征空间中的相关阵为Principal component analysis is a typical unsupervised algorithm, which is often used to solve the linear problem of the original space. In order to solve the nonlinear problem of the original space with a linear method in the feature space, B. Scholkopf et al. proposed the kernel principal component analysis ( Kernel Principal Component Analysis, KPCA). Define a nonlinear mapping from the original space Rn to the feature space F: If a given sample X={x1 ,...,xN },xi ∈Rn , then by Mapping can get a set of vectors Suppose the set of vectors satisfies Then the correlation matrix in the feature space is
如果该组向量则可令可知满足条件， 代替式中的则KPCA问题可以转换为求特征空间中相关阵的特征值λ即特征向量If the set of vectors can make know Satisfy the condition, replace the Then the KPCA problem can be transformed into finding the correlation matrix in the feature space The eigenvalue λ of is the eigenvector
其中，是样本的线性组合，令ɑ＝[ɑ1,…,ɑN]T，则当不 能显式获得的时候，引入核函数，设首先需要计算：in, is a linear combination of samples, let ɑ＝[ɑ1 ,…,ɑN ]T , then when When it cannot be obtained explicitly, introduce a kernel function, set First you need to calculate:
K＝ΦTΦ；K = ΦT Φ;
其中，矩阵K是NxN的矩阵，也称核矩阵。则问题转换为：Among them, the matrix K is an NxN matrix, also called a kernel matrix. Then the problem translates to:
Kɑ＝Nλɑ；Kɑ=Nλɑ;
其中，ɑ＝[ɑ1,…,ɑN]。当中心化的过程可以直接在K上运算：Among them, ɑ=[ɑ1 ,…,ɑN ]. when The centralization process can operate directly on K:
其中，满足是一个NxN的1矩阵。假设得到的 特征值λ1≥λ2≥…λn及其对应的特征向量ɑ1,ɑ2,…,ɑN，在特征空间中的第k个特征向量ɑi,k表示的第k个特征向量的第i个值，由归一化得变量x在归一化之后的第k个特征向量方向的投影为第k个主分量， 公式为：in, Satisfy is an NxN matrix of 1s. suppose to get The eigenvalues λ1≥λ2≥…λn and their corresponding eigenvectors ɑ1 ,ɑ2 ,…,ɑN , the kth eigenvector in the feature space ɑi,k means The i-th value of the k-th eigenvector of , given by normalized the k-th eigenvector of the variable x after normalization The projection of the direction is the k-th principal component, and the formula is:
本发明选择一种常用的核函数，径向基核函数来进行计算。当选择径向基核函数时，会 有明显的过学习问题，主要是由于径向基核函数对应的特征空间是无线维的，通过该方法得 出的主分量的次数与给定样本的维数无关，而与样本数量有关。为解决这个问题，引入了稀 疏化方法，即为稀疏核主分量分析(Sparse Kernel Principal ComponentAnalysis,SKPCA)。The present invention selects a commonly used kernel function, the radial basis kernel function, for calculation. When the radial basis kernel function is selected, there will be obvious over-learning problems, mainly because the feature space corresponding to the radial basis kernel function is wireless dimensional, and the number of principal components obtained by this method is related to the dimension of the given sample is not related to the number of samples, but to the number of samples. To solve this problem, a sparse method, namely Sparse Kernel Principal Component Analysis (SKPCA), is introduced.
在核主分量分析中，特征向量能用样本表示为也就 是说特征向量是全部样本的线性组合，这为特征向量的稀疏化提供了一种思路。通过近似 基求解算法，来求的近似基，从而得到的稀疏化方法。In Kernel Principal Component Analysis, the eigenvectors can be used as samples Expressed as That is, the feature vector is a linear combination of all samples, which provides an idea for the sparseness of feature vectors. Through the approximate basis solving algorithm, to find approximation basis, so that we get sparse method.
近似基求解的具体步骤是：The specific steps of approximate basis solution are:
A.建立集合为近似最大无关组，XA＝φ；A. Create a collection is an approximate maximum independent group, XA = φ;
B.对于k＝2,...,N的求极小值value；B. For k=2,...,N Find the minimum value;
其中，in,
C.如果得到极小值value≤ε，则把对应的元素加到XA当中，否则为Xl。ε是线性相关截尾 误差，在有限的样本中求无限维空间的基几乎没有稀疏性，因此选择近似计算。C. If the minimum value value≤ε is obtained, add the corresponding element to XA , otherwise it is Xl . ε is the linear correlation truncation error. There is almost no sparsity in finding the basis of infinite dimensional space in limited samples, so approximate calculation is chosen.
D.返回步骤B，直至完成所有的计算，计算结束。D. Return to step B until all calculations are completed, and the calculation ends.
其中，步骤B中的求解过程如下：Among them, the solution process in step B is as follows:
令make
由Lagrange条件得-2K0+2Kλ＝0，其中K0＝(k(x1,xk),…,k(xl,xk))T，K是方阵， Kij＝k(xi,xj)。当核函数是高斯径向基核函数时，矩阵K正定，得λmin＝K-1K0即为f(λ)的极小 值点。Condition by Lagrange Get -2K0 +2Kλ=0, where K0 =(k(x1 ,xk ),...,k(xl ,xk ))T , K is a square matrix, Kij =k(xi ,xj ). When the kernel function is a Gaussian radial basis kernel function, the matrix K is positive definite, and λmin =K-1 K0 is the minimum value point of f(λ).
假设求得的一组近似基为用表示近似基构成的 基向量，特征向量可以表示为问题转成：Assuming that a set of approximate bases obtained is use Represents the basis vector formed by the approximate basis, and the eigenvector can be expressed as The problem turns into:
两边同乘记得Multiply both sides remember have to
可知是特征值与特征向量的问题，即令则It can be seen that it is a problem of eigenvalues and eigenvectors, that is, make but
推导得：Deduced:
其中，K(m,:)表示核矩阵K的第m行，K(:,n)表示第n列，1N＝[1,...,1]表示1行N列的行 向量。问题也就转换成了(KI)-1Ksα＝λα，为典型的特征值与特征向量问题。Among them, K(m,:) represents the mth row of the kernel matrix K, K(:,n) represents the nth column, and 1N =[1,...,1] represents a row vector of 1 row and N columns. The problem is transformed into (KI )-1 Ks α=λα, which is a typical eigenvalue and eigenvector problem.
虽然稀疏化可以控制学习机的学习能力，防止过学习，但是稀疏化方法仍然可能会丢失 数据集的某些重要特性，即欠学习问题。因此，有必要引入集成方法，称为集成稀疏核主分 量分析(Integration Sparse Kernel Principal Component Analysis,ISKPCA)。Although sparsification can control the learning ability of the learning machine and prevent over-learning, the sparsification method may still lose some important characteristics of the dataset, that is, the under-learning problem. Therefore, it is necessary to introduce an integration method, called Integrated Sparse Kernel Principal Component Analysis (ISKPCA).
由于核主分量分析是一种无监督学习，不能从外部获得指导，即不能对某次学习结果进 行奖赏或惩罚，所以，本文选择简单平均的方法。SKPCA的集成方法具体步骤流程如下：Since kernel principal component analysis is a kind of unsupervised learning, it cannot obtain guidance from the outside, that is, it cannot reward or punish a certain learning result, so this paper chooses the simple average method. The specific steps of the integration method of SKPCA are as follows:
A.设置重复的次数Re；A. Set the number of repetitions Re;
B.计算核矩阵K＝ΦTΦ；B. Calculate the kernel matrix K=ΦT Φ;
C.计算样本集的Re组近似基ΦI1,…,ΦIRe(在求解第k组近似基ΦIk的时候，前一组近似基ΦIk-1剩余的向量为第k组的优先向量)；C. Computational sample set The approximate basis of Re group ΦI1 ,...,ΦIRe (when solving the k-th group of approximate basis ΦIk , the remaining vector of the previous group of approximate basis ΦIk-1 is the priority vector of the k-th group);
D.根据上述的SKPCA方法，对各个组的近似基ΦIk求特征值λk，特征向量ɑk，共获得Re 组特征值λ1,…,λRe和特征向量ɑ1,…,ɑRe；D. According to the above SKPCA method, calculate the eigenvalue λk and the eigenvector ɑk for the approximate basis ΦIk of each group, and obtain the eigenvalues λ1 ,…,λRe and the eigenvectors ɑ1 ,…,ɑ of the Re group.Re ;
E.对前n个特征值λ1≥…≥λn和对应的特征向量，分别求集成特征值集 成特征向量其中λik是第k组近似基ΦIk所张成的子空间为解空间的求得的第i 个特征值，是对应的特征向量；E. For the first n eigenvalues λ1 ≥...≥λn and the corresponding eigenvectors, find the integrated eigenvalues respectively integrated feature vector where λik is the i-th eigenvalue obtained when the subspace spanned by the k-th approximate basis ΦIk is the solution space, is the corresponding eigenvector;
F.对集成特征值和集成特征向量进行归一化，即得集成稀疏核主分量分析的特征值 和特征向量。F. On the integrated eigenvalues and the integrated feature vector Normalization is performed to obtain the eigenvalues and eigenvectors of the integrated sparse kernel principal component analysis.
将集成稀疏核主分量分析方法用于心电RdR散点图的分类识别，需要先建立集成稀疏核 主分量分析模型，然后选择控制指标。本发明选择一种常用的控制指标，即平方预测误差 (Squared Prediction Error,SPE)。To use the integrated sparse kernel principal component analysis method for the classification and identification of the ECG RdR scattergram, it is necessary to first establish the integrated sparse kernel principal component analysis model, and then select the control indicators. The present invention selects a commonly used control index, that is, the squared prediction error (Squared Prediction Error, SPE).
平方预测误差的具体方法：The specific method of squared prediction error:
对于第i个样本Xi，假设经过SKPCA求得其前n个主分量为t1,…,tn，对应的特征值为 λ1,…,λn。用表示特征空间N个主成分的重构向量同理则 样本X的SPE定义为：For the i-th sample Xi , it is assumed that the first n principal components are obtained by SKPCA as t1 ,...,tn , and the corresponding eigenvalues are λ1 ,...,λn . use Reconstruction vector representing the N principal components of the feature space Similarly Then the SPE of sample X is defined as:
其中，in,
为便于叙述，使用MIT-BIH数据库中的心电数据，以此数据库的数据作示意性分析介绍。For the convenience of description, the ECG data in the MIT-BIH database is used, and the data in this database is used as a schematic analysis and introduction.
图2是系统结构框图，该系统的硬件主要由采集电路、微控制器电路、存储电路、RAM 电路、JTAG电路、通信电路、警报电路、LED运行指示电路、电源管理电路和云服务器等组 成。为了降低系统的复杂度，本系统在远程监测终端部分选择B/S(浏览器/服务器)模式来 开发。本系统通过互联网连接成一个整体，且为符合便携式特点，采用单导联设计。采集电 路负责采集心电信号，在微处理器上进行数据滤波，然后LCD显示波形、心率等信息，可以 利用通信电路将数据上传至云服务器，服务器端接收到数据，对数据进行处理，并得出诊断 分析报告，根据诊断分析结果，判断是否需要发送警报信息等操作。医生也可以通过浏览器 查看用户心电数据，并给出诊断意见。Figure 2 is a block diagram of the system structure. The hardware of the system is mainly composed of acquisition circuit, microcontroller circuit, storage circuit, RAM circuit, JTAG circuit, communication circuit, alarm circuit, LED operation indication circuit, power management circuit and cloud server. In order to reduce the complexity of the system, the system selects B/S (browser/server) mode in the remote monitoring terminal part to develop. The system is connected as a whole through the Internet, and in order to meet the portable characteristics, it adopts a single-lead design. The acquisition circuit is responsible for collecting the ECG signal, filtering the data on the microprocessor, and then displaying the waveform, heart rate and other information on the LCD. The communication circuit can be used to upload the data to the cloud server. The server receives the data, processes the data, and obtains the result. A diagnostic analysis report is generated, and according to the diagnostic analysis results, it is determined whether it is necessary to send alarm information and other operations. Doctors can also view the user's ECG data through the browser and give diagnostic opinions.
图3是通过MATLAB来绘制MIT-BIH心电数据的RdR散点图示意图。RdR散点图可以有效 反映心搏间期随时间的变化趋势，蕴含大量的临床信息，能显示不同的心搏特征。实验显示， 使用本发明方法来对其进行识别分类，效果较好。Figure 3 is a schematic diagram of the RdR scatter plot of MIT-BIH ECG data drawn by MATLAB. The RdR scatter diagram can effectively reflect the change trend of the heartbeat interval with time, which contains a lot of clinical information and can display different heartbeat characteristics. Experiments show that using the method of the present invention to identify and classify it has a better effect.
图4是从测试样本中随机选择某一类的心电数据样本，如图是选择到实际类别为“3”的 一些测试样本集，将其分别与各个类别计算其SPE的差值，结果如图(4)，可以看到其与类别 为“3”的SPE差值最小，分类全部正确；Figure 4 is a random selection of a certain type of ECG data samples from the test samples. The figure is to select some test sample sets whose actual category is "3", and calculate the difference of their SPE with each category. The results are as follows As shown in Figure (4), it can be seen that the difference between it and the SPE of category "3" is the smallest, and the classification is all correct;
图5是依据本发明方法，部分分类的结果展示图，实验显示，准确率较高，图中，参数 true是实际的样本类别，predicted是分类器预测的类别，图中显示的图像是RdR散点图， 可以看到分类结果全部正确。Fig. 5 is a graph showing the results of partial classification according to the method of the present invention. The experiment shows that the accuracy rate is high. In the graph, the parameter true is the actual sample category, predicted is the category predicted by the classifier, and the image shown in the graph is the RdR dispersion Click on the graph, you can see that the classification results are all correct.
综上所述，本发明利用集成稀疏核主分量分析方法，通过计算样本数据与使用核主分量 分析映射数据之间的差值来研究数据之间的最大相关性，并以此来判断心电数据类别，研究 自动识别分类RdR散点图。使用本发明方法来对心率变异性作分析，可以为实现自动化诊断、 缓解紧缺的医疗资源、减少医疗资源的浪费、提高就诊效率提供一些基础。To sum up, the present invention uses the integrated sparse kernel principal component analysis method to study the maximum correlation between the data by calculating the difference between the sample data and the mapping data using the kernel principal component analysis, and judges the ECG based on this. Data categories, study automatic identification of categorical RdR scatterplots. Using the method of the present invention to analyze the heart rate variability can provide some basis for realizing automatic diagnosis, alleviating the shortage of medical resources, reducing the waste of medical resources, and improving the efficiency of medical treatment.
以上所述仅是本发明的优选实施方式，本发明的保护范围并不仅局限于上述实施例，凡 属于本发明思路下的技术方案均属于本发明的保护范围。应当指出，对于本技术领域的普通 技术人员来说，在不脱离本发明原理前提下的若干改进和润饰，这些改进和润饰也应视为本 发明的保护范围。The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above-mentioned embodiments, and all technical solutions that belong to the idea of the present invention belong to the protection scope of the present invention. It should be pointed out that for those skilled in the art, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810921525.7ACN109157211A (en) | 2018-08-14 | 2018-08-14 | A kind of portable cardiac on-line intelligence monitoring diagnosis system design method |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810921525.7ACN109157211A (en) | 2018-08-14 | 2018-08-14 | A kind of portable cardiac on-line intelligence monitoring diagnosis system design method |
| Publication Number | Publication Date |
|---|---|
| CN109157211Atrue CN109157211A (en) | 2019-01-08 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810921525.7APendingCN109157211A (en) | 2018-08-14 | 2018-08-14 | A kind of portable cardiac on-line intelligence monitoring diagnosis system design method |
| Country | Link |
|---|---|
| CN (1) | CN109157211A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110403600A (en)* | 2019-07-26 | 2019-11-05 | 武汉海星通技术股份有限公司 | Intelligent analysis method and system for paroxysmal atrial fibrillation based on difference time scattergram |
| CN111920397A (en)* | 2020-08-07 | 2020-11-13 | 江苏师范大学 | Arteriosclerosis degree detection method based on sparse least square support vector machine |
| CN114496209A (en)* | 2022-02-18 | 2022-05-13 | 青岛市中心血站 | Blood donation intelligent decision method and system |
| CN115969384A (en)* | 2022-11-10 | 2023-04-18 | 江苏科技大学 | A Fetal ECG Detection Method Based on Kernel Principal Component Analysis |
| CN116261424A (en)* | 2020-12-30 | 2023-06-13 | 三星电子株式会社 | Electronic device and electronic device control method |
| CN117373036A (en)* | 2023-10-24 | 2024-01-09 | 东南大学附属中大医院 | Data analysis processing method based on intelligent AI |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106073755A (en)* | 2016-05-27 | 2016-11-09 | 成都信汇聚源科技有限公司 | The implementation method that in a kind of miniature holter devices, atrial fibrillation identifies automatically |
| CN106725426A (en)* | 2016-12-14 | 2017-05-31 | 深圳先进技术研究院 | A kind of method and system of electrocardiosignal classification |
| CN107292292A (en)* | 2017-07-20 | 2017-10-24 | 浙江好络维医疗技术有限公司 | A kind of QRS complex sorting technique based on SVMs |
| CN206792400U (en)* | 2017-01-16 | 2017-12-26 | 吉林东华原医疗设备有限责任公司 | HRV detection means |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106073755A (en)* | 2016-05-27 | 2016-11-09 | 成都信汇聚源科技有限公司 | The implementation method that in a kind of miniature holter devices, atrial fibrillation identifies automatically |
| CN106725426A (en)* | 2016-12-14 | 2017-05-31 | 深圳先进技术研究院 | A kind of method and system of electrocardiosignal classification |
| CN206792400U (en)* | 2017-01-16 | 2017-12-26 | 吉林东华原医疗设备有限责任公司 | HRV detection means |
| CN107292292A (en)* | 2017-07-20 | 2017-10-24 | 浙江好络维医疗技术有限公司 | A kind of QRS complex sorting technique based on SVMs |
| Title |
|---|
| 甘良志: "核学习算法与集成方法研究", 《中国博士学位论文全文数据库(信息科技辑)》* |
| 陆宏伟等: "基于ＲｄＲ新型散点图心率变异性研究", 《生物医学工程学杂志》* |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110403600A (en)* | 2019-07-26 | 2019-11-05 | 武汉海星通技术股份有限公司 | Intelligent analysis method and system for paroxysmal atrial fibrillation based on difference time scattergram |
| CN110403600B (en)* | 2019-07-26 | 2022-02-08 | 武汉海星通技术股份有限公司 | Intelligent analysis method and system for paroxysmal atrial fibrillation based on difference time scatter diagram |
| CN111920397A (en)* | 2020-08-07 | 2020-11-13 | 江苏师范大学 | Arteriosclerosis degree detection method based on sparse least square support vector machine |
| CN116261424A (en)* | 2020-12-30 | 2023-06-13 | 三星电子株式会社 | Electronic device and electronic device control method |
| US12393642B2 (en) | 2020-12-30 | 2025-08-19 | Samsung Electronics Co., Ltd. | Electronic devices and controlling method of the same |
| CN114496209A (en)* | 2022-02-18 | 2022-05-13 | 青岛市中心血站 | Blood donation intelligent decision method and system |
| CN114496209B (en)* | 2022-02-18 | 2022-09-27 | 青岛市中心血站 | Intelligent decision-making method and system for blood donation |
| CN115969384A (en)* | 2022-11-10 | 2023-04-18 | 江苏科技大学 | A Fetal ECG Detection Method Based on Kernel Principal Component Analysis |
| CN117373036A (en)* | 2023-10-24 | 2024-01-09 | 东南大学附属中大医院 | Data analysis processing method based on intelligent AI |
| CN117373036B (en)* | 2023-10-24 | 2024-06-11 | 东南大学附属中大医院 | Data analysis and processing method based on intelligent AI |
| Publication | Publication Date | Title |
|---|---|---|
| CN109157211A (en) | A kind of portable cardiac on-line intelligence monitoring diagnosis system design method | |
| Sridhar et al. | Accurate detection of myocardial infarction using non linear features with ECG signals | |
| Yıldırım et al. | Arrhythmia detection using deep convolutional neural network with long duration ECG signals | |
| Zhao et al. | ECG classification using deep CNN improved by wavelet transform | |
| Dhyani et al. | Analysis of ECG-based arrhythmia detection system using machine learning | |
| Tadesse et al. | DeepMI: Deep multi-lead ECG fusion for identifying myocardial infarction and its occurrence-time | |
| Gan et al. | Parallel classification model of arrhythmia based on DenseNet-BiLSTM | |
| CN114512239B (en) | Cerebral apoplexy risk prediction method and system based on transfer learning | |
| CN114521900B (en) | Arrhythmia classification and identification method based on transfer learning | |
| Prakash et al. | A system for automatic cardiac arrhythmia recognition using electrocardiogram signal | |
| CN111564218A (en) | A cardiovascular disease risk monitoring system based on big data | |
| Geng et al. | Identification of major depression patients using machine learning models based on heart rate variability during sleep stages for pre-hospital screening | |
| Jiang et al. | Visualization deep learning model for automatic arrhythmias classification | |
| Khan et al. | A novel hybrid GRU-CNN and residual bias (RB) based RB-GRU-CNN models for prediction of PTB Diagnostic ECG time series data | |
| Yu et al. | Spa-Tem MI: A spatial–temporal network for detecting and locating myocardial infarction | |
| Shen et al. | Risk prediction for cardiovascular disease using ECG data in the China Kadoorie Biobank | |
| Philip et al. | Identifying arrhythmias based on ecg classification using enhanced-PCA and enhanced-SVM methods | |
| Sun et al. | A dynamic learning-based ECG feature extraction method for myocardial infarction detection | |
| Liu et al. | Automatic classification of arrhythmias using multi-branch convolutional neural networks based on channel-based attention and bidirectional LSTM | |
| Wang et al. | Multiscale residual network based on channel spatial attention mechanism for multilabel ECG classification | |
| He et al. | Localization of myocardial infarction using a multi-branch weight sharing network based on 2-D vectorcardiogram | |
| Raghukumar et al. | Enhancing Myocardial Infarction Diagnosis: Insights from ECG Image Analysis and Machine Learning | |
| Nath et al. | Quantum annealing for automated feature selection in stress detection | |
| TWI688371B (en) | Intelligent device for atrial fibrillation signal pattern acquisition and auxiliary diagnosis | |
| Kanna et al. | A multi-branch multi-scale convolutional neural network using automatic detection of fetal arrhythmia |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | Application publication date:20190108 | |
| WD01 | Invention patent application deemed withdrawn after publication |