阅读说明：本技术 一种基于高斯混合模型的心电信号识别算法 (Electrocardiosignal recognition algorithm based on Gaussian mixture model ) 是由 陈少杰 于 2020-05-18 设计创作，主要内容包括：本发明涉及一种基于高斯混合模型心电信号识别算法,该方法具体过程是：首先确定读取原始心电图像,应用心电提取算法获取连续周期心电信号,然后根据心电数据点导数(包括一阶和二阶导数),对周期心电信号进行分割,提取单个周期的心电数据。并利用插值法,实现对缺失数据进行平滑。再此基础上,将相应心电数据进行分组,并通过训练集上的心电数据进行训练。训练过程中,根据高斯混合模型的最优解的数字特征,通过先对模型参数进行初始化,而后依据二阶平方差最小,获取隐函数的Q表达,继而根据相应的公式,对原始参数进行相应的调整,重复上述过程,使得模型参数的逐步向最优值靠近。当误差在一定范围后,停止训练,并使用交叉验证集上的数据,对模型参数进行调优。最终获得稳定的模型参数。该算法方法不仅快速、高效,而且能够有效克服局部心电图因扭曲、噪声等因素造成的失真,在心电信号自动分析过程中具有重要的应用价值。(The invention relates to a Gaussian-mixed-model-based electrocardiosignal recognition algorithm, which comprises the following specific steps: firstly, determining to read an original electrocardio image, obtaining continuous periodic electrocardio signals by applying an electrocardio extraction algorithm, then segmenting the periodic electrocardio signals according to electrocardio data point derivatives (including first-order derivatives and second-order derivatives), and extracting single-period electrocardio data. And the missing data is smoothed by an interpolation method. And on the basis, grouping the corresponding electrocardiogram data, and training by the electrocardiogram data on the training set. In the training process, according to the digital characteristics of the optimal solution of the Gaussian mixture model, initializing the model parameters, then obtaining Q expression of the implicit function according to the minimum second-order square difference, then correspondingly adjusting the original parameters according to a corresponding formula, and repeating the process to enable the model parameters to gradually approach to the optimal value. And when the error is within a certain range, stopping training, and using the data on the cross validation set to adjust and optimize the model parameters. Finally, stable model parameters are obtained. The algorithm method is fast and efficient, can effectively overcome distortion of local electrocardiograms caused by factors such as distortion, noise and the like, and has important application value in the automatic analysis process of the electrocardiosignals.)

1. A Gaussian mixture model-based electrocardiosignal recognition algorithm is characterized by comprising the following steps:

step A: reading an original electrocardio image, and obtaining continuous periodic electrocardio signals by applying an electrocardio extraction algorithm.

And B: according to the electrocardio data point derivatives (including first and second derivatives), the periodic electrocardio signals are segmented, and the electrocardio data of a single period are extracted.

And C: and aiming at the missing condition of adjacent data points, the missing electrocardio data are supplemented by adopting an interpolation mode by combining the electrocardio data on adjacent positions in the same period, so that the completeness of the electrocardio period is ensured.

Step D: and randomly grouping a plurality of complete electrocardiographic cycle data to respectively obtain a training group, a cross validation group and test group data.

Step E: initializing a Gaussian mixture model, then training by using electrocardiosignals of a training group, then correcting a training result by using cross validation group data, and repeating the steps until the error is within a specified range.

Step F: and testing the test group data by using the trained Gaussian mixture model to verify the effectiveness of the algorithm.

2. The gaussian mixed model-based cardiac signal recognition algorithm of claim 1, wherein the training and cross-validation process operates according to the following equation:

wherein, formula (1) represents a probability density function obeying Gaussian distribution, and formula (2) represents a Gaussian mixture model composed of K Gaussian mixture distributions. Equation (3, 4) indicates that given a set of N data X ═ X₁，x₂，...，x_N]And satisfies the probability of occurrence under the i.i.d. condition. Wherein the hidden variable (Z) is used in formula (4) instead of the Gaussian coefficient in formula (3). Equation (5) represents the likelihood function for this function under different parameters θ. Equations (6, 7) represent the optimal solution of the gaussian distribution parameters obtained after using an optimization algorithm given a set of training data. Equation (8, 9) represents the posterior distribution Q of the hidden variable z, and the solution of the objective function- -P distribution- -the parametric representation of the normal distribution.

3. The algorithm for recognizing the electrocardiosignal based on the Gaussian mixed model as claimed in claim 1, wherein in the whole training algorithm process, the Q distribution cannot be solved through the traditional maximum likelihood estimation, and finally, the solution of an objective function cannot be solved. In the algorithm process, the traditional EM algorithm is used, namely, an initial expression P of a solution of an objective function is assumed, and then parameters and Q expression are sequentially solved according to formulas (6, 7 and 8).

4. The algorithm of claim 1, wherein after Q expression is obtained, a new expression of P is obtained according to equation (9). The updated P is closer to the optimal solution than the starting value.

5. The algorithm for recognizing a cardiac signal based on a Gaussian mixture model as claimed in claim 1, wherein: in step E, after iteration, the training process is stopped when the error on the training set is within the allowable range.

6. The algorithm for recognizing a cardiac signal based on a Gaussian mixture model as claimed in claim 1, wherein: in the step E, after continuous iteration, after gaussian distribution parameters and P representation are obtained, the corresponding optimal solution needs to be properly adjusted through a cross validation set, and a final reasonable expression is obtained.

Technical Field

The invention relates to the technical field of electrocardiosignal automatic processing, in particular to an electrocardiosignal identification algorithm based on a Gaussian mixture model

Background

The automatic analysis of electrocardio is one of the research hotspots in the Internet + era under the background of artificial intelligence. The internet is used as a link, the intelligent terminal is used as a carrier, the user is linked with the central electrocardiogram monitoring, the electrocardiogram signal is automatically analyzed by means of an artificial intelligence technology while the electrocardiogram quick acquisition is realized, so that the timely and accurate objective evaluation of the heart blood health condition is provided for the user, and tragedies such as sudden death and the like can be effectively avoided. The method has very important significance in theory and social practical value. The algorithm of the invention belongs to the field of automatic processing of digital electrocardiosignals. The algorithm takes a digital electrocardiosignal image as a processing object and applies a digital image processing technology to extract the digital electrocardiosignal from an electrocardio digital image. On the basis, the extracted electrocardiosignals are identified by using a Gaussian mixture model, so that a doctor can be assisted to diagnose the corresponding cardiovascular diseases.

With the continuous development and improvement of artificial intelligence technology, the application range of artificial intelligence technology is also continuously expanded. The automatic electrocardio-recognition combines artificial intelligence technology and electrocardio-analysis application, and utilizes multilayer neural network models (such as a restrictive Boltzmann model RBM, an automatic encoder model AE, a variation automatic encoder model VAE and the like). Firstly, a group of electrocardiosignals is used as input to train a multilayer neural network model. And then, the unknown electrocardiosignals are subjected to feature extraction through a hidden layer in the multilayer neural network model. After the electrocardio characteristics are obtained, whether the cardiovascular diseases occur or not can be judged by correspondingly analyzing the electrocardio characteristics, and the types of the cardiovascular diseases can be further deduced on the basis, so that a basis is provided for subsequent diagnosis and treatment clinically.

The automatic electrocardio analysis technology has wide application prospect. Currently, with home broadband, WIFI, 3G, 4G, etc., networks are already widely available in common homes and in many public places. People can obtain related latest consultation information at any time and any place through intelligent terminal equipment such as a mobile phone. However, for electrocardiographic acquisition, offline acquisition by professional electrocardiographic acquisition equipment is still the main reason, and clinicians mainly perform electrocardiographic analysis on patients through printed electrocardiographic images. The current situation mainly has the following defects:

the human heart has a strong compensatory mechanism, and the electrocardiogram of the early stage of the disease is often short in duration for some patients with cardiovascular diseases. This presents certain difficulties in "capturing" these diseases by electrocardiograms. Once the short time of the disease is missed, the compensatory mechanism of the heart can lead the function of the cardiovascular system to be normal, and the electrocardiogram becomes normal correspondingly. Therefore, the transient electrocardiosignal detection is difficult to accurately detect certain cardiovascular diseases.

The electrocardiogram is an external manifestation of cardiac function and is also an important basis for clinicians to diagnose cardiovascular diseases. The clinician can infer the cardiovascular function status from the distribution of the electrocardiogram. However, on one hand, China is a country with high cardiovascular disease incidence, and a large number of patients need to be treated by professional doctors in time clinically; on the other hand, the number of clinical professionals in China is relatively small, so that more and more patients with cardiovascular diseases cannot be treated in time. The common people generally have insufficient understanding of cardiovascular diseases, and the common people may be misled by other information, thereby causing misdiagnosis or missing the best treatment time.

Based on the above background, it is urgent to develop an automatic electrocardiographic acquisition and analysis system, which comprehensively utilizes an artificial intelligence technology, an internet technology, a cloud computing technology and a big data technology to acquire, process, transmit, store, identify and analyze electrocardiographic signals. On one hand, the propaganda of cardiovascular disease knowledge is strengthened, and the immunity of common people to pseudo science is improved by scientific knowledge. On the other hand, the electrocardiosignals are collected in real time by means of the intelligent terminal, and the collected electrocardiosignals are transmitted to the cloud electrocardio processing unit through the internet. And after the electrocardiosignals are processed by the cloud, the processing results are finally issued to the mobile phone terminal of the user. Besides, the system can also combine the user history record and the relevant big data information of the disease incidence of the area, and provide more accurate disease diagnosis and prediction service for the user. The algorithm effectively solves the difficult problem that the Gaussian mixture model can not be solved theoretically by taking deep learning as the background and introducing hidden variables. By using a certain amount of training data, the model can be quickly and effectively stabilized, and the automatic identification of the electrocardiosignal is realized on the basis, so that the possibility is provided for the automatic diagnosis of the cardiovascular disease, and the sudden death and other sudden cardiovascular diseases are effectively solved.

Disclosure of Invention

The invention aims to provide an electrocardiosignal recognition algorithm based on a Gaussian mixture model, which is beneficial to overcoming the defects of the traditional electrocardiosignal recognition process, and effectively solves the problem of difficult convergence point in the traditional electrocardiosignal model training process by introducing the distribution of implicit parameters. The accuracy and the stability of the automatic identification of the electrocardiosignals are improved.

In order to achieve the purpose, the technical scheme of the invention is as follows: an electrocardiosignal recognition algorithm based on a Gaussian mixture model comprises the following steps:

step A: reading an original electrocardio image, and obtaining continuous periodic electrocardio signals by applying an electrocardio extraction algorithm.

And B: according to the electrocardio data point derivatives (including first and second derivatives), the periodic electrocardio signals are segmented, and the electrocardio data of a single period are extracted.

And C: and aiming at the missing condition of adjacent data points, the missing electrocardio data are supplemented by adopting an interpolation mode by combining the electrocardio data on adjacent positions in the same period, so that the completeness of the electrocardio period is ensured.

Step D: and randomly grouping a plurality of complete electrocardiographic cycle data to respectively obtain a training group, a cross validation group and test group data.

Step E: initializing a Gaussian mixture model, and then training by using the electrocardiosignals of the training set. Until the error meets the corresponding requirement.

Step F: and then correcting the training result by using the cross validation group data, and repeating the steps until the error is within the specified range.

Step G: and testing the test group data by using the trained Gaussian mixture model to verify the effectiveness of the algorithm. Formula (1) represents a probability density function obeying gaussian distribution, and formula (2) represents a gaussian mixture model composed of K gaussian mixture distributions. Equation (3, 4) indicates that given a set of N data X ═ X₁，x₂，...，x_NAnd satisfy the probability of occurrence under the i.i.d. condition. Wherein the hidden variable (Z) is used in the formula (4) instead of the formula(3) The gaussian coefficient. Equation (5) represents the likelihood function for this function under different parameters θ. Equations (6, 7) represent the optimal solution of the gaussian distribution parameters obtained after using an optimization algorithm given a set of training data. Equation (8, 9) represents the posterior distribution Q of the hidden variable z, and the solution of the objective function- -P distribution- -the parametric representation of the normal distribution. (ii) a

Drawings

FIG. 1 is a flow chart of the algorithm of the present invention.

Detailed Description

The implementation process of the method is further explained by combining the actual electrocardiogram and the algorithm flow.

FIG. 1 is a flow chart of an implementation of the Gaussian mixture model-based electrocardiosignal recognition algorithm of the present invention. As shown in fig. 1, the method comprises the steps of:

step A: and selecting a proper algorithm to extract the electrocardiosignals from the electrocardio image.

And B: and carrying out operations such as interpolation, smoothing and the like on the extracted electrocardiosignals so as to enable the electrocardiosignals to be continuous, and realizing the division of the electrocardiosignals according to the cardiac cycle on the basis.

And C: grouping cardiac signals of a single cardiac cycle: training, cross validation and testing groups.

Step D: a Gaussian mixture model is initialized, including Gaussian distribution parameters. The posterior distribution Q of the hidden variable z is obtained according to equation (8).

Step E: the distribution P and gaussian parameters are calculated according to equation (9) and the training process is repeated until the error falls within the allowable range.

Step F: and E, according to the result of the step E, correspondingly correcting the Gaussian mixture model by using the electrocardiosignals on the cross validation set, and eliminating overfitting possibly caused in the training process. It should be emphasized that the examples described herein are illustrative and not restrictive, and thus the present invention includes, but is not limited to, those examples described in the detailed description, and that other embodiments derived from the teachings of the present invention by those skilled in the art are also within the scope of the present invention.

The above description is only a preferred embodiment of the present invention, and all changes and modifications that come within the scope of the invention as defined by the appended claims fall within the scope of the invention.