WO2022016638A1 - 基于样本分布改进的卷积神经网络的训练方法及模型 - Google Patents

基于样本分布改进的卷积神经网络的训练方法及模型 Download PDF

Info

Publication number
WO2022016638A1
WO2022016638A1 PCT/CN2020/109294 CN2020109294W WO2022016638A1 WO 2022016638 A1 WO2022016638 A1 WO 2022016638A1 CN 2020109294 W CN2020109294 W CN 2020109294W WO 2022016638 A1 WO2022016638 A1 WO 2022016638A1
Authority
WO
WIPO (PCT)
Prior art keywords
neural network
convolutional neural
sample distribution
layer
heartbeat
Prior art date
Application number
PCT/CN2020/109294
Other languages
English (en)
French (fr)
Inventor
朱俊江
王雨轩
陈红岩
朱志超
张顺宇
Original Assignee
上海数创医疗科技有限公司
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 上海数创医疗科技有限公司 filed Critical 上海数创医疗科技有限公司
Priority to EP20946488.2A priority Critical patent/EP4187435A4/en
Publication of WO2022016638A1 publication Critical patent/WO2022016638A1/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
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
    • A61B5/346—Analysis of electrocardiograms
    • A61B5/349—Detecting specific parameters of the electrocardiograph cycle
    • A61B5/364—Detecting abnormal ECG interval, e.g. extrasystoles, ectopic heartbeats
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
    • A61B5/346—Analysis of electrocardiograms
    • A61B5/349—Detecting specific parameters of the electrocardiograph cycle
    • A61B5/366—Detecting abnormal QRS complex, e.g. widening
    • 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
    • A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
    • 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
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • G06F2218/14—Classification; Matching by matching peak patterns
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • G06F2218/16—Classification; Matching by matching signal segments

Definitions

  • the present application belongs to the technical field of heartbeat type screening by artificial neural networks, and in particular relates to a training method and model of a convolutional neural network improved based on sample distribution.
  • the ECG signal is a comprehensive reflection of the electrical activity of the myriad cardiomyocytes of the heart.
  • the cardiomyocytes of the sinoatrial node, atrium, and ventricle depolarize in turn, forming a heart beat that includes a signal segment of a P wave, a QRS wave, and a T wave.
  • the combination of multiple heart beats forms an ECG signal.
  • Premature ventricular beat refers to the beat formed by the depolarization of the ventricle caused by an electrical impulse from any part of the ventricle or an ectopic rhythm point in the ventricular septum before the sinus node impulse reaches the ventricle.
  • the technical problem to be solved by the present invention is to provide a training method and model of a convolutional neural network improved based on sample distribution in order to solve the deficiencies in the prior art.
  • An improved convolutional neural network training method based on sample distribution comprising the following steps:
  • Step S31 Collect enough lead II cardiac beat signals in the marked 12-lead ECG signals to form an initial data set, and the II lead cardiac beat signals include premature ventricular beat signals and non-premature ventricular beat signals each accounting for half of the total number , set the label of the premature ventricular beat signal to a, and the label of the non-premature heart beat signal to b;
  • Step S33 The following judgments are made on all heartbeat data in each data: Obtain the number of established heartbeat data sets, and form a new heartbeat data set after eliminating the heartbeat data with less than 3 established conditions in the initial data;
  • Step S34 Repeat step S32 until the heartbeat data set does not change, and use the heartbeat data set as training data;
  • Step S35 input the training data and its corresponding labels into the convolutional neural network for training, and obtain a convolutional neural network model with improved sample distribution;
  • the sampling frequency of the heart beat signal of lead II is 500 Hz, and is filtered by a Butterworth bandpass filter of 0.5-100 Hz.
  • each heartbeat signal is an electrocardiogram signal from 0.38s before the peak of the R wave to 0.5s after the peak of the R wave.
  • the improved convolutional neural network model of the sample distribution is composed of 10 layers of networks, namely layer1-layer9 and a classifier; wherein layer1-layer5 is composed of a volume
  • the stacking layer and a pooling layer are composed of the LeakyReLU activation function;
  • the convolutional layer in layer1 contains 32 kernels, the size of the convolution kernel is 7, and the stride and kernel size in the pooling layer in layer1 are both 2;
  • layer2 The convolutional layer contains 32 kernels, and the size of the convolution kernel is 5.
  • the stride and kernel size in the pooling layer in layer2 are both 2; the convolutional layer of layer3 contains 32 kernels, and the size of the convolution kernel is 3. , the stride and kernel size in the pooling layer in layer3 are both 2; the convolutional layer in layer4 contains 32 kernels, the convolution kernel size is 5, and the stride and kernel size in the pooling layer in layer4 are both 2 ; The convolution layer of layer5 contains 32 kernels, and the size of the convolution kernel is 5. The stride and kernel size in the pooling layer in layer5 are both 2; the convolution layer of layer6 contains 32 kernels, and the size of the convolution kernel is all 2 is 5; layer7 is a bidirectional long short-term memory network layer.
  • the number of input layer neurons is the same as the number of output features of layer6.
  • the forward output neurons are 30 neurons, and the backward output neurons are 30 neurons.
  • the short-term memory network layer outputs a total of 60 neurons; layer8 is the attention mechanism layer, which undertakes the output of layer7; layer9 is the fully connected layer, which undertakes the output of layer8, and the number of output neurons is 4.
  • the present application also provides an improved convolutional neural network model based on sample distribution, which is obtained by training the above-mentioned training method of the improved convolutional neural network based on sample distribution.
  • the present application also provides a method for using the improved convolutional neural network model based on sample distribution
  • the training method and model of the improved convolutional neural network based on the sample distribution provided by the present application by calculating the average value x avg , the maximum value x MAX , the minimum value x MIN , the peak value x F , and the kurtosis value x Q of the heart beat in advance, Whether the conditions are met to screen the samples, when When the number of established in the model is less than 3, it is considered that the training data used by the model does not conform to the sample distribution, and these heartbeat signals that do not meet the requirements are eliminated, thereby reducing the misjudgment of the heartbeat type.
  • FIG. 1 is a flowchart of a method for recognizing heart beat types based on an improved convolutional neural network based on sample distribution according to an embodiment of the present application.
  • FIG. 2 is a schematic structural diagram of an improved convolutional neural network model based on sample distribution according to an embodiment of the present application.
  • This embodiment provides a method for training a convolutional neural network improved based on sample distribution, as shown in FIG. 1 , including the following steps:
  • Step S31 Collect enough lead II cardiac beat signals in the marked 12-lead ECG signals to form an initial data set, and the II lead cardiac beat signals include premature ventricular beat signals and non-premature ventricular beat signals each accounting for half of the total number , set the label of the premature ventricular beat signal to a, and the label of the non-premature heart beat signal to b;
  • Step S33 The following judgments are made on all heartbeat data in each data: Obtain the number of established heartbeat data sets, and form a new heartbeat data set after eliminating the heartbeat data with less than 3 established conditions in the initial data;
  • Step S34 Repeat step S32 until the heartbeat data set does not change, and use the heartbeat data set as training data;
  • Step S35 Input the training data and its corresponding labels into the convolutional neural network for training, and obtain a convolutional neural network model with improved sample distribution.
  • the sampling frequency of the heart beat signal of lead II in step S31 is 500 Hz, and is filtered by a Butterworth bandpass filter of 0.5-100 Hz. If not, it needs to be resampled to 500 Hz first.
  • Each heartbeat signal is an ECG signal from 0.38s before the R wave peak to 0.5s after the R wave peak.
  • the improved convolutional neural network model for sample distribution consists of a 10-layer network and a 10-layer network, namely layer1-layer9 and a classifier; layer1-layer5 consists of a convolutional layer and a pooling layer, all using the LeakyReLU activation function; layer1
  • the middle convolutional layer contains 32 kernels, and the size of the convolution kernel is 7.
  • the stride and kernel size in the pooling layer of layer1 are both 2; the convolutional layer of layer2 contains 32 kernels, and the size of the convolution kernel is 5.
  • the stride and kernel size in the pooling layer in layer2 are both 2;
  • the convolutional layer in layer3 contains 32 kernels, the convolution kernel size is 3, and the stride and kernel size in the pooling layer in layer3 are both 2 ;
  • the convolutional layer of layer4 contains 32 kernels, and the size of the convolution kernel is 5.
  • the stride and kernel size in the pooling layer in layer4 are both 2; the convolutional layer of layer5 contains 32 kernels, and the size of the convolution kernel is all is 5, the stride size and kernel size in the pooling layer in layer5 are both 2; the convolutional layer in layer6 contains 32 kernels, and the convolution kernel size is 5; layer7 is the input layer neuron of the bidirectional long short-term memory network layer Consistent with the number of output features of layer6, the forward output neurons are 30 neurons, the backward output neurons are 30 neurons, and the bidirectional long short-term memory network layer outputs a total of 60 neurons; layer8 is the attention mechanism layer, to undertake the output of layer7; layer9 is a fully connected layer, to undertake the output of layer8, and the number of output neurons is 4.
  • the unknown type of heart beat is considered to be a premature ventricular beat; when the output value is less than or equal to 0.5 , that the unknown type of heart beat is a non-premature heart beat signal.
  • This embodiment provides an improved convolutional neural network model based on sample distribution, which is characterized in that it is obtained by training the improved convolutional neural network model based on sample distribution described in Embodiment 1.
  • This embodiment provides a method for using an improved convolutional neural network model based on sample distribution, including:
  • Step S1 collecting ECG 12 unknown type electrical conductivity in lead II ECG beat signal heartbeat, calculating the average heartbeat x avg, maximum x MAX, the minimum value x MIN, peak x F, and kurtosis value x Q ;
  • Step S2 Judgment The number established in
  • Step S3 If there are 3 or more established in step S2, then input into the improved convolutional neural network model based on the sample distribution of Embodiment 2 to judge the heartbeat type, otherwise it is considered that the unknown type of heartbeat is "the type that cannot be judged";
  • Step S4 when the output value of the improved convolutional neural network model based on the sample distribution is close to a, the unknown type of heart beat is considered to be a premature ventricular beat; when the output value is close to b, the unknown type of heart beat is considered to be a non-premature ventricular beat signal;
  • step S4 when the output value of the improved convolutional neural network model based on the sample distribution is greater than 0.5, the unknown type of heart beat is considered to be a premature ventricular beat; the output value is less than When it is equal to 0.5, the unknown type of heartbeat is considered to be a non-premature heartbeat signal.
  • the embodiments of the present application may be provided as a method, a system, or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • computer-usable storage media including, but not limited to, disk storage, CD-ROM, optical storage, etc.
  • These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions
  • the apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Artificial Intelligence (AREA)
  • Public Health (AREA)
  • Physics & Mathematics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Pathology (AREA)
  • Cardiology (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Veterinary Medicine (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Signal Processing (AREA)
  • Physiology (AREA)
  • Psychiatry (AREA)
  • Evolutionary Computation (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Fuzzy Systems (AREA)
  • Databases & Information Systems (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Power Engineering (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

本申请涉及一种本申请提供的基于样本分布改进的卷积神经网络的训练方法及模型，通过先行计算心拍的平均值xavg，最大值xMAX，最小值xMIN，峰值xF，和峭度值xQ，是否符合条件来筛选样本，当 aa，bb，cc，dd，ee 中成立的个数小于3个时，则认为模型所使用的训练数据不符合样本分布，剔除掉这些不符合要求的心拍信号，以此减少心拍类型的误判。

Description

基于样本分布改进的卷积神经网络的训练方法及模型 技术领域
本申请属于人工神经网络在心跳类型筛选技术领域，尤其是涉及一种基于样本分布改进的卷积神经网络的训练方法及模型。
背景技术
心电信号是心脏的无数心肌细胞电活动的综合反映。正常情况下，窦房结、心房、心室的心肌细胞依次除极，形成一个包含一段P波，QRS波，T波的信号段的心拍，多个心拍组合形成一段心电信号。室性早搏心拍指的是在窦房结冲动尚未抵达心室之前，由心室中的任何一个部位或室间隔的异位节律点提前发出电冲动引起心室的除极所形成的心拍。从一段长时间的心电信号中筛查出所有室性早搏心拍对判别病人情况极为重要，但又耗时耗力，因此利用计算机筛查出心电信号中的室早心拍意义重大。然而由于心脏内可以发出电冲击的位置很多，电冲击位置不同时，所展示出的心拍形状不同，而且存在很多干扰，因此在对心拍进行分类时，想要找到所有的室早心拍以及“其他”类型心拍的模式和数据对深度学习模型进行训练几乎是不可能的，因而容易导致对心拍类型的误判。
发明内容
本发明要解决的技术问题是：为解决现有技术中的不足，从而提供一种基于样本分布改进的卷积神经网络的训练方法及模型。
本发明解决其技术问题所采用的技术方案是：
一种基于样本分布改进的卷积神经网络的训练方法，包括以下步骤：
步骤S31：收集标记好的12导联心电信号中的II导联心拍信号充分多条形成初始数据集，II导联心拍信号包括各占总数量一半的室早心拍信号和非室早心拍信号，将室早心拍信号的标签设置为a，非室早心拍信号的标签设置为b；
步骤S32：对收集到的数据集中的所有心拍数据进行统计，统计出每个心拍数据的均值、最大值、最小值、峰值和峭度值分别形成均值数组AVG＝{a1,a2,a3…aN}、最大值数组MAX＝{m1,m2,m3…mN}、最小值数组MIN＝{c1,c2,…cN}、峰值数组F＝{f1,f2,…fN}和峭度值数组Q＝{q1,q2,…qN}，再次分别求取均值数组、最大值数组、最小值数组、峰值数组和峭度值数组的均值和均方差值分别记为
Figure PCTCN2020109294-appb-000001
步骤S33：对每个数据中的所有心拍数据均进行以下判断：
Figure PCTCN2020109294-appb-000002
Figure PCTCN2020109294-appb-000003
获取其中成立的个数，将初始数据中小于3个成立条件的心拍数据剔除后形成新的心拍数据集；
步骤S34：重复执行步骤S32，直至心拍数据集不发生变化时，将心拍数据集作为训练数据；
步骤S35：将训练数据及其对应的标签输入到卷积神经网络进行训练，得到样本分布改进的卷积神经网络模型；
优选地，本发明的基于样本分布改进的卷积神经网络的训练方法，II导联心拍信号的采样频率500Hz，并经过0.5-100Hz的巴特沃兹带通滤波器进行滤波。
优选地，本发明的基于样本分布改进的卷积神经网络的训练方法，每个心拍信号为从R波波峰之前的0.38s到R波峰后0.5s的心电图信号。
优选地，本发明的基于样本分布改进的卷积神经网络的训练方法，样本分布改进的卷积神经网络模型由10层网络即layer1-layer9和一个分类器组成；其中layer1-layer5均由一个卷积层和一个池化层组成，均使用LeakyReLU激活函数；layer1中卷积层包含32个核，卷积核大小均为7，layer1中池化层中的步长和核大小均为2；layer2的卷积层包含32个核，卷积核大小均为5，layer2中池化层中的步长和核大小均为2；layer3的卷积层包含32个核，卷积核大小均为3，layer3中池化层中的步长和核大小均为2；layer4的卷积层包含32个核，卷积核大小均为5，layer4中池化层中的步长和核大小均为2；layer5的卷积层包含32个核，卷积核大小均为5，layer5中池化层中的步长和核大小均为2；layer6的卷积层包含32个核，卷积核大小均为5；layer7为双向长短期记忆网络层的输入层神经元与layer6的输出特征个数一致，向前输出神经元为30个神经元，向后输出神经元为30个神经元，双向长短期记忆网络层共输出60个神经元；layer8为注意力机制层，承接layer7的输出；layer9为全连接层，承接layer8的输出，输出的神经元个数为4个。
优选地，本发明的基于样本分布改进的卷积神经网络的训练方法，a＝1，b＝0，当基于样本分布改进的卷积神经网络模型的输出值大于0.5时，认为未知类型的心拍为室早心拍；输出值小于等于0.5时，认为未知类型的心拍为非室早心拍信号。
本申请还提供一种基于样本分布改进的卷积神经网络模型，由上述的基于样本分布改进的卷积神经网络的训练方法训练得到。
本申请还提供一种基于样本分布改进的卷积神经网络模型的使用方法，
包括以下步骤：
采集未知类型的12导联心电信号中的II导联心拍信号中的心拍，计算该心拍的平均值x avg，最大值x MAX，最小值x MIN，峰值x F，和峭度值x Q；判断
Figure PCTCN2020109294-appb-000004
Figure PCTCN2020109294-appb-000005
中成立的个数，若有3个及以上成立时，则输入上述的基于样本分布改进的卷积神经网络模型中，根据基于样本分布改进的卷积神经网络模型的输出结果判断心拍信号的类型。
本发明的有益效果是：
本申请提供的基于样本分布改进的卷积神经网络的训练方法及模型，通过先行计算心拍的平均值x avg，最大值x MAX，最小值x MIN，峰值x F，和峭度值x Q，是否符合条件来筛选样本，当
Figure PCTCN2020109294-appb-000006
Figure PCTCN2020109294-appb-000007
Figure PCTCN2020109294-appb-000008
中成立的个数小于3个时，则认为模型所使用的训练数据不符合样本分布，剔除掉这些不符合要求的心拍信号，以此减少心拍类型的误判。
附图说明
下面结合附图和实施例对本申请的技术方案进一步说明。
图1是本申请实施例的基于样本分布改进的卷积神经网络的心拍类型识别方法的流程图。
图2是本申请实施例的基于样本分布改进的卷积神经网络模型的结构示意图。
具体实施方式
需要说明的是，在不冲突的情况下，本申请中的实施例及实施例中的特征可以相互组合。
下面将参考附图并结合实施例来详细说明本申请的技术方案。
实施例1
本实施例提供一种基于样本分布改进的卷积神经网络的训练方法，如图1所示，包括以下步骤：
步骤S31：收集标记好的12导联心电信号中的II导联心拍信号充分多条形成初始数据集，II导联心拍信号包括各占总数量一半的室早心拍信号和非室早心拍信号，将室早心拍信号的标签设置为a，非室早心拍信号的标签设置为b；
步骤S32：对收集到的数据集中的所有心拍数据进行统计，统计出每个心拍数据的均值、最大值、最小值、峰值和峭度值分别形成均值数组AVG＝{a1,a2,a3…aN}、最大值数组MAX＝{m1,m2,m3…mN}、最小值数组MIN＝{c1,c2,…cN}、峰值数组F＝{f1,f2,…fN}和峭度值数组Q＝{q1,q2,…qN}，再次分别求取均值数组、最大值数组、最小值数组、峰值数组和峭度值数组的均值和均方差值分别记为
Figure PCTCN2020109294-appb-000009
步骤S33：对每个数据中的所有心拍数据均进行以下判断：
Figure PCTCN2020109294-appb-000010
Figure PCTCN2020109294-appb-000011
获取其中成立的个数，将初始数据中小于3个成立条件的心拍数据剔除后形成新的心拍数据集；
步骤S34：重复执行步骤S32，直至心拍数据集不发生变化时，将心拍数据集作为训练数据；
步骤S35：将训练数据及其对应的标签输入到卷积神经网络进行训练，得到样本分布改进的卷积神经网络模型。
步骤S31中的II导联心拍信号的采样频率500Hz，并经过0.5-100Hz的巴特沃兹带通滤波器进行滤波，如果不是，则需要先将其重采样至500Hz。每个心拍信号为从R波波峰之前的0.38s到R波峰后0.5s的心电图信号。
样本分布改进的卷积神经网络模型由10层网络10层网络即layer1-layer9和一个分类器组成；其中layer1-layer5均由一个卷积层和一个池化层组成，均使用LeakyReLU激活函数；layer1中卷积层包含32个核，卷积核大小均为7，layer1中池化层中的步长和核大小均为2；layer2的卷积层包含32个核，卷积核大小均为5，layer2中池化层中的步长和核大小均为2；layer3的卷积层包含32个核，卷积核大小均为3，layer3中池化层中的步长和核大小均为2；layer4的卷积层包含32个核，卷积核大小均为5，layer4中池化层中的步长和核大小均为2；layer5的卷积层包含32个核，卷积核大小均为5，layer5中池化层中的步长和核大小均为2；layer6的卷积层包含32个核，卷积核大小均为5；layer7为双向长短期记忆网络层的输入层神经元与layer6的输出特征个数一致，向前输出神经元为30个神经元，向后输出神经元为30个神经元，双向长短期记忆网络层共输出60个神经元；layer8为注意力机制层，承接layer7的输出；layer9为全连接层，承接layer8的输出，输出的神经元个数为4个。
a和b的值可以选择为a＝1，b＝0，当基于样本分布改进的卷积神经网络模型的输出值大于0.5时，认为未知类型的心拍为室早心拍；输出值小于等于0.5时，认为未知类型的心拍为非室早心拍信号。
实施例2
本实施例提供一种基于样本分布改进的卷积神经网络模型，其特征在于，由实施例1所述的基于样本分布改进的卷积神经网络的训练方法训练得到。
实施例3
本实施例提供基于样本分布改进的卷积神经网络模型的使用方法，包括：
步骤S1：采集未知类型的12导联心电信号中的II导联心拍信号中的心拍，计算该心拍的平均值x avg，最大值x MAX，最小值x MIN，峰值x F，和峭度值x Q；
步骤S2：判断
Figure PCTCN2020109294-appb-000012
Figure PCTCN2020109294-appb-000013
Figure PCTCN2020109294-appb-000014
中成立的个数；
步骤S3：如果步骤S2中有3个及以上成立则输入到实施例2的基于样本分布改进的卷积神经网络模型来判断心拍类型，否则认为未知类型的心拍是“无法判断的类型”；
步骤S4：当基于样本分布改进的卷积神经网络模型的输出值接近a时，认为未知类型的心拍为室早心拍；输出值接近b时，认为未知类型的心拍为非室早心拍信号；
a和b的值可以选择为a＝1，b＝0，步骤S4中当基于样本分布改进的卷积神经网络模型的输出值大于0.5时，认为未知类型的心拍为室早心拍；输出值小于等于0.5时，认为未知类型的心拍为非室早心拍信号。
以上述依据本申请的理想实施例为启示，通过上述的说明内容，相关工作人员完全可以在不偏离本项申请技术思想的范围内，进行多样的变更以及修改。本项申请的技术性范围并不局限于说明书上的内容，必须要根据权利要求范围来确定其技术性范围。
本领域内的技术人员应明白，本申请的实施例可提供为方法、系统、或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本申请是参照根据本申请实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。

Claims (7)

  1. 一种基于样本分布改进的卷积神经网络的训练方法，其特征在于，包括以下步骤：
    步骤S31：收集标记好的12导联心电信号中的II导联心拍信号充分多条形成初始数据集，II导联心拍信号包括各占总数量一半的室早心拍信号和非室早心拍信号，将室早心拍信号的标签设置为a，非室早心拍信号的标签设置为b；
    步骤S32：对收集到的数据集中的所有心拍数据进行统计，统计出每个心拍数据的均值、最大值、最小值、峰值和峭度值分别形成均值数组AVG＝{a1,a2,a3…aN}、最大值数组MAX＝{m1,m2,m3…mN}、最小值数组MIN＝{c1,c2,…cN}、峰值数组F＝{f1,f2,…fN}和峭度值数组Q＝{q1,q2,…qN}，再次分别求取均值数组、最大值数组、最小值数组、峰值数组和峭度值数组的均值和均方差值分别记为
    Figure PCTCN2020109294-appb-100001
    σ avg，
    Figure PCTCN2020109294-appb-100002
    σ MAX，
    Figure PCTCN2020109294-appb-100003
    σ MIN，
    Figure PCTCN2020109294-appb-100004
    σ F，
    Figure PCTCN2020109294-appb-100005
    σ Q；
    步骤S33：对每个数据中的所有心拍数据均进行以下判断：
    Figure PCTCN2020109294-appb-100006
    Figure PCTCN2020109294-appb-100007
    Figure PCTCN2020109294-appb-100008
    获取其中成立的个数，将初始数据中小于3个成立条件的心拍数据剔除后形成新的心拍数据集；
    步骤S34：重复执行步骤S32，直至心拍数据集不发生变化时，将心拍数据集作为训练数据；
    步骤S35：将训练数据及其对应的标签输入到卷积神经网络进行训练，得到样本分布改进的卷积神经网络模型。
  2. 根据权利要求1所述的基于样本分布改进的卷积神经网络的训练方法，其特征在于，II导联心拍信号的采样频率500Hz，并经过0.5-100Hz的巴特沃兹带通滤波器进行滤波。
  3. 根据权利要求1或2所述的基于样本分布改进的卷积神经网络的训练方法，其特征在于，每个心拍信号为从R波波峰之前的0.38s到R波峰后0.5s的心电图信号。
  4. 根据权利要求1或2所述的基于样本分布改进的卷积神经网络的训练方法，其特征在于，样本分布改进的卷积神经网络模型由10层网络即layer1-layer9和一个分类器组成；其中layer1-layer5均由一个卷积层和一个池化层组成，均使用LeakyReLU激活函数；layer1中卷积层包含32个核，卷积核大小均为7，layer1中池化层中的步长和核大小均为2；layer2的卷积层包含32个核，卷积核大小均为5，layer2中池化层中的步长和核大小均为2；layer3的卷积层包含32个核，卷积核大小均为3，layer3中池化层中的步长和核大小均为2；layer4的卷积层包含32个核，卷积核大小均为5，layer4中池化层中的步长和核大小均为2；layer5的卷积层包含32个核，卷积核大小均为5，layer5中池化层中的步长和核大小均为2；layer6的卷积层包含32个核，卷积核大小均为5；layer7为双向长短期记忆网络层的输入层神经元与layer6的输出特征个数一致，向前输出神经元为30个神经元，向后输出神经元为30个神经元，双向长短期记忆网络层共输出60个神经元；layer8为注意力机制层，承接layer7的输出；layer9为全连接层，承接layer8的输出，输出的神经元个数为4个。
  5. 根据权利要求1-3任一项所述的基于样本分布改进的卷积神经网络的训练方法，其特征在于，a＝1，b＝0，当基于样本分布改进的卷积神经网络模型的输出值大于0.5时，认为未知类型的心拍为室早心拍；输出值小于等于0.5时，认为未知类型的心拍为非室早心拍信号。
  6. 一种基于样本分布改进的卷积神经网络模型，其特征在于，由权利要求1-5任一项所述的基于样本分布改进的卷积神经网络的训练方法训练得到。
  7. 一种基于样本分布改进的卷积神经网络模型的使用方法，其特征在于，
    包括以下步骤：
    采集未知类型的12导联心电信号中的II导联心拍信号中的心拍，计算该心拍的平均值x avg，最大值x MAX，最小值x MIN，峰值x F，和峭度值x Q；判断
    Figure PCTCN2020109294-appb-100009
    Figure PCTCN2020109294-appb-100010
    Figure PCTCN2020109294-appb-100011
    中成立的个数，若有3个及以上成立时，则输入权利要求6所述的基于样本分布改进的卷积神经网络模型中，根据基于样本分布改进的卷积神经网络模型的输出结果判断心拍信号的类型。
PCT/CN2020/109294 2020-07-22 2020-08-14 基于样本分布改进的卷积神经网络的训练方法及模型 WO2022016638A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
EP20946488.2A EP4187435A4 (en) 2020-07-22 2020-08-14 TRAINING METHOD AND MODEL FOR CONVOLUTIONAL NEURAL NETWORK BASED ON SAMPLE DISTRIBUTION IMPROVEMENT

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010708023.3 2020-07-22
CN202010708023.3A CN111956202B (zh) 2020-07-22 2020-07-22 基于样本分布改进的卷积神经网络的训练方法及模型

Publications (1)

Publication Number Publication Date
WO2022016638A1 true WO2022016638A1 (zh) 2022-01-27

Family

ID=73364143

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/109294 WO2022016638A1 (zh) 2020-07-22 2020-08-14 基于样本分布改进的卷积神经网络的训练方法及模型

Country Status (3)

Country Link
EP (1) EP4187435A4 (zh)
CN (1) CN111956202B (zh)
WO (1) WO2022016638A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111956201A (zh) * 2020-07-22 2020-11-20 上海数创医疗科技有限公司 基于卷积神经网络的心拍类型识别方法和装置
CN114451898A (zh) * 2022-03-15 2022-05-10 西安理工大学 短时训练卷积神经网络的心电信号分类方法

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190069795A1 (en) * 2017-03-10 2019-03-07 Qatar University Personalized ecg monitoring for early detection of cardiac abnormalities
CN109620207A (zh) * 2018-12-26 2019-04-16 上海数创医疗科技有限公司 基于改进卷积神经网络的室性早搏心跳识别方法及装置
CN110141218A (zh) * 2019-06-17 2019-08-20 东软集团股份有限公司 一种心电信号分类方法、装置及程序产品、存储介质

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019144311A1 (zh) * 2018-01-24 2019-08-01 悦享趋势科技（北京）有限责任公司 规则嵌入式人工神经网络系统及其训练方法

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190069795A1 (en) * 2017-03-10 2019-03-07 Qatar University Personalized ecg monitoring for early detection of cardiac abnormalities
CN109620207A (zh) * 2018-12-26 2019-04-16 上海数创医疗科技有限公司 基于改进卷积神经网络的室性早搏心跳识别方法及装置
CN110141218A (zh) * 2019-06-17 2019-08-20 东软集团股份有限公司 一种心电信号分类方法、装置及程序产品、存储介质

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
See also references of EP4187435A4 *
ZHOU FEI-YAN, JIN LIN-PENG, DONG JUN: "PVC Recognition Algorithm Based on Ensemble", ACTA ELECTRONICA SINICA, ZHONGGUO DIANZI XUEHUI, CN, vol. 45, no. 2, 25 February 2017 (2017-02-25), CN , pages 501 - 507, XP055891198, ISSN: 0372-2112, DOI: 10.3969/j.issn.0372-2112.2017.02.032 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111956201A (zh) * 2020-07-22 2020-11-20 上海数创医疗科技有限公司 基于卷积神经网络的心拍类型识别方法和装置
CN111956201B (zh) * 2020-07-22 2022-09-06 上海数创医疗科技有限公司 基于卷积神经网络的心拍类型识别方法和装置
CN114451898A (zh) * 2022-03-15 2022-05-10 西安理工大学 短时训练卷积神经网络的心电信号分类方法
CN114451898B (zh) * 2022-03-15 2023-06-30 西安理工大学 短时训练卷积神经网络的心电信号分类方法

Also Published As

Publication number Publication date
EP4187435A1 (en) 2023-05-31
CN111956202A (zh) 2020-11-20
CN111956202B (zh) 2021-09-03
EP4187435A4 (en) 2024-02-28

Similar Documents

Publication Publication Date Title
He et al. Automatic cardiac arrhythmia classification using combination of deep residual network and bidirectional LSTM
WO2021057328A1 (zh) 基于特征选取的st段分类卷积神经网络及其使用方法
CN108478209B (zh) 心电信息动态监护方法和动态监护系统
CN111990989A (zh) 一种基于生成对抗及卷积循环网络的心电信号识别方法
CN110151138B (zh) 基于卷积神经网络的睡眠呼吸暂停片段检测方法、设备
WO2020006939A1 (zh) 基于生成式对抗网络的心电图生成及分类方法
WO2022016638A1 (zh) 基于样本分布改进的卷积神经网络的训练方法及模型
CN109077714B (zh) 信号识别方法、装置、设备和存储介质
CN110717416B (zh) 基于特征选取的st段分类识别用神经网络训练方法
CN113095302B (zh) 用于心律失常分类的深度模型、利用该模型的方法及装置
CN110367968B (zh) 一种右束支阻滞检测方法、装置、设备及存储介质
CN106066933A (zh) 获取房颤识别人工神经网络权重值矩阵的方法
Kumari et al. Heart rhythm abnormality detection and classification using machine learning technique
RU2016146176A (ru) Способ и система неинвазивной скрининговой оценки физиологических параметров и патологий
CN109620206B (zh) 包含异位心跳判断的房颤人工智能识别方法及装置
CN113384277B (zh) 心电数据的分类方法及分类系统
CN114847905B (zh) 一种心率失常数据检测识别方法及系统
Chen et al. Heart murmurs clustering using machine learning
Cai et al. Rule-based rough-refined two-step-procedure for real-time premature beat detection in single-lead ECG
CN111956201B (zh) 基于卷积神经网络的心拍类型识别方法和装置
CN112603324B (zh) 基于改进损失函数的神经网络的训练方法
CN112022140B (zh) 一种心电图的诊断结论自动诊断方法及系统
CN113349753A (zh) 一种基于便携式动态心电监护仪的心律失常检测方法
US9192315B2 (en) Patient signal analysis and characterization based on late potentials
CN110353657B (zh) 一种基于双选机制的多种波形类型筛选方法及装置

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: 20946488

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2020946488

Country of ref document: EP

Effective date: 20230222