CN105550659A - Real-time electrocardiogramclassification method based on random projection - Google Patents
Real-time electrocardiogramclassification method based on random projection Download PDFInfo
- Publication number
- CN105550659A CN105550659A CN201510988099.5A CN201510988099A CN105550659A CN 105550659 A CN105550659 A CN 105550659A CN 201510988099 A CN201510988099 A CN 201510988099A CN 105550659 A CN105550659 A CN 105550659A
- Authority
- CN
- China
- Prior art keywords
- data
- classification
- real time
- accidental projection
- electrocardiogram
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
Classifications
-
- 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
Abstract
The invention requests to protect a real-time electrocardiogram classification method based on random projection, and aims to solve the problems of data acquisition and computation and power consumption transmission with which a remote electrocardiogram monitoring system faces and the problem that electrocardiogram cannot be classified in real time. Five types of heartbeat are classified into normal pulsation, atrial premature beat, ventricular premature beat, left bundle branch block and right bundle branch block. The method comprises the following steps: (1) data preprocessing; (2) characteristic extraction: on the basis of a compressed sensing principle, compressing data, calculating an RR interval and an RR weight, and splicing characteristic vectors to form second characteristics; (3) classification: dividing secondary characteristic data into training data and test data, wherein the training data and the test data are independently used for modeling ant testing; and (4) decision classification: carrying out multiple-lead classification result data fusion. The step of data preprocessing comprises the following specific steps: 1) filtering an electrocardiosignal, and removing interference; 2) carrying out waveform detection and segmentation; 3) carrying out data standardization. The electrocardiogram data classification method provided by the invention is accurate in a classification result, and improves data processing capability.
Description
Technical field
The present invention relates to crossing domain that is biomedical and computer science, in particular to a kind of method of classifying to dynamic electrocardiogram (ECG) data.
Background technology
The World Health Organization's recent statistics data show, heart disease is cause the dead main cause of disease always.Although traditional electrocardiogram monitor system effectively can reduce the mortality ratio of cardiac, because it can not remote monitoring, so cannot the electrocardiosignal of monitored patient in real time.But heart disease has disguise and latency, being difficult to when not falling ill show from cardiogram, is again of short duration during morbidity, often only continues the time of tens seconds, waits to arrive in hospital and have an electro-cardiogram when checking, symptom disappears, and cardiogram recovers normal.Doctor cannot diagnose in time to patient, incurs loss through delay disease treatment.Therefore some patient need carry 24 hours Holter, within 24 hours, gathers electrocardiogram (ECG) data.Further, along with the development of current information science and medical science, the health of increasing people to individual and family puts forward higher requirement, expects can grasp oneself health status at any time, finds aura as early as possible.Telediagnosis of Electrocardiogram Signals system, wearable electrocardiogram acquisition equipment arises at the historic moment.Measurement and the storage of mass data are not becoming problem, but doctor will to googol like this according to analyze, and finding out the abnormal heart and clap, is a very hard work.In order to save the time of doctor greatly, improve diagnosis efficiency, extract and produce the potential essential information of these data, the automatic classification method of design stability is only basic goal.
Rarefaction representation and compressive sensing theory obtain efficiently in unconventional mode and process data, for research high dimensional data provides a tight mathematical framework, for the immanent structure disclosed in high dimensional data provides new thinking.This theory can be understood as and simulated data saving is converted to compressed digital form, thus avoids the waste of resource.Be and while sampling, data compressed, be equivalent to and find minimum coefficient to represent signal in the process of sampling, and with suitable restructing algorithm, raw data can be recovered.So, a small amount of low-dimensional data after compression contains the full detail of original high dimensional data.Theoretically, the data after compression meet classification demand.
Summary of the invention
The invention reside in and a kind of electrocardiosignal sorting technique is provided, to solve the contradiction between feature extraction complexity and accuracy, and cannot the problem of real-time grading.
The invention reside in and provide a kind of heart real time Modulation recognition method based on sparse projection, described method comprises:
(1) data prediction:
carrying out filtering to leading electrocardiosignal more, removing the noise such as Hz noise and baseline wander;
wave test, waveform partition;
to the heartbeat data standardization split;
(2) feature extraction: the principle based on compressed sensing is compressed data, reduces characteristic dimension, calculates RR interval and RR weights, and splicing proper vector, forms quadratic character;
(3) classify: quadratic character data are divided into training data and test data, and training data does classification model construction, and the sorter after test data puts into modeling is tested;
(4) Decision Classfication: multi-lead classification results data fusion, uses probability function to do last classification.The electrocardiogram (ECG) data sorting technique that the present invention proposes, classification results is relatively more accurate, utilizes compressed sensing to reduce data characteristics dimension simultaneously, improves efficiency of algorithm and data-handling capacity.
The beneficial effect that the present invention compared with prior art exists is:
sparse projection coefficient is adopted to have better robustness as the proper vector that electrocardiosignal is classified;
by carrying out standardization to monocycle electrocardiosignal, can the impact of guiding principle amount between different variable;
classifying quality can be improved to RR characteristic weighing coefficient;
multi-lead classification results data fusion can reduce False Rate.
Accompanying drawing explanation
Fig. 1 is the process flow diagram of the heart real time sorting technique based on sparse projection.
Fig. 2 is the design sketch of Filtering of ECG Signal.
Fig. 3 is R spot check mapping.
Fig. 4 is that single-unit claps heartbeat figure.
Fig. 5 is sparse projection feature ten folding cross validation test pattern.
Fig. 6 is ten folding cross validation test patterns of sparse projection feature+weights RR feature.
Fig. 7 is multi-lead beat classification process flow diagram.
Embodiment
Below in conjunction with accompanying drawing, the heart real time Modulation recognition method based on sparse projection given by the present invention is described in detail.
Fig. 1 is the process flow diagram of heartbeat signal sorting technique, comprises following step:
(1) data prediction: often lead electrocardiosignal carry out filtering to leading more, removes Hz noise and baseline wander; Wave test, waveform partition; To the heartbeat data standardization split;
(2) feature extraction: sparse projection feature+weights RR spaced features
(3) classify: characteristic is divided into training data and test data, training data does classification model construction, and the sorter after test data puts into modeling does simulation test;
(4) Decision Classfication: multi-lead classification results data fusion, uses probability function to do last classification.
Fig. 2 is the design sketch of Filtering of ECG Signal, and upper figure is the original electro-cardiologic signals that singly leads collected, and middle figure removes the electrocardiosignal after noise with wavelet transformation, and figure below is the electrocardiosignal utilizing wavelet transformation to remove baseline wander.The wavelet transformation of signal is equivalent to signal and leads to and low-pass filtering at the band of different scale, and decomposable asymmetric choice net obtains approximation component and the details coefficients of signal.The noisy part of electrocardiosignal mainly concentrates in high-frequency wavelet coefficient.Further, the wavelet coefficient amplitude of useful signal is comparatively large, and number is less; And wavelet coefficient amplitude corresponding to noise is little and number is more.Therefore, we adopt threshold value method to process wavelet coefficient, remove noise.Baseline wander is mainly low frequency component, by the approximation component zero setting of high yardstick, just obtains the signal of baseline wander.Concrete operations are as follows: we select db5 to carry out 3 layers of decomposition as wavelet function, utilize and estimate threshold value, according to the threshold denoising chosen and reconstruction signal without partial likelihood; We select db5 to carry out 6 layers of wavelet decomposition as wavelet function, by the approximation component zero setting of high yardstick, remove baseline wander.
Fig. 3 is R spot check mapping, and concrete operations are as follows: carry out multiscale analysis to signal, and when sudden change appears in signal, the coefficient after its wavelet transformation has modulus maximum value, thus can by determining to the detection of modulus maximum point the time point that fault occurs.
Fig. 4 is that single-unit claps heartbeat figure, and concrete operations are as follows: with the R detected point for basic point gets forward 100 points, get 200 points backward, obtain the heartbeat of 300 sampled points.Subsequently, in order to eliminate the impact of guiding principle amount between different variable, clap heartbeat group to single-unit and carry out standardization, namely average is 0, and variance is 1.
Fig. 5 is sparse projection feature ten folding cross validation test pattern, and we are to five class beat classification, is normal pollex heartbeat, atrial premature beats, premature ventricualr contraction, left bundle branch block, right bundle branch block respectively.Can see that measuring accuracy is stabilized in about 95%, and concrete operations are as follows when data dimension drops to about 10 from 300 time: to every class heartbeat, Stochastic choice 50 samples, and to each sample mark label.Adopt the mode of accidental projection, while on the low n-dimensional subspace n point set in higher-dimension theorem in Euclid space being embedded into a Stochastic choice, guarantee that Euclidean distance between any two points that former high-dimension space point is concentrated is maintained lower dimensional space is approximate.To the quadratic character with label, adopt the mode of SVM classifier and ten folding cross validations, that is, quadratic character data set be divided into ten parts, in turn will wherein nine parts as training data, a as test data, test.Each test all can draw corresponding accuracy.The accuracy mean value of the result of 10 times is as the estimation to arithmetic accuracy, and the validity of feature is extracted in inspection.
Fig. 6 is ten folding cross validation test patterns of sparse projection feature+weights RR feature, can find out that degree of accuracy can reach 98.8% when RR weights are 8,9.Concrete operations are as follows: detect and calculate front RR interval, this RR interval, rear RR interval, front RR interval refers to: the interval between current R point and previous R point; This RR interval refers to: the equispaced of current R point and rear 10th R point; Rear RR interval refers to: the interval between current R point and a rear R point.Then, selection dimension is the accidental projection feature of 30, by the feature after accidental projection and weights RR spaced features, is spliced into proper vector, forms quadratic character.To the quadratic character with label, adopt the mode of SVM classifier and ten folding cross validations, the accuracy mean value of the result of 10 times is as the estimation to arithmetic accuracy, and the validity of feature is extracted in inspection.
Fig. 7 is multi-lead classification process figure.Concrete operations are as follows: the multi-lead quadratic character with label is randomly drawed 30% as training data, and 70% as test data.Select gaussian kernel function to carry out SVM training to training data, the sorter after test data puts into modeling does simulation test, draws classification results.Multi-lead classification results data fusion, does final decision classification, and namely the classification results of holocentric jumping multi-lead merges.If each classification results led is consistent, then get this classification results; If result is divided into the number of leading of category-A more than half, be then judged to category-A; If which kind of classification results do not have more cross 1/2 and to lead number, then refused to sentence.
Claims (5)
1., based on the heart real time sorting technique of accidental projection, it is characterized in that, comprise the following steps:
Step 1, data prediction:
filtering is carried out to electrocardiosignal, removes the undesired signal such as Hz noise and baseline wander;
wave test, waveform partition;
to the heartbeat data standardization split;
Step 2, feature extraction: compress data based on compressed sensing, reduce characteristic dimension, calculates RR interval, calculates RR weights, and splicing proper vector, forms quadratic character;
Step 3, classification: quadratic character data are divided into training data and test data, and training data does classification model construction, and the sorter after test data puts into modeling is tested;
Step 4, Decision Classfication: use probability function to do last classification to multi-lead classification results.
2. the heart real time sorting technique based on accidental projection according to claim 1, is characterized in that: the mode adopting accidental projection in step 2, the point set in higher-dimension theorem in Euclid space is embedded into the low n-dimensional subspace n of a Stochastic choice.
3. the heart real time sorting technique based on accidental projection according to claim 1, is characterized in that: the accidental projection adopted in step 2 refers to
Y
i=ΦΧ
i
Wherein, Φ is the stochastic matrix of a M × N, M<N; X
ibe through the N × 1 heartbeat sample of step 1, routine X
1={ a
1, a
2... a
n-1, a
n; Y
ithe heart bat short data records to be sorted of M × 1, routine Y
1={ b
1,b
2,... b
m-1,b
m.
4. the heart real time sorting technique based on accidental projection according to claim 1, it is characterized in that: calculate RR weights in step 2 and refer to, weights are changed within the scope of 0-100, export the nicety of grading of test set, according to best result class precision, choose best initial weights.
5. the heart real time sorting technique based on accidental projection according to claim 1, is characterized in that: the electrocardiosignal gathered in step 4 carries out Decision Classfication and comprises finger: if each classification results led is consistent, then get this classification results; If result is divided into the number of leading of category-A more than half, be then judged to category-A; If which kind of classification results do not have more cross 1/2 and to lead number, then refused to sentence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510988099.5A CN105550659A (en) | 2015-12-25 | 2015-12-25 | Real-time electrocardiogramclassification method based on random projection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510988099.5A CN105550659A (en) | 2015-12-25 | 2015-12-25 | Real-time electrocardiogramclassification method based on random projection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105550659A true CN105550659A (en) | 2016-05-04 |
Family
ID=55829842
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510988099.5A Pending CN105550659A (en) | 2015-12-25 | 2015-12-25 | Real-time electrocardiogramclassification method based on random projection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105550659A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106037714A (en) * | 2016-05-24 | 2016-10-26 | 华中科技大学 | Heart beat monitoring device |
| CN106108880A (en) * | 2016-06-28 | 2016-11-16 | 吉林大学 | A kind of heart claps automatic identifying method and system |
| CN106377247A (en) * | 2016-09-10 | 2017-02-08 | 天津大学 | Feature selection-based arrhythmia classification method |
| CN106963360A (en) * | 2017-03-06 | 2017-07-21 | 四川大学 | A kind of eigenmatrix design method recognized for electrocardio |
| CN107958207A (en) * | 2017-11-09 | 2018-04-24 | 四川大学 | A kind of electrocardiogram recognition method of extremely low data volume |
| CN108537123A (en) * | 2018-03-08 | 2018-09-14 | 四川大学 | Electrocardiogram recognition method based on multi-feature extraction |
| CN109077715A (en) * | 2018-09-03 | 2018-12-25 | 北京工业大学 | A kind of electrocardiosignal automatic classification method based on single lead |
| CN110226140A (en) * | 2017-01-25 | 2019-09-10 | Ntn株式会社 | State monitoring method and state monitoring apparatus |
| CN110367936A (en) * | 2019-08-05 | 2019-10-25 | 广州视源电子科技股份有限公司 | Electrocardiograph signal detection method and device |
| CN111759298A (en) * | 2020-07-10 | 2020-10-13 | 齐鲁工业大学 | Method for reducing arrhythmia false alarm rate of multi-parameter monitor |
| CN113011462A (en) * | 2021-02-22 | 2021-06-22 | 广州领拓医疗科技有限公司 | Classification and device of tumor cell images |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1951320A (en) * | 2006-11-16 | 2007-04-25 | 上海交通大学 | Abnormal electrocardiogram recognition method based on ultra-complete characteristics |
| US20120101401A1 (en) * | 2009-04-07 | 2012-04-26 | National University Of Ireland | Method for the real-time identification of seizures in an electroencephalogram (eeg) signal |
-
2015
- 2015-12-25 CN CN201510988099.5A patent/CN105550659A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1951320A (en) * | 2006-11-16 | 2007-04-25 | 上海交通大学 | Abnormal electrocardiogram recognition method based on ultra-complete characteristics |
| US20120101401A1 (en) * | 2009-04-07 | 2012-04-26 | National University Of Ireland | Method for the real-time identification of seizures in an electroencephalogram (eeg) signal |
Non-Patent Citations (2)
| Title |
|---|
| 刘训利: "基于线性表示模型的人脸识别研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 商卫波: "心电信号自动分析与诊断处理方法研究", 《中国优秀博硕士学位论文全文数据库(硕士)医药卫生科技辑》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106037714A (en) * | 2016-05-24 | 2016-10-26 | 华中科技大学 | Heart beat monitoring device |
| CN106108880A (en) * | 2016-06-28 | 2016-11-16 | 吉林大学 | A kind of heart claps automatic identifying method and system |
| CN106377247A (en) * | 2016-09-10 | 2017-02-08 | 天津大学 | Feature selection-based arrhythmia classification method |
| CN110226140A (en) * | 2017-01-25 | 2019-09-10 | Ntn株式会社 | State monitoring method and state monitoring apparatus |
| CN106963360A (en) * | 2017-03-06 | 2017-07-21 | 四川大学 | A kind of eigenmatrix design method recognized for electrocardio |
| CN107958207A (en) * | 2017-11-09 | 2018-04-24 | 四川大学 | A kind of electrocardiogram recognition method of extremely low data volume |
| CN108537123A (en) * | 2018-03-08 | 2018-09-14 | 四川大学 | Electrocardiogram recognition method based on multi-feature extraction |
| CN109077715A (en) * | 2018-09-03 | 2018-12-25 | 北京工业大学 | A kind of electrocardiosignal automatic classification method based on single lead |
| CN109077715B (en) * | 2018-09-03 | 2021-09-17 | 北京工业大学 | Electrocardiosignal automatic classification method based on single lead |
| CN110367936A (en) * | 2019-08-05 | 2019-10-25 | 广州视源电子科技股份有限公司 | Electrocardiograph signal detection method and device |
| CN111759298A (en) * | 2020-07-10 | 2020-10-13 | 齐鲁工业大学 | Method for reducing arrhythmia false alarm rate of multi-parameter monitor |
| CN113011462A (en) * | 2021-02-22 | 2021-06-22 | 广州领拓医疗科技有限公司 | Classification and device of tumor cell images |
| CN113011462B (en) * | 2021-02-22 | 2021-10-22 | 广州领拓医疗科技有限公司 | Classification and device of tumor cell images |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105550659A (en) | Real-time electrocardiogramclassification method based on random projection | |
| CN107714023B (en) | Static electrocardiogram analysis method and device based on artificial intelligence self-learning | |
| US11529103B2 (en) | Artificial intelligence self-learning-based automatic electrocardiography analysis method and apparatus | |
| US11234629B2 (en) | Method and device for self-learning dynamic electrocardiography analysis employing artificial intelligence | |
| US11476000B2 (en) | Methods and systems using mathematical analysis and machine learning to diagnose disease | |
| Thomas et al. | Automatic ECG arrhythmia classification using dual tree complex wavelet based features | |
| CN109171712B (en) | Atrial fibrillation identification method, atrial fibrillation identification device, atrial fibrillation identification equipment and computer readable storage medium | |
| EP3672474A1 (en) | A method of detecting abnormalities in ecg signals | |
| CN109077715A (en) | A kind of electrocardiosignal automatic classification method based on single lead | |
| Kaya et al. | Feature selection using genetic algorithms for premature ventricular contraction classification | |
| CN113095302B (en) | Depth model for arrhythmia classification, method and device using same | |
| CN109948396B (en) | Heart beat classification method, heart beat classification device and electronic equipment | |
| CN112971795B (en) | Electrocardiosignal quality evaluation method | |
| Bruun et al. | Automatic atrial fibrillation detection: A novel approach using discrete wavelet transform and heart rate variability | |
| Zairi et al. | Intelligent system for detecting cardiac arrhythmia on FPGA | |
| Kar et al. | A technical review on statistical feature extraction of ECG signal | |
| Sathawane et al. | Prediction and analysis of ECG signal behaviour using soft computing | |
| Hussein et al. | Detection of electrocardiogram QRS complex based on modified adaptive threshold | |
| Jiang et al. | Heartbeat classification system based on modified stacked denoising autoencoders and neural networks | |
| CN110507299A (en) | Heart rate signal detection device and method | |
| Nanarkar et al. | A Survey on Classification and identification of Arrhythmia using Machine Learning techniques | |
| Saeed et al. | A Slantlet based statistical features extraction for classification of normal, arrhythmia, and congestive heart failure in electrocardiogram | |
| Vázquez-Seisdedos et al. | A comparison of different classifiers architectures for electrocardiogram artefacts recognition | |
| Jokić et al. | An efficient approach for heartbeat classification | |
| Zhang et al. | Evaluation of single-lead ECG signal quality with different states of motion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20160504 |