CN107622259B

CN107622259B - T wave detection method, electrocardiogram data analysis method and device

Info

Publication number: CN107622259B
Application number: CN201711004462.0A
Authority: CN
Inventors: 魏守水; 尚海霞; 刘飞飞
Original assignee: Shandong University
Current assignee: Shandong University
Priority date: 2017-10-20
Filing date: 2017-10-20
Publication date: 2019-12-24
Anticipated expiration: 2037-10-20
Also published as: CN107622259A

Abstract

The invention provides a T wave detection method based on a sliding window integral area. The search window boundary setting rule is improved to be adaptive. First, three piecewise functions are generated after analyzing the relationship between RR intervals and RTon intervals (from R peak to the start of T wave) and the relationship between RR intervals and RToff intervals (from R peak to the end of T wave) using a k-means clustering method based on an artificially labeled MIT QT database. A grid search is then used to determine a combination of parameters that fits the sliding window integration area method. Finally, we evaluated the effectiveness of the method on the QT database and the newly labeled european ST-T database, selecting 100ms as the time margin for true positives, and using the traditional sliding window integral area method as a comparison method.

Description

T wave detection method, electrocardiogram data analysis method and device

Technical Field

The invention relates to a T wave detection method, an electrocardiogram data analysis method and an electrocardiogram data analysis device.

Background

The electrocardiogram data mainly comprises an Electrocardiogram (ECG) characterized by a plurality of waveforms such as P wave, QRS wave, T wave and the like, and basic physiological and pathological data of clinical diagnosis can be provided by reasonably reading the ECG. In recent years, the advent of many wearable monitoring devices has made it possible to monitor ECG signals in everyday life, and thus a large amount of ECG data is generated each day, but it is impossible for a physician to manually review or diagnose each ECG signal. Therefore, it is crucial to develop accurate methods for automatically analyzing cardiac electrical signals, especially for ambulatory ECG monitoring.

The electrocardiogram mainly comprises several waveforms, such as a P wave, a QRS wave and a T wave, and time interval information between starting points and ending points of different waveforms, however, accurate or robust T wave detection (including the starting points and the ending points of the T waves) is still challenging due to large changes of T wave states. Although many T-wave detection methods based on different technologies exist at present, the detection method based on wavelet transform is sensitive to noise and is easily influenced by noise; mathematical model-based methods require the creation of a robust generic ECG template, but when the waveform varies widely, it is not possible to create a generic template.

Zhang et al first proposed a sliding window integral area method to detect the T-wave endpoint and confirmed the effectiveness of this method on the QT database. Subsequently, Song et al improved this method for detecting T-wave origins. However, the sliding window integration method proposed by Zhang team or Song team is non-adaptive in parameter setting, and they adopt empirical values and do not include a parameter optimization step, so that the existing sliding window integration region method has limitation in application.

Disclosure of Invention

Aiming at the defects in the prior art, the invention provides a T wave detection method based on the integral area of a sliding window, which is applied to electrocardiogram data detection, clusters a k mean value of a data mining technology and is used for setting a search boundary, optimizes parameters by using a grid search strategy, and has better performance in a test display improvement method on a QT database and a European ST-T database.

The technical scheme of the invention is as follows:

a T wave detection method based on sliding window integral area is applied to electrocardio data detection and comprises the following steps:

establishing a decision function for determining a sliding window search boundary by adopting a clustering method;

training the decision function by using a known database, and determining a variable parameter combination of the decision function;

and giving the variable parameter combination to a decision function, determining a search boundary of the sliding window according to the decision function, and when the area enclosed by the sliding window and the electrocardiographic waveform in the search boundary range reaches the maximum value, determining the starting point/end point of the search boundary as the starting point/end point of the T wave.

Further, determining the variable parameter combination comprises:

training is carried out by using a QT database, the parameter combination is selected when the obtained F1 measure value is the highest, and the F1 measure is as follows:

wherein TP is true positive, FP is false positive FP, and FN is false negative.

Further, the clustering method adopts a k-means mean clustering method, and the establishment of the decision function for determining the sliding window search boundary by adopting the k-means mean clustering method comprises the following steps:

RR interval and RT are obtained by adopting a k-means mean clustering method_onExtent of intervals, and RR intervals and RT_offA range of intervals;

according to RR interval and RT_onEstablishing a starting point decision function of a sliding window searching boundary in the interval range;

according to RR interval and RT_offEstablishing an end point decision function of a sliding window searching boundary in the interval range;

wherein RT is_onThe interval refers to the time interval between the R peak and the start of the T wave, RT_offThe interval refers to the time interval between the R peak and the end of the T wave.

Further, selecting a variable parameter combination in a starting point decision function of the sliding window search boundary by using a grid search method; and selecting variable parameter combinations in the endpoint decision function of the sliding window search boundary by using a grid search method.

On the basis of the T wave detection method, the invention also provides an electrocardiogram data analysis method, which comprises the following steps: the electrocardio data are received, the starting point and the end point of the P wave, the Q wave, the R wave and the S wave are respectively detected, the starting point and the end point of the T wave are detected by adopting the method, and the electrocardio characteristic values among the waves are calculated by utilizing the starting point and the end point positions of the R wave, the P wave, the T wave, the Q wave and the S wave.

Further, filtering and denoising are carried out on the received electrocardio data.

Further, the filtering and denoising method comprises median filtering, wavelet filtering or amplitude-limiting filtering.

The invention also provides a storage device, which stores a plurality of instructions, wherein the instructions are loaded by a processor and execute the following processing:

the electrocardio data are received, the starting point and the end point of the P wave, the Q wave, the R wave and the S wave are respectively detected, the starting point and the end point of the T wave are detected by adopting the method, and the electrocardio characteristic values among the waves are calculated by utilizing the starting point and the end point positions of the R wave, the P wave, the T wave, the Q wave and the S wave.

Further, another technical scheme of the invention is an electrocardiogram data analysis device, which comprises a processor, a storage unit and a processing unit, wherein the processor is used for realizing each instruction; and storage means for storing a plurality of instructions, the instructions being loaded by the processor and performing the following:

The invention also provides another storage device, which stores a plurality of instructions, wherein the instructions are loaded by a processor and execute the following processing:

Further, another technical solution of the present invention is a T-wave detection apparatus based on a sliding window integration area, including a processor for implementing each instruction; and storage means for storing a plurality of instructions, the instructions being loaded by the processor and performing the following:

The invention has the beneficial effects that:

the invention provides an improved sliding window integral area method which is used for detecting a starting point and an end point of a T wave, has self-adaptive parameter setting, clusters a data mining technology k mean value and is used for setting a search boundary, and optimizes parameters by using a grid search strategy. Tests on the QT database and the european ST-T database showed better performance of the improved method.

For the QT database T wave start detection in the invention, the F1 measurement of the first channel is improved from 54.70% to 70.46%, and the F1 measurement of the second channel is improved from 54.05% to 72.94%. For QT database T-wave endpoint detection, the F1 measure for the two ECG channels increased from 87.83% to 93.73% and from 86.73% to 94.75%, respectively. When tested on the european ST-T database, the F1 measure for the first channel increased from 41.02% to 84.13% and the F1 measure for the second channel increased from 44.33% to 87.62% for T-wave onset detection. For T-wave endpoint detection, the F1 measure for the first channel increased from 98.83% to 99.57%, and the F1 measure for the second channel increased from 91.76% to 98.29%. On the detection of two databases, the method provided by the invention has higher F1 measurement than the F1 measurement of the traditional method.

Drawings

FIG. 1 is a graph of the probability density distribution of the T-wave start/end and R-wave peak time intervals labeled in the present invention (A) the QT database (B) the European ST-T database;

FIG. 2 is a diagram illustrating the use of the SWIA method for T-wave start point detection

FIG. 3 is a diagram illustrating the use of the conventional SWIA method for T-wave endpoint detection

FIG. 4 shows the clustering results of T-wave feature points (A) of the start point of T-wave and (B) of the end point of T-wave

FIG. 5 is a flow chart of a method of T-wave start and end point detection

FIG. 6 marked with solid dots (& cndot.) is the result of our improved method, open circles: (O and O) Is the result of the conventional approach, the shaded portion is acceptable for the "TP" case, (a) e 0107; (B) e 0111; (C) e0118.

The specific implementation mode is as follows:

the invention will be further illustrated with reference to the following examples and drawings:

it should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.

The invention finds particular application in two databases, including the QT database, which provides 105 two-channel ECG recordings at a 15 minute sampling rate of 250Hz, and the european ST-T database, which has 43 recordings with artificially labeled T-wave start points and 103 recordings with artificially labeled T-wave end points. Typically 30 to 100 representative beats are labeled in each labeled record. Table 1 gives the label information for the summarized QT database.

The european ST-T database is intended for evaluation of methods related to analyzing ST-segment and T-wave changes. The database was recorded from 90 electrocardiograms at a sampling rate of 250Hz for 2 channels at 2 hours, without manual labeling of the start and end points of the T wave. Thus in this study we picked the first 5 minute signal of 23 recordings and manually labeled the start and end points of the T wave by an experienced clinical expert. If there is a serious signal quality problem in the first 5 minutes of recording, the recording after 5 minutes is used. Table 1 gives the detailed annotation information for the European ST-T database.

To verify the consistency of the labeling information between the two databases, we analyzed the time interval between the T-wave start/end and the R-wave peak position of the two databases. Figure 1 shows the probability density distribution of these time intervals in the two databases. As shown in FIG. 1, the T-wave start/end points of our manually labeled European ST-T database have similar probability density distributions to the annotations in the QT database.

TABLE 1 summary of QT database and European ST-T database annotation information

Description of the drawings: where Dis _ qrs represents the time interval between the T-wave start/end and the R-wave position of the current beat,

SD represents the standard deviation.

The description of the detailed annotation information for the QT database comes from the web pages:

https://www.physionet.org/physiobank/database/qtdb/doc/index.shtml.

the SWAI (sliding window integrated area) method detects the start and end points of a T wave by calculating the area of the ECG waveform within a sliding window (the algorithm is described with a normal form of the T wave). When the area of the sliding window within the search range fixed in advance reaches the maximum value, the start/end point of the T wave within the current sliding window is detected. Next, we will illustrate the mechanism of the method (see fig. 2 and 3) by way of example (taking a forward T wave as an example). In addition, in the actual process, since the form type of the T wave is not known in advance, a threshold r needs to be set in the subsequent calculation to determine the waveform form of the T wave.

Fig. 2 illustrates a detection mechanism of the start point of the T wave. First, the R-wave peak is located, and then the left and right boundaries T1, T2 of the T-wave origin search window are determined based on the value of the current RR interval:

wherein R isR_iIs the ith RR interval and Ri is the ith position of the R peak. At a fixed sliding window [ t, t + w ]]In, At is defined as follows:

where w is 0.12s the window width, t is from t1 to t2, Sj is the waveform amplitude of the jth sample point,is the local average amplitude (window where p ═ 0.016 s), and is defined as:

as shown in fig. 2, when t is t_onAt reaches its maximum value.

Fig. 3 shows the detection of the T-wave endpoint. First, the location of the R-wave peak is located, and then the left and right boundaries T3 and T4 of the T-wave endpoint search range are determined based on the value of the current RR interval:

then, within a fixed sliding window [ t-w t ], the waveform area At is calculated:

where w is 0.128s, t is a number from t3 to t4, Sj,this is illustrated in equation (2). When t ═ tend, as shown in fig. 3, At reaches its maximum value.

One key issue when using the SWIA method is to determine the search boundaries accurately, but the setting of the search boundaries is closely related to the size of the RR interval. The conventional SWIA method uses two piecewise functions and predefined parameters to setA boundary is searched. To more accurately model the relationship between RR-intervals and search boundaries, we use clustering methods to analyze, particularly preferably k-means clustering, RR-intervals and RT_on(RT_on: representing the time interval between the R peak and the start of the T wave) and the RR interval and RT_off(RT_off: representing the time interval between the R peak and the T wave offset).

Fig. 4 shows a scatter plot with optimal k-means clustering (k-3). RR intervals and RT_onAnd RR intervals and RT_offThe relationship between them is as follows:

therefore, the decision functions obtained by the three piecewise function wave start and end point detections for determining the search boundary of T are as follows:

then, the optimal combination of parameters in functions (8) and (9) is determined using a grid search method. The variable parameter combinations at the start of the T-wave are ald 0.4, alu 0.2, ard 0.32, aru 0.32, amd 0.3, amu 0.0

The variable parameter combination at the end of the T-wave is ald 0.2, alu 0.1, ard 0.1, aru 0.0, amd 0.0, amu 0.1.

For T-wave origin detection, we set r to 6; for T-wave endpoint detection, we set r to 5.

TABLE 2 parameter set information for detecting T-wave

The method of the invention can determine the True Positive (TP), False Positive (FP) and False Negative (FN) detected with a threshold of 100ms, as shown in FIG. 5. Using sensitivity (Se), forward prediction (P +) and F1 measures as evaluation indices:

fig. 6 gives an example of the detection of the proposed method as well as the conventional method.

We first tested the method proposed by the present invention on the QT database, compared to Song's T-wave start detection method and Zhang's T-wave end detection method.

Table 3 gives the database of start and end detection results for QT, using two ECG channel signals (first and second channels), respectively.

As shown in table 3, the method provided by the present invention significantly improves the detection accuracy of the start point and the end point detection. For the T-wave onset detection, the F1 measure for the first channel increased from 54.70% to 70.46%, and the F1 measure for the second channel increased from 54.05% to 72.94%. For T-wave endpoint detection, the F1 measure for the first channel increased from 87.83% to 93.73%, and the F1 for the second channel increased from 86.73% to 94.75%, respectively. In addition, detection errors were also analyzed. As expected, the improved swaa method reported less detection error than the conventional method, except that the T-wave endpoint detection for the second ECG channel was slightly increased (conventional 0.027 ± 31.85ms vs. improvement of 2.45 ± 33.98 ms). However, it is noted that all of the Se, P + and F1 indices have risen from 86% to 94%.

TABLE 3T-wave detection of QT database

Table 4 shows the results of the start and end point detection of the European ST-T database. The method provided by the invention can be used for more remarkably improving the T wave starting point detection. The F1 measure for the first channel increased from 41.02% to 84.13%, and the F1 measure for the second channel from 44.33% to 87.62%. The average detection error of the two channels is significantly reduced from 19.52ms to 7.04ms and from 26.27ms to 6.35ms respectively. Although the improvement in endpoint detection performance was slight, our improvement was effective because the F1 measurements for both channels increased from 98.83% to 99.57% and 91.76% to 98.29%, respectively. However, the average detection error increases slightly when running the new method but this is not critical to the problem.

TABLE 4. T-wave detection information of European database

As can be seen from tables 3-4, the method proposed by the present invention has better performance than the conventional method for the detection of the start and end points of T-waves, which indicates that the setup involving the use of clustering techniques for searching for boundaries in SWAI method helps to improve the accuracy of detection. Clustering is a statistical-based technique that finds independent parts belonging to different populations by quantitative analysis and comparison of multiple features. Furthermore, we note that the detection of the start of the T-wave, neither the conventional SWIA nor the modified version, results in better performance than the detection of the end of the T-wave. One possible reason is that clustering methods determine search boundaries based on statistical techniques. Therefore, the accuracy of the clustering result is related to the size of the data volume. However, the labeled T-wave start points in the QT database are much smaller than the number of labeled T-wave end points (1371 vs 3452). Therefore, the relationship found by cluster analysis between RR intervals and RTon is not strong (see fig. 4(a)) and the relationship between RR intervals and RToff (see fig. 4 (B)). A significant difference between the SWIA of the present invention and the conventional SWIA is that the present invention can more broadly and adaptively determine the search window, which implements the boundaries by using k-means clustering and grid search strategies. However, the conventional swaa uses only predefined parameters, and does not give any detailed description.

In addition, we also note that the standard deviation of the T-wave endpoint detected by the improved method is 25.82ms, and that the standard deviation detected by the conventional swaa is 21.19 ms. They were all electrocardiogram working groups (CSE) below the acceptable threshold (30.6ms) set forth in the general standard. And Martinez et al also adopts the same method as Zhang to calculate indexes when evaluating a wavelet-based method in a QT database to detect T waves. In addition, the index was calculated by verifying the result on the QT database using the same method based on the low-pass differential method, the hidden markov model-based method, the partial gibbs sample and bayes method, and the TU complex analysis method (their results are given in table 5). However, when calculating Se and P +, they do not give their time tolerance, nor do they state how to deal with the mean and standard deviation of error errors in the calculations. For comparison with their results, we also used the same method to calculate some indicators, using the annotated feature points minus the detection points, and then obtain the mean and standard deviation of the error. Based on the signals of the two channels, each feature point is located twice, so that two positions can be obtained. Then, we select the detected position with smaller error. If the absolute value of the error is greater than 100ms, we ignore the error for the current beat. Furthermore, the above methods have similarities, and their training and test sets of data are all from the same database. However, in the present invention, the QT database is selected as the training set and the European ST-T database is used only as the test set. The present invention uses the F1 measure as the final determinant rather than the mean and deviation of the error as the determinant. To show the results of the invention more clearly, the error between annotation and automatic detection by the method was also calculated, as well as the mean and absolute standard deviation of the data in error set with error values less than or equal to 100ms were calculated.

TABLE 5 comparison (sampling rate 250Hz) of unknown end-point detection methods for T-waves using QT database

Based on the above method, another embodiment of the present invention is a storage device storing a plurality of instructions, the instructions being loaded by a processor and performing the following:

establishing a decision function for determining a search boundary of a sliding window by adopting a k-means mean clustering method;

and giving the variable parameter combination to a decision function, determining a search boundary of the sliding window according to the decision function, and when the area enclosed by the sliding window and the electrocardiographic waveform in the search boundary range reaches the maximum value, determining the starting point and the end point of the search boundary as the starting point and the end point of the T wave.

Further, another embodiment of the present invention is a T-wave detecting apparatus based on a sliding window integration area, comprising a processor for implementing each instruction; and storage means for storing a plurality of instructions, the instructions being loaded by the processor and performing the following:

The invention also provides an electrocardio data analysis method based on the T wave detection method, which comprises the following steps: the electrocardio data are received, the starting point and the end point of the P wave, the Q wave, the R wave and the S wave are respectively detected, the starting point and the end point of the T wave are detected by adopting the method, and the electrocardio characteristic values among the waves are calculated by utilizing the starting point and the end point positions of the R wave, the P wave, the T wave, the Q wave and the S wave.

Filtering and denoising the received electrocardiographic data may be performed prior to processing the electrocardiographic data.

The filtering and denoising method includes median filtering, wavelet filtering or clipping filtering, but is not limited to the above-mentioned filtering and denoising method.

On the basis of the electrocardio-data analysis method, the invention also provides a storage device, which stores a plurality of instructions, and the instructions are loaded by a processor and execute the following processing:

Further, another embodiment of the present invention is an electrocardiographic data analyzing apparatus, including a processor for implementing instructions; and storage means for storing a plurality of instructions, the instructions being loaded by the processor and performing the following:

The starting point and the end point of the P wave, the Q wave, the R wave and the S wave can be selected by adopting a mode identification method, and then the starting point and the end point can be obtained by a positioning method.

The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A T wave detection method based on sliding window integral area is applied to electrocardio data detection and is characterized by comprising the following steps:

the variable parameter combination is endowed to a decision function, a search boundary of the sliding window is determined according to the decision function, and when the area enclosed by the sliding window and the electrocardiographic waveform in the search boundary range reaches the maximum value, the starting point and the end point of the search boundary are the starting point and the end point of the T wave;

the clustering method is a k-means mean clustering method, and the step of establishing a decision function for determining the search boundary of the sliding window by adopting the k-means mean clustering method comprises the following steps:

obtaining RR interval and RT according to clustering result_onEstablishing a starting point decision function of a sliding window searching boundary in the interval range;

obtaining RR interval and RT according to clustering result_offEstablishing an end point decision function of a sliding window searching boundary in the interval range;

wherein RT is_onThe interval refers to the time interval between the R peak and the start of the T wave, RT_offThe interval refers to the time interval between the R peak and the T wave offset.

2. The method of claim 1, wherein a first variable parameter combination in the starting point decision function of the sliding window search boundary is selected using a grid search method; and selecting a second variable parameter combination in the endpoint decision function of the sliding window search boundary by using a grid search method.

3. An electrocardiogram data analysis method comprises the following steps: receiving electrocardiogram data, and respectively detecting the starting point and the end point of P wave, Q wave, R wave and S wave, which is characterized in that: the method according to claim 1, wherein the start point and the end point of the T wave are detected, and the electrocardio characteristic values among the waves are calculated by using the start point and the end point positions of the R wave, the P wave, the T wave, the Q wave and the S wave.

4. The method of claim 3, wherein the received electrocardiographic data is filtered to reduce noise.

5. The method of claim 4, wherein the filtering for denoising comprises median filtering, wavelet filtering, or clipping filtering.

6. A memory device storing a plurality of instructions, the instructions being loaded by a processor and performing the following:

receiving electrocardiogram data, respectively detecting the starting point and the end point of P waves, Q waves, R waves and S waves, detecting the starting point and the end point of T waves by adopting the method as claimed in claim 1, and calculating electrocardiogram characteristic values among the waves by utilizing the positions of the starting points and the end points of the R waves, the P waves, the T waves, the Q waves and the S waves.

7. An electrocardiogram data analysis device comprises a processor, a storage unit and a processing unit, wherein the processor is used for realizing instructions; and storage means for storing a plurality of instructions, wherein the instructions are loaded by the processor and perform the following:

8. A memory device storing a plurality of instructions, the instructions being loaded by a processor and performing the following:

9. A T wave detection device based on a sliding window integral area comprises a processor, a first detection module, a second detection module and a third detection module, wherein the processor is used for realizing instructions; and storage means for storing a plurality of instructions, wherein the instructions are loaded by the processor and perform the following: