JP2001218747A - Electrocardiogram analysis system and computer readable recording medium having program therefor recorded thereon - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an electrocardiogram analysis system capable of automatically extracting a target waveform from electrocardiogram waveforms, reducing the time and labor of electrocardiogram analysis and improving the reliability and a computer readable recording medium having a program recorded thereon. SOLUTION: This analysis system is provided with a transformation and extraction part 11 for obtaining a feature amount by wavelet-transforming electrocardiogram waveform data and extracting a maximum value or a minimum value from the feature amount and a pattern analysis part 12 for extracting the target waveform included in the electrocardiogram waveform by pattern-analyzing the maximum value or the minimum value extracted in the transformation and extraction part 11. By the system, the target waveform is automatically extracted from the electrocardiogram waveform, the time and labor of the electrocardiogram analysis are reduced and the reliability is improved by performing objective analysis.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an electrocardiogram analyzing system for extracting various target waveforms such as R wave and T wave from electrocardiogram waveform data. Further, the present invention relates to a computer-readable recording medium on which a program for realizing such extraction of various target waveforms is recorded.

[0002]

2. Description of the Related Art In order to diagnose and analyze various states of the human body, it is important to grasp the state of the heart, which is the source of the human body. As one method for this, acquisition of an electrocardiogram has been conventionally performed. In the electrocardiogram, a potential difference (action potential) of an electromotive force generated due to expansion and contraction of each part of the heart is derived from the surface of the skin, and is amplified and recorded. FIG. 16 shows an example of this electrocardiogram. As shown in FIG. 16, an electrocardiogram waveform (electrocardiogram waveform)
Is generally known to be composed of a plurality of fundamental waves such as a P wave, a Q wave, an R wave, an S wave, and a T wave.
The interval between adjacent R waves is the heartbeat interval, the interval between adjacent Q waves and T
The mutual spacing of the waves is called the QT spacing. The heartbeat interval is a reference when measuring the heart rate (heart rate = the number of heartbeat intervals per minute), and the QT interval represents the duration of the action potential.

[0003] It is known that these heartbeat intervals and QT intervals constantly fluctuate, and indirectly reflect the mechanism of the biological regulation system of the human body. Specifically, the QT interval can be qualitatively determined as a function of the heartbeat interval (Baze
tt equation), and the heartbeat interval is controlled by the autonomic nerves of the human body. Therefore, by observing the heartbeat interval and the QT interval, it is possible to know one end of the biological function including the state of the autonomic nerve without putting a burden on the subject.

Here, in order to obtain the heartbeat interval and the QT interval, naturally, the R wave, the Q wave, and the T wave are used.
It is necessary to specify where each waveform of the wave is located in the electrocardiogram waveform. Further, since other P waves and S waves also serve as indices of biological functions, there is a demand to specify where these P waves and S waves are located in the electrocardiogram waveform (hereinafter, P waves, Q waves, The waveform of a fundamental wave such as an R wave, an S wave, and a T wave is referred to as a target waveform). Specifically, it is necessary to specify the position of the top of the target waveform (the generation position of the target waveform) and its start point and end point on the time axis of the electrocardiogram waveform (hereinafter, the time of the electrocardiogram waveform as described above). Specifying all or a part of the position, the start point, and the end point of the vertex of the target waveform on the axis is referred to as “extracting the target waveform”).

As a method of extracting a target waveform from an electrocardiogram waveform, various methods have been conventionally studied. For example, waveform data (hereinafter, referred to as an ECG waveform)
Attempts have been made to extract a target waveform based on the value or slope of each part of the electrocardiogram waveform data). In recent years, a method that applies a wavelet transform (Wavelet Transform) has been proposed as a method of extracting a target waveform from an electrocardiogram waveform. In this wavelet transform, first, a basic waveform (hereinafter, referred to as a wavelet function)
The waveform data (hereinafter referred to as “basic waveform data”) of the basic waveform is scale-converted to obtain waveform data (converted waveform data) of a plurality of similar waveforms (hereinafter referred to as “converted waveforms”). Then, pattern matching between the plurality of converted waveform data and the electrocardiogram waveform data is performed, and an index value indicating a characteristic of the electrocardiogram waveform data is extracted (hereinafter, an index value obtained by wavelet transform is referred to as a feature amount). ).

[0006]

However, in such a conventional ECG analysis system, it is difficult to extract a target waveform from an ECG waveform for various reasons. One of the reasons is that the electrocardiogram waveform itself changes under the influence of various factors. That is, the baseline of the electrocardiogram waveform fluctuates due to changes in the posture and respiration of the subject, or when the subject is moving, the myoelectricity resulting from the operation of the muscles of each part mixes into the electrocardiogram waveform and becomes noise. Or, a steep step-like signal intervenes in the ECG waveform, or
Since there are variations in electrocardiogram data due to individual differences, it has been difficult to uniformly identify a target waveform.

Therefore, as described above, it is actually extremely difficult to simply extract a target waveform based on the value or gradient of the electrocardiogram waveform data. In addition, in the method using the wavelet transform, although it has become possible to some extent to check the discontinuity point of the target waveform, there is still much room for debate in the handling of feature amounts obtained by pattern matching. It was still difficult to extract the target waveform. Under such circumstances, the actual situation is that a doctor or the like inspects and specifies the target waveform based on his or her own experience, and the analysis of the electrocardiogram requires a great deal of time, and the judgment criteria vary. As a result, there is a problem that the reliability of the analysis result is reduced. Therefore, there has been a demand for a system capable of automatically and reliably extracting a waveform.

SUMMARY OF THE INVENTION The present invention has been made in view of the above-described problems, and enables a target waveform to be automatically extracted from an electrocardiogram waveform, thereby reducing the labor of the electrocardiogram analysis and improving its reliability. An object of the present invention is to provide an electrocardiogram analysis system. Another object of the present invention is to provide a computer-readable recording medium on which a program for realizing such extraction of various target waveforms is recorded.

[0009]

In order to achieve this object, the present invention according to claim 1 relates to an electrocardiogram analyzing system, in which waveform data of an electrocardiogram waveform to be analyzed is subjected to wavelet transform. The target waveform included in the electrocardiogram waveform is obtained by performing a pattern analysis on the maximum value or the minimum value extracted by the conversion extraction unit that extracts the maximum value or the minimum value from the feature amount and the conversion extraction unit. And a pattern analysis means for extracting.

In this system, electrocardiogram waveform data is subjected to a wavelet transform to extract its characteristic amount, and a maximum value or a minimum value is extracted from this characteristic amount. The present inventor has clarified that a certain tendency exists between the maximum values or the minimum values, and the like.Therefore, by analyzing the pattern of the maximum value or the minimum value, each maximum value or the minimum value is analyzed. It is possible to clarify which of the target waveforms the minimum value corresponds to. Here, since the maximum value and the minimum value correspond to the vertices of each target waveform, the positions of the vertices and the like can be specified, and the target waveform can be extracted. According to this analysis system, the target waveform can be automatically extracted from the electrocardiogram waveform, thereby reducing the labor of the electrocardiogram analysis and improving the reliability by performing an objective analysis. it can.

The present invention described in claim 2 is the same as the claim 1.
In the present invention described above, the conversion and extraction unit converts the predetermined basic waveform data into a plurality of scale orders to obtain a plurality of conversion waveform data, and the conversion conversion unit obtains a plurality of conversion waveform data. Wavelet transform between the plurality of transformed waveform data and the waveform data of the electrocardiogram waveform, thereby obtaining a plurality of feature amounts, and a plurality of feature amounts acquired by the wavelet transform unit. Extreme value extracting means for extracting a local maximum value or a local minimum value of the target waveform, wherein the pattern analysis means determines a change tendency of a plurality of local maximum values or local minimum values extracted by the local value extracting means for each of the target waveforms. The above-mentioned target waveform is extracted by comparing with the change tendency of the above.

This shows the configuration of the conversion extracting means and the pattern analyzing means more specifically. According to this configuration, the basic waveform data is scale-converted by a plurality of scale orders to obtain a plurality of converted waveform data, and the plurality of converted waveform data is used to perform a wavelet transform, thereby obtaining a plurality of feature amounts. Is obtained. Then, a plurality of maximum values or minimum values are extracted from the plurality of feature amounts. By obtaining a plurality of maximum values or minimum values in this manner, a target waveform can be obtained based on a certain tendency existing in these maximum values or minimum values.

Further, the present invention described in claim 3 is based on claim 1.
Or in the present invention according to 2, the pattern analysis means comprises a range of scale orders at which the maximum value or the minimum value is obtained, a change pattern of the maximum value or the minimum value corresponding to the change of the scale order, a value of the maximum value or the minimum value. And extracting at least one of the target waveforms based on at least one of the deviations between the scale orders.

[0014] This clearly shows the standard of pattern analysis in the pattern analysis means. In other words, as described above, the present inventor has clarified that there is a certain tendency between the maximum value and the minimum value and the like, and this tendency is roughly classified into three criteria and applied to pattern analysis. By doing so, the target waveform can be extracted easily and reliably. All of these three criteria may be used, or only some arbitrary criteria may be used. The specific contents of each criterion will be described later.

The present invention according to claim 4 provides the present invention as claimed in claim 1.
In the present invention as set forth in any one of Items 1 to 3, the pattern analysis means may include:
It is characterized in that an R-wave of a target waveform is first extracted, and a time position relative to the R-wave is used as one of criteria for extracting waves other than the R-wave of the target waveform. I have.

This shows the extraction order of the target waveform more clearly. The first extraction of the R wave in this manner is relatively easy because the R wave has the largest peak value in the target waveform, and the R wave is located at the temporal center position in the target waveform. Therefore, by extracting this R wave first, it becomes easier to extract another target waveform. However, it is not always necessary to extract the R wave first, and other target waveforms may be extracted first.
Alternatively, it is also possible to extract in parallel.

According to a fifth aspect of the present invention, there is provided a computer-readable recording medium on which a program for analyzing an electrocardiogram is recorded, wherein a waveform data of an electrocardiogram waveform to be analyzed is subjected to a wavelet transform. The target waveform contained in the electrocardiogram waveform is obtained by performing a pattern analysis on the maximum value or the minimum value extracted in the conversion extraction procedure for extracting the maximum value or the minimum value from the feature amount, and extracting the maximum value or the minimum value from the feature amount. It is configured as a computer-readable recording medium that records a program for sequentially executing a pattern analysis procedure to be extracted.

By causing the computer to read the recording medium and executing the program, the ECG waveform data is subjected to a wavelet transform to extract a feature amount thereof, and a maximum value or a minimum value is extracted from the feature amount. Then, by performing a pattern analysis of these local maximum values or local minimum values, it is possible to specify the vertices of each target waveform and extract the waveform. Since such a procedure can be automatically executed by a computer, a target waveform can be automatically extracted from an electrocardiogram waveform, so that the labor of the electrocardiogram analysis can be reduced and objective analysis can be performed. By doing so, the reliability can be improved.

The present invention according to claim 6 provides the invention according to claim 5
In the present invention described above, the conversion and extraction procedure includes a scale conversion procedure of obtaining a plurality of converted waveform data by performing scale conversion on predetermined basic waveform data by a plurality of scale orders; A wavelet transform is performed between the plurality of converted waveform data obtained and the waveform data of the electrocardiogram waveform to obtain a plurality of feature amounts, and a maximum is obtained from the plurality of feature amounts obtained by the wavelet transform. And an extreme value extraction procedure for extracting a value or a minimum value.
The pattern analysis procedure is a procedure for extracting the target waveform by comparing a change tendency of a plurality of maximum values or minimum values extracted in the extreme value extraction procedure with a known change tendency for each of the target waveforms. It is configured as a feature.

This shows the contents of execution of the conversion extraction procedure and the pattern analysis procedure more specifically. According to this procedure, the basic waveform data is scale-converted by a plurality of scale orders to obtain a plurality of converted waveform data, and a plurality of feature amounts are obtained. Then, a plurality of maximum values or minimum values are extracted. By obtaining a plurality of maximum values or minimum values in this manner, a target waveform can be obtained based on a certain tendency existing in these maximum values or minimum values. Since such a procedure can be automatically executed by a computer, a target waveform can be easily extracted.

The present invention according to claim 7 provides the present invention as claimed in claim 5.
Or in the present invention according to 6, the pattern analysis procedure includes a range of a scale order at which a maximum value or a minimum value is obtained, a change pattern of a maximum value or a minimum value corresponding to a change of the scale order, and a deviation information between scale orders. It is characterized in that it is a procedure for extracting the target waveform on the basis of at least one.

This clearly shows the standard of pattern analysis in the pattern analysis procedure. As described above, since there is a certain tendency between the maximum value and the minimum value and the like, this tendency is roughly classified into three criteria and applied to the pattern analysis. Since such a procedure can be automatically executed by a computer, a target waveform can be easily and reliably extracted based on a clear standard.

The present invention according to claim 8 provides the present invention according to claim 5.
In the present invention as set forth in any one of claims 7 to 7, the pattern analysis procedure comprises:
A procedure for extracting an R wave of the target waveform first, and extracting a wave other than the R wave in the target waveform by using a temporal position with respect to the R wave as one of criteria. It is configured.

This clearly shows the extraction order of the target waveform. As described above, since the R wave has the largest peak value among the target waveforms, the extraction is easy, and since the R wave is at the temporal center position of the target waveform, by extracting this R wave first, Extraction of another target waveform can be performed more easily. Since such a procedure can be automatically executed by the computer, the target waveform can be performed more quickly and more reliably.

[0025]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of an electrocardiogram analysis system according to the present invention and a computer-readable recording medium on which a program for realizing the same is recorded will be described in detail with reference to the drawings. I do. The present invention is not limited by the embodiment.

(Outline of Analysis) The electrocardiogram analysis processing according to this embodiment is performed roughly as follows. First,
The feature amount is calculated by performing a wavelet transform on the electrocardiogram waveform data, and further, a maximum value or a minimum value of the feature amount is extracted. Here, the wavelet transform of the electrocardiogram waveform data is performed using a plurality of transformed waveform data obtained by performing a scale transform of the basic waveform data by a plurality of scale orders. Therefore, a plurality of maximum values or minimum values of the feature amount are extracted corresponding to a plurality of scale orders.
The inventors of the present application have clarified that there is a certain pattern different for each target waveform between a plurality of maximum values or minimum values extracted in this manner. Therefore, pattern matching using this pattern is performed. Thus, a target waveform can be extracted from the electrocardiogram waveform.

(System Configuration) The configuration of a system for extracting such a target waveform will be described. FIG.
FIG. 1 is a block diagram showing a configuration of an electrocardiogram analysis system according to the embodiment. As shown in FIG. 1, the system includes an arithmetic control unit 1 and a random access memory (RAM).
2, ROM (Read Only Memory) 3, HD (Hard Disk)
4, an input control interface (input control IF) 5, an output control interface (output control IF) 6, a keyboard 7, a mouse 8, and a monitor 9. These components perform data communication via a bus 10. Connected as possible. This system can be realized by, for example, a personal computer.

The arithmetic and control unit 1 controls each unit of the present system, and is conceptually provided with a conversion extracting unit 11 and a pattern analyzing unit 12. The conversion extraction unit 11 performs scale conversion of basic waveform data, wavelet conversion of electrocardiogram waveform data, and
The maximum value or the minimum value is extracted. These processes are respectively performed by a scale conversion unit 13, a wavelet conversion unit 14, and an extreme value extraction unit 1 provided in the conversion extraction unit 11.
5 is performed. Details of the processing by these units will be described later. Further, the pattern analysis unit 12 extracts a target waveform included in the electrocardiogram waveform by performing pattern analysis on the feature amount extracted by the conversion extraction unit 11.

All or any part of each unit of the arithmetic control unit 1 can be realized by a CPU and a program interpreted and executed by the CPU. That is, the OS (Operating System) is stored in the ROM 3 or the HD 4.
A computer program for giving instructions to the CPU in cooperation with em) and performing various processes is stored. This computer program is executed by being loaded into the RAM 2 as a main storage unit, and configures each processing unit in cooperation with the CPU. However, this computer program can be stored in an application program server or the like connected to the present system via an arbitrary network, and can be downloaded and executed in whole or in part as necessary. . Or,
All or any part of each processing unit can be realized as hardware such as wired logic. In addition, in addition to the above-mentioned program, electrocardiogram waveform data described later can be recorded on the HD 4.

In addition, the keyboard 7 and the mouse 8
This is an input unit for inputting arbitrary input contents such as an instruction to start an electrocardiogram analysis process. The keyboard 7 and the mouse 8 are connected to the bus 10 via the input control IF 5. In addition, electrocardiogram waveform data output from an arbitrarily configured electrocardiograph is directly or indirectly acquired via the input control IF 5. Also, the monitor 9
Output means for outputting an analysis result or the like. The monitor 9 is connected to the bus 10 via the output control IF 6. The monitor 9 realizes a pointing device function in cooperation with the mouse 8.

(Electrocardiogram analysis processing) The electrocardiogram analysis processing in the present system configured as described above will be described.
FIG. 2 is a main flowchart of the electrocardiogram analysis processing by the present system, FIG. 3 is a flowchart of the conversion extraction processing of FIG. 2, and FIGS. 4 and 5 are flowcharts of the pattern analysis processing of FIG. In the following processing, the purpose is to automatically specify the position of the top of the target waveform on the time axis of the electrocardiogram waveform (the generation position of the target waveform).

(Electrocardiogram analysis processing-capture of waveform data) First, as shown in FIG. 2, this processing is performed when an instruction to start the electrocardiogram analysis processing is issued via the keyboard 7 and the mouse 8 (step S2-1). The acquisition of the electrocardiogram waveform data is performed (step S2-2). The electrocardiogram waveform data is obtained by deriving and amplifying an action potential generated along with expansion and contraction of each part of the heart from the skin surface (see FIG. 16 described above) and A / D converting this. It is. The derivation and amplification of the action potential are performed by an electrocardiograph, and the A / D conversion is performed using an A / D converter such as an A / D conversion card. The electrocardiogram waveform data obtained in this way is taken into the system in real time via the input control IF 5 and recorded on the HD 4. Alternatively, the waveform data may be pre-fetched and stored in the HD 4 before the start of this processing.

Here, the acquisition of the electrocardiogram waveform data in step S2-2 and the later-described step S2-3,
The processing in S2-4 is performed for each data set composed of data corresponding to the sampling rate at the time of A / D conversion (for example, when the sampling rate = 1 kHz,
8 bits × 1000 ≒ 8192 = 2 ¹³ ) and the data sets overlap each other. FIG. 6 shows a timing chart for storing and processing such data. In FIG. 6, a data set Data1 that is first captured and processed is overlapped with a data set Data2 that is secondly captured and processed, and the subsequent data sets Data3 to Datan are similarly overlapped. It is done repeatedly. And among the results obtained in each process,
Only the portions Out1 to Outn indicated by oblique lines are extracted and connected to each other, and output as a final result. This is because, if the processing is simply performed sequentially for each data set of the number corresponding to the sampling rate (every 8192 data sets), the continuity of data between the data sets cannot be maintained, and so on as a whole. This is because correct results may not be obtained. However, when it is not necessary to consider such a problem, it is possible to simply take in and process sequentially from the first data.

(Electrocardiogram analysis processing-conversion extraction processing) After the electrocardiogram waveform data is fetched in this way, the conversion extraction section 11
Under this control, the conversion and extraction processing of the electrocardiogram waveform data is performed (step S2-3). In this conversion extraction process, as shown in FIG. 3, first, it is determined whether or not the potential (amplitude) of the electrocardiogram waveform data is an appropriate magnitude (step S3-1). Here, "appropriate size"
This means that even when the wavelet transform is performed using the transformed waveform data at the scale order j = 3 to 4 in the wavelet transform process described later, only the feature amount of the R wave remains without disappearing or converging. That's it. For example, by obtaining in advance a potential corresponding to such a magnitude from an experimental value or the like and setting it as a reference value, and comparing the potential of the electrocardiogram waveform data with this reference value,
Appropriateness of the size can be determined. This step S
If it is determined in 3-1 that the size is not appropriate, the level is adjusted by amplification or the like so as to obtain an appropriate size (step S3-2). By performing such processing, it is possible to reduce adverse effects due to variations between electrocardiogram waveform data. Since the inner product value of the electrocardiogram waveform data and the basic waveform data is calculated in step S3-4 described later, step S3-
In 2, the same result can be obtained by adjusting the level of the basic waveform data instead of the electrocardiogram data. In particular, since the number of basic waveform data is smaller than that of electrocardiogram data, it is preferable to adjust the level of the basic waveform data in step S3-2 in that the amount of calculation can be reduced.

(Electrocardiogram analysis processing-conversion extraction processing-wavelet conversion processing) Next, scale conversion processing (step S3-3) and wavelet conversion processing (step S3-
4), extreme value extraction processing (step S3-5) is sequentially performed. These are performed by the scale conversion unit 13, the wavelet conversion unit 14, and the extreme value extraction processing unit 15, respectively. First, before describing the scale transformation, the wavelet transformation processing will be described. As a premise theory, a pattern matching theory (matched filter theory) for detecting a generation time of a target signal from a signal containing noise is known (“Prospects of time-frequency analysis [II to IV]). ”, IEICE Journal Vol.79 No.6 pp.
597-602 June 1996, No.7 pp.746-751 July 1996, No.8
pp.820-830 August 1996). Here, a signal including noise = x (t), a template signal to be compared = g
(T) When the position of the template signal on the time axis = τ, a convolution operation (calculation of a cross-correlation function, pattern matching) represented by the following equation (1) is performed to generate a target signal. It is shown that the time can be detected.

(Equation 1)

However, the noise-containing signal x (t)
If the waveform of each component constituting is unknown, a method of collecting appropriate waveforms (generally called a wavelet function system) and comparing this waveform with data to be analyzed is adopted.
This method is the wavelet transform in step S3-4. This wavelet transform W (a, τ) is expressed by equation (2),
The wavelet function system (a, τ) is shown in Expression (3) (where a = scale conversion coefficient).

(Equation 2)

(Equation 3)

(Electrocardiogram analysis processing-conversion extraction processing-scale conversion processing) The wavelet function system used in this wavelet conversion is constructed by performing scale conversion of predetermined basic waveform data. For this reason, in step S3-3 in FIG. 3, a scale conversion process is performed in advance to obtain a wavelet function system.
In the scale conversion in step S3-3, a Mexican hat well-known as a wavelet function is used as a basic waveform. This Mexican hat is used because it is known to be excellent in peak detection of a waveform (“Configuration of an appropriate analyzing wavelet in single-shot waveform processing”, Medical Electronics and Biotechnology, Vol. 37, Special Issue) (1999) p
p.227). This Mexican hat is a function represented by the second derivative of the Gaussian function as shown in the following equation (4).

(Equation 4) This Mexican hat is shown in FIG. However, any waveform other than the Mexican hat can be used as the basic waveform.

There are various methods for performing the scale conversion on the basic waveform. In this embodiment, the scale conversion is performed by a power of 2 (2- ^j ) by the scale order j (discrete binary wavelet conversion). The discrete conversion is performed in order to reduce the number of data to an appropriate number. If it is not necessary to consider this point, another continuous conversion can be performed. In this scale conversion, a plurality of converted waveform data is obtained by changing the scale order j. FIG. 8 shows the ECG waveform H
W, each converted waveform TW _{0 at} scale order j = 0 to 3
To TW ₃ and feature amounts WTD _{0 to} WTD ₃ extracted by the wavelet transform process (the feature amounts will be described later). As shown in FIG. 8, each conversion waveform TW ₀ ~TW ₃ in scale orders j = 0 to 3 is converted gradually steep waveform. Incidentally, in FIG. 8 shows only the conversion waveform TW ₀ ~TW ₃ in scale orders j = 0 to 3, in this process, the scale degree j =
The scale conversion is performed using −2 to 4, and the converted waveform TW
_{−2 to} TW ₄ are obtained (the reason why such a scale order is selected will be described later).

(Results of Electrocardiogram Analysis Processing-Conversion Extraction Processing-Wavelet Transformation Processing) Wavelet conversion processing based on the conversion waveform data of the conversion waveforms TW _{-2 to} TW ₄ thus obtained and the waveform data of the electrocardiogram waveform HW ( As a result of the convolution operation shown by the above equation (2), the feature amounts WTD _{-2 to} WTD ₄ are extracted (as described above, the scale order j =
Since each converted waveform data in −2 to 4 is obtained,
In this wavelet transform process, actually, FIG.
Unlike the contents shown in (1), the feature amounts WTD _{-2 to} WTD4 corresponding to the scale orders j = _{−2 to 4} are extracted.)
This calculation can be performed simply as the calculation of the inner product value when the time axis shift conversion is not considered.

(Electrocardiogram analysis processing-conversion extraction processing-extreme value extraction processing) Next, extreme value extraction processing is performed (step S3).
-5). Here, the maximum value and the minimum value of each feature value are extracted by sampling the feature values WTD _{-2 to} WTD ₄ obtained by the wavelet transform with the maximum value and the minimum value (hereinafter, the maximum value and the minimum value). The minimum value is collectively referred to as an extreme value as needed). Extracting both the maximum value and the minimum value in this manner is, as is clear from FIG. 16, the R wave, the T wave,
Also, since the P wave has a positive potential with respect to the base line, the feature value has a positive value, and the Q wave and the S wave have a negative potential, and the feature value has a negative value. Therefore, for example, when only the R wave needs to be extracted, only the local maximum value of each feature amount may be extracted. FIG. 9 shows an example of the maximum value among the extreme values extracted in this way.

(Electrocardiogram Analysis Process-Pattern Analysis Process-Outline) As shown in FIG. 2, after the conversion extraction process is completed, a pattern analysis process is performed (step S2-4). The reason for performing this pattern analysis processing is as follows. That is, since the actually recorded electrocardiogram waveform has individual differences, the individual extreme values themselves extracted in the conversion extraction processing are not constant. Therefore, it is difficult to extract a target waveform based on the magnitude of each extreme value. However, it has been confirmed by the inventor of the present application that the extreme value change pattern is consistent for many electrocardiogram waveforms. Therefore, a target waveform can be extracted by performing a pattern analysis using the change tendency of the extreme values.

The change tendency will be described. FIG.
Is a graph showing the tendency of the change of the extreme value for each target waveform, and is a graph showing the relationship between each scale order (horizontal axis) and the extreme value (vertical axis) extracted from the feature amount. FIG. 10 shows the maximum values R _{-2 to} R _{4 of} the R wave, the maximum values T _{-2 to} T _{3 of} the T wave,
The maximum values P _{-2 to} P ₁ of the P wave are shown (note that the value of this extreme value changes according to the result of the level adjustment in step S3-2). In addition, the minimum values of the Q wave and the S wave are obtained in a similar manner, but are not shown.

The tendency of the change of the extremum for each target waveform as shown in FIG. 10 corresponds to the tendency (first tendency) relating to the range of the scale order j at which the maximum value or the minimum value is obtained, and the change of the scale order j. It is roughly classified into a tendency relating to a change pattern of the maximum value or the minimum value (second tendency), a tendency relating to the numerical value of the maximum value or the minimum value, and a tendency relating to the deviation between the scale orders j (third tendency). As a first tendency, for example, as shown in FIG. 10, only R ₄ exists as a maximum value at the scale order j = 4, and T ₄ and P ₄
Does not exist or has converged.
Also, as a second tendency, for example, the maximum value R
_-1 > the maximum value R _-2 tends to be strong, while the T wave has a strong tendency of the maximum value T _-1 <the maximum value T _-2 . The third tendency is, for example, that R ₁ > 30 or R ₁ > T ₁ > P ₁ .

These first to third tendencies are consistent with respect to many ECG waveforms. Therefore, step S3
By performing pattern analysis on all or any part of these trends with respect to the extremum obtained in -5, it is specified which extremum of the target waveform each extremum is. You. Since the position of the extremum is the position of the vertex of each target waveform, the position of the vertex of each target waveform can be specified.

Before specifically describing the contents of the pattern analysis, the basis for setting the range of the scale order j = −2 to 4 used in the above-described scale conversion processing will be described. Further, in this pattern analysis processing, various problems that have made the electrocardiogram analysis difficult so far have been solved, and this point will be described together. First, as a premise theory, a method of reconstructing the original waveform from the wavelet transform maximum value (inverse wavelet transform) is known (Stephane Mallat and Wen Liang Hwang: Singularity)
Detection and Processing with Wavelets IEEE Tran
s. on info. theory, Vol. 38, No2, March 1992 (617-
643)). Here, it is shown that the original waveform can be reconstructed based on the local maximum value extracted from the waveform.
FIG. 11 shows an example of this reconfiguration. In FIG. 11,
(A) is an original waveform, (b) is a converted waveform (W (2 ¹ , τ) to W (2 ⁶ , τ)) at a scale order j = 1 to ⁶ , and
The low-frequency component S (2 ⁶ , τ) left in the converted waveform at the scale order j larger than the scale order j = 6 is shown. (C) shows the position of the local maximum value extracted from each converted waveform, and (d) shows the signal reconstructed from the local maximum value of (c) by adding a low frequency component S (2 ⁶ , τ) to the signal. FIG.

Here, as a general problem in the electrocardiogram analysis, it is pointed out that the electrocardiogram waveform includes "swaying of the base line", "mixing of myoelectricity", or "intervention of a steep step signal". . Therefore, it is considered to appropriately set the scale order j in the scale conversion process and to remove these problem elements by performing an appropriate process in the pattern analysis. Of these, "baseline fluctuation" means that the baseline of the electrocardiogram waveform fluctuates due to the influence of a change in the blood pressure or the like on the change in the posture of the subject, and appears as low-frequency swell in the electrocardiogram waveform data. Since this low-frequency undulation corresponds to the low-frequency component S (2 ⁶ , τ) in FIG. 11 described above, the high-frequency scale order j that generates such a low-frequency component S (2 ⁶ , τ) By eliminating local maxima or minima,
It can be excluded from ECG waveform data.

Thus, by setting the range of the higher-order scale order j to, for example, 6 or less, the fluctuation of the baseline can be eliminated. However, as shown in FIG.
At the scale order j = 4, only the maximum value R ₄ of the R wave exists, and at the scale order j higher than that,
In this embodiment, the range of the higher-order scale order j is finally set to 4 because there is a high possibility that the maximum value of all waves disappears or converges. Next, the lower-order scale order j
Think of the range. As shown in FIG. 10, as the scale order j = −2 to the scale order j = −1,
While the maximum values R _-2 and R _-1 of the R wave tend to increase,
The maximum values T _-2 and T _-1 of the T wave tend to decrease. Such a noticeable difference is important for distinguishing both waves. However, in the lower scale order j higher than that, it is not appropriate because the local maximum value and the local minimum value may not be present. Therefore, in this embodiment, the range of the lower order scale order j is finally changed. -2. That is, in conclusion
In the present embodiment, converted waveform data is obtained by performing skewing conversion in the range of scale order j = −2 to 4, and wavelet conversion is performed using the converted waveform data and electrocardiogram waveform data.

Further, "mixing of electromyogram" means that the electrocardiogram waveform slightly fluctuates due to the influence of muscle activity or the like, and appears as fine high-frequency noise in electrocardiogram waveform data. This high-frequency noise can be excluded from the electrocardiogram waveform data by excluding the one having a small value among the maximal values obtained using the low-order scale order j corresponding to the high frequency. Therefore, in the present embodiment, this processing is performed in step S4-1 described below.

Further, "the presence of a steep step signal" appears as a remarkably steep peak value in the electrocardiogram waveform data. This step-like signal has a scale order j
Since they appear as the same extreme value between the two, or as an extreme value having almost no deviation between the scale orders, such values can be excluded from the electrocardiogram waveform data. Therefore, in the present embodiment, this processing is performed in step S4-5 described later.

(Electrocardiogram analysis processing-Pattern analysis processing-Removal of small extremum) Each step of the pattern analysis processing performed based on such grounds will be described.
First, among the extreme values obtained in step S3-5,
The maximum value or the minimum value smaller than the predetermined value is removed (step S4-1). This is to remove myoelectric contamination as described above. Here, the predetermined value can be set to, for example, a value of about 5 in absolute value.

(Electrocardiogram analysis processing-Pattern analysis processing-R
Thereafter, an R-wave extraction process is performed. As shown in FIG. 16, the first extraction of the R wave is easy because the R wave has the largest peak value among the target waveforms, and the R wave is easily extracted from the target waveform. This is because extracting the R wave first makes it easier to extract another target waveform. However, it is not always necessary to extract the R wave first, and other target waveforms may be extracted first, or it is also possible to extract them in parallel.

In the R-wave extraction processing, first, among the maximum values at the scale order j = 1 previously extracted, the maximum value that is equal to or greater than a predetermined value and is temporally the next position (first position) In the processing, the local maximum at the earliest position) is R
It is extracted as a candidate for a wave vertex (R ₁ ) (step S4)
-2, see FIG. 9). As described above, this is a reference corresponding to the tendency that the R wave has the largest peak value (third tendency) among the target waveforms at the scale order j = 1. Here, the predetermined value is, for example, a maximum value = 30
Is set. For example, when the maximum value of the scale order j = 1 are extracted in the order of 18,70,5,17,68,6 is maximum value 70 is extracted as a candidate for R _1. The reason why the maximum value of the scale order j = 1 is selected is that there is no possibility that the maximum value of the R wave disappears and converges with this scale order. It is also possible to select from local maxima.

Next, among the maximum values at the scale order j other than 1 (here, J = −2 to 0, 2 to 4), the maximum value R ₁ extracted at step S4-2 is A maximum value or a minimum value within a predetermined range (for example, within ± 20 ms) is extracted as a candidate of an R-wave vertex in each scale order (step S4-3, see FIG. 9). That is, the maximum values R _-2 , R _-1 , R ₀ , R ₂ , R ₃ , and R ₄ are further extracted. These extracted maximum values are stored in a one-dimensional array. Incidentally, setting such a predetermined range, would otherwise among other scales orders of local maxima, but we want to extract the maximum value of the closest position relative to the maximum value R _1, the scale degree Depending on the maximum value, there is a possibility that the maximum value disappears or converges. Therefore, the extraction is limited to the maximum value within a predetermined range.

Then, the maximum value R _-2 extracted so far
All respective to R ₄ is greater than or not is judged 0 (Step S4-4). This is a criterion corresponding to the fact that the local maximums R _{-2 to} R ₄ all tend to be larger than 0 (third trend) as shown in FIG. In particular, P
Wave to tend not substantially exist in the scale of the above R _1, it is possible to exclude P waves by determining the reference.

If all of the maximum values R _{-2 to} R ₄ are larger than 0, it is further determined whether or not the relationship between the maximum values R _{-2 to} R ₄ matches a predetermined pattern (step S4). −
5). Specifically, the maximum value R _-1 > the maximum value R _-2 , the maximum value R ₀
> Maximum value R ₂ , Maximum value R ₂ > Maximum value R ₃ , and Maximum value R ₁
> Whether they meet all the conditions of the maximum value R ₃ is judged. These are based on the second tendency, which is particularly clear.

[0056] For example, since it can be said that the maximum value R _-1 is the maximum value R _-2 greater tendency, and a maximum value R _-1> maximum value R _-2 and one of the conditions. In particular, it is difficult to separate the R wave and the T wave because the patterns are similar as a whole. However, as described above, the R wave has a strong tendency of the maximum value R ₋₁ > the maximum value R ₋₂ , Since the T wave has a strong tendency of the maximum value T _-1 <the maximum value T _-2 , the T wave can be excluded by setting the maximum value R _-1 > the maximum value R _-2 as one of the conditions. Also, by judging such a condition, the case where the local maximum value is the same between the scale orders is excluded, so that the problem of “the presence of a steep step-like signal” can be excluded as described above. . Here, other relations between local maxima may be used, but those having a relatively small difference are not considered as conditions. For example, comparing the local maximum value R ₀ and the local maximum value R ₁ , the local maximum value R ₁ tends to be slightly large, but this difference is small and is excluded from the conditions.

Next, whether or not the maximum value R ₄ is greater than a predetermined value is determined (step S4-6). As this predetermined value, for example, about 20 can be used. this is,
At the scale order j = 4, the P wave disappears or converges, and the T wave also tends to converge to less than 10 (first tendency). Because it can be excluded.

Thereafter, it is determined whether or not there is another R wave within a predetermined range before and after the temporal position of the maximum value R _{-2 to} R ₄ extracted as a candidate of the R wave apex. (Step S4-7). This is performed in order to exclude the R wave, which may rarely occur in the near range. As the predetermined range, for example, about ± 300 ms is set. Steps S4-4 to S4
In -7, if any of the judgment conditions is not met, the maximum values R _{-2 to} R ₄ extracted so far are discarded, and the process returns to step S4-2 again. Then, R ₁ candidates are extracted again from the local maximum value after R ₁ extracted earlier.

Meanwhile, if it is determined that meets the conditions in step S4-7, of the electrocardiogram wave data, before and after the basis of the temporal position of the maximum value R _-2 to R _4, which are extracted so far The highest potential E1 within the predetermined range is extracted, and this potential E1 is specified as the peak of the R wave (step S4-8, see FIG. 9). This is performed in order to eliminate the deviation since the position of each local maximum value tends to slightly deviate from the position of the original waveform due to the scale conversion processing. As this predetermined range, for example, about ± 10 ms is set. Step S4-2 so far
Although the positions of the vertices of the R wave can be specified by the processes from S4 to S4-8, some of these steps may be omitted, or further processes may be added. For example, even when only the two processes of steps S4-5 and S4-6 are performed, it is known that the accuracy of extracting the R wave is high to some extent, so that only these two processes are performed to speed up the process. Can be achieved.

Thereafter, with reference to the position of the vertex of the R wave specified in steps S4-2 to S4-7, the P wave,
The positions of the vertices of the T wave, the Q wave, and the S wave can be specified. First, at the scale order j = 0, the maximum value P ₀ immediately before the vertex of the R wave is specified as the position of the vertex of the P wave, and the potential E2 corresponding to this vertex is determined.
Is identified as the peak of the P-wave. (Step S5-
1, see FIG. 9). The scale order need not be 0, and a local maximum value in an arbitrary scale order at least where the P wave does not disappear or converge can be used. Also in this case, the difference between the local maximum value and the electrocardiogram waveform data may be eliminated by performing the same processing as in step S4-8. Or you may extract further by applying the same reference | standard as step S4-2 to S4-8.

Thereafter, similarly, the T wave, the Q wave, and
The position of the top of the S wave can be specified (step S
5-2 to S5-4). Note that in specifying the vertices of the Q wave and the S wave, a minimum value is used instead of a maximum value for the above-described reason. Further, as the reference scale order, a scale order corresponding to the characteristics of each wave can be used as shown in each step of FIG. Thus, the first pattern analysis processing ends. Thereafter, as shown in FIG. 2, steps S2-2 to S2-4 are repeated until the processing of all data sets is completed (step S2-5), and this processing ends.

(Extraction of T Wave End Point) In the above-described analysis processing, the position of the vertex of the target waveform can be automatically obtained. This shows that the heartbeat interval (interval between the vertices of adjacent R waves), which is one of the important indices of the electrocardiogram analysis, can be automatically obtained. However, since the QT interval, which is another important index, is defined as the mutual interval between the adjacent Q wave (start point) and T wave (end point), the start point of the Q wave and the end point of the T wave are defined. If automatic extraction is not possible, the QT interval cannot be determined automatically. Of these, the Q wave is smaller than the other target waveforms, so its starting point does not matter much, but the T wave is a relatively large wave and its ending point becomes a problem.

Therefore, the inventor of the present application has also studied a process for extracting the end point of the T wave. FIG. 12 shows an outline of this processing. In FIG. 12, (a),
(B) and (c) respectively show (a) and (b) of FIG.
This is almost the same as (b) and (c) (however, the low-frequency component S (2 ⁶ , τ) is omitted). In FIG. 12C, the converted waveform W at the scale order j = 1 to 6 is shown.
Looking at the feature quantity extracted from (2 ¹ , τ) to W (2 ⁶ , τ), this feature quantity has disappeared at the scale order 6, and this disappearance position almost coincides with the end point of the waveform. You can see that it is. Therefore, the end position of the T wave can be specified based on such a disappearance position of the feature amount (or a disappearance position of the local maximum value or the local minimum value). Not the result
This is the result of the wavelet analysis included in the analysis method of the present application). The specific processing for specifying the end position of the T-wave can be performed in series with the above-described processing for specifying the apex of the target waveform under the processing of the conversion extraction unit 11 and the pattern analysis unit 12, or It can be performed as a typical process. In addition, the end point of the target waveform other than the T wave can be similarly extracted. Also, for example,
The temporal position of the local minimum for the scale order j = 0-2,
It is also possible to define the end point of the T wave or the like based on the extreme value. For example, an average value of the extremum T ₁ and extreme T _2, the temporal position of the average value may be used as the end point of the T-wave.

(Electrocardiogram Analysis Result) An example of an electrocardiogram analysis result by the analysis system described above will be described. The present inventor has 11 syncope patients (60 ± 16 years old)
On the other hand, an ECG was sampled at 90 at a sampling rate of 1 [kHz].
(S) Recorded. Then, by analyzing the electrocardiogram waveform data of the electrocardiogram thus obtained by the present system, the P wave, the Q wave, the R wave, the S wave, and the T wave are automatically recognized, and the recognition results are monitored. Is displayed. This automatic recognition result is shown in FIG. In this FIG.
(A) is an analysis result of an electrocardiogram in which the baseline is violent, (b)
(C) is an analysis result of an electrocardiogram including a steep step-like waveform, and (c) is an analysis result of an electrocardiogram including a steep step-like waveform. In each of these figures, the position indicated by the arrow is the position of the automatically recognized target waveform (the positions of the vertices of the P, Q, R, S, and T waves are denoted by Pp, Qp, Rp, Sp, Tp). indicates the end point of the T-wave as T _E).

Further, the electrocardiogram is divided into a waveform in which noise or fluctuation of the baseline is noticeable (a poor recording example) and a waveform in which the noise or baseline fluctuation is not good (a good recording example). Method). The result is shown in FIG. As can be seen from FIG. 14, an extremely high recognition rate of 99% or more was obtained for all the waveforms in the good recording example and 94% or more except for the end point of the T wave in the poor recording example. The end point of the T wave (T-en
Regarding d), a high recognition rate of 100% was obtained in the good recording example and 80.1% in the poor recording example.

(Application to Electrocardiogram Display, etc.) Since the peaks and end points of the target waveform can be recognized at a high recognition rate, the heartbeat interval and the QT interval can be automatically recognized. It is now possible to display a new form of electrocardiogram that could not be performed. Hereinafter, an example of the electrocardiogram display will be described. In order to examine the usefulness of this system, the present inventor set up a standing test (Head up
tilt: HUT) test was performed. This standing test is a test for examining the response of the living body, including the autonomic nervous system, to the stress of a change in posture, and is known as a typical test for identifying syncope patients. Here, a syncope patient refers to a person who, when standing for a long time, loses the sympathetic / parasympathetic nervous system (autonomic nervous system) balance and falls into an anemia (low blood pressure) state.

In this standing test, 10 healthy males in their early twenties (mean 22 years ± 0.70) and 2 syncope patients
Examples were targeted. The state of sleeping at first (rest) was maintained for 5 minutes, the state of standing (80 °) was maintained for 30 minutes, and the state of sleeping again was maintained for 5 minutes, during which the electrocardiogram was continuously recorded. A target waveform is automatically extracted by the present system from the electrocardiogram waveform data of the electrocardiogram, and a heartbeat interval and a QT interval are automatically calculated.
These changes are displayed on the monitor 9. FIG. 15 shows a typical example of the relationship between the heartbeat interval (horizontal axis) and the QT interval (vertical axis) between healthy subjects and syncope patients. In FIG. 15, the bandwidths of the heartbeat interval and the QT interval are clearly different from each other, but this is due to the individual difference and has no relationship with the medical condition.

As can be seen from FIG. 15, first, the following facts were confirmed for healthy subjects. First, in the sleeping (resting) state, only a change in the heartbeat interval is remarkable (A in FIG. 15).
In the standing state, both the heartbeat interval and the QT interval are shortened (B in FIG. 15). After that, when sleeping again, it does not return to the original state (even after 20 minutes) (see FIG. 15).
C, however, it returns to its original state when it is set at 5 °). Further, the change in the QT interval accompanying the heartbeat interval until each state was stabilized drawn a counterclockwise trajectory. On the other hand, the following facts were confirmed for syncope patients. First, the behavior of the change of the QT interval accompanying the heartbeat interval was the same as that of a healthy person up to 14 minutes after standing (D → E in FIG. 15). afterwards,
The heartbeat interval and the QT interval gradually began to elongate (dashed line in FIG. 15), and the heartbeat interval suddenly exceeded 3000 ms (heart rate 20), resulting in a syncope state (F in FIG. 15).

As described above, according to the present system, the target waveform can be reliably extracted, so that an easy and objective electrocardiogram analysis can be performed. Furthermore, since the heartbeat interval and the QT interval can also be automatically recognized, the relationship between the heartbeat interval and the QT interval can be displayed as described above. Can do it. In particular, since these analyzes can be performed in real time, for example, it is possible to predict syncope by monitoring the heartbeat interval, and to perform a diagnosis or the like that has been impossible so far. In addition, the automatically recognized target waveform can be output / analyzed by any method and used for diagnosis from various angles.

[0070]

As described above, according to the electrocardiogram analysis system according to the first aspect of the present invention, the feature value is obtained by wavelet transforming the electrocardiogram waveform data, and the maximum value or the minimum value is obtained from the feature value. And a pattern analysis means for extracting a target waveform included in the electrocardiogram waveform by pattern-analyzing the maximum value or the minimum value extracted by the conversion extraction means. By analyzing the pattern of the maximum value or the minimum value, it is possible to clarify whether each maximum value or the minimum value corresponds to one of the target waveforms, and thus to extract the target waveform. Can be. According to this analysis system, the target waveform can be automatically extracted from the electrocardiogram waveform, thereby reducing the labor of the electrocardiogram analysis and improving the reliability by performing an objective analysis. it can.

According to the electrocardiogram analyzing system of the second aspect, the conversion extracting means includes a scale converting means for obtaining a plurality of converted waveform data, a wavelet converting means for obtaining a plurality of feature amounts, An extreme value extracting means for extracting a local minimum value, wherein the pattern analyzing means extracts the target waveform by comparing a change tendency of the plurality of local maximum values or the local minimum value with a known change tendency. By obtaining the minimum value, a target waveform can be obtained based on a certain tendency existing in the maximum value or the minimum value.

According to the electrocardiogram analyzing system of the third aspect, the pattern analyzing means includes a range of the scale order at which the maximum value or the minimum value is obtained, and a change pattern of the maximum value or the minimum value corresponding to the change of the scale order. Since the target waveform is extracted based on at least one of the maximum value or the minimum value and the deviation between the scale orders, the target waveform can be easily and reliably extracted based on a clear reference.

According to the electrocardiogram analysis system of the fourth aspect, the pattern analysis means first extracts the R wave out of the target waveform and uses the time position with respect to this R wave as one of the criteria. Then, since a wave other than the R wave in the target waveform is extracted, by extracting the R wave first, it is possible to more easily extract another target waveform.

According to a computer-readable recording medium having recorded thereon the program according to the fifth aspect, the ECG waveform data is subjected to wavelet transform to extract a local maximum value or a local minimum value, and the conversion extraction procedure is performed in the conversion extraction procedure. By performing a pattern analysis on the maximum value or the minimum value obtained, a program for sequentially executing a pattern analysis procedure for extracting a target waveform is recorded, so that the recording medium is read by a computer to execute the program. It is possible to clarify which maximum value or minimum value corresponds to which of the target waveforms, and to extract the target waveform. According to this analysis system, the target waveform can be automatically extracted from the electrocardiogram waveform, thereby reducing the labor of the electrocardiogram analysis and improving the reliability by performing an objective analysis. it can.

According to a computer-readable recording medium having recorded thereon the program according to claim 6, the conversion and extraction procedure includes a scale conversion procedure for obtaining a plurality of converted waveform data and a wavelet for obtaining a plurality of feature values. A conversion procedure and an extreme value extraction procedure for extracting a local maximum value or a local minimum value are sequentially performed.The pattern analysis procedure compares a change trend of a plurality of local maximum values or local minimum values with a known change tendency, and compares the target waveform with a target waveform. , The target waveform can be obtained based on a certain tendency existing in these local maximum values or local minimum values. Since such a procedure can be automatically executed by a computer, a target waveform can be easily extracted.

According to a computer-readable recording medium on which the program according to claim 7 is recorded, the pattern analysis procedure includes a range of a scale order at which a maximum value or a minimum value is obtained, and a maximum value corresponding to a change in the scale order. Since the procedure is for extracting the target waveform based on at least one of the change pattern of the value or the minimum value and the deviation information between the scale orders, such a procedure can be automatically executed by a computer. Therefore, the target waveform can be easily extracted based on a clear criterion.

According to a computer-readable recording medium having recorded thereon the program according to the eighth aspect, the pattern analysis procedure first extracts an R wave from a target waveform, and determines a time position with respect to the R wave. Since it is a procedure for extracting a wave other than the R wave in the target waveform by using one of the criteria, it is easy to extract the other target waveforms by extracting the R wave first. it can. Since such a procedure can be automatically executed by the computer, the target waveform can be performed more quickly and more reliably.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an electrocardiogram analysis system according to an embodiment of the present invention.

FIG. 2 is a main flowchart of an electrocardiogram analysis process by the present system.

FIG. 3 is a flowchart of a conversion extraction process of FIG. 2;

FIG. 4 is a flowchart of a pattern analysis process of FIG. 2;

FIG. 5 is a flowchart of a pattern analysis process of FIG. 2;

FIG. 6 is a timing chart of data storage and processing.

FIG. 7 is a diagram showing a Mexican hat used as a basic waveform.

FIG. 8 is a diagram showing an electrocardiogram waveform, converted waveform data at scale orders j = 0 to 3, and feature amounts extracted by wavelet transform processing.

FIG. 9 shows an example of an extreme value extracted in the extreme value extracting process.

FIG. 10 is a graph showing a tendency of change of an extreme value for each target waveform, and showing a relationship between each scale order and an extreme value extracted from a feature amount.

FIG. 11 is a diagram illustrating an example of waveform reconstruction by inverse wavelet transform.

FIG. 12 is a diagram illustrating an outline of a process of extracting an end point of a T wave.

FIG. 13 is a diagram showing automatic recognition results of a P wave, a Q wave, an R wave, an S wave, and a T wave.

FIG. 14 is a diagram showing the result of comparison of the coincidence rate between the result of automatic recognition by the present system and the judgment by inspection by an electrocardiologist.

FIG. 15 is a diagram showing a typical example of a relationship between a heartbeat interval between a healthy person and a syncope patient and a QT interval.

FIG. 16 is a diagram showing an example of an electrocardiogram.

[Explanation of symbols]

 Reference Signs List 1 arithmetic control unit 2 RAM 3 ROM 4 HD 5 input control interface 6 output control interface 7 keyboard 8 mouse 9 monitor 10 bus 11 conversion extraction unit 12 pattern analysis unit 13 scale conversion unit 14 wavelet conversion unit 15 extreme value extraction unit

Claims (8)

[Claims] 1. A conversion extracting means for obtaining a characteristic amount by performing a wavelet transform on waveform data of an electrocardiogram waveform to be analyzed, and extracting a maximum value or a minimum value from the characteristic amount, and a maximum value extracted by the conversion extracting means. Pattern analysis means for extracting a target waveform included in the electrocardiogram waveform by pattern-analyzing the value or the minimum value. 2. The scale conversion means for obtaining a plurality of converted waveform data by performing scale conversion on predetermined basic waveform data by a plurality of scale orders; By performing a wavelet transform between the plurality of converted waveform data and the waveform data of the electrocardiogram waveform, a wavelet transform unit for acquiring a plurality of feature amounts, and a plurality of feature amounts acquired by the wavelet transform,
An extreme value extracting means for extracting a plurality of local maximum values or local minimum values, wherein the pattern analysis means determines a change tendency of the local maximum values or local minimum values extracted by the extreme value extracting means for each of the target waveforms. The electrocardiogram analysis system according to claim 1, wherein the target waveform is extracted by comparing the target waveform with a known change tendency. 3. The pattern analysis means according to claim 1, wherein a range of the scale order at which the maximum value or the minimum value is obtained, a change pattern of the maximum value or the minimum value corresponding to the change of the scale order, a value of the maximum value or the minimum value, and a scale order. 3. The electrocardiogram analysis system according to claim 1, wherein the target waveform is extracted based on at least one of the deviations between the two. 4. The pattern analysis means first extracts an R wave of a target waveform, and uses a time position with respect to the R wave as one of criteria for determining an R wave of the target waveform other than the R wave. The electrocardiogram analysis system according to claim 1, wherein a wave is extracted. 5. A conversion extraction procedure for obtaining a feature amount by performing a wavelet transform on waveform data of an electrocardiogram waveform to be analyzed and extracting a maximum value or a minimum value from the feature amount, and a maximum value extracted in the conversion extraction procedure. Alternatively, a computer-readable recording medium recording a program for sequentially executing a pattern analysis procedure of extracting a target waveform included in the electrocardiogram waveform by performing a pattern analysis of the minimum value. 6. The conversion and extraction procedure includes the steps of: converting a predetermined basic waveform data by a plurality of scale orders to obtain a plurality of converted waveform data; By performing a wavelet transform between a plurality of converted waveform data and the waveform data of the electrocardiogram waveform, a wavelet transform procedure for acquiring a plurality of feature amounts, and a plurality of feature amounts acquired by the wavelet transform,
And an extremum extraction procedure for extracting a plurality of maxima or minima is sequentially performed. The pattern analysis procedure comprises: determining a change tendency of the plurality of maxima or minima extracted in the extremum extraction procedure; 6. A computer-readable recording medium storing a program according to claim 5, wherein said procedure is a procedure for extracting said target waveform by comparing with a known change tendency for each waveform. 7. The pattern analysis procedure according to claim 1, wherein at least one of a range of the scale order at which the maximum value or the minimum value is obtained, a change pattern of the maximum value or the minimum value corresponding to the change of the scale order, and deviation information between the scale orders. 7. A computer-readable recording medium on which the program according to claim 5 or 6 is a procedure for extracting the target waveform with reference to the following. 8. The pattern analysis procedure according to claim 1, wherein an R wave of the target waveform is first extracted, and a temporal position with respect to the R wave is used as one of criteria for determining an R wave of the target waveform other than the R wave. 6. A procedure for extracting waves.
A computer-readable recording medium recording the program according to claim 7.