JP3224569B2 - Cardiac arrhythmia detection device based on ECG signal - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an electrocardiograph or ECG apparatus, and more particularly to an apparatus for detecting an arrhythmia based on characteristic information of a QRS complex included in a signal generated from a cardiovascular system of a living body.

[0002]

FIG. 1 shows a portion of a typical ECG waveform, showing two normal heartbeats and one abnormal heartbeat. The QRS shape in the ECG signal is quantified by using a small number of numbers and feature information, and the ventricular extrasystole (PV
Various methods and devices for detecting C) and the like and quantifying cardiac arrhythmias are already known. In general, by grouping heartbeats into similar groups using QRS waveform information, feature information obtained from heartbeats that are similar in shape can be obtained by:
This can be used to determine whether a given heartbeat is normal or abnormal such as PVC. In the existing QRS feature information extraction device, QR
S is detected. That is, a QRS candidate is first detected, and the ECG included in the window including the detection time period is detected.
Part of the data is retrieved. Next, the ECG data in the window, such as the QRS pattern vector,
It is converted into feature information representing a QRS shape such as an S feature information vector. Until now, various techniques for converting the QRS pattern vector into a small number of types of feature information have been proposed and implemented. For example, U.S. Pat. No. 4,336,810
In the method / apparatus for analyzing arrhythmia of ECG recording disclosed in the above publication, a magnetically recorded ECG tape is scanned to detect a QRS complex. The scanned ECG signal is digitally converted and compared with known QRS patterns grouped as templates. If a pattern similar to the QRS complex compared this time is not found,
A new template is created, and the next QRS group is scanned and grouped. The grouping of each QRS complex is
Q including the width of the QRS complex and the area above and below the baseline of the QRS complex
This is performed based on characteristic information of the shape of the RS group. Each QR
Group S is normal, supraventricular extrasystole, ventricular extrasystole or unknown (un
known). In a wearable portable ECG analyzer / recorder disclosed in U.S. Pat. No. 4,583,553, a patient's ECG signal is digitized and then filtered by a hardware circuit, followed by QRS complex detection, grouping, and so on. Used as input to an algorithm for incorporation of similar QRS complexes into the population and reporting on the results. This algorithm operates in real time and operates on 43 types (event
It is designed to detect s). In addition, the use of a two-channel ECG minimizes errors associated with grouping. US Patent 4,742,4
No. 58 discloses a pattern recognition analysis method / apparatus for filtering an ECG signal and analyzing a QRS complex included in the ECG signal by a hardware circuit. By using the (sign) change point of the slope and the interval between the (sign) change points, each of which is characteristic information of the QRS complex, an index (signiture) for specifying the ECG signal is created,
When the QRS groups are detected and grouped, the QRS groups are incorporated into groups of similar QRS groups, and if there is no such matching group, a new group is created. U.S. Pat. No. 4,589,420 discloses an ECG that uses an algorithm to detect QRS complexes contained in an ECG signal and group them to determine characteristics of a patient's condition.
A rhythm analysis method / apparatus is disclosed. Several different QRS complex feature information, such as QRS complex width, RR interval, instantaneous and average heart rate, are extracted from the ECG signal. The recognition and grouping of the QRS group are performed based on the amount of noise included in the QRS group. The heart beats are grouped in two stages, first each new QRS complex is compared with the previous QRS complex and then examined by a finite state machine process based on the comparison.

[0003]

SUMMARY OF THE INVENTION The main object of the present invention is to:
An object of the present invention is to provide an apparatus for classifying a QRS complex to distinguish between a normal QRS complex and an abnormal QRS complex. Another object of the present invention is to provide a normal QRS group candidate based on N types of feature information plotted in an N-dimensional feature information space.
An object of the present invention is to provide an apparatus for determining an S group. It is another object of the present invention to provide an apparatus that distinguishes a QRS complex obtained from an ECG device into a normal QRS complex and an abnormal QRS complex based on a cluster of the QRS complex located in the feature information space. Another object of the present invention is to provide an apparatus that distinguishes a QRS complex from a normal one and an abnormal one based on the number of QRS complexes included in a cluster located in the feature information space. It is another object of the present invention to provide an apparatus for locating QRS complexes without being affected by noise and baseline drift.

[0004]

SUMMARY OF THE INVENTION The above and other objects and advantages are attained in accordance with the present invention by obtaining at least one continuous analog signal obtained from an ECG device and providing at least one digital signal based on the analog signal. Generating a plurality of scalar signals from the at least one digital signal, extracting a plurality of feature information from the scalar signal, and generating a plurality of scalar signals in a feature information space having a dimension equal to the number of extracted feature information. This is achieved by plotting the extracted feature information. The normal QRS group is determined based on the number of QRS groups included in the cluster of the QRS feature information in the feature information space and the occurrence time. Subsequent QRS complexes obtained after the determination of the normal QRS complex are the normal QRS complexes.
Groups are grouped based on the relative positional relationship with the group.

[0005]

According to the present invention, the QRS complex is determined to be normal or abnormal based on the number of QRS complexes included in the cluster of the QRS complex located in the N-dimensional feature information space and the occurrence time. Are divided into groups. Then, the presence or absence of arrhythmia is determined from a signal having the QRS group belonging to the normal group. Therefore, the arrhythmia is accurately detected.

[0006]

The above and other objects, features and advantages achieved by the present invention are disclosed and clarified by the following description of embodiments. In the drawings referred to for the description of the following embodiments, corresponding elements are generally designated by the same reference numerals. FIG. 2 shows an arrhythmia detecting device 1 of the present invention. Device 1 is connected to an existing ECG device and receives an ECG signal from that device,
Second and third channel connectors 10, 12, and 14 are included. The device 1 also receives an obtained ECG signal via a multiplexer (MUX) 16 and a filter bank 18 and converts it into a digital signal, and an analog / digital conversion circuit (ADC) 20 for processing and quantifying the digital signal. Microprocessor-22 and annotated EC
And a display 32 for displaying G data. The microprocessor 22 has an input / output (I / O) circuit 24 for receiving the digital signal and outputting the processed ECG data.
, A central processing unit (CPU) 26, a ROM 28 for storing a program command and a predetermined constant, an intermediate result of a digital signal and a program, and a variable constant (variable
and a RAM 30 for storing the variable constants.

[0007] The I / O circuit 24 outputs to the MUX 16 a timing signal T for controlling the timing of collecting ECG signals. According to the signal T, each channel connector 10, 1
The ECG signal is sampled at a frequency between 2,14 and about 250 Hertz (Hz). The sampling frequency by the signal T may be set to 300 Hz so as to cooperate with an anti-aliasing filter described later provided in the filter bank 18. The ADC circuit 20
At least 12-bit data with a unit resolution of less than 2.4 microvolts and a nominal operating voltage of ± 5 volts
This is a conventional ADC circuit that generates data.

[0008] The filter bank 18 receives the received EC.
Includes a conventional low-pass anti-aliasing filter for removing high frequency noise from the G signal. The filter bank 18 further converts extraneous AC line frequencies from the received ECG signal.
cies), and may include a conventional notch filter to remove induced noise and its harmonic noise. The AC line frequency notch filter may be configured by software and executed by the microprocessor 22. The display 32 is a conventional recording device that records ECG data, that is, an ECG waveform with a mark on a chart sheet. The display 32 may cooperate with the microprocessor 22 to display ECG data in a table.

The device 1 shown in FIG. 2 has a channel connector 1 as a means for simultaneously acquiring three analog ECG signals.
0, 12, and 14. An ECG device that supplies these three ECG signals is an ECG device (not shown) similar to a conventional one. If three ECG signals cannot be obtained from the ECG device, the microprocessor 22 can be used to output one or two signals. Two ECs
If a G signal is available, the microprocessor 22
Generates an interpolated ECG signal based on the difference between the two ECG signals. Also, if only one ECG signal is available, the microprocessor 22 uses the 0.
A first extrapolation signal equal to five times and a second extrapolation ECG signal equal to -10 times the signal are generated.

The operation of the arrhythmia detecting device according to the first embodiment of the present invention will be described below with reference to FIGS. In step S10 of FIG. 3 (hereinafter simply referred to as S10; the same applies to other steps), the operation of the apparatus 1 starts, and an initial process is performed. In this initial processing, the feature information space is cleared, and the value of the counter COUNT is reset to zero. Further, the flag FF is set to 1 to indicate the initial state.
And the counter WARM-UP-COU
The value of NT is set to a value equal to the predetermined value LEARNCOUNT. The value of LEARNCOUNT is 20
Any value may be used as long as the value is in the range from to the 500th place, but the 128th place is desirable.

Next, at S20, an ECG signal is fetched. Channel connector 10, 12, 14, MUX1
6. The filter bank 18 and the ADC circuit 20 cooperate to obtain an analog ECG signal, convert the analog signal into a digital signal, and store the digital signal in the RAM 30. The MUX 16 and the I / O circuit 24 work together to
The G signal is sampled at a predetermined sampling frequency of about 250 Hz to 300 Hz. Thus, the digital signals respectively corresponding to the ECG signals obtained via the connectors 10, 12, 14 are vectors e ₁ (t), e ₂ (t),
It is extracted as e ₃ (t) and stored in the RAM 30.

In S30, Q is added to the stored digital signal.
It is checked whether or not the RS group candidate is included. FIG. 4 illustrates specific steps of the QRS group candidate specifying routine. First, in S301, a digital signal vector is converted into a scalar signal. The scalar signal M (t) is obtained using the following equation. m (t) = | e ₁ (t) −e ₁ (t-1)] [e ₂ (t) −e ₂ (t-1)] [e ₃ (t) −e ₃ (t-1)] | M (t) = | m (t-1) + m (t) + m (t + 1) | where M (t) is a scalar signal, t is time, and number 1 is the previous or next sampling cycle EC obtained by
Used to indicate G sample. For example, if each sample is acquired at a sampling frequency of 250 Hz and at intervals of 4 milliseconds (msec), the symbol of (t-1) is the time (t)
Indicates a sample obtained 4 msec before the sample obtained in (1), and the symbol (t + 1) indicates a sample obtained 4 msec after the sample obtained at time (t). The scalar signal M (t) can also be obtained using the following equation. M (t) = | [e ₁ (t + 1) −e ₁ (t-2)] [e ₂ (t + 1) −e ₂ (t-2)] [e ₃ (t + 1) −e ₃ (t-2)] | In terms of signal processing, the scalar signal M (t) is
Conventional 3 point (12msec etc.) box car filter (3 po
3 processed by a filter consisting of an int boxcar filter
It is the absolute value of the product of the ECG signals. Other filters, such as a conventional matched filter, can be used to amplify the QRS complex and suppress noise in the digital signal, as described in more detail below.

At S302, the scalar signal M (t) is set to a predetermined threshold FIRST-THRESH (usually, units of the least significant digital signal).
By identifying a point that intersects with 64) from below in terms of bit), QRS group candidates are detected. When the scalar signal M (t) crosses the predetermined threshold from below, the window MAX-FWD-
The signal M (t) within the range of SEARCH (usually 264 msec)
Is obtained. In order to prove whether or not the maximum value of the signal M (t) has been accurately determined, the position of the maximum value can be confirmed using a moving window. The maximum value of the signal M (t) determined in this way is regarded as a QRS group candidate, and its occurrence time T _qrs is stored for use in the following steps. Next, S40 of FIG.
At 30, it is determined whether a QRS group candidate has been identified. If the result of this determination is negative, S2 is performed again.
0 is executed, and when affirmative, S50 is executed.

In S50, characteristic information is extracted from the scalar signal obtained in S301. In S501 of FIG. 5, a scalar signal around each QRS complex candidate is analyzed,
It is determined whether the QRS complex should be considered noise. For the ith QRS complex, T _qrs (i) −w _noise
The signal M (t) in the section of ≦ t ≦ T _qrs (i) + w _noise is extracted. Therefore, w _noise corresponds to half the width of the window used for noise analysis. Thus, T _qrs
It is determined whether or not a part N _i (t) of the signal M (t) centered on (i) is noise. In this analysis, the number of times the signal N _i (t) crosses the noise threshold candidate is counted for crossings in both directions. This noise threshold candidate may change as described later. The number of times of this intersection is a predetermined value MAX-SIGN-C
If less than HANGES (usually 24)
The RS group is regarded as not noise, and the noise threshold candidate becomes the determined noise threshold. On the other hand, when the number of crossings is equal to or greater than the predetermined value MAX-SIGN-CHANGES, the noise threshold candidate is determined to be a predetermined value IJ-INCREASE (usually, converted to the unit of the least significant digit by 2). ) And the noise analysis is performed again. If the number of noise threshold candidates is increased a predetermined number of times, the number of intersections becomes a predetermined value M
Before falling below AX-SIGN-CHANGES,
When the noise threshold candidate reaches the predetermined maximum noise threshold, the signal portion N _i (t) is regarded as noise, and the current analysis ends.

The noise threshold candidate changes from a small value to a maximum limit value θ _noise ^max . The small value is set to a value equal to a predetermined ratio of the evaluation value at the occurrence time T _qrs of the scalar signal M (t), and this value is a moving average with the previous noise threshold determined for the previous heartbeat ( θ)
_This value is averaged as _noise ^ave . The maximum limit value is a value obtained by _dividing the evaluation value at the generation time T _qrs of the scalar signal M (t) by the value PEAK-FACTOR (normally, 6). From these descriptions, other noise threshold modification techniques will be apparent to those skilled in the art that can be used as a means to find threshold candidates that meet the noise threshold crossing criteria.

Next, in S502, one of the characteristic information of the scalar signal is extracted. First, a time series for each QRS group k is obtained by the following equation. Here, i = 1, 2,. . . , 2W _qrs-width and W
_qrs-width represents a half interval of the window used for analyzing the _width of the QRS complex. F () is about 12
shows a conventional box car filter of msec, F ⁵ ()
This means that the box car filter is applied five times in succession. If it is determined in step S501 that the QRS group candidate is noise, the width of the QRS group is determined using the following equation. Next, the width threshold θ _qrs-width is set to a minimum value such that the width threshold crosses less than eight times. Further, the width width [k] of the k-th QRS group is determined using the following equation. width [k] = T _offset [k]-T _onset [k] where T _offset [k] and T _onset [k] are e
_width [i] Determined from the time series and the threshold θ _qrs-width .

When the width width [k] is smaller than a predetermined constant value MIN-WIDTH, or when the value θ
_qrs-width is the peak value of the value e _width (i) and the value PEAK−
PORTION-FRACTION / PEAK-POR
If it is greater than the product of TION (usually 6/8), the heartbeat is considered bad and is excluded from the following steps. Each read signal is again applied to the boxcar filter F (), and the output signal of the filter gives the second derivative. A threshold crossing criterion for finding a noise heartbeat is applied to each lead and the sum of all leads, respectively, to determine whether the heartbeat candidate should be considered noise, i.e., removed from the following steps as an inappropriate heartbeat. A decision is again made as to whether or not to do so.

Next, in S503, a plurality of areas are determined for each QRS group. Each area is determined by integrating the difference between the QRS complex and the baseline of the QRS complex. The baseline is determined by the following equation. B _i [k] = (e _i [T _onset [k]] + e _i [T _offset [k]]) / 2 where _i represents the channel number of the ECG signal and k represents the QRS complex. The signed first half area for the digital signal channel is determined by integrating the difference between the baseline and the QRS complex from the start of the QRS complex to its center. Here, the center of the QRS complex is the start of the QRS complex (on
set) and the end (offset), and the unsigned first half area obtained by integrating the difference between the baseline and the QRS group from the start to the center of the QRS group becomes the baseline and Q from the start to the end point.
This is a point in time when the area of the unsigned area obtained by integrating the difference between the RS groups becomes about の of the total area. The unsigned full area and unsigned first half area are determined by adding the unsigned full area and unsigned first half area for all digital signal channels, respectively. The center of the QRS complex can also be determined by systematically searching for the candidate value. The first half area without a sign is obtained by integrating the absolute value of the difference between the baseline and the QRS complex. The second half area with the sign and the second half area without the sign are obtained in the same manner as the first half area.

Under certain conditions, the heart rate is determined to be inappropriate and is excluded from further processing. If the difference between the values of the raw ECG signal values at the start and end of the QRS complex is greater than or equal to the sum of the absolute values of the front and back heights in each lead, the heart rate is determined to be inappropriate. You. The front height and back height values include two calculated values for each lead processed. Usually, three leads are processed, so three front heights and three rear heights are required. Here, the front height refers to the maximum amplitude (may be positive or negative) of the front half area for an arbitrary lead. The rear height refers to the maximum amplitude of the latter half area for an arbitrary lead. Also, when the number of intersections at the last threshold value selected for determining the width of the QRS complex is zero,
The heart rate is determined to be inappropriate. Furthermore, the width is the value MIN
If it is smaller than -WIDTH, the heartbeat is determined to be inappropriate. Parameters that have not been defined so far are defined below. Next, the program proceeds to S60.

In S60 of FIG. 3, the feature information extracted from the scalar signal is plotted in an N-dimensional feature information space. The dimension N matches the number of extracted feature information. As a result, a cluster of the QRS group is formed. The formed cluster is updated in this step as described later. 3 channels are received from the ECG signal,
Since four areas are determined for each channel in addition to the width of the QRS complex obtained by interpolation or extrapolation, a feature information space having a total of 13 dimensions is formed. However, the feature information of each QRS complex is
It is not an indicator of whether or not those QRS complexes are normal. For that determination, it is compared with another QRS complex. That is, each QRS complex is compared with the QRS complex included in the cluster in the feature information space in terms of shape and time of occurrence.

The feature information extracted from each QRS group is compared with the feature information extracted from the previous QRS group by a method called clustering or grouping described below. The RAM 30 stores a list of existing clusters. The list includes information on the position of each cluster in the feature information space (for example, the average value of the feature information of the QRS group included in the cluster), and the normal forward and backward connection of the QRS group located in the cluster. The interval (for example, the average value), the number of QRS groups located in the cluster, and the label or group name of the QRS group included in the cluster are included. Here, the front and rear connection intervals refer to each time interval between each QRS group and the QRS groups before and after it. The list further includes the time Tqrs of the first QRS complex located in each cluster and the latest QRS complex, two QRSs in each cluster.
The number of consecutive additions, which is the number of times the S group was added consecutively, and UNKNOWN, NORMAL, PVC, DEL
A group name indicating the type of cluster such as ETE is included. E
At the start of the analysis of the CG recording, no cluster exists. Clusters are formed in the following manner.

First, for each detected QRS group candidate, the characteristic information of the QRS group candidate is compared with the characteristic information of each cluster in the cluster list. If no cluster exists in this list, a new cluster is created, and the cluster has a position in the space corresponding to the characteristic information of the QRS complex candidate and a forward and backward connection interval of the QRS complex candidate. The occurrence time of the first and last QRS groups assigned and positioned in the cluster is set as the occurrence time of the QRS group candidate. Also, the number of QRS groups included in the cluster is set to 1, and the group name UNKNOWN
Is given. This new cluster is then added to the list.

When a cluster is present in the list, the distance between the feature information of the QRS group in the feature information space and the position of each cluster is determined, and the cluster closest to the QRS group is determined. If the distance to this recent cluster is smaller than the value MAX-DX,
RSs are added to the recent cluster. In this manner, the number of QRS groups included in the cluster, that is, the number increases, and the information is obtained by calculating the average value of the previous position of the cluster in the feature information space and the feature information of the QRS group. Update the position of the cluster in space (eg, give the previous average position of the cluster a weight of 7/8, and give the feature information of the new QRS complex a weight of 1/8); Update the forward and backward connection intervals to increase the number of consecutive additions. If the cluster closest to the QRS complex is greater than or equal to MAX-DX (usually 400), a new cluster is formed and added to the cluster list as described above.

As clusters are formed one after another, the cluster list grows. When the length of the list reaches a predetermined length, each time a new cluster is added, the old cluster is removed from the list, thereby keeping the length of the list constant. Up to 16 clusters can be listed
The accuracy can be improved without lowering the analysis efficiency. The cluster to be deleted is QRS
The time when the group was added is the oldest, and the smaller number of QRS groups among the listed clusters is 3
It is included in one part. Subsequently, S70 is executed, and the value of the counter COUNT is increased to only one.

In S80, it is determined whether or not the flag FF is 0. If YES, step S90 described later
Is executed. If NO, S110 is executed, and a sufficient number, that is, a predetermined number of QRS group candidates of WARM-UP-COUNT since the start of the operation of the apparatus has been collected and plotted in the feature information space. It is determined whether or not. Initially, the QRSs are used for forming clusters in the feature information space, and no appropriate group name is given to the clusters or individual QRSs. . if,
When the value of the counter COUNT is the value WARM-UP-COUNT
If it is less than T, S20 is executed and the subsequent QRS complex is collected as described above. The total number of QRS groups plotted in the feature information space is the value WARM-UP-C
If it is greater than or equal to OUNT, the judgment in S110 is affirmed, and in S120, each cluster is analyzed and the normal Q is analyzed.
A determination is made as to whether the data corresponds to the RS group.

In S120, first, the maximum number of Qs
The cluster containing the RSs is selected for the merging process. Since other nearby clusters may have similar feature information, the distance between the largest cluster and the other cluster is determined. The determined distance is a predetermined value MAX-DX
If so, the two clusters take the average of their respective averages, take the maximum of their respective maximums, take the minimum of their respective minimums, and give the last The addition of the QRS complex is made by adding the last QRS complex to the new cluster.

When the number of QRS groups included in the largest cluster is extremely large in comparison with the other clusters, for example, when the number is a predetermined number MIN-INI-N
If it is larger than SR-POP (usually a half of several WARM-UP-COUNTs), the cluster is determined to be normal and a normal group name is assigned. On the other hand, the normal heart rhythm state is affirmed. Q included in the largest cluster
If the number of RS groups is less than a few MIN-INI-NSR-POP, there is a possibility of a two-stage pulse. The two-stage vein is performed by determining the largest cluster and the cluster including the second largest number of QRS groups, and determining the average value of the backward connection intervals for each cluster. Here, the backward connection interval refers to a time interval between the current QRS group and the previous QRS group. The shorter rear connection interval of the two rear connection intervals is less than 0.75 times the longer rear connection interval, and the average value of the two rear connection intervals (both the shorter and the longer ones) (The average value of the backward connection intervals) is less than 7/8, and the number of both clusters is a predetermined number MAX.
-INI-BGM-TOG (usually several WARM-UP-
COUNT less than 1/8) and the total number of both clusters is a predetermined number MIN-INI-BG
If it is less than M-TOG, there is a possibility of a double vein. In this case, the cluster having the longer backward connection interval is regarded as normal, and the other clusters are regarded as PVC heartbeats. In addition, the two-stage pulse state is affirmed, and the normal heart rhythm state is denied. If the above condition for grouping the clusters is not satisfied, (i) the number POP-SUM equal to the total number of all clusters is the number WARM {UP} COUNT
Or (ii) the number of largest clusters is greater than twice the number of second largest clusters, or (ii)
i) If the maximum number of clusters is larger than half of the number POP-SUM, it is determined that the maximum cluster is normal. In this case, the two-stage pulse state is denied, and the normal heart rhythm state is affirmed. If a normal cluster is not found as a result of S130, the value WA is set in S150.
After the value LEARN-COUNT is added to RM-UP-COUNT, step S20 is executed again, and a new QRS group having the number of values WARM-UP-COUNT is collected. If the result in S130 is affirmative, the flag FF is set to zero in S140, and then S20 is executed.

When the cluster "learning" is completed by setting the flag FF to zero in S140, a plurality of additional one-time initialization steps (not shown) are executed. Except for the case where there are two-stage veins, the number of QRS groups in all clusters except for the normal cluster is set to one. When there are two-stage veins, the number of QRS groups in all clusters except for the normal cluster and the PVC cluster is set to one. Normal QRS
The group parameters are determined. That is, the average normal QRS complex width, signed area, average NN time interval, and average RR time interval are determined. The average width of the normal cluster is used as the width of the average normal QRS complex. The average area of the normal cluster is used as the signed area of the average normal QRS complex. Average N of average normal QRS complex
The N time interval is the average value of the average backward connection interval and the average forward connection interval of the normal cluster when there are no two-stage veins, and the average of the longer cluster when there are two-stage veins. This is the average value of the backward connection interval and the average backward connection interval of the shorter cluster. The average RR time interval of the average normal QRS complex is the average NN time interval, which is the R wave interval, and is equal to the average NN time interval. After the cluster "learning", when new QRS group candidates are detected, they are classified into clusters and given group names. It is possible that the number of new QRS groups in a cluster that has not yet been given a group name will grow to a sufficient number to allow the cluster to be reliably given a group name.
There are a plurality of group names, for example, normal (NORMA
L), premature ventricular contraction (PVC), deletion (DELETE)
D). When one QRS group belonging to a cluster to which a group name has not been given is grouped as normal, the entire cluster may be grouped as normal by a rule described later. However, this is done because the total number of QRS groups in the cluster is the value MIN-
If the number of normal QRS groups included in the cluster is greater than NEW-NORMAL-POP (usually 10) and is greater than half the total number of QRS groups in the cluster, or the number of normal QRS groups included in the cluster is Only when the number is larger than four times the number of abnormal QRS groups included in the cluster. One QR belonging to a cluster for which a group name has not yet been given
When group S is grouped as DELETED,
The entire cluster may be grouped as "deleted" according to rules described later. However, this is done because the total number of QRS groups in the cluster is the value MIN
Only if it is greater than NEW-DELETE-POP (usually 6) and the number of “deleted” QRS groups included in the cluster is greater than half the total number of QRS groups in the cluster. After cluster learning, new QRS complexes may be detected due to external effects such as patient movement. In this case, when the patient has a very messy rhythm, a number of new clusters are formed due to the noise included in the data, which replaces the existing clusters. In this state, the Q included in each cluster
The number of RS groups is set to 1, and the cluster name given to the cluster is deleted. Thereafter, an operation of giving a group name to the cluster is performed again as a new QRS group candidate is detected. As new QRS group candidates are detected and given group names after cluster learning, the average NN
Overall rhythm / shape information including time, average NW time, average NSS time, and average RR time is updated. Each time a heartbeat is grouped as normal, the average NN is updated. Average N
The value of N hours is updated to (7 × average NN + RR) / 8. In this equation, the average NN time is the current value of the average NN time, and the RR time is the time between normal heartbeats. Each time the heartbeat is grouped as normal, the average NW time and average NSS time are updated. The value of the average NW time is
(7 × average NW + width) / 8 is updated. In this equation, the width is the measured width of the current normal heartbeat. Average NSS
Is updated to (7 × average NSS + SAS) / 8. In the formula, SAS is the total area having the sign of the normal heartbeat this time, that is, the sum of the first half area and the second half area for all leads. Further, when a heartbeat is detected, the average RR time is updated to (7 × average RR + rr) / 8 regardless of whether the heartbeat is normal or a PVC heartbeat. In this equation, rr is the time between the current heartbeat and the previous heartbeat. The rhythm of the patient may be (1) a normal one with a fairly regular connection interval, (2) an irregular one with a fluctuating connection interval, or (3) a fluctuating connection interval. Rare clusters are formed (un
or (4) a large number of clusters are newly formed due to large fluctuations in the connection interval (bizarre
) Is evaluated. The variability of the connection interval is determined based on the RR interval, not the NN interval. If the criteria for determining the condition of an individual patient are met, the condition is affirmed; otherwise, it is denied. Each state is ranked from "normal" to "random", with "random" having the first rank. Thus, if the criteria for "messy" are met, a "messy" state is declared, whether rare or not. Each time a condition is recognized, this recognition is shown as an output and the number constant RHYTHM-COUNT is set. Upon the first detection of each state, that state is set to this value. Each time a new heartbeat is processed, the state count is decreased. When the state count for each state decreases to zero and the criteria for the other state are met, a new output indication is issued. Only the group-named heartbeats (not including those with the group-name extension deferred) are processed and used to determine the patient's rhythmic state. The cluster list is consulted for heart rate processing to determine the rate of cluster formation. The cluster formation speed is determined by counting the number of clusters included in the cluster list, which is smaller than the number RECENT-SHAPES (usually 32), within a predetermined period. If the patient is in an unusual rhythm ("messy" or "rare") and the criteria for the unusual rhythm (ie, many new clusters have been formed and the intervals are irregular) are no longer met Then, the rhythm state is changed to a normal rhythm state, and relearning is started. “Messy”
Neither the normal state nor the "rare" state is negated, and the count RECENT-SHAPES has the value shape-count.
t-LO-THRESHOLD (usually 5) and the count RR-difference is equal to the value R
If it is smaller than Rdif-LO-THRESHOLD (normally, 5), relearning is appropriate. If the re-learning is appropriate and the “messy” state is affirmed, the re-learning is started, the number of QRS groups in the cluster is set to 1, and
The “normal”, “abnormal”, and “delete” counts are set to zero. Thereafter, the "normal" state is changed to the value RHYTHM-COUN.
T (usually 100), the "irregular" state is negated, the "rare" state is negated, and the "messy" state is negated. The second rank is a test for a "random" condition.
The "random" state is denied and the count RECENT-SH
APE is the value shape-count-HI-THRES
HOLD (usually 10) and count R
R-difference is the value RRdif-count
If -HI-THRESHOLD (usually 9) or more, the "messy" shape is set to the value RHYTHM-COUNT, the "normal" state is negated, the "irregular" state is negated, and the "rare" state Is denied. In addition, the "rare" state or the "random" state is denied, and the count RECEN
T-SHAPE has the value shape-count-LO-T
Greater than HRESHOLD and count RR-d
if the value is RRdif-count-LO
If not less than -THRESHOLD, the "rare" state is set to the value RHYTHM-COUNT, the "normal" state is negated, the "irregular" state is negated, and the "random" state is negated. Furthermore, the "irregular" state is denied,
The count RR-difference is equal to the value RRdif-
count-LO-THRESHOLD or more, and the count RECENT-SHAPE is the value shape.
If less than -count-LO-THRESHOLD, an "irregular" state and relearning can be set. In an "irregular" state and a state in which re-learning can be set, and
If the “random” state is set, re-learning is started, the number of QRS groups in the cluster is set to 1, and the “normal”, “abnormal”, and “delete” counts are set to zero. Also, the "irregular" state is set to the value RHYTHM-COUNT,
The "normal" state is negated, the "rare" state is negated, and the "messy" state is negated. The above reference test is performed for each of the valid heartbeats after grouping. After performing the test, the "ragged" status count decreases if greater than zero, the "rare" status count decreases if greater than zero, and the "random" status count decreases if greater than zero. Finally, if the "messy" or "rare" condition is asserted, an unnormal condition is asserted. When new QRS complex candidates (heartbeats) are detected after cluster learning, rhythm and shape limits are determined before analyzing each heartbeat. The parameters early and late are defined as 7 × average NN / 8 and 3 × average NN / 2, respectively. In addition, the parameters early-fusion, late-
fusion is 7 × average NN / 8 and 9 × average N
Defined as N / 8. Further, the parameter low-c-
Pause and hi-c-pause are defined as 29 × average NN / 16 and (average NN + late), respectively.
The parameters wide and narrow are defined as 5 × average NW / 4 and 5 × average NW / 8, respectively. The parameter Twv-RR is the maximum time during which a T-wave is generated, and is usually set to 350 msec. The parameter PZ-DUR is the length of time between heartbeats required to qualify as a pause and is typically set to 2.5 seconds. The parameters NN-TOL and NNZ-TOL are tolerances for comparing the RR interval with the average NN for the non-noise heartbeat and the noise heartbeat, respectively, and are typically 48 s.
ec and 96 sec. When a new QRS group candidate is detected, some of the QRS group candidates may be determined to be inappropriate and may be excluded from further processing. The RR interval for each heartbeat is less than the value earlyT,
The assertion LowHeight () is affirmed, and the number of sign changes determined at the time of noise detection is represented by the value SIGNC.
The number of sign changes in the second derivative of the signal in the window used for the width determination is greater than H-NOBEAT (typically 5) and greater than the value SUM123-NOBEAT (typically 13); If the total unsigned area for the lead is greater than the value SUMU-NOBEAT (typically 12500), then the heartbeat is deemed inappropriate. The area is the sum of multiple samples for a time interval, and the sampling rate is typically 250 Hz
It is. The analog signal obtained from the lead is converted to a digital signal after being amplified, and the amplification gain is adjusted by automatic gain control. Also, assertion Lo
wHeight () is affirmed and the width is the value WIDTH2-
If it is greater than NOBEAT (typically 160 msec), the heartbeat is deemed inappropriate. Further, when the RR interval is 2
If it is smaller than 00 msec and the previous heartbeat was normal, the heartbeat is determined to be inappropriate. Also, the previous heart rate is normal, the RR interval is smaller than the value earlyT, and the width is the parameter narrow + NW-T.
OL (usually 16 msec), the assertion LowHeight () is affirmed, and the number of sign changes determined at the time of noise detection is the value SIGNCH-NOBE.
If it is greater than AT (usually 5), the heart rate is deemed inappropriate. The parameter NWZ-TOL is usually twice the parameter NW-TOL. In S90 of FIG.
A plurality of rules such as the rules shown in Tables 1 to 4 below are applied to the QRS group candidates included in the digital signal, and each QRS group is grouped, and a normal QRS group and an abnormal QRS group (That is, a group other than the normal or deleted QRS group), a group name such as a deleted QRS group is given.

[0029]

[Table 1]

[Table 2]

[Table 3]

[Table 4]

These rules are weighted so that a QRS group included in a cluster to which a group name has already been assigned can be given a group name of that cluster preferentially. Thus, each QRS group is given a group name upon reception. However, the grouping of the QRS groups cannot be performed immediately, and the grouping of the QRS groups may be delayed until a subsequent QRS group is received, and the QRS group may be put into a standby state. The maximum number of QRS groups that can be put on standby is a predetermined number such as five, and in that case, the QRS groups supplied later are added, and the maximum number of uncropped QRS groups is six. When six or more QRS groups are made to be on standby, the oldest standby QRS group is deleted with the same effect as the deleted QRS group. 6 uncropped QRs in total
If there are S groups, the six QRS groups are grouped as follows. The grouping rules in Tables 1 to 4 are applied in the order in the table according to the number of QRS groups waiting. For example, in the case of one beat, there is no waiting heartbeat, and in the case of two beats, the present heartbeat and one waiting heartbeat are included. Each item in the left part of Tables 1 to 4 is an assertion, and the number in parentheses indicates the QRS group to which the assertion is applied. The QRS complex indicated by the smallest number is the oldest QRS complex. For example, assertion NormL (1) asks whether the oldest QRS complex is included in the normal cluster, and assertion NormL (5) asks whether the fifth standby QRS complex is included in the normal cluster. Ask. Further, the assertion NormT (3-2) includes the third and second standby QRs.
It is asked whether or not the difference between the occurrence times of the S group is within the normal range.
Expressions such as NormT (2) are NormT (2-0)
Where the zeroth QRS complex is the QRS complex that occurred immediately before the heartbeat sequence in question.
In Tables 1 to 4, the symbol “&&” is a logical product (AN
D) and the symbol “||” represents a logical sum (OR). The assertion includes a cluster including the QRS group, a noise component of the QRS group, a shape of the QRS group, a generation time of the QRS group, and
Handles the overall rhythm of a digital signal.

Each heartbeat grouped according to the plurality of rules in Tables 1 to 4 is categorized as one of normal, T-wave, deleted, abnormal, or ungrouped. The grouping instructions in the right part of Tables 1 to 4 are grouped according to various group names, and all grouping instructions other than standby, normal, T-wave, deleted, and ungrouped are classified as abnormal. For example, the group SVPCWP (supraventricular extrasystole with pause) is abnormally categorized. In addition, group SVPC (supraventricular extrasystole), VentESC (ventricular escape beat), PVC (ventricular extrasystole), SupV
(Super ventricular), fusion, and funny are all classified as abnormal heartbeats. The symbol SetQ () is not a grouping instruction but a function described later.

Assertions related to the positioning of a cluster include assertions NormL (), NoL (), PV
C () and DelL () whose assertions are normal, ungrouped, PVC
Or ask if you are positioned in a cluster grouped for deletion. In addition, assertion SameCl
The user (a, b) asks whether the QRS groups a and b belong to the same cluster, and issues an assertion SmallPo.
p () asks whether the number of clusters including the QRS group is less than a predetermined value MIN-NEW-NORMAL-POP. Assertions related to noise components are assertion Noisy () and NotNoisy.
(), And as described above, it is asked whether or not the portion including the QRS complex is determined to be noise.

Assertions handling the shape of the QRS complex include assertions WideW (), NormW (), N2W
() And NNoiseW (). Assertion W
ideW () indicates that the width of the QRS complex is (5 × average NW) /
Affirmative if greater than the value Wide defined as 4. The assertion NormW () is affirmative if the QRS complex width is between the value wide and the value narrow. However, the value narrow is initially (5 × average NW) / 8
And the average NW is updated to (7 × average NW + width) / 8 after two consecutive normal heartbeats are captured. Assertion N2W () and NNoiseW () are QRS
Affirmative if the group widths are within the average NW ± NW-TOL and average NW ± NWZ-TOL, respectively. The value NW-TOL and the value NWZ-TOL are predetermined allowable error values (typically, 16 ms and 32 ms, respectively).
ec). The assertion LowHigh () is
The height of the QRS complex is a predetermined value MIN-HIGH
(Normally 300 times the minimum AD step height). Also, the assertion DiffAre
aSign (a, b) indicates that Q
This is true when the sign of the signed first half area or the second half area of each of the RS groups a and b is different.

Assertion AbnormalRhyth
m () is a rarely asserted assertion that is affirmed when the patient is in an abnormal rhythmic state (when many small clusters are newly formed). For example, affirmative if the patient's cardiac rhythm is "messy" or "rare". The assertion related to the time of occurrence deals with the RR interval of the QRS complex. 6 and 7 show the average N used to evaluate time handling assertions.
A plurality of sections are represented around N. In the figure, symbols below the line indicate endpoint values based on the average NN. The upper section of the line is derived from the lower value of the line. For example, the assertion NormT () indicates that the RR interval has the value e.
It is affirmed if there is between early (7/8 * average NN) and value late (3/2 * average NN). Assertion Earl
yT () and LateT () have the RR interval value ear
If the value is less than ly and greater than the value late, respectively, the result is affirmed. Assertion GoodT () is affirmed if the RR interval is between the values early and late (if the occurrence time is good). Assertion Fusi
onT () is a value in which the RR interval is early-fusion.
Affirmed when it is between and the value late-fusion,
Assertion OKT () is affirmed if the RR interval is between the value early and late-fusion, and assertion Cpause () is that the RR interval is low-value.
It is affirmed if it is between c-pause and the value hi-c-pause. In addition, assertion LT-c-Pau
se () is affirmed if the RR interval is less than the value low-c-pause. Assertion RunT ()
Is affirmed if the RR interval is less than the average NN + early. Also, some time assertions
It deals with whether the RR interval is within a predetermined tolerance of the average NN. For example, assertions NNT () and ZNNT () are the same, and the RR interval is the average NN
± NNNTOL (value NNTOL is about 48 msec) and average NN ± NNTZTOL (value NNZTOL is about 96
msec) is affirmative. In addition, assertions SlateT () and SearlyT ()
Is affirmative if the RR interval is within the range of average NN ± (NNTOL / 4). Finally, assertion LongP
auseT () is a value in which the RR interval is a predetermined value PZ−
An affirmation is made if the value is greater than DUR (for example, about 2.5 msec), and the assertion TwaveT () indicates that the RR interval is a predetermined value Twv-RR (for example, about 350 m).
Yes if less than sec).

In S90 of FIG. 3, the number of QRS groups on standby is determined, and a rule suitable for the number is applied to the QRS groups requiring grouping. An exception is the case of 6 beats, which will be described below. When all the assertions shown in the respective lines in the left part of Tables 1 to 4 are affirmed, the group name shown in the right part of the same line is given to the QRS group. For example, if the assertion OKT (3) is affirmed in the case of three beats, the newest QRS group is a normal QRS group, and the remaining two QRS groups are a deleted QRS group. The latest QRS complex in the case of three beats is the third heartbeat. D () shown in the right part,
Grouping indication symbols such as N (), PVC (), F () indicate that the () th QRS group is deleted, normal, PV
C, indicates that a fusion group name should be given. A grouping indication symbol such as D () indicates that a cluster including the QRS group should be grouped according to a cluster update rule described below.

The symbol SetQ () indicates that the heartbeat is not classified,
A group of additional rules to indicate grouping when the heartbeats before and after that heartbeat may not have been categorized. The rule denoted by the symbol SetQ (p) is understood using the following definitions. That is, the value RR is the heart rate p
And the time interval between heart rate p-1. Also, the value CRR
Is the time RR corrected for the heart rate center already described
It is. Further, the value RR2 is determined by the heart rate p-1 and the heart rate p + 1.
Is the time interval between Assertion NotNoisy
If (p) && NotNoL (p) is affirmed, heart rate p
Gets the group name of the cluster that contains it. Otherwise, the next assertion (5 × average NN / 8 <RR <5 × average NN / 4) &&
(NNoiseW (p)) && (Average NN-NNZTO
L <CRR <average NN + NNZTOL) && (average NW
-NNZTOL <heart rate p width <average NW + NNZTO
If L) is affirmative, the heartbeat p is determined to be normal. Otherwise, the next assertion (7 × average NN / 4 <RR2 <5 × average NN / 2) &&
(RR <7 × average NN / 8) && Not (average NW−T
If (OL <width of heartbeat p <average NW + TOL) is affirmed, heartbeat p is determined to be PVC. Where the value T
OL is 5 for the sampling frequency. Otherwise, if the abnormal condition is denied, a determination is made as to whether the occurrence time is so fast that it can be considered a T-wave. If the time interval RR between the current heartbeat candidate and the previous heartbeat is less than 350 msec, the heartbeat candidate is determined to be a T wave. Then, if the T wave is found, a group name of "delete" is given. Otherwise, if NotAbnormalRhythm is affirmative, the heartbeat p is given a group name of "suspicious". Otherwise, if RR <7 × average NN / 8 is affirmed, the heartbeat cluster is updated as if the heartbeat were a PVC heartbeat. Otherwise, if RR <3 × average NN / 2 is true, the heartbeat cluster is updated as if the heartbeat was a normal heartbeat. Otherwise, the heartbeats are regrouped as PVC heartbeats and the heartbeat cluster is updated accordingly. A cluster to which a group name is assigned may include a QRS group having a group name different from the group name.

When grouping six QRS groups, an additional rule is applied in order to repeatedly perform the grouping of the QRS groups. First, check Norm
L (6) && NormL (5) && GoodT (6-5) is applied, and when this is affirmed, SetQ (1), SetQ (2), SetQ (3), SetQ are applied to those six heartbeats, respectively.
It is instructed to give group names of Q (4), N (5), and N (6). On the other hand, if this check is denied, it is confirmed whether the heartbeat is located in the cluster to which the group name has already been given in order from the latest QRS group. This is an iterative process. The heartbeat which is a problem this time is assumed to be a heartbeat n. It is determined whether the check NormL (n) is affirmed for each QRS complex (heart rate n). If affirmative, the group name N (n) is given to those heartbeats, and the process proceeds to the next heartbeat. If the check NormL (n) is denied, it is determined whether or not those heartbeats are actually T waves of the ECG signal. Check NotGoodT (n- (n-1)) (where n represents the current QRS group and n-1 represents the previous QRS group)
If both and CheckLowHeight (n) are affirmative, the nth QRS complex is a T-wave and is grouped as "deleted".

Finally, a check is made as to whether the QRS complex is within the normal range. (I) Check Noisy (n)
Is affirmed and check PVCL (n) is denied, or
Or, when the width of the n-th QRS complex is a value between the value narrow and the value (wide + NW-TOL) (where NW-TOL is a predetermined allowable error value), or (ii) The sum of the areas with sign for all leads is (3 × average NSS) / 4 and (5 ×
When the value is between (average NSS) / 4 (however, the average NSS)
S is the average value of the sum of the signed first half area and the signed second half area of the QRS groups included in the normal cluster), and those QRS groups are considered to be within the normal range. If the heartbeats are within the normal range, it is determined whether those heartbeats may be grouped as "normal". QRS
The difference between the RR intervals of the group n and the QRS group m is the value (ea)
(rly-NNTOL) is searched for (where the value NN-TOL is a predetermined value),
If the QRS group n and the QRS group m are adjacent and therefore m = n-1, a group name N (n) is assigned. However, when the QRS group candidates exist between the heartbeats m and n, the following rules are applied to those intermediate QRS groups.
Rules apply. Check Noisy () or Low
If Height () is affirmative, the QRS group is given a group name D (). Otherwise, if the check AbnormalRhythm is affirmed, the group name PVC () is given. Otherwise, if the cluster to which the heartbeat belongs has a group name, the heartbeat is given the same group name as the group name of the cluster. Otherwise, the QRS complex is given a group name of "suspicious". As described above, when six heartbeats are repeatedly processed from the latest heartbeat to the oldest heartbeat, the second stage of the grouping operation is started, and the six heartbeats are again repeatedly subjected to the work. Is done. The process starts with the oldest heartbeat and ends with the latest heartbeat. The heartbeat processed this time is defined as a heartbeat p. If the current heartbeat has already been given a group name, the processing of the current heartbeat is omitted and the processing of the next heartbeat is started. This time heart rate p is p <6, check NotNo
rmW (p) is affirmed and check EarlyT (p)
Is affirmative, check NotNoisy (p) is affirmative, and the interval between heartbeats p and p + 1 is the value early-
If greater than fusion, group name PVC
(P) is given. If the heart rate p is p <6 and the check NoL (p + 1) is affirmed, the group name N (p) is given. Otherwise, check NoL (p + 1) is affirmed and check NotNor
mW (p) is affirmed and the check NotNormW (p
+1) is affirmed, and the check TwaveT ((p +
1), p) are affirmed and the check DiffAre
If aSign (p, (p + 1)) is affirmative, the group names PVC (p) and D (p + 1) are provided. Otherwise, if the check Nosy (p) is affirmed, or if the check LowHeight (p) is affirmed, the group name D (p) is given. Otherwise, if the check AbnormalRhythm is affirmed, the group name PVC (p) is given.
Otherwise, if the check NoL (p) is negative, the heartbeat is given the group name of the cluster containing that heartbeat. Otherwise, the number of normal heartbeats or the number of abnormal heartbeats in the cluster containing that heartbeat is the value MIN
If the number of abnormal heartbeats is larger than the number of normal heartbeats when the number of abnormal heartbeats is larger than the number of abnormal heartbeats, the heartbeats are grouped as "abnormal". Group with "normal". Otherwise, group the heartbeat as "undecided". Then, the operation proceeds to S100. After the heartbeat grouping instruction is determined, the cluster to which the heartbeat belongs may be updated according to the classification of the group name of the heartbeat. That is, if the heart rate is normal, the cluster is also “normal”, all other true QRS groups including PVC are “abnormal”, and false QRS groups such as T-wave or noise are “abnormal”. "Delete". When updating the group name of the cluster according to the normal heartbeat, the number of normal heartbeats of the cluster is increased, and a test is applied as to whether or not the cluster can be made “normal”. This test is the same as the test applied when a group name is not assigned to a cluster as described later. When updating the group name of the cluster according to the abnormal heartbeat, the number of abnormal heartbeats of the cluster is increased, and a test is performed as to whether or not the cluster can be made “abnormal”. This test is the same as the test applied when a group name is not assigned to a cluster as described later. When updating the group name of the cluster according to the deleted heartbeat, the number of deleted heartbeats of the cluster is increased, and a test is performed as to whether or not the cluster can be “deleted”. This test is the same as the test applied when a group name is not assigned to a cluster as described later.

In S100, one or more Qs
After the grouping of the RS groups, clustering is performed according to the cluster update rule. When a QRS complex is determined to be normal, all Qs included in the cluster
The number of RS groups is a predetermined value MIN-NEW-NO
If the number of normal QRS groups included in the cluster is larger than RMAL-POP (for example, the number of QRS groups grouped as “normal”) is greater than half the number of all QRS groups, the cluster is “ "Normal". The number of all QRS groups in the cluster is the value MIN-NEW.
If it is larger than -NORMAL-POP and the number of normal QRS groups of the cluster is more than four times the number of abnormal QRS groups, the cluster is determined to be "normal". When a certain QRS group is PVC, the number of all QRS groups included in the cluster is determined by a predetermined value MI.
If it is larger than N-NEW-ABNORMAL-POP and the number of PVCQRS groups included in the cluster is larger than half the number of all QRS groups, the cluster is determined to be “PVC”. Also, the number of all QRS groups in the cluster is larger than the value MIN-NEW-ABNORMAL-POP, and the number of PVC QRS groups is normal QR.
If it is larger than twice the number of S groups, the cluster is set to “PVC”. Finally, when a certain QRS group is deleted, the number of all the QRS groups included in the cluster is set to a predetermined value MIN-NEW-DELETE-
If the number is larger than the POP and the number of deleted QRS groups included in the cluster is larger than half of the number of all QRS groups, the cluster is determined to be “deleted”. Thereafter, the operation of the program returns to S20 and below, and processes a new ECG signal.

Next, a second embodiment of the present invention will be described. Also in this embodiment, the apparatus 1 having the configuration shown in FIG. 2 is used, and the entire operation is performed according to the flowchart of FIG. 3, but in this embodiment, S30 of FIG. Is done.

[0041] In S311 of FIG. 8, the reference point evaluation function Q (i) has the formula Q (i) = Q (f a (i), f b (i), f c (i), f
_d (i)). Where f _a (i), f _b
(i), f _c (i), and f _d (i) are each the output of a filtered ECG signal that generates a base vector, as described in more detail below. The reference point evaluation function Q (i) of this embodiment is given by the following equation. Q (i) = A | f a (i) | + B | f b (i) | + C | f c (i) | + D | f d (i) | where, A, B, C, D are determined in advance Is a constant.

FIG. 9 shows the flow of information in the operation step of S311. The QRS detection / feature information extraction device 2 can be constituted by software stored in the ROM 28 of the microprocessor 22, and includes first, second, third, and fourth filters 101, 10
3, 105, and 107. Each filter outputs characteristic information based on the received digital signal. Apparatus 2 also includes first, second, third, and fourth full-wave rectifiers 109, 1
Each full-wave rectifier receives an output signal of a corresponding filter and extracts a characteristic information signal. Apparatus 2 further includes an adder 117, which combines the rectified feature information signals into a single output, a combined feature information signal or function Q (i). Device 2
Also, each of the filters 101, 103, 105, 107
And output the characteristic information signal for the same processing as that of S60, including output terminals 119, 121, 123, and 125. Each filter 101, 103, 105, 107
Generate different basis vectors, such as the basis vectors shown in FIG. Filter 101,
Other basis vectors can be generated by 103, 105, 107 and the filters 101, 1
Several filters (not shown) on 03, 105 and 107
Can generate a larger number of basis vectors. In the present embodiment, the device 2 is configured by software provided in the microprocessor 22, but may be configured by using a dedicated circuit element.

The basis vectors are selected to generate a filter output corresponding to the shape of the QRS complex included in the digital signal. In FIG. 9, each filter receives the same digital signal but produces a characteristic output different from the output of all other filters. The basis vectors are partially linear and may include a small number of line segments. These basic vectors remove noise such as AC line noise and baseline drift. The basis vectors are also generated at a frequency of 250 Hz, with windows in the range of about 100-200 msec. Thus,
The filters 101, 103, 105, and ₁₀₇ generate outputs at all points of the input digital signal, and can select a reference point such as the time _Tqrs based on the characteristic information. Each rectifier 109, 111, 113, 115
Generate a rectified feature information signal, ie, its absolute value, of the output from the corresponding filter. The outputs of all the rectifiers are collected in the adder 117, and generate an added output of a characteristic information signal such as a function Q (i).

In S312, the reference point evaluation function Q (i)
Is determined from the function Q (i). The time series Q (i) of this embodiment is represented by S30
Functions in the same way as the function M (t) given by
Is determined and used. The reference point T _qrs is specified for each specified maximum value. Thus, the reference point T
_qrs is obtained by the maximum value of the feature information signal obtained by the basic vector. The program then proceeds to S in FIG.
Proceed to 40. The operation of the apparatus 1 in the second embodiment is the same as that in the remaining steps of FIG.

The device 2 of FIG. 9 can be used together with another ECG evaluation device or arrhythmia detection device. Also,
The output of device 2 is a plurality of feature information signals, which can be plotted in N-dimensional feature information space and used to form clusters, as described above.

In addition, the present invention can be carried out in various modified and improved forms based on the knowledge and skills of those skilled in the art. Therefore, only a limited embodiment will be described here, but many embodiments can be considered without departing from the spirit of the present invention.

[Brief description of the drawings]

FIG. 1 illustrates a portion of a typical ECG waveform.

FIG. 2 is a schematic diagram showing an arrhythmia detection device of the present invention.

3 shows a high-level flowchart of the operating steps performed by the device of FIG. 2 to group QRS complexes.

FIG. 4 is a low-level flowchart of operation steps executed in step S30 of FIG. 3 according to the first embodiment of the present invention.

FIG. 5 is a low-level flowchart of operation steps executed in step S50 of FIG. 3 according to the first embodiment of the present invention.

FIG. 6 is a diagram showing a timing pattern of RS group evaluation in step S90 of FIG. 3;

FIG. 7 is a diagram showing an average QRS group width and a normal range.

FIG. 8 is a low-level flowchart of operation steps executed in step S30 of FIG. 3 according to another embodiment of the present invention.

FIG. 9 is a diagram showing a flow of information in an operation step S30 of FIG. 3 in still another embodiment.

FIG. 10 is a schematic diagram conceptually showing an output signal of the basic filter of FIG. 9;

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-2-198538 (JP, A) JP-A-54-124591 (JP, A) US Patent 4,894,920 (US, A) US Patent 4,853,553 (US, A) U.S. Pat. No. 4,742,458 (US, A) U.S. Pat. No. 4,433,810 (US, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) A61B 5/0472

Claims (3)

(57) [Claims] 1. An arrhythmia detection device, comprising: acquiring means for acquiring at least one analog ECG signal at a predetermined sampling cycle; and at least one analog ECG signal connected to the acquiring means. Converting means for converting the digital signal into digital signals; processing means for generating diagnostic information based on the at least one digital signal; and display means for connecting to the processing means and displaying the diagnostic information. A first means for determining a normal QRS group based on a plurality of first QRS groups; and a second means for grouping each of a plurality of second QRS groups based on the normal QRS group Means for generating a plurality of scalar signals based on the at least one digital signal; and Means for determining an occurrence time of each of the plurality of first QRS groups by detecting a plurality of peak values included in the scalar signal; and for each of the first QRS groups, a QRS width, a plurality of first half areas, a plurality of second half areas, Means for calculating the first half area having a sign obtained by integrating the difference between the base line and the QRS group, and the absolute value of the difference between the base line and the QRS group.
The plurality of second-half areas obtained by integrating the difference between the baseline and the QRS complex.
Absolute value of the difference between the second half area and the base line and the QRS complex with that code
And a second half area having no sign obtained by integrating . 2. The method according to claim 1, wherein the at least one ECG device is provided with at least one
An arrhythmia detection device that detects an arrhythmia based on at least one digital signal obtained from the plurality of analog ECG signals, the device including a processor having an input circuit that obtains the at least one digital signal, Determining means for determining a noisy or non-noisy noise attribute for the at least one digital signal based on a generation time of the QRS group candidate according to a threshold crossing criterion; and a generation time of the QRS group candidate and the noise attribute. Extracting means for extracting N types of feature information from the at least one digital signal based on the at least one digital signal; plotting means for plotting the QRS group candidates in an N-dimensional feature information space according to the extracted N types of feature information; A plurality of first QRS groups are generated by repeatedly operating the determining means, the extracting means, and the plotting means. Means for generating at least one cluster from the plurality of first QRS group candidates by plotting complements, means for determining a normal cluster based on the at least one cluster, the determining means and the extraction Means for plotting a plurality of second QRS complex candidates by repeatedly operating the means and the plotting means, thereby adding or not adding at least one second QRS complex candidate to the at least one cluster.
Means for at least one of generating a plurality of clusters, and each of the plurality of second QRS group candidates included in the digital signal is classified into the N clusters extracted from the normal cluster and each of the second QRS group candidates. Means for grouping based on the characteristic information. 3. At least one of the ECG devices supplied from the
An arrhythmia detection device that detects an arrhythmia based on at least one digital signal obtained from the plurality of analog ECG signals, the device including a processor having an input circuit that obtains the at least one digital signal, Extracting means for extracting N types of feature information from the at least one digital signal based on the occurrence time of the QRS group candidate; and storing the QRS group candidate in an N-dimensional feature information space according to the extracted N types of feature information. Plotting means for plotting; means for generating at least one cluster from the plurality of first QRS group candidates by repeatedly operating the extracting means and the plotting means to plot the plurality of first QRS group candidates; Means for determining a normal cluster based on the at least one cluster; By repeatedly operating the plotting means and plotting a plurality of second QRS group candidates, it is determined whether the second QRS group candidate is added to the at least one cluster or at least one new cluster is generated. Means for performing at least one of: the maximum number of the plurality of second QRS complex candidates including both the true QRS complex and the non-QRS complex signal; and the N extracted from the normal cluster and each of the second QRS complex candidates. An apparatus comprising: means for performing grouping based on the type of characteristic information and removing the non-QRS group signal at that time; and means for displaying diagnostic information based on the grouped QRS group. .