JPH03218730A - Electrocardiogram waveform recognizing device - Google Patents

Abstract

PURPOSE:To improve the recognition accuracy and the maximum likelihood of a partition point by using two kinds of neural networks, and executing the recognition of a partition by taking notice of an electrocardiogram waveform of that time and a timewise structure of the partition. CONSTITUTION:At the time of recognizing a partition at the time point (t), a value of an electrocardiogram waveform to be recognized and a prediction Ft of the partition of the time point (t) of a partition buffer 14 are inputted to a pattern associative type neural network 12. Consequently, a recognition result Kt of the partition of the time point (t) is obtained as the output of the pattern associative type neural network 12. This recognition result Kt becomes the input of a recurrent neural network 13. Thus, the prediction Ft+1 of the partition of a time point (t+1) is obtained as the output of the recurrent neural network 13. This prediction Ft+1 is stored in the partition buffer 14, and used for recognizing the partition at the time point (t+1). The partition buffer for holding the prediction of the partition of the next time point is initialized prior to the recognition.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an electrocardiogram waveform recognition device, and more particularly to an electrocardiogram waveform recognition device that recognizes the classification of electrocardiogram waveforms.

[Conventional technology]

FIG. 8 is a waveform diagram showing an example of a general electrocardiogram waveform and its classification.

Conventional electrocardiogram waveform recognition devices use a low-pass filter or hand-pass filter to remove noise from the input signal, perform differentiation on the result, and calculate the difference in order to recognize the division points of the waveform shown in FIG. The point where the value becomes O is detected as the peak point of the R wave. Furthermore, in order to detect the peak point of the R wave, it is also possible to perform second order differentiation on the output of the filter and detect the time point at which the output reaches the maximum as the peak point of the R wave. After finding the time point of the peak point of the R wave, the dividing point between the R wave, Q wave, and S wave is found using the point where the input waveform intersects with 0. After that, using the division points of the R wave, Q wave, and S wave and the point where the input waveform crosses the zero point again, the division point of the S wave and T wave, and the division point of the Q wave and P wave are determined. find.

[Problem to be solved by the invention]

The conventional electrocardiogram waveform recognition device described above recognizes other division points based on finding the peak point of the R wave, without paying attention to the temporal structure of the electrocardiogram waveform. Therefore, if the peak point of the R wave is incorrectly recognized, all segmentation points will be incorrectly recognized.

The purpose of the present invention is to improve both the precision and accuracy of recognition of division points by using two types of neural networks to recognize divisions by focusing on the electrocardiogram waveform at that time and the temporal structure of the divisions. An object of the present invention is to provide an improved electrocardiogram waveform recognition device.

[Means to solve the problem]

The electrocardiogram waveform recognition device of the first invention includes a standard waveform pattern that is a standard electrocardiogram waveform pattern and a standard waveform division pattern memory section that stores the divisions of the waveform, and the electrocardiogram waveform pattern and predicted one time. A pattern associative neural network that recognizes the waveform division at the current time from the previous division, and a current neural network that predicts the division of the time from the time series of the division pattern up to one time ago, and the predicted division. It has a partitioned buffer for holding, and a learning/recognition control unit that controls learning and recognition operations of the pattern associative neural network and the recurrent Φ neural network.

Further, in addition to the electrocardiogram waveform recognition device of the first invention, the electrocardiogram waveform recognition device of the second invention includes, during learning, a time near the current time as input to the pattern associative neural network of the first invention. It is configured with an input waveform buffer that holds standard waveform patterns.

[Effect]

Since the electrocardiogram waveform is determined by the structure of the heart,
Generally has a chronological structure. In the present invention, this time-series structure is used in an electrocardiogram waveform recognition device. Recognition of the electrocardiogram waveform can be performed more accurately if the predicted classification is included, rather than relying only on the electrocardiogram waveform at the time point to be recognized. This recognition is achieved using a pattern associative neural network. Furthermore, prediction of the electrocardiogram classification at the time point to be recognized is achieved using a recurrent neural network. A recurrent neural network can learn the structure of a general time series, and has a function of predicting the classification of a point in time based on recognition results up to that point in time.

The pattern associative neural network used here includes, for example, "Nikkei Electronics" magazine issue 427 (1
(August 987), pages 115 to 124, ``Using neural networks for pattern recognition, signal processing, and knowledge processing.''
The pattern associative neural network described in detail in the article entitled (hereinafter referred to as cited document 1) can be used.

FIG. 7 is a structural explanatory diagram showing an example of the structure of a pattern associative neural network. As shown in Figure 7,
This pattern associative neural network has a hierarchical structure including an input layer 71, an intermediate layer 72, and an output layer 73. Note that although the intermediate layer 72 is one layer in FIG. 7, it may be multiple layers of two or more layers.

The output of a node in each layer of a pattern associative neural network is assigned a weight W to the node connected to that node.
The value of the sum of the products multiplied by is converted by a nonlinear function. In this way, the conversion characteristics of the pattern associative neural network are determined by the weights W. The value of weight W is determined by learning. The learning method can be executed using, for example, backward propagation, which is explained in detail in Cited Document 1.

The recurrent neural network used here includes, for example, Servan-Schreiber D. e
t. al. , “Encoding Sequentla
l Structure in Simple Rec
current NetWOrkS″Technical
Report CMU-OS-88-183, Car
neg1e Mellon Un1versity,1
988 (hereinafter referred to as cited document 2), it is explained in detail, and a current neural network can be used.

FIG. 8 is a structural explanatory diagram showing an example of the structure of a current fist neural network. As shown in FIG. 8, this recurrent neural network has an input layer 81.
, a context layer 82, an intermediate layer 83, and an output layer 84 have a hierarchical structure. Note that although the intermediate layer 83 is one layer in this figure, it may be multiple layers of two or more layers.

The output of a node in each layer of the recurrent Φ neural network is obtained by converting the sum of the nodes connected to that node by a weight W using a nonlinear function. In this way, the conversion characteristics of the recurrent neural network are determined by the weights W. Further, the value of the weight W is determined by learning. The learning method can be implemented using, for example, backward Φ propagation, which is explained in detail in Cited Document 1.

〔Example〕

Next, an embodiment of the first invention of the present application will be described with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the electrocardiogram waveform recognition device of the first invention.

The standard waveform ● division pattern memory section 11 shown in FIG.
It holds N sets of standard electrocardiogram waveforms and their divisions. Therefore, the learning recognition control unit 15 extracts the value St of the electrocardiogram waveform at the current time point t of the electrocardiogram waveform and its division Kt from the standard waveform division pattern memory unit 11, and calculates the current φ by the pattern associative neural network. Used for learning internal parameters with neural networks.

FIG. 2 is a block diagram showing an example of the signal flow in the learning phase of the first invention. The learning phase of the first invention will be explained below using FIG. 1 and FIG. 2.

The pattern associative neural network 12 divides the electrocardiogram waveform value St at time point t into a division Kt-1 of time point 1-1.
is taken as an input signal, and the classification Kt of time point t is taken as a teacher signal. The internal parameters of the pattern associative neural network 12 are updated based on these input signals and the teacher signal.

This update can be performed using backward propagation, which is explained in detail in Reference 1. The above-described learning operation is performed for the electrocardiogram waveform values St and divisions Kt held in the standard waveform ● division pattern memory section 11.

If the error after learning is not small enough, the learning operation described above is repeated until the error becomes small enough.

This completes the learning of the pattern associative neural network 12. Through this learning, the pattern associative neural network 12 can... It has a function of recognizing the division Kt of the time point t based on the value St of the electrocardiogram waveform of the time point t and the division Kt-■ of the time point t-i.

Further, the recurrent φ neural network 13 uses the division K t-t of the time point t-1 as an input signal, and uses the time t-1 as an input signal.
Let the classification K, be the teacher signal. Based on these input signals and teacher signals, internal parameters of the recurrent neural network 13 are updated. This update can be performed using backward propagation, which is explained in detail in Reference 1. The above-described learning operation is performed for the division Kt held in the standard waveform ● division parameter memory section 11. If the error after learning is not small enough, the learning operation described above is repeated until the error becomes small enough. This completes the learning of the recurrent neural network 13. Through this learning, the recurrent neural network 13 has the time point t-1
It has a function of predicting the division Kt of time point t from the time series of division Kt-■.

FIG. 3 is a block diagram showing an example of the signal flow in the recognition phase of the first invention.

Next, the phase of recognizing the divisions of an electrocardiogram using the trained pattern associative neural network 12 and the trained glue current neural network 13 will be described with reference to FIGS. 1 and 3.

In recognizing the division at time point t, the value St of the electrocardiogram waveform to be recognized and the prediction Ft of the division at time point t in the division buffer 14 (this prediction will be described later) are input to the pattern associative neural network 12. With these inputs, pattern associative neural network 1
As the output of step 2, a recognition result Kt of the division of time point t is obtained.

This recognition result Kt is input to the recurrent neural network 13. This input allows recurrent ●
As the output of the neural network 13, the time point t+
The predicted F tit for a partition of 1 is obtained. This predicted value Ft
Put +1 in partitioned buffer 14? and is used to recognize the division at time point t+i. The partition buffer 14, which holds the prediction of the partition for the next point in time, is initialized prior to recognition.

Next, an embodiment of the second invention of the present application will be described with reference to the drawings.

FIG. 4 is a block diagram showing an embodiment of the electrocardiogram waveform recognition device of the second invention.

The standard waveform division pattern memory section 41 shown in FIG.
It holds N sets of standard electrocardiogram waveforms and their divisions. Here, the time point near the current time point t is t−e
■* t+62 (e■≧1, e2≧1).

Generally, it is possible to set e1≠eI2, but here, for the sake of simplicity, ex=82 (=e). The input waveform buffer 46 stores electrocardiogram values St-etSt-e+1+...Sty...St+ at time points near the current time point t.
. -1sSt+e is maintained.

The learning/recognition control unit 45 determines the current time point t of the electrocardiogram waveform.
The value of the electrocardiogram waveform near S t-a s S t-e+
1 +...St,...S,+. 1■tSt+. and its classification Kt? It is used for learning internal parameters of the pattern associative neural network 42 and the recurrent neural network 43.

FIG. 5 is a block diagram showing an example of the signal flow in the learning phase of the second invention. The learning phase of the second invention will be explained below with reference to FIGS. 4 and 5.

The pattern associative neural network 42 uses the electrocardiogram waveform value St- corresponding to the time point t-et t+e near the current time point t stored in the input waveform buffer 46. sst-a+1+...St+...Ste.
, -1+St+. and the section Kt-■ of the time point t-1 are taken as input signals, and the section Kt of the time point t is taken as the teacher signal. The internal parameters of the pattern associative bilateral network 42 are updated based on these input signals and the teacher signal. This update can be performed using the backward bropage formula described in detail in Reference 1. The above-described learning operation is performed using the standard waveform ● the value St- of the electrocardiogram waveform corresponding to the time point t-13*t+e in the vicinity of the time point t stored in the division Bavan memory section 41. s S t-ee1,・”S t+”
" S t+e-1 + S the and the classification Kt. If the error after learning is not small enough,
The learning operation described above is repeated until the error becomes sufficiently small. With the above steps, the learning of the pattern associative neural network 42 is completed.

Through this learning, the pattern associative neural network 42 is able to calculate the time point i emt+e near the time point t.
The value of the electrocardiogram waveform corresponding to S t-atS t-e+1
,・”S t +”” S toe−1 + S
From the and the division Kt-1 of the time point t-i, the time point t
It has a function of recognizing the classification Kt of .

The learning of the recurrent neural network 43 uses the section Kt-1 of the time point t-1 as an input signal and the section Kt of the time point t as a teacher signal. Based on the teacher signals of these input signals, the internal parameters of the recurrent competitive network 43 are updated. This update can be performed using the backward propagation mechanism, which is explained in detail in Reference 1. The learning operation described above is performed using the standard waveform●division pattern memory section 41
This is performed for the classification Kt held in . If the error after learning is not small enough, the learning operation described above is repeated until the error becomes small enough. This completes the learning of the recurrent neural network 43. Through this learning, the recurrent neural network 43
A function of predicting the classification Kt of time point t from the time series of t-1 is performed.

FIG. 6 is a block diagram showing an example of the signal flow in the recognition phase of the second invention.

Next, the phase of recognizing electrocardiogram segments using the trained pattern associative neural network 42 and the trained glue current neural network 43 will be described with reference to FIGS. 4 and 6.

Upon recognition of the segment at time point t, the input waveform is temporarily stored in the input waveform buffer 46.

In recognizing the division at time point t, time points t-e, which are taken out from the input waveform buffer 46 and near time point t,
Value of electrocardiogram waveform corresponding to t+e S t-eys t-
ee1,・"S t+"' S tea-Its t
he and the prediction Ft(
This prediction will be described later) is input to the pattern associative neural network 42. By this,
In the pattern associative neural network 42, the recognition result Kt of the classification of time point t is obtained as an output. This recognition result Kt is recurrent ●Two s-ral network 4
3 input. As a result, the recurrent neural network 43 obtains a prediction Ft+1 for the section at time point t+1. This prediction Ft+1 is stored in the classification buffer 44 and used for recognition of the classification at time point t+1. The partition buffer 44, which holds the prediction of the next time point partition, is initialized prior to recognition.

〔Effect of the invention〕

The electrocardiogram waveform recognition device of the present invention recognizes segmentation points by paying attention to the temporal structure of the electrocardiogram waveform, thereby improving both the precision and accuracy of segmentation point recognition compared to conventional electrocardiogram waveform recognition devices. It has the effect that it can be improved.

[Brief explanation of drawings]

Fig. 1 is a block diagram showing an embodiment of the electrocardiogram waveform recognition device of the first invention, Fig. 2 is a block diagram showing the signal flow in the learning phase of the first invention, and Fig. 3 is a block diagram showing an embodiment of the electrocardiogram waveform recognition device of the first invention. FIG. 4 is a block diagram showing an example of the signal flow in the recognition phase of the second invention. FIG. 5 is a block diagram showing an example of the electrocardiogram waveform recognition device of the second invention. FIG. FIG. 6 is a block diagram showing an example of the signal flow in the recognition phase of the second invention. FIG. 7 is a structural explanatory diagram showing an example of the structure of a pattern associative neural network. FIG. Current: A structural explanatory diagram showing an example of the structure of a neural network. FIG. 9 is a waveform diagram showing an example of a general electrocardiogram waveform and its division. 11...Standard waveform●Division pattern memory section, 12...
・Pattern associative neural network, 13...
Recurrent neural network, 14... Segmented buffer, 15... Learning order recognition control unit, 41... Standard waveform ● Segmented pattern memory unit, 42... Pattern associative neural network, 43... Recurrent neural network
Neural network, 44...partition buffer, 4
5...Learning recognition control unit, 46...Input waveform buffer, 71...Input layer, 72...Middle layer, 73...
Output layer, 81... Input layer, 82... Context layer, 83... Middle layer, 84... Output layer.

Claims (1)

[Scope of Claims] 1. A standard waveform/segmentation pattern memory unit that holds a standard waveform pattern that is a standard electrocardiogram waveform pattern and its waveform divisions, and a standard waveform/segmentation pattern memory unit that stores the electrocardiogram waveform pattern and the predicted segmentation one time before. A pattern associative neural network that recognizes the waveform segment at the current time from , a recurrent neural network that predicts the segment at that time from the time series of segment patterns up to one time before, and a segment buffer that holds the predicted segment. , a learning/recognition control unit that controls learning and recognition operations of the pattern associative neural network and the recurrent neural network. 2. Electrocardiogram waveform recognition according to claim 1, characterized in that during learning, the input of the pattern associative neural network according to claim 1 includes an input waveform buffer that holds a standard waveform pattern at a time in the vicinity of the current time. Device.