JP2730910B2 - Waveform analysis method using ambiguous automata - Google Patents

Description

The present invention relates to a method of analyzing a waveform to be analyzed, using a so-called ambiguous automaton, to determine whether or not a characteristic of a predetermined reference waveform pattern exists in the waveform to be analyzed. To analyze a waveform to be analyzed.

2. Description of the Related Art Conventional techniques and problems such as electrocardiographic and pulse waveforms analyzed for diagnosing cardiovascular diseases, voice waveforms analyzed for voice recognition, and signal waveforms identified during telecommunications. There is a so-called pattern matching method for performing analysis based on whether or not a predetermined reference waveform pattern exists in a waveform to be analyzed. DP (dynamic programming) matching is known as one type of such a pattern matching method. Such DP matching is
For example, a technique widely used in the field of voice recognition, which detects a matching between a reference waveform pattern and an analyzed waveform while relatively moving the analyzed waveform on the time axis. There is a feature that can absorb expansion and contraction. However, according to this DP matching,
If an error is included in the analyzed waveform, matching becomes difficult, and there is a disadvantage that the reason for the obtained result or an element that can clearly explain the causal relationship is lacking.

As another type of pattern matching method, a hidden Markov process is known. This method attempts to analyze the waveform by expressing the analyzed waveform with a stochastic model, so it can give meaning to the state and give some reason to the obtained result. However, it is inevitable that the reason is somewhat ambiguous to give the state as a black box.

That is, when the DP matching or the Hidden Markov process is used for waveform analysis, reliability of the obtained result may not be reliably obtained.

The present inventors have conducted various studies on the background of the above circumstances, and as a result, by applying fuzzy processing to the analyzed waveform, to obtain vague fuzzy data including errors but guaranteeing a certain degree of certainty. The ambiguous fuzzy data is converted into a finite automaton having a predetermined finite number of states for detecting the characteristics of the reference waveform pattern and a condition range for transition from one state to another state defined by the fuzzy data expression. When inputting and determining the presence or absence of the feature of the analyzed waveform based on the certainty of the state at the input end point of the fuzzy data, it has been found that the data processing is simplified and a highly reliable waveform analysis result is easily obtained. . The present invention has been made based on such findings.

Means for Solving the Problems That is, the gist of the present invention is a method of analyzing a part or all of a waveform to be analyzed based on whether or not a feature of a predetermined reference waveform pattern exists. (A) a fuzzy converting step of generating fuzzy data by classifying the analyzed waveform into predetermined plural degrees of gradient for each predetermined small section;
(B) a plurality of predetermined states for detecting the characteristics of the reference waveform pattern, and a range of transition conditions from one state to another state is determined based on the gradients at the plurality of stages; Using an ambiguous automaton
A processing step of sequentially processing the fuzzy data;
The processing in the processing step determines whether or not the feature of the reference waveform pattern exists in the analyzed waveform based on the certainty of the state of the fuzzy automaton at the end of the input of the fuzzy data. And a process.

Operation In this way, in the fuzzification process, the analyzed waveform including the error is classified into a plurality of predetermined gradients for each predetermined small section, so that a certain degree of certainty is guaranteed. After being converted to ambiguous fuzzy data, in a processing step, the fuzzy data has a plurality of predetermined states for detecting the characteristics of the reference waveform pattern and changes from one state to another state. The range of the transition condition is processed using the vague automaton defined based on the gradients of the plurality of stages, and in the determination step, based on the certainty of the state of the vague automaton at the end of the input of the fuzzy data. It is determined whether or not the feature of the reference waveform pattern exists in the analyzed waveform.

Effect of the Invention As described above, according to the present invention, the analyzed waveform is converted into ambiguous fuzzy data in which a certain degree of certainty is assured at a plurality of predetermined gradients for each predetermined small section. After that, the plurality of states and the transition conditions between the states are analyzed using the definite vague automaton, so that the data processing becomes relatively simple and a highly reliable waveform analysis result is obtained.

Embodiment Hereinafter, an application example of the present invention will be described in detail with reference to the drawings.

In FIG. 1, a pulse wave sensor 10 is mounted on an ecological artery such as a radial artery, a dorsal artery, and a carotid artery, and is pressed by an artery at an optimal pressure for efficiently detecting a pulse wave. And a pulse wave signal SM representing a pulse wave generated from the artery. The magnitude of the pulse wave signal SM corresponds to the blood pressure value in the artery. The pulse wave signal SM is
The data is sequentially input to the CPU 14 through the A / D converter 12. CPU14
Configures a microcomputer together with the ROM 16 and the RAM 18, processes input signals in accordance with a program stored in the ROM 16 in advance while using the storage function of the RAM 18, and outputs a display signal to the display device 22 via the interface 20. In the case of this device, the physiological age is determined from the pulse wave signal SM according to the above program.

The operation will be described below with reference to the flowchart of FIG.

First, by performing steps S1 and S2,
Among the pulse wave signals SM stored in the RAM 18, a pulse wave signal SM corresponding to, for example, one cycle is read out, for example, for 25 ms.
The average value of each interval is calculated. Pulse wave represented by a series of average data bp _t indicating the average value are those shown in FIG. 3, the sampling period of the A / D converter 12 (1 ms)
The pulse wave signal shown in FIG. 4 at the point corresponding to
There is almost no difference compared to the SM waveform (analyzed waveform). The data compression processing in step S2 enables a large amount of waveforms to be handled with a small amount of data, thereby reducing the time consumed for the waveform diagnosis processing and the storage area occupied in the ROM 16 or the RAM 18. 25 in the above case
Since the average value is obtained for each section at ms intervals, the average value is reduced to 1/25.

In step S3, a series of average data bp _t determined at step S1, a series of difference data bpd _t is determined according to the following equation (1), to represent the object to be analyzed pulse wave by the difference data bpd _t . Here, bp _t is the t-th average data, bpd _t is the t-th of the difference data. Also, delta indicates the time interval of the section average data bp _t is calculated. Reason for the analyzed pulse wave is represented by the difference data bpd _t is the degree of change, that is, better shown in inclination, it can represent a trend of the pulse wave more accurately.

By attaching classifying the degree of inclination of the series of multiple stages differential data bpd _t (gradient) in _{bpd t = (bp t -bp t} -1) / Δ ··· (1) Step S4,
Create fuzzy data. That is, the difference data bpd _t
Is not absolutely correct because it contains errors of measuring instruments, errors at the time of measurement, errors from the measurement target, etc., but it can be handled as correct data containing some ambiguity, in this embodiment, the value of the differential data bpd _t indicating the slope of the analyzed waveform for each small section, for example a rapid increase as shown in table 1, considerable increase, increase, no change, decreased, considerable Decrease, sharp decrease 7
It is converted into a fuzzy value corresponding to the stage.

The analyzed waveform input by the pulse wave signal SM is represented by the fuzzy data as described above, so that it is represented by an expression close to a word commonly used. For example, the pulse wave of the childhood shown in FIG.
It may be expressed that there is only a small portion of no change, then a significant decrease followed by a significant increase, and then after such a change again, continues to decrease significantly. Since the pulse wave signal SM contains an error, the above-mentioned fuzzy value simply indicates that the probability of belonging to the degree of the change is high. For example, in the case of an increase, the probability of an increase state is the highest, but there is also a probability of a considerable increase or no change. This probability is called the membership value (= confidence),
A relationship between the fuzzy value and the membership value is called a membership function (= fuzzy function). Sixth
The figure exemplifies a membership function for a fuzzy value of “0”, and other fuzzy values have certainty factors indicated by similar membership functions.

In step S5, fuzzy data is sequentially input to an ambiguous automaton which is an automatic judgment algorithm stored in advance. This ambiguous automaton is devised in order to accurately represent a part or all of the features of the reference waveform pattern, since the direct comparison between the analyzed pulse wave converted into the fuzzy data and the reference waveform pattern is complicated. Since it has a finite state for indicating the characteristics of the reference waveform pattern, it is also called a finite state automaton. Each state of this ambiguous automaton indicates a certain characteristic of the pulse wave or reference pulse wave pattern. Seventh
FIG. 8, FIG. 8 and FIG. 9 are examples of the vague automaton stored in the ROM 16 in advance. The automaton 1 for examining the number of mountains after the maximum peak (systolic blood pressure) of the pulse wave, the maximum of the pulse wave An automaton 2 for examining the number of flat portions of the peak number, and an automaton 3 for examining that one flat portion and one flat or rising portion exist in order while continuing to fall after the maximum peak of the pulse wave. Is shown.

Here, each state S _i of the fuzzy automata are those determined to obscure in response to the input of the fuzzy data. That is, each state S _i has a certainty factor indicating the degree of existence in that state, and when a certain fuzzy data input is given, the range of the transition condition to which state transitions is made. Is predetermined. A state where the certainty factor is not zero is said to be activated, and the input of fuzzy data can be accepted. When transitioning from one state S _i to another state S _{i + 1} by the input a, the fuzzy function indicating the certainty of the fuzzy value indicated in the transition condition range and the fuzzy function indicating the certainty of the input a Of the smaller value is determined (MIN-MAX operation)
Then, the maximum value is set as the certainty factor of the state S _{i + 1} . Further, when a transition is made from a plurality of states to the same state, the maximum certainty of the transition certainty is adopted as the certainty of the transition destination state.

For example, in the automaton 1 of FIG. 7, only the state S0 is activated with the certainty factor 1 in the expected state. At this time, if the fuzzy value “−1” is input, the state S0 is maintained at the certainty factor 1.0 because the fuzzy value “−1” is included in the maintenance state of the state S0, and the fuzzy value “−1” Is included in the transition condition range -3 to -1 from the state S0 to the state S1, the transition to the state S2 is performed, and the certainty factor of the state S1 is set to 1.0. However, when the fuzzy value “0” is input, the state S0 is included in the condition for maintaining the state S0, so that the state S0 has a certainty factor of 1.0.
And the transition to the state S1 is performed even though the fuzzy value “0” is not included in the transition condition range -3 to −1 from the state S0 to the state S1. The degree is assumed to be 0.5. That is, the fuzzy value “0” itself is ambiguous and its certainty is as shown by the fuzzy function F ₀ in FIG. 6, and the transition condition range itself is also ambiguous and its certainty is As shown in the functions F _-1 to F _-3 , as shown in FIG. 11, the result of the MIN-MAX operation is 0.5, and the certainty factor of the state S1 is 0.5.

In this way, when the fuzzy values for one pulse wave are sequentially input to the automaton, the certainty degree of the specific state in each automaton is matched to the pattern matching that seems to be the most similar among the pattern matchings that reach it. You can think of it as giving a degree. That is, when the certainty factor of the state S5 of the automaton 1 is 1.0,
This indicates that two peaks exist after the maximum peak in the pulse wave corresponding to the input fuzzy data. When the certainty factor of the state S5 of the automaton 2 is 1, it indicates that two flat portions exist after the maximum peak in the pulse wave shape corresponding to the input fuzzy data.
When the certainty factor of the state S5 of the automaton 3 is 1.0, one flat portion and one flat or rising portion are present in order after the maximum peak in the pulse wave shape corresponding to the input fuzzy data. It is shown that.

In step S6, the result of inputting the fuzzy data to the fuzzy automaton is evaluated according to a previously determined rule, and what characteristics are present in the pulse wave is determined.In step S7, the evaluation is performed. The result is output on the display 22. The above rule is
For example, those shown in the following (a) and (b).

(A) If the reliability of state S5 of automaton 1 is 1.0 and the reliability of state S5 of automata 2 and 3 is not 1.0,
The pulse wave pattern is that of a child.

(B) The reliability of state S5 of automata 1 and 3 is 1.0,
If the reliability of automaton 2 is not 1.0, the pulse wave pattern is adolescent.

That is, the judgment formula corresponding to the above rule is
This determination formula is executed in step S6, and the actual physiological age of the pulse wave is displayed in step S7. Indicator of Fig. 1
Reference numeral 22 indicates this state.

FIG. 12 uses the automata 1, 2, 3 and the rules (a) and (b) and the automaton and the rule (not shown) for determining how flat the vicinity of the highest peak of the pulse wave is. The results of the experiments performed are shown.

As described above, one application example of the present invention has been described, but the present invention is also applied to other aspects.

For example, in the above application example, data compression is performed by calculating average data every 25 ms,
Step S2 for executing data compression may not necessarily be provided as long as some operation time and required storage capacity can be tolerated.

Further, in (1) of the aforementioned application example, although the difference data bpd _t by the difference between the average data (bp _t -bp _t-1) is divided by the time interval Δ is calculated, it is known period since the average data bp _t is the mean data from the fact that the calculated difference (bp _t -bp _t-1) itself at delta represents a slope, it may not be divided necessarily by the time interval delta.

In addition, in the above-described application example, the pulse wave signal SM is averaged for each small section every 25 ms. However, the pulse wave signal SM may not be necessarily every 24 ms, and an important section for extracting the characteristics of the pulse wave. May be averaged in a section finer than the other sections.

Further, in the above-described application example, after the pulse wave signal SM is temporarily stored in the RAM 18, a portion corresponding to one pulse wave is read out and compressed, but the pulse wave signal SM is transmitted through the A / D converter 12. S
An average value may be obtained in real time while M is being input, and the average value may be stored in the RAM 18.

Further, in the above-described application example, a portion corresponding to one pulse wave of the pulse wave signal SM is analyzed. However, when a location where a feature appears is specified, only that portion may be analyzed.

Further, in the above-described application example, a plurality of vague automata are used to obtain the physiological age, but one automaton may be used depending on the characteristics of the waveform.

Further, in the application example described above, a function having an isosceles triangle shape is used as the fuzzy function, but a curve such as a normal distribution may be used.

Further, in the above-described application example, as shown in FIG. 1, the pulse wave signal SM from the pulse wave sensor 10 was directly input to the waveform analyzer. The pulse wave signal SM is recorded once in a storage device such as a recorder, and is stored in a main body arranged in a place different from the pulse wave sensor 10 from the storage device.
May be input and the waveform may be analyzed.

In the above-described application example, the analysis of the pulse wave collected from the living body has been described. However, the analysis of other waveforms such as the analysis of an electrocardiographic waveform for diagnosis and the analysis of a voice waveform for voice recognition has been described. It goes without saying that it may be used for In this case, automata and rules for extracting the features of each reference pattern are prepared.

The above is merely an application example of the present invention, and the present invention can be variously modified without departing from the spirit thereof.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing an example of a pulse wave analyzer to which the present invention is applied. FIG. 2 is a flowchart showing the operation of the apparatus shown in FIG. FIG. 3 is a diagram showing a pulse wave represented by the average data. FIG. 4 is a diagram showing a pulse wave represented by a pulse wave signal. FIG. 5 is a diagram showing an example of a pulse wave shape in a childhood. FIG. 6 is a diagram illustrating an example of a fuzzy function when the fuzzy value is “0”. 7, 8, and 9 are diagrams showing examples of the vague automaton used in the apparatus of FIG. 1, respectively. FIG. 10 is a view showing a fuzzy function of a fuzzy value indicating a transition condition for transition from the state S0 to the state S1 in the automaton 1 in FIG. FIG. 11 is a diagram showing a certainty factor determined when a transition is made from state S0 to S1 when a fuzzy value “0” is input to automaton 1 in FIG. FIG. 12 is a chart showing experimental results of pulse wave analysis performed in the apparatus of FIG.

──────────────────────────────────────────────────続 き Continuation of the front page (51) Int.Cl. ⁶ Identification number Office reference number FI Technical display location G10L 3/00 551 A61B 5/02 310Z 9/18 301 5/04 312C

Claims (1)

(57) [Claims] 1. A method for analyzing a part or all of a waveform to be analyzed based on whether or not a characteristic of a predetermined reference waveform pattern exists, wherein the waveform to be analyzed is determined for each predetermined small section. A fuzzification process of creating fuzzy data by classifying the data into a plurality of predetermined gradients; and a plurality of predetermined states for detecting a feature of the reference waveform pattern; and A step of inputting the fuzzy data to an ambiguous automaton in which a range of a transition condition from one state to another state is determined based on the gradients of the plurality of stages, and sequentially processing the fuzzy data; By the process according to the process, based on the certainty of the state of the fuzzy automaton at the end of the input of the fuzzy data, the reference in the analyzed waveform Waveform analysis method using the fuzzy automaton, characterized in that it comprises a and a determination step of determining whether exists wherein the shape pattern.