JP2021040969A - Method for generating determination algorithm, determination algorithm, determination system, determination method, program, and recording medium - Google Patents

Abstract

To generate determination algorithm capable of reducing a calculation load on blood pressure surge determination.SOLUTION: A method for generating determination algorithm includes: classifying a plurality of feature amounts capable of characterizing a waveform of a blood pressure surge to define a feature amount category group, as a primary candidate, including at least a feature amount category on amplitude, a feature amount category on rising, and a feature amount category on falling (S11); for each feature amount category, creating a plurality of sets of feature amounts and thresholds (S12); taking a plurality of blood pressure variations protruding upward included in teacher data Dte as targets, using, one by one as a classifier, all sets of feature amounts and thresholds in the feature amount category group as the primary candidate to determine whether or not the blood pressure variations are classified into a blood pressure surge (S13); calculating an evaluation index for each determination result as a classifier for each set (S14); and determining a set of a feature amount and a threshold providing the best value of the calculated evaluation indices, as a set of a first feature amount and a first threshold to be an element of determination algorithm (S15).SELECTED DRAWING: Figure 7A

Description

The present invention relates to a method for generating a determination algorithm, and more particularly to a method for generating a determination algorithm for determining whether or not the Yamagata blood pressure fluctuation included in the time series data of blood pressure is a blood pressure surge. The present invention also relates to such a determination algorithm. The present invention also relates to a determination system and a determination method for determining whether or not the Yamagata blood pressure fluctuation included in the time series data of blood pressure is a blood pressure surge by using such a determination algorithm. The present invention also relates to a program that causes a computer to execute such a determination method. The present invention also relates to a recording medium on which such a program is recorded.

In patients suffering from sleep apnea syndrome (SAS), blood pressure rises over, for example, 5 to 15 seconds when resuming breathing after apnea, and then falls to its original level. It is known. In the present specification, such a chevron blood pressure fluctuation is referred to as a "blood pressure surge" (or simply "surge"). From a study of 24-hour ambulatory blood pressure monitoring (ABPM) (several hourly blood pressure measurements (this is called "spot measurement")), nighttime blood pressure fluctuations Is known to be associated with the risk of cardiovascular events (angina, myocardial infarction, ischemic heart failure, etc.). For this reason, nocturnal blood pressure surges are considered to be clinically important. Therefore, it is also important to determine whether or not the Yamagata blood pressure fluctuation included in the blood pressure time series data is a blood pressure surge (this is appropriately referred to as "blood pressure surge determination"). In addition, in this specification, "blood pressure" is referred to as "blood pressure measured from the pressure pulse wave for each heartbeat (hereinafter," blood pressure for each beat "(Beat By Beat BP)", or simply "blood pressure". .) Means.

Conventionally, for example, in Patent Document 1 (Japanese Unexamined Patent Publication No. 2018-153232), FIG. 9 (an example of a blood pressure surge waveform is shown by a curve C) as a feature amount representing a blood pressure surge waveform for determining a blood pressure surge. As shown in the inside, the difference between the blood pressure value at the peak point P2 of the blood pressure surge and the blood pressure value at the start point P1 of the blood pressure surge (fluctuation amount during rise) L1 and the blood pressure value at the peak point P2 of the blood pressure surge. The difference between the blood pressure value at the end point P4 of the blood pressure surge (fluctuation amount during descent) L3 and the time difference between the time at the peak point P2 of the blood pressure surge and the time at the start point P1 of the blood pressure surge (rise time) T1 , The difference (descending time) T3 between the time at the peak point P2 of the blood pressure surge and the time at the end point P4 of the blood pressure surge is listed. In addition to the above four features, some other features are listed in the same document. Then, using these features, it is determined whether or not the Yamagata blood pressure fluctuation included in the blood pressure time series data is a blood pressure surge.

Japanese Unexamined Patent Publication No. 2018-153232

Overnight blood pressure time series data include beat-corresponding peaks (corresponding to systolic blood pressure (SBP) or diastolic blood pressure (DBP)) for approximately 30,000 beats. It is said that hundreds of blood pressure surges can occur overnight.

Here, in Patent Document 1 (Japanese Unexamined Patent Publication No. 2018-153232), since a large number of features are used to determine whether or not there is a blood pressure surge, each feature for surge determination has a large calculation load. There is a problem that the usefulness of the quantity is not known.

Therefore, an object of the present invention is a determination algorithm for determining whether or not the Yamagata blood pressure fluctuation included in the time-series blood pressure data is a blood pressure surge, which can reduce the calculation load of the blood pressure surge determination. Is to provide a method of generating a determination algorithm capable of generating. The present invention also provides such a determination algorithm. The present invention also provides a determination system and a determination method for performing blood pressure surge determination using such a determination algorithm. The present invention also provides a program that causes a computer to execute such a determination method. The present invention also provides a recording medium on which such a program is recorded.

In order to solve the above problem, the method of generating the judgment algorithm of this disclosure is
It is a method of generating a judgment algorithm for judging whether or not the Yamagata blood pressure fluctuation included in the time series data of blood pressure is a blood pressure surge.
A plurality of features capable of characterizing the waveform of the blood pressure surge are classified, and the feature amount of the primary candidate including at least the feature amount category related to amplitude, the feature amount category related to ascending, and the feature amount category related to descending. Define categories
For each of the feature amount categories, a set of the feature amount and the threshold value is formed by combining a plurality of feature amounts included in the feature amount category and a threshold value of one or more steps for each of the plurality of feature amount amounts. Create multiple sets and
Multiple Yamagata blood pressure fluctuations derived from blood pressure time-series data, for which it is known in advance whether or not each is a blood pressure surge, are prepared as teacher data.
Targeting the blood pressure fluctuations of the plurality of Yamagata included in the teacher data, the feature amounts and thresholds of all the sets in the feature amount category group of the primary candidates are used as a classifier one by one to generate a blood pressure surge. Judge whether it is classified or not,
Regarding the judgment result as a classifier for each of the above sets, an evaluation index was obtained for each set of the above-mentioned primary candidates in the feature amount category group.
Among the evaluation indexes obtained over all the sets in the feature amount category group of the primary candidate, the set of the feature amount and the threshold value giving the best value is the first feature amount which is an element of the determination algorithm. It is characterized in that it is determined as a set of the first threshold value and the first threshold value.

In the present specification, the "primary candidate" and the "secondary candidate" and "tertiary candidate" described later represent the order in which the feature amounts are sequentially selected. Further, the "primary candidate feature amount category group" includes at least three feature amount categories of a feature amount category related to amplitude, a feature amount category related to ascending, and a feature amount category related to descending, and four or more features. It may include a quantity category.

Further, the "Yamagata blood pressure fluctuation" is expressed as a chevron waveform indicated by a envelope connecting the peaks of the blood pressure (systolic blood pressure (SBP) or diastolic blood pressure (DBP)) for each beat. This "Yamagata blood pressure fluctuation" refers to each mountain as a whole, and does not refer to only one peak point (apex).

The "plurality of chevron blood pressure fluctuations" contained in the teacher data is typically assumed to be a significant number of chevron blood pressure fluctuations sufficient for statistical processing. Regarding the fluctuation of blood pressure in Yamagata, "it is known whether or not it is a blood pressure surge" means, for example, that a doctor or the like (including a person who has substantially the same level of ability as a doctor in determining a blood pressure surge) in Yamagata. It means that it has been determined whether or not it is a blood pressure surge by looking at the waveform of blood pressure fluctuation (time change waveform of blood pressure).

"Use the feature amount and threshold value" and "use as a classifier" for a plurality of Yamagata blood pressure fluctuations is determined by comparing the value of the feature amount indicated by the blood pressure fluctuation of each Yamagata with the above threshold value. Means to do.

The “evaluation index” refers to, for example, a recall rate (Recall), a precision rate (Precision), an F value (F-measure), or the like.

In the method of generating the determination algorithm of this disclosure, first, a plurality of features capable of characterizing the waveform of the blood pressure surge are classified, and the feature category related to the amplitude, the feature category related to the rise, and the feature related to the fall are classified. A feature quantity category group of primary candidates including at least a quantity category is defined. For each of the feature amount categories, a set of the feature amount and the threshold value is formed by combining a plurality of feature amounts included in the feature amount category and a threshold value of one or more steps for each of the plurality of feature amount amounts. Create multiple sets. As a result, multiple sets of features and thresholds in the feature amount category related to amplitude, multiple sets of feature amounts and thresholds in the feature amount category related to ascending, and multiple sets of feature amounts and thresholds in the feature amount category related to descending , At least created. In addition, multiple Yamagata blood pressure fluctuations derived from time-series data of blood pressure (typically, a considerable number sufficient for statistical processing), for which it is known in advance whether or not each is a blood pressure surge. , Prepare as teacher data.

Next, in this method, for the blood pressure fluctuations of the plurality of Yamagata included in the teacher data, the feature amounts and the thresholds of all the sets in the feature amount category group of the primary candidate are used as a classifier one set at a time. It is used to determine whether or not each is classified as a blood pressure surge. As a result, a plurality of sets of classifiers (features and features) in each feature category included in the feature category group of the primary candidate for the blood pressure fluctuations of the plurality of chevrons included in the teacher data. Judgment results for each of the thresholds) are obtained.

Next, in this method, an evaluation index is obtained for each of the above-mentioned determination results as a set of classifiers for all the sets in the above-mentioned primary candidate feature amount category group. Then, among the evaluation indexes obtained over all the sets in the feature amount category group of the primary candidate, the set of the feature amount and the threshold value giving the best value is the first element of the determination algorithm. It is determined as a set of the feature amount and the first threshold value. As a result, the best value of the evaluation index among all the sets (feature amount and threshold value) including the feature amount and the threshold value of a plurality of sets in each feature amount category included in the feature amount category group of the primary candidate. Is determined as a set of a first feature amount and a first threshold value.

In this way, as an element of the determination algorithm for determining the blood pressure surge, a determination algorithm can be generated that includes only the pair of the first feature amount and the first threshold value as useful. According to the generated determination algorithm, for example, by comparing the value of the first feature amount indicated by the blood pressure fluctuation of the Yamagata to be determined with the first threshold value, the blood pressure fluctuation of the Yamagata to be determined is compared. Can be determined if is a blood pressure surge.

In the method of generating the determination algorithm of one embodiment,
Among the plurality of Yamagata blood pressure fluctuations included in the teacher data, the data on the Yamagata blood pressure fluctuation classified as a blood pressure surge according to the combination of the first feature amount and the first threshold value is intermediate data. Specified as
From the feature amount category group of the primary candidate, the feature amount category given the first feature amount is excluded, and all the remaining feature amount categories are specified as the feature amount category group of the secondary candidate.
Targeting the blood pressure fluctuations of a plurality of chevrons included in the intermediate data, the feature amounts and threshold values of all the sets in the feature amount category group of the secondary candidates are classified into blood pressure surges one by one as a classifier. Judge whether or not it will be done,
Regarding the judgment result as a classifier for each of the above sets, an evaluation index was obtained for each of the above sets of the feature amount category group of the secondary candidates.
Among the evaluation indexes obtained over all the sets in the feature amount category group of the secondary candidate, the set of the feature amount and the threshold value giving the best value is the second feature amount which is an element of the determination algorithm. It is characterized in that it is determined as a set of a second threshold value and a second threshold value.

In the method of generating the determination algorithm of this one embodiment, the blood pressure surge is caused by the combination of the first feature amount and the first threshold value among the plurality of chevron blood pressure fluctuations included in the teacher data. The data on the classified Yamagata blood pressure fluctuations are specified as intermediate data. If the above teacher data includes a considerable number of Yamagata blood pressure fluctuations, this intermediate data may still contain a considerable number of Yamagata blood pressure fluctuations sufficient for statistical processing. Further, the feature amount category giving the first feature amount is excluded from the feature amount category group of the primary candidate, and all the remaining feature amount categories are specified as the feature amount category group of the secondary candidate.

Next, in this method, for a plurality of Yamagata blood pressure fluctuations included in the intermediate data, all the features and thresholds in the feature category group of the secondary candidates are used as a classifier one by one. It is determined whether or not each of them is classified as a blood pressure surge. As a result, a plurality of sets of classifiers (features and thresholds) in each feature category included in the feature category group of the secondary candidate for the blood pressure fluctuations of the plurality of chevrons included in the intermediate data. The judgment result by each of) is obtained.

Next, in this method, the evaluation index is obtained for each of the above-mentioned judgment results as a set of classifiers for all the sets in the above-mentioned secondary candidate feature amount category group. Then, among the evaluation indexes obtained over all the sets in the feature amount category group of the secondary candidate, the set of the feature amount and the threshold value giving the best value is the second element of the determination algorithm. It is determined as a set of the feature amount and the second threshold value. As a result, the best value of the evaluation index among all the sets (feature amount and threshold value) including the feature amount and the threshold value of a plurality of sets in each feature amount category included in the feature amount category group of the secondary candidate. Is determined as a set of a second feature amount and a second threshold value.

In this way, as elements of the determination algorithm for determining the blood pressure surge, a set of the first feature amount and the first threshold value, and a set of the second feature amount and the second threshold value. A determination algorithm can be generated that includes only as useful. According to the generated determination algorithm, for example, the value of the first feature amount indicated by the blood pressure fluctuation of the Yamagata to be determined is compared with the first threshold value, and the blood pressure fluctuation of the Yamagata to be determined is compared. By comparing the value of the second feature amount indicated by the above with the second threshold value, it is possible to determine whether or not the blood pressure fluctuation of the Yamagata to be determined is a blood pressure surge by the AND calculation.

In the method of generating the determination algorithm of one embodiment,
Of the plurality of Yamagata blood pressure fluctuations included in the intermediate data, the data on the Yamagata blood pressure fluctuation classified as a blood pressure surge according to the combination of the second feature amount and the second threshold value is used as the final data. Identify and
From the feature amount category group of the secondary candidate, the feature amount category given the second feature amount is excluded, and all the remaining feature amount categories are specified as the feature amount category group of the tertiary candidate.
Targeting multiple Yamagata blood pressure fluctuations included in the final data, all sets of features and thresholds in the feature category group of the tertiary candidates are classified into blood pressure surges by using one set as a classifier. Judge whether or not it will be done,
Regarding the judgment result as the classifier for each of the above sets, the evaluation index was obtained for each set in the feature amount category group of the above tertiary candidates.
Among the evaluation indexes obtained over all the sets in the feature amount category group of the tertiary candidate, the set of the feature amount and the threshold value giving the best value is the third feature amount which is an element of the determination algorithm. It is characterized in that it is determined as a set of a third threshold value and the third threshold value.

In the method of generating the determination algorithm of this one embodiment, among the plurality of chevron blood pressure fluctuations included in the intermediate data, the blood pressure surge is classified according to the combination of the second feature amount and the second threshold value. The data on the blood pressure fluctuation in Yamagata will be specified as the final data. If the above teacher data includes a considerable number of Yamagata blood pressure fluctuations, this final data may still contain a considerable number of Yamagata blood pressure fluctuations sufficient for statistical processing. Further, the feature amount category to which the second feature amount is given is excluded from the feature amount category group of the secondary candidate, and all the remaining feature amount categories are specified as the feature amount category group of the tertiary candidate. The number of feature amount categories included in the feature amount category group of the tertiary candidate may be one.

Next, in this method, for a plurality of Yamagata blood pressure fluctuations included in the final data, all the features and thresholds in the feature category group of the tertiary candidate are used as a classifier one by one. It is determined whether or not each of them is classified as a blood pressure surge. As a result, a plurality of sets of classifiers (features and thresholds) in each feature category included in the feature category group of the tertiary candidate for the blood pressure fluctuations of the plurality of chevrons included in the final data. The judgment result by each of) is obtained.

Next, in this method, the evaluation index is obtained for each of the above-mentioned determination results as a classifier for each set of the above-mentioned tertiary candidates over all the sets in the feature amount category group. Then, among the evaluation indexes obtained over all the sets in the feature amount category group of the tertiary candidate, the set of the feature amount and the threshold value giving the best value is the third element of the determination algorithm. It is determined as a set of the feature amount and the third threshold value. As a result, the best value of the evaluation index among all the sets (feature amount and threshold value) including the feature amount and the threshold value of a plurality of sets in each feature amount category included in the feature amount category group of the tertiary candidate. Is determined as a set of a third feature amount and a third threshold value.

In this way, as elements of the determination algorithm for determining the blood pressure surge, the pair of the first feature amount and the first threshold value, the set of the second feature amount and the second threshold value, and , A determination algorithm may be generated that includes only the combination of the third feature amount and the third threshold value as useful. According to the generated determination algorithm, for example, the value of the first feature amount indicated by the blood pressure fluctuation of the Yamagata to be determined is compared with the first threshold value, and the blood pressure fluctuation of the Yamagata to be determined is compared. Compares the value of the second feature amount indicated by the above with the second threshold value, and the value of the third feature amount indicated by the blood pressure fluctuation of Yamagata to be determined and the third threshold value. By comparing the above, it is possible to determine whether or not the blood pressure fluctuation in Yamagata, which is the object of the determination, is a blood pressure surge by the AND calculation.

In the method of generating the determination algorithm of one embodiment,
When creating a plurality of sets of the feature amount and the threshold value for each of the feature amount categories, the threshold values to be combined with the feature amount are set in a plurality of steps in a certain range over a certain range.
After determining the above-mentioned sets that form the elements of the above-mentioned determination algorithm, a search range narrower than the above-mentioned range including the threshold value is set for the threshold values forming the above-determined sets, and the search range is finer than the above-mentioned step. It is characterized in that each of the above thresholds is adjusted by performing a grid search in increments.

In the method of generating the determination algorithm of this one embodiment, since each of the above threshold values is finely adjusted, the accuracy of blood pressure surge determination by the generated determination algorithm can be improved.

In another aspect, the determination algorithm of this disclosure
It is a judgment algorithm using the first feature amount determined by the method of generating the above judgment algorithm.
By comparing the value of the first feature amount indicated by the blood pressure fluctuation of the judgment target Yamagata included in the time series data of the blood pressure with the first threshold value, the blood pressure fluctuation of the judgment target Yamagata can be obtained. It is characterized in determining whether or not it is a blood pressure surge.

In the determination algorithm of this disclosure, only one feature quantity can be used for determining the blood pressure surge. Therefore, the calculation load for blood pressure surge determination can be reduced as compared with the conventional example. Moreover, the set of the first feature amount and the first threshold value is a set of the feature amount and the threshold value that gives the best value of the evaluation index through all the sets in the feature amount category group of the primary candidate. is there. Therefore, according to this determination algorithm, the accuracy of blood pressure surge determination can be improved.

In another aspect, the determination algorithm of this disclosure
It is a judgment algorithm using the first and second feature quantities determined by the method of generating the above judgment algorithm.
The value of the first feature amount indicated by the blood pressure fluctuation of Yamagata to be determined included in the time series data of the blood pressure is compared with the first threshold value, and the blood pressure fluctuation of Yamagata to be determined is By comparing the value of the second feature amount shown with the second threshold value, it is determined whether or not the blood pressure fluctuation of the Yamagata to be determined is a blood pressure surge.

In the determination algorithm of this disclosure, only two features can be used for determining the blood pressure surge. Therefore, the calculation load for blood pressure surge determination can be reduced as compared with the conventional example. Further, the feature amount categories that give the first and second feature amounts are different feature amount categories from each other. Therefore, as an element of the determination algorithm for determining the blood pressure surge, the feature amounts of two categories can be included, and the blood pressure surge determination can be performed from the two viewpoints. Moreover, the set of the first feature amount and the first threshold value is a set of the feature amount and the threshold value that gives the best value of the evaluation index through all the sets in the feature amount category group of the primary candidate. is there. Further, the set of the second feature amount and the second threshold value is a set of the feature amount and the threshold value that gives the best value of the evaluation index through all the sets in the feature amount category group of the secondary candidate. is there. As a result, the accuracy of blood pressure surge determination can be further improved.

In another aspect, the determination algorithm of this disclosure
A determination algorithm that uses the first, second, and third features determined by the method for generating the determination algorithm.
The value of the first feature amount indicated by the blood pressure fluctuation of Yamagata to be determined included in the time-series data of the blood pressure is compared with the first threshold value, and the blood pressure fluctuation of Yamagata to be determined is The value of the second feature amount shown is compared with the second threshold value, and the value of the third feature amount indicated by the blood pressure fluctuation of Yamagata to be determined is compared with the third threshold value. By comparing, it is characterized in that it is determined whether or not the blood pressure fluctuation of the Yamagata to be determined is a blood pressure surge.

In the determination algorithm of this disclosure, the number of features used for determining the blood pressure surge can be limited to three. Therefore, the calculation load for blood pressure surge determination can be reduced as compared with the conventional example. Further, the feature amount categories that give the first, second, and third feature amounts are different feature amount categories from each other. Therefore, as an element of the determination algorithm for determining the blood pressure surge, the feature quantities of the three categories can be included, and the blood pressure surge determination can be performed from the three viewpoints. Moreover, the set of the first feature amount and the first threshold value is a set of the feature amount and the threshold value that gives the best value of the evaluation index through all the sets in the feature amount category group of the primary candidate. is there. Further, the set of the second feature amount and the second threshold value is a set of the feature amount and the threshold value that gives the best value of the evaluation index through all the sets in the feature amount category group of the secondary candidate. is there. Further, the set of the third feature amount and the third threshold value is a set of the feature amount and the threshold value that gives the best value of the evaluation index through all the sets in the feature amount category group of the tertiary candidate. is there. As a result, the accuracy of blood pressure surge determination can be further improved.

Also, in another aspect, the determination algorithm of this disclosure
A determination algorithm that uses the first, second, and third features determined by the method for generating the determination algorithm.
The first, second, and third feature quantities were set as explanatory variables x1, x2, x3, respectively, the judgment result of whether or not the blood pressure was surge was set as the objective variable y, and α1, α2, α3, and β were used as coefficients. When
y = α1, x1 + α2, x2 + α3, x3 + β ... (Eq.1)
It is characterized in that it is determined whether or not the Yamagata blood pressure fluctuation included in the time series data of blood pressure is a blood pressure surge by linear discriminant analysis using a straight line represented by.

In the determination algorithm of this disclosure, for example, if y ≧ 0, the blood pressure fluctuation of the Yamagata to be determined is a blood pressure surge, while if y <0, the blood pressure fluctuation of the Yamagata to be determined is a non-blood pressure surge. It is judged as follows. The coefficients α1, α2, α3, and β are naturally determined in the process of applying this determination algorithm to the Yamagata blood pressure fluctuation included in the teacher data.

In this determination algorithm, the number of features used for determining the blood pressure surge can be limited to three. Therefore, the calculation load for blood pressure surge determination can be reduced as compared with the conventional example. Further, according to this determination algorithm, the accuracy of blood pressure surge determination can be improved.

Moreover, in another aspect, the determination algorithm of this disclosure
A determination algorithm that uses the first, second, and third features determined by the method for generating the determination algorithm.
When the first, second, and third features are the explanatory variables x1, x2, x3, the probability of whether or not there is a blood pressure surge is y, and α1, α2, α3, β are coefficients.
y = 1 / {-exp (α1 ・ x1 + α2 ・ x2 + α3 ・ x3 + β)}
… (Eq.2)
It is characterized in that it is determined whether or not the Yamagata blood pressure fluctuation included in the time series data of blood pressure is a blood pressure surge by logistic regression analysis using the formula represented by.

Here, exp (α1, x1 + α2, x2 + α3, x3 + β) represents an exponential function.

In the determination algorithm of this disclosure, for example, when a certain threshold th is set, if y ≧ th, the blood pressure fluctuation of the Yamagata to be determined is a blood pressure surge, while if y <th, the Yamagata to be determined is determined. It is determined that the blood pressure fluctuation is a non-blood pressure surge. The coefficients α1, α2, α3, and β are naturally determined in the process of applying this determination algorithm to the Yamagata blood pressure fluctuation included in the teacher data.

In this determination algorithm, the number of features used for determining the blood pressure surge can be limited to three. Therefore, the calculation load for blood pressure surge determination can be reduced as compared with the conventional example. Further, according to this determination algorithm, the accuracy of blood pressure surge determination can be improved.

In yet another aspect, the determination system of this disclosure
It is a judgment system that determines whether or not the Yamagata blood pressure fluctuation included in the blood pressure time series data is a blood pressure surge.
A storage unit that stores one of the above determination algorithms,
A data acquisition unit that acquires data representing the blood pressure fluctuation of the judgment target Yamagata included in the blood pressure time series data,
A feature amount calculation unit that calculates the feature amount to be used by the judgment algorithm for the blood pressure fluctuation of Yamagata, which is the judgment target, and a feature amount calculation unit.
It is characterized by including a surge determination unit that determines whether or not the blood pressure fluctuation of the Yamagata to be determined is a blood pressure surge based on the calculated feature amount using the stored determination algorithm. And.

In the determination system of this disclosure, the storage unit stores any of the above determination algorithms. The data acquisition unit acquires data representing the blood pressure fluctuation of Yamagata to be determined, which is included in the time series data of blood pressure. The feature amount calculation unit calculates the feature amount to be used by the determination algorithm for the blood pressure fluctuation of the Yamagata to be determined. The surge determination unit determines whether or not the blood pressure fluctuation of the Yamagata to be determined is a blood pressure surge based on the calculated feature amount using the stored determination algorithm. This makes it possible to determine whether or not the Yamagata blood pressure fluctuation included in the blood pressure time series data is a blood pressure surge. Here, since the determination algorithm of the present disclosure uses the determination algorithm described above, the calculation load of the blood pressure surge determination can be reduced and the accuracy of the blood pressure surge determination can be improved as compared with the conventional example. ..

In yet another aspect, the method of determining this disclosure is
It is a judgment method to judge whether or not the blood pressure fluctuation of Yamagata included in the time series data of blood pressure is a blood pressure surge.
A storage unit that stores one of the above determination algorithms is provided.
Acquire the data representing the blood pressure fluctuation of the judgment target Yamagata included in the time series data of blood pressure,
For the blood pressure fluctuation of Yamagata to be judged, the feature amount to be used by the judgment algorithm is calculated.
Using the stored determination algorithm, it is characterized in that it is determined whether or not the blood pressure fluctuation of the Yamagata to be determined is a blood pressure surge based on the calculated feature amount.

Since the determination algorithm described above is used in the determination method of this disclosure, the calculation load of the blood pressure surge determination can be reduced and the accuracy of the blood pressure surge determination can be improved as compared with the conventional example.

In yet another aspect, the program of this disclosure is a program for causing a computer to execute the above-mentioned determination method.

The above determination method can be implemented by causing a computer to execute the program of this disclosure.

In yet another aspect, the recording medium of this disclosure is a computer-readable recording medium that stores the program.

The determination method can be carried out by causing a computer to read and execute the program stored in the recording medium of the disclosure.

As is clear from the above, according to the method of generating the determination algorithm of this disclosure, it is a determination algorithm for determining whether or not the Yamagata blood pressure fluctuation included in the time series data of blood pressure is a blood pressure surge. Therefore, it is possible to generate a product that can reduce the calculation load of blood pressure surge determination. Further, according to the determination algorithm, determination system and determination method of this disclosure, the calculation load of blood pressure surge determination can be reduced. Further, the above-mentioned determination method can be carried out by causing a computer to execute the program of this disclosure. Further, the above-mentioned determination method can be carried out by causing a computer to read and execute the program stored in the recording medium of this disclosure.

It is a block diagram which shows one Embodiment which configured the determination system of this invention as a system on a network. It is a perspective view which illustrates the wearing state of the sphygmomanometer included in the said system. It is sectional drawing which illustrates the wearing state of the said blood pressure monitor. It is a figure which shows the block structure of the said sphygmomanometer. It is a figure which shows the block structure of the server included in the said system. FIG. 6A is a diagram showing time-series data of the overnight blood pressure of the subject. FIG. 6B is an enlarged view of a part of FIG. 6A, showing Yamagata's blood pressure fluctuation, which is a candidate for blood pressure surge included in the time-series data, and the waveform of the detected blood pressure surge. is there. It is a figure which shows the flow (a part) of the method of generating the determination algorithm of one Embodiment of this invention, which is executed by the control part of the said server. It is a figure which shows the flow (part) of the method of generating the said determination algorithm. It is a figure which shows the flow (part) of the method of generating the said determination algorithm. It is a figure which shows the data table which recorded the teacher data. It is a figure which illustrates the waveform of the blood pressure surge. In the figure schematically showing the range in which the six feature categories (I), (II), ..., (VI) are related, which are defined by the method for generating the above determination algorithm on the waveform of the blood pressure surge. is there. The figure which shows the set of the plurality of feature quantities created for each of the said feature quantity categories (I), (II), ..., (VI), and the 10-step threshold value for each of the said plurality of feature quantities. Is. By the method of generating the above determination algorithm, all sets (feature amount and threshold value) in the feature amount category group of the primary candidate (including six feature amount categories (I) to (VI) in this example). It is a figure which shows the evaluation index table which recorded the recall rate (Recall) obtained over (the set). It is a figure which shows the frequency distribution around the 1st threshold Th1 of the 1st feature amount Fe1 shown by the blood pressure fluctuation of a plurality of chevrons included in the teacher data. By the method of generating the above determination algorithm, all sets (feature amount and threshold value) in the feature amount category group of the secondary candidate (including five feature amount categories (II) to (VI) in this example). It is a figure which shows the evaluation index table which recorded the recall rate (Recall) obtained over (the set). A second feature amount Fe2 indicated by a plurality of chevron blood pressure fluctuations (a chevron blood pressure fluctuation classified as a blood pressure surge according to a combination of a first feature amount Fe1 and a first threshold Th1) included in the intermediate data. , It is a figure which shows the frequency distribution around the 2nd threshold Th2. It is a figure which shows typically the image which sequentially selects a feature amount one by one from each feature amount category (I)-(VI) by the method of generating the above-mentioned determination algorithm. It is a figure which shows the recall data (Recall) when up to 6 features are sequentially selected one by one from each feature category (I) to (VI) as an element of the judgment algorithm for blood pressure surge judgment. is there. It is a figure which shows the flow of the blood pressure surge determination method executed by the said system. It is a figure which shows the flow of the method of generating the judgment algorithm which forms a different kind of classifier.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

(Outline configuration of system)
FIG. 1 shows an example in which the determination system of the present invention is configured as a system of one embodiment on a network (the whole is indicated by reference numeral 100). This system 100 is a hospital terminal having a tonometry type sphygmomanometer 200, a server 300 that detects and analyzes a blood pressure surge from time-series data of blood pressure acquired by the sphygmomanometer 200, and a display 420 as a display screen. It includes 400A and 400B. These sphygmomanometers 200, servers 300, and hospital terminals 400A and 400B can communicate with each other via a network 900, which is a local area network (LAN) in this example. Communication via the network 900 may be either wireless or wired. In this embodiment, the network 900 is an in-hospital LAN (Local Area Network), but is not limited to this, and may be another type of network such as the Internet, or a USB cable or the like is used. It may be one-to-one communication. Although only two hospital terminals 400A and 400B (hereinafter collectively referred to by reference numerals 400) are shown in this example, three or more hospital terminals may be provided. Similarly, although only one sphygmomanometer 200 is shown in this example, two or more sphygmomanometers may be provided.

(Structure of blood pressure monitor)
The sphygmomanometer 200 shown in FIG. 1 includes, for example, a tonometry type sphygmomanometer as disclosed in Japanese Patent Application Laid-Open No. 2018-42606. FIG. 2 illustrates how the sphygmomanometer 200 is attached to the wrist W of the subject. Further, FIG. 3 illustrates a state in which the sphygmomanometer 200 attached to the wrist W of the subject is performing blood pressure measurement. The sphygmomanometer 200 illustrated in FIGS. 2 and 3 continuously measures the pressure pulse wave of the radial artery TD running along the radius 10 for each beat.

FIG. 4 illustrates the block configuration of the sphygmomanometer 200. In this example, the sphygmomanometer 200 includes a blood pressure device 210, a motion sensor 220, an operation unit 230, a communication unit 240, a memory 250, and a control unit 260. Further, the blood pressure device 210 includes a pressure sensor 211 and a pressing mechanism 212.

The pressing mechanism 212 applies a pressing force to the portion to be measured, as shown by an arrow in FIG. When the pressing mechanism 212 applies a pressing force to the measurement site, the pressure sensor 211 continuously detects the pressure pulse wave of the radial artery TD for each beat by the tonometry method. The tonometry method is a method in which a pressure sensor 211 measures a pressure pulse wave to determine blood pressure by compressing a blood vessel using a pressing mechanism 212. If a blood vessel is regarded as a circular tube of uniform thickness, the internal pressure (blood pressure) of the blood vessel and the external pressure of the blood vessel are followed according to Laplace's law, considering the blood vessel wall regardless of the blood flow in the blood vessel and the presence or absence of pulsation. The relational expression with (pressure of pressure pulse wave) can be derived. Under the condition that the blood vessel is compressed on the pressing surface by this relational expression, the pressure of the pressure pulse wave and the blood pressure can be approximated to be equal by approximating the radii of the outer wall and the inner wall of the blood vessel. Therefore, the pressure of the pressure pulse wave becomes the same value as the blood pressure. As a result, the sphygmomanometer 200 measures the blood pressure value of the measurement site for each heartbeat. Then, the sphygmomanometer 200 generates time-series data 801 of blood pressure in which the measurement time (time) and the blood pressure are associated with each other, as shown in FIG. 6A, for example. The overnight blood pressure time series data 801 contains beat-corresponding peaks for about 30,000 beats. In FIG. 6B, the peak of systolic blood pressure (SBP) for each beat is shown by the envelope LPS, and the peak of diastolic blood pressure (DBP) is shown by the envelope LPD. The chevron blood pressure fluctuation formed by the peaks that give the maximum of the envelope LPS and LPD (typically, the peaks PS1, PS2, PS3, ... That give the maximum of the envelope LPS) is a candidate for the blood pressure surge.

The motion sensor 220 shown in FIG. 4 includes a 3-axis acceleration sensor in this example, and outputs data representing the influence of disturbance (noise, etc.) such as body movement during blood pressure measurement.

The operation unit 230 receives an instruction (input) from a user (typically, a medical person such as a doctor or a nurse, a subject, or the like, which may be a system administrator). The operation unit 230 is composed of, for example, a plurality of buttons. The communication unit 240 transmits and receives various data. The communication unit 240 is connected to the network 900 shown in FIG. The memory 250 shown in FIG. 4 stores various data. For example, the memory 250 can store the measured value (time series data 801 of blood pressure) measured by the blood pressure device 210, the SBP value / DBP value for each beat, the pulse rate, and the like. The memory 250 includes a RAM (Random Access Memory), a ROM (Read Only Memory), and the like. For example, various programs are stored in the memory 250 so as to be changeable.

The control unit 260 includes a CPU (Central Processing Unit) in this example. For example, the control unit 260 reads each program and each data stored in the memory 250. Further, the control unit 260 controls each unit 210, 230, 240, 250 according to the read program to execute a predetermined operation (function). Further, the control unit 260 performs a predetermined calculation, analysis, processing, etc. in the control unit 260 according to the read program. It should be noted that a part or all of each function executed by the control unit 260 may be configured in hardware by one or a plurality of integrated circuits or the like.

The sphygmomanometer 200 is a blood pressure time series data 801 (see FIG. 6 (A)) in which the measurement time (time) and the blood pressure are associated with each other, and the SBP value / DBP value, pulse rate, and motion sensor 220 for each beat. The output or the like is output as measurement data to another device (server 300 in this example).

(Server configuration)
FIG. 5 illustrates a block configuration of the server 300. In this example, the server 300 includes a communication unit 310, a display 320, an operation unit 330, a storage unit 340, and a control unit 350.

The communication unit 310 transmits and receives various data. The communication unit 310 is connected to the network 900 shown in FIG. The communication unit 310 receives, for example, the measurement data transmitted from the sphygmomanometer 200 via the network 900. Further, the communication unit 310 transmits various output data generated by the control unit 350 in the server 300 to the hospital terminal 400 via the network 900.

The display 320 shown in FIG. 5 has a display screen for displaying various images. The display 320 can visually display the results of various analyzes in the control unit 350 and the like. In addition, the display 320 can visually display predetermined information according to a request from the user via the operation unit 330. For example, the display 320 may visually display the information (data) stored in the storage unit 340. For example, as the display 320, a liquid crystal display, an organic EL (Electro Luminescence) display, or the like can be adopted.

The operation unit 330 receives a predetermined operation (instruction) from the user. For example, the operation unit 330 is composed of a mouse, a keyboard, and the like. Here, when a touch panel type monitor is adopted as the display 320, the display 320 has not only a display function but also a function as an operation unit 330.

The storage unit 340 stores various data. For example, the storage unit 340 can store the measured values (time series data 801 of blood pressure) measured by the blood pressure device 210, and the measured data such as the SBP value / DBP value for each beat and the pulse rate. The storage unit 340 can also store various output data generated by the control unit 350. The storage unit 340 includes a RAM, a ROM, and the like. Further, various programs are stored in the storage unit 340 so as to be changeable. In this example, the storage unit 340 stores a program for executing a method for generating a determination algorithm, a determination method, and the like, which will be described later. As a storage medium of the auxiliary storage device for assisting the storage area of the storage unit 340, a magnetic disk (HD (Hard Disk), FD (Flexible Disk)), an optical disk (CD (Compact Disc), DVD (Digital Versatile Disk)) ), BD (Blu-ray (registered trademark) Disc)), an optical magnetic disk (MO (Magneto-Optical disk)), a semiconductor memory (memory card, SSD), or the like may be used.

The control unit 350 includes a CPU in this example. For example, the control unit 350 reads each program and each data stored in the storage unit 340. Further, the control unit 350 controls each unit 310, 320, 330, 340 according to the read program to execute a predetermined operation (function). Further, the control unit 350 performs a predetermined calculation, analysis, processing, etc. in the control unit 350 according to the read program. A part or all of each function executed by the control unit 350 may be configured in hardware by one or a plurality of integrated circuits or the like. The specific operation of the control unit 350 will be described later.

(Hospital terminal configuration)
The hospital terminal 400 shown in FIG. 1 is composed of a general personal computer in this example. Here, the hospital terminal 400 may be a mobile terminal such as a tablet instead of the personal computer.

The display 420 included in the hospital terminal 400 has a display screen for displaying various images. The display 420 visually displays, for example, images corresponding to various output data received from the server 300. In addition, the display 420 can visually display predetermined information according to the operation by the user. For example, as the display 420, a liquid crystal display, an organic EL display, or the like can be adopted.

(Generation of judgment algorithm by server)
In this system 100, the server 300 executes the "method of generating the determination algorithm" of one embodiment, and the Yamagata blood pressure fluctuation included in the blood pressure time series data (see FIG. 6 (A)) is a blood pressure surge. Generate a judgment algorithm to judge whether or not there is.

7A-7C show a flow of a method of generating a determination algorithm executed by the control unit 350 of the server 300.

First, as shown in step S11 of FIG. 7A, the control unit 350 classifies a plurality of features capable of characterizing the waveform of the blood pressure surge, and classifies a plurality of feature amounts of the primary candidate (in this example, the feature amount category group of the primary candidate). Includes 6 feature categories (I)-(VI)). Here, the "primary candidate" and the "secondary candidate" and "tertiary candidate" described later represent the order in which the feature amounts are sequentially selected according to this flow.

In this example, as shown in FIG. 11, the feature amount category (I) is a feature amount category related to the reaction stage (Reactivity Phase), and the feature amount is "rise time" (corresponding to T1 in FIG. 9). , "Rising beats" (corresponding to the number of beats during the rising time), etc. are included.

The feature amount category (II) is a feature amount category related to the recovery phase, and the feature amounts are "descending time" (corresponding to T3 in FIG. 9) and "descending beat rate" (during the descending time). (Equivalent to the number of beats of), etc. are included.

The feature quantity category (III) is a category of feature quantities related to amplitude, and the feature quantities include "rising fluctuation amount" (corresponding to L1 in FIG. 9) and "descending fluctuation amount" (FIG. 9). (Equivalent to L3 inside), etc. are included.

In addition, the feature amount category (IV) is a category of feature amount related to ascending (Upward), and the feature amount includes "inclination at the time of ascending" (corresponding to the gradient from the point P1 to the point P2 in FIG. 9) and "ascending". The area of time ”(corresponding to the area formed between the waveform C from the point P1 to the point P2 in FIG. 9 and the horizontal line passing through the point P1) and the like are included.

Further, the feature amount category (V) is a category of the feature amount related to the descent (Downward), and the feature amount includes "inclination at the time of descent" (corresponding to the gradient from the point P2 to the point P4 in FIG. 9) and "descending". The area of time ”(corresponding to the area formed between the waveform C from the point P2 to the point P4 in FIG. 9 and the horizontal line passing through the point P1) and the like are included.

The feature amount category (VI) is a feature amount category related to the whole duration, and the feature amount is "total time" (corresponding to the total of T1 and T3 in FIG. 9) and "total number of beats". "(Equivalent to the number of beats during the entire time), etc. are included.

In addition, in FIG. 10, on the waveform of the blood pressure surge, the range related to the feature quantity categories (I), (II), ..., (VI) is the same as the category number (I), (II), ... , (VI) are schematically shown by double-headed arrows.

In this example, various feature quantities such as "rising time" and "rising beat" are generally represented by the symbol "Fe". Further, the threshold value of one or more steps for each feature amount Fe is generally represented by the symbol “Th”. Further, the set of the feature amount Fe and the threshold value Th is generally represented by the symbol "Cm".

Next, as shown in step S12 of FIG. 7A, the control unit 350 has a plurality of feature quantities Fe included in the feature quantity categories for each of the feature quantity categories (I), (II), ..., (VI). And a threshold value Th of one or more steps for each of the plurality of feature amounts Fe are combined to create a plurality of sets Cm of the feature amount Fe and the threshold value Th.

For example, as shown in FIG. 11, in the feature amount category (I), 1 to 10 [sec. ] Over the range of 1 [sec. ] In increments, prepare 10-step threshold Th. The feature amount, "rise time", and 1 to 10 [sec. ] In combination with the 10-step threshold Th, 10 sets Cm11 to Cm20 are created. Further, for the feature amount of "rising beats", a threshold value Th of 10 steps is prepared in 1 [beat] increments over a range of 1 to 10 [beats]. Ten sets of Cm21 to Cm30 are created by combining the feature amount "rising beat number" and the threshold value Th in 10 steps in the range of 1 to 10 [beats].

Further, in the feature amount category (II), 1 to 10 [sec. ] Over the range of 1 [sec. ] In increments, prepare 10-step threshold Th. The feature amount, "descending time", and 1 to 10 [sec. ] In combination with the 10-step threshold Th, 10 sets Cm31 to Cm40 are created. Further, for the feature amount of "falling beats", a threshold value Th of 10 steps is prepared in 1 [beat] increments over a range of 1 to 10 [beats]. Ten sets of Cm41 to Cm50 are created by combining the feature amount "falling beats" and the threshold value Th in 10 steps in the range of 1 to 10 [beats].

Further, in the feature amount category (III), a threshold value Th of 10 steps is prepared in increments of 4 [mmHg] over a range of 20 to 40 [mmHg] for the feature amount “variation amount at the time of rising”. A set of 10 Cm51 to Cm60 is created by combining the feature amount "fluctuation amount during ascent" and the threshold value Th in 10 steps in the range of 20 to 40 [mmHg]. Further, for the feature amount of "fluctuation amount during descent", a threshold value Th of 10 steps is prepared in increments of 4 [mmHg] over a range of 20 to 40 [mmHg]. Ten sets of Cm61 to Cm70 are created by combining the feature amount "fluctuation amount during descent" and the threshold value Th in 10 steps in the range of 20 to 40 [mmHg].

Further, in the feature amount category (IV), 0 to 3.0 [mmHg / sec. ] In the range of 0.3 [mmHg / sec. ] In increments, prepare 10-step threshold Th. The feature amount, "fluctuation amount during ascent", and 0 to 3.0 [mmHg / sec. ] In combination with the 10-step threshold Th, 10 sets Cm71 to Cm80 are created. In addition, due to the feature amount of "area when rising", 10 to 200 [mmHg · sec. ] Over the range of 10 [mmHg · sec. ] In increments, prepare 10-step threshold Th. The feature amount, "area when rising", and 10 to 200 [mmHg · sec. ] In combination with the threshold value Th of 10 steps, 10 sets Cm81 to Cm90 are created.

Further, in the feature amount category (V), -3.0 to 0 [mmHg / sec. ] In the range of 0.3 [mmHg / sec. ] In increments, prepare 10-step threshold Th. The feature amount, "fluctuation amount during descent", and -3.0 to 0 [mmHg / sec. ] In combination with the threshold value Th of 10 steps, 10 sets Cm91 to Cm100 are created. Further, due to the feature amount "area when descending", 10 to 200 [mmHg · sec. ] Over the range of 10 [mmHg · sec. ] In increments, prepare 10-step threshold Th. The feature amount, "area when rising", and 10 to 200 [mmHg · sec. ] In combination with the 10-step threshold Th, 10 sets Cm101 to Cm110 are created.

Further, in the feature amount category (VI), 1 to 30 [sec. ] Over the range of 3 [sec. ] In increments, prepare 10-step threshold Th. The feature amount, "total time", and 1 to 30 [sec. ] In combination with the 10-step threshold Th, 10 sets Cm111-Cm120 are created. Further, for the "total number of beats" which is a feature amount, a threshold value Th of 10 steps is prepared in 3 [beat] increments over a range of 1 to 30 [beats]. A set of 10 Cm121-Cm130 is created by combining the feature amount "total number of beats" and the threshold value Th in 10 steps in the range of 1 to 30 [beats].

In this example, in this way, for each feature category (I), (II), ..., (VI), (feature quantity) × (threshold number) = (feature quantity number) × 10 [pieces]. Create a set Cm. In this example, the threshold value Th of 10 steps is combined with any of the plurality of feature quantities Fe, but the present invention is not limited to this. For each feature Fe, the number of steps of the threshold Th may be different.

Further, for a certain feature amount Fe, either an upper limit threshold value (this is represented by the symbol "ThU") or a lower limit threshold value (this is represented by the symbol "ThL") may be set. .. Both the upper limit threshold value ThU and the lower limit threshold value ThL may be set. It is desirable to set a flag in the storage unit 340 indicating whether the threshold value is ThU, which is the upper limit, or ThL, which is the lower limit.

Next, the control unit 350 prepares the teacher data Dte as shown in FIG. 7A. In this example, the teacher data Dte is a plurality of chevron blood pressure fluctuations derived from the blood pressure time series data 801 (see FIG. 6 (A)), and it is known in advance whether or not each of them is a blood pressure surge. It shall be.

Specifically, in this example, the teacher data Dte is stored in the storage unit 340 in the form of a data table as shown in FIG. In this data table, the mountain-shaped blood pressure fluctuations with respect to systolic blood pressure (SBP) (for example, the mountain-shaped blood pressure fluctuations formed by the peaks PS1, PS2, PS3, ... ) Occurred "time", characteristic quantities such as "rising fluctuation amount", "surge peak value", "rising time", "falling time" of blood pressure fluctuation of each Yamagata, and blood pressure fluctuation of each Yamagata is blood pressure "Teacher labels" indicating information on whether or not there is a surge (in this example, "surge" or "non-surge") are recorded in association with each other.

In this example, the distinction between "teacher label" "surge" and "non-surge" is that, under the supervision of a doctor, a person who is substantially at the same level as the doctor in determining blood pressure surge is the blood pressure fluctuation of each Yamagata. It was determined by looking at the waveform (the time-varying waveform of blood pressure as illustrated in FIG. 9). Judgment results (separate between "surge" and "non-surge") of Yamagata's blood pressure fluctuations, which are sufficient for statistical processing (hundreds in this example), are sent to the storage unit 340 via the operation unit 330. It was recorded. In the following, the blood pressure fluctuation of Yamagata recorded as "surge" in the "teacher label" column is called "true surge", and the blood pressure fluctuation of Yamagata recorded as "non-surge" in the "teacher label" column is called. Called "true non-surge".

In this example, the control unit 350 reads out such a data table as the teacher data Dte from the storage unit 340 and prepares it.

It should be noted that steps S11 and 12 of FIG. 7A and preparation of the teacher data Dte may be carried out first, or may be carried out in parallel with each other.

Next, as shown in step S13 of FIG. 7A, the control unit 350 targets the blood pressure fluctuations of a plurality of chevrons included in the teacher data Dte, and the feature amount category group of the primary candidate (6 in this example). The feature amounts Fe and the threshold value Th of all the sets in the feature amount categories (I) to (VI) of the above are used as a classifier for each set of Cm, and it is determined whether or not each of them is classified as a blood pressure surge. ..

In this example, for a plurality of chevron blood pressure fluctuations included in the teacher data Dte, the entire set of the feature amounts Fe and the threshold Th in the feature amount categories (I) to (VI) of the primary candidates Cm11 to Cm30. , ..., Cm31 to Cm50, ..., Cm51 to Cm70, ..., Cm71 to Cm90, ..., Cm91 to Cm110, ..., Cm111 to Cm130, ... Determine if it is classified as a blood pressure surge.

Here, for a set Cm of a certain feature amount Fe and a certain threshold value Th, the threshold value Th and the value of the corresponding feature amount of the Yamagata blood pressure fluctuation included in the teacher data Dte (this is represented by the symbol “x”). When comparing with (represented), the upper limit threshold ThU is classified as a blood pressure surge if the inequality (x <ThU) is satisfied, while the non-blood pressure surge is classified if the inequality (x ≧ ThU) is satisfied. Classify into. Further, the threshold value ThL, which is the lower limit, is classified as a blood pressure surge if the inequality (x> ThL) is satisfied, and is classified as a non-blood pressure surge if the inequality (x ≦ ThL) is satisfied.

Assuming that the number of blood pressure fluctuations in Yamagata included in the teacher data Dte is n1 (n1 is several hundreds), n1 × (number of features in feature categories (I) to (VI)) × (for each feature) The determination result of (threshold number) is obtained.

Next, as shown in step S14 of FIG. 7A, the control unit 350 sets an evaluation index for each set of Cm as a classifier over all sets of Cm in the feature amount category group of the primary candidate. Ask.

This evaluation index may be a general one that represents the performance of the classifier, such as recall, precision, and F-measure. In this example, it is assumed that the recall rate (Recall) is obtained as an evaluation index. The recall rate is expressed as TP for the number of True Positives and FN for the number of False Negatives.
Recall = TP / (TP + FN)
Is defined as. Here, the true positive number TP represents the number at which the true surge can be correctly determined to be a blood pressure surge by the classifier. The false negative number FN represents the number determined by the classifier to be a non-blood pressure surge, even though it is a true surge.

As a result, for example, the results shown in the evaluation index table of FIG. 12 can be obtained. This evaluation index table includes a "feature amount category" column, a "feature amount" column, a "threshold value" column, and a "recall rate (Recall)" column. In the "feature amount category" column, feature amount categories (I), (II), (III), ... Are specified. In the "feature amount" column, the feature amount Fe included in the feature amount category is specified for each feature amount category in the "feature amount category" column. For example, with respect to the feature amount category (I), in the "feature amount" column, "rising time", "rising beat number [beat]" and the like are specified. In the "threshold value" column, the threshold value Th for the feature amount Fe is specified for each feature amount Fe specified in the "feature amount" column. For example, regarding the "rise time" which is the feature amount included in the feature amount category (I), the threshold values 1, 2, 3, 4, ... [Sec. ] Has been identified. Similarly, for the "rising beats", which is the feature amount included in the feature amount category (I), the threshold values 1, 2, 3, 4, ... [Beats] are specified. In addition, the threshold values 20, 22, 24, 26, ... [MmHG] are specified for the "variation amount during ascent" included in the feature amount category (III).

In the "Recall" column, for each set Cm of the feature amount Fe specified in the "feature amount" column and the threshold value Th specified in the "threshold value" column, the recall rate (Recall) for the set Cm. Is recorded. For example, "rise time" and threshold 1 [sec. ] And the recall rate (Recall) for the set Cm is 0.61, the "rise time" and the threshold value 2 [sec. ], The recall rate (Recall) for the set Cm is 0.82, and so on. The recall rate (Recall) for the set Cm of the "rising beat" and the threshold value 1 [beat] is 0.52, and the recall rate (Recall) for the set Cm of the "rising beat" and the threshold value 2 [beat] is 0.52. The result of 0.74, ... is obtained. The recall rate (Recall) for the set Cm of the "rising fluctuation amount" and the threshold value 20 [mmHG] is 0.66, and the recall rate for the set Cm of the "rising fluctuation amount" and the threshold value 22 [mmHG]. The result of (Recall) is 0.65, ...

In this example, the recall rate (Recall) as an evaluation index is obtained by (the number of features in the feature categories (I) to (VI)) × (the number of thresholds for each feature).

Next, as shown in step S15 of FIG. 7A, the control unit 350 is among the evaluation indexes (recall rate (Recall) in the above example) obtained over the entire set Cm in the feature amount category group of the primary candidate. The set Cm of the feature amount Fe and the threshold value Th that gives the best value is determined as the set Cm1 of the first feature amount Fe1 and the first threshold value Th1 that form an element of the determination algorithm.

In the example shown in the evaluation index table of FIG. 12 above, among the recalls (Recall) obtained over all sets of Cm in the feature amount categories (I) to (VI) of the primary candidates, the feature amount category (I) "Rise time", which is a feature amount contained in, and a threshold value of 4 [sec. ] And the recall rate (Recall) for the set Cm14 gives the best value B1 such as 0.93. Therefore, in this example, the “rise time” and the threshold value 4 [sec. ] And the set Cm14 are determined as the set Cm1 of the first feature amount Fe1 and the first threshold value Th1 which are the elements of the determination algorithm.

In this way, a set of the first feature amount Fe1 and the first threshold Th1 is used as an element for determining whether or not the Yamagata blood pressure fluctuation included in the blood pressure time series data 801 is a blood pressure surge. A determination algorithm can be generated that includes only Cm1 as useful.

However, in this embodiment, the flow proceeds to FIG. 7B in order to add an element included in the determination algorithm.

First, in step S16 of FIG. 7B, the control unit 350 uses the set Cm1 of the first feature amount Fe1 and the first threshold value Th1 among the plurality of chevron blood pressure fluctuations included in the teacher data Dte to cause a blood pressure surge. The data on the blood pressure fluctuation of Yamagata classified in is specified as the intermediate data Dt2.

For example, FIG. 13 shows the frequency distribution of the first feature amount Fe1 indicated by the blood pressure fluctuations of a plurality of chevrons included in the teacher data Dte around the first threshold Th1. In FIG. 13, the true non-surge frequency is shaded from the upper right to the lower left, the true surge frequency is shaded from the upper left to the lower right, and the true non-surge frequency is defined as the true non-surge frequency. Cross-hatching is attached to the portion where the true surge frequency overlaps (the same applies to FIG. 15 described later). In this example, most of the true non-surge is contained in the group G1 whose value is smaller than the first threshold Th1. On the other hand, most of the true surge is contained in the group G2 whose value is larger than the first threshold Th1. In this case, the control unit 350 specifies the data for the group G2 whose value is larger than the first threshold Th1 among the plurality of chevron blood pressure fluctuations included in the teacher data Dte as the intermediate data Dt2. In the above example, since the teacher data Dte includes a considerable number of Yamagata blood pressure fluctuations (n1 is several hundreds), this intermediate data Dt2 still contains a considerable number of Yamagata blood pressure fluctuations sufficient for statistical processing. Can be included. As can be seen from FIG. 13, the intermediate data Dt2 still contains true non-surge data.

Further, in step S17 of FIG. 7B, the control unit 350 has a first feature from the feature amount category group of the primary candidate (in this example, six feature amount categories (I) to (VI) are included). The feature amount category to which the quantity Fe1 is given is excluded, and all the remaining feature amount categories are specified as the feature amount category group of the secondary candidate.

In the above example, since the "rise time" included in the feature amount category (I) becomes the first feature amount Fe1, the feature amount category (I) is excluded from the feature amount category group of the primary candidates. Will be done. As a result, the five feature amount categories (II) to (VI) are specified as the feature amount category group of the secondary candidate.

In addition, step S16 and step S17 of FIG. 7B may be carried out first, or may be carried out in parallel with each other.

Next, as shown in step S18 of FIG. 7B, the control unit 350 targets a plurality of chevron blood pressure fluctuations included in the intermediate data Dt2, and features a secondary candidate feature category group (five in this example). The feature amount Fe and the threshold value Th of all sets Cm in the feature amount categories (II) to (VI) of the above are used as classifiers for each set of Cm to determine whether or not they are classified as blood pressure surges. To do.

In this example, for a plurality of chevron blood pressure fluctuations (group G2) included in the intermediate data Dt2, all of the feature amounts Fe and the threshold Th in the feature amount categories (II) to (VI) of the secondary candidates. Blood pressure surges using sets Cm31 to Cm50, ..., Cm51 to Cm70, ..., Cm71 to Cm90, ..., Cm91 to Cm110, ..., Cm111 to Cm130, ... (See FIG. 11) as classifiers for each set of Cm. Judge whether or not it is classified into.

Assuming that the number of Yamagata blood pressure fluctuations included in the intermediate data Dt2 is n2 (usually n2 <n1), n2 × (the number of features in the feature categories (II) to (VI)) × ( The determination result of (the number of threshold values for each feature amount) is obtained.

Next, as shown in step S19 of FIG. 7B, the control unit 350 sets an evaluation index for each set of Cm as a classifier over all sets of Cm in the feature amount category group of the secondary candidate. Ask.

In this example, the recall rate (Recall) is obtained as an evaluation index for all the sets of Cm in the feature amount categories (II) to (VI) of the secondary candidates for the judgment result as a classifier for each set of Cm.

Then, for example, the result shown in the evaluation index table of FIG. 14 is obtained. Similar to the evaluation index table of FIG. 12, this evaluation index table includes a "feature amount category" column, a "feature amount" column, a "threshold value" column, and a "recall" column. .. In the evaluation index table of FIG. 14, the data for the excluded feature amount category (I) is omitted.

In the "Recall" column, for each set Cm of the feature amount Fe specified in the "feature amount" column and the threshold value Th specified in the "threshold value" column, the recall rate (Recall) for the set Cm. Is recorded. For example, "descending time" and threshold value 1 [sec. ] And the recall rate (Recall) for the set Cm is 0.82, the "descending time" and the threshold value 2 [sec. ], The recall rate (Recall) for the set Cm is 0.88, and so on. The recall rate (Recall) for the set Cm of the "falling beats" and the threshold value 1 [beat] is 0.55, and the recall rate (Recall) for the set Cm of the "falling beats" and the threshold value 2 [beats] is 0.55. The result of 0.76, ... is obtained. The recall rate (Recall) for the set Cm of the "rising fluctuation amount" and the threshold value 20 [mmHG] is 0.69, and the recall rate for the set Cm of the "rising fluctuation amount" and the threshold value 22 [mmHG]. The result of (Recall) is 0.63, ...

In this example, the recall rate (Recall) as an evaluation index is obtained by (the number of features in the feature categories (II) to (VI)) × (the number of thresholds for each feature).

Next, as shown in step S20 of FIG. 7B, the control unit 350 is among the evaluation indexes (recall in the above example) obtained over the entire set Cm in the feature amount category group of the secondary candidate. The set Cm of the feature amount Fe and the threshold value Th that gives the best value is determined as the set Cm2 of the second feature amount Fe2 and the second threshold value Th2 that form an element of the determination algorithm.

In the example shown in the evaluation index table of FIG. 14 above, among the recalls (Recall) obtained over all sets of Cm in the feature amount categories (II) to (VI) of the secondary candidates, the feature amount category (II) "Descent time", which is a feature quantity included in, and a threshold value of 4 [sec. ] And the recall rate (Recall) for the set Cm34 is 0.89, which gives the best value B2. Therefore, in this example, the "descending time" and the threshold value 4 [sec. ] And the set Cm34 are determined as the set Cm2 of the second feature amount Fe2 and the second threshold value Th2, which are elements of the determination algorithm.

In this way, a set of the first feature amount Fe1 and the first threshold Th1 is used as an element for determining whether or not the Yamagata blood pressure fluctuation included in the blood pressure time series data 801 is a blood pressure surge. A determination algorithm can be generated that includes only Cm1 and the set Cm2 of the second feature amount Fe2 and the second threshold Th2 as useful.

However, in this embodiment, the flow proceeds to FIG. 7C in order to further add the elements included in the determination algorithm.

First, in step S21 of FIG. 7C, the control unit 350 uses the set Cm2 of the second feature amount Fe2 and the second threshold value Th2 among the plurality of chevron blood pressure fluctuations included in the intermediate data Dt2 to cause a blood pressure surge. The data on the blood pressure fluctuation of Yamagata classified into the above is specified as the final data Dt3.

For example, FIG. 15 shows the frequency distribution of the second feature amount Fe2 indicated by the blood pressure fluctuations of the plurality of chevrons included in the intermediate data Dt2 around the second threshold Th2. In this example, most of the true non-surge is contained in group G3, which has a value greater than the second threshold Th2. On the other hand, most of the true surge is contained in the group G4 whose value is smaller than the second threshold Th2. In this case, the control unit 350 specifies the data for the group G4 whose value is smaller than the second threshold value Th2 among the plurality of chevron blood pressure fluctuations included in the intermediate data Dt2 as the final data Dt3. In the above example, since the teacher data Dte includes a considerable number of Yamagata blood pressure fluctuations (n1 is several hundreds), this final data Dt3 still contains a considerable number of Yamagata blood pressure fluctuations sufficient for statistical processing. Can be included. As can be seen from FIG. 15, the final data Dt3 still contains a small amount of true non-surge data.

Further, in step S22 of FIG. 7C, the control unit 350 has a second feature from the feature amount category group of the secondary candidate (including five feature amount categories (II) to (VI) in this example). The feature amount category to which the quantity Fe2 is given is excluded, and all the remaining feature amount categories are specified as the feature amount category group of the tertiary candidate.

In the above example, since the "descending time" included in the feature amount category (II) becomes the second feature amount Fe2, the feature amount category (II) is excluded from the feature amount category group of the secondary candidate. Will be done. As a result, the four feature amount categories (III) to (VI) are specified as the feature amount category group of the tertiary candidate.

In addition, either step S21 and step S22 of FIG. 7C may be carried out first, or may be carried out in parallel with each other.

Next, as shown in step S23 of FIG. 7C, the control unit 350 targets a plurality of chevron blood pressure fluctuations included in the final data Dt3, and features a third-order candidate feature category group (four in this example). The feature amount Fe and the threshold value Th of all sets Cm in the feature amount categories (III) to (VI) of the above are used as classifiers for each set of Cm to determine whether or not they are classified as blood pressure surges. To do.

In this example, for a plurality of Yamagata blood pressure fluctuations (group G4) included in the final data Dt3, all of the feature amounts Fe and the threshold Th in the feature amount categories (III) to (VI) of the tertiary candidates. Whether or not each set of Cm51 to Cm70, ..., Cm71 to Cm90, ..., Cm91 to Cm110, ..., Cm111 to Cm130, ... (See FIG. 11) is classified as a blood pressure surge by using each set of Cm as a classifier. Is determined.

Assuming that the number of Yamagata blood pressure fluctuations included in the final data Dt3 is n3 (usually n3 <n2), n3 × (the number of features in the feature categories (III) to (VI)) × ( The determination result of (the number of threshold values for each feature amount) is obtained.

Next, as shown in step S24 of FIG. 7C, the control unit 350 sets an evaluation index for each set of Cm as a classifier over all sets of Cm in the feature amount category group of the tertiary candidate. Ask.

In this example, the recall rate (Recall) is obtained as an evaluation index for all the sets of Cm in the feature quantity categories (III) to (VI) of the tertiary candidates for the judgment result as a classifier for each set of Cm.

In this example, the recall rate (Recall) as an evaluation index is obtained by (the number of features in the feature categories (III) to (VI)) × (the number of thresholds for each feature).

Next, as shown in step S25 of FIG. 7C, the control unit 350 is among the evaluation indexes (recall in the above example) obtained over the entire set Cm in the feature amount category group of the tertiary candidate. The set Cm of the feature amount Fe and the threshold value Th that gives the best value is determined as the set Cm3 of the third feature amount Fe3 and the third threshold value Th3 that form the elements of the determination algorithm.

As a result, the evaluation index is included in all the sets of Cm (feature amount Fe and threshold value Th) including a plurality of sets of feature amount Fe and threshold value Th in each feature amount category (III) to (VI) of the tertiary candidate. The set Cm that gives the best value of (this is referred to as B3) is determined as the set Cm of the third feature amount Fe3 and the third threshold value Th3. In this example, the set Cm51 of the "rising amount fluctuation amount" which is the feature amount included in the feature amount category (III) and the threshold value 20 [mmHg] is the third feature amount Fe3 and the third feature amount Fe3 which are the elements of the determination algorithm. It is assumed that the set Cm3 with the threshold value Th3 of 3 is determined.

In this way, as elements of the determination algorithm for determining the blood pressure surge, the set Cm1 of the first feature amount Fe1 and the first threshold value Th1 and the set Cm2 of the second feature amount Fe2 and the second threshold value Th2. , And a determination algorithm can be generated that includes only the set Cm3 of the third feature amount Fe3 and the third threshold value Th3 as useful.

(Parameter adjustment)
After determining the sets Cm1 to Cm3 that form the elements of the determination algorithm, the first threshold value Th1, the second threshold value Th2, and the third threshold value Th3 in the determined sets Cm1 to Cm3 are further adjusted (parameter adjustment). You may.

Specifically, for each threshold value Th1, Th2, Th3 forming the set Cm1 to Cm3, a search range narrower than the initial range (the range shown in FIG. 11 for each threshold value) including the threshold value is set. A known grid search is performed over this search range in steps finer than the initial steps, and the above-mentioned threshold values Th1, Th2, and Th3 are adjusted.

For example, the first threshold Th1 (lower limit 4 [sec.] In the above example) for the first feature Fe1 (“rise time” in the above example) forming the set Cm1 is shown in FIG. As shown in 1 to 10 [sec. ] Over the range of 1 [sec. ] It is determined from the values set in 10 steps in increments. In this case, for example, the threshold value 4 [sec. ], Which is narrower than the initial range (1-10 [sec.]) Of 3-5 [sec.]. ] Set the search range. Over this search range, finer increments than the initial increments (1 [sec.]), For example 0.2 [sec.]. ] In increments, a new candidate threshold group of a plurality of stages (10 stages in this example) is set.

Similarly, the second threshold Th2 (lower limit 4 [sec.] In the above example) for the second feature Fe2 (“descent time” in the above example) forming the set Cm2 is shown in FIG. As shown in the inside, 1 to 10 [sec. ] Over 1 [sec. ] It is determined from the values set in 10 steps in increments. In this case, for example, the threshold value 4 [sec. ], Which is narrower than the initial range (1-10 [sec.]) Of 3-5 [sec.]. ] Set the search range. Over this search range, finer increments than the initial increments (1 [sec.]), For example 0.2 [sec.]. ] In increments, a new candidate threshold group of a plurality of stages (10 stages in this example) is set.

Further, the third threshold value Th3 (in the above example, the lower limit of 20 [mmHg]) for the third feature amount Fe3 (“increase amount fluctuation amount” in the above example) forming the set Cm3 is shown in FIG. As shown in the inside, it was determined from the values set in 10 steps in 2 [mmHg] increments over the range of 20 to 40 [mmHg]. In this case, for example, a search range of 18 to 22 [mmHg], which includes the threshold value of 20 [mmHg] and is narrower than the initial range (20 to 40 [mmHg]), is set. Over this search range, a new candidate threshold group of a plurality of steps (10 steps in this example) is set in steps finer than the initial step (2 [mmHg]), for example, in steps of 0.4 [mmHg].

In this way, for each threshold value Th1 to Th3 (these are the three axes), a plurality of stages of candidate threshold value groups are set in a finer step than the initial step over a search range narrower than the initial range. Then, a known grid search is performed on the combination of the candidate threshold values set for each of the three axes (each set formed by combining the candidate thresholds of the three axes serves as a grid point). Specifically, the performance value of surge determination is calculated over all of the above grid points. Of the performance values obtained over all of the above grid points, the grid point that gives the best value is obtained. The set of the three-axis candidate thresholds formed by the grid points is determined as a set of adjusted new thresholds. Thereby, each threshold value Th1, Th2, Th3 is adjusted. In this example, the first threshold Th1 (lower limit) for the first feature Fe1 (“rise time” in the above example) forming the set Cm1 is 4.8 [sec. ] It shall be adjusted. Further, the second threshold value Th2 (lower limit) for the second feature amount Fe2 (“descent time” in the above example) forming the set Cm2 is 3.2 [sec. ] It shall be adjusted. Further, the third threshold value Th3 (lower limit) for the third feature amount Fe3 (in the above example, the “increased amount fluctuation amount”) forming the set Cm3 was adjusted to 20 [mmHg] (as in the initial state). (It became the value of).

In this case, since the respective threshold values Th1, Th2, and Th3 are finely adjusted (parameter adjustment), the accuracy of blood pressure surge determination by the generated determination algorithm can be improved. Moreover, in this example, the set Cm1 of the first feature amount Fe1 and the first threshold value Th1, the set Cm2 of the second feature amount Fe2 and the second threshold value Th2, and the third feature amount Fe3 and the third feature amount Fe3. After the set Cm3 with the threshold value Th3 of 3 is once determined, the grid search is performed again to adjust the parameters, so that the overall optimization can be performed. The parameters may be adjusted each time the sets Cm1, Cm2, and Cm3 that form the elements of the determination algorithm are determined.

(Generated judgment algorithm)
Depending on the method for generating the determination algorithm described above, the elements of the determination algorithm include, for example, the "rise time" as the first feature amount Fe1 and the threshold value of 4.8 [sec. ] And the set Cm1 is determined. Further, the "descending time" as the second feature amount Fe2 and the threshold value 3.2 [sec. ] And the set Cm2 is determined. Further, it is assumed that the set Cm3 of the "rising fluctuation amount" as the third feature amount Fe3 and the threshold value 20 [mmHg] which is the lower limit as the third threshold value Th3 is determined.

In that case, the control unit 350 sets the determination algorithm for determining the blood pressure surge, for example, as follows. In this example, the value of the "rise time" indicated by the blood pressure fluctuation of Yamagata to be determined is x1 [sec. ], The value of "descending time" is x2 [sec. ], The value of "fluctuation amount at the time of rising" is x3 [mmHg]. At this time, in the judgment algorithm as an example,
x1 ≧ 4.8 [sec. ],And,
x2 ≧ 3.2 [sec. ],And,
x3 ≧ 20 [mmHg]
When the condition (AND condition) is satisfied, the blood pressure fluctuation in Yamagata is determined to be a blood pressure surge. Otherwise (when even one inequality is not satisfied), the Yamagata blood pressure fluctuation is determined to be a non-blood pressure surge.

In this determination algorithm, the number of features used for determining the blood pressure surge can be limited to three. Therefore, the calculation load for blood pressure surge determination can be reduced as compared with the conventional example. Further, the feature amount categories (in the above example, the feature amount categories (I), (II), (III)) giving the first, second, and third feature amounts Fe1, Fe2, and Fe3 are different from each other. The quantity category. Therefore, the feature amount Fe of three categories can be included as an element of the determination algorithm for determining the blood pressure surge, and the blood pressure surge determination can be performed from the three viewpoints. Moreover, the set Cm1 of the first feature amount Fe1 and the first threshold value Th1 is an evaluation index (in the above example, the recall rate (in the above example, the recall rate) through all the sets Cm in the feature amount categories (I) to (VI) of the primary candidates. Recall). The same applies hereinafter.) This is a set Cm of a feature amount Fe and a threshold value Th that gives the best value B1. Further, the set Cm2 of the second feature amount Fe2 and the second threshold value Th2 is the feature amount Fe and the threshold value Th that give the best value B2 of the evaluation index through all the sets Cm in the feature amount category of the secondary candidate. It is a set Cm. Further, the set Cm3 of the third feature amount Fe3 and the third threshold value Th3 is the feature amount Fe and the threshold value Th that give the best value B3 of the evaluation index through all the sets Cm in the feature amount category of the tertiary candidate. It is a set Cm. As a result, the accuracy of blood pressure surge determination can be improved.

In this example, the control unit 350 stores the determination algorithm set in this way in the storage unit 340. In the following description, it is assumed that such a determination algorithm is stored in the storage unit 340 of the server 300.

In the method for generating the above-mentioned determination algorithm, as schematically shown in FIG. 16, the feature quantities are sequentially one by one from the feature quantity categories (I) to (VI) included in the feature quantity category group of the primary candidate. Select. Therefore, the blood pressure surge can be determined from a plurality of viewpoints. However, it is not always necessary to simply select a large number of features. For example, FIG. 17 shows a recall rate (Recall) when up to 6 feature quantities are sequentially selected one by one from each feature quantity category (I) to (VI) as an element of a determination algorithm for determining a blood pressure surge. The data of is shown. In this example, the recall rate was highest when the number of selected features was two or three. Even when only one feature was selected, the recall rate was not so inferior. On the other hand, when the number of selected features was 4, the recall rate (Recall) began to decrease slightly, and when the number of selected features was 5 or more, the recall rate (Recall) decreased significantly. The reason for this is that the method of generating the above-mentioned judgment algorithm is a kind of supervised learning, and increasing the number of features leads to overfitting (there is no generalization performance for evaluation data and newly acquired data). It is believed that there is.

In conclusion, considering the effect of reducing the calculation load of blood pressure surge determination, it is desirable to select one, two, or three features one by one from each feature category (I) to (VI). It can be said that.

(Method of determining blood pressure surge)
As a determination system, the system 100 determines that the Yamagata blood pressure fluctuation (the Yamagata blood pressure fluctuation to be determined) included in the blood pressure time series data 801 (see FIG. 6 (A)) is a blood pressure surge in response to the user's request. Whether or not there is is determined as follows.

Here, for a certain subject, measurement data including time-series data 801 of overnight blood pressure, SBP value / DBP value for each beat, pulse rate, output of motion sensor 220, etc. are already measured by the sphygmomanometer 200. It is assumed that the data is stored in the storage unit 340 of the server 300 via the network 900. The sphygmomanometer 200 may once transmit the measurement data to any hospital terminal 400, and the hospital terminal 400 may transmit the measurement data to the server 300.

First, as shown in step S101 of FIG. 18, the control unit 350 of the server 300 reads out the overnight blood pressure time series data 801 from the storage unit 340.

Next, as shown in step S102, the control unit 350 of the server 300 functions as a preprocessing unit, and uses a smoothing, noise removal, and low-pass filter using a moving average known for the blood pressure time series data 801. Pretreatment such as removal of high frequency components is performed.

Further, the control unit 350 of the server 300 extracts an effective section to be the target of detection / analysis of the blood pressure surge from the blood pressure time series data 801. That is, based on the output of the motion sensor 220, the section of the blood pressure time series data 801 that is affected by the disturbance (noise, etc.) is excluded as an invalid section, and the section that is not affected by the disturbance is set as the valid section. leave. As a result, the risk of erroneously detecting fluctuations due to noise as a blood pressure surge can be reduced. In addition, the processing time for detecting and analyzing blood pressure surges can be shortened.

The server 300 may preprocess the overnight blood pressure time series data 801 in advance in step S102, and store the preprocessed data in the storage unit 340. Then, when the user requests the blood pressure surge determination, the server 300 reads the preprocessed data (for simplicity, represented by the same code 801 as the original time series data ...) from the storage unit 340. You may. As a result, the processing time for detecting and analyzing the blood pressure surge can be shortened.

Next, as shown in steps S103 to S104, the control unit 350 of the server 300 acts as a data acquisition unit, and the Yamagata blood pressure fluctuation of the determination target included in the preprocessed data 801 (in this example, the systole period). Data representing the Yamagata blood pressure fluctuation of blood pressure (SBP)) is acquired. Specifically, the time and blood pressure value of the peak point P2 as shown in FIG. 9 are acquired for the preprocessed data 801 using the sliding window (step S103). Further, a search range is set before and after the peak point P2, and the time and blood pressure value of each of the start point P1 and the end point P4 are acquired (step S104). As a result, the blood pressure fluctuation of Yamagata to be determined is specified.

Next, as shown in step S105 of FIG. 18, the control unit 350 of the server 300 can function as a feature amount calculation unit to characterize the blood pressure surge waveform for each of the Yamagata blood pressure fluctuations to be determined. Of the plurality of feature quantities, in this example, at least the first, second and third feature quantities Fe1, Fe2 and Fe3 to be used by the determination algorithm are calculated.

In this example, the value of the "rise time" as the calculated first feature amount Fe1 is x1 [sec. ], The value of the "descending time" as the second feature amount Fe2 is x2 [sec. ], It is assumed that the value of the "variation amount during ascent" as the third feature amount Fe3 is x3 [mmHg].

In addition, for statistical processing of blood pressure surge, at this stage, the blood pressure fluctuation of the Yamagata to be determined is calculated together with the features other than the first, second and third features Fe1, Fe2 and Fe3. You may.

The control unit 350 of the server 300 reads out the determination algorithm stored in the storage unit 340. As described above, the determination algorithm has, as the elements of the determination algorithm for determining the blood pressure surge, the "rise time" as the first feature amount Fe1 and the threshold value of 4.8 [as the lower limit as the first threshold value Th1. sec. ] And the set Cm1. Further, the "descending time" as the second feature amount Fe2 and the threshold value 3.2 [sec. ] And the set Cm2. Further, the set Cm3 of the “rising fluctuation amount” as the third feature amount Fe3 and the threshold value 20 [mmHg] which is the lower limit as the third threshold value Th3 is included.

Next, as shown in step S106, the control unit 350 of the server 300 functions as a data acquisition unit, and the determination algorithm stored in the storage unit 340 is used to calculate the blood pressure fluctuation of the Yamagata to be determined. Based on the first, second, and third feature quantities Fe1, Fe2, and Fe3, it is determined whether or not the blood pressure fluctuation of the Yamagata to be determined is a blood pressure surge, respectively.

In this example, the determination algorithm has, as elements for determining the blood pressure surge, an "rise time" as the first feature amount Fe1 and a threshold value of 4.8 [sec. ] And the set Cm1 is included. Further, the "descending time" as the second feature amount Fe2 and the threshold value 3.2 [sec. ] And the set Cm2 is included. Further, it includes a set Cm3 of the "rising fluctuation amount" as the third feature amount Fe3 and the threshold value 20 [mmHg] which is the lower limit as the third threshold value Th3. In response to this, in this example, the control unit 350 sets the value of the "rise time" indicated by the blood pressure fluctuation of the Yamagata to be determined x1 [sec. ], The value of "descending time" is x2 [sec. ], Assuming that the value of "fluctuation amount when rising" is x3 [mmHg],
x1 ≧ 4.8 [sec. ],And,
x2 ≧ 3.2 [sec. ],And,
x3 ≧ 20 [mmHg]
When the condition (AND condition) is satisfied, the Yamagata blood pressure fluctuation is determined to be a blood pressure surge. Otherwise (when even one inequality is not satisfied), the control unit 350 determines that the chevron blood pressure fluctuation is a non-blood pressure surge.

After that, as shown in step S107, the control unit 350 of the server 300 stores the determination result of whether or not there is a blood pressure surge for each of the blood pressure fluctuations in Yamagata in the storage unit 340.

For statistical processing of blood pressure surges, for each of the above-mentioned Yamagata blood pressure fluctuations, feature quantities other than the first, second, and third feature quantities Fe1, Fe2, and Fe3 are also stored in the storage unit 340. You may memorize it.

The blood pressure surge determination result stored in the storage unit 340 of the server 300 is provided from the server 300 to the hospital terminal 400 via the network 900 in response to a request from the hospital terminal 400. As a result, the user can know the result of the blood pressure surge determination on the display screen of the display 420 of the hospital terminal 400. For example, on the display screen of the display 420, the number of occurrences of blood pressure surges included in the overnight blood pressure time series data 801 and the statistical values of the feature amounts (mean value calculated by the above statistical processing, variance, etc.) are displayed. You can know. Further, as shown in FIG. 6B, the waveform of the detected blood pressure surge (indicated by the rectangular frame 803, 803, ... With a broken line) can be confirmed.

As described above, according to this system 100, it is possible to determine whether or not the Yamagata blood pressure fluctuation included in the blood pressure time series data 801 is a blood pressure surge. Here, since the determination algorithm described above is used in this system 100, the calculation load of the blood pressure surge determination can be reduced and the accuracy of the blood pressure surge determination can be improved as compared with the conventional example.

Further, in the above-mentioned determination algorithm, the first, second, and third threshold values Th1, Th2, and Th3 can be adjusted (variably reset) independently of each other after the fact. For example, once a useful feature has been identified in a clinical trial, it is possible to make adjustments focusing on the threshold value for that feature.

(Modification example 1 of judgment algorithm)
The elements of the above-mentioned determination algorithm may be only the set Cm1 of the first feature amount Fe1 and the first threshold value Th1 and the set Cm2 of the second feature amount Fe2 and the second threshold value Th2. In that case, according to the above example, the determination algorithm sets the value of the "rise time" indicated by the blood pressure fluctuation of the Yamagata to be determined x1 [sec. ], The value of "descending time" is x2 [sec. ] When
x1 ≧ 4.8 [sec. ],And,
x2 ≧ 3.2 [sec. ]
When the condition (AND condition) is satisfied, the Yamagata blood pressure fluctuation is determined to be a blood pressure surge. Otherwise (when even one inequality is not satisfied), the control unit 350 determines that the chevron blood pressure fluctuation is a non-blood pressure surge. In this determination algorithm, only two features can be used for determining the blood pressure surge. Therefore, the calculation load for blood pressure surge determination can be further reduced as compared with the case where the number of features used for blood pressure surge determination is three.

(Modification example 2 of the judgment algorithm)
Further, the element of the above-mentioned determination algorithm may be only the set Cm1 of the first feature amount Fe1 and the first threshold value Th1. In that case, according to the above example, the determination algorithm sets the value of the "rise time" indicated by the blood pressure fluctuation of the Yamagata to be determined x1 [sec. ] When
x1 ≧ 4.8 [sec. ]
When the above condition is satisfied, the Yamagata blood pressure fluctuation is determined to be a blood pressure surge. Otherwise, the control unit 350 determines that the chevron blood pressure fluctuation is a non-blood pressure surge. In this determination algorithm, only one feature quantity can be used for determining the blood pressure surge. Therefore, the calculation load for blood pressure surge determination can be further reduced as compared with the case where the number of features used for blood pressure surge determination is two.

(Two examples of judgment algorithms that make up different types of classifiers)
In the determination algorithm of the above example, the set Cm1 of the first feature amount Fe1 and the first threshold value Th1, the set Cm2 of the second feature amount Fe2 and the second threshold value Th2, and the third feature amount Fe3 The blood pressure surge was determined under the three-element AND condition of the set Cm3 with the third threshold Th3. However, it is not limited to this. As the determination algorithm, there may be a configuration in which different types of classifiers are formed.

For example, as an example of such a determination algorithm, the first, second, and third feature quantities Fe1, Fe2, and Fe3 are set as explanatory variables x1, x2, x3, respectively, and the determination result of whether or not there is a blood pressure surge is determined. When the objective variable y is set and α1, α2, α3, β are used as coefficients,
y = α1, x1 + α2, x2 + α3, x3 + β ... (Eq.1)
It may be configured to determine whether or not the Yamagata blood pressure fluctuation included in the blood pressure time series data 801 is a blood pressure surge by linear discriminant analysis using the straight line represented by.

In this determination algorithm, for example, if y ≧ 0, the blood pressure fluctuation of the Yamagata to be determined is a blood pressure surge, while if y <0, the blood pressure fluctuation of the Yamagata to be determined is a non-blood pressure surge. Is judged to be. The coefficients α1, α2, α3, and β are naturally determined in the process of applying this determination algorithm to the Yamagata blood pressure fluctuation included in the teacher data Dte. That is, as shown in the flow of FIG. 19, first, the teacher data Dte is prepared, and the first, second, and third feature quantities are prepared according to the flow (method of generating the determination algorithm) shown in FIGS. 7A to 7C. Select Fe1, Fe2, and Fe3 (step S201 in FIG. 19). Then, the coefficients α1, α2, α3, β are determined by applying the linear discriminant analysis using the above straight line (Eq.1) to the blood pressure fluctuation of Yamagata included in the teacher data Dte (parameter adjustment). As a result, the straight line (Eq.1) is determined (step S202 in FIG. 19).

In this determination algorithm, the number of features used for determining the blood pressure surge can be limited to three. Therefore, the calculation load for blood pressure surge determination can be reduced as compared with the conventional example. Further, according to this determination algorithm, the accuracy of blood pressure surge determination can be improved.

Further, as another example of the determination algorithm forming a different type of classifier, the first, second, and third feature quantities Fe1, Fe2, and Fe3 are set as explanatory variables x1, x2, and x3, respectively, and whether or not the blood pressure surge occurs. When the probability of is y and α1, α2, α3, β are coefficients,
y = 1 / {-exp (α1 ・ x1 + α2 ・ x2 + α3 ・ x3 + β)}
… (Eq.2)
It may be configured to determine whether or not the Yamagata blood pressure fluctuation included in the time series data of blood pressure is a blood pressure surge by logistic regression analysis using the formula represented by.

In this determination algorithm, for example, when a certain threshold th is set, if y ≧ th, the blood pressure fluctuation of the Yamagata to be determined is a blood pressure surge, while if y <th, the blood pressure fluctuation of the Yamagata to be determined is a blood pressure fluctuation. Is determined to be a non-blood pressure surge. The coefficients α1, α2, α3, and β are naturally determined in the process of applying this determination algorithm to the Yamagata blood pressure fluctuation included in the teacher data Dte, as in the above example.

In this determination algorithm, as in the above example, the number of features used for determining the blood pressure surge can be limited to only three. Therefore, the calculation load for blood pressure surge determination can be reduced as compared with the conventional example. Further, according to this determination algorithm, the accuracy of blood pressure surge determination can be improved.

(Other variant 1)
In the above example, in step S11 of FIG. 7A, the feature amount category group of the primary candidate includes six feature amount categories (I) to (VI), but the present invention is not limited to this. For example, the feature category (I) related to the reaction phase (Reactivity Phase) is integrated into the feature category (IV) related to the ascending (Upward), and the feature category (II) related to the recovery phase (Recovery Phase) is related to the descending (Downward). It may be integrated into the feature category (V), and the feature category (VI) for the whole duration may be abolished. In that case, the feature category groups of the primary candidates are the feature category (III) related to Amplitude, the feature category (IV) related to rising (Upward), and the feature category (V) related to descending (Downward). It includes three feature categories of. Also in this case, the set Cm1 of the first feature amount Fe1 and the first threshold value Th1, the set Cm2 of the second feature amount Fe2 and the second threshold value Th2, and the third feature amount Fe3 and the third feature amount Fe3. The set Cm3 with the threshold Th3 can be determined, and the determination algorithm can be appropriately generated. Further, by using the generated determination algorithm, the accuracy of blood pressure surge determination can be improved.

(Other variant 2)
In the above example, in step S12 of FIG. 7A, when creating the set Cm of the feature amount Fe and the threshold value Th, either or both of the upper limit threshold value ThU and the lower limit threshold value ThL are immediately set. However, it is not limited to this. After confirming the frequency distribution of the value x of each feature amount indicated by the blood pressure fluctuations of a plurality of chevrons included in the teacher data Dte (see, for example, FIG. 13), the threshold ThU is the upper limit ThU or the lower limit. The threshold value ThL may be determined manually or automatically so that, for example, the frequency of being classified as a blood pressure surge increases.

(Other variant 3)
In the above example, the control unit 350 of the server 300 generated the determination algorithm, but the present invention is not limited to this. Instead of the server 300, the determination algorithm may be generated by a general personal computer. In that case, it is desirable to store the generated determination algorithm in the storage unit 340 of the server 300 via, for example, the network 900. In such a case, the system 100 described above determines whether or not the Yamagata blood pressure fluctuation (the Yamagata blood pressure fluctuation to be determined) included in the blood pressure time series data 801 is a blood pressure surge, according to the user's request. Can be determined at any time.

(Other variant 4)
Further, the above-mentioned determination method is recorded as software (computer program) on a recording medium such as a CD (compact disc), a DVD (digital universal disc), or a flash memory that can store data non-transitory. You may. By installing the software recorded on such a recording medium on a substantial computer device such as a personal computer, a PDA (Personal Digital Assistance), or a smartphone, the above-mentioned determination method is executed on those computer devices. Can be made to.

(Other variant 5)
Further, in the above-described embodiment, the sphygmomanometer 200 is a sphygmomanometer of the tonometry type, but the present invention is not limited to this. The sphygmomanometer 200 includes a light emitting element that irradiates light toward an artery passing through a corresponding portion of the measurement site, and a light receiving element that receives reflected light (or transmitted light) of the light, and has an arterial pulse. Blood pressure may be detected continuously based on the change in volume of the wave (photoelectric method). Further, the sphygmomanometer 200 is provided with a piezoelectric sensor in contact with the part to be measured, detects strain due to pressure of an artery passing through the corresponding part of the part to be measured as a change in electrical resistance, and changes in this electrical resistance. Blood pressure may be continuously detected based on (piezoelectric method). Further, the blood pressure monitor 200 includes a transmitting element that sends a radio wave (transmitted wave) toward an artery passing through a corresponding portion of the measured portion, and a receiving element that receives a reflected wave of the radio wave, and has an arterial pulse. The change in the distance between the artery and the sensor due to the wave may be detected as the phase shift between the transmitted wave and the reflected wave, and the blood pressure may be continuously detected based on this phase shift (radio wave irradiation method). ). Further, if a physical quantity capable of calculating blood pressure can be observed, a method other than these may be applied.

The above embodiment is an example, and various modifications can be made without departing from the scope of the present invention. The plurality of embodiments described above can be established independently, but combinations of the embodiments are also possible. Further, although various features in different embodiments can be established independently, it is also possible to combine features in different embodiments.

100 System 200 Sphygmomanometer 260,350 Control unit 300 Server 320,420 Display 340 Storage unit 400, 400A, 400B, ... Hospital terminal

Claims (13)

It is a method of generating a judgment algorithm for judging whether or not the Yamagata blood pressure fluctuation included in the time series data of blood pressure is a blood pressure surge.
A plurality of features capable of characterizing the waveform of the blood pressure surge are classified, and the feature amount of the primary candidate including at least the feature amount category related to amplitude, the feature amount category related to ascending, and the feature amount category related to descending. Define categories
For each of the feature amount categories, a set of the feature amount and the threshold value is formed by combining a plurality of feature amounts included in the feature amount category and a threshold value of one or more steps for each of the plurality of feature amount amounts. Create multiple sets and
Multiple Yamagata blood pressure fluctuations derived from blood pressure time-series data, for which it is known in advance whether or not each is a blood pressure surge, are prepared as teacher data.
Targeting the blood pressure fluctuations of the plurality of Yamagata included in the teacher data, the feature amounts and thresholds of all the sets in the feature amount category group of the primary candidates are used as a classifier one by one to generate a blood pressure surge. Judge whether it is classified or not,
Regarding the judgment result as a classifier for each of the above sets, an evaluation index was obtained for each set of the above-mentioned primary candidates in the feature amount category group.
Among the evaluation indexes obtained over all the sets in the feature amount category group of the primary candidate, the set of the feature amount and the threshold value giving the best value is the first feature amount which is an element of the determination algorithm. A method of generating a determination algorithm, characterized in that it is determined as a set of a first threshold value and a first threshold value. In the method of generating the determination algorithm according to claim 1,
Among the plurality of Yamagata blood pressure fluctuations included in the teacher data, the data on the Yamagata blood pressure fluctuation classified as a blood pressure surge according to the combination of the first feature amount and the first threshold value is intermediate data. Specified as
From the feature amount category group of the primary candidate, the feature amount category given the first feature amount is excluded, and all the remaining feature amount categories are specified as the feature amount category group of the secondary candidate.
Targeting the blood pressure fluctuations of a plurality of chevrons included in the intermediate data, the feature amounts and threshold values of all the sets in the feature amount category group of the secondary candidates are classified into blood pressure surges one by one as a classifier. Judge whether or not it will be done,
Regarding the judgment result as a classifier for each of the above sets, an evaluation index was obtained for each of the above sets of the feature amount category group of the secondary candidates.
Among the evaluation indexes obtained over all the sets in the feature amount category group of the secondary candidate, the set of the feature amount and the threshold value giving the best value is the second feature amount which is an element of the determination algorithm. A method of generating a determination algorithm, characterized in that it is determined as a set of a second threshold value and a second threshold value. In the method for generating the determination algorithm according to claim 2,
Of the plurality of Yamagata blood pressure fluctuations included in the intermediate data, the data on the Yamagata blood pressure fluctuation classified as a blood pressure surge according to the combination of the second feature amount and the second threshold value is used as the final data. Identify and
From the feature amount category group of the secondary candidate, the feature amount category given the second feature amount is excluded, and all the remaining feature amount categories are specified as the feature amount category group of the tertiary candidate.
Targeting multiple Yamagata blood pressure fluctuations included in the final data, all sets of features and thresholds in the feature category group of the tertiary candidates are classified into blood pressure surges by using one set as a classifier. Judge whether or not it will be done,
Regarding the judgment result as the classifier for each of the above sets, the evaluation index was obtained for each set in the feature amount category group of the above tertiary candidates.
Among the evaluation indexes obtained over all the sets in the feature amount category group of the tertiary candidate, the set of the feature amount and the threshold value giving the best value is the third feature amount which is an element of the determination algorithm. A method of generating a determination algorithm, characterized in that it is determined as a set of a third threshold value and a third threshold value. In the method for generating the determination algorithm according to any one of claims 1 to 3,
When creating a plurality of sets of the feature amount and the threshold value for each of the feature amount categories, the threshold values to be combined with the feature amount are set in a plurality of steps in a certain range over a certain range.
After determining the above-mentioned sets that form the elements of the above-mentioned determination algorithm, a search range narrower than the above-mentioned range including the threshold value is set for the threshold values forming the above-determined sets, and the search range is finer than the above-mentioned step. A method of generating a determination algorithm, which comprises performing a grid search in increments and adjusting each of the above thresholds. A determination algorithm using the first feature amount determined by the method for generating the determination algorithm according to claim 1 or 4.
By comparing the value of the first feature amount indicated by the blood pressure fluctuation of the judgment target Yamagata included in the time series data of the blood pressure with the first threshold value, the blood pressure fluctuation of the judgment target Yamagata can be obtained. A determination algorithm characterized by determining whether or not there is a blood pressure surge. A determination algorithm using the first and second feature quantities determined by the method for generating the determination algorithm according to claim 2 or 4.
The value of the first feature amount indicated by the blood pressure fluctuation of Yamagata to be determined included in the time series data of the blood pressure is compared with the first threshold value, and the blood pressure fluctuation of Yamagata to be determined is A determination algorithm characterized in that it is determined whether or not the blood pressure fluctuation of the Yamagata to be determined is a blood pressure surge by comparing the value of the second feature amount shown with the second threshold value. .. A determination algorithm using the first, second, and third features determined by the method for generating the determination algorithm according to claim 3 or 4.
The value of the first feature amount indicated by the blood pressure fluctuation of Yamagata to be determined included in the time-series data of the blood pressure is compared with the first threshold value, and the blood pressure fluctuation of Yamagata to be determined is The value of the second feature amount shown is compared with the second threshold value, and the value of the third feature amount indicated by the blood pressure fluctuation of Yamagata to be determined is compared with the third threshold value. A determination algorithm characterized by determining whether or not the blood pressure fluctuation in Yamagata, which is the object of determination, is a blood pressure surge by comparison. A determination algorithm using the first, second, and third feature quantities determined by the method for generating the determination algorithm according to claim 3.
The first, second, and third feature quantities were set as explanatory variables x1, x2, x3, respectively, the judgment result of whether or not the blood pressure was surge was set as the objective variable y, and α1, α2, α3, and β were used as coefficients. When
y = α1, x1 + α2, x2 + α3, x3 + β ... (Eq.1)
A determination algorithm characterized by determining whether or not the Yamagata blood pressure fluctuation included in the time series data of blood pressure is a blood pressure surge by linear discriminant analysis using a straight line represented by. A determination algorithm using the first, second, and third feature quantities determined by the method for generating the determination algorithm according to claim 3.
When the first, second, and third features are the explanatory variables x1, x2, x3, the probability of whether or not there is a blood pressure surge is y, and α1, α2, α3, β are coefficients.
y = 1 / {-exp (α1 ・ x1 + α2 ・ x2 + α3 ・ x3 + β)}
… (Eq.2)
A determination algorithm characterized by determining whether or not the Yamagata blood pressure fluctuation included in the time series data of blood pressure is a blood pressure surge by logistic regression analysis using the formula represented by. It is a judgment system that determines whether or not the Yamagata blood pressure fluctuation included in the blood pressure time series data is a blood pressure surge.
A storage unit that stores the determination algorithm according to any one of claims 5 to 9.
A data acquisition unit that acquires data representing the blood pressure fluctuation of the judgment target Yamagata included in the blood pressure time series data,
A feature amount calculation unit that calculates the feature amount to be used by the judgment algorithm for the blood pressure fluctuation of Yamagata, which is the judgment target, and a feature amount calculation unit.
It is characterized by including a surge determination unit that determines whether or not the blood pressure fluctuation of the Yamagata to be determined is a blood pressure surge based on the calculated feature amount using the stored determination algorithm. Judgment system. It is a judgment method to judge whether or not the blood pressure fluctuation of Yamagata included in the time series data of blood pressure is a blood pressure surge.
A storage unit that stores the determination algorithm according to any one of claims 5 to 9 is provided.
Acquire the data representing the blood pressure fluctuation of the judgment target Yamagata included in the time series data of blood pressure,
For the blood pressure fluctuation of Yamagata to be judged, the feature amount to be used by the judgment algorithm is calculated.
A determination method characterized by determining whether or not the blood pressure fluctuation in Yamagata, which is the determination target, is a blood pressure surge, based on the calculated feature amount, using the determination algorithm stored above. A program for causing a computer to execute the determination method according to claim 11. A computer-readable recording medium that stores the program according to claim 12.

Images