Dementia determination device, dementia determination method, and program

Abstract

The present invention comprises an acquisition unit that acquires information about the heart rate of a subject, a feature quantity estimation unit that estimates a time-series feature quantity for heart rate from the information about the heart rate acquired by the acquisition unit, a determination unit that detects decrease in homeostasis of heartbeat on the basis of the time-series feature quantity for heart rate estimated by the feature quantity estimation unit and thereby determines whether the subject has dementia or whether the subject has a high likelihood of developing dementia, and an outputting unit that outputs the results of the determination made by the determination unit.

Description

Dementia assessment device, dementia assessment method, and program

[0001] The present invention relates to a dementia assessment device, a dementia assessment method, and a program for assessing dementia by acquiring a heart rate.

[0002] The number of dementia patients has been increasing in recent years. Approximately half of dementia cases are Alzheimer's disease. No complete prevention method or fundamental treatment has been found for Alzheimer's disease, and since it takes approximately 10 to 20 years for the initial symptoms to appear, by the time Alzheimer's disease is discovered, it is often difficult to treat. For this reason, early detection and early response to Alzheimer's disease have become important issues.

[0003] The gold standard for diagnosing Alzheimer's disease is the Mini-Mental State Examination (MMSE), which is conducted when dementia is suspected. In this assessment method, the subject answers questions posed by the interviewer one by one about the test items. The test takes approximately 10 to 15 minutes.

[0004] However, oral question-based screening tests such as the MMSE place a psychological burden on test subjects, making regular testing difficult. Furthermore, in the early stages of dementia, when subjects behave in a manner similar to that of healthy individuals, it is difficult to find an opportunity to administer the MMSE. Therefore, attempts have been made to diagnose dementia based on sleep duration and wake-up frequency, which are factors found in daily life, but it is difficult to definitively distinguish between elderly people with Alzheimer's disease and those without Alzheimer's disease.

[0005] To solve these problems, the inventors of the present application have developed a method for estimating the circadian rhythm of a subject using their heart rate and distinguishing between Alzheimer's disease and healthy elderly people based on the estimated circadian rhythm (see Patent Document 1).

[0006] International Publication No. 2022 / 186275

[0007] According to the method described in Patent Document 1, it is possible to accurately diagnose dementia using the predicted amplitude ratio of sine waves and cosine waves of circadian rhythms and non-circadian rhythms, but there has been a demand for even higher accuracy in dementia diagnosis. In particular, conventional methods have the problem that the accuracy of circadian rhythms varies depending on the length of sleep time, so the accuracy varies from day to day, and the accuracy decreases as the sleep time becomes shorter.

[0008] The present invention aims to provide a dementia assessment device, a dementia assessment method, and a program that are capable of highly accurate dementia assessment regardless of the length of sleep time.

[0009] The dementia assessment device of the present invention comprises an acquisition unit that acquires information regarding the heart rate of a subject, a feature estimation unit that estimates time-series features of the heart rate from the information regarding the heart rate acquired by the acquisition unit, a determination unit that determines whether the subject has dementia or is in a state in which the subject is likely to develop dementia by detecting a decrease in the heart rate constancy based on the time-series features of the heart rate estimated by the feature estimation unit, and an output unit that outputs the result of the determination by the determination unit.

[0010] In addition, the dementia diagnosis method of the present invention includes an acquisition process for acquiring information regarding the heart rate of a subject, a feature estimation process for estimating time-series features of the heart rate from the information regarding the heart rate acquired by the acquisition process, and a determination process for determining whether the subject has dementia or whether the subject is in a state where there is a high possibility of developing dementia by detecting a decrease in the homeostasis of the heart rate based on the time-series features of the heart rate estimated by the feature estimation process.

[0011] The program of the present invention causes a computer to execute the steps of the dementia assessment method.

[0012] According to the present invention, dementia can be diagnosed with high accuracy based on the heart rate.

[0013] FIG. 3 is a configuration diagram showing an example of a dementia diagnosis device according to an embodiment of the present invention. FIG. 3A is a diagram showing an example of a circadian rhythm of a heart rate estimated from heart rate information according to an embodiment of the present invention. FIG. 3B is a diagram showing an example of a time-series feature of a heart rate according to an embodiment of the present invention. FIG. 3B is a flowchart showing an example of a diagnosis process using time-series feature estimation of a heart rate according to an embodiment of the present invention. FIG. 3C is a diagram showing an example of a time-series feature estimation process of a heart rate and a calculation process of a time-series feature according to an embodiment of the present invention. FIG. 3D is a diagram showing an example of a time-series feature estimation process of a heart rate according to an embodiment of the present invention. FIG. 3E is a diagram showing an example of a time-series feature calculation process of a heart rate according to an embodiment of the present invention. FIG. 3F is a diagram showing an example of a time-series feature estimation process of a heart rate according to an embodiment of the present invention. FIG. 3G is a diagram showing an example of a time-series feature estimation process of a heart rate according to an embodiment of the present invention.

[0014] Hereinafter, a dementia assessment device, a dementia assessment method, and a program according to one embodiment of the present invention (hereinafter referred to as "the present embodiment") will be described with reference to the accompanying drawings.

[0015] 1 shows the configuration of a dementia assessment device 1 of this example. The dementia assessment device 1 of this example includes an acquisition unit 11, a first estimation unit 13, a second estimation unit 14, a determination unit 15, a storage unit 16, and an output unit 17. Note that the dementia assessment device 1 of this example can be configured as a computer including, for example, a CPU and memory that perform calculations based on a program, and each processing unit such as the first estimation unit 13, the second estimation unit 14, and the determination unit 15 is a functional processing unit configured on the memory by executing a program.

[0016] Explaining the configuration shown in FIG. 1 , a biological information measuring device 2 is connected to the dementia assessment device 1 during measurement. The biological information measuring device 2 is, for example, a mat-type sensor, a smart watch equipped with a sensor that detects biological information, or a wearable terminal equipped with a sensor that detects biological information. The biological information measuring device 2 measures data related to at least the subject's heart rate while sleeping, and outputs the measured data to the dementia assessment device 1. In the following explanation, an example will be described in which the biological information measuring device 2 is, for example, a mat-type sensor. In the case of a mat-type sensor, the data related to the heart rate is biological vibration data.

[0017] The dementia assessment device 1 and the biological information measurement device 2 are connected via a wired or wireless connection. The biological information measurement device 2 includes a communication unit or interface for data output to the dementia assessment device 1. The acquisition unit 11 performs an acquisition process to acquire data related to heart rate (hereinafter referred to as "heart rate data") from the biological information measurement device 2. The first estimation unit (circadian rhythm estimation unit) 13 performs a process to estimate the circadian rhythm of the heart rate (hereinafter referred to as "circadian rhythm") from the heart rate data. One method for estimating this circadian rhythm is based on the method previously proposed by the inventor of the present application in Patent Document 1. However, while the method previously proposed in Patent Document 1 estimated the circadian rhythm and non-circadian rhythms, other methods include a method that determines the circadian rhythm alone (e.g., a method using the sine and cosine coefficients of the circadian rhythm) and a method that uses the instability of the circadian rhythm and the presence or absence of a rhythm. Since the assessment process using the circadian rhythm, such as the above, has already been proposed in Patent Document 1, details will be omitted in this specification.

[0018] The second estimation unit (feature estimation unit) 14 performs feature estimation processing to estimate time-series feature values ​​of the heart rate from the heart rate data. Details of the processing to estimate time-series feature values ​​of the heart rate will be described later. The determination unit 15 determines whether or not the subject has dementia, or whether or not the subject is in a state where there is a high possibility of developing dementia, based on at least one or both of the estimation results from the first estimation unit 13 and the second estimation unit 14. In the following description, the determination made by the determination unit 15 is described as determining whether or not the subject has dementia, but this determination also includes a case where the subject is determined to be in a state where there is a high possibility of developing dementia. Details of the determination processing by the determination unit 15 will also be described later.

[0019] The storage unit 16 stores the mathematical formulas, thresholds, etc. used by the determination unit 15. The storage unit 16 may also store data related to the heart rate during sleep. The output unit 17 outputs the determination result obtained by the determination unit 15 to an external device. Examples of the external device include an image display device, a printing device, a personal computer, a tablet terminal, a smartphone, and a dedicated terminal. The output unit 17 includes a communication unit or an interface.

[0020] Fig. 2 shows an example configuration of the biological information measuring device 2. As shown in Fig. 2, a mat-type pressure sensor is installed on the bed on which the subject a sleeps, and the biological information measuring device 2 detects information containing a heart rate component from pressure changes detected by the mat-type pressure sensor, thereby obtaining information related to the heart rate. The mat-type pressure sensor is installed, for example, under the bed mattress, in a position corresponding to the upper body of the subject a. The information related to the heart rate obtained by the biological information measuring device 2 is transmitted to the dementia assessment device 1.

[0021] [Outline of Two Determination Methods] FIG. 3 shows an outline of the circadian rhythm of the heart rate estimated by the first estimation unit 13 ( FIG. 3A ) and the time-series feature of the heart rate estimated by the second estimation unit 14 ( FIG. 3B ). The graphs in FIGS. 3A and 3B show the fluctuations in the heart rate per minute of a certain subject over a three-day period. As shown by the line fluctuating with a period of approximately 24 hours in FIG. 3A , human heart rates typically have a circadian rhythm with a period of approximately 24 hours. The first estimation unit 13 detects this circadian rhythm of the heart rate. The determination unit 15 determines whether or not the subject has dementia based on the results of the circadian rhythm determination. This determination using the circadian rhythm is made by viewing the heart rate from a macroscopic perspective. Processes for determining dementia based on the circadian rhythm include a method using the sine and cosine coefficients of the circadian rhythm, a method using the instability of the circadian rhythm and the presence or absence of a rhythm, and a determination method using not only the circadian rhythm but also a non-circadian rhythm that deviates from the circadian rhythm.

[0022] Meanwhile, the second estimation unit 14 extracts heart rates at relatively short intervals (e.g., one minute) over a predetermined time window (e.g., a 10-minute time window), detects the time-series feature values ​​shown in FIG. 3B , and performs processing to obtain information necessary for dementia assessment from a microscopic perspective from the time-series feature values. The assessment unit 15 then assesses whether or not dementia exists from a microscopic perspective based on the time-series feature values. This process of assessing whether or not dementia exists from a microscopic perspective based on the time-series feature values ​​corresponds to detecting a decrease in heart rate constancy based on the time-series feature values ​​estimated by the second estimation unit 14, and assesses whether or not dementia exists when a decrease in heart rate constancy is detected. Below, the assessment process by the first estimation unit 13 using circadian rhythms and the assessment process by the second estimation unit 14 using time-series feature values ​​will be described in order.

[0023] [Dementia Assessment Process Using Time-Series Features] Next, the assessment process by estimation using time-series features performed by the second estimation unit 14 will be described. FIG. 4 is a flowchart showing the procedure for dementia assessment using time-series features performed by the dementia assessment device 1. First, the acquisition unit 11 acquires heart rate data during sleep, and the second estimation unit 14 extracts the heart rate over a predetermined time window (step S11). Here, for example, extraction is performed every minute over a 10-minute time window. Note that the heart rate data during sleep acquired in step S11 is data measured over a relatively long period of time, such as data from at least one night's sleep. Then, the second estimation unit 14 divides the extracted heart rate into sections (step S12). Here, the division into sections is performed using, for example, a correlation coefficient with the Benford distribution.

[0024] Thereafter, the second estimation unit 14 calculates time-series features for each divided section (step S13). The determination unit 15 then determines whether each divided section corresponds to dementia (AD) or healthy individuals (step S14). After performing the determination for each section in step S14 for all sections, the determination unit 15 acquires the proportion of all sections that are determined to be dementia (step S15). The determination unit 15 then determines whether the proportion of sections that are determined to be dementia in step S15 is equal to or greater than a preset threshold (step S16).

[0025] If the proportion of people judged to have dementia in step S16 is equal to or greater than the threshold (YES in step S16), the determination unit 15 determines that the subject has dementia (AD) (step S17). On the other hand, if the proportion of people judged to have dementia in step S16 is not equal to or greater than the threshold (NO in step S16), the determination unit 15 determines that the subject is a healthy person (step S18). After the processes of steps S17 and S18, the determination unit 15 ends the process.

[0026] [Processing by the Second Estimation Unit] Next, the estimation process using the time-series feature values ​​will be described in detail. Fig. 5 shows an overview of the determination process by estimation using the time-series feature values. First, as shown in the upper part of Fig. 5, the second estimation unit 14 acquires heart rates over a relatively long period of time, such as data from a single sleep session, and extracts the heart rates over a time window of, for example, 10 minutes (600 seconds), shifting the time window by one minute.

[0027] Furthermore, the second estimation unit 14 divides the heart rate for each time window into sections using the correlation coefficient xbec(x) with the Benford distribution, as shown in the second row from the top of Figure 5. Here, the section division is not limited to dividing the heart rate into multiple sections in the time direction, but also includes multiple division methods, as described below, which are used to divide the heart rate into multiple sections in the time direction. The reason for providing multiple division methods is that heart rate fluctuations vary from person to person, which in turn results in different correlations with the Benford distribution, so that dementia can be accurately diagnosed even in various heart rate fluctuation states. However, providing multiple division methods is merely an example, and only one division method may be used.

[0028] 5, the second estimator 14 calculates a time-series feature amount for each divided section using the lag-2 partial autocorrelation coefficient xpac(x) and the number of peaks xncp(x) obtained by CWT (Continuous Wavelet Transform).

[0029] Next, the details of the heart rate interval segmentation process performed by the second estimator 14 will be described. FIG. 6 shows an overview of the heart rate interval segmentation process. Here, the correlation between the distribution of the first digit of the heart rate value (the first digit of the heart rate value per minute) and the Benford distribution is utilized. The Benford distribution (Benford's law) states that the distribution of the first digit (1 to 9) of many numbers in the natural world follows a monotonically decreasing distribution of log10(1 + 1 / 1) to log10(1 + 1 / 9).

[0030] In this example, when the heart rate data shown in the upper part of FIG. 6 is obtained, the second estimation unit 14 performs multiple methods of cutting out multiple heart rate intervals based on the correlation coefficient xbec(x) with the Benford distribution, as shown in the lower part of FIG. 6. Specifically, the second estimation unit 14 divides the area into regions that satisfy the correlation coefficient xbec(x) > (max(xbec) - min(xbec)) × α%. Here, the conditional expression α for cutting out is cut out in four ways: 90%, 70%, 50%, and 30%, and these are used together. In the graph shown in the lower part of FIG. 6, the areas with dots in the background are the cut-out intervals. As shown in FIG. 6, not only the intervals in the time axis direction but also the values ​​of α (the vertical axis of the graph in FIG. 6) are cut out in 90%, 70%, 50%, and 30% ways. The values ​​of 90%, 70%, 50%, and 30% are merely examples, and the present invention is not limited to these values. Also, cutting out using a plurality of methods is merely an example, and the division may be performed using only one method.

[0031] Next, the second estimator 14 calculates time-series features for the heart rate within the cut-out section. In this example, the second estimator 14 uses the lag-2 partial autocorrelation coefficient xpac(x). That is, the lag-2 partial autocorrelation coefficient xpac(x) is obtained from the following equation (1), which is the correlation when the influence of the heart rate HR(t-1) at lag 1 (time t-1) is removed from the heart rates HR(t) and HR(t-2) at a certain time t [seconds] and time t-2 (this is lag 2). In the case of sampling at one-second intervals, the correlation becomes greater when the lag is set to two for a signal with a two-second period.

[0032]

[0033] Furthermore, the number of peaks xncp(x) is obtained by continuous wavelet transform (CWT). FIG. 7 shows an example of obtaining the number of heart rate peaks by continuous wavelet transform (CWT). The graph in FIG. 7 shows an example of heart rates after continuous wavelet transform. In FIG. 7, the points indicated by white circles are heart rate peaks, and the number of peaks xncp(x) is obtained. Note that the use of lag-2 partial autocorrelation takes advantage of the knowledge gained from experiments conducted by the inventors of the present application and verification of the experimental results, which is that the heart rates of patients with dementia often fluctuate up and down over a period of two seconds or more (decreased heart rate constancy). The knowledge gained from experiments conducted by the inventors of the present application and verification of the experimental results also shows that this tendency is not observed in healthy individuals, and that because heart rate constancy is maintained (heart rate is kept constant), any fluctuations quickly return to the original value.

[0034] [Determination Process in Determination Unit for Data Obtained Using Time-Series Feature Amounts] Next, the determination process in the determination unit 15 for data obtained using time-series feature amounts in this manner will be described. FIG. 8 shows the determination process in the determination unit 15. First, the determination unit 15 obtains the average value for each interval of the lag-2 partial autocorrelation coefficient xpac(x) and the average value for each interval of the number of heart rate peaks xncp(x), as described above. Then, the determination unit 15 determines whether or not dementia exists for each interval using random forest, a machine learning technique, based on the average value for each interval of the lag-2 partial autocorrelation coefficient xpac(x) and the average value for each interval of the number of heart rate peaks xncp(x). Random forest is a machine learning algorithm that combines decision trees and ensemble learning, and determines whether or not dementia exists for each interval by comparing it with similar learned data.

[0035] The bottom part of Figure 8 shows the results of using random forest to determine whether each cutout section is dementia or healthy. The cutout sections marked with a circle are sections that the random forest determined to be likely dementia, and the cutout sections marked with an x ​​are sections that the random forest determined to be likely healthy.

[0036] Here, the determination unit 15 determines whether the ratio of sections determined to have a high possibility of dementia among all cutout sections exceeds a preset threshold. For example, suppose there are 15 cutout sections, and 8 of these sections are determined to have a high possibility of dementia. In this case, the ratio is 8 / 15 = 0.53. If the threshold is, for example, 0.8, the ratio of 0.53 is below the threshold, and the determination unit 15 determines that the subject is healthy.

[0037] Figure 11 shows an overview of the random forest decision process using two data: the average value of the lag-2 partial autocorrelation coefficient xpac(x) for each interval and the average value of the heart rate peak count xncp(x) for each interval. Using these two data, the average value of the lag-2 partial autocorrelation coefficient xpac(x) for each interval and the average value of the heart rate peak count xncp(x) for each interval, enables the two time series features to complement each other. For example, as shown in Figure 11, at the top of the trained random forest (RF) decision tree, AD and healthy subjects are separated by the average value of the lag-2 partial autocorrelation coefficient xpac(x) for each interval. While the right (without vibration for 2 seconds) is 100% classified as healthy subjects, the left (with vibration for 2 seconds) contains a mixture of AD patients and healthy subjects. Therefore, in the second row, AD and healthy subjects are distinguished by the average value of the heart rate peak count xncp(x) for each interval, with the left (when the peak count is small) representing AD patients and the right (when the peak count is large) representing healthy subjects. In other words, the misidentification based on the average value of the lag-2 partial autocorrelation coefficient xpac(x) for each interval is corrected using the average value of the lag-2 partial autocorrelation coefficient xpac(x). Note that healthy subjects were identified as having AD using the average value of the lag-2 partial autocorrelation coefficient xpac(x) for each interval because, although rare, some healthy subjects experience 2-second oscillations (the area surrounded by the ellipse in the waveform diagram on the left of Figure 11).

[0038] In the explanation so far, dementia is estimated using time-series features using both the average value for each interval of the lag-2 partial autocorrelation coefficient xpac(x) and the average value for each interval of the number of heart rate peaks xncp(x). However, it is also possible to determine whether or not dementia exists using only one of the average value for each interval of the lag-2 partial autocorrelation coefficient xpac(x) and the average value for each interval of the number of heart rate peaks xncp(x).

[0039] [Process in which the determination unit performs determination using estimation based on circadian rhythm and estimation based on time-series feature amount] The explanation so far has been given of an example in which the determination unit 15 performs determination processing based on the estimation result based on circadian rhythm, and an example in which the determination unit 15 performs determination processing based on the estimation result based on time-series feature amount (processing shown in the flowchart in FIG. 4). Alternatively, the determination unit 15 may perform determination by combining these two processes.

[0040] 9 illustrates an example in which the determination unit 15 makes a determination based on the results of estimation by the first estimation unit 13 and the results of estimation by the second estimation unit 14. To explain the process illustrated in the flowchart of Fig. 9, first, the second estimation unit 14 extracts heart rates within a predetermined time window (step S11), separates the extracted heart rates into intervals (step S12), and calculates time-series feature amounts for each of the separated intervals (step S13), in the same manner as in the example of the flowchart of Fig. 4.

[0041] The assessment unit 15 then performs dementia assessment for each cutout section using a random forest (step S14'). During dementia assessment, the random forest learns using the average values ​​for each section of the lag-2 partial autocorrelation coefficient xpac(x), the average values ​​for each section of the number of heart rate peaks xncp(x), and the estimated value of the circadian rhythm. The assessment unit 15 then performs dementia assessment using the random forest using the average values ​​for each section of the lag-2 partial autocorrelation coefficient xpac(x), the average values ​​for each section of the number of heart rate peaks xncp(x), and the estimated value of the circadian rhythm. As already explained, there are various assessment methods using circadian rhythm, and values ​​based on the method applied in this example are used. Furthermore, when using the random forest here, using both the average values ​​for each section of the lag-2 partial autocorrelation coefficient xpac(x) and the average values ​​for each section of the number of heart rate peaks xncp(x) is merely an example, and either value may be used.

[0042] The determination unit 15 then determines the proportion of dementia in the entire section based on the determination results using the random forest for each cutout section (step S15') and determines whether the proportion is equal to or greater than a threshold (step S16'). Here, the determination unit 15 determines that the subject has dementia if the proportion is equal to or greater than the threshold (step S17'). This determination may also include whether or not there is a high probability of developing dementia. Furthermore, the determination unit 15 determines that the subject does not have dementia if the proportion is not equal to or greater than the threshold (step S18').

[0043] [Experimental Examples of Accuracy of Each Determination Process] Figure 10 shows the accuracy of AD determination by each determination process of this example. Here, determinations were made on a large number of subjects who were previously known to be dementia patients or healthy elderly people, and the accuracy, recall, precision, specificity, and F-value were calculated. Here, accuracy is the proportion of correct determinations for all subjects, recall is the proportion of correct determinations for AD patients, precision is the proportion of correct determinations when determining that someone has AD, specificity is the proportion of correct determinations for healthy people, and F-value is the harmonic mean of recall and precision.

[0044] The circadian rhythm AD determination method shown in the first row of Fig. 10 is a case where determination is made from the results estimated based on the circadian rhythm by the first estimator 13. The heartbeat time series feature AD determination method shown in the second row of Fig. 10 is a case where determination is made from the results estimated from the time series feature by the second estimator 14. The combined determination method shown in the third row of Fig. 10 is a case where the results estimated by the first estimator 13 and the results estimated by the second estimator 14 are combined.

[0045] As can be seen from the comparison of accuracy in Figure 10, the heart rate time series feature AD assessment method has higher accuracy than the circadian rhythm AD assessment method. Furthermore, the combined assessment method, which combines the circadian rhythm AD assessment method and the heart rate time series feature AD assessment method, has even higher accuracy than the heart rate variability AD assessment method. Therefore, the dementia diagnosis device of this example can perform highly accurate dementia assessment based on the heart rate during sleep, etc.

[0046] [Modification] In the above-described embodiment, the second estimator 14, which is a feature estimator, extracts the heart rate at regular intervals within a time window of a predetermined width, divides the heart rate in each extracted time window into a plurality of intervals based on the range in which the heart rate decreases continuously, and estimates a time-series feature of the heart rate for each of the divided intervals. This feature estimation process is one example, and the second estimator 14, which is a feature estimator, may estimate other time-series feature amounts as long as the time-series feature amount of the heart rate is capable of detecting a decrease in heart rate constancy.

[0047] In this regard, in the above-described embodiment, the second estimator 14, which serves as a feature estimator, calculates correlation using a lag-2 partial autocorrelation coefficient. However, using a lag-2 partial autocorrelation coefficient is merely an example, and correlation may be calculated using other processes. FIG. 12 shows a list of representative examples of features obtained by the second estimator 14. The "label" column in FIG. 12 indicates the type of feature, and lists 15 types of feature in descending order of the average match rate (ave_acc_without_bias_data_unm) after removing the bias of the number of dementia patients and healthy individuals. The average match rate shown in the table in FIG. 12 is 100% when the value is 1, and, for example, 0.9 indicates a match rate of 90%.

[0048] The types shown in FIG. 12 are explained below. Here, the value or meaning of "x" corresponds to the numerical value or meaning shown in FIG. 12. For example, "partial_autocorrelation__lag_2_mean" indicates the mean of the partial autocorrelation coefficients when x = 2, i.e., lag 2. Also, "fft_aggregated__aggtype_"x"" indicates an example in which variance is used when x = variance in FIG. 12.

[0049] ・partial_autocorrelation__lag_x: The average of the partial autocorrelation coefficients at lag x. ・autocorrelation__lag_x: The average of the autocorrelation coefficients at lag x. ・number_cwt_peaks__n_x: The average number of CWT peaks. This is the count of peaks when smoothing the value at a certain time (t) and the x values ​​to the left and right (tx, t-x+1,..., t+x-1, t+x) using CWT (Continuous Wavelet Transform). ・number_peaks__n_x: The average number of peaks. This is the count of the maximum value of x among the values ​​at a certain time (t) and the x values ​​to the left and right (tx, t-x+1,..., t+x-1, t+x). ・fourier_entropy__bins_x: The average entropy of x power spectral densities calculated using the Welch algorithm. fft_aggregated__aggtype_"x": Either the mean, variance, skew, or kurtosis of the absolute Fourier transform spectrum (x=mean, variance, skew, or kurtosis; in the example of FIG. 12, x is the variance).

[0050] As shown in Figure 12, the average partial autocorrelation coefficients for lag 2 [partial_autocorrelation_lag_2_mean] in the top row shows that the average match rate between dementia patients and healthy individuals exceeds 90%, indicating that highly accurate judgment is possible. When the remaining 14 types of feature amounts shown in Figure 12 are used, the results are lower than the average partial autocorrelation coefficients for lag 2, but still show that highly accurate judgment is possible.

[0051] Furthermore, in the above-described embodiment, an example of using heart rate data as biometric information has been described, but this is not limiting. For example, the biometric information may be pulse rate data. That is, in the embodiment of the present invention, "information regarding heart rate" corresponds to "heart rate data" or "pulse rate data." Furthermore, in the above-described embodiment, the extracted heart rate for each time window is divided into multiple sections based on a correlation coefficient with the Benford distribution, but each section may be divided into multiple sections using a correlation coefficient other than the Benford distribution.

[0052] Furthermore, a program for realizing all or part of the functions of the dementia assessment device 1 shown in Fig. 1 may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into a computer and executed to perform all or part of the processing performed by the dementia assessment device 1. The program here is a program that executes the processing steps shown in the flowchart of Fig. 4 or Fig. 9. Note that when combining a circadian rhythm assessment process and a time-series feature assessment process, the combination shown in the flowchart of Fig. 9 is just one example of the combined process, and the two processes may be combined in other ways. For example, dementia assessment may be performed using each process, and the two assessment results may be compared to ultimately determine whether or not dementia is present.

[0053] DESCRIPTION OF SYMBOLS 1... Dementia determination device 2... Biological information measurement device 11... Acquisition unit 13... First estimation unit (circadian rhythm estimation unit) 14... Second estimation unit (feature amount estimation unit) 15... Determination unit 16... Storage unit 17... Output unit

Claims

1. A dementia diagnosis device comprising: an acquisition unit that acquires information regarding a subject's heart rate; a feature estimation unit that estimates time-series features of the heart rate from the information regarding the heart rate acquired by the acquisition unit; a judgment unit that detects a decrease in heart rate constancy based on the time-series features of the heart rate estimated by the feature estimation unit and thereby judges whether or not the subject has dementia, or whether or not the subject is in a state in which the subject is likely to develop dementia; and an output unit that outputs the result of the judgment made by the judgment unit.

2. The dementia diagnosis device of claim 1, wherein the feature estimation unit extracts the heart rate at regular intervals in a time window of a predetermined width, divides the heart rate in each extracted time window into a plurality of intervals based on the range in which the heart rate decreases continuously, and estimates a time series feature of the heart rate for each divided interval, and the judgment unit judges whether or not there is a possibility of dementia for each interval based on the time series feature, and makes the judgment from the proportion of intervals judged to have a possibility of dementia.

3. The dementia diagnosis device according to claim 2, wherein the division into the plurality of sections is performed using a plurality of different division methods.

4. The dementia diagnosis device according to claim 1, wherein the feature estimation unit estimates the time series feature of the heart rate from the correlation between the heart rate at a certain time and the heart rate two seconds prior to that time using a partial autocorrelation coefficient that removes the influence of the heart rate one second prior.

5. The dementia diagnosis device according to claim 1, wherein the feature estimation unit estimates a time series feature of the heart rate from the heart rate at a certain time and the number of peaks in the heart rate a predetermined time before that.

6. The dementia diagnosis device according to claim 1, wherein the feature estimation unit estimates the time series feature of the heart rate from a correlation between the heart rate at a certain time and the heart rate a specified time before that, using a lag-2 partial autocorrelation coefficient, and estimates the time series feature of the heart rate from the heart rate at a certain time and the number of heart rate peaks a specified time before that.

7. The dementia diagnosis device according to claim 2, wherein the feature estimation unit divides the interval into a plurality of sections using a plurality of different division methods and estimates the time series feature of the heart rate for each divided interval, and estimates the time series feature of the heart rate from a correlation between the heart rate at a certain time and the heart rate a predetermined time before that using a lag-2 partial autocorrelation coefficient, and also from the heart rate at a certain time and the number of peaks in the heart rate a predetermined time before that.

8. The dementia diagnosis device of claim 1, further comprising a rhythm estimation unit that estimates a circadian rhythm of the heart rate from information about the heart rate acquired by the acquisition unit, and the judgment unit judges whether or not the subject has dementia based on the time series features of the heart rate estimated by the feature estimation unit and the circadian rhythm of the heart rate estimated by the rhythm estimation unit.

9. A dementia diagnosis method comprising: an acquisition process for acquiring information regarding a subject's heart rate; a feature estimation process for estimating time-series features of the heart rate from the information regarding the heart rate acquired by the acquisition process; and a judgment process for detecting a decrease in heart rate homeostasis based on the time-series features of the heart rate estimated by the feature estimation process, thereby determining whether or not the subject has dementia, or whether or not the subject is in a state where there is a high possibility of developing dementia.

10. A program that causes a computer that has acquired information about a subject's heart rate to execute a calculation to determine dementia from the information about the heart rate, the program causing the computer to execute a feature estimation procedure that estimates time-series features of the heart rate from the information about the heart rate, and a determination procedure that determines whether the subject has dementia, or whether the subject is in a state where he or she is likely to develop dementia, by detecting a decrease in heart rate homeostasis based on the time-series features of the heart rate estimated by the feature estimation procedure.

Images