Biological information analysis system and biological information analysis program

Abstract

A biological information analysis system according to the present invention comprises: a heartbeat cycle calculation unit that calculates, as a heartbeat cycle, an interval of heartbeat activity of a subject from biological information acquired from the subject; a unit of analysis determination unit that divides the heartbeat cycle using a preset segment length as a unit of analysis ; a clustering unit that performs clustering of the heartbeat cycle for the unit of analysis using a probability distribution model relating to the heartbeat cycle in the unit of analysis; and a determination unit that performs determination of arrhythmia on the basis of a result of the clustering.

Description

Biological information analysis system and biological information analysis program

[0001] The present invention relates to a series of biological information analysis systems and biological information analysis programs for determining arrhythmia based on biological information acquired from a subject.

[0002] The heart contracts and expands repeatedly, acting as a pump to send blood throughout the body. This activity (hereafter referred to as cardiac activity) is maintained regular by weak electrical stimulation of cardiac muscle cells, but abnormalities in this electrical stimulation can also cause abnormalities in cardiac activity. This is called arrhythmia. There are various types of arrhythmia, some of which are highly lethal, causing cardiac arrest, and some of which can lead to serious diseases such as cerebral infarction. Therefore, early detection of the onset of arrhythmia is extremely important. For example, techniques that use biological information obtained from a subject using a biological information measuring device are known to determine arrhythmia.

[0003] Non-Patent Document 1 discloses a technique relating to a method for determining arrhythmia. Specifically, among various types of arrhythmia, atrial fibrillation (AF) is targeted, and determination is made based on the magnitude of variability in periodic information on the occurrence of heartbeats (hereinafter referred to as cardiac cycle). FIG. 19 is a diagram showing an example of the distribution of cardiac cycles in atrial fibrillation. Note that the curve shown in FIG. 19 represents an approximate curve (fitting curve). Atrial fibrillation is an arrhythmia that occurs continuously over a long period of time and is characterized by irregular cardiac cycles. Therefore, atrial fibrillation can be determined based on the magnitude of variability in cardiac cycles, based on the relationship between cardiac cycles and probability density as shown in FIG. 19.

[0004] However, the method of Non-Patent Document 1 raises concerns about erroneous determination of atrial fibrillation during periods when arrhythmic premature contractions (PVCs, PACs) also occur frequently. Figures 20A and 20B are diagrams showing an example of the distribution of cardiac cycles during premature contractions. The curves shown in Figures 20A and 20B are approximate curves (fitting curves). Among premature contractions, compensatory premature contractions are arrhythmias in which the heartbeat occurs early from the normal rhythm, resulting in large variations in cardiac cycles and leading to erroneous determination of atrial fibrillation (see Figure 20A). To address this issue, Patent Document 1 assumes multiple distributions of cardiac cycles and determines premature contractions in addition to atrial fibrillation based on the number of distributions and the magnitude of variation (see Figure 20B).

[0005] B Logan, J Healey. Robust Detection of Atrial Fibrillation for a Long Term Telemonitoring System, Computers in Cardiology 2005;32:619-622.

[0006] JP 2009-089883 A

[0007] As in Non-Patent Document 1 and Patent Document 1, arrhythmia can be determined from the distribution of cardiac cycles, but conventional techniques have low accuracy in detecting the distribution of cardiac cycles, and it is expected that, for example, in atrial fibrillation, the distribution will be divided into multiple parts, or in premature contractions, distributions that should be divided into multiple parts will be mixed together, which can lead to erroneous determination. Therefore, there is a demand for highly accurate detection of the distribution of cardiac cycles to determine arrhythmia.

[0008] The present invention aims to solve the above problems and provide a biological information analysis system and a biological information analysis program that can classify arrhythmias with high accuracy.

[0009] The biological information analysis system of the present invention that achieves the above object comprises the following components: (1) A biological information analysis system that uses biological information of a subject to determine arrhythmia, comprising: a cardiac cycle calculation unit that calculates an interval between cardiac activity of the subject as a cardiac cycle from the biological information acquired from the subject, an analysis unit determination unit that divides the cardiac cycle into analysis units of predetermined segment lengths, a clustering unit that performs clustering of the cardiac cycles for each analysis unit using a probability distribution model related to the cardiac cycle of the analysis unit, and a determination unit that performs arrhythmia determination based on the results of the clustering.

[0010] (2) The biological information analysis system according to (1), wherein the probability distribution model is generated by learning about the cardiac cycle of the analysis unit.

[0011] (3) The biological information analysis system according to (1), wherein the probability distribution model uses a Gaussian mixture distribution model.

[0012] (4) The biological information analysis system according to (1), wherein the clustering unit selects the probability distribution model using an information criterion and performs clustering using the selected probability distribution model.

[0013] (5) The bioinformation analysis system described in (1) above, wherein the determination unit determines arrhythmia based on the number of clusters in the clustering results, the number of cardiac cycles belonging to each cluster, spread information of the cardiac cycles belonging to each cluster, and / or time information of the cardiac cycles belonging to each cluster.

[0014] (6) The biological information analysis system according to any one of (1) to (3), wherein the clustering unit comprises: a first clustering unit that performs clustering of the cardiac cycles for the analysis unit using a probability distribution model for the cardiac cycles of the analysis unit; and a second clustering unit that performs additional clustering when two clusters exist in the cardiac cycles as a result of the clustering; and an arrhythmia reference information generating unit that generates arrhythmia reference information based on the result of the additional clustering, wherein the second clustering unit calculates a summed cardiac cycle using a cardiac cycle belonging to one of the two clusters having a smaller number of cardiac cycles and either before or after that cardiac cycle, and performs the additional clustering on the cardiac cycles of the analysis unit and the summed cardiac cycle using the probability distribution model; and the determining unit determines whether an arrhythmia exists based on the result of the clustering and the arrhythmia reference information.

[0015] (7) The biological information analysis system according to any one of (1) to (3), further comprising an arrhythmia reference information generating unit that generates arrhythmia reference information when three clusters exist in the cardiac cycle as a result of the clustering, wherein the arrhythmia reference information generating unit determines, among the three clusters, a longest cluster having the longest time information of the cardiac cycles belonging to the cluster and a shortest cluster having the shortest time information of the cardiac cycles belonging to the cluster, and generates arrhythmia reference information based on the similarity between the number of cardiac cycles belonging to the longest cluster and the number of cardiac cycles belonging to the shortest cluster, and the determination unit determines arrhythmia based on the result of the clustering and the arrhythmia reference information.

[0016] (8) The biological information analysis system according to any one of (1) to (3), further comprising: an arrhythmia reference information generation unit that generates arrhythmia reference information when there are three clusters in the cardiac cycle as a result of the clustering; wherein the arrhythmia reference information generation unit determines, among the three clusters, a longest cluster having the longest time information of the cardiac cycle belonging to the cluster, a shortest cluster having the shortest time information of the cardiac cycle belonging to the cluster, and other intermediate clusters; calculates total time information using the time information of the cardiac cycle belonging to the longest cluster and the time information of the cardiac cycle belonging to the shortest cluster; and generates arrhythmia reference information based on a similarity between the total time information and the time information of the cardiac cycle belonging to the intermediate cluster; and wherein the determination unit performs arrhythmia determination based on the result of the clustering and the arrhythmia reference information.

[0017] (9) A bioinformation analysis system according to any one of (1) to (8), further comprising: performing baseline correction on the cardiac cycle of the analysis unit; and performing clustering of the cardiac cycle on the analysis unit after the baseline correction.

[0018] (10) A bioinformation analysis program for determining arrhythmia using a subject's bioinformation, the bioinformation analysis program causing a computer to execute a series of processes: calculating the interval between cardiac activity of the subject as a cardiac cycle from the bioinformation acquired from the subject; reading a preset segment length from a memory unit; dividing the cardiac cycle using the segment length as an analysis unit; clustering the cardiac cycle for the analysis unit using a probability distribution model for the cardiac cycle of the analysis unit; and determining arrhythmia based on the results of the clustering.

[0019] According to the present invention, cardiac cycles in which arrhythmia characteristics appear are clustered using a probability distribution model, and the results of this clustering are utilized to classify arrhythmias with high accuracy.

[0020] FIG. 1 is a diagram illustrating an example of the configuration of a biological information analysis system according to a first embodiment. FIG. 2 is a diagram illustrating a detailed example of an electrocardiogram signal used for analysis. FIG. 3 is a flowchart illustrating arrhythmia determination processing according to the first embodiment. FIG. 4 is a flowchart illustrating model construction processing according to the first embodiment. FIG. 5A is a diagram (part 1) illustrating the model construction processing according to the first embodiment. FIG. 5B is a diagram (part 2) illustrating the model construction processing according to the first embodiment. FIG. 5C is a diagram (part 3) illustrating the model construction processing according to the first embodiment. FIG. 6 is a diagram illustrating an example of calculation of the Bayes Information Criterion by the clustering unit. FIG. 7A is a diagram (part 1) illustrating an example of an electrocardiogram signal used in the determination processing by the determination unit. FIG. 7B is a diagram (part 2) illustrating an example of an electrocardiogram signal used in the determination processing by the determination unit. FIG. 7C is a diagram (part 3) illustrating an example of an electrocardiogram signal used in the determination processing by the determination unit. FIG. 8A is a diagram (part 1) illustrating an example of a clustering result used in the determination processing by the determination unit. FIG. 8B is a diagram (part 2) illustrating an example of a clustering result used in the determination process by the determination unit. FIG. 8C is a diagram (part 3) illustrating an example of a clustering result used in the determination process by the determination unit. FIG. 9 is a diagram illustrating a configuration example of a biological information analysis system according to embodiment 2. FIG. 10 is a flowchart illustrating arrhythmia determination process according to embodiment 2. FIG. 11 is a flowchart illustrating arrhythmia reference information generation process according to embodiment 2. FIG. 12 is a diagram illustrating a configuration example of a biological information analysis system according to embodiment 3. FIG. 13 is a flowchart illustrating arrhythmia reference information generation process according to embodiment 3. FIG. 14 is a diagram illustrating a configuration example of a biological information analysis system according to embodiment 4. FIG. 15 is a flowchart illustrating arrhythmia determination process according to embodiment 4. FIG. 16 is a flowchart illustrating second clustering process according to embodiment 4. FIG. 17A is a diagram (part 1) illustrating an example of first clustering process by the first clustering unit. FIG. 17B is a diagram (part 2) illustrating an example of first clustering process by the first clustering unit.Fig. 18A is a diagram (part 1) for explaining an example of the second clustering process by the second clustering section. Fig. 18B is a diagram (part 2) for explaining an example of the second clustering process by the second clustering section. Fig. 19 is a diagram showing an example of the distribution of cardiac cycles in atrial fibrillation. Fig. 20A is a diagram (part 1) showing an example of the distribution of cardiac cycles in premature systoles. Fig. 20B is a diagram (part 2) showing an example of the distribution of cardiac cycles in premature systoles.

[0021] Hereinafter, embodiments of the bioinformation analysis system according to the present invention will be described in detail with reference to the drawings. Note that the present invention is not limited to these embodiments. Furthermore, the individual embodiments of the present invention are not independent and can be implemented in combination with each other as appropriate.

[0022] 1 is a diagram showing an example of the configuration of a biological information analysis system according to embodiment 1 of the present invention. The biological information analysis system includes an electrocardiogram signal measuring device 2, which is a biological information measuring device, an analysis system 3, a receiving terminal 4, and a learning device 5.

[0023] The biological information measuring device is not limited and can be freely determined depending on the biological information to be acquired. It is also assumed that the biological information measuring device includes a biological information output mechanism, such as a connector for electrically connecting to the analysis system 3, a communication device having means for communicating with the biological information analysis system, or an input port for media into which the acquired biological information is written. In the first embodiment, the biological information is an electrocardiographic signal 10, the biological information measuring device is an electrocardiographic signal measuring device 2, the electrocardiographic signal measuring device 2 includes an input port for media into which the acquired biological information is written, and the subject 1 is a human. In the first embodiment, an example is described in which the electrocardiographic signal 10 is acquired to monitor the condition of the subject 1 (whether or not the subject 1 has an arrhythmia). However, in addition to the electrocardiographic signal 10, any other device capable of estimating the cardiac cycle may be used, such as a pulse wave signal or heart sounds.

[0024] 1, an electrocardiogram signal 10 to be analyzed is acquired from a subject 1 via an electrocardiogram signal measuring device 2. The subject 1 is not particularly limited to a human being or an animal. The receiving terminal 4 is, for example, a smartphone carried by the subject 1 or a guard (for example, a surgeon such as a doctor, a factory hygiene manager, or a nearby worker).

[0025] A wearable electrocardiograph is an example of the electrocardiogram signal measuring device 2. Specifically, the electrocardiogram signal measuring device 2 includes a garment body to be worn by the subject 1, a plurality of electrodes, and an electrocardiograph 100 electrically connected to each electrode. The electrocardiograph 100 and the like are fixed to the garment body using a band or the like.

[0026] The electrocardiograph 100 is an example of a measuring device that acquires an electrocardiographic signal 10. The electrocardiograph 100 has a function of continuously acquiring the electrocardiographic signal 10 of the subject 1, a function of storing the acquired electrocardiographic signal 10, and a function of transferring data via communication with the analysis system 3. Note that the electrocardiograph 100 may be configured to transfer data including the electrocardiographic signal 10 to a server device, and the server device may transfer the electrocardiographic signal 10 to the analysis system 3. In the following description, the subject 1 refers to a person wearing an electrocardiographic signal measuring device 2.

[0027] In the first embodiment, an electrocardiogram signal 10 acquired from a subject 1 includes information about the subject 1, the acquisition date and time, the location, and the acquisition result. The acquisition result is waveform data 11 obtained by plotting the horizontal axis as time and the vertical axis as voltage (see FIG. 1 ).

[0028] 2 is a diagram showing a detailed example of an electrocardiographic signal 10 used for analysis. The electrocardiographic signal 10 may include waveform components called P waves 11a, Q waves 11b, R waves 11c, S waves 11d, and T waves 11e, and the shapes of these components are used as criteria for determining arrhythmia. For example, the waveform data 11 is composed of repeated patterns of these waveform components. Note that the shape of the waveform data 11 included in the electrocardiographic signal 10 may change depending on the subject and the way the electrocardiograph 100 is worn.

[0029] Returning to FIG. 1, the analysis system 3 includes a communication unit 311, a cardiac cycle calculation unit 312, an analysis unit determination unit 313, a clustering unit 314, a judgment unit 315, an input / output unit 316, a control unit 317, and a memory unit 318.

[0030] The communication unit 311 can communicate with the electrocardiogram signal measuring device 2 and the receiving terminal 4 via a communication network. The communication network here is configured using, for example, an existing public line network, a local area network (LAN), a wide area network (WAN), etc., and may be wired or wireless. The communication unit 311 is configured, for example, by a connector that electrically connects to a communication target, a communication device that has means for communicating with the electrocardiogram signal measuring device 2, or an input port for a medium on which data is stored.

[0031] The cardiac cycle calculation unit 312 detects cardiac beats from the biological information and calculates the cardiac cycle, which is the interval between the beats.

[0032] The analysis unit determination section 313 divides the cardiac cycle calculated by the cardiac cycle calculation section 312 into predetermined segments, and sets these as analysis units.

[0033] The clustering unit 314 selects a model to be used for clustering, performs clustering of the cardiac cycles using the selected model, and obtains the number of clusters and information about each cluster (the number of cardiac cycles belonging to each cluster, spread information, time information, etc.). In the first embodiment, the model used is the model generated by the learning device 5.

[0034] Based on the clustering result, the determining unit 315 determines whether the subject 1 is experiencing arrhythmia. If the subject 1 is experiencing arrhythmia, the determining unit 315 determines whether the subject 1 is experiencing atrial fibrillation / premature contraction.

[0035] The input / output unit 316 is configured by devices having input and output functions, and outputs various information under the control of the control unit 317. The input / output unit 316 includes a user interface such as a keyboard, mouse, microphone, etc., and has an input function of receiving input of the electrocardiogram signal 10 output from the electrocardiogram signal measuring device 2. Furthermore, the input / output unit 316 has output functions such as a display made of, for example, a liquid crystal or organic EL (Electro Luminescence) display, and a speaker that outputs sound.

[0036] Furthermore, it is assumed that the input function of the input / output unit 316 includes an input mechanism for biometric information, such as an input port for a medium on which biometric information is written. If the input function of the input / output unit 316 overlaps with the function of the communication unit 311, it is sufficient that either one of them has that function.

[0037] Furthermore, for the biometric information to be input to the analysis system 3, in addition to using previously acquired electrocardiogram signals 10, the acquisition of the electrocardiogram signals 10 by the electrocardiogram signal measuring device 2 and the input to the analysis system 3 may be carried out in parallel. In this case, it is assumed that the electrocardiogram signal measuring device 2 will constantly transmit values ​​acquired in a continuous manner, or will transmit values ​​acquired in a batch manner in batches at regular intervals, and there is no particular limitation.

[0038] The control unit 317 comprehensively controls the operation of the analysis system 3. The control unit 317 also transmits the determination result by the determination unit 315 together with the wearer information to the receiving terminal 4 via the communication unit 311. The wearer information is information for identifying the subject 1, including personal information such as the name and sex of the subject 1, and the management number of the wearable device or peripheral equipment worn by the subject 1. The control unit 317 may also output the determination result by the determination unit 315 using an output function of the input / output unit 316.

[0039] The storage unit 318 stores various programs for operating the analysis system 3 and various data including data generated by each unit (e.g., segment length). The various programs include a program for arrhythmia determination processing executed using a trained model. The storage unit 318 is configured using a read-only memory (ROM) in which the various programs are pre-installed, a random access memory (RAM), a hard disk drive (HDD), a solid state drive (SSD), etc., which store calculation parameters and data for each process.

[0040] The various programs can be widely distributed by recording them on computer-readable recording media such as HDDs, flash memories, CD-ROMs, DVD-ROMs, Blu-ray (registered trademark), etc. The communication unit 311 can also acquire the various programs via a communication network.

[0041] The analysis system 3 having the above-described functional configuration is a computer configured using one or more pieces of hardware such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), etc. The analysis system 3 may be configured separately from the electrocardiogram signal measuring device 2, or may be configured integrally with the electrocardiogram signal measuring device 2.

[0042] The receiving terminal 4 includes a communication unit 41 , an output unit 42 , and a control unit 43 .

[0043] The communication unit 41 receives, via the communication network, information output from the analysis system 3. The communication unit 41 receives, for example, the determination result by the determination unit 315 and wearer information from the analysis system 3.

[0044] The output unit 42 is configured by an output device such as a display device or a printer, and outputs various information under the control of the control unit 43. The output unit 42 has output functions such as a display made of, for example, a liquid crystal or organic EL display, and a speaker.

[0045] The control unit 43 controls the overall operation of the receiving terminal 4. The control unit 43 also outputs the determination result of the determination unit 315 through the output function of the output unit .

[0046] The receiving terminal 4 having the above functional configuration is a computer configured using one or more pieces of hardware such as a CPU, a GPU, an ASIC, an FPGA, etc. The receiving terminal 4 may also have a user interface.

[0047] Next, an arrhythmia determination process based on biological information obtained from the subject 1 will be described. Fig. 3 is a flowchart showing the arrhythmia determination process according to the first embodiment. In the arrhythmia determination process, first, the communication unit 311 or the input / output unit 316 acquires an electrocardiographic signal 10 obtained from the subject 1 (step S101). The control unit 317 processes waveform data 11 of the acquired electrocardiographic signal 10 as necessary to generate data for the determination process. In the first embodiment, an example will be described in which the input / output unit 316 has an input port for a medium to which the electrocardiographic signal 10 is written, and reads the electrocardiographic signal 10 stored in the medium after acquisition of the electrocardiographic signal 10 has been completed.

[0048] After acquiring the electrocardiographic signal 10, the cardiac cycle calculation unit 312 detects cardiac beats from the electrocardiographic signal 10 and estimates the cardiac cycle, which is the interval between the beats (step S102). As described in Fig. 2, the electrocardiographic signal 10 may include waveform components called P waves, Q waves, R waves, S waves, and T waves, which are generated by cardiac beats. The cardiac cycle calculation unit 312 detects these waveform components (especially R waves) to determine the timing of cardiac beats and calculates an estimated cardiac cycle from the interval between the beats.

[0049] After estimating the cardiac cycle, the analysis unit determination unit 313 divides the cardiac cycle by a predetermined segment length to determine the analysis unit (step S103). The analysis unit determination unit 313 divides the calculated cardiac cycle by a predetermined segment length and sets the result as the analysis unit. The segment unit may be, but is not limited to, a time unit obtained by dividing the cardiac cycle by a predetermined time length or a beat unit obtained by dividing the cardiac cycle every time a certain number of heartbeats are detected. However, in order to determine atrial fibrillation, it is desirable to set the analysis unit to a time unit of 30 seconds or more or a beat unit equivalent thereto. Furthermore, overlap can be freely set for the analysis unit. In this embodiment 1, an example will be described in which the analysis unit determination unit 313 divides the cardiac cycle into time units of 60 seconds without setting overlap, and sets the analysis unit as the analysis unit.

[0050] Thereafter, the clustering unit 314 determines a model to be used for clustering (step S104).

[0051] Here, the model determined by the clustering unit 314 will be described. This model is constructed by the learning device 5, and specifically, is a probability distribution model. The probability distribution model may be a Gaussian Mixture Model (GMM) or a Poisson Mixture Model (PMM), but is not particularly limited to this. In the first embodiment, the Gaussian Mixture Model (GMM) will be used as an example. FIG. 4 is a flowchart showing the model construction process according to the first embodiment.

[0052] Prior to constructing a model, the learning device 5 sets the number of clusters to be classified as a hyperparameter of the Gaussian mixture distribution model (step S201). In the first embodiment, the number of clusters to be classified is described as one to three, but is not particularly limited thereto. This means that the cardiac cycles in the analysis unit are classified into one of three cluster numbers, and an independent model is constructed for each cluster number. Below, the Gaussian mixture distribution model with one cluster number will be described as model (1), the Gaussian mixture distribution model with two clusters as model (2), and the Gaussian mixture distribution model with three clusters as model (3).

[0053] After setting the number of clusters, the learning device 5 sets the initial parameters of the model (step S202). Specifically, the learning device 5 sets the initial values ​​of the parameters (mean, covariance matrix, and mixing coefficients) of the Gaussian distribution for each number of clusters. Possible methods for determining the initial values ​​of the parameters include, but are not limited to, a method using a predetermined value, a method using a random value, or a method using a value calculated by a clustering method such as k-means.

[0054] After the initial setting, the learning device 5 optimizes the model using the EM (Expectation-Maximization) algorithm (step S203). The EM algorithm is a method for performing maximum likelihood estimation of the parameters of a probability distribution model, and is composed of an Expectation Step (E-step) that calculates the likelihood, which is the probability that each piece of data belongs to a Gaussian distribution in each cluster, and a Maximization Step (M-step) that re-estimates the parameters of the Gaussian distribution in each cluster based on the calculated likelihood. The EM algorithm loops through the E-step and M-step until a convergence criterion is satisfied. Examples of the convergence criterion include, but are not limited to, that the amount of change in the parameters or likelihood in each loop is less than a threshold, or that the number of loops is equal to or greater than a threshold.

[0055] 5A to 5C are diagrams for explaining the model construction process according to the first embodiment. FIG. 5A shows the Gaussian distribution in model (1). FIG. 5B shows the Gaussian distribution in model (2). FIG. 5C shows the Gaussian distribution in model (3). From FIGS. 5A to 5C, it can be seen that the Gaussian distribution is optimized according to the respective number of clusters. The model construction process by the learning device 5 is completed through the flow described above.

[0056] Returning to FIG. 3 , the clustering unit 314 determines a model to be used for clustering from each model constructed by the learning device 5. Specifically, in the first embodiment, three models, model (1) to model (3), are prepared, and the clustering unit 314 selects a model to be used for clustering from these. Methods for selecting a model include, but are not limited to, artificial selection and selection based on an information criterion such as the Bayesian Information Criterion or the Akaike's Information Criterion. In the first embodiment, an example will be described in which the clustering unit 314 selects a model based on the smallest Bayesian Information Criterion (hereinafter, BIC), which is one of the information criteria. The BIC is a function that can evaluate the fit to data using a likelihood function, and selecting the model with the smallest BIC allows the optimal model to be selected in many cases.

[0057] Fig. 6 is a diagram for explaining an example of BIC calculation by the clustering unit. Fig. 6 shows an example of BIC calculation for each model shown in Figs. 5A to 5C. In Fig. 6, model (3) has a smaller BIC than the other models, so it can be determined to be the optimal model, and is selected as the model to be used for clustering, and the number of clusters to be classified is determined. This completes the model determination process by the clustering unit 314.

[0058] 3, the clustering unit 314 performs clustering of cardiac cycles using the determined model (step S105). As a result of the clustering, the clustering unit 314 acquires the number of clusters and information about each cluster (the number of cardiac cycles belonging to each cluster, spread information, time information, etc.).

[0059] After the clustering process, the determination unit 315 executes a process of determining whether or not the subject 1 has an arrhythmia based on the clustering result (step S106). Specifically, the determination unit 315 determines whether the subject 1 has atrial fibrillation, premature contraction, or something else.

[0060] 7A to 7C are diagrams illustrating an example of an electrocardiographic signal 10 used in the determination process by the determination unit. FIGS. 8A to 8C are diagrams illustrating an example of a clustering result used in the determination process by the determination unit. FIG. 7A shows an electrocardiographic signal 10A without arrhythmia. FIG. 8A shows an example of the clustering result of cardiac cycles in the electrocardiographic signal 10A. In the case of the electrocardiographic signal 10A without arrhythmia, the cardiac cycle is constant, and the clustering result is a single cluster with small variance. FIG. 7B shows an electrocardiographic signal 10B in which atrial fibrillation has occurred. FIG. 8B shows an example of the clustering result of cardiac cycles in the electrocardiographic signal 10B. In the case of the electrocardiographic signal 10B in which atrial fibrillation has occurred, the cardiac cycle is irregular, and the clustering result is a single cluster with large variance. FIG. 7C shows an electrocardiographic signal 10C in which a premature ventricular contraction has occurred. FIG. 8C shows an example of the clustering result of cardiac cycles in the electrocardiographic signal 10C. In the case of the electrocardiogram signal 10C in which a premature ventricular contraction has occurred, the cardiac cycle before the premature ventricular contraction is short and the cardiac cycle after the premature ventricular contraction is long, so the clustering results in three clusters with little variance.

[0061] As shown in Figures 7A to 7C and Figures 8A to 8C, the number of clusters and the information about each cluster (the number of cardiac cycles belonging to each cluster, spread information, time information, etc.) show characteristics of atrial fibrillation and premature contractions, so it is possible to determine arrhythmia based on these characteristics.

[0062] In the first embodiment, for example, the determination unit 315 determines "atrial fibrillation" when the number of clusters is one to three and the standard deviation of the cluster with the largest number of cardiac cycles is 0.064 [sec] or more. Furthermore, the determination unit 315 determines "premature contraction" when the number of clusters is two or three and the standard deviation of each cluster is less than 0.064 [sec]. In such cases, the determination unit 315 determines "other" for an arrhythmia-free electrocardiogram signal 10A shown in FIG. 7A , "atrial fibrillation" for an atrial fibrillation-occurring electrocardiogram signal 10B shown in FIG. 7B , and "premature contraction" for an ventricular premature contraction-occurring electrocardiogram signal 10C shown in FIG. 7C .

[0063] Thereafter, the control unit 317 transmits information relating to the determination result by the determination unit 315 and information about the wearer to the receiving terminal 4 (step S107). Information about the condition of the subject 1 (whether or not the subject 1 has arrhythmia) is displayed on the receiving terminal 4. For example, a guard carrying the receiving terminal 4 checks the determination result.

[0064] According to the first embodiment described above, cardiac cycles in which arrhythmia characteristics appear are clustered using a probability distribution model, and the results of this clustering are utilized to classify arrhythmias with high accuracy.

[0065] According to this first embodiment, as a result of clustering using a probability distribution model, the number of clusters and information about each cluster (such as the number of cardiac cycles belonging to each cluster, spread information, and time information) can be obtained, which makes it possible to classify arrhythmias with high accuracy using these.

[0066] Furthermore, according to the first embodiment, highly accurate clustering using a probability distribution model is possible, and therefore, premature contractions whose cardiac cycles are divided into a plurality of clusters can be classified with high accuracy.

[0067] In this embodiment 1, a learning device 5 is provided separately from the analysis system 3, and a model is constructed using the learning device 5. However, the analysis system 3 may be configured to have the functions of the learning device 5, or the model construction process in the arrhythmia determination process may be skipped by directly reading the trained model.

[0068] Furthermore, in the first embodiment, the cardiac cycle estimated from the electrocardiogram signal 10 is directly clustered, but other processing may be used to process the data. One example of data processing is baseline correction. In baseline correction, the clustering unit 314 removes the baseline in the cardiac cycle using a method such as a moving average or a cutoff in frequency space. This removes heart rate fluctuations caused by exercise or mental excitement, enabling highly accurate classification of arrhythmias. While the data processing described above is expected to be performed by the clustering unit 314, it may also be performed by the analysis unit determination unit 313, the control unit 317, or the like, and is not limited thereto.

[0069] (Embodiment 2) Next, a description will be given of embodiment 2 of the present invention. In embodiment 2, an analysis system 3A that acquires position information is provided instead of the analysis system 3 according to embodiment 1. Below, the same components as in embodiment 1 are given the same reference numerals, and differences will be described.

[0070] 9 is a diagram showing an example of the configuration of a biological information analysis system according to the present embodiment 2. The biological information analysis system according to the present embodiment 2 includes an electrocardiogram signal measuring device 2, an analysis system 3A, a receiving terminal 4, and a learning device 5.

[0071] The analysis system 3A includes a communication unit 311, a cardiac cycle calculation unit 312, an analysis unit determination unit 313, a clustering unit 314, a judgment unit 315, a first arrhythmia reference information generation unit 319, an input / output unit 316, a control unit 317, and a memory unit 318.

[0072] The first arrhythmia reference information generating unit 319 performs a determination process based on the number of cardiac cycles belonging to each cluster, and generates arrhythmia reference information, which is a parameter indicating whether or not a premature contraction can be determined.

[0073] Next, an arrhythmia determination process according to the second embodiment will be described. Fig. 10 is a flowchart showing the arrhythmia determination process according to the second embodiment. In the arrhythmia determination process according to the second embodiment, similar to steps S101 to S105 (see Fig. 3) of the first embodiment described above, a cardiac cycle is estimated from the acquired electrocardiographic signal 10, an analysis unit is determined, and then a cardiac cycle clustering process is performed using a model (steps S301 to S305).

[0074] In the second embodiment, after the clustering process, the control unit 317 determines whether the number of clusters is three as a result of the clustering process (step S306). If the control unit 317 determines that the number of clusters is not three (step S306: No), the control unit 317 proceeds to step S308. On the other hand, if the control unit 317 determines that the number of clusters is three (step S306: Yes), the control unit 317 proceeds to step S307.

[0075] In step S307, the first arrhythmia reference information generating unit 319 generates arrhythmia reference information. If the number of clusters is three as a result of clustering, the first arrhythmia reference information generating unit 319 generates arrhythmia reference information, which is a parameter indicating whether or not a premature contraction can be determined, based on the number of cardiac cycles belonging to each cluster.

[0076] 11 is a flowchart showing the arrhythmia reference information generation process according to the second embodiment. First, the first arrhythmia reference information generator 319 determines the longest cluster (hereinafter referred to as "cluster Long") and the shortest cluster (hereinafter referred to as "cluster Short") among the three clustered clusters, which have the longest cardiac cycle time information (step S401). At this time, if a compensatory extra-systole occurs among the extra-systoles in the analysis unit, the shortened cardiac cycle before the extra-systole is considered to belong to cluster Short, and the lengthened cardiac cycle after the extra-systole is considered to belong to cluster Long. The time information here may be, for example, the average of the cardiac cycles belonging to each cluster.

[0077] Next, the first arrhythmia reference information generating unit 319 acquires the numbers of cardiac cycles belonging to the cluster Long and the cluster Short, respectively, and evaluates the similarity between the numbers of cardiac cycles belonging to the two clusters (step S402). In the second embodiment, the method for evaluating the similarity between the numbers of cardiac cycles belonging to the two clusters is assumed to use relative error, but is not particularly limited to this. The relative error can be obtained, for example, by dividing the absolute value of the difference between two values ​​by the larger of the two values. In the second embodiment, the relative error is obtained by dividing the absolute value of the difference between two values ​​by the larger of the two values.

[0078] The first arrhythmia reference information generating unit 319 then generates arrhythmia reference information based on the similarity between the numbers of cardiac cycles belonging to the two evaluated clusters (step S403). Specifically, the arrhythmia reference information is a parameter indicating whether or not an analysis unit is a premature systole candidate, and in the second embodiment, it is assumed to have a binary value, with "1" indicating that the analysis unit can be determined to be a premature systole candidate and "0" indicating that the analysis unit cannot be determined to be a premature systole candidate, but this is not particularly limited.

[0079] In the second embodiment, for example, if the relative error between the number of cardiac cycles belonging to cluster Long and the number of cardiac cycles belonging to cluster Short is 0.1 or more, the first arrhythmia reference information generating unit 319 determines the error as 0, i.e., "not a premature contraction candidate," and if it is less than 0.1, determines the error as 1, i.e., "a premature contraction candidate." This is because, among premature contractions, a compensatory premature contraction is an arrhythmia in which a heartbeat occurs in advance of a normal rhythm. Therefore, if the analysis unit is assumed to be a premature contraction, it can be determined that, of the three clusters, the cardiac cycles belonging to cluster Long are calculated from the waveform in which the premature contraction occurred and the waveform thereafter, and that the cardiac cycles belonging to cluster Short are calculated from the waveform in which the premature contraction occurred and the waveform immediately before it, and it can be determined that the numbers of cardiac cycles belonging to these clusters are similar to the premature contraction occurring in the analysis unit.

[0080] If the control unit 317 determines that the number of clusters is not three (step S306: No), the first arrhythmia reference information generating unit 319 may generate a predetermined value as the arrhythmia reference information. For example, in the second embodiment, if the number of clusters is not three as a result of clustering, the first arrhythmia reference information generating unit 319 generates a "missing value" as the arrhythmia reference information.

[0081] With the above, the arrhythmia reference information generating process by the first arrhythmia reference information generating unit 319 is completed.

[0082] 10 , in step S308, after generating the arrhythmia reference information, the determination unit 315 executes a process of determining whether or not the condition of the subject 1 is arrhythmia based on the clustering result. Specifically, the determination unit 315 determines whether or not the condition is atrial fibrillation, premature contraction, or something else.

[0083] Thereafter, the control unit 317 transmits information relating to the determination result by the determination unit 315 and information about the wearer to the receiving terminal 4 (step S309). Information about the condition of the subject 1 (whether or not the subject 1 has arrhythmia) is displayed on the receiving terminal 4. For example, a guard carrying the receiving terminal 4 checks the determination result.

[0084] According to the second embodiment described above, similar to the first embodiment, clustering is performed using a probability distribution model on cardiac cycles in which arrhythmia characteristics appear, and the results of this clustering are utilized to classify arrhythmias with high accuracy.

[0085] Furthermore, in this embodiment 2, if the number of clusters determined as a result of clustering is three, the first arrhythmia reference information generation unit 319 generates arrhythmia reference information based on the number of cardiac cycles belonging to each cluster, thereby providing a biological information analysis system that can classify arrhythmias with high accuracy.

[0086] In the second embodiment, the cluster information of the cluster Long and cluster Short is used as is, but data processing may be performed by other processes with reference to the first embodiment.

[0087] Another example of data processing is the elimination of intermittent peaks. Premature systoles include compensatory peaks, which are heartbeats that occur before the normal rhythm, and intermittent peaks, which are heartbeats that interrupt the normal rhythm. When an intermittent peak occurs, the number of cardiac cycles belonging to cluster Long does not increase, and only the number of cardiac cycles belonging to cluster Short increases. As a result, the number of cardiac cycles belonging to cluster Long and cluster Short becomes dissimilar to the premature systoles that occurred in the analysis unit. As a result, in the arrhythmia determination process according to the second embodiment, the cardiac cycles are not determined as premature systoles, which may result in a decrease in the accuracy of arrhythmia classification. As a method for excluding intermittent peaks, when two or more consecutive cardiac cycles belonging to cluster Short are detected in the time series in the analysis unit, the intermittent peaks are determined to be intermittent peaks and are excluded from the determination target in advance, and the determination process of the first arrhythmia reference information generating unit 319 is performed. However, this is not particularly limited. This allows the first arrhythmia reference information generating unit 319 to target only compensatory peaks, thereby enabling highly accurate classification of arrhythmias.

[0088] (Embodiment 3) Next, embodiment 3 of the present invention will be described. In embodiment 3, an analysis system 3B that acquires position information is provided instead of the analysis system 3 according to embodiment 1. Below, the same components as in embodiment 1 are given the same reference numerals, and differences will be described.

[0089] 12 is a diagram showing an example of the configuration of a biological information analysis system according to the present embodiment 3. The biological information analysis system according to the present embodiment 3 includes an electrocardiogram signal measuring device 2, an analysis system 3B, a receiving terminal 4, and a learning device 5.

[0090] The analysis system 3B includes a communication unit 311, a cardiac cycle calculation unit 312, an analysis unit determination unit 313, a clustering unit 314, a judgment unit 315, a second arrhythmia reference information generation unit 320, an input / output unit 316, a control unit 317, and a memory unit 318.

[0091] If the number of clusters is three as a result of clustering, the second arrhythmia reference information generation unit 320 performs a judgment process based on the time information of the cardiac cycles belonging to each cluster, and generates arrhythmia reference information, which is a parameter indicating whether or not a premature contraction can be determined.

[0092] The arrhythmia determination process in the third embodiment is performed according to the flowchart shown in Fig. 10. In the third embodiment, in step S307, the second arrhythmia reference information generating unit 320 generates arrhythmia reference information according to the flowchart shown in Fig. 13.

[0093] 13 is a flowchart showing the arrhythmia reference information generation process according to the third embodiment. First, the second arrhythmia reference information generator 320 determines the longest cluster (hereinafter referred to as "cluster Long"), the shortest cluster (hereinafter referred to as "cluster Short"), and the remaining cluster (hereinafter referred to as "cluster Center") from among the three clustered clusters, based on the time information of the cardiac cycles to which they belong (step S501). The time information here refers to, for example, the average of the cardiac cycles belonging to each cluster. In this case, for the compensatory premature contraction in the analysis unit, the shortened cardiac cycle before the premature contraction is considered to belong to cluster Short, the lengthened cardiac cycle after the premature contraction is considered to belong to cluster Long, and the normal cardiac cycle unrelated to the premature contraction is considered to belong to cluster Center.

[0094] Next, the second arrhythmia reference information generating unit 320 calculates total time information from the cluster Long and the cluster Short (step S502). The total time information is calculated based on the time information of the cardiac cycles belonging to the cluster Long and the time information of the cardiac cycles belonging to the cluster Short, and a value that represents the gap between the two clusters is calculated. In the third embodiment, it is assumed that the total time information is an average of the time information of the cardiac cycles belonging to the cluster Long and the time information of the cardiac cycles belonging to the cluster Short, but this is not particularly limited.

[0095] Next, the second arrhythmia reference information generating unit 320 evaluates the similarity between the summed time information and the time information of the cardiac cycles belonging to the cluster Center (step S503). In the third embodiment, the method of evaluating the similarity is assumed to be the use of relative error, but is not particularly limited thereto. The relative error can be obtained, for example, by dividing the absolute value of the difference between two values ​​by the larger of the two values. In the third embodiment, the relative error is obtained by dividing the absolute value of the difference between two values ​​by the larger of the two values.

[0096] Then, the second arrhythmia reference information generating unit 320 generates arrhythmia reference information, which is a parameter indicating whether or not a premature contraction can be determined, based on the similarity between the evaluated total time information and the time information of the cardiac cycles belonging to the cluster Center (step S504). Specifically, the arrhythmia reference information is a parameter indicating whether or not an analysis unit is a premature contraction candidate, and in the third embodiment, it is assumed to have a binary value, with "1" indicating that the analysis unit can be determined to be a premature contraction candidate and "0" indicating that the analysis unit cannot be determined to be a premature contraction candidate, but this is not particularly limited.

[0097] In the third embodiment, for example, if the relative error between the total time information and the time information of the cardiac cycle belonging to cluster Center is 0.2 or more, the second arrhythmia reference information generating unit 320 determines the error as "0," i.e., "not a premature contraction candidate," and if it is less than 0.2, determines the error as "1," i.e., "a premature contraction candidate." This is because, among premature contractions, a compensatory premature contraction is an arrhythmia in which a heartbeat occurs that is ahead of a normal rhythm. Therefore, if the analysis target is a premature contraction, it can be determined that the average of the cardiac cycles belonging to the cluster with the longest time information of the cardiac cycle and the cluster with the shortest time information is similar to the cardiac cycle belonging to the remaining cluster, i.e., the cluster showing a normal rhythm.

[0098] If the control unit 317 determines that the number of clusters is not three (step S306: No), the second arrhythmia reference information generating unit 320 may generate a predetermined value as the arrhythmia reference information. For example, in the third embodiment, if the number of clusters is not three as a result of clustering, the second arrhythmia reference information generating unit 320 generates a "missing value" as the arrhythmia reference information.

[0099] This completes the arrhythmia reference information generation process by the second arrhythmia reference information generating unit 320. The determining unit 315 performs a process of determining whether or not the subject 1 has an arrhythmia based on the clustering result. In the third embodiment, the determining unit 315 determines the condition as "atrial fibrillation" when the number of clusters is one to three and the standard deviation of the cluster containing the largest number of cardiac cycles is 0.064 [sec] or more, and determines the condition as "premature contraction" when the number of clusters is two and the standard deviation of each cluster is less than 0.064 [sec], or when the number of clusters is three and the standard deviation of each cluster is less than 0.064 [sec] and the arrhythmia reference information is "1".

[0100] According to the third embodiment described above, similar to the first embodiment, clustering is performed using a probability distribution model on cardiac cycles in which arrhythmia characteristics appear, and the results of this clustering are utilized to classify arrhythmias with high accuracy.

[0101] Furthermore, in this embodiment 3, if the number of clusters determined as a result of clustering is three, the second arrhythmia reference information generating unit 320 generates arrhythmia reference information based on the time information of the cardiac cycles belonging to each cluster, thereby providing a biometric information analysis system that can classify arrhythmias with high accuracy.

[0102] In the third embodiment, the information of each cluster is used as is, but data processing may be performed by other processes with reference to the first and second embodiments.

[0103] (Fourth embodiment) Next, a fourth embodiment of the present invention will be described. This fourth embodiment includes an analysis system 3C that acquires position information instead of the analysis system 3 according to the first embodiment. Below, the same components as those in the first embodiment are denoted by the same reference numerals, and differences will be described.

[0104] 12 is a diagram showing an example of the configuration of a biological information analysis system according to the present embodiment 4. The biological information analysis system according to the present embodiment 4 includes an electrocardiogram signal measuring device 2, an analysis system 3C, a receiving terminal 4, and a learning device 5.

[0105] The analysis system 3C includes a communication unit 311, a cardiac cycle calculation unit 312, an analysis unit determination unit 313, a clustering unit 314A, a judgment unit 315, a third arrhythmia reference information generation unit 321, an input / output unit 316, a control unit 317, and a memory unit 318.

[0106] The clustering unit 314A has a first clustering unit 314a and a second clustering unit 314b.

[0107] Similar to the clustering unit 314, the first clustering unit 314a selects a model to be used for clustering, performs clustering of the cardiac cycles using the selected model, and obtains the number of clusters and information about each cluster.

[0108] If the number of clusters is two as a result of the clustering by the first clustering unit 314a, the second clustering unit 314b performs additional clustering processing using the cardiac cycles in the analysis unit.

[0109] The third arrhythmia reference information generating unit 321 generates arrhythmia reference information based on the result of the additional clustering performed by the second clustering unit 314b.

[0110] Next, an arrhythmia determination process according to the fourth embodiment will be described. Fig. 15 is a flowchart showing the arrhythmia determination process according to the fourth embodiment. In the arrhythmia determination process according to the fourth embodiment, similarly to steps S101 to S105 (see Fig. 3) of the first embodiment described above, a cardiac cycle is estimated from the acquired electrocardiographic signal 10, an analysis unit is determined, and then the first clustering unit 314a performs a clustering process (first clustering process) of the cardiac cycles using a model (steps S601 to S605).

[0111] In the fourth embodiment, after the clustering process, the control unit 317 determines whether the number of clusters is two as a result of the first clustering process (step S606). If the control unit 317 determines that the number of clusters is not two (step S606: No), the control unit 317 proceeds to step S609. On the other hand, if the control unit 317 determines that the number of clusters is two (step S606: Yes), the control unit 317 proceeds to step S607.

[0112] In step S607, the second clustering unit 314b executes additional clustering processing (second clustering processing). Fig. 16 is a flowchart showing the second clustering processing according to the fourth embodiment.

[0113] First, the second clustering unit 314b detects, one by one, cardiac cycles belonging to the cluster with the fewer number of cardiac cycles from among the two clusters clustered by the first clustering unit 314a (step S701).

[0114] Next, the second clustering unit 314b replaces the detected cardiac cycle and any cardiac cycle before or after the detected cardiac cycle with a summed cardiac cycle on the time series in the analysis unit (step S702). At this time, which cardiac cycle before or after the detected cardiac cycle to use is determined as follows: First, if the time information of a cardiac cycle belonging to a cluster with a smaller number of cardiac cycles is shorter than that of the other cluster, the cardiac cycle after the detected cardiac cycle is used. Conversely, if the time information is longer than that of the other cluster, the cardiac cycle before the detected cardiac cycle is used. The time information here is, for example, the average of the cardiac cycles belonging to each cluster. Note that in the fourth embodiment, it is assumed that the summed cardiac cycle uses the average of the detected cardiac cycle and any cardiac cycle before or after the detected cardiac cycle, but this is not particularly limited.

[0115] The two cardiac cycles used in this calculation are deleted from the analysis unit, and instead the calculated summed cardiac cycle is added to the analysis unit to replace the cardiac cycle.

[0116] The above operation is performed for all cardiac cycles that belong to the detected clusters with a small number of cardiac cycles.

[0117] Next, the second clustering unit 314b constructs a model for the cardiac cycle in the analysis unit after the replacement, with the number of clusters set to 1 (step S703). In the model construction process of step S703, specifically, in step S201 (see FIG. 4) of the first embodiment described above, the number of clusters to be classified is set to 1, and the model construction process is performed in the same manner as in steps S202 to S203 (see FIG. 4).

[0118] The second clustering unit 314b then clusters the cardiac cycles using the constructed model and obtains cluster information (such as the number of cardiac cycles belonging to each cluster, spread information, and time information) as the second clustering result (step S704). The spread information here is, for example, the standard deviation of the distribution of each cluster, and the time information is, for example, the average of the cardiac cycles belonging to each cluster.

[0119] This completes the second clustering process by the second clustering unit 314b.

[0120] 10 , after the second clustering process, the third arrhythmia reference information generating unit 321 generates arrhythmia reference information based on the results of the second clustering process (step S608). Specifically, the arrhythmia reference information is a parameter indicating whether or not an analysis unit is a premature contraction candidate. In the fourth embodiment, the arrhythmia reference information is assumed to have a binary value, with "1" indicating that the analysis unit can be determined to be a premature contraction candidate and "0" indicating that the analysis unit cannot be determined to be a premature contraction candidate, but this is not limited thereto. In the fourth embodiment, for example, the third arrhythmia reference information generating unit 321 determines that the cluster is not a premature contraction candidate by setting the value to 0 when the standard deviation of the cluster is 0.064 [sec] or more, i.e., "not a premature contraction candidate," and determines that the cluster is not a premature contraction candidate by setting the value to 1 when the standard deviation is less than 0.064 [sec], i.e., "a premature contraction candidate."

[0121] The reason for this determination will be explained with reference to FIGS. 17A to 18B. FIGS. 17A and 17B are diagrams illustrating an example of the first clustering process performed by the first clustering unit 314a. FIGS. 18A and 18B are diagrams illustrating an example of the second clustering process performed by the second clustering unit. FIG. 17A shows the result of clustering of cardiac cycles (first clustering process) performed by the first clustering unit 314a for an electrocardiographic signal 10D (not shown) in which atrial fibrillation has occurred. FIG. 18A shows the result of second clustering process of cardiac cycles performed by the second clustering unit 314b for an electrocardiographic signal 10D in which atrial fibrillation has occurred. FIG. 17B shows the result of clustering of cardiac cycles (first clustering process) performed by the first clustering unit 314a for an electrocardiographic signal 10E (not shown) in which a premature ventricular contraction has occurred. FIG. 18B shows the result of the second clustering process of cardiac cycles performed by the second clustering unit 314b in the electrocardiogram signal 10E in which a premature ventricular contraction has occurred.

[0122] 17B , the clustering of cardiac cycles by the first clustering unit 314a for the electrocardiogram signal 10E in which a premature ventricular contraction has occurred should generally result in a cluster count of 3, but the cluster Long, to which the cardiac cycles that become longer after the premature contraction belong, and the cluster Center, to which normal cardiac cycles unrelated to the premature contraction belong, overlap to form a single cluster (hereinafter referred to as "cluster LC"), resulting in a cluster count of 2. This type of clustering of cardiac cycles by the first clustering unit 314a, where clusters for premature contractions overlap to form a single cluster, is expected to occur when the model construction by the learning device 5 or the model selection by the first clustering unit 314a is insufficient, and may result in an erroneous determination of atrial fibrillation, etc.

[0123] 18B, when the second clustering unit 314b processes an electrocardiogram signal 10E in which a premature ventricular contraction has occurred, the clusters of cardiac cycles after replacement with summed cardiac cycles have smaller variations. On the other hand, when the second clustering unit 314b processes an electrocardiogram signal 10D in which atrial fibrillation has occurred, the clusters of cardiac cycles after replacement with summed cardiac cycles have larger variations. This is because, among premature contractions, compensatory premature contractions are arrhythmias in which the heartbeat occurs in advance of the normal rhythm, and therefore the shortened cardiac cycle before the premature contraction and the lengthened cardiac cycle after the premature contraction are adjacent to each other on the time series in the analysis unit. Therefore, based on the cluster Short to which the shortened cardiac cycle before the premature contraction belongs, the elements of cluster L hidden in cluster LC can be identified, and furthermore, it can be determined that the average is similar to cluster C, i.e., a normal cardiac cycle that is not related to a premature contraction. For example, as a result of additional clustering of cardiac cycles by the second clustering unit 314b, if the standard deviation of the cluster is less than 0.064 [sec], it can be determined to be a "premature contraction candidate."

[0124] If the control unit 317 determines that the number of clusters is not two (step S606: No), the third arrhythmia reference information generating unit 321 may generate a default value as the arrhythmia reference information. In the fourth embodiment, if the number of clusters is not two as a result of clustering, the third arrhythmia reference information generating unit 321 generates a "missing value" as the arrhythmia reference information.

[0125] With the above, the arrhythmia reference information generating process by the third arrhythmia reference information generating unit 321 is completed.

[0126] In step S609, the determination unit 315 performs a determination process of whether the state of the subject 1 is arrhythmia based on the clustering result. Specifically, the determination unit 315 determines whether the state is atrial fibrillation, premature contraction, or something else. In the fourth embodiment, the determination unit 315 determines the state as "atrial fibrillation" when, for example, the number of clusters is 1 or 3 and the standard deviation of the cluster to which the largest number of cardiac cycles belongs is 0.064 [sec] or more, or the number of clusters is 2 and the standard deviation of the cluster to which the largest number of cardiac cycles belongs is 0.064 [sec] or more, and the arrhythmia reference information is 0. The determination unit 315 determines the state as "premature contraction" when, for example, the number of clusters is 2 and the standard deviation of each cluster is less than 0.064 [sec] and the arrhythmia reference information is 1, or the number of clusters is 3 and the standard deviation of each cluster is less than 0.064 [sec].

[0127] Thereafter, the control unit 317 transmits information relating to the determination result by the determination unit 315 and information about the wearer to the receiving terminal 4 (step S610). Information about the condition of the subject 1 (whether or not the subject 1 has arrhythmia) is displayed on the receiving terminal 4. For example, a guard carrying the receiving terminal 4 checks the determination result.

[0128] According to the fourth embodiment described above, similar to the first embodiment, clustering is performed using a probability distribution model on cardiac cycles in which arrhythmia characteristics appear, and the results of this clustering are utilized to classify arrhythmias with high accuracy.

[0129] Furthermore, according to the fourth embodiment, even if the accuracy of clustering by the first clustering unit 314a is insufficient, the second clustering unit 314b and the third arrhythmia reference information generating unit 321 can generate arrhythmia reference information based on the information of each cluster, thereby providing a biological information analysis system that can classify arrhythmias with high accuracy.

[0130] In the fourth embodiment, the information of each cluster is used as is, but data processing may be performed by other processes with reference to the first and second embodiments.

[0131] Other Embodiments While the embodiments for carrying out the present invention have been described above, the present invention should not be limited to the above-described embodiments. For example, although the analysis systems 3, 3A to 3C have been described as being provided separately from the electrocardiogram signal measuring device 2, the analysis systems 3, 3A to 3C may be configured integrally with the electrocardiogram signal measuring device 2.

[0132] REFERENCE SIGNS LIST 1 Subject 2 Electrocardiogram signal measuring device 3, 3A to 3C Analysis system 4 Receiving terminal 5 Learning device 10, 10A to 10E Electrocardiogram signal 11 Waveform data 41, 311 Communication unit 42 Output unit 43, 317 Control unit 312 Cardiac period calculation unit 313 Analysis unit determination unit 314, 314A Clustering unit 314a First clustering unit 314b Second clustering unit 315 Determination unit 316 Input / output unit 318 Storage unit 319 First arrhythmia reference information generation unit 320 Second arrhythmia reference information generation unit 321 Third arrhythmia reference information generation unit

Claims

1. A bioinformation analysis system for determining arrhythmia using bioinformation of a subject, comprising: a cardiac cycle calculation unit that calculates the interval between cardiac activity of the subject as a cardiac cycle from the bioinformation acquired from the subject; an analysis unit determination unit that divides the cardiac cycle into analysis units of predetermined segment lengths; a clustering unit that clusters the cardiac cycles for the analysis units using a probability distribution model related to the cardiac cycles of the analysis units; and a determination unit that determines arrhythmia based on the results of the clustering.

2. The biological information analysis system according to claim 1, wherein the probability distribution model is generated by learning about the cardiac cycle of the analysis unit.

3. The bioinformation analysis system according to claim 1, wherein the probability distribution model uses a Gaussian mixture distribution model.

4. The bioinformation analysis system according to claim 1, wherein the clustering unit selects the probability distribution model using an information criterion, and performs clustering using the selected probability distribution model.

5. The bioinformation analysis system of claim 1, wherein the determination unit determines arrhythmia based on the number of clusters in the clustering results, the number of cardiac cycles belonging to each cluster, spread information of the cardiac cycles belonging to each cluster, and / or time information of the cardiac cycles belonging to each cluster.

6. The biological information analysis system according to any one of claims 1 to 3, wherein the clustering unit comprises: a first clustering unit that performs clustering of the cardiac cycles for the analysis unit using a probability distribution model for the cardiac cycles of the analysis unit; and a second clustering unit that performs additional clustering when two clusters exist in the cardiac cycles as a result of the clustering; and an arrhythmia reference information generation unit that generates arrhythmia reference information based on the result of the additional clustering, wherein the second clustering unit calculates a summed cardiac cycle using a cardiac cycle that belongs to one of the two clusters having a smaller number of cardiac cycles and either the cardiac cycle before or after that cardiac cycle, and performs the additional clustering on the cardiac cycles of the analysis unit and the summed cardiac cycle using the probability distribution model, and the determination unit determines whether an arrhythmia has occurred based on the result of the clustering and the arrhythmia reference information.

7. A bioinformation analysis system as described in any one of claims 1 to 3, further comprising an arrhythmia reference information generation unit that generates arrhythmia reference information when three clusters exist in the cardiac cycle as a result of the clustering, wherein the arrhythmia reference information generation unit determines, among the three clusters, the longest cluster having the longest time information of the cardiac cycles belonging to the cluster and the shortest cluster having the shortest time information of the cardiac cycles belonging to the cluster, and generates arrhythmia reference information based on the similarity between the number of cardiac cycles belonging to the longest cluster and the number of cardiac cycles belonging to the shortest cluster, and wherein the determination unit determines arrhythmia based on the result of the clustering and the arrhythmia reference information.

8. The biological information analysis system according to any one of claims 1 to 3, further comprising an arrhythmia reference information generation unit that generates arrhythmia reference information when three clusters exist in the cardiac cycle as a result of the clustering, wherein the arrhythmia reference information generation unit determines, from the three clusters, the longest cluster having the longest time information of the cardiac cycle belonging to the cluster, the shortest cluster having the shortest time information of the cardiac cycle belonging to the cluster, and any other intermediate clusters, calculates total time information using the time information of the cardiac cycle belonging to the longest cluster and the time information of the cardiac cycle belonging to the shortest cluster, and generates arrhythmia reference information based on the similarity between the total time information and the time information of the cardiac cycle belonging to the intermediate cluster, and wherein the determination unit performs arrhythmia determination based on the result of the clustering and the arrhythmia reference information.

9. The biological information analysis system according to claim 1, wherein baseline correction is performed on the cardiac cycle of the analysis unit, and clustering of the cardiac cycle is performed on the analysis unit after the baseline correction.

10. A bioinformation analysis program for diagnosing arrhythmia using a subject's bioinformation, the bioinformation analysis program causing a computer to execute a series of processes: calculating the interval between cardiac activity of the subject as a cardiac cycle from the bioinformation acquired from the subject; reading a preset segment length from a memory unit; dividing the cardiac cycle using the segment length as an analysis unit; clustering the cardiac cycle for the analysis unit using a probability distribution model for the cardiac cycle of the analysis unit; and diagnosing arrhythmia based on the results of the clustering.

Images