Respiratory monitoring system

Abstract

Computer 3 performs the following steps to accurately detect the number of breaths using data obtained from radio waves transmitted from multiple transmitters and received by multiple receivers: a tensor data acquisition step in which data transmitted from multiple transmitters and received by multiple receivers is measured at minute time intervals and formatted into tensor data obtained per unit time; a tensor decomposition step in which the tensor data is reduced in dimension using time-series data per unit time to generate a group of 2D graphs; a respiratory data extraction step in which respiratory data is extracted from the group of 2D graphs using a respiratory motion model that has been pre-trained with first training data; a respiratory rate detection step in which the respiratory rate is detected from the respiratory data using a respiratory rate model that has been pre-trained with second training data; and an output step in which the respiratory rate detected in the respiratory rate detection step is output.

Description

Technical Field

[0001] The present invention relates to a respiration monitoring system that detects respiratory movements using tensor data obtained from a human body by a radar.

Background Art

[0002] As disclosed in Patent Documents 1 to 4, the respiration rate can be measured non - contact with a human body by using a radar.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Patent Document 3

Patent Document 4

Summary of the Invention

Problems to be Solved by the Invention

[0004] By using a radar, high - precision movements can be measured. However, even when a person is lying in bed, movements such as turning over occur, so the received radio waves do not necessarily reflect respiratory movements. In order to obtain radio waves that reflect respiratory movements even when there are movements in the human body, it is preferable to use radio waves transmitted from a plurality of transmitters and received by a plurality of receivers. However, using radio waves transmitted from a plurality of transmitters and received by a plurality of receivers results in an enormous amount of data. The inventors of the present invention found that respiratory - corresponding data exists in a group of two - dimensional graphs generated by reducing the dimension of the acquired tensor data using time - series data per unit time.

[0005] The present invention aims to provide a respiratory monitoring system that can accurately detect the respiratory rate using data obtained from radio waves transmitted from multiple transmitters and received by multiple receivers.

[0006] The respiratory monitoring system of the present invention as described in claim 1 is a respiratory monitoring system that detects respiratory movement using data obtained from a human body 2 in a lying-down position by radar 1, characterized in that a computer 3 performs the following steps: a tensor data acquisition step in which the data at a unit time measured at minute intervals, transmitted from a plurality of transmitters and received by a plurality of receivers, is formatted into tensor data; a tensor decomposition step in which the tensor data acquired in the tensor data acquisition step is reduced in dimension using the time series data of the unit time to generate a group of two-dimensional graphs; a respiratory data extraction step in which a respiratory movement model pre-trained with first training data is used to extract respiratory data from the group of two-dimensional graphs obtained in the tensor decomposition step; a respiratory rate detection step in which the respiratory rate is detected from the respiratory data extracted in the respiratory data extraction step using a respiratory rate model pre-trained with second training data; and an output step in which the respiratory rate detected in the respiratory rate detection step is output. The present invention as described in claim 2 is a respiratory monitoring system as described in claim 1, wherein the respiratory rate extracted in the respiratory rate detection step is stored together with acquisition time data obtained from the human body 2, the computer 3 performs a monitoring step in which it monitors the respiratory rate during a first predetermined period including a plurality of unit times, and in the monitoring step, if it is determined that the respiratory rate during the first predetermined period is below or above a predetermined number at a predetermined ratio, the output step outputs an alarm. The present invention as described in claim 3 is a respiratory monitoring system as described in claim 1, wherein the respiratory rate extracted in the respiratory rate detection step is stored together with acquisition time data obtained from the human body 2, the computer 3 performs a monitoring step in which it monitors the respiratory rate in a second predetermined period including a plurality of unit times and the respiratory rate in a third predetermined period including a plurality of unit times, and in the monitoring step, if it is determined that the respiratory rate in the third predetermined period is a change that exceeds a threshold with respect to the respiratory rate in the second predetermined period, the output step outputs an alarm. The present invention as described in claim 4 is a respiratory monitoring system as described in claim 1, wherein the respiratory data extracted in the respiratory data extraction step is stored together with acquisition time data obtained from the human body 2, the computer 3 performs a sleep stage prediction step in which it determines the sleep onset phase, non-REM sleep phase, and REM sleep phase from the respiratory data stored together with the acquisition time data using a sleep stage model that has been pre-trained with third training data, and in the output step, the sleep onset phase, non-REM sleep phase, and REM sleep phase determined in the sleep stage prediction step are output together with the acquisition time data. [Effects of the Invention]

[0007] According to the present invention, a group of 2D graphs is generated by reducing the dimensionality of acquired tensor data using time-series data per unit time, respiratory data is extracted from the generated group of 2D graphs using a respiratory motion model, and the respiratory rate is detected from the extracted respiratory data using a respiratory rate model. This makes it possible to accurately detect the respiratory rate using data obtained from radio waves transmitted from multiple transmitters and received by multiple receivers. [Brief explanation of the drawing]

[0008] [Figure 1] A flowchart illustrating computer processing in a respiratory monitoring system according to one embodiment of the present invention. [Figure 2] Equations showing tensor data and tensor decomposition [Figure 3] This figure shows the respiratory data extracted in the respiratory data extraction step. [Figure 4] Graph showing verification results [Modes for carrying out the invention]

[0009] The respiratory monitoring system according to the first embodiment of the present invention includes the following steps: a tensor data acquisition step in which a computer formats data measured at minute intervals in a unit time, transmitted from multiple transmitters and received by multiple receivers, into tensor data; a tensor decomposition step in which the tensor data acquired in the tensor data acquisition step is reduced in dimension using time-series data of a unit time to generate a group of two-dimensional graphs; a respiratory data extraction step in which a respiratory motion model pre-trained with first training data is used to extract respiratory data from the group of two-dimensional graphs obtained in the tensor decomposition step; a respiratory rate detection step in which the respiratory rate is detected from the respiratory data extracted in the respiratory data extraction step using a respiratory rate model pre-trained with second training data; and an output step in which the respiratory rate detected in the respiratory rate detection step is output. According to this embodiment, a group of 2D graphs is generated by reducing the dimensionality of the acquired tensor data using time-series data per unit time. From the generated group of 2D graphs, respiratory data is extracted using a respiratory motion model. The respiratory rate is then detected from the extracted respiratory data using a respiratory rate model. This allows for the accurate detection of the respiratory rate using data obtained from radio waves transmitted from multiple transmitters and received by multiple receivers.

[0010] A second embodiment of the present invention is a respiratory monitoring system according to the first embodiment, in which the respiratory rate extracted in the respiratory rate detection step is stored together with acquisition time data obtained from the human body, and a computer performs a monitoring step in which it monitors the respiratory rate in a first predetermined period including multiple unit times, and in the monitoring step, if it is determined that the respiratory rate in the first predetermined period is below or above a predetermined number at a predetermined ratio, an alarm is output in the output step. According to this embodiment, abnormalities in cardiopulmonary function can be detected at an early stage.

[0011] A third embodiment of the present invention is a respiratory monitoring system according to the first embodiment, in which the respiratory rate extracted in the respiratory rate detection step is stored together with acquisition time data obtained from the human body, and a computer performs a monitoring step in which it monitors the respiratory rate in a second predetermined period including multiple unit time and the respiratory rate in a third predetermined period including multiple unit time, and in the monitoring step it determines that the respiratory rate in the third predetermined period has changed to exceed a threshold with respect to the respiratory rate in the second predetermined period, and outputs an alarm in the output step. According to this embodiment, it is possible to predict viral and bacterial infections and changes in disease symptoms at an early stage.

[0012] A fourth embodiment of the present invention is a respiratory monitoring system according to the first embodiment, in which the respiratory data extracted in the respiratory data extraction step is stored together with acquisition time data obtained from the human body, and a computer performs a sleep stage prediction step in which it determines the sleep onset phase, non-REM sleep phase, and REM sleep phase from the respiratory data stored together with the acquisition time data using a sleep stage model that has been pre-machine-trained with third training data, and in the output step, the sleep onset phase, non-REM sleep phase, and REM sleep phase determined in the sleep stage prediction step are output together with the acquisition time data. According to this embodiment, health status and other factors can be predicted from sleep stages. [Examples]

[0013] A respiratory monitoring system according to an embodiment of the present invention will be described below. FIG. 1 is a flowchart showing computer processing in the respiratory monitoring system according to this embodiment, and FIG. 2 is an equation showing tensor data and tensor decomposition.

[0014] The respiratory monitoring system according to this embodiment detects a respiratory movement using data obtained from a human body 2 lying in bed by a radar 1. In this embodiment, a millimeter-wave radar is used as the radar 1, and data transmitted from three transmitters and received by four receivers is used. The data is recorded every 10 milliseconds over 60 seconds. Therefore, the number of 10-millisecond time points is 6000. At each time point, measurements of three transmitters (0, 1, 2) × four receivers (0, 1, 2, 3) are performed four times for 12 phases (0, ω,..., 11ω). Therefore, the number of points per time point is 3 × 4 × 4 = 48 points every 10 milliseconds. Each time point is a power spectrum of 256 frequencies with complex numerical values, and these are formatted as tensor data.

[0015] Note that in this embodiment, the unit time is 60 seconds and the micro time is 10 milliseconds. In the tensor data acquisition step S11, the computer 3 formats data transmitted from a plurality of transmitters, received by a plurality of receivers, measured every micro time, and obtained in unit time as tensor data.

[0016] Equation 1 shown in FIG. 2 represents the i3-th value of the i1-th transmitter and the i2-th receiver at the recording time of the i4-th 10-ms time point, the i5-th element of the complex number (i5 = 1: real number, i5 = 2: imaginary number), and the i6-th frequency.

[0017] In the tensor decomposition step S12, computer 3 generates a set of 2D graphs by reducing the dimensionality of the tensor data acquired in the tensor data acquisition step S11 using time series data per unit time.

[0018] Equation 2, shown in Figure 2, is an equation that demonstrates tensor decomposition. In practice, since obtaining the expanded matrix (Equation 3) and then obtaining the complete tensor decomposition takes time, we use Equation 4.

[0019] In the respiratory data extraction step S13, computer 3 uses a respiratory movement model that has been pre-trained using the first training data to extract respiratory data from the group of 2D graphs obtained in the tensor decomposition step S12. For the first training data, for example, data obtained from the polysomnography testing device Alice (Philips), as described below, can be used.

[0020] Figure 3 shows the respiratory data extracted in the respiratory data extraction step. As shown in Figure 3, the group of two-dimensional graphs obtained in tensor decomposition step S12 includes respiratory data. In this embodiment, data obtained from a human body in a supine position is used.

[0021] Figure 4 is a graph showing the verification results. In Figure 4, the horizontal axis represents the predicted target variable (Us), and the vertical axis represents the actual target variable. Figure 4(a) uses linear regression (Person correlation coefficient: 0.775, P=1.41×10⁻⁵), and Figure 4(b) uses Gaussian regression (Person correlation coefficient: 0.934, P=7.53×10⁻¹¹). The actual target variable used is the Alice data. The Alice data consists of three components: P.flow, THOR, and ABDM, recorded every 10 milliseconds, to record respiratory movements. The Alice data itself is composed of 6,000 10-millisecond time points.

[0022] In the respiratory rate detection step S14, computer 3 uses a respiratory rate model that has been pre-trained using second training data to detect the respiratory rate from the respiratory data extracted in the respiratory data extraction step S13. For the second training data, for example, respiratory rate data from electrocardiograms corresponding to the respiratory data shown in Figure 3 can be used. In output step S15, the respiratory rate detected in respiratory rate detection step S14 is output.

[0023] In this way, by generating a set of 2D graphs by reducing the dimensionality of the acquired tensor data using time-series data per unit time, extracting respiratory data from the generated set of 2D graphs using a respiratory motion model, and detecting the respiratory rate from the extracted respiratory data using a respiratory rate model, it is possible to accurately detect the respiratory rate using data obtained from radio waves transmitted from multiple transmitters and received by multiple receivers.

[0024] Computer 3 stores the respiratory rate extracted in the respiratory rate detection step S14 together with the acquisition time data obtained from the human body 2 (S16), and then executes the monitoring step S17. In monitoring step S17, the number of breaths during a first predetermined period including multiple unit times is obtained (S17a). If it is determined that the number of breaths during the first predetermined period falls below or exceeds a predetermined number at a predetermined ratio (Yes in S17b), an alarm is output in output step S18. In monitoring step S17, if the respiratory rate does not fall below or exceed a predetermined number of respiratory rates in a predetermined ratio relative to the respiratory rate in the first predetermined period (No in S17b), the respiratory rate in the next first predetermined period is monitored. This monitoring step S17 allows for the early detection of abnormalities in cardiopulmonary function.

[0025] Animals, including humans, maintain their activity by using oxygen in their cells to break down organic compounds, converting them into energy and carbon dioxide. In other words, for an individual or cell, respiratory movement is an essential movement that supplies the energy necessary for maintaining life. Therefore, the presence or absence of respiration is one of the most important events for determining life or death, and it is necessary to detect when the respiratory rate falls below a predetermined number at a predetermined rate. Furthermore, respiration is also important for checking cardiopulmonary function. The "respiratory movement" of an individual animal, including humans, is the movement to take in oxygen into the blood and expel carbon dioxide, and it consists of three elements of lung function: inhalation, exhalation, and gas exchange (exchange of oxygen and carbon dioxide). When the oxygen supply mechanism is impaired and the balance between oxygen absorption and carbon dioxide expulsion is disrupted, it is a condition called respiratory failure, which is observed in severe cases of all diseases, including senility. Therefore, it is necessary to detect when respiratory rates fall below or exceed a predetermined rate at a predetermined ratio.

[0026] In the monitoring step S19, computer 3 obtains the number of breaths during a second predetermined period including multiple unit times and the number of breaths during a third predetermined period including multiple unit times (S19a). If it determines that the number of breaths during the third predetermined period exceeds a threshold relative to the number of breaths during the second predetermined period (Yes in S19b), it outputs an alarm in the output step S20. In monitoring step S19, if the respiratory rate in the third predetermined period does not exceed the threshold relative to the respiratory rate in the second predetermined period (No in S19b), then the system monitors whether the respiratory rate in the next third predetermined period exceeds the threshold relative to the respiratory rate in the next second predetermined period.

[0027] This monitoring step S19 allows for early prediction of viral and bacterial infections and changes in disease symptoms. For example, in infectious diseases such as the common cold, influenza, and COVID-19, inflammation often generates heat, disrupting the balance of oxygen supply and causing abnormalities in respiratory function. If the inflammation extends to the lungs, the oxygen supply mechanism is impaired, leading to respiratory failure. In heart and vascular diseases, insufficient blood can be delivered to the lungs, disrupting the balance of oxygen supply and causing abnormalities in respiratory function. On the other hand, even in daily life, when there are rapid and significant changes in physical activity, such as sports, energy is required, and respiratory function increases. During activities that place a heavy load on the body, respiratory function is promoted, increasing the volume of inhalation, exhalation, and gas exchange per unit of time, and raising the respiratory rate and heart rate. Therefore, large changes in respiratory rate are an important indicator for evaluating the state of physical activity.

[0028] Furthermore, it is important for confirming the assessment of the patient's condition. Abnormalities in "respiratory movement" are observed in many diseases, including not only lung diseases but also heart diseases, and measuring respiratory movement is important for the early detection of these changes. Changes in respiratory movement are said to precede changes in oxygen saturation (SpO2), and are useful for the early detection of changes in the patient's condition. Quickly detecting large changes in respiratory rate can increase the chances of saving a patient's life. Thus, as respiratory movement measurement data from hospitalized patients is accumulated, it is expected that the accuracy of evaluating and diagnosing abnormalities in respiratory movement will improve, and the possibility of predicting sudden changes in the condition of hospitalized patients will increase. Therefore, large changes in respiratory rate are an important indicator.

[0029] Computer 3 stores the respiratory data extracted in the respiratory data extraction step S13 together with the acquisition time data from which the respiratory data was obtained from the human body 2 (S21), and then executes the sleep stage prediction step S22. In the sleep stage prediction step S22, a "sleep stage model" pre-trained using third-party training data is used to determine the sleep onset phase, non-REM sleep phase, and REM sleep phase from the respiratory data stored along with the acquisition time data. The sleep onset phase, non-REM sleep phase, and REM sleep phase determined in the sleep stage prediction step S22 are then output along with the acquisition time data (S23).

[0030] During sleep, respiratory movements are primarily controlled by the respiratory center. Therefore, in non-REM sleep (sleep stages N2 or N3), fluctuations in both tidal volume and respiratory frequency decrease significantly, resulting in very regular breathing. On the other hand, while REM and sleep onset phases exhibit irregular respiratory movements, they show different periodic changes. Furthermore, during REM sleep, diaphragmatic movement is maintained, but the activity of intercostal muscles and other accessory respiratory muscles decreases significantly, leading to a dominance of abdominal breathing and differences in changes between the rib cage and abdominal wall. Thus, because the patterns of respiratory movements differ during sleep onset, non-REM sleep, and REM sleep, a sleep stage model pre-trained using these patterns as third-party training data can be used to determine sleep onset, non-REM sleep, and REM sleep. By outputting the sleep onset, non-REM sleep, and REM sleep phases determined in sleep stage prediction step S22 along with the acquisition time data, health status and other factors can be predicted.

[0031] In sleep disorders, "central apnea" and "obstructive apnea" are problematic, and it is necessary to be able to distinguish between the qualitative changes during apnea. In "central apnea," the movement of both the thoracic and abdominal walls, as well as the ventilation volume, changes become small, and "respiratory movement" almost ceases. On the other hand, in "obstructive apnea," ventilation ceases, but the movement of the body walls continues, and is even enhanced. For the diagnosis of sleep apnea syndrome, it is essential to be able to identify the presence of "obstructive apnea" throughout the sleep phase, regardless of whether it is non-REM or REM sleep. On the other hand, "obstructive apnea" during REM sleep has a significant impact on complications (hypertension, fatal vascular damage, etc.), so it is necessary to distinguish it from the non-REM phase. By determining the sleep stage using the sleep stage prediction step S22 and monitoring in the monitoring steps S17 and S19, it may be possible to identify the presence or absence of "central apnea" or "obstructive apnea" at each stage. In this embodiment, data obtained from a person in a lying position was used, but it is possible that respiratory movements can be similarly detected using data obtained from a person in a seated position, such as while driving. [Industrial applicability]

[0032] According to the present invention, accurate respiratory rate can be detected using data obtained from radio waves transmitted from multiple transmitters and received by multiple receivers, enabling early prediction of cardiopulmonary abnormalities, viral and bacterial infections, and changes in disease conditions. Furthermore, health status can be predicted from sleep stages. [Explanation of Symbols]

[0033] 1. Radar (millimeter-wave radar) 2 human body 3 Computers

Claims

1. A respiratory monitoring system that detects respiratory movements using data obtained from the human body by radar, Computers A tensor data acquisition step involves formatting the data, which is transmitted from multiple transmitters and received by multiple receivers, and measured at minute intervals, into tensor data. A tensor decomposition step is performed to generate a group of 2D graphs by reducing the dimensionality of the tensor data acquired in the tensor data acquisition step using the time series data of the unit time, A respiratory data extraction step is performed to extract respiratory data from the group of 2D graphs obtained in the tensor decomposition step, using a respiratory movement model that has been pre-machine-trained with the first training data. A respiratory rate detection step that detects the respiratory rate from the respiratory data extracted in the respiratory data extraction step, using a respiratory rate model that has been pre-trained using second training data, An output step which outputs the respiratory rate detected in the respiratory rate detection step. Execute A respiratory monitoring system characterized by the following features.

2. The respiratory rate extracted in the respiratory rate detection step is stored together with the acquisition time data from which the respiratory rate was obtained from the human body. The aforementioned computer, A monitoring step is performed to monitor the number of breaths during a first predetermined period that includes multiple units of time. In the monitoring step, if it is determined that the number of respiratory respirations during the first predetermined period falls below or exceeds a predetermined number at a predetermined ratio, the output step outputs an alarm. The respiratory monitoring system according to feature 1.

3. The respiratory rate extracted in the respiratory rate detection step is stored together with the acquisition time data from which the respiratory rate was obtained from the human body. The aforementioned computer, A monitoring step is performed to monitor the number of breaths during a second predetermined period including a plurality of the aforementioned unit times, and the number of breaths during a third predetermined period including a plurality of the aforementioned unit times. If the monitoring step determines that the respiratory rate in the third predetermined period exceeds a threshold value relative to the respiratory rate in the second predetermined period, the output step outputs an alarm. The respiratory monitoring system according to feature 1.

4. The respiratory data extracted in the respiratory data extraction step is stored together with the acquisition time data from which the respiratory data was obtained from the human body. The aforementioned computer, Using a sleep stage model pre-trained with third training data, a sleep stage prediction step is performed to determine the sleep onset phase, non-REM sleep phase, and REM sleep phase from the respiratory correspondence data stored together with the acquired time data. In the output step, the sleep onset period, the non-REM sleep period, and the REM sleep period, as determined in the sleep stage prediction step, are output along with the acquisition time data. The respiratory monitoring system according to feature 1.

Images