WO2020179664A1

WO2020179664A1 - Blood sugar level measurement device, blood sugar level measurement method, and probe

Info

Publication number: WO2020179664A1
Application number: PCT/JP2020/008305
Authority: WO
Inventors: 廣瀬　明; セコウ長江
Original assignee: 国立大学法人東京大学
Priority date: 2019-03-01
Filing date: 2020-02-28
Publication date: 2020-09-10
Also published as: JPWO2020179664A1

Abstract

Provided is a blood sugar level measurement device 100 that measures blood sugar levels from measurement signals from a 3-port probe comprising a transmission port, detection port, and reference port for a waveguide that transmits electromagnetic waves. A detection port signal input unit 20 obtains detection port signals from the probe. A reference port signal input unit 30 obtains reference port signals from the probe. A correction unit 40 generates a corrected signal that has a reference port signal subtracted from a detection port signal. A complex neural network 60: uses a pair comprising a corrected signal and a glucose concentration as teacher data and learns weighting; and estimates glucose concentration as output data, by using the generated corrected signal as input data and calculating an output using a non-linear activation function for a weighted sum, in each layer. Provided is a probe comprising: a transmission port and detection port for a waveguide for electromagnetic waves; and a section that sandwiches the skin of a person, in a path from the transmission port in the waveguide and the detection port. A dielectric material is filled inside the waveguide.

Description

Blood glucose measuring device, blood glucose measuring method, and probe

The present invention relates to a blood glucose measurement technique.

In the conventional invasive blood glucose measurement method, it is necessary to collect blood from the human body, which places a heavy burden on the patient and has a high risk of infection. Therefore, a non-invasive blood glucose measuring method is desirable.

As a non-invasive method for measuring blood glucose level, there is a method for measuring blood glucose level based on the relationship between glucose concentration and permittivity because electromagnetic waves can penetrate human tissues.

Since human blood glucose levels are very low, very high sensitivity is required for non-invasive measurement methods.

In order to measure the blood glucose level using the transmitted wave transmitted through the tissues of the human body, waveguides are placed on both sides of the sample, and the amplitude and phase of the radio waves are acquired as data. In our previous research, we demonstrated that the measurement accuracy in the practical blood glucose range can be improved by learning the amplitude and phase data of the transmitted wave by using a millimeter-wave probe and using a complex neural network [16], [ 17].

However, the transmitted wave has ripple due to the influence of resonance, and there are cases where the amplitude and phase change of the transmitted wave are not monotonous with respect to the glucose concentration. Since such data instability affects the estimation accuracy of glucose concentration, it was necessary to design a more appropriate probe.

The present invention has been made in such a situation, and one of the exemplary purposes of the embodiment is to provide a technique for improving the measurement accuracy of blood glucose level.

One aspect of the present invention is a blood glucose level measuring device that measures a blood glucose level from a measurement signal from a probe including a transmission port, a detection port, and a reference port of a waveguide that transmits electromagnetic waves. This device calculates the difference between the detection port signal input unit that acquires the detection port signal from the probe, the reference port signal input unit that acquires the reference port signal from the probe, and the detection port signal and the reference port signal. A complex neural network in which weights are learned using a correction unit that generates a correction signal and a pair of correction signal and glucose concentration as training data, and the generated correction signal is used as input data to make the weighted sum non-linear in each layer. It includes a complex neural network that estimates the glucose concentration as output data by calculating the output by the activation function.

Another aspect of the present invention is a probe. This probe includes a transmission port and a detection port of an electromagnetic wave waveguide, and a portion that sandwiches the human skin in the path from the transmission port of the waveguide to the detection port, and a dielectric material is provided inside the waveguide. The material is filled. The dielectric material may be silicon. The probe may further include a reference port.

It should be noted that any combination of the above components and the conversion of the expression of the present invention between devices, methods, systems, recording media, computer programs, etc. are also effective as aspects of the present invention.

According to the present invention, the accuracy of measuring the blood glucose level can be improved.
Can be provided.

FIG. 1 (a) is a schematic diagram of how to use the 3-port probe, and FIG. 1 (b) shows the structure of the probe and the outer ear model. 2 (a), 2 (b), and 2 (c) show the amplitude, phase, and unwrapped phase of the detection port signal, respectively. 3 (a), 3 (b), and 3 (c) show the amplitude, phase, and unwrapped phase of the reference port signal, respectively. 4 (a) and 4 (b) show the difference in amplitude between the detection port signal and the reference port signal, and the difference in unwrapped phase, respectively. 1 shows a configuration of a complex neural network used in the embodiment of the present invention. 6 shows a teacher signal given to the complex neural network of FIG. 3 shows a taper structure of silicon filled in a waveguide. 8 (a) and 8 (b) show the moving average and unwrapped phase of the amplitude of the detection port signal measured by the probe not filled with silicon. 9 (a) and 9 (b) show the moving average and unwrapped phase of the amplitude of the detection port signal measured by the probe filled with silicon. It is a block diagram of a blood glucose level measurement system. It is a flowchart which shows the blood glucose level measurement procedure.

Hereinafter, the present invention will be described based on preferred embodiments with reference to the drawings. The same or equivalent components, members, and processes shown in the drawings shall be designated by the same reference numerals, and redundant description will be omitted as appropriate. Further, the embodiment is not limited to the invention but is an example, and all the features and combinations thereof described in the embodiment are not necessarily essential to the invention.

1. Introduction Diabetes is a global, life-threatening disease. Daily blood glucose control is necessary for diabetics. In the conventional measurement method, it is necessary to collect blood, which imposes a great burden on the patient. Therefore, a non-invasive measurement method is desirable. However, since the blood glucose level in humans is very low, the non-invasive measurement method requires very high sensitivity.

Since electromagnetic waves can penetrate human tissues, various methods have been proposed [1]. All of these methods are based on the relationship between glucose concentration and permittivity [2]-[4]. So far, several experiments using aqueous glucose solutions have been reported [5]-[7]. In recent studies, some groups have proposed methods for measuring blood glucose levels using patch antennas [8], [9]. The results showed the possibility of measuring blood glucose levels using electromagnetic waves.

In order to measure the blood glucose level using transmitted waves, waveguides are placed on both sides of the sample, and the amplitude and phase of the radio waves are acquired as data. Previous studies have proved that the amplitude and phase data of radio waves include information on glucose concentration [10]-[14]. Some research groups have conducted animal experiments using millimeter waves [9], [15]. The latest research has demonstrated that learning amplitude and phase data using complex neural networks can improve measurement accuracy in the practical blood glucose range [16], [17]. However, all the problems have not been solved yet, and there is a need for a method that can eliminate data instability. In this specification, we propose a new probe design to solve this problem.

2. Proposed method Figure 1(a) shows the conceptual diagram and structure of the probe we propose. This section describes frequency selection, tissue model and probe design.

2. 1 Frequency selection Previous studies have shown that high frequency millimeter waves are sensitive to changes in glucose concentration [18]. The complex permittivity ε _r (ω, χ) can be expressed as follows by the debai relaxation equation.

Here, ω is the angular frequency, χ is the concentration, ε _status (χ) and ε _∞ (χ) are the permittivity at very low and high frequencies, and τ (χ) is the relaxation time.

Previous studies have shown that a frequency range of 60 GHz to 80 GHz is appropriate [17]. The frequency range used in this paper is 55 GHz to 85 GHz.

2.2 Organizational model

We constructed a histological model of the outer ear and finger gaps for numerical analysis. The model is shown in FIG. 1 (b). Since the depth of millimeter waves that can penetrate the human body is about a few millimeters, only thin tissues such as ear lobes can penetrate. For modeling, it is assumed that thin human tissue consists only of skin, fat and blood. The dielectric properties of human tissue are so complex that it cannot be completely reproduced. In a previous study, we constructed a model of the dielectric properties of human tissue [19]. Its characteristics can be expressed by the following formula [19].

Here, Δε _n is Δε _n = (ε _status −ε _∞ ) n, α _n is a distribution parameter, σ _i is an ionic conductivity, and ε ₀ is a permittivity in free space.

Dielectric constants of dry and moist skin are different. The dielectric properties of dry skin are used herein. The dielectric properties of fat have also been shown in previous studies [19]. The treatment of blood is different from that of skin and fat. The model of blood needs to depend on glucose concentration. Another previous study constructed such a model. [20] The glucose concentration range applicable to this model is 70 to 150 mg/dL. The model can be expressed by the following formula [20].

Here, χ is a glucose concentration (70 to 150 mg/dL).

Generally, the outer skin of the tissue is the skin, and the fat is underneath. Blood is distributed in both skin and fat. The average thickness of the ear lobe is 5 to 6 mm. The outermost layer is the skin layer. The fat layer was 4 layers and the blood layer was 5 layers. The thickness of one fat layer was 0.75 mm, the thickness of one blood layer was 0.32 mm, and the thickness of the skin layer was 0.2 mm. The total thickness of the model is 5 mm.

2.3 Probe design In our previous study, we experimented with a glucose-aqueous solution in a petri dish [17]. Ripple was present in the experimental data. The cause is resonance of electromagnetic waves by the container. Appropriate probes are needed to solve the problem. We propose a probe with a dielectric in the waveguide.

2.3.1 Dielectric Material Human skin contains water that reflects millimeter waves. A suitable material is needed that matches the tissue impedance to reduce reflections. Here we choose silicon. Silicon has a relative permittivity of 11.9, which matches well with dry skin. Therefore, silicon is a suitable dielectric.

2.3.2 3-port probe Previous studies [8] and [17] used only the transmission coefficient. However, the problem is the instability of this transmission coefficient. Unstable data has a great effect on the concentration measurement results. In this specification, a 3-port probe is proposed to solve this problem. Since the frequency range used is 60 GHz to 80 GHz, the size of the waveguide is 4 mm×2 mm. FIG. 1(a) is a schematic view of a method of using a 3-port probe. Here, the probe is sandwiched between the ear lobes for measurement. The structure of the probe is also shown in FIG. The probe is composed of a transmission port (Port 1), a measurement (detection) port (Port 2), and a reference port (Port 3). The outer ear consists of a skin layer, a blood layer, and a fat layer. All probes are in direct contact with the skin to avoid air reflections. Our proposed probe can improve the data problem and increase the measurement accuracy.

2.4 Results of numerical analysis HFSS (Ansys Co.) was used for the numerical analysis. The result of the numerical analysis includes a transmission coefficient (transmission data) S ₂₁ from the transmission port to the detection port and a transmission coefficient (reference data) S ₃₁ from the transmission port to the reference port. The results are shown in FIGS. 2 and 3. FIGS. 2A and 2B and FIGS. 3A and 3B show amplitude and phase data of S ₂₁ and S ₃₁ , respectively. The horizontal axis of the graph is the frequency of millimeter waves (unit: GHz), and the vertical axis is amplitude (unit: dB) or phase (unit: rad). In order to see the phase difference, the phase data of S ₂₁ and S ₃₁ are unwrapped and a straight line is fitted based on the data. Then, by taking the difference between the unwrapped phase data and the straight line, the phase deviation between S ₂₁ and S ₃₁ can be obtained. 2 (c) and 3 (c) show the phase deviations of S ₂₁ and S ₃₁ . Here, unwrap has a periodicity in which the phase value goes around at 2π and returns, so to speak, it is "folded (wrapped)", but this is defined as 0, 2π, 4π, 6π, ... , Is to unfold this fold.

Looking at the phase deviation, it can be seen that the change with glucose concentration is non-monotonic. That is, in both cases of the S ₂₁ characteristic (transmission characteristic to the detection port) and the S ₃₁ characteristic (transmission characteristic to the detection port), the phase deviation monotonically increases in the measured frequency band as the glucose concentration increases. You can see that it is not. This non-monotonic change affects the estimation of glucose concentration and may lead to inaccurate estimation of glucose concentration. However, by taking the difference in the phase deviation between S ₂₁ and S _31, the change due to the difference in glucose concentration can be made monotonous. The result is shown in FIG. As with the phase, the difference (ratio) in amplitude between S ₂₁ and S ₃₁ is also taken. The result is shown in FIG. It can be seen that the differences in amplitude are monotonically arranged. That is, if the difference between S ₂₁ and S ₃₁ is calculated, it is understood that the difference in amplitude and the difference in phase deviation monotonically increase in the measured frequency band as the glucose concentration increases. By using the processed data (i.e. corrected data by subtracting the S ₃₁ signal (see port signal) from S ₂₁ signal (detection port signal)), higher accuracy of concentration estimation is obtained. These results showed that the reference port was effective for millimeter wave blood glucose measurement.

2.5 Conclusion In the present specification, we have proposed a 3-port probe for a millimeter-wave blood glucose measurement system. The results of the numerical analysis showed the effectiveness of using the reference port. As a result of the data processing, by using the reference port, the changes in phase and amplitude due to the difference in glucose concentration were successfully arranged monotonically. This probe is adaptable to various measurements. The data processed using the third port as a reference port is effectively used for estimating the glucose concentration.

3. Configuration of complex neural network Higher accuracy by learning a complex neural network by using a detection port signal corrected by a reference port signal as input data using a 3-port probe with a reference port and giving glucose concentration as training data. Glucose concentration can be estimated.

Complex neural networks (Complex-valued neural networks (CVNNs)) replace all real numbers in conventional neural networks with complex numbers [21] and have generalization ability in the complex domain. The j-th neuron in the l-th layer receives an input from the i-th neuron in the previous layer or the input terminal. The weight is defined by w _lji and has amplitude |w _lji | and phase θ _lji . The weighted sum is calculated by the following equation.

Here, the output of the neuron zlj is calculated by the following equation by the non-linear function fap.

A complex neural network is advantageous for applications that process electromagnetic fields because it can express amplitude and phase with a single complex number. By designing an optimal complex neural network, the complex neural network can reduce noise and improve the robustness of the system.

FIG. 5 shows the structure of a complex neural network used in the embodiment of the present invention. In this embodiment, as shown in FIG. 5, a configuration of a complex neural network having an input terminal, one hidden layer, and a single output neuron is used. Single-power neurons are used because they are suitable for expressing continuous concentration changes. In the previous study [17], the S ₂₁ signal (detection port signal) was given as the input signal of the complex neural network of FIG. 5, but in the embodiment of the present invention, the S ₂₁ signal (detection port signal) is the S ₃₁ signal. Note that the correction signal corrected by (reference port signal) is given as input data. A signal obtained by pre-processing the correction signal of S ₂₁ of the millimeter wave frequencies ω ₁ , ω ₂ ,..., ω _I is applied to the input terminal. For the pretreatment, the treatment shown in the previous study [17] is used.

FIG. 6 shows a teacher signal given to the complex neural network of the single output neuron of FIG. As shown in FIG. 6, the teacher signal is a unit vector having an angle in the range of 0 to π in the complex plane, and the concentration in the range of 0 mg / dL to 300 mg / dL is mapped to the angle from 0 to π. ..

As described above, in our previous study [17], S ₂₁ was given to the input of the complex neural network on the premise of a 2-port probe without a reference port, but in the embodiment of the present invention, a 3-port probe is used to make a complex. The difference is that the difference between S ₂₁ and S ₃₁ is given to the input of the neural network, but since the learning method of the complex neural network is the same as that of the previous study [17], refer to Reference [17] for the learning method and the learning result. The description is omitted here. In the embodiment of the present invention, by using the difference between S ₂₁ and S ₃₁ as the input signal, the influence of reflection can be removed from the input signal, the adverse effect of reflection is reduced, and the accuracy of concentration measurement is improved. be able to. This is because the effect of reflection is the same between the S ₂₁ signal and the S ₃₁ signal, so the reflection component is removed by taking the difference between S ₂₁ and S ₃₁ , while the component related to the glucose concentration in the blood of the human body. Is left.

4. Composition of Dielectric Material It is desirable to fill the waveguide with a dielectric material that matches the tissue impedance of the human body to reduce reflections. As a suitable dielectric material, silicon having a dielectric constant close to that of dry skin is considered as one of the candidates, but any dielectric material having a dielectric constant close to that of dry skin can be used. It can be used.

FIG. 7 shows a taper structure of silicon with which the waveguide is filled. When a millimeter wave is input to the waveguide using a transmitter, reflection also occurs at the boundary between silicon and air. To reduce this reflection, silicon is gradually tapered as shown in FIG. The permittivity changes from air to silicon. Thereby, the reflection at the boundary between silicon and air can be reduced.

8(a) and 8(b) show the moving average of the amplitude of the detection port signal (S ₂₁ signal) and the unwrapped phase measured by the probe not filled with silicon. Here, the experimental results using a 2-port probe without a reference port are shown. The results measured immediately after meal, 30 minutes after meal, and 1 hour after meal are shown. As shown in FIG. 8 (a), there is almost no difference in amplitude between 30 minutes after meal and 1 hour after meal, and as shown in FIG. 8 (b), the phases are immediately after meal, 30 minutes after meal, and 1 hour after meal. It can be seen that there is almost no difference between the two.

9(a) and 9(b) show the moving average of the amplitude of the detection port signal (S ₂₁ signal) and the unwrapped phase measured by the probe filled with silicon. The results measured with a 2-port probe without a reference port are shown 30 minutes after meal and 1 hour after meal. As shown in FIG. 9 (a), there is a significant difference in amplitude between 30 minutes after meal and 1 hour after meal, and as shown in FIG. 9 (b), the phase is 30 minutes after meal and 1 hour after meal. It can be seen that there is a significant difference. By using the waveguide filled with silicon in this way, it becomes possible to eliminate the influence of reflection and detect the difference in glucose concentration that should be originally detected with higher accuracy.

FIG. 10 is a block diagram of the blood sugar level measuring system. The blood glucose level measuring system includes a 3-port probe 10 and a blood glucose level measuring device 100.

The 3-port probe 10 includes the transmission port, the detection port, and the reference port as described above. The millimeter wave is input to the transmission port while changing the frequency of the millimeter wave in a predetermined frequency band, and output signals are acquired from the detection port and the reference port. The signal S ₂₁ measured at the detection port is supplied to the detection port signal input unit 20, and the signal S ₃₁ measured at the reference port is supplied to the reference port signal input unit 30.

The detection port signal input unit 20 gives the detection port signal of each frequency to the correction unit 40.

The reference port signal input unit 30 gives a reference port signal of each frequency to the correction unit 40.

The correction unit 40 generates a correction signal by calculating the difference between the amplitude of the detection port signal and the reference port signal, and gives the correction signal to the preprocessing unit 50.

The preprocessing unit 50 adjusts the correction signal to the input data format of the complex neural network 60 and gives it to the input terminal of the complex neural network 60.

The complex neural network 60 has weights learned in advance by the teacher data stored in the teacher data storage unit 70. As described in FIG. 6, the teacher data is a unit vector having an angle in the range of 0 to π in the complex plane, and the angle from 0 to π corresponds to the concentration in the range of 0 mg / dL to 300 mg / dL. There is.

The complex neural network 60 calculates the output of the hidden layer from the input data using the learned weights, outputs the output data by the weighted sum of the outputs of the hidden layer, and gives it to the glucose concentration output unit 80. The output data is an angle in the range of 0 to π.

The glucose concentration output unit 80 converts the output data into glucose concentration and outputs it. The output data indicating the angle in the range of 0 to π is mapped to the glucose concentration in the range of 0 mg / dL to 300 mg / dL and output.

FIG. 11 is a flowchart showing a blood glucose level measuring procedure.

While changing the millimeter wave frequency, the detection port signal of the 3-port probe is acquired (S10), and at the same time, the reference port signal is acquired (S20).

A correction signal is generated by calculating the difference between the detection port signal and the reference port at each frequency (S30).

The correction signal at each frequency is preprocessed to generate input data to the complex neural network (S40).

Convert input data to output data by a complex neural network with learned weights (S50).

The output data of the complex neural network is converted into glucose concentration and output (S60).

In the blood glucose level measuring system according to the embodiment of the present invention, the detection signal is corrected by the reference signal using the configuration of the 3-port probe provided with the reference port, and the calibration is performed to determine the amplitude and phase of the transmitted wave. It can be arranged monotonically with respect to the glucose concentration. Further, by filling the inside of the waveguide of the probe with a dielectric material such as silicon, it is possible to make a remarkable difference in the amplitude and phase of the transmitted wave with respect to the difference in glucose concentration. In addition, by learning using a complex neural network, the glucose concentration can be adaptively estimated from the amplitude and phase of the transmitted wave. By these means, the estimation accuracy of the glucose concentration using millimeter waves can be further improved.

As described above, two effective methods for eliminating the influence of reflection are to use a 3-port probe provided with a reference port and to fill the inside of the waveguide with a dielectric material such as silicon. Is. Depending on the part of the human body to which the probe is attached, it may not be possible to provide three reference ports by providing a reference port. In that case, means for filling the inside of the waveguide with a dielectric material such as silicon can be used. In addition, by implementing both means of providing a reference port to form a 3-port probe and filling the inside of the waveguide with a dielectric material, the effects of both are combined to further improve the measurement accuracy. Can be made to.

References
[1] C.-F. So, et al., "Recent advances in noninvasive glucose monitoring," Medical Devices, vol.5, pp.45.52, 2012.
[2] T. Karacolak, et al., "Cole.cole model for glucose-dependent dielectric properties of blood plasma for continuous glucose monitoring," Microwave and Optical Technology Letters, vol.55, no.5, pp.1160.1164, 2013 ..
[3] V. Meriakri, et al., "Dielectric properties of glucose solutions in the millimeter-wave range and a problem of glucose content in blood control," Microwave & Telecommunication Technology, 2005 15th International Crimean Conference, vol.2IEEE, pp .853.854 2005.
[4] V. Meriakri, et al., "Dielectric properties of water solutions with small content of glucose in the millimeter-wave band and the determination of glucose in blood," Physics and Engineering of Microwaves, Millimeter and Submillimeter Waves (MSMW) and Workshop on Terahertz Technologies, vol.2IEEE, pp.873.875 2007.
[5] DV Blackham and RD Pollard, "An improved technique for permittivity measurements using a coaxial probe," IEEE Transactions on Instrumentation and Measurement, vol.46, no.5, pp.1093.1099, 1997.
[6] M. Hofmann, et al., "A microwave sensing system for aqueous concentratrion measurements based on a microwave reflectometer," IEEE MTT-S, pp.1.3, 2012.
[7] Y. Nikawa and D. Someya, "Application of millimeter waves to measure blood sugar level," Asia-Pacific Microwave Conference, vol.3IEEE, pp.1303.1306 2001.
[8] S. Saha, et al., "A glucose sensing system based on transmission measurements at millimetre waves using micro strip patch antennas," Scientific Reports, vol.7, no.1, p.6855, 2017.
[9] H. Cano-Garcia, et al., "Reflection and transmission measurements using 60 GHz patch antennas in the presence of animal tissue for non-invasive glucose sensing," European Conference on Antennas and Propagation (EuCAP)IEEE, pp. 1.3 2016.
[10] H. Cano-Garcia, et al., "Detection of glucose variability in saline solutions from transmission and reflection measurements using V-band waveguides," Measurement Science and Technology, vol.26, no.12, p.125701, 2015.
[11] Y. Nikawa and T. Michiyama, "Blood-sugar monitoring by reflection of millimeter wave," Asia-Pacific Microwave Conference (APMC)IEEE, pp.1.4 2007.
[12] Y. Nikawa and S. Nakamura, "Microwave application in medical sensing," International Symposium on Medical Information and Communication Technology (ISMICT)IEEE, pp.131.133 2015.
[13] T. Kurabayashi, et al., "Reflection spectroscopy on solutions of biological materials in millimeter wave frequency," International Conference on Infrared, Millimeter, and Terahertzwaves (IRMMW-THz), 2015IEEE, pp.1.2 2015.
[14] PH Siegel, et al., "Compact non-invasive millimeter-wave glucose sensor," Infrared, Millimeter, and Terahertz waves (IRMMW-THz), 2015 40th International Conference on IEEE, pp.1.3 2015.
[15] PH Siegel, et al., "Millimeter-wave non-invasive monitoring of glucose in anesthetized rats," International Conference on Infrared, Millimeter, and Terahertz waves (IRMMW-THz) IEEE, pp.1.2 2014.
[16] S. Hu and A. Hirose, "Proposal of millimeter-wave adaptive glucose-concentration estimation system using complex-valued neural networks," IGARSS 2018 --2018 IEEE International Geoscience and Remote Sensing Symposium, pp.4074.4077, 07 2018.
[17] S. Hu, S. Nagae, and A. Hirose, "Millimeter-wave adaptive glucose concentration estimation with complex-valued neural networks," IEEE Transactions on Biomedical Engineering, vol. 66, no. 7, pp 2065-2071 , July 2019.
[18] M. Hofmann, et al., "Non-invasive glucose monitoring using open electromagnetic waveguides," European Microwave Conference (EuMC)IEEE, pp.546.549 2012.
[19] S. Gabriel, RW Lau, and C. Gabriel, "The dielectric properties of biological tissues: III. Parametric models for the dielectric spectra of tissues," Physics in Medicine and Biology, vol.41, pp.2271.2293, Nov . 1996.
[20] B. Freer, "Feasibility of a non-invasive wireless blood glucose monitor,"Master's thesis, Rochester Institute of Technology, 2011. https://scholarworks.rit.edu/theses/7149
[21] A. Hirose, Complex-valued neural networks. Springer Science & Business Media, 2006.

The present invention can be used for blood sugar level measurement.

10 3-port probe, 20 detection port signal input unit, 30 reference port signal input unit, 40 correction unit, 50 preprocessing unit, 60 complex neural network, 70 teacher data storage unit, 80 glucose concentration output unit, 100 blood sugar level measuring device ..

Claims

A blood glucose level measuring device for measuring a blood glucose level from a measurement signal from a probe having a waveguide transmission port for transmitting electromagnetic waves, a detection port and a reference port,
A detection port signal input section for obtaining a detection port signal from the probe,
A reference port signal input unit that acquires a reference port signal from the probe,
A correction unit that generates a correction signal by calculating the difference between the detection port signal and the reference port signal,
A complex neural network in which weights are learned using a pair of correction signal and glucose concentration as training data, and the output by the non-linear activation function of the weighted sum is calculated in each layer using the generated correction signal as input data. A blood glucose level measuring device including a complex neural network that estimates a glucose concentration as output data.
The blood glucose level measuring device according to claim 1, wherein a dielectric material is filled inside the waveguide of the probe.
The blood glucose level measuring device according to claim 2, wherein the dielectric material filled in the waveguide of the probe has a tapered structure.
Electromagnetic wave waveguide transmission port and detection port,
The path from the transmission port to the detection port of the waveguide is provided with a portion that sandwiches the skin of the human body.
A probe in which a dielectric material is filled inside the waveguide.
The probe according to claim 4, wherein the dielectric material is silicon.
The probe according to claim 4 or 5, further comprising a reference port.
A blood glucose level measuring method for measuring a blood glucose level from a measurement signal from a probe having a waveguide transmission port for transmitting electromagnetic waves, a detection port and a reference port,
The step of acquiring the detection port signal from the probe and
Obtaining a reference port signal from the probe,
Generating a correction signal by calculating the difference between the detection port signal and the reference port signal,
Using a complex neural network whose weights have been learned using the pair of correction signal and glucose concentration as training data, the output by the non-linear activation function of the weighted sum is calculated for each layer using the generated correction signal as input data. A method for measuring a blood glucose level, which comprises a step of estimating a glucose concentration as output data.
A program for measuring a blood glucose level from a measurement signal from a probe having a transmission port of a waveguide for transmitting electromagnetic waves, a detection port and a reference port,
Obtaining a detection port signal from the probe,
Obtaining a reference port signal from the probe,
Generating a correction signal by calculating the difference between the detection port signal and the reference port signal,
Using a complex neural network whose weights have been learned using the pair of correction signal and glucose concentration as training data, the output by the non-linear activation function of the weighted sum is calculated for each layer using the generated correction signal as input data. A program characterized by having a computer perform a step of estimating a glucose concentration as output data.