Miniature communication and sensing integrated terahertz non-invasive blood glucose monitoring system
By combining terahertz total reflection spectroscopy with deep residual networks, non-invasive and real-time blood glucose monitoring has been achieved, solving the problem of real-time high-precision monitoring in existing technologies and promoting the development of terahertz wearable devices.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- THE 54TH RESEARCH INSTITUTE OF CHINA ELECTRONICS TECHNOLOGY GROUP CORPORATION
- Filing Date
- 2023-04-26
- Publication Date
- 2026-05-12
AI Technical Summary
Existing blood glucose testing technologies cannot achieve non-invasive, real-time, and high-precision monitoring, and are difficult to integrate with communication systems, thus failing to meet the real-time assessment and treatment needs of chronic or acute patients.
This method employs terahertz total reflection spectroscopy combined with a miniaturized detection chip and a deep residual network to achieve non-invasive blood glucose monitoring through the interaction between terahertz waves and human skin. It also utilizes the deep residual network for blood glucose value prediction and integrates a communication system for real-time data transmission.
It enables non-invasive, real-time, and continuous blood glucose monitoring, improving the accuracy and convenience of testing. It is applicable to a variety of wearable devices and meets the needs of smart healthcare.
Smart Images
Figure CN116458878B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of terahertz human characteristic monitoring technology, and in particular to an integrated communication and sensing system in the terahertz frequency band, which is applicable to non-invasive human blood glucose monitoring. Background Technology
[0002] Blood glucose is an important energy indicator for the human body. Maintaining normal physiological levels is a prerequisite for daily and high-intensity physical activities, mental state, and rational judgment. However, these indicators are affected by individual dietary intake, physical exertion, trauma, infection, stress, or extreme external environments. Excessive or insufficient levels of any of these indicators can lead to fainting, sudden loss of vision or hearing, severe stress, infection, or trauma. Real-time monitoring of blood glucose levels allows for objective assessment of physical condition and illness, especially for timely evaluation and treatment of chronic or acute cases, thereby further improving the emergency response capabilities of the medical system. In recent years, with the rapid development of smart healthcare and remote communication technologies, portable non-invasive monitoring devices have gradually been applied to the civilian market. However, blood glucose testing still requires a blood glucose meter, fine needle, and test strips, involving three steps: finger puncture, blood collection, and blood glucose testing. Professional nurses need at least one minute to obtain blood glucose results, making real-time monitoring impossible. Therefore, developing non-invasive blood glucose monitoring with remote transmission capabilities is crucial. Currently, non-invasive blood glucose testing products have appeared on the market. They generally use the principle of interaction between the human body surface and traditional electromagnetic wave frequencies such as infrared, microwave or thermal radiation. However, they are far from meeting the requirements of real-time and accurate monitoring in terms of environmental constraints, result waiting time, and measurement accuracy. Summary of the Invention
[0003] The technical problem this invention aims to solve is addressing the need for real-time monitoring of blood glucose as a vital sign. It addresses the issues of existing detection methods being non-real-time, invasive, and difficult to integrate. The invention proposes a non-invasive detection module based on terahertz wave total reflection spectroscopy. It breaks through key technologies such as detection techniques based on the correlation between terahertz waves and human epidermis and internal health indicators, terahertz miniaturized detection chip technology, low-loss, high-precision hybrid integration process for miniaturized terahertz detection device packaging technology, and prediction methods based on deep residual networks. Employing fully domestically produced processes, the invention achieves the development of a miniaturized prototype, completes large-sample medical data comparison, and realizes high-accuracy, miniaturized, multi-data terahertz non-invasive vital sign detection, thus promoting the rapid development of terahertz wearable devices.
[0004] The present invention adopts the following technical solution:
[0005] A miniature terahertz non-invasive blood glucose monitoring system integrating communication and sensing includes a transmitter chip, a detector communication chip, and a remote processing system;
[0006] The transmitting chip includes a resonant tunneling oscillator and an on-chip antenna, serving as a terahertz transmitting front-end;
[0007] The detection and communication chip is used to receive signals reflected back from human skin to the receiving end, measure changes in electromagnetic field intensity, and thus obtain spectral information related to the water content of the skin surface; and wirelessly transmit the spectral information to the remote processing system.
[0008] Based on the terahertz measurement experiment, the remote processing system acquires a large amount of dataset information. After processing the dataset through data preprocessing and clustering methods, it uses a deep residual network to train a blood glucose prediction model.
[0009] The transmitting chip emits a terahertz band detection wave towards the human skin, and the detection and communication chip receives the spectral information reflected back from the human skin and wirelessly transmits the spectral information to the remote processing system; the remote processing system collects the received spectral information, processes the data and obtains the final blood glucose prediction model.
[0010] Furthermore, the data processing flow of the remote processing system specifically includes step 1, data preprocessing, and step 2, prediction model construction;
[0011] The specific steps of the data preprocessing are as follows:
[0012] Step 101: The received spectral information is converted into photoelectric data, and the converted data is saved as a CSV file. Then, the data is normalized.
[0013] Step 102: Read the data and extract its features, converting it into an image matrix with width W, height H, and dimension 3. The corresponding true blood glucose values are then converted into W×H label images.
[0014] The specific steps for constructing the prediction model are as follows:
[0015] Step 201: Input the W×H×3 measurement data image matrix into the model, obtain the blood glucose prediction value with dimension W×H through model mapping, and calculate the difference between it and the label image, which is called the cross-entropy loss function.
[0016] Step 202: Optimize the model parameters based on the cross-entropy loss function calculation results, train the model using a deep residual network, continuously improve the model's prediction accuracy, and finally form a corresponding functional relationship between the input value and the blood glucose value, thereby achieving accurate prediction of blood glucose values.
[0017] Furthermore, each dimension of the label image corresponds to an input feature: the first dimension represents the voltage change, the second dimension represents the noise floor during the test, and the third dimension represents the temperature.
[0018] Furthermore, it also includes a lens assembly for receiving terahertz signals and for attenuating total internal reflection at the upper surface of the lens and the part to be measured, and then transmitting the reflected signal.
[0019] Furthermore, the lens group comprises lenses or lens assemblies supported by materials including but not limited to high-resistivity silicon.
[0020] The advantages of this invention compared to the prior art are:
[0021] (1) By utilizing the terahertz carrier wave to map the correlation between biological signs and skin surface water content, a non-invasive, real-time, and continuous sign detection method is achieved by adopting the interdisciplinary approach of semiconductor physics, microelectronics, and biology. It is easy to integrate with communication systems and other types of sensing systems and expand its applications.
[0022] (2) A miniaturized non-invasive detection chip is used to realize an active radiation chip for single-tube terahertz wave based on quantum tunneling mechanism, with a radiation frequency greater than 100 GHz. At the same time, the chip is further used to realize terahertz wave detection, realizing a highly integrated function of "one chip for two purposes".
[0023] (3) By adopting a low-loss, high-precision polymer packaging process, a high-performance, high-reliability micro-modular system for miniaturized non-invasive vital sign detection modules is realized, which is suitable for various wearable applications such as wristbands, fabrics, and helmets, and meets the urgent needs of intelligent medical development. Attached Figure Description
[0024] Figure 1 This is a structural diagram of a miniature terahertz vital signs monitoring system according to an embodiment of the present invention.
[0025] Figure 2 The following is the training process for the blood glucose prediction model in an embodiment of the present invention;
[0026] Figure 3 This is a schematic diagram illustrating the usage of the detection device in an embodiment of the present invention.
[0027] In the diagram: 1. Transmitting chip, 2. Lens assembly, 3. Detection and communication chip. Detailed implementation method:
[0028] The present invention will now be described in further detail with reference to the accompanying drawings.
[0029] This invention proposes a real-time vital sign monitoring system suitable for daily use, which uses the THz frequency band (electromagnetic waves in the range of 0.3-30THz) in the electromagnetic spectrum for non-invasive body surface monitoring.
[0030] Compared to optical and microwave frequencies, using terahertz waves for non-invasive vital sign detection has the following advantages:
[0031] (1) Terahertz waves are completely non-ionized, and their photon energy is low, which is harmless to most biological cells and is suitable for continuous long-term uninterrupted characteristic monitoring of the human body.
[0032] (2) Terahertz waves are penetrable and can penetrate various clothing such as cotton, linen, and down, enabling them to play an important role in fields such as remote wireless monitoring of human body surface signs.
[0033] (3) Terahertz waves are excellent broadband information carriers. In addition to improving detection sensitivity, they can also be extended to become important applications of integrated communication and sensing such as body area networks.
[0034] (4) Terahertz waves have a high signal-to-noise ratio in the time domain spectrum. Through sampling measurement technology, the interference of background radiation noise can be effectively prevented, which makes terahertz technology very suitable for detection applications.
[0035] Therefore, terahertz non-invasive blood glucose detection technology with independent intellectual property rights in my country can promote the application of terahertz technology in detection, communication, and integrated sensing fields. Vital sign detection is the scientific basis for doctors to diagnose diabetes and formulate and adjust treatment plans. Non-invasive vital sign detection technology can meet the needs of painless, frequent testing, and real-time monitoring, without causing patients the pain of blood collection or the risk of wound infection, which is beneficial for better control of the condition of both chronic and acute patients.
[0036] Terahertz technology has made significant progress in many fields such as imaging, detection, and wireless communication. There will be a huge market demand for miniaturized terahertz sources and terahertz detectors. Portable, miniaturized, room-temperature operable, and easily integrated terahertz vital sign detection chips will be one of the important development directions in the future bioelectronics field.
[0037] Therefore, research on miniature terahertz non-invasive blood glucose detection technology has an urgent application need for assessing physical condition and injury status, evaluating patient conditions online, and dynamically deploying treatment plans. At the same time, this technology has important research significance for expanding its application in the medical system and for cross-integration in the field of bioelectronics. Therefore, it is urgent to carry out the research work of this project, conduct medical sample comparison during the project development cycle, and complete the effectiveness evaluation of online monitoring.
[0038] like Figure 1 As shown, a miniature communication and sensing integrated terahertz non-invasive vital sign monitoring system includes a transmitting chip, a lens group, a detection chip, and a remote data processing system.
[0039] The system employs terahertz total internal reflection spectroscopy for non-invasive vital sign detection. A detection chip receives signals reflected from human skin and measures changes in electromagnetic field intensity to obtain spectral information related to skin surface moisture content. A resonant tunneling oscillator and on-chip antenna are used to independently implement the terahertz transmitting front-end. The receiving antenna and diode are integrated using low-loss, high-precision micro-packaging technology. Both transmission and detection are implemented using RTD chips, ultimately packaged into an independent vital sign detection module. This module includes communication components, allowing unreceived signals to be directly and wirelessly transmitted to the signal processing unit.
[0040] Terahertz attenuating total internal reflection lenses are lenses or lens combinations made of, but not limited to, high-resistivity silicon materials, used to receive terahertz signals and attenuate total internal reflection at the upper surface of the lens and the part to be measured, and then transmit the reflected signal as a medium.
[0041] It uses terahertz total reflection spectroscopy to achieve non-invasive detection of vital signs.
[0042] Changes in blood glucose levels in the human body cause physiological changes in cells, tissues, and skin. The cell membrane is a leakage medium; under any circumstances, such as exercise or diet, it can affect the cell membrane potential and the values of potassium, calcium, magnesium, and sodium ions on both sides, thus affecting the overall water distribution. Specifically, when blood glucose levels rise, the water content on the skin surface decreases, and this change in skin water content causes a change in the skin's complex refractive index.
[0043] At such high frequencies as terahertz, the dielectric properties of many media are no longer single and constant, but rather frequency-dependent, such as water, solutions, and biological tissues. Therefore, based on the principle that dielectric polarization under high-frequency electromagnetic waves requires a certain relaxation time, Debye established the Debye model to describe the single relaxation effect of a medium. This relaxation effect mainly manifests as molecular translation and rotational diffusion, and hydrogen bond recombination. When terahertz waves act on the skin surface, if the water content in the skin tissue changes, the dielectric properties of the skin tissue obtained by the Debye model will differ. To better reflect the dielectric properties of skin in the terahertz band, a double Debye model can be used, considering both the slow relaxation effects of hydrogen bond breaking and formation, and the fast relaxation effects of hydrogen bond network recombination. The dielectric properties of skin depend on its complex refractive index, which can be analytically determined. The relationship between the complex refractive index and the dielectric constant of skin is as follows:
[0044]
[0045] Where: n 样 ε is the complex refractive index of the skin; 样 denoted as ν, where n is the dielectric constant of the skin; ν is the refractive index of the skin; and k is the extinction coefficient.
[0046] The real part of the skin's complex refractive index is the refractive index itself, which increases with increasing water content. Similarly, the imaginary part of the skin's complex refractive index is the extinction coefficient, which also increases with increasing water content. Since the water content of human skin is affected by the concentration of blood glucose in the body, the skin's water content can be determined by detecting its refractive index and extinction coefficient, and thus, changes in blood glucose levels can be predicted.
[0047] According to the reflection spectrum formula
[0048]
[0049] In the formula: R prism F is the relative reflectivity; F{} is the Fourier transform; E 样 (t) and E 参 (t) represents the terahertz time-domain signal when the finger is on and off the prism, respectively; r 样 and r 参 The Fresnel reflection coefficients are given when a finger is placed on and off the prism, respectively. Where r... 样 and r 参 It can also be expressed as
[0050]
[0051]
[0052] Where: n y and n l γ represents the refractive index of the finger skin and the refractive index of the prism, respectively; θ is the incident angle of the terahertz wave at the contact surface between the prism and the finger; y The terahertz wave exits at the contact surface between the prism and the finger; n k γ is the air refractive index of the prism base without a finger; k The terahertz wave exits at the interface between the prism and the air at an angle of incidence.
[0053] According to Fresnel's law:
[0054]
[0055]
[0056] Organize, get
[0057]
[0058] In the formula: ε′ is the real part of the complex permittivity, that is, the relative permittivity of the sample; ε″ is the imaginary part of the complex permittivity, which represents the loss generated by the repeated polarization of the sample.
[0059] The non-invasive blood glucose detection system using terahertz attenuated total internal reflection spectroscopy receives signals reflected back from human skin via a detection chip 3, measures changes in electromagnetic field intensity, and thus obtains spectral information related to skin surface moisture content. A resonant tunneling oscillator and an on-chip antenna are used to independently implement the terahertz transmitting front-end. The receiving antenna and diode are integrated using low-loss, high-precision micro-packaging technology. Both the transmitting chip 1 and the detection / communication chip 3 are implemented as RTD chips, ultimately packaged into an independent detection module. The detection chip employs a low-loss, high-precision hybrid integration method. By integrating various functional substrates and chips with low loss and high precision, the system achieves its function. A single InP resonant tunneling oscillator chip can realize an integrated transmitting source or detector, a feature previously limited to discrete devices.
[0060] The chip is an active radiation chip that realizes single-tube radiation of terahertz waves based on the quantum tunneling mechanism, with a radiation frequency greater than 100 GHz. At the same time, the chip is further used to realize the detection of terahertz waves, realizing a highly integrated function of "one chip for two purposes". It is the first fully integrated terahertz chip in the field of human vital sign detection.
[0061] The quantum resonant tunneling chip is one of the core components of this technology. Its function is to realize an integrated transmitter source of traditional discrete devices through an InP resonant tunneling oscillator chip. The RTO is a combination of an RTD device and a slot antenna, including: an RTD device, a reflector, a slot antenna, a vibration damping resistor, and a heat sink.
[0062] Based on the previously proposed terahertz blood glucose measurement method and medical theory, a crucial step in this project is to achieve a mapping from terahertz measurement experimental values to actual blood glucose values by collecting a large number of terahertz measurement experimental values and actual blood glucose values from individuals. To achieve this goal, the remote data processing system employs a deep residual network-based approach for vital sign prediction, as follows: Figure 2 As shown, based on the terahertz measurement experiment, a large amount of dataset information was obtained. The dataset was processed through data preprocessing and clustering methods to solve the problems of multi-dimensional and individual differences in the data, thereby improving the accuracy of subsequent model training. A deep residual network was used for training, and then a vital sign prediction model was developed to solve the mapping problem from experimental measurements to real vital sign values.
[0063] Data processing employs a deep learning-based blood glucose prediction method, which can be divided into two parts: data preprocessing and model building.
[0064] To better train the model and achieve more accurate prediction results, the acquired data needs to be preprocessed. First, the data storage format is addressed; for easier and faster data retrieval, the data is uniformly saved as a CSV file. Next, the data is normalized. Although deep learning frameworks introduce bias to avoid the impact of different feature measurement scales on model accuracy, when the value ranges of two features differ too much, it can affect the model's convergence speed.
[0065] Finally, data transformation is performed. After reading the data, in order to better extract features from the input model, it is transformed into an image matrix with a width of W, a height of H, and a dimension of 3. The corresponding blood glucose true value is transformed into a W×H label image. The specific values of W and H are adjusted according to the size of the dataset. Each dimension of the image corresponds to an input feature. The first dimension represents the change in voltage, the second dimension represents the noise floor during the test, and the third dimension represents the temperature. This transforms the problem of processing a single data point into the problem of processing an image. Such transformation has the following advantages: (1) When adding constraints later, only the dimensions need to be changed at the two positions of data transformation and model data input, which facilitates further research and allows different weights to be assigned to each dimension so that different constraints have different constraints; (2) By combining the data together and inputting it into the model for processing, it is beneficial to extract the common features of the data, which can help the model converge faster; (3) At present, the research in the field of image processing is relatively comprehensive and in-depth. After converting the data into image-like data, we can refer to the relevant research in the field of deep learning image processing to further improve the model effect.
[0066] After data preprocessing, the final model is built. The residual network consists of vertically stacked residual blocks. Based on ordinary convolution, it addresses network degradation by using identity mapping as short-circuit connections. Identity mapping is not data-driven and requires no weight parameter control.
[0067] Deep networks can integrate features from high-, medium-, and low-depth networks, thus enriching the feature hierarchy. Residual networks can avoid the gradient explosion and gradient vanishing problems that occur when the network depth increases. Therefore, blood glucose value prediction can be achieved by constructing deep residual network models.
[0068] Overfitting can occur during model training, meaning the model has a small loss function and high prediction accuracy on training data, but a large loss function and low prediction accuracy on test data. To avoid this, the Dropout algorithm is used during training. In each training batch of a deep neural network, Dropout significantly reduces the probability of overfitting by ignoring half of the feature detectors (i.e., hidden layer nodes with values of 0). During forward propagation, it causes the activation values of certain neurons to stop working with a certain probability, reducing the model's dependence on certain local features and improving its generalization ability.
[0069] Model training process: Input the W×H×n (n is 3 here) measurement data image matrix into the model, obtain the blood glucose prediction value with dimension W×H through model mapping, calculate the difference between it and the label image (the blood glucose value image, which also has dimension W×H), called the cross-entropy loss function, and then optimize the model parameters based on the loss function calculation results (using a deep residual network for training), continuously improve the model prediction accuracy, and finally form a corresponding functional relationship between the input value and the blood glucose value, which is then solidified into the remote system to achieve accurate prediction of blood glucose value.
[0070] When the subject places a part of their body against the upper surface of the lens, the detection chip of the testing module converts the received terahertz spectral signal into an electrical signal and transmits it to the remote system. The remote system (which can be a mobile phone or other terminal device) calculates the subject's blood glucose value based on the fixed corresponding function relationship.
[0071] Through the above steps, this paper can realize the construction of a blood glucose prediction model based on deep learning. This model is then embedded in a remote monitoring terminal, and the monitoring device can accurately predict blood glucose levels when a part of the human body comes into contact with the detection module.
[0072] The resulting miniature communication and sensing integrated terahertz non-invasive blood glucose monitoring system can be used with either wired or remote monitoring. Figure 3 Four optional terahertz non-invasive blood glucose meter detection methods are presented.
Claims
1. A miniature communication and sensing integrated terahertz non-invasive blood glucose monitoring system, characterized in that, It includes a transmitting chip, a detection and communication chip, a lens group, and a remote processing system; the lens group is used to receive terahertz signals and attenuate total reflection at the upper surface of the lens and the part to be measured, and then transmit the reflected signal; The transmitting chip includes a resonant tunneling oscillator and an on-chip antenna, serving as a terahertz transmitting front-end; The detection and communication chip is used to receive signals reflected back to the receiver from human skin, measure changes in electromagnetic field intensity, and thus obtain spectral information related to the water content of the skin surface. And wirelessly transmit the spectral information to a remote processing system; Based on the terahertz measurement experiment, the remote processing system acquires a large amount of dataset information. After processing the dataset through data preprocessing and clustering methods, it uses a deep residual network to train a blood glucose prediction model. The transmitting chip emits a terahertz band detection wave into the human skin, and the detection and communication chip receives the spectral information reflected back from the human skin and wirelessly transmits the spectral information to the remote processing system; the remote processing system collects the received spectral information, processes the data and obtains the final blood glucose prediction model. The data processing flow of the remote processing system specifically includes step 1, data preprocessing, and step 2, prediction model construction. The specific steps of the data preprocessing are as follows: Step 101: The received spectral information is converted into photoelectric data, and the converted data is saved as a CSV file. Then, the data is normalized. Step 102: Read the data and extract its features, converting it into an image matrix with width W, height H, and dimension 3. The corresponding true blood glucose values are then converted into W×H label images. The specific steps for constructing the prediction model are as follows: Step 201: Input the W×H×3 measurement data image matrix into the model, obtain the blood glucose prediction value with dimension W×H through model mapping, and calculate the difference between it and the label image, which is called the cross-entropy loss function. Step 202: Optimize the model parameters based on the cross-entropy loss function calculation results, train the model using a deep residual network, continuously improve the model's prediction accuracy, and finally form a corresponding functional relationship between the input value and the blood glucose value, thereby achieving accurate prediction of blood glucose values.
2. The miniature communication and sensing integrated terahertz non-invasive blood glucose monitoring system according to claim 1, characterized in that, Each dimension of the labeled image corresponds to an input feature: the first dimension represents the change in voltage, the second dimension represents the noise floor during the test, and the third dimension represents the temperature.
3. The miniature communication and sensing integrated terahertz non-invasive blood glucose monitoring system according to claim 1, characterized in that, The lens group consists of lenses or lens combinations supported by materials including but not limited to high-resistivity silicon.