Medical examination device and monitoring algorithm for portable non-invasive continuous blood glucose monitoring

Abstract

The application discloses a medical examination equipment and a monitoring algorithm for portable noninvasive blood glucose continuous monitoring, relates to the technical field of medical monitoring, and the algorithm collects blood glucose related physical signals through a quantum dot enhanced sensor and a metabolic resonance exciter of a physical signal collection layer, processes the signals through a quantum signal processing layer, processes the signals through a spatiotemporal graph convolutional neural network which contains a spatiotemporal graph convolution module, a metabolic kinetics modeling module and a personalized adaptive layer in an intelligent algorithm core, and outputs blood glucose values and confidence levels, the equipment integrates key components and has a dynamic calibration mechanism, the algorithm core outputs the blood glucose values and the confidence levels to a calibration decision module, whether to trigger calibration is determined according to the confidence level, a micro flow control system cooperates with tissue fluid collection and electrochemical detection to realize parameter optimization, the overall design is a portable structure, daily use is facilitated, and precise noninvasive blood glucose continuous monitoring is realized.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring technology, and more specifically, to portable, non-invasive, continuous blood glucose monitoring medical testing equipment and monitoring algorithms. Background Technology

[0002] In the field of medical testing, blood glucose monitoring is of paramount importance for the diagnosis, treatment, and management of diseases such as diabetes. Traditional blood glucose monitoring methods mainly rely on invasive testing methods such as finger-prick blood sampling. While this method can provide relatively accurate blood glucose values, it also has some obvious limitations. On the one hand, frequent blood sampling causes physical pain and psychological stress for patients, especially for diabetic patients who need long-term blood glucose monitoring, where this pain is continuous and difficult to ignore. On the other hand, invasive testing can only provide single-point blood glucose data and cannot achieve continuous, real-time monitoring of blood glucose levels, making it difficult to capture dynamic changes in blood glucose in daily life. This is extremely detrimental to precise disease management and treatment plan adjustments.

[0003] With the continuous advancement of technology, non-invasive continuous glucose monitoring (NIG) has become a research hotspot and direction. This technology aims to achieve continuous glucose monitoring through non-invasive methods, such as utilizing physical signals on the human body surface, to overcome the shortcomings of traditional methods. However, some existing NUGs still have many problems in terms of accuracy, reliability, and individual adaptability. For example, some technologies are sensitive to environmental factors, and slight changes in temperature and humidity may interfere with the monitoring results; some methods have difficulty effectively extracting specific signals related to glucose, resulting in low measurement accuracy; and some technologies fail to fully consider the physiological differences between individuals, leading to significant fluctuations in monitoring accuracy across different individuals. Against this backdrop, a NUG algorithm based on quantum dot-enhanced sensors and metabolic resonance exciters to collect physical signals, combined with intelligent algorithms such as spatiotemporal graph convolutional neural networks for processing, along with corresponding medical testing equipment, is proposed. This has significant innovative value and practical application significance, bringing breakthrough progress to the field of glucose monitoring, effectively solving the above-mentioned problems in existing technologies, providing more accurate and convenient monitoring methods for the management of diseases such as diabetes, and improving patients' treatment experience and quality of life.

[0004] Therefore, existing non-invasive blood glucose monitoring technologies have problems such as sensitivity to environmental factors, difficulty in extracting specific signals, and poor individual adaptability, resulting in insufficient monitoring accuracy and reliability. Summary of the Invention

[0005] In order to overcome the problems of existing non-invasive blood glucose monitoring technologies, such as sensitivity to environmental factors, difficulty in extracting specific signals, and poor individual adaptability, which lead to insufficient monitoring accuracy and reliability, this invention discloses a portable non-invasive continuous blood glucose monitoring medical testing device and monitoring algorithm that can effectively solve the above-mentioned problems.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0007] A non-invasive continuous blood glucose monitoring algorithm includes the following steps:

[0008] Physical signals related to blood glucose are acquired by a quantum dot enhancement sensor and a metabolic resonance exciter in the physical signal acquisition layer. The quantum dot enhancement sensor uses phenylboronic acid-functionalized core-shell quantum dots, and the metabolic resonance exciter emits terahertz waves to excite the metabolic resonance effect.

[0009] The collected physical signals are transmitted to the quantum signal processing layer for processing;

[0010] The spatiotemporal graph convolutional neural network in the core of the intelligent algorithm is used to process the processed signal. The spatiotemporal graph convolutional neural network includes a spatiotemporal graph convolution module, a metabolic kinetic modeling module, and a personalized adaptation layer. The spatiotemporal features are extracted by the spatiotemporal graph convolution module, the parameters are self-optimized by combining the metabolic kinetic modeling module with deep learning, and the parameters are adapted based on personalized physiological parameters by the personalized adaptation layer, outputting blood glucose value and confidence level.

[0011] Preferably, the physical signal acquisition layer acquires signals in the following specific manner:

[0012] Metabolic resonance effect is excited at a characteristic frequency using a metabolic resonance exciter.

[0013] Quantum signals are detected by utilizing the fluorescence quenching effect generated by the specific binding of phenylboronic acid and glucose in a quantum dot-enhanced sensor, thus achieving dual-mode signal collaborative acquisition.

[0014] Preferably, the spatiotemporal graph nodes of the spatiotemporal graph convolutional neural network include: nodes with terahertz resonance response at characteristic frequencies of 0.1-10 THz, nodes with quantum dot fluorescence intensity of 550-650 nm, quantum dot fluorescence lifetime nodes, epidermal temperature nodes, local blood flow velocity nodes, tissue impedance nodes of 50-500 kHz, and environmental temperature and humidity nodes.

[0015] Preferably, the metabolic kinetic modeling adopts the following formula:

[0016] Parameter self-optimization is achieved by combining deep learning, where G is the blood glucose level and t is the time. , The correlation coefficient is... These are insulin-related parameters.

[0017] Preferably, the personalized physiological parameters embedded in the personalized adaptation layer include BMI index, skin melanin index, basal metabolic rate, and insulin sensitivity index.

[0018] Preferably, it also includes a dynamic calibration mechanism, specifically:

[0019] The algorithm's core outputs blood glucose levels and confidence scores. To the calibration decision module;

[0020] The calibration decision module determines the confidence level. Is it greater than 0.15? If the confidence level is greater than 0.15, the confidence assessment module will trigger a calibration command.

[0021] After receiving the calibration command, the microfluidic system uses a microneedle array to collect tissue fluid, collecting 0.1-0.5 μL samples.

[0022] Electrochemical detection was performed on the collected tissue fluid samples to obtain reference blood glucose values. Feedback is sent to the algorithm core;

[0023] The core of the algorithm is based on reference blood glucose values. Update personalized parameters using the parameter optimizer; the update formula is:

[0024]

[0025] in, For parameter update amount, For learning rate, To predict blood sugar levels, For personalized parameters.

[0026] Preferably, the dynamic calibration mechanism includes three levels of confidence thresholds: < 0.10 indicates high confidence; 0.10 ≤ ≤0.15 is considered a medium confidence level; > 0.15 indicates a low confidence level.

[0027] Preferably, a portable, non-invasive, continuous blood glucose monitoring medical testing device includes:

[0028] The physical signal acquisition layer includes the quantum dot enhanced sensor and metabolic resonance exciter mentioned above, and is used to acquire physical signals related to blood glucose.

[0029] The quantum signal processing layer is used to process signals acquired by the physical signal acquisition layer;

[0030] The core of the intelligent algorithm includes the spatiotemporal graph convolutional neural network mentioned above, which is used to analyze and process the processed signal and output blood glucose value and confidence level.

[0031] The user interaction layer is used to display information such as blood glucose levels and interact with the user.

[0032] Preferably, it also includes dynamic calibration-related components, including a calibration decision module, a confidence assessment module, a microfluidic system, a tissue fluid acquisition module, an electrochemical detection module, and a parameter optimizer, for implementing the dynamic calibration mechanism described above.

[0033] Preferably, the device is designed to be portable, making it easy for users to carry and use on a daily basis.

[0034] Compared with existing technologies, the beneficial effects of this invention are as follows: Existing non-invasive blood glucose monitoring technologies struggle to extract sufficiently specific blood glucose-related physical signals. The physical signal acquisition layer of this invention employs a quantum dot-enhanced sensor and a metabolic resonance exciter working in tandem. The quantum dot-enhanced sensor utilizes core-shell quantum dots functionalized with phenylboronic acid. Phenylboronic acid can specifically bind to glucose to produce a fluorescence quenching effect. This characteristic makes the sensor highly sensitive to changes in blood glucose levels, enabling it to accurately capture weak signal changes related to blood glucose concentration. Simultaneously, the metabolic resonance exciter emits terahertz waves to excite the metabolic resonance effect, enhancing blood glucose-related metabolic signals from another dimension. The synergistic effect of various signal acquisition methods not only improves signal strength but also enhances signal characteristics, thereby solving the problem of extracting specific signals in existing technologies and providing a foundation for subsequent accurate blood glucose monitoring. Current non-invasive blood glucose monitoring technologies are sensitive to environmental factors; changes in environmental temperature and humidity, skin temperature, etc., can affect the accuracy of monitoring results. In contrast, the spatiotemporal graph nodes of spatiotemporal graph convolutional neural networks include multiple nodes such as terahertz resonance response, quantum dot fluorescence intensity, fluorescence lifetime, skin temperature, local blood flow velocity, tissue impedance, and environmental temperature and humidity. By comprehensively considering the information from these nodes, a comprehensive spatiotemporal feature model can be constructed, enabling real-time capture and analysis of multiple... The network analyzes the physical signal characteristics of different dimensions, such as environmental temperature and humidity, and comprehensively analyzes nodes related to these changes, as well as nodes related to skin temperature and tissue impedance, to accurately determine the degree of impact of these changes on the monitoring signal. It then performs corresponding corrections and filtering, thus distinguishing environmental interference from specific blood glucose monitoring signals and effectively reducing the impact of environmental factors on the monitoring results, making the results more stable and reliable. Existing non-invasive blood glucose monitoring technologies have poor individual adaptability, failing to meet the physiological characteristics and metabolic differences of different individuals, leading to biased monitoring results. This solution embeds the BMI index into the personalized adaptation layer. Personalized physiological parameters such as skin melanin index, basal metabolic rate, and insulin sensitivity index are used to adjust and optimize the monitoring model based on these personalized parameters of different users when processing and analyzing signals. For example, for individuals with a high skin melanin index, the algorithm will adjust the weights and processing methods of epidermal temperature and tissue impedance signals to more accurately reflect their actual blood glucose level. In this way, the monitoring algorithm can adapt to the physiological characteristics and metabolic differences of different individuals, solve the problem of poor individual adaptability of existing technologies, and improve the accuracy and reliability of monitoring results.Existing non-invasive blood glucose monitoring technologies suffer from insufficient accuracy and reliability due to the aforementioned problems. This solution's dynamic calibration mechanism enables real-time dynamic calibration and optimization of monitoring results. The algorithm core outputs blood glucose values ​​and confidence levels to the calibration decision module. By setting three confidence thresholds, the reliability of monitoring results can be assessed in real time. When the confidence level falls below the threshold, the dynamic calibration mechanism is activated. The microfluidic system uses a microneedle array to collect tissue fluid and performs electrochemical detection to obtain a reference blood glucose value, which is then fed back to the algorithm core. The algorithm core updates personalized parameters based on the reference blood glucose value through a parameter optimizer, achieving real-time dynamic calibration and optimization of the monitoring model. This dynamic calibration process can promptly detect and correct monitoring errors caused by factors such as sensor drift and changes in individual physiological states, ensuring that the monitoring algorithm outputs high-precision blood glucose values ​​stably over the long term, thus solving the problems of insufficient accuracy and reliability in existing technologies. Attached Figure Description

[0035] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other embodiments can be derived from the provided drawings without creative effort.

[0036] Figure 1 This is a flowchart of the non-invasive blood glucose monitoring algorithm of the present invention;

[0037] Figure 2 This is a structural diagram of the portable device of the present invention. Detailed Implementation

[0038] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent.

[0039] To better illustrate this embodiment, some parts in the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions;

[0040] It will be understood by those skilled in the art that certain well-known structures and their descriptions may be omitted in the accompanying drawings.

[0041] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0042] Example

[0043] A non-invasive continuous blood glucose monitoring algorithm includes the following steps:

[0044] Physical signals related to blood glucose are acquired by a quantum dot enhancement sensor and a metabolic resonance exciter in the physical signal acquisition layer. The quantum dot enhancement sensor uses phenylboronic acid-functionalized core-shell quantum dots, and the metabolic resonance exciter emits terahertz waves to excite the metabolic resonance effect.

[0045] The collected physical signals are transmitted to the quantum signal processing layer for processing;

[0046] The spatiotemporal graph convolutional neural network in the core of the intelligent algorithm is used to process the processed signal. The spatiotemporal graph convolutional neural network includes a spatiotemporal graph convolution module, a metabolic kinetic modeling module, and a personalized adaptation layer. The spatiotemporal features are extracted by the spatiotemporal graph convolution module, the parameters are self-optimized by combining the metabolic kinetic modeling module with deep learning, and the parameters are adapted based on personalized physiological parameters by the personalized adaptation layer, outputting blood glucose value and confidence level.

[0047] The specific method by which the physical signal acquisition layer acquires signals is as follows:

[0048] Metabolic resonance effect is excited at a characteristic frequency using a metabolic resonance exciter.

[0049] Quantum signals are detected by utilizing the fluorescence quenching effect generated by the specific binding of phenylboronic acid and glucose in a quantum dot-enhanced sensor, thus achieving dual-mode signal collaborative acquisition.

[0050] The spatiotemporal graph nodes of the spatiotemporal graph convolutional neural network include: nodes with terahertz resonance response at characteristic frequencies of 0.1-10 THz, nodes with quantum dot fluorescence intensity of 550-650 nm, nodes with quantum dot fluorescence lifetime, nodes with epidermal temperature, nodes with local blood flow velocity, nodes with tissue impedance of 50-500 kHz, and nodes with ambient temperature and humidity.

[0051] The metabolic kinetics modeling uses the following formula:

[0052] Parameter self-optimization is achieved by combining deep learning, where G is the blood glucose level and t is the time. , The correlation coefficient is... These are insulin-related parameters.

[0053] The personalized physiological parameters embedded in the personalized adaptation layer include BMI index, skin melanin index, basal metabolic rate, and insulin sensitivity index.

[0054] It also includes a dynamic calibration mechanism, specifically:

[0055] The algorithm's core outputs blood glucose levels and confidence scores. To the calibration decision module;

[0056] The calibration decision module determines the confidence level. Is it greater than 0.15? If the confidence level is greater than 0.15, the confidence assessment module will trigger a calibration command.

[0057] After receiving the calibration command, the microfluidic system uses a microneedle array to collect tissue fluid, collecting 0.1-0.5 μL samples.

[0058] Electrochemical detection was performed on the collected tissue fluid samples to obtain reference blood glucose values. Feedback is sent to the algorithm core;

[0059] The core of the algorithm is based on reference blood glucose values. Update personalized parameters using the parameter optimizer; the update formula is:

[0060]

[0061] in, For parameter update amount, For learning rate, To predict blood sugar levels, For personalized parameters.

[0062] The dynamic calibration mechanism sets three confidence thresholds: < 0.10 indicates high confidence; 0.10 ≤ ≤0.15 is considered a medium confidence level; A value of >0.15 indicates low confidence (triggering calibration).

[0063] Please see Figure 1 When the device is started, the control system of the quantum dot enhancement sensor first performs state detection and calibration on the core-shell structure quantum dots functionalized with phenylboronic acid. By applying excitation light of known intensity to the sensor, its fluorescence output response is measured to ensure that the fluorescence performance of the quantum dots is stable and in the best working state. At the same time, the biocompatible gel layer on the surface of the sensor is heated to the normal human body temperature range (36-37℃) to ensure that glucose molecules can diffuse freely on the skin surface and fully contact the phenylboronic acid on the surface of the quantum dots.

[0064] The frequency tuning module of the metabolic resonance exciter selects a set of initial terahertz wave transmission frequencies from the stored frequency library based on the user's skin type (pre-detected by a skin optical sensor), such as 1.5THz, 3THz, 6THz, etc., and sets the transmission power to a low-power mode (initial power of about 1mW) to reduce the thermal effect on the user's skin, while ensuring that the metabolic resonance effect can be effectively excited. The microstrip antenna array performs self-testing and calibration, adjusting the antenna's directivity and radiation efficiency to ensure that the terahertz wave can uniformly cover the monitoring area on the skin surface.

[0065] After initialization, the quantum dot enhancement sensor and the metabolic resonance exciter start working simultaneously. The metabolic resonance exciter emits terahertz waves at a set frequency and power to excite molecules related to glucose metabolism in the superficial skin tissue to produce a resonance response. The quantum dot enhancement sensor detects the fluorescence quenching effect produced by the specific binding of phenylboronic acid and glucose in real time, and collects the fluorescence intensity (wavelength range 550-650nm) and fluorescence lifetime change signals of quantum dots. At the same time, the sensor also collects auxiliary physical signals such as epidermal temperature, local blood flow velocity, tissue impedance (frequency range 50-500kHz), and ambient temperature and humidity. These signals are sent to the quantum signal processing layer for processing in sequence through the multiplex analog switch inside the sensor.

[0066] After receiving various physical signals from the physical signal acquisition layer, the quantum signal processing layer first amplifies the weak fluorescence signal using a low-noise amplifier (LNA). Then, it uses a bandpass filter (for fluorescence intensity signals, the passband of the bandpass filter is the frequency range corresponding to 550-650nm; for terahertz resonance response signals, the passband is 0.1-10THz) to remove high-frequency noise and low-frequency drift interference from the signal, thereby improving the signal-to-noise ratio (SNR) to about 30dB.

[0067] After amplification and filtering, the analog signals are sequentially fed into a high-precision analog-to-digital converter (ADC) for digital conversion. The sampling frequency of the ADC is set according to different signal types. For fluorescence intensity signals and terahertz resonance response signals, the sampling frequency is 1MHz; for epidermal temperature, local blood flow velocity, tissue impedance, and ambient temperature and humidity signals, the sampling frequency is 100Hz. The converted digital signals are temporarily stored in the data buffer for further processing.

[0068] The digital signals in the buffer are organized and preprocessed according to spatiotemporal graph nodes of different dimensions. For terahertz resonance response signals, 10 equally spaced characteristic frequency points are selected in the frequency range of 0.1-10THz, such as 0.1THz, 0.3THz, 0.5THz, 0.7THz, 0.9THz, 2THz, 4THz, 6THz, 8THz, and 10THz as nodes. The amplitude and phase information of the signal at each frequency point are extracted to form a frequency-time series feature matrix.

[0069] For quantum dot fluorescence intensity signals, five equally spaced characteristic wavelength points are selected in the wavelength range of 550-650nm, such as 550nm, 575nm, 600nm, 625nm, and 650nm, as nodes. The intensity change and fluorescence lifetime decay curve features of the signal at each wavelength point are extracted to form a wavelength-time series feature matrix.

[0070] For the signals of skin temperature, local blood flow velocity, tissue impedance, and ambient temperature and humidity, the rate of change, mean, variance, and periodic features related to blood glucose changes in their time series are extracted to form corresponding feature vectors.

[0071] The spatiotemporal graph convolutional neural network (STGCN) is used to perform convolution operations on the feature matrices and feature vectors of the above dimensions to extract spatiotemporal features. The size of the convolution kernel is designed according to the spatiotemporal correlation of the signal features. For example, a 3×3 convolution kernel is used for the frequency-time series feature matrix, and a 2×3 convolution kernel is used for the wavelength-time series feature matrix. Through multiple convolution, pooling and activation operations, deep spatiotemporal correlation features in the signal are gradually extracted, and finally a spatiotemporal feature vector that integrates multi-dimensional information is obtained.

[0072] Based on formula Establish a blood glucose metabolism kinetic model, where G represents blood glucose level and t represents time. , Correlation coefficient and insulin-related parameters The initial values ​​are preset based on the average level of the population and stored in the parameter storage area of ​​the algorithm core.

[0073] The spatiotemporal feature vectors extracted by the spatiotemporal graph convolution module are used as input. Combined with deep learning algorithms, such as the Long Short-Term Memory (LSTM) network, the parameters of the metabolic dynamics model are self-optimized. By minimizing the mean squared error (MSE) loss function between the predicted blood glucose value and the actual blood glucose value (labeled using invasive blood glucose data during the training phase), the values ​​of correlation coefficients and insulin-related parameters are continuously adjusted using the gradient descent method. In actual monitoring, as user data accumulates, the model parameters will be continuously updated and optimized to better adapt to the individual metabolic characteristics of users.

[0074] The personalized adaptation layer first reads the user's personalized physiological parameters such as BMI, skin melanin index, basal metabolic rate, and insulin sensitivity index from the user storage area. These parameters are obtained by the user input (basic information such as height and weight are used to calculate BMI) and detection by the skin optical sensor (skin melanin index) when the device is used for the first time. The basal metabolic rate and insulin sensitivity index are initially estimated by empirical formulas based on the user's age, gender, body composition, and other information.

[0075] Based on personalized physiological parameters, the outputs of the spatiotemporal graph convolutional neural network and metabolic kinetic model are adjusted accordingly. For example, for users with high skin melanin index, the weight of tissue impedance signal in blood glucose prediction is appropriately increased, because high skin melanin content may affect the transmission of photoelectric signals and metabolic processes. For users with low insulin sensitivity index, the dynamic range of insulin-related parameters in the metabolic kinetic model is adjusted to more accurately reflect changes in the metabolic rate of blood glucose in the body. The adjusted results are output as the final blood glucose prediction value and confidence level to the user interaction layer and related modules of the dynamic calibration mechanism.

[0076] The blood glucose value and confidence level output by the intelligent algorithm are transmitted to the calibration decision module in real time. The calibration decision module first evaluates the confidence level and judges the reliability of the current monitoring result based on the preset three-level confidence threshold (< 0.10 is high confidence; 0.10≤≤0.15 is medium confidence; > 0.15 is low confidence). When the confidence level is greater than 0.15, the calibration decision module sends a trigger signal to the confidence evaluation module to start the calibration process.

[0077] After receiving the trigger signal, the confidence assessment module further analyzes the trend and cause of the confidence level change. For example, if the confidence level suddenly drops and fluctuates greatly, it may be due to external environmental factors, such as sudden temperature changes or poor sensor contact. If the confidence level gradually decreases but remains relatively stable, it may be due to changes in the user's physiological state, such as changes in insulin sensitivity or model parameter drift. Based on different confidence level change characteristics, the confidence assessment module will adjust the priority and frequency of subsequent calibrations to ensure that calibration is initiated when it is most needed.

[0078] After receiving the calibration command, the microfluidic system first controls the microneedle array drive module to quickly penetrate the skin surface and enter the dermis to collect tissue fluid at an appropriate insertion depth (pre-set according to the user's skin thickness, generally about 0.5-1mm). The microneedle array is arranged in an array and contains 10-20 tiny needles, each with a diameter of about 50-100μm, to ensure efficient collection of a sufficient volume of tissue fluid sample (0.1-0.5μL).

[0079] The micropump inside the microfluidic chip is activated, guiding the collected tissue fluid sample from the microneedle array to the reaction chamber of the electrochemical detection chip through pressure difference. In the reaction chamber, glucose in the tissue fluid undergoes an oxidation reaction under the catalysis of glucose oxidase, generating a current signal proportional to the glucose concentration. The electrochemical detection chip amplifies and measures the current signal to obtain a reference blood glucose value, and converts the reference blood glucose value into a digital signal through analog signal and feeds it back to the algorithm core.

[0080] The parameter optimizer adjusts the parameters according to the difference between the reference blood glucose value and the predicted blood glucose value, using the formula... The parameter update amount is calculated, where the learning rate (η) is dynamically adjusted based on the calibration history. The initial learning rate is 0.01, which is gradually reduced to approximately 0.001 after multiple consecutive calibrations to ensure the stability and convergence of the parameter updates. The parameter update amount (Δθ) is based on the predicted blood glucose value. Compared with reference blood glucose level The error and the current value of the personalized parameter (θ) are calculated, and the updated personalized parameter is fed back to the personalized adaptation layer and metabolic kinetic modeling module of the core intelligent algorithm for model calibration and optimization. The updated model parameters will be used for subsequent blood glucose prediction, thereby improving the accuracy and reliability of monitoring results.

[0081] A portable, non-invasive, continuous blood glucose monitoring medical testing device, comprising:

[0082] The physical signal acquisition layer includes the quantum dot enhanced sensor and metabolic resonance exciter mentioned above, and is used to acquire physical signals related to blood glucose.

[0083] The quantum signal processing layer is used to process signals acquired by the physical signal acquisition layer;

[0084] The core of the intelligent algorithm includes the spatiotemporal graph convolutional neural network mentioned above, which is used to analyze and process the processed signal and output blood glucose value and confidence level.

[0085] The user interaction layer is used to display information such as blood glucose levels and interact with the user.

[0086] It also includes dynamic calibration-related components, including a calibration decision module, a confidence assessment module, a microfluidic system, a tissue fluid acquisition module, an electrochemical detection module, and a parameter optimizer, to implement the dynamic calibration mechanism described above.

[0087] The device is designed to be portable, making it easy for users to carry and use in daily life.

[0088] Please see Figure 2 This non-invasive continuous blood glucose monitoring device adopts an integrated portable design with dimensions of 100mm×60mm×20mm. It can be designed as a watch or armband, or other shapes that are convenient for users to carry and use.

[0089] The overall structure of the device consists of two outer shells, upper and lower, which are injection molded from medical-grade polycarbonate material. They are sturdy, lightweight, waterproof, and resistant to electromagnetic interference. The upper and lower shells are tightly connected by ultrasonic welding technology to form a sealed device cavity. The internal structure integrates various functional modules such as a physical signal acquisition layer, a quantum signal processing layer, an intelligent algorithm core, a user interaction layer, and dynamic calibration-related components. The modules are electrically connected and transmit signals through flexible printed circuit boards (FPCs) and multi-pin connectors.

[0090] A 15mm diameter circular groove is designed at the bottom center of the lower outer shell of the device as a mounting base for the quantum dot enhancement sensor. The sensor chip is soldered onto the FPC using a flip-chip process and connected to the quantum signal processing layer via FPC leads. The sensor surface is covered with a biocompatible gel layer with a thickness of approximately 0.5mm. The gel layer has tiny pores evenly distributed in it, which ensures good adhesion between the skin and the sensor and allows glucose molecules to diffuse freely to the quantum dot surface and bind with phenylboronic acid. Miniature optical lenses and filter groups are set around the sensor. The optical lenses are used to focus the excitation light and collect the fluorescence signal. The filter group uses narrow-band filters with a center wavelength of 550-650nm and a bandwidth of ±10nm to ensure that fluorescence signals within a specific wavelength range can pass through, thereby improving the specificity and sensitivity of signal acquisition.

[0091] The terahertz wave emitting chip of the metabolic resonance exciter is installed in an independent mounting bracket next to the quantum dot enhancement sensor. The mounting bracket is electromagnetically shielded with a metal shield to prevent the terahertz wave from interfering with other electronic components. The emitting chip is connected to the quantum signal processing layer via an FPC. Its microstrip antenna array extends to the side of the device housing. The antenna surface is covered with a transparent polyimide protective film to ensure that the terahertz wave can be effectively radiated to the skin surface. The microstrip antenna array consists of 8 equally spaced antenna elements, each measuring 3mm × 3mm. The operating frequency covers the range of 0.1-10THz. The directional emission and focusing of the terahertz wave are achieved through phase control and power combining technology, thereby improving excitation efficiency and signal strength.

[0092] The quantum signal processing layer uses a multilayer printed circuit board with dimensions of 50mm×30mm×5mm, which is installed in the middle of the lower shell of the device. The PCB integrates a low-power, high-performance quantum signal processing chip (manufactured using 28nm CMOS process). The chip contains 8 analog-to-digital converters (ADCs), each with a resolution of 16 bits and a sampling frequency of up to 2MHz, which can meet the requirements for synchronous acquisition and processing of various physical signals. The quantum signal processing chip is surrounded by peripheral circuit components such as low-noise amplifiers (LNAs), bandpass filters, and data buffers (SRAMs). Through optimized wiring design and power management scheme, high precision and low power consumption of signal processing are ensured. The PCB is connected to the quantum dot enhancement sensor and metabolic resonance exciter of the physical signal acquisition layer through the FPC, and is also connected to the AIP processor of the intelligent algorithm core through the FPC on the other side, realizing efficient signal transmission and collaborative processing.

[0093] The dedicated Artificial Intelligence Processor (AIP), the core of the intelligent algorithm, is installed in the middle of the upper casing of the device. It uses a 40mm×25mm×3mm BGA package chip, soldered onto another multi-layer PCB. The AIP processor has powerful parallel computing capabilities and deep learning acceleration functions. It has 128 built-in neural network computing cores and supports mixed operations of INT8, FP16 and FP32 multi-precision data formats. It can efficiently run the Space-Time Graph Convolutional Neural Network (STGCN) algorithm and metabolic dynamics modeling algorithm. The AIP processor is connected to the PCB of the quantum signal processing layer through a high-speed data bus to realize rapid data interaction and processing. At the same time, the AIP processor is also connected to the device's storage module through a PCIe interface to store information such as algorithm model parameters, user data and historical monitoring records. A miniature heat sink made of graphene composite material is integrated next to the AIP processor to dissipate the heat generated by the processor during operation in a timely manner through heat conduction and natural heat dissipation, ensuring stable operation of the device in high-temperature environments.

[0094] A 2.0-inch square area is designed on the front of the upper casing of the device, housing a high-definition color LCD display. Capacitive touch technology enables touch button functionality. The display is connected to the AIP processor via an FPC, capable of displaying real-time blood glucose levels, historical blood glucose change curves, confidence levels, device operating status, and calibration prompts. The touch button module provides a simple and intuitive operating interface through a customized user interface design. Users can perform operations such as starting and stopping the device, viewing historical data, and setting alarm thresholds using virtual buttons on the touchscreen. Simultaneously, four physical buttons (power button, back button, increase button, and decrease button) are located around the display for basic operations when the touchscreen malfunctions or is inconvenient for the user, improving the device's usability and reliability.

[0095] A small Bluetooth antenna mounting slot is designed on the side of the upper shell of the device, integrating a low-power Bluetooth 5.0 chip. It is connected to the AIP processor via FPC. The Bluetooth module is used to wirelessly connect with users' smartphones or tablets and other mobile devices, with a transmission distance of up to 10 meters. Users can remotely view blood glucose data, receive alarm information, perform data analysis and management, etc. on their mobile devices through the accompanying mobile application. At the same time, the Bluetooth module also supports data synchronization with cloud servers. Users can upload long-term blood glucose monitoring data to the cloud, which facilitates doctors to make remote diagnoses and adjust treatment plans.

[0096] The microfluidic system is integrated into a separate microfluidic chip module, installed on one side of the lower outer shell of the device. The microfluidic chip is fabricated using multilayer micro-nano processing technology and includes components such as a microneedle array, micropump, microvalve, electrochemical detection chip, and microfluidic channels. The microneedle array consists of 16 tiny needles made of stainless steel with a passivated surface to ensure the sharpness and biocompatibility of the needles. The micropump and microvalve use piezoelectric actuation technology to precisely control the amount of tissue fluid collected and the fluid transport path. The electrochemical detection chip is integrated into the reaction chamber of the microfluidic chip and uses a carbon electrode prepared by screen printing technology with a surface modified with glucose oxidase, enabling rapid and accurate detection of glucose concentration in the tissue fluid. The microfluidic chip module is connected to the AIP processor, the core of the intelligent algorithm, via an FPC to realize the transmission of control signals and detection data.

[0097] The calibration decision module, confidence assessment module, and parameter optimizer, as auxiliary functional modules of the core intelligent algorithm, are all integrated into the software algorithm of the AIP processor. The logic control and data processing functions of these modules are implemented on the AIP processor. The calibration decision module receives blood glucose values ​​and confidence data output by the algorithm core in real time, judges according to the preset confidence threshold, and triggers calibration instructions. The confidence assessment module analyzes the trend of confidence changes and adjusts the priority and frequency of calibration. The parameter optimizer calculates the parameter update based on the difference between the reference blood glucose value and the predicted blood glucose value, and feeds it back to the algorithm core for model calibration. These modules work closely with the hardware resources of the AIP processor to ensure the efficient operation and real-time response of the dynamic calibration mechanism.

[0098] Workflow: When the user presses the power button to start the device, the device first performs a system self-test, including the status detection and initialization of the quantum dot enhancement sensor, metabolic resonance exciter, quantum signal processing layer, intelligent algorithm core, user interaction layer, and dynamic calibration-related components. The self-test process lasts for about 10 seconds, during which the display shows the startup screen and the self-test progress bar. If any abnormality is found during the self-test, the device will issue an audible and visual alarm to prompt the user and automatically enter the fault diagnosis mode.

[0099] After the self-test is completed, the quantum dot enhancement sensor and metabolic resonance exciter in the physical signal acquisition layer start working. The metabolic resonance exciter emits terahertz waves at a preset frequency and power to excite the metabolic resonance effect of the superficial skin tissue. The quantum dot enhancement sensor detects the fluorescence quenching signal generated by the combination of phenylboronic acid and glucose in real time, and at the same time collects auxiliary physical signals such as epidermal temperature, local blood flow velocity, tissue impedance and ambient temperature and humidity. The collected physical signals are transmitted to the quantum signal processing layer through FPC for signal amplification, filtering and analog-to-digital conversion, and effective signal features are extracted and stored in the data buffer.

[0100] The quantum signal processing layer sends the processed digital signal to the AIP processor, the core of the intelligent algorithm. The AIP processor runs the Space-Time Graph Convolutional Neural Network (STGCN) algorithm to extract the spatiotemporal features of the signal. The metabolic dynamics modeling module combines deep learning to self-optimize the model parameters. The personalized adaptation layer adjusts the results according to the user's personalized physiological parameters and finally outputs the real-time blood glucose value and confidence score. The blood glucose value and confidence score are displayed to the user intuitively through the display screen of the user interaction layer. At the same time, the data is synchronized to the user's mobile device APP and cloud server via Bluetooth module.

[0101] The blood glucose value and confidence level output by the algorithm core are transmitted to the calibration decision module in real time. When the confidence level is greater than 0.15, the calibration decision module triggers a calibration command. After receiving the command, the microfluidic system uses a microneedle array to collect tissue fluid samples and perform electrochemical detection to obtain a reference blood glucose value, which is then fed back to the algorithm core. The parameter optimizer updates personalized parameters based on the reference blood glucose value to achieve dynamic calibration of the monitoring model, ensuring the accuracy and reliability of subsequent monitoring data. After calibration is completed, the device automatically returns to normal monitoring mode and displays a calibration success message on the screen.

[0102] During operation, the device stores the user's blood glucose monitoring data, calibration records, and other relevant information in the eMMC flash memory chip in real time. Users can view historical data and generated data reports through touch buttons or mobile device APP. When the blood glucose value exceeds the normal range preset by the user, such as higher than 7.8 mmol / L or lower than 3.9 mmol / L, the device will issue an audible and visual alarm to prompt the user and display alarm information and corresponding handling suggestions on the display screen, such as eating in time or seeking medical treatment.

[0103] Through the above algorithms and device implementations, this non-invasive continuous blood glucose monitoring system achieves efficient integration of the entire process from signal acquisition and processing to intelligent analysis and dynamic calibration. The portable design and high-performance component integration of the device make it easy to apply to the daily blood glucose monitoring of diabetic patients, providing a precise, reliable and convenient new method for diabetes management, significantly improving the quality of life of diabetic patients and improving blood glucose control.

[0104] The same or similar labels correspond to the same or similar parts;

[0105] The terms used to describe positional relationships in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent.

[0106] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all implementation methods here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A non-invasive continuous glucose monitoring algorithm, characterized in that, Includes the following steps: Physical signals related to blood glucose are acquired by a quantum dot enhancement sensor and a metabolic resonance exciter in the physical signal acquisition layer. The quantum dot enhancement sensor uses phenylboronic acid-functionalized core-shell quantum dots, and the metabolic resonance exciter emits terahertz waves to excite the metabolic resonance effect. The collected physical signals are transmitted to the quantum signal processing layer for processing; The spatiotemporal graph convolutional neural network in the core of the intelligent algorithm is used to process the processed signal. The spatiotemporal graph convolutional neural network includes a spatiotemporal graph convolution module, a metabolic kinetic modeling module, and a personalized adaptation layer. The spatiotemporal features are extracted by the spatiotemporal graph convolution module, the parameters are self-optimized by combining the metabolic kinetic modeling module with deep learning, and the parameters are adapted based on personalized physiological parameters by the personalized adaptation layer, outputting blood glucose value and confidence level.

2. The non-invasive continuous glucose monitoring algorithm of claim 1, wherein, The specific method by which the physical signal acquisition layer acquires signals is as follows: Metabolic resonance effect is excited at a characteristic frequency using a metabolic resonance exciter. Quantum signals are detected by utilizing the fluorescence quenching effect generated by the specific binding of phenylboronic acid and glucose in a quantum dot-enhanced sensor, thus achieving dual-mode signal collaborative acquisition.

3. The non-invasive continuous glucose monitoring algorithm of claim 1, wherein, The spatiotemporal graph nodes of the spatiotemporal graph convolutional neural network include: nodes with terahertz resonance response at characteristic frequencies of 0.1-10 THz, nodes with quantum dot fluorescence intensity of 550-650 nm, nodes with quantum dot fluorescence lifetime, nodes with epidermal temperature, nodes with local blood flow velocity, nodes with tissue impedance of 50-500 kHz, and nodes with ambient temperature and humidity.

4. The non-invasive continuous blood glucose monitoring algorithm according to claim 1, characterized in that, The metabolic kinetics modeling uses the following formula: Parameter self-optimization is achieved by combining deep learning, where G is the blood glucose level and t is the time. , The correlation coefficient, These are insulin-related parameters.

5. The non-invasive continuous blood glucose monitoring algorithm according to claim 1, characterized in that, The personalized physiological parameters embedded in the personalized adaptation layer include BMI index, skin melanin index, basal metabolic rate, and insulin sensitivity index.

6. The non-invasive continuous blood glucose monitoring algorithm according to claim 1, characterized in that, It also includes a dynamic calibration mechanism, specifically: The algorithm's core outputs blood glucose levels and confidence scores. To the calibration decision module; The calibration decision module determines the confidence level. Is it greater than 0.15? If the confidence level is greater than 0.15, the confidence assessment module will trigger a calibration command. After receiving the calibration command, the microfluidic system uses a microneedle array to collect tissue fluid, collecting 0.1-0.5 μL samples. Electrochemical detection was performed on the collected tissue fluid samples to obtain reference blood glucose values. Feedback is sent to the algorithm core; The core of the algorithm is based on reference blood glucose values. Update personalized parameters using the parameter optimizer; the update formula is: ； in, For parameter update amount, For learning rate, To predict blood glucose levels, For personalized parameters.

7. The non-invasive continuous blood glucose monitoring algorithm according to claim 6, characterized in that, The dynamic calibration mechanism sets three confidence thresholds: < 0.10 indicates high confidence; 0.10 ≤ ≤0.15 is considered a medium confidence level; > 0.15 indicates a low confidence level.

8. A portable, non-invasive, continuous blood glucose monitoring medical testing device, characterized in that, include: A physical signal acquisition layer comprising the quantum dot enhanced sensor and metabolic resonance exciter as described in any one of claims 1-7, for acquiring physical signals related to blood glucose; The quantum signal processing layer is used to process signals acquired by the physical signal acquisition layer; The core of the intelligent algorithm includes the spatiotemporal graph convolutional neural network as described in any one of claims 1-7, used to analyze and process the processed signal and output blood glucose value and confidence level; The user interaction layer is used to display blood glucose information and interact with the user.

9. The portable, non-invasive, continuous blood glucose monitoring medical testing device according to claim 8, characterized in that, It also includes dynamic calibration-related components, including a calibration decision module, a confidence assessment module, a microfluidic system, a tissue fluid acquisition module, an electrochemical detection module, and a parameter optimizer, for implementing the dynamic calibration mechanism described in any one of claims 1-7.

10. The portable, non-invasive, continuous blood glucose monitoring medical testing device according to claim 8, characterized in that, The device is designed to be portable, making it easy for users to carry and use in daily life.

Images