Noninvasive blood glucose monitoring system and method based on multi-modal data fusion

Abstract

The invention relates to the technical field of biomedical detection, and provides a noninvasive blood glucose monitoring system based on multi-modal data fusion, which comprises a gas acquisition channel, a gas sensing module, a physiological signal acquisition module, a microprocessor and a display unit which work cooperatively in sequence, and a wireless communication module and a power supply module can be additionally arranged; the gas collection channel is used for guiding a user to exhale gas and ensuring that the gas is stably transmitted to a sensing area; the gas sensing module is arranged in the gas collecting channel and is used for detecting the concentration value of acetone in exhaled gas; the physiological signal acquisition module is used for acquiring at least one physiological parameter of a user and providing a correction basis for blood glucose calculation; the microprocessor receives the acetone concentration value and the physiological parameters and calculates the blood glucose value through a fusion model. The method supports dual-mode operation of a local basic algorithm and a cloud AI algorithm, is high in accuracy, is low in cost, is suitable for daily monitoring of diabetic patients and metabolism management of healthy people, and has a remarkable industrial application value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biomedical detection, in particular to a non-invasive blood glucose monitoring system and method based on multi-modal data fusion. BACKGROUND

[0002] Blood glucose monitoring is the core demand of daily management of diabetic patients and metabolic monitoring of healthy people. Traditional invasive or minimally invasive monitoring methods have problems such as trauma, infection risk and high cost. Non-invasive blood glucose monitoring technology has become the industry development trend due to its safety and convenience. With the improvement of people's health management awareness, higher requirements are put forward for the accuracy, stability and diversity of application scenarios of non-invasive blood glucose monitoring, especially the precise monitoring under different physiological conditions.

[0003] In the related technology of non-invasive blood glucose monitoring, existing solutions attempt to calculate blood glucose value by detecting acetone concentration in exhaled gas, and the core relies on the linear relationship between acetone concentration and blood glucose to achieve monitoring. Some solutions only provide a single gas sensing module and do not correct other physiological parameters. The device uses a single algorithm mode and lacks subsequent upgrade optimization capability.

[0004] The existing non-invasive blood glucose monitoring technology generally has the problem of insufficient accuracy. The acetone concentration in exhaled gas is easily affected by factors such as exercise state, dietary structure, individual metabolic differences, etc. Simply relying on the linear relationship between acetone concentration and blood glucose leads to large monitoring errors under different physiological conditions, especially after exercise. At the same time, existing technologies mostly use a single algorithm, which cannot achieve precise optimization according to user individual characteristics and data accumulation. The clinical recognition and market acceptance are limited, and it is difficult to meet the actual needs of daily monitoring and health management. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application aims to provide a non-invasive blood glucose monitoring system and method based on multi-modal data fusion, which effectively improves the accuracy and stability of non-invasive blood glucose monitoring by introducing physiological parameters as correction factors, while taking into account the immediate landing and future upgrade capability of the product.

[0006] To achieve the above purpose, the present application realizes the following technical scheme: a non-invasive blood glucose monitoring system based on multi-modal data fusion, comprising a gas collection channel, a gas sensing module, a physiological signal acquisition module, a microprocessor, a display unit, and a wireless communication module and a power module can be additionally provided, which work cooperatively in sequence. The gas collection channel is used to guide the user's exhaled gas to ensure stable transmission of the gas to the sensing area. The gas sensing module is arranged in the gas collection channel to detect the acetone concentration value in the exhaled gas. The physiological signal acquisition module acquires at least one physiological parameter of the user, providing a correction basis for blood glucose calculation. The microprocessor receives the acetone concentration value and the physiological parameter, and calculates the blood glucose value through a fusion model. The display unit outputs the blood glucose value, and the wireless communication module realizes data uploading and model iteration.

[0007] Preferably, the gas sensing module is a metal oxide semiconductor sensor or a photoacoustic ring-down spectroscopy sensor, and the acetone detection range is 1-1000ppm.

[0008] Preferably, the physiological signal acquisition module includes a photoplethysmogram sensor integrated on the surface of the shell, or an external wearable device connected through wireless communication, which can acquire physiological parameters such as heart rate, body temperature, and blood oxygen; wherein the heart rate detection range is 30-200bpm, divided into three intervals of resting heart rate, normal activity heart rate, and post-exercise heart rate.

[0009] Preferably, the microprocessor uses a low-power ARM Cortex-M4 core microcontroller, with 128KB Flash memory and 32KB RAM built-in, supporting local basic algorithm and cloud AI algorithm dual-mode operation; the fusion model includes a primary calibration algorithm based on linear regression or polynomial fitting, and a senior calibration algorithm based on a neural network model issued by the cloud.

[0010] Preferably, the wireless communication module supports Bluetooth 5.0 and above or WiFi communication, and can upload data such as acetone concentration value, physiological parameter, blood glucose value, measurement time, and user identification to the cloud server, with data transmission using AES-256 encryption algorithm.

[0011] Preferably, the display unit is a 0.96-inch and above OLED display screen, which can clearly display the blood glucose value, measurement state, and abnormal prompt; the power module uses a rechargeable lithium battery, and the shell adopts a handheld design, with a size suitable for ergonomics, and is provided with a mouthpiece, an operation button, a charging interface, and a PPG sensor area, facilitating user holding and operation.

[0012] A non-invasive blood glucose monitoring method based on multi-modal data fusion, comprising the following steps: Step one, the user blows evenly into the gas collection channel for 3-5 seconds, and the gas sensing module detects the acetone concentration value in the exhaled gas and transmits it to the microprocessor; Step two, the physiological signal acquisition module synchronously acquires the physiological parameters of the user and transmits the data to the microprocessor; Step 3: The microprocessor calls the fusion model, inputs the acetone concentration value and physiological parameters into the model, and calculates the blood glucose value through the corresponding calibration algorithm. The blood glucose value calculation formula is: Blood glucose value = a × acetone concentration + b × heart rate correction coefficient + c, where a, b, and c are constant coefficients determined through clinical trials. Step 4: The display unit outputs the blood glucose prediction results. If a wireless communication module is provided, the relevant data can be encrypted and uploaded to the cloud server. Step 5: The cloud server trains an LSTM neural network model based on massive user data, and the lightweight model is pushed to the device via OTA to achieve algorithm upgrade.

[0013] Preferably, in step three, the microprocessor calls different calibration coefficients according to the user's heart rate range to correct the effect of acetone concentration on blood glucose calculation, effectively distinguishing between high acetone caused by exercise and high acetone caused by high blood glucose.

[0014] Preferably, the primary calibration algorithm in step three is implemented based on a pre-stored two-dimensional calibration table, with the X-axis representing acetone concentration and divided into 10 intervals, and the Y-axis representing heart rate intervals, quickly outputting blood glucose results.

[0015] Preferably, the cloud-based AI model training data in step five includes acetone concentration, heart rate, time series features, and basic user information. It can also incorporate auxiliary information such as meal type and exercise status selectively input by the user to minimize prediction error.

[0016] This invention provides a non-invasive blood glucose monitoring system and method based on multimodal data fusion. It has the following beneficial effects: 1. This invention introduces physiological parameters such as heart rate as correction factors and combines them with dynamic adjustment of calibration coefficients for heart rate zones to effectively distinguish between increased acetone concentration caused by exercise and hyperglycemia, significantly reducing monitoring errors under different physiological conditions and significantly reducing errors compared with venous blood glucose.

[0017] 2. This invention supports dual-mode operation of local basic algorithms and cloud AI algorithms. The basic algorithm ensures that the device is plug-and-play and meets immediate monitoring needs; the cloud AI algorithm continuously improves accuracy over time through massive data training and personalized optimization, realizing the upgrade from basic monitoring to precise customization and adapting to the needs of different user groups.

[0018] 3. This invention is completely non-invasive and painless, avoiding the trauma and infection risks of invasive monitoring; the device is small in size, lightweight, easy to operate, has a short measurement time, and a long battery life, making it suitable for frequent daily monitoring and carrying around, thus lowering the barrier to entry for users.

[0019] 4. This invention is not only suitable for daily blood glucose monitoring for diabetic patients, but also meets the metabolic health management needs of healthy people, and is adaptable to multiple scenarios such as home monitoring, exercise health management, and personalized medicine. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the non-invasive blood glucose monitoring system provided in an embodiment of the present invention; Figure 2 A system workflow diagram provided for embodiments of the present invention; Figure 3 A logical block diagram of the multimodal data fusion algorithm provided in this embodiment of the invention; Figure 4 This is a schematic diagram of the device's external structure provided in an embodiment of the present invention.

[0021] The components include: 1. Housing; 2. Mouthpiece; 3. Gas acquisition channel; 4. Gas sensing module; 5. Microprocessor; 6. Physiological signal acquisition module; 7. Display unit; 8. Wireless communication module; and 9. Power supply module. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] As one aspect of the present invention, please refer to the appendix. Figure 1 - Appendix Figure 4 This invention provides a non-invasive blood glucose monitoring system based on multimodal data fusion, comprising a gas acquisition channel 3, a gas sensing module 4, a physiological signal acquisition module 6, a microprocessor 5, and a display unit 7 that work in sequence and in concert. A wireless communication module 8 and a power supply module 9 may also be added. The gas acquisition channel 3 guides the user's exhaled gas, ensuring stable gas transmission to the sensing area. Gas sensing module 4 is located within gas acquisition channel 3 to detect the concentration of acetone in exhaled breath. Gas sensing module 4 is a metal oxide semiconductor sensor or an optical cavity ring-down spectroscopy sensor, with an acetone detection range of 1-1000 ppm; The physiological signal acquisition module 6 collects at least one physiological parameter from the user to provide a basis for correcting blood glucose calculation. The physiological signal acquisition module 6 includes a photoplethysmography (PPG) sensor integrated on the surface of the housing 1, or an external wearable device connected via wireless communication, which can collect physiological parameters such as heart rate, body temperature, and blood oxygen. Its heart rate detection range is 30-200 bpm, divided into three intervals: resting heart rate, normal activity heart rate, and post-exercise heart rate. Microprocessor 5 receives acetone concentration values ​​and physiological parameters, and calculates blood glucose values ​​through a fusion model. Microprocessor 5 uses a low-power ARM Cortex-M4 core microcontroller, with 128KB Flash memory and 32KB RAM, and supports dual-mode operation of local basic algorithms and cloud AI algorithms. The fusion model includes a primary calibration algorithm based on linear regression or polynomial fitting, and an advanced calibration algorithm based on a neural network model distributed from the cloud. Display unit 7 outputs blood glucose values, and wireless communication module 8 enables data uploading and model iteration. Wireless communication module 8 supports Bluetooth 5.0 and above or WiFi communication, and can upload data such as acetone concentration, physiological parameters, blood glucose values, measurement time, and user identification to the cloud server. Data transmission uses AES-256 encryption algorithm. Display unit 7 is a 0.96-inch or larger OLED display screen, which can clearly display blood glucose values, measurement status, and abnormal prompts. Power module 9 uses a rechargeable lithium battery. The housing 1 adopts a handheld design with an ergonomic size, and is equipped with a mouthpiece, operation buttons, charging interface, and PPG sensor area for easy user grip and operation.

[0024] As another aspect of the present invention, embodiments of the present invention provide a non-invasive blood glucose monitoring method based on multimodal data fusion, comprising the following steps: Step 1: The user blows air evenly into the gas collection channel 3 for 3-5 seconds. The gas sensing module 4 detects the acetone concentration in the exhaled air and transmits it to the microprocessor 5. Step 2: The physiological signal acquisition module 6 synchronously acquires the user's physiological parameters and transmits the data to the microprocessor 5; Step 3: Microprocessor 5 calls the fusion model, inputs the acetone concentration value and physiological parameters into the model, and calculates the blood glucose value through the corresponding calibration algorithm. The blood glucose value calculation formula is: Blood glucose value = a × acetone concentration + b × heart rate correction coefficient + c, where a, b, and c are constant coefficients determined through clinical trials. Microprocessor 5 calls different calibration coefficients according to the user's heart rate interval to correct the influence of acetone concentration on blood glucose calculation, effectively distinguishing between high acetone caused by exercise and high acetone caused by high blood glucose. The primary calibration algorithm is implemented based on a pre-stored two-dimensional calibration table, with the X-axis representing acetone concentration and divided into 10 intervals, and the Y-axis representing the heart rate interval, quickly outputting the blood glucose result. Step 4: Display unit 7 outputs the blood glucose prediction result. If a wireless communication module 8 is provided, the relevant data can be encrypted and uploaded to the cloud server. Step 5: The cloud server trains an LSTM neural network model based on massive user data. The lightweight model is pushed to the device via OTA to upgrade the algorithm. The cloud AI model training data includes acetone concentration, heart rate, time series features, and basic user information. It can also incorporate auxiliary information such as meal type and exercise status selectively input by the user to minimize prediction error.

[0025] The following description, in conjunction with specific embodiments, will be provided. Example 1: Basic Hardware Implementation I. Equipment Structural Design: The device features a handheld housing 1 with overall dimensions of 120mm (length) × 40mm (width) × 25mm (height), making it easy for users to hold and operate. One end of the device has a nozzle 2 (gas collection channel 3), and the surface of the housing 1 has a transparent window for displaying results. The side of the housing 1 has operation buttons and a charging port.

[0026] II. Core Component Configuration: Gas sensing module 4: A mature MOS (metal oxide semiconductor) gas sensor (such as the Figaro FIGAROTGS822 sensor) is selected, with the cost controlled within 20 yuan. The sensitivity is moderate and it can detect the acetone concentration in the range of 5-500ppm.

[0027] Physiological signal acquisition module 6: A PPG (photoplethysmography) sensor (such as the Sonix Technology SN32F700 series) is integrated on the side of the housing 1 to acquire the user's heart rate signal at a sampling frequency of 100Hz.

[0028] Microprocessor 5: A microcontroller with a low-power ARM Cortex-M4 core (such as the STMicroelectronics STM32L4 series), with 128KB Flash memory and 32KB RAM.

[0029] Display Unit 7: Uses a 0.96-inch OLED display with a resolution of 128×64 pixels.

[0030] III. Workflow Design: Step 1: The user holds the device and naturally presses the PPG sensor on the side with their finger. Step 2: The user blows air evenly into mouthpiece 2 for 3-5 seconds; Step 4: Microprocessor 5 calls the pre-stored calibration algorithm to calculate the blood glucose value; Step 5: The display screen shows the final blood glucose monitoring results.

[0031] IV. Basic Algorithm Implementation: A two-dimensional calibration table is pre-stored in the microprocessor 5 and stored in the Flash memory; The X-axis represents the acetone concentration (unit: ppm), divided into 10 intervals; The Y-axis represents heart rate zones, divided into three zones: resting heart rate (<70 bpm), normal activity heart rate (70-90 bpm), and post-exercise heart rate (>90 bpm). Based on the user's current heart rate zone, different calibration coefficients are used to correct the acetone value, thereby obtaining the blood glucose value; The calculation formula is: Blood glucose level = a × acetone concentration + b × heart rate correction factor + c Where a, b, and c are constant coefficients determined through clinical trials.

[0032] Example 2: Cloud AI Upgrade Path (High Value Demonstration) I. Data Collection and Upload: The device connects to a mobile app via Bluetooth 5.0. After each measurement, the raw data (acetone concentration, heart rate, measurement time, user ID) is uploaded to the cloud server; Users can selectively input auxiliary information such as meal type (fasting, post-meal) and exercise status.

[0033] II. Cloud-based AI model training: The cloud server adopts a distributed computing architecture and uses the TensorFlow framework to build deep learning models; The model structure uses an LSTM neural network, which can process time series data and capture the metabolic change patterns of users; Training data includes: acetone concentration, heart rate, time series features, and basic user information (age, gender, weight). Training objective: Minimize the error between the predicted blood glucose value and the actual blood glucose value manually entered by the user (from a traditional blood glucose meter).

[0034] III. Model Optimization and Deployment: The trained lightweight model (with a size of less than 50KB) is pushed to the user's device via OTA. The device supports model version management and can be rolled back to previous versions; The model update frequency can be dynamically adjusted based on the accumulation of user data. Initially, it will be updated once a month, and later it will be adjusted as needed.

[0035] IV. Personalized Service Implementation: The longer the device is used, the better the AI ​​model understands the user's metabolic characteristics, and the higher the prediction accuracy. It supports providing customized models for different user groups (such as type 1 diabetes, type 2 diabetes, and gestational diabetes).

[0036] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A non-invasive blood glucose monitoring system based on multimodal data fusion, characterized in that, It includes a gas acquisition channel (3), a gas sensing module (4), a physiological signal acquisition module (6), a microprocessor (5), and a display unit (7) that work in sequence and in coordination. A wireless communication module (8) and a power supply module (9) may also be added. The gas collection channel (3) is used to guide the user to exhale gas and ensure that the gas is stably transmitted to the sensing area; The gas sensing module (4) is located in the gas collection channel (3) to detect the acetone concentration in the exhaled gas; The physiological signal acquisition module (6) acquires at least one physiological parameter from the user to provide a basis for blood glucose calculation correction; the microprocessor (5) receives the acetone concentration value and physiological parameters and calculates the blood glucose value through the fusion model; the display unit (7) outputs the blood glucose value, and the wireless communication module (8) realizes data uploading and model iteration.

2. The non-invasive blood glucose monitoring system based on multimodal data fusion according to claim 1, characterized in that, The gas sensing module (4) is a metal oxide semiconductor sensor or an optical cavity ring-down spectral sensor, with an acetone detection range of 1-1000ppm.

3. The non-invasive blood glucose monitoring system based on multimodal data fusion according to claim 1, characterized in that, The physiological signal acquisition module (6) includes a photoplethysmography sensor integrated on the surface of the housing (1) or an external wearable device connected via wireless communication, which can collect physiological parameters such as heart rate, body temperature, and blood oxygen. The heart rate detection range is 30-200 bpm, divided into three zones: resting heart rate, normal activity heart rate, and post-exercise heart rate.

4. The non-invasive blood glucose monitoring system based on multimodal data fusion according to claim 1, characterized in that, The microprocessor (5) adopts a low-power ARM Cortex-M4 core microcontroller with 128KB Flash memory and 32KB RAM, and supports dual-mode operation of local basic algorithms and cloud AI algorithms; the fusion model includes a primary calibration algorithm based on linear regression or polynomial fitting, and an advanced calibration algorithm based on a neural network model distributed from the cloud.

5. The non-invasive blood glucose monitoring system based on multimodal data fusion according to claim 1, characterized in that, The wireless communication module (8) supports Bluetooth 5.0 and above or WiFi communication, and can upload data such as acetone concentration, physiological parameters, blood glucose value, measurement time, and user identification to the cloud server. The data transmission adopts the AES-256 encryption algorithm.

6. The non-invasive blood glucose monitoring system based on multimodal data fusion according to claim 1, characterized in that, The display unit (7) is an OLED display screen of 0.96 inches or larger, which can clearly display blood glucose value, measurement status and abnormal prompts; the power module (9) adopts a rechargeable lithium battery, and the housing (1) adopts a handheld design with an ergonomic size. It is equipped with a mouthpiece (2), operation buttons, charging interface and PPG sensor area, which is convenient for users to hold and operate.

7. A non-invasive blood glucose monitoring method based on multimodal data fusion, using a non-invasive blood glucose monitoring system based on multimodal data fusion as described in any one of claims 1-6, characterized in that, Includes the following steps: Step 1: The user blows air evenly into the gas collection channel (3) for 3-5 seconds. The gas sensing module (4) detects the acetone concentration in the exhaled air and transmits it to the microprocessor (5). Step 2: The physiological signal acquisition module (6) synchronously acquires the user's physiological parameters and transmits the data to the microprocessor (5). Step 3: The microprocessor (5) calls the fusion model, inputs the acetone concentration value and physiological parameters into the model, and calculates the blood glucose value through the corresponding calibration algorithm. The blood glucose value calculation formula is: blood glucose value = a × acetone concentration + b × heart rate correction coefficient + c, where a, b, and c are constant coefficients determined through clinical trials. Step 4: The display unit (7) outputs the blood glucose prediction results. If a wireless communication module (8) is provided, the relevant data can be encrypted and uploaded to the cloud server. Step 5: The cloud server trains an LSTM neural network model based on massive user data, and the lightweight model is pushed to the device via OTA to achieve algorithm upgrade.

8. The non-invasive blood glucose monitoring method based on multimodal data fusion according to claim 7, characterized in that, In step three, the microprocessor (5) calls different calibration coefficients according to the user's heart rate interval to correct the effect of acetone concentration on blood glucose calculation, effectively distinguishing between high acetone caused by exercise and high acetone caused by high blood glucose.

9. A non-invasive blood glucose monitoring method based on multimodal data fusion according to claim 7, characterized in that, The primary calibration algorithm in step three is based on a pre-stored two-dimensional calibration table. The X-axis represents the acetone concentration and is divided into 10 intervals, while the Y-axis represents the heart rate interval, quickly outputting blood glucose results.

10. A non-invasive blood glucose monitoring method based on multimodal data fusion according to claim 7, characterized in that, In step five, the cloud-based AI model training data includes acetone concentration, heart rate, time series features, and basic user information. It can also incorporate auxiliary information such as meal type and exercise status selectively input by the user to minimize prediction error.

Images