Wearable intelligent noninvasive blood glucose monitoring method and system based on microfluidic technology

Abstract

The invention discloses a wearable intelligent noninvasive blood glucose monitoring method and system based on a micro-fluidic technology, and the method comprises the steps: responding to a blood glucose monitoring instruction, detecting the state of wearable equipment, collecting sweat to a plurality of micro-channels through a micro-fluidic chip, carrying out an oxidase reaction, obtaining a reaction spectrum, and combining the physiological index characteristics of a user to obtain a blood glucose monitoring result; the blood glucose monitoring result is accurately determined. By means of the design, noninvasive, real-time and portable blood glucose monitoring is achieved, operation is easy and convenient, environment adaptability is high, and high application value is achieved.

Description

Technical Field

[0001] The present invention relates to the field of health monitoring, and in particular to a wearable intelligent non-invasive blood glucose monitoring method and system based on microfluidics technology. Background Art

[0002] Most existing blood glucose monitoring methods rely on blood sampling by needle puncture, which is not only painful for patients, but also has the risk of infection, and it is difficult to achieve long-term, real-time blood glucose monitoring. In recent years, with the development of microfluidics technology, non-invasive blood glucose monitoring technology has gradually attracted attention. However, existing non-invasive blood glucose monitoring technology still faces problems such as difficulty in sample collection, difficulty in preservation, and insufficient monitoring accuracy. Summary of the invention

[0003] The purpose of the present invention is to provide a wearable intelligent non-invasive blood glucose monitoring method and system based on microfluidics technology.

[0004] In a first aspect, an embodiment of the present invention provides a wearable intelligent non-invasive blood glucose monitoring method based on microfluidics technology, comprising:

[0005] In response to a blood glucose monitoring instruction, detecting a device state of a wearable device; the wearable device is configured with a microfluidic chip, and the microfluidic chip includes a plurality of microchannels;

[0006] On the basis that the device state is characterized as normal, collecting sweat of the target user into the multiple microchannels through the microfluidic chip of the wearable device;

[0007] performing an oxidase reaction on the sweat collected by the plurality of microchannels to obtain an oxidase reaction spectrum;

[0008] The blood glucose monitoring result corresponding to the target user is determined according to the oxidase reaction spectrum and the physiological indicator characteristics of the target user.

[0009] In a second aspect, an embodiment of the present invention provides a server system, including a server, wherein the server is used to execute the method described in the first aspect.

[0010] Compared with the prior art, the beneficial effects provided by the present invention include: using a wearable intelligent non-invasive blood glucose monitoring method and system based on microfluidic technology disclosed by the present invention, including: by responding to blood glucose monitoring instructions, detecting the state of the wearable device, using a microfluidic chip to collect sweat to multiple microchannels, performing oxidase reactions and obtaining reaction spectra, and combining the user's physiological index characteristics to accurately determine the blood glucose monitoring results. Such a design realizes non-invasive, real-time, and portable blood glucose monitoring, is easy to operate, has strong environmental adaptability, and has high application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative work.

[0012] Figure 1 A schematic diagram of the steps of a wearable intelligent non-invasive blood glucose monitoring method based on microfluidics technology provided by an embodiment of the present invention;

[0013] Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0015] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0016] In order to solve the technical problems in the aforementioned background technology, Figure 1 This is a flow chart of a wearable intelligent non-invasive blood glucose monitoring method based on microfluidic technology provided in an embodiment of the present disclosure. The wearable intelligent non-invasive blood glucose monitoring method based on microfluidic technology is introduced in detail below.

[0017] Step S201, in response to a blood glucose monitoring instruction, detecting a device state of a wearable device; the wearable device is configured with a microfluidic chip, and the microfluidic chip includes a plurality of microchannels;

[0018] Step S202, on the basis that the device state is characterized as normal, collecting the sweat of the target user into the multiple microchannels through the microfluidic chip of the wearable device;

[0019] Step S203, performing an oxidase reaction on the sweat collected by the plurality of microchannels to obtain an oxidase reaction spectrum;

[0020] Step S204: determining the blood glucose monitoring result corresponding to the target user according to the oxidase reaction spectrum and the physiological indicator characteristics of the target user.

[0021] In an embodiment of the present invention, illustratively, after waking up in the morning, the user sends a blood glucose monitoring instruction through an application on a smart phone. This instruction is transmitted to a remote server via a network. After receiving the instruction, the server immediately communicates with the intelligent non-invasive blood glucose monitoring device worn on the user's wrist. The server first checks the status of the device, including whether the device is online, whether the battery is sufficient, whether the microfluidic chip and the sensor are working properly, etc. For example, the server sends a status query request to the device via a Bluetooth connection, and the device replies to confirm that all systems are operating normally, the battery is full, and the microfluidic chip is ready. After confirming that the device is in a normal state, the server sends a command to start monitoring to the wearable device. The microfluidic chip built into the wearable device is immediately started, and the microchannels covered on its surface start working. These microchannels are exquisitely designed and can efficiently guide the natural sweat on the surface of the user's skin to flow in. With the user's slight activity or the natural perspiration of the skin, the sweat is gradually absorbed into the multiple microchannels of the microfluidic chip. The whole process does not require active operation by the user and is fully automated. Once the sweat is collected in the microchannel, the oxidase layer in the microfluidic chip begins to react with the glucose in the sweat. This oxidase reaction changes the optical properties of sweat, especially the absorption spectrum. The optical sensor built into the microfluidic chip monitors these changes in real time and generates corresponding oxidase reaction spectrum data. These data are then converted into digital signals and transmitted to the server wirelessly. After receiving the oxidase reaction spectrum data from the wearable device, the server uses complex algorithms to analyze it in combination with the physiological indicator characteristics of the user (such as age, gender, weight, historical blood sugar data, etc.) stored in advance on the server. These algorithms are based on a large amount of clinical data and machine learning models, and can quickly and accurately parse blood glucose concentration information from the oxidase reaction spectrum. After calculation, the server finally obtains the user's real-time blood glucose monitoring results and notifies the user instantly through a mobile phone application. For example, after analysis, the server shows that the user's current blood glucose value is 100mg / dL, which is within the normal range, and pushes a message to the mobile phone to remind the user to continue to maintain a healthy lifestyle. The above scenario example fully demonstrates the various steps of the wearable intelligent non-invasive blood glucose monitoring method based on microfluidic technology. From command reception, device status detection, sweat collection, oxidase reaction to result analysis, each step is closely connected to achieve non-invasive and real-time monitoring of the user's blood glucose level.

[0022] In the embodiment of the present invention, the aforementioned step S204 may be implemented through the following example execution.

[0023] Acquire optical signal characteristics in the oxidase reaction spectrum corresponding to the target user, and acquire physiological index characteristics corresponding to the target user;

[0024] Performing feature integration processing on the optical signal feature and the physiological index feature in the same feature domain to obtain a first integrated feature, and determining a first variable conversion unit corresponding to the optical signal feature from a plurality of variable conversion units according to the first integrated feature; the plurality of variable conversion units are respectively configured with feature integration standards of different scales for integrating multi-modal integrated features;

[0025] In the first variable conversion unit, a variable conversion operation is performed on the optical signal feature according to the physiological indicator feature to obtain a first variable conversion feature;

[0026] Performing feature integration processing on the physiological indicator feature and the first variable conversion feature in the same feature domain to obtain a second integrated feature, and determining a second variable conversion unit corresponding to the physiological indicator feature from among the multiple variable conversion units according to the second integrated feature;

[0027] In the second variable conversion unit, a variable conversion operation is performed on the physiological indicator feature according to the first variable conversion feature to obtain a second variable conversion feature;

[0028] Obtain a category label, and determine a blood glucose monitoring result corresponding to the target user according to the category label, the first variable conversion feature, and the second variable conversion feature.

[0029] In an embodiment of the present invention, illustratively, the server extracts the oxidase reaction spectrum from the data received by the wearable device and analyzes the optical signal characteristics therein, such as the light absorption intensity of a specific wavelength. At the same time, the server retrieves the physiological index characteristics of the user from the user profile, including age, gender, weight, BMI index, and historical blood sugar data. The server uses a machine learning algorithm to map the optical signal characteristics and the physiological index characteristics to the same feature domain and perform feature integration processing. For example, through feature splicing or feature fusion technology, the feature vector of the optical signal and the feature vector of the physiological index are combined into a new feature vector, that is, the first integrated feature. The server is preset with multiple variable conversion units, each of which is configured with feature integration standards of different scales. According to the first integrated feature, the server selects the most suitable variable conversion unit as the first variable conversion unit through a matching algorithm for further variable conversion of the optical signal feature. In the first variable conversion unit, the server uses the physiological index feature as the conversion basis, performs nonlinear transformation or scale transformation on the optical signal feature, and obtains the first variable conversion feature. This step is intended to make the optical signal feature more consistent with the spatial distribution of the physiological index feature and improve the accuracy of subsequent analysis. The server maps the physiological indicator feature and the first variable conversion feature to the same feature domain again for feature integration processing. A new feature vector, namely the second integrated feature, is generated through feature splicing or fusion technology. Based on the second integrated feature, the server selects the most suitable variable conversion unit as the second variable conversion unit through a matching algorithm again, which is used to further convert the physiological indicator feature. In the second variable conversion unit, the server uses the first variable conversion feature as the conversion basis to perform nonlinear transformation or scale transformation on the physiological indicator feature, thereby obtaining the second variable conversion feature. This step is intended to further optimize the physiological indicator feature so that it is more coordinated with the optical signal feature. The server obtains category labels related to blood sugar levels from the database, such as "normal", "high", "low", etc. Then, the server uses a machine learning classifier or regression model to perform comprehensive analysis and judgment based on the category label, the first variable conversion feature and the second variable conversion feature, and finally determines the blood sugar monitoring result corresponding to the target user. For example, the server outputs the user's blood sugar monitoring result as "normal" and notifies the user of the result instantly through a mobile phone application.

[0030] In the embodiment of the present invention, the step of obtaining the optical signal characteristics in the oxidase reaction spectrum corresponding to the target user may be implemented through the following examples.

[0031] Acquire an oxidase reaction spectrum corresponding to a target user, and cut the oxidase reaction spectrum into a plurality of local oxidase reaction spectra;

[0032] Performing a feature extraction operation on the multiple local oxidase reaction spectra to obtain local variable conversion features corresponding to the multiple local oxidase reaction spectra;

[0033] Performing focus enhancement processing on the local variable conversion features corresponding to the multiple local oxidase reaction spectra to obtain feature importance coefficients corresponding to the multiple local oxidase reaction spectra;

[0034] The local variable conversion features corresponding to the multiple local oxidase reaction spectra and the feature importance coefficients corresponding to the multiple local oxidase reaction spectra are linearly superimposed to obtain the optical signal features corresponding to the oxidase reaction spectra.

[0035] In an embodiment of the present invention, exemplarily, the server extracts a complete oxidase reaction spectrum from the data received by the wearable device. In order to analyze the information in the spectrum more carefully, the server uses a sliding window technique or a fixed interval cutting method to cut the complete oxidase reaction spectrum into multiple local spectral segments, each of which contains spectral information within a certain wavelength range. The server performs a feature extraction operation on each local oxidase reaction spectrum, which may include calculating the peak, valley, slope, curvature of the spectrum or performing a wavelet transform, etc. Through these operations, the server extracts local variable conversion features that can represent its characteristics from each local spectrum. The server performs focus enhancement processing on each local variable conversion feature, which involves using an attention mechanism, a weight assignment, or a feature selection algorithm. Through these methods, the server can evaluate the importance of each local variable conversion feature to the overall blood glucose monitoring task and assign a feature importance coefficient to each feature. The server multiplies each local variable conversion feature by its corresponding feature importance coefficient, and then linearly superimposes all the product results. In this way, the server can obtain an optical signal feature vector that integrates all local variable conversion features and their importance. This vector will be used for subsequent blood glucose monitoring result analysis. Through the above steps, the server can analyze the information in the oxidase reaction spectrum more finely and extract the optical signal features that are more critical to the task of blood glucose monitoring. This will help improve the accuracy and reliability of blood glucose monitoring.

[0036] In the embodiment of the present invention, the step of obtaining the physiological indicator characteristics corresponding to the target user may be implemented through the following examples.

[0037] Acquire multiple blood sugar related indicators corresponding to the target user, and use the multiple blood sugar related indicators as a blood sugar related indicator set;

[0038] The blood sugar related indicator set is loaded into a dense network architecture, and the dense network architecture is used to perform feature conversion processing on the blood sugar related indicator set to obtain physiological indicator features corresponding to the target user.

[0039] In an embodiment of the present invention, illustratively, the server retrieves multiple blood sugar indicators related to the user from the user's health record or historical monitoring data. These indicators may include the user's past blood sugar measurement value, glycosylated hemoglobin level, eating habits, exercise frequency, etc. The server integrates these indicators into a blood sugar related indicator set for subsequent analysis and processing. The server uses deep learning technology, especially dense network architecture (DenseNet), to perform feature conversion processing on the blood sugar related indicator set. The dense network architecture can effectively extract and utilize the complex relationship between indicators through the dense connection mechanism. The server takes the blood sugar related indicator set as input and obtains a series of high-dimensional physiological indicator features through the forward propagation process of the dense network architecture. These features can more comprehensively reflect the user's physiological state, especially aspects related to blood sugar levels. Through the above steps, the server can obtain the physiological indicator features corresponding to the target user, and these features will be used for subsequent blood sugar monitoring result analysis. Using the dense network architecture for feature conversion processing can more deeply mine and utilize the user's physiological information, and improve the accuracy and personalization of blood sugar monitoring.

[0040] In the embodiment of the present invention, the feature integration processing is performed on the optical signal feature and the physiological indicator feature in the same feature domain to obtain the first integrated feature, which can be implemented through the following examples.

[0041] Loading the optical signal feature and the physiological index feature into a Transformer network architecture, and using a first multi-layer perceptron in the Transformer network architecture to perform feature domain alignment on the optical signal feature to obtain an optical signal conversion feature;

[0042] Using the second multi-layer perceptron in the Transformer network architecture to align the feature domains of the physiological indicator features to obtain physiological indicator conversion features; the optical signal conversion features and the physiological indicator conversion features have the same feature domain;

[0043] The optical signal conversion feature and the physiological indicator conversion feature are subjected to feature integration processing to obtain a first integrated feature.

[0044] In an embodiment of the present invention, illustratively, the server has extracted the optical signal features and physiological indicator features of the target user. Now, the server loads these features into the pre-trained Transformer network architecture, ready for feature domain alignment and integration processing. The Transformer network architecture is very suitable for the integration of multimodal features due to its powerful sequence processing capabilities and self-attention mechanism. In the Transformer network architecture, the server uses the first multi-layer perceptron (MLP) to align the optical signal features with the feature domain. The multi-layer perceptron maps the optical signal features to a feature domain that is the same as the physiological indicator features through a series of nonlinear transformations. In this way, the optical signal conversion features have the same feature domain as the physiological indicator features, laying the foundation for subsequent feature integration. Similarly, the server uses the second multi-layer perceptron in the Transformer network architecture to align the physiological indicator features with the feature domain. This multi-layer perceptron also maps the physiological indicator features to the same feature domain as the optical signal conversion features. After such processing, the optical signal conversion features and the physiological indicator conversion features have the same feature domain, and feature integration can be performed directly. After obtaining the optical signal conversion feature and the physiological index conversion feature, the server integrates the two features using feature integration technology (such as feature splicing, feature fusion, etc.). Through the integration process, the server obtains a first integrated feature that contains the optical signal and physiological index information. This integrated feature will be used for subsequent variable conversion and blood glucose monitoring result analysis.

[0045] In the embodiment of the present invention, the optical signal conversion feature and the physiological indicator conversion feature are subjected to feature integration processing to obtain a first integrated feature, which can be implemented through the following examples.

[0046] Acquire a first quadratic mean corresponding to the optical signal conversion feature, and perform standardization processing on the optical signal conversion feature according to the first quadratic mean to obtain an optical signal standardized feature;

[0047] Performing nonlinear transformation on the standardized feature of the optical signal to obtain a nonlinear feature of the optical signal, and performing truncated mean processing on the nonlinear feature of the optical signal to obtain a first truncated mean feature;

[0048] Acquire a second quadratic mean corresponding to the physiological indicator conversion feature, and perform standardization processing on the physiological indicator conversion feature according to the second quadratic mean to obtain a physiological indicator standardized feature;

[0049] Performing nonlinear transformation on the physiological indicator standardized feature to obtain the physiological indicator nonlinear feature, and performing truncated mean processing on the physiological indicator nonlinear feature to obtain a second truncated mean feature;

[0050] The first truncated mean feature and the second truncated mean feature are superimposed to obtain a comprehensive truncated mean feature, and the comprehensive truncated mean feature is subjected to feature conversion processing according to the relationship strength vector corresponding to the affine transformation network architecture in the Transformer network architecture to obtain the first integrated feature.

[0051] In an embodiment of the present invention, exemplarily, the server calculates the quadratic mean of the optical signal conversion feature (which has been aligned with the feature domain by the multi-layer perceptron in the Transformer), that is, calculates the average of the squares of all eigenvalues. Then, the server uses this quadratic mean to standardize the optical signal conversion feature so that the processed eigenvalues ​​have a uniform scale, and obtains the optical signal standardized feature. The server applies a nonlinear transformation function (such as a Sigmoid function or a ReLU function) to the optical signal standardized feature to obtain the nonlinear feature of the optical signal. Then, the server performs a truncated mean process on this nonlinear feature, that is, calculates the mean after removing a certain proportion of the maximum and minimum values ​​to obtain the first truncated mean feature. Similar to the processing of the optical signal conversion feature, the server calculates the quadratic mean of the physiological indicator conversion feature (which has also been aligned with the feature domain by the multi-layer perceptron in the Transformer), and uses this quadratic mean to standardize the physiological indicator conversion feature to obtain the physiological indicator standardized feature. The server applies a nonlinear transformation function to the physiological indicator standardized feature to obtain the nonlinear feature of the physiological indicator. Then, the server performs a truncated mean process on this nonlinear feature to obtain the second truncated mean feature. The server simply superimposes (or weighted averages) the first truncated mean feature and the second truncated mean feature to obtain a comprehensive truncated mean feature that integrates the optical signal and physiological indicator information. In the Transformer network architecture, there is an affine transformation network architecture for learning the relationship strength between different features. The server uses this relationship strength vector to perform feature transformation processing on the comprehensive truncated mean feature to obtain the first integrated feature. This first integrated feature will be used for subsequent blood glucose monitoring result analysis.

[0052] In an embodiment of the present invention, determining the first variable conversion unit corresponding to the optical signal feature from a plurality of variable conversion units according to the first integrated feature includes:

[0053] Using the classifier in the Transformer network architecture to perform type identification on the first integrated feature, and obtain unit adaptation coefficients corresponding to multiple variable conversion units;

[0054] The variable conversion unit corresponding to the largest unit adaptation coefficient is used as the first variable conversion unit corresponding to the optical signal feature.

[0055] In an embodiment of the present invention, illustratively, the server has obtained a first integrated feature, which integrates the information of optical signals and physiological indicators. Now, the server uses the classifier in the Transformer network architecture to identify the type of this first integrated feature. The role of the classifier is to calculate a unit adaptation coefficient for each possible variable conversion unit, which represents the degree of adaptation between the first integrated feature and each variable conversion unit. Specifically, assume that the server has 5 different variable conversion units, which are used to process different types of optical signal features. The classifier will calculate a unit adaptation coefficient for each of the 5 variable conversion units, such as 0.2, 0.5, 0.7, 0.3 and 0.1. After obtaining the unit adaptation coefficients corresponding to multiple variable conversion units, the server will find the largest coefficient. In the above example, the largest unit adaptation coefficient is 0.7. The server will use the variable conversion unit corresponding to the largest unit adaptation coefficient as the first variable conversion unit corresponding to the optical signal feature. That is to say, in this example, the server will select the variable conversion unit with a unit adaptation coefficient of 0.7 as the first variable conversion unit for subsequent optical signal feature processing.

[0056] In an embodiment of the present invention, the first variable conversion unit includes a feature weighting unit; in the first variable conversion unit, a variable conversion operation is performed on the optical signal feature according to the physiological indicator feature to obtain a first variable conversion feature, which can be implemented through the following example.

[0057] Fusing the optical signal feature and the physiological index feature to obtain a first fusion feature;

[0058] Using the feature weighting unit to perform focus enhancement processing on the first fused feature to obtain a first focus enhancement feature;

[0059] A preset first number of features among the first focus enhancement features are used as first variable conversion features.

[0060] In an embodiment of the present invention, illustratively, the server has determined the first variable conversion unit corresponding to the optical signal feature, and the first variable conversion unit includes a feature weighting unit. Now, the server needs to fuse the optical signal feature and the physiological indicator feature. The fusion process can be to splice two feature vectors, or to merge them into a new feature vector in some way. After the fusion process, the server obtains a first fusion feature, which contains information of both the optical signal and the physiological indicator. In the first variable conversion unit, there is a feature weighting unit, which is used to perform focus enhancement processing on the fused feature. Focus enhancement processing can be to weight certain elements in the feature vector so that these elements are more important in subsequent processing. The server uses the feature weighting unit to perform focus enhancement processing on the first fusion feature to obtain a first focus enhancement feature. This feature is more prominent in some aspects and is helpful for the subsequent analysis of blood glucose monitoring results. In the first focus enhancement feature, many feature elements can be included. However, not all feature elements are equally important to the subsequent analysis of blood glucose monitoring results. Therefore, the server presets a front number, such as selecting the first 10 or 20 feature elements. The server uses these preset first number of feature elements as the first variable conversion feature for subsequent blood sugar monitoring result analysis. This processing method can simplify the subsequent analysis process while retaining the most important feature information.

[0061] In the embodiment of the present invention, the use of the feature weighting unit to perform focus enhancement processing on the first fusion feature to obtain a first focus enhancement feature can be implemented through the following examples.

[0062] According to the mapping relationship strength vector corresponding to the feature weighting unit, the first fused feature is converted into a first index vector, a first identifier vector and a first attribute vector;

[0063] Constructing a first undetermined vector according to a cross product between the first index vector and the optical signal feature, and a cross product between the first index vector and the physiological indicator feature;

[0064] Constructing a second undetermined vector according to a mapping vector of a cross product between the first identifier vector and the optical signal feature, and a mapping vector of a cross product between the first identifier vector and the physiological indicator feature;

[0065] constructing a third undetermined vector according to the cross product between the first attribute vector and the optical signal feature, and the cross product between the first attribute vector and the physiological indicator feature;

[0066] Taking the cross product between the first undetermined vector and the second undetermined vector as a fourth undetermined vector, and obtaining the number of elements of the optical signal feature;

[0067] The ratio of the fourth undetermined vector to the square root of the number of elements is normalized to obtain a self-feature importance coefficient vector, and the cross product between the self-feature importance coefficient vector and the third undetermined vector is used as the first focus enhancement feature.

[0068] In an embodiment of the present invention, exemplarily, the feature weighting unit in the server has a corresponding mapping relationship strength vector, which defines how to convert the fused features into different vector representations. The server uses this mapping relationship strength vector to convert the first fused feature into three different vectors: a first index vector, a first identifier vector, and a first attribute vector. These three vectors represent different aspects of the fused feature for subsequent focus enhancement processing. The server calculates the cross product between the first index vector and the optical signal feature, and the cross product between the first index vector and the physiological indicator feature. Then, the server combines the two cross product results in some form (such as splicing or adding) to construct a first pending vector. This vector contains the interaction information between the optical signal feature and the physiological indicator feature in terms of the first index vector. Similar to constructing the first pending vector, the server calculates the mapping vector of the cross product between the first identifier vector and the optical signal feature, and the mapping vector of the cross product between the first identifier vector and the physiological indicator feature. Then, the server combines the two mapping vectors to construct a second pending vector. This vector contains the interaction information between the optical signal feature and the physiological indicator feature in terms of the first identifier vector. The server calculates the cross product between the first attribute vector and the optical signal feature, and the cross product between the first attribute vector and the physiological indicator feature. Then, the server combines the two cross product results to construct a third undetermined vector. This vector contains the interactive information between the optical signal feature and the physiological indicator feature in terms of the first attribute vector. The server calculates the cross product between the first undetermined vector and the second undetermined vector to obtain a fourth undetermined vector. This vector combines the interactive information between the optical signal feature and the physiological indicator feature in terms of the first index vector and the first identifier vector. At the same time, the server obtains the number of elements of the optical signal feature, which represents the number of characteristic elements contained in the optical signal feature. The server calculates the ratio of the fourth undetermined vector to the square root of the number of elements of the optical signal feature, and performs normalization to obtain a self-feature importance coefficient vector. This vector represents the importance of each characteristic element in the optical signal feature. Then, the server calculates the cross product of the self-feature importance coefficient vector and the third undetermined vector to obtain a first focus enhancement feature. This feature strengthens the important aspects of the optical signal feature and integrates the information of the physiological indicator feature for subsequent analysis of blood glucose monitoring results.

[0069] In an embodiment of the present invention, the first variable conversion unit includes a feature weighting unit;

[0070] In the first variable conversion unit, performing a variable conversion operation on the optical signal feature according to the physiological indicator feature to obtain a first variable conversion feature includes:

[0071] Using the feature weighting unit, converting the optical signal feature into a second index vector, and converting the physiological indicator feature into a second identifier vector and a second attribute vector;

[0072] Performing data encoding on the cross product between the second index vector and the optical signal feature, and the mapping vector of the cross product between the second identifier vector and the physiological indicator feature, to obtain a fifth vector to be determined, and acquiring the number of elements of the optical signal feature;

[0073] Standardizing the ratio between the fifth undetermined vector and the square root of the number of elements to obtain a multidimensional feature importance coefficient vector, and performing data encoding on the multidimensional feature importance coefficient vector and the cross product between the second attribute vector and the physiological indicator feature to obtain a multidimensional gating feature;

[0074] A preset first number of features in the multi-dimensional gating features are used as first variable conversion features.

[0075] In an embodiment of the present invention, exemplarily, the feature weighting unit in the server is responsible for converting the input features into different vector representations. In this step, the server converts the optical signal feature into a second index vector, which represents the representation of the optical signal feature in a certain index space. At the same time, the server converts the physiological indicator feature into a second identifier vector and a second attribute vector, which respectively represent the representation of the physiological indicator feature in the identifier space and the attribute space. The server calculates the cross product between the second index vector and the optical signal feature, and the mapping vector of the cross product between the second identifier vector and the physiological indicator feature. Then, the server performs data encoding on the two cross product results to obtain a fifth undetermined vector. This vector integrates the interactive information of the optical signal feature and the physiological indicator feature in the second index vector and the second identifier vector. At the same time, the server obtains the number of elements of the optical signal feature, which represents the number of feature elements contained in the optical signal feature. The server calculates the ratio of the fifth undetermined vector to the square root of the number of elements of the optical signal feature, and performs normalization processing to obtain a multi-dimensional feature importance coefficient vector. This vector represents the importance of the optical signal feature and the physiological indicator feature in each dimension. The server calculates the cross product result of the multidimensional feature importance coefficient vector and the cross product between the second attribute vector and the physiological indicator feature, and performs data encoding on the result to obtain a multidimensional gating feature. This feature integrates the interactive information of the optical signal feature and the physiological indicator feature in terms of the multidimensional feature importance coefficient vector and the second attribute vector. In the multidimensional gating feature, many feature elements can be included. However, not all feature elements are equally important for the subsequent analysis of blood glucose monitoring results. Therefore, the server will preset a top number, such as selecting the top 10 or top 20 feature elements. The server uses these preset top number feature elements as the first variable conversion features for subsequent analysis of blood glucose monitoring results. This processing method can simplify the subsequent analysis process while retaining the most important feature information.

[0076] In an embodiment of the present invention, the first variable conversion unit includes a bottleneck feature weighting unit;

[0077] In the first variable conversion unit, performing a variable conversion operation on the optical signal feature according to the physiological indicator feature to obtain a first variable conversion feature includes:

[0078] Acquire correlation features, fuse the correlation features with the physiological indicator features to obtain second fusion features; the correlation features are used to establish correlation relationships between features of different dimensions;

[0079] Using the first gated recurrent network architecture in the bottleneck feature weighting unit, the second fused feature is subjected to focus enhancement processing to obtain a second focus enhancement feature, and a preset number of first features in the second focus enhancement feature are used as candidate variable conversion features;

[0080] The optical signal feature and the candidate variable conversion feature are fused to obtain a third fused feature, and the third fused feature is subjected to focus enhancement processing by using the second gated recurrent network architecture in the bottleneck feature weighting unit to obtain a third focus enhancement feature;

[0081] A preset first number of features in the third focus enhancement feature are used as first variable conversion features.

[0082] In an embodiment of the present invention, exemplarily, the server first obtains a correlation feature, which is used to establish a correlation relationship between features of different dimensions, and can be a pre-trained correlation matrix or a correlation vector calculated by a certain algorithm. Then, the server fuses the correlation feature with the physiological indicator feature, such as by splicing or a certain fusion algorithm, to obtain a second fusion feature. This fusion feature contains both the information of the physiological indicator and the correlation information between the features of different dimensions. The bottleneck feature weighting unit in the server includes a first gated recurrent network architecture, which can perform focus enhancement processing on the input features. The server inputs the second fusion feature into the first gated recurrent network architecture, and the network performs a series of calculations, such as the switch operation of the gate mechanism, to focus on the fusion feature, so that some important features are more prominent. After processing, the server obtains a second focus enhancement feature. In the second focus enhancement feature, many feature elements may be included. The server will preset a top number, such as selecting the top 10 or top 20 feature elements. The server uses these preset top number feature elements as candidate variable conversion features, which may be more important in the subsequent analysis of blood glucose monitoring results. The server fuses the optical signal feature and the candidate variable conversion feature, for example, through splicing or some fusion algorithm, to obtain a third fused feature. This fused feature contains both the information of the optical signal and the information of the candidate variable conversion feature. The bottleneck feature weighting unit in the server also includes a second gated recurrent network architecture, which can also perform focus enhancement processing on the input features. The server inputs this third fused feature into the second gated recurrent network architecture, and the network performs focus enhancement on the fused feature through a series of calculations. After processing, the server obtains the third focus enhancement feature. The third focus enhancement feature can also include many feature elements. The server will again preset a top number, such as selecting the top 10 or top 20 feature elements. The server uses these preset top number feature elements as the first variable conversion features, which are the most important features in the subsequent analysis of blood glucose monitoring results.

[0083] In an embodiment of the present invention, the first variable conversion unit includes a physiological key indicator conversion unit; in the first variable conversion unit, a variable conversion operation is performed on the optical signal feature according to the physiological indicator feature to obtain a first variable conversion feature, which can be implemented through the following example.

[0084] Acquire an optical signal standardized feature corresponding to the optical signal feature, and a physiological index standardized feature corresponding to the physiological index feature;

[0085] Using the first quantum neural network in the key physiological indicator conversion unit, the optical signal characteristics are quantum entangled and interacted to obtain first quantum entanglement and interaction characteristics;

[0086] Using the second quantum neural network in the key physiological index conversion unit, the characteristics of the physiological index are quantum entangled and interacted to obtain second quantum entanglement and interaction characteristics;

[0087] The second quantum entanglement and interaction feature is subjected to truncated mean processing to obtain a third truncated mean feature, and the first quantum entanglement and interaction feature and the third truncated mean feature are superimposed to obtain a first variable conversion feature.

[0088] In an embodiment of the present invention, illustratively, the server first obtains optical signal features and physiological index features from a data source. Then, these features are standardized, such as by removing the mean and dividing by the standard deviation, so that the feature data conforms to the standard normal distribution. After standardization, the server obtains standardized features of optical signals and standardized features of physiological indicators. The key physiological index conversion unit in the server includes a first quantum neural network, which can perform quantum entanglement and interaction processing on the input features. The server inputs the standardized features of the optical signal into the first quantum neural network, and the network converts the optical signal features by simulating the process of quantum entanglement and interaction to obtain the first quantum entanglement and interaction features. This feature contains the entanglement and interaction information of the optical signal at the quantum level. Similarly, the key physiological index conversion unit in the server also includes a second quantum neural network. The server inputs the standardized features of the physiological indicators into the second quantum neural network, and the network converts the physiological index features by simulating the process of quantum entanglement and interaction to obtain the second quantum entanglement and interaction features. This feature contains the entanglement and interaction information of the physiological indicators at the quantum level. The server performs truncated mean processing on the second quantum entanglement and interaction features. This processing method is to remove the extreme values ​​​​in the feature (such as the maximum and minimum values ​​​​), and then calculate the mean of the remaining values ​​​​. After processing, the server obtains the third truncated mean feature, which removes the influence of extreme values ​​​​and is more stable. Finally, the server superimposes the first quantum entanglement and interaction feature and the third truncated mean feature. This superposition can be a simple addition or a more complex fusion method. After superposition processing, the server obtains the first variable conversion feature, which contains both the entanglement and interaction information of optical signals and physiological indicators at the quantum level and the stability information after removing extreme values ​​​​. This feature will be used for subsequent blood glucose monitoring results analysis.

[0089] In an embodiment of the present invention, the first variable conversion unit includes a key dimension retention unit;

[0090] In the first variable conversion unit, a variable conversion operation is performed on the optical signal feature according to the physiological indicator feature to obtain a first variable conversion feature, which can be implemented through the following examples.

[0091] The physiological indicator feature is deleted, and the optical signal feature is used as the first variable conversion feature output by the key dimension retention unit.

[0092] In an embodiment of the present invention, exemplarily, when processing data, the server first identifies the physiological indicator features in the data set. These features may be certain physiological indicators related to the user's blood sugar level, such as heart rate, blood pressure, etc. The server performs an operation to delete these physiological indicator features from the data set, that is, no longer consider the impact of these features on subsequent analysis. After deleting the physiological indicator features, the server will focus on the optical signal features. These optical signal features may be data collected by a certain optical sensor and directly related to the user's blood sugar level. The server regards these optical signal features as key dimensions and directly uses these features as the output of the first variable conversion unit, that is, the first variable conversion features. These features will be used for subsequent analysis of blood sugar monitoring results. Through such processing, the server is actually performing a feature selection operation, which believes that the optical signal features are more important for the analysis of blood sugar monitoring results, while the physiological indicator features can be ignored at this stage. This method is suitable for situations where the optical signal features have a strong correlation with blood sugar levels and the physiological indicator features have a weak correlation.

[0093] In the embodiment of the present invention, determining the blood glucose monitoring result corresponding to the target user according to the category label, the first variable conversion feature and the second variable conversion feature can be implemented through the following example.

[0094] fusing the first variable conversion feature and the second variable conversion feature to obtain a fourth fused feature, configuring the category label on an initial element of the fourth fused feature to obtain a variable conversion feature;

[0095] The variable conversion feature is subjected to focus enhancement processing by using a gated recurrent network architecture and a focus enhancement network architecture to obtain a feature weighted feature, and the feature weighted feature is subjected to feature conversion processing to obtain a blood glucose monitoring result corresponding to the target user.

[0096] In an embodiment of the present invention, illustratively, the server has obtained a first variable conversion feature and a second variable conversion feature, which respectively represent the optical signal feature and the physiological index feature (or some variant thereof) after different conversion processing. In order to make comprehensive use of these two features, the server fuses them, such as by splicing or some fusion algorithm, merging the two features into a new feature vector, i.e., the fourth fusion feature. After obtaining the fourth fusion feature, the server also needs to consider the information of the category label. The category label represents different categories of the user's blood sugar level, such as normal, high, low, etc. The server configures the category label to the initial element position of the fourth fusion feature, so that each fusion feature will carry a corresponding category label to form the final variable conversion feature. In order to further enhance the expressive power and importance of the feature, the server uses the gated recurrent network architecture focus enhancement network architecture to focus on the variable conversion feature. This network architecture can perform weighted processing on the feature through a series of calculations, such as the switch operation of the gate mechanism, so that some important features are more prominent. After processing, the server obtains the feature weighted feature. Finally, the server performs feature conversion processing on the feature weighted feature. This processing can be achieved through a classification algorithm or a regression algorithm, the purpose of which is to convert the feature weighted features into specific blood glucose monitoring results. After processing, the server obtains the blood glucose monitoring result corresponding to the target user, which can be a specific blood glucose value or a category label of a blood glucose level.

[0097] In the embodiment of the present invention, after the aforementioned step S204, the following implementation is further provided.

[0098] Obtaining a blood sugar monitoring result from a target user, and obtaining a first target blood sugar health state based on the blood sugar monitoring result;

[0099] Based on the reliability of the user data corresponding to the target user, obtaining a matching blood sugar health state related to the first target blood sugar health state from a preset database; and

[0100] A blood glucose management plan corresponding to the matching blood glucose health status is searched from the preset database.

[0101] In the embodiment of the present invention, illustratively, the server has determined the blood glucose monitoring result corresponding to the target user through the previous steps, and this result can be a specific blood glucose value or a category label of a blood glucose level. The server further determines the current blood glucose health state of the target user, such as "normal", "high" or "low" based on the blood glucose monitoring result, through a certain algorithm or rule, and uses this state as the first target blood glucose health state. A preset database is maintained in the server, which stores different blood glucose health states and various information related thereto. The server evaluates the reliability of the target user data, and this reliability can be calculated based on factors such as the integrity, accuracy, and timeliness of the user data. Then, the server searches for a matching blood glucose health state related thereto from the preset database based on the reliability of the first target blood glucose health state and the user data. For example, if the first target blood glucose health state is "high" and the user data has a high reliability, the server may find "hyperglycemia" as a matching blood glucose health state. After finding a matching blood glucose health state, the server continues to search for a blood glucose management plan corresponding to this state in the preset database. This plan may include dietary advice, exercise advice, drug treatment advice, etc. For example, if the matching blood sugar health status is "high blood sugar", the server may find a blood sugar management plan that includes suggestions such as reducing sugar intake and increasing aerobic exercise. The server will provide this plan as the final output result to the target user or relevant medical staff for reference.

[0102] In an embodiment of the present invention, the preset database includes a plurality of blood sugar health status relationship pairs, and the blood sugar health status relationship pairs include association conditions between blood sugar health status and the blood sugar health status.

[0103] In an embodiment of the present invention, exemplarily, the preset database in the server not only includes the corresponding relationship between the blood sugar health state and the blood sugar management plan, but also includes multiple blood sugar health state relationship pairs. These relationship pairs describe the association between different blood sugar health states. For example, the preset database may include the following blood sugar health state relationship pairs: "Normal blood sugar" is associated with "low risk", indicating that the user's blood sugar level is within the normal range and the related health risk is low. "Hyperglycemia" is associated with "high risk" and "may require drug treatment", indicating that the user's blood sugar level is high, there is a high health risk, and may require drug treatment. "Hypoglycemia" is associated with "high risk" and "needs immediate treatment", indicating that the user's blood sugar level is low, there is also a high health risk, and immediate measures need to be taken to deal with it. Through such a design, the server can have a more comprehensive understanding of the association between different blood sugar health states, and consider these associated factors when providing blood sugar management plans for users, thereby providing more personalized and accurate suggestions. For example, when the server determines that the target user's first target blood sugar health status is "hyperglycemia", it will not only look for a blood sugar management plan corresponding to "hyperglycemia", but will also consider other factors associated with "hyperglycemia", such as "high risk" and "may require drug treatment", and incorporate these factors into the blood sugar management plan provided to ensure the comprehensiveness and effectiveness of the plan.

[0104] In the embodiment of the present invention, the obtaining of a matching blood sugar health state related to the first target blood sugar health state from a preset database based on the reliability of the user data corresponding to the target user can be implemented through the following examples.

[0105] On the basis that the reliability of the user data is completely reliable, obtaining a first blood sugar health state relationship pair including the first target blood sugar health state from the plurality of blood sugar health state relationship pairs; and

[0106] Based on the first blood sugar health state relationship pair, a matching blood sugar health state related to the first target blood sugar health state is obtained.

[0107] In an embodiment of the present invention, illustratively, the server has determined the first target blood sugar health state of the target user, such as "hyperglycemia". At the same time, the server also evaluates the reliability of the target user data and determines that the user data is "completely reliable". Next, the server needs to find a matching blood sugar health state related to this "hyperglycemia" state in the preset database. Since the user data is completely reliable, the server decides to use multiple blood sugar health state relationship pairs in the preset database to assist the matching process. The server searches for the first blood sugar health state relationship pair containing "hyperglycemia" as the first target blood sugar health state in the preset database. It finds a relationship pair: "hyperglycemia" is associated with "high risk" and "may require drug treatment". Based on this first blood sugar health state relationship pair, the server determines that the matching blood sugar health state related to the "hyperglycemia" state includes not only "hyperglycemia" itself, but also the associated "high risk" and "may require drug treatment". Finally, when providing a blood sugar management plan, the server will consider these matching blood sugar health states to ensure that the plan is not only for the "hyperglycemia" state, but also takes into account the risks and possible drug treatment needs associated with it. In this way, the server can provide more comprehensive, personalized and accurate blood sugar management suggestions.

[0108] In the embodiments of the present invention, the following implementation modes are also provided.

[0109] For each of the matching blood sugar health states, obtaining a second blood sugar health state relationship pair related to the matching blood sugar health state based on the multiple blood sugar health state relationship pairs;

[0110] On the basis that the number of the second blood sugar health status relationship pairs obtained exceeds the first number, a new matching blood sugar health status is obtained based on the second blood sugar health status relationship pairs.

[0111] In an embodiment of the present invention, illustratively, the server has obtained matching blood sugar health states from a preset database based on the target user's first target blood sugar health state "hyperglycemia" and completely reliable user data, including "hyperglycemia" itself and the associated "high risk" and "may require drug treatment". Next, the server further analyzes each matching blood sugar health state. Taking "high risk" as an example, the server searches for a second blood sugar health state relationship pair associated with "high risk" in the preset database. It finds multiple relationship pairs, such as "high risk" associated with "regular monitoring required", "high risk" associated with "recommended lifestyle adjustments", etc. The server counts the number of obtained second blood sugar health state relationship pairs. If it is found that this number exceeds the preset first number (such as 3), the server will obtain a new matching blood sugar health state based on these second blood sugar health state relationship pairs. In this example, since the number of second blood sugar health state relationship pairs associated with "high risk" exceeds the first number, the server also adds the blood sugar health states in these relationship pairs, such as "regular monitoring required" and "recommended lifestyle adjustments", to the list of matching blood sugar health states. Finally, when providing blood sugar management solutions, the server will take into account all new matching blood sugar health states, including the original "high blood sugar", "high risk" and "may need medication", as well as the newly added "need regular monitoring" and "recommended lifestyle adjustments". In this way, the server can ensure that the blood sugar management solutions provided are more comprehensive, detailed and personalized.

[0112] In the embodiment of the present invention, the status details of the blood sugar health status include the blood sugar health status category and the intervention level;

[0113] Wherein, obtaining a matching blood sugar health state related to the first target blood sugar health state from a preset database based on the reliability of the user data corresponding to the target user can be implemented through the following examples.

[0114] On the basis that the reliability of the user data is not completely reliable, based on the interference level of the blood sugar health status included in each of the first blood sugar health status relationship pairs, obtaining the interference level of each of the first blood sugar health status relationship pairs;

[0115] Based on the status details of the target user, obtaining a target interferability level corresponding to the target user;

[0116] Based on the target intervenability level and the intervenability levels of each of the first blood sugar health status relationship pairs, obtaining a third blood sugar health status relationship pair from the first blood sugar health status relationship pair; and

[0117] Based on the third blood sugar health state relationship pair, a matching blood sugar health state related to the first target blood sugar health state is obtained.

[0118] In the embodiment of the present invention, illustratively, when processing the blood sugar data of the target user, the server has determined that the first target blood sugar health state is "hyperglycemia". However, this time when the server evaluates the reliability of the target user's data, it is judged as "not completely reliable". In this case, the server decides to use a different method to obtain a matching blood sugar health state related to "hyperglycemia". First, it checks the first blood sugar health state relationship pairs in the preset database, which not only contain blood sugar health states, but also contain the interference level of each state. For example, "hyperglycemia" is associated with "high risk" and "may need drug treatment", where "may need drug treatment" has a higher interference level because it directly involves medical decision-making. Next, the server obtains the target interference level based on the status details of the target user. Assume that the status details of the target user show that they have tried drug treatment before, but the effect is not good, so their target interference level may focus more on non-drug treatment or lifestyle adjustment. Now, the server screens from the first blood sugar health state relationship pairs based on the target interference level and the interference level of each of the first blood sugar health state relationship pairs. It selects those relationship pairs whose intervention levels match or are close to the target intervention levels as the third blood glucose health state relationship pairs. For example, it may select the relationship pairs associated with "hyperglycemia" and "recommended lifestyle adjustments" and "regular monitoring required" because these recommendations do not require drug treatment and are consistent with the target intervention level of the target user. Finally, based on the third blood glucose health state relationship pair, the server obtains matching blood glucose health states related to the "hyperglycemia" state, this time including "hyperglycemia" itself and the associated "recommended lifestyle adjustments" and "regular monitoring required". When providing blood glucose management solutions, the server will give special consideration to these matching blood glucose health states to ensure that the solutions are targeted at the "hyperglycemia" state and meet the current target intervention level and lifestyle adjustment requirements of the target user. In this way, even if the user data is not completely reliable, the server can provide personalized and practical blood glucose management suggestions.

[0119] In the embodiments of the present invention, the following implementation modes are also provided.

[0120] For each of the matching blood sugar health states, obtaining a fourth blood sugar health state relationship pair related to the matching blood sugar health state based on the multiple blood sugar health state relationship pairs;

[0121] On the basis that the number of the fourth blood sugar health status relationship pairs obtained exceeds the second number, obtaining a fifth blood sugar health status relationship pair from the fourth blood sugar health status relationship pairs based on the target intervenability level and the intervenability levels of the fourth blood sugar health status relationship pairs; and

[0122] A new matching blood sugar health state is obtained based on the fifth blood sugar health state relationship pair.

[0123] In an embodiment of the present invention, illustratively, the server has obtained matching blood sugar health states from a preset database based on the first target blood sugar health state "hyperglycemia" of the target user, the unreliable user data, and the target interference level, including "hyperglycemia" itself and the associated "recommended lifestyle adjustments" and "regular monitoring required". Next, the server further analyzes and expands each matching blood sugar health state. Taking "recommended lifestyle adjustments" as an example, the server searches for the fourth blood sugar health state relationship pair related to "recommended lifestyle adjustments" in the preset database. It finds multiple relationship pairs, such as "recommended lifestyle adjustments" associated with "recommended low-sugar diet", "recommended lifestyle adjustments" associated with "increased aerobic exercise", etc. The server counts the number of fourth blood sugar health state relationship pairs obtained. If it is found that this number exceeds the preset second number (such as 2), the server will obtain the fifth blood sugar health state relationship pair based on the target interference level and the respective interference levels of these fourth blood sugar health state relationship pairs. In this example, since the number of fourth blood sugar health state relationship pairs associated with "recommended lifestyle adjustments" exceeds the second number, and these recommendations are consistent with the target intervention level of the target user (non-drug treatment or lifestyle adjustments), the server also adds the blood sugar health states in these relationship pairs, such as "recommended low-sugar diet" and "increased aerobic exercise", to the list of matching blood sugar health states. Ultimately, when providing a blood sugar management plan, the server will consider all new matching blood sugar health states, including the original "high blood sugar", "recommended lifestyle adjustments" and "regular monitoring required", as well as the newly added "recommended low-sugar diet" and "increased aerobic exercise". In this way, even if the user data is not completely reliable, the server can provide more comprehensive, detailed and personalized blood sugar management recommendations by continuously expanding and analyzing blood sugar health state relationship pairs.

[0124] In the embodiment of the present invention, the obtaining of blood glucose monitoring results from the target user and obtaining the first target blood glucose health state based on the blood glucose monitoring results can be implemented through the following examples.

[0125] Acquire the blood sugar monitoring result, and extract a first blood sugar semantic representation from the blood sugar monitoring result;

[0126] Processing the first blood glucose semantic representation using the blood glucose knowledge graph to obtain a second blood glucose semantic representation; and

[0127] Based on the second blood glucose semantic representation, the first target blood glucose health state is obtained.

[0128] In an embodiment of the present invention, illustratively, the server receives blood glucose monitoring results from a target user, which is a series of measurement data about the user's blood glucose level. In order to extract useful information from these data, the server first performs a preprocessing step, that is, extracts a first blood glucose semantic representation from the blood glucose monitoring results. This representation may include basic information such as the value of the blood glucose level and the measurement time. Next, the server uses a pre-built blood glucose knowledge graph to process the first blood glucose semantic representation. The blood glucose knowledge graph is a structured knowledge base containing a large amount of information about blood glucose health status, the relationship between blood glucose levels and various factors. By querying and reasoning in this knowledge graph, the server can convert the first blood glucose semantic representation into a richer and more specific second blood glucose semantic representation. This representation may include the range of the user's blood glucose level, the corresponding health status description, possible risk factors, etc. Finally, based on this second blood glucose semantic representation, the server can determine the first target blood glucose health state of the target user. For example, if the second blood glucose semantic representation shows that the user's blood glucose level is continuously high, then the server will use "hyperglycemia" as the first target blood glucose health state. Through this processing, the server can accurately extract useful information from the user's blood sugar monitoring results and convert it into specific blood sugar health status, laying the foundation for the subsequent provision of personalized blood sugar management plans.

[0129] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned wearable intelligent non-invasive blood glucose monitoring method based on microfluidics technology. Figure 2 As shown, Figure 2 The block diagram of the computer device 100 provided in the embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112 and a communication unit 113. To achieve data transmission or interaction, the memory 111, the processor 112 and the communication unit 113 are directly or indirectly electrically connected to each other. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0130] For illustrative purposes, the foregoing description is made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise form disclosed. Numerous modifications and variations are possible in accordance with the above teachings. These embodiments are selected and described in order to best illustrate the principles of the present disclosure and its practical application, so that those skilled in the art can best utilize the present disclosure and utilize various embodiments with different modifications to suit the intended specific application.

Claims

1. A wearable intelligent non-invasive blood glucose monitoring method based on microfluidics technology, characterized in that: include: In response to a blood glucose monitoring instruction, detecting a device state of the wearable device; The wearable device is equipped with a microfluidic chip, and the microfluidic chip includes a plurality of microchannels; On the basis that the device state is characterized as normal, collecting sweat of the target user into the multiple microchannels through the microfluidic chip of the wearable device; performing an oxidase reaction on the sweat collected by the plurality of microchannels to obtain an oxidase reaction spectrum; The blood glucose monitoring result corresponding to the target user is determined according to the oxidase reaction spectrum and the physiological indicator characteristics of the target user.

2. The method according to claim 1, characterized in that Determining the blood glucose monitoring result corresponding to the target user according to the oxidase reaction spectrum and the physiological indicator characteristics of the target user includes: Acquire an oxidase reaction spectrum corresponding to a target user, and cut the oxidase reaction spectrum into a plurality of local oxidase reaction spectra; Performing a feature extraction operation on the multiple local oxidase reaction spectra to obtain local variable conversion features corresponding to the multiple local oxidase reaction spectra; Performing focus enhancement processing on the local variable conversion features corresponding to the multiple local oxidase reaction spectra to obtain feature importance coefficients corresponding to the multiple local oxidase reaction spectra; Performing linear superposition processing on the local variable conversion features corresponding to the multiple local oxidase reaction spectra and the feature importance coefficients corresponding to the multiple local oxidase reaction spectra to obtain the optical signal features corresponding to the oxidase reaction spectra; Acquire multiple blood sugar related indicators corresponding to the target user, and use the multiple blood sugar related indicators as a blood sugar related indicator set; Loading the blood sugar related indicator set into a dense network architecture, and performing feature conversion processing on the blood sugar related indicator set using the dense network architecture to obtain physiological indicator features corresponding to the target user; Loading the optical signal feature and the physiological index feature into a Transformer network architecture, and using a first multi-layer perceptron in the Transformer network architecture to perform feature domain alignment on the optical signal feature to obtain an optical signal conversion feature; Using the second multi-layer perceptron in the Transformer network architecture to align the feature domains of the physiological indicator features to obtain physiological indicator conversion features; the optical signal conversion features and the physiological indicator conversion features have the same feature domain; Acquire a first quadratic mean corresponding to the optical signal conversion feature, and perform standardization processing on the optical signal conversion feature according to the first quadratic mean to obtain an optical signal standardized feature; Performing nonlinear transformation on the standardized feature of the optical signal to obtain a nonlinear feature of the optical signal, and performing truncated mean processing on the nonlinear feature of the optical signal to obtain a first truncated mean feature; Acquire a second quadratic mean corresponding to the physiological indicator conversion feature, and perform standardization processing on the physiological indicator conversion feature according to the second quadratic mean to obtain a physiological indicator standardized feature; Performing nonlinear transformation on the physiological indicator standardized feature to obtain the physiological indicator nonlinear feature, and performing truncated mean processing on the physiological indicator nonlinear feature to obtain a second truncated mean feature; The first truncated mean feature and the second truncated mean feature are superimposed to obtain a comprehensive truncated mean feature, and according to the relationship strength vector corresponding to the affine transformation network architecture in the Transformer network architecture, the comprehensive truncated mean feature is subjected to feature conversion processing to obtain a first integrated feature; Using the classifier in the Transformer network architecture to perform type identification on the first integrated feature, and obtain unit adaptation coefficients corresponding to multiple variable conversion units; The variable conversion unit corresponding to the largest unit adaptation coefficient is used as the first variable conversion unit corresponding to the optical signal feature; the plurality of variable conversion units are respectively configured with feature integration standards of different scales for integrating multi-modal integration features; In the first variable conversion unit, a variable conversion operation is performed on the optical signal feature according to the physiological indicator feature to obtain a first variable conversion feature; Performing feature integration processing on the physiological indicator feature and the first variable conversion feature in the same feature domain to obtain a second integrated feature, and determining a second variable conversion unit corresponding to the physiological indicator feature from among the multiple variable conversion units according to the second integrated feature; In the second variable conversion unit, a variable conversion operation is performed on the physiological indicator feature according to the first variable conversion feature to obtain a second variable conversion feature; Acquire a category label, fuse the first variable conversion feature and the second variable conversion feature to obtain a fourth fused feature, configure the category label on an initial element of the fourth fused feature to obtain a variable conversion feature; The variable conversion feature is subjected to focus enhancement processing by using a gated recurrent network architecture and a focus enhancement network architecture to obtain a feature weighted feature, and the feature weighted feature is subjected to feature conversion processing to obtain a blood glucose monitoring result corresponding to the target user.

3. The method according to claim 2, characterized in that The first variable conversion unit includes a feature weighting unit; In the first variable conversion unit, performing a variable conversion operation on the optical signal feature according to the physiological indicator feature to obtain a first variable conversion feature includes: Fusing the optical signal feature and the physiological index feature to obtain a first fusion feature; According to the mapping relationship strength vector corresponding to the feature weighting unit, the first fused feature is converted into a first index vector, a first identifier vector and a first attribute vector; Constructing a first undetermined vector according to a cross product between the first index vector and the optical signal feature, and a cross product between the first index vector and the physiological indicator feature; Constructing a second undetermined vector according to a mapping vector of a cross product between the first identifier vector and the optical signal feature, and a mapping vector of a cross product between the first identifier vector and the physiological indicator feature; constructing a third undetermined vector according to the cross product between the first attribute vector and the optical signal feature, and the cross product between the first attribute vector and the physiological indicator feature; Taking the cross product between the first undetermined vector and the second undetermined vector as a fourth undetermined vector, and obtaining the number of elements of the optical signal feature; Normalizing the ratio between the fourth undetermined vector and the square root of the number of elements to obtain a self-feature importance coefficient vector, and taking the cross product between the self-feature importance coefficient vector and the third undetermined vector as the first focal enhancement feature; A preset first number of features among the first focus enhancement features are used as first variable conversion features.

4. The method according to claim 2, characterized in that: The first variable conversion unit includes a feature weighting unit; In the first variable conversion unit, performing a variable conversion operation on the optical signal feature according to the physiological indicator feature to obtain a first variable conversion feature includes: Using the feature weighting unit, converting the optical signal feature into a second index vector, and converting the physiological indicator feature into a second identifier vector and a second attribute vector; Performing data encoding on the cross product between the second index vector and the optical signal feature, and the mapping vector of the cross product between the second identifier vector and the physiological indicator feature, to obtain a fifth vector to be determined, and acquiring the number of elements of the optical signal feature; Standardizing the ratio between the fifth undetermined vector and the square root of the number of elements to obtain a multidimensional feature importance coefficient vector, and performing data encoding on the multidimensional feature importance coefficient vector and the cross product between the second attribute vector and the physiological indicator feature to obtain a multidimensional gating feature; A preset first number of features in the multi-dimensional gating features are used as first variable conversion features.

5. The method according to claim 2, characterized in that: The first variable conversion unit includes a bottleneck feature weighting unit; In the first variable conversion unit, performing a variable conversion operation on the optical signal feature according to the physiological indicator feature to obtain a first variable conversion feature includes: Acquire correlation features, fuse the correlation features with the physiological indicator features to obtain second fusion features; the correlation features are used to establish correlation relationships between features of different dimensions; Using the first gated recurrent network architecture in the bottleneck feature weighting unit, the second fused feature is subjected to focus enhancement processing to obtain a second focus enhancement feature, and a preset number of first features in the second focus enhancement feature are used as candidate variable conversion features; The optical signal feature and the candidate variable conversion feature are fused to obtain a third fused feature, and the third fused feature is subjected to focus enhancement processing by using the second gated recurrent network architecture in the bottleneck feature weighting unit to obtain a third focus enhancement feature; A preset first number of features in the third focus enhancement feature are used as first variable conversion features.

6. The method according to claim 2, characterized in that The first variable conversion unit includes a physiological key indicator conversion unit; In the first variable conversion unit, performing a variable conversion operation on the optical signal feature according to the physiological indicator feature to obtain a first variable conversion feature includes: Acquire an optical signal standardized feature corresponding to the optical signal feature, and a physiological index standardized feature corresponding to the physiological index feature; Using the first quantum neural network in the key physiological indicator conversion unit, the optical signal characteristics are quantum entangled and interacted to obtain first quantum entanglement and interaction characteristics; Using the second quantum neural network in the key physiological index conversion unit, the characteristics of the physiological index are quantum entangled and interacted to obtain second quantum entanglement and interaction characteristics; The second quantum entanglement and interaction feature is subjected to truncated mean processing to obtain a third truncated mean feature, and the first quantum entanglement and interaction feature and the third truncated mean feature are superimposed to obtain a first variable conversion feature.

7. The method according to claim 2, characterized in that The first variable conversion unit includes a key dimension retention unit; In the first variable conversion unit, performing a variable conversion operation on the optical signal feature according to the physiological indicator feature to obtain a first variable conversion feature includes: The physiological indicator feature is deleted, and the optical signal feature is used as the first variable conversion feature output by the key dimension retention unit.

8. The method according to claim 1, characterized in that After determining the blood glucose monitoring result corresponding to the target user according to the oxidase reaction spectrum and the physiological indicator characteristics of the target user, the method further includes: Acquire the blood sugar monitoring result, and extract a first blood sugar semantic representation from the blood sugar monitoring result; Processing the first blood glucose semantic representation using the blood glucose knowledge graph to obtain a second blood glucose semantic representation; and Based on the second blood glucose semantic representation, obtaining a first target blood glucose health state; Based on the reliability of the user data corresponding to the target user, obtaining a matching blood sugar health state related to the first target blood sugar health state from a preset database; and A blood glucose management plan corresponding to the matching blood glucose health status is searched from the preset database.

9. The method according to claim 8, characterized in that The preset database includes a plurality of blood sugar health status relationship pairs, wherein the blood sugar health status relationship pairs include associations between blood sugar health statuses and the blood sugar health statuses, and obtaining a matching blood sugar health status related to the first target blood sugar health status from the preset database based on the reliability of user data corresponding to the target user includes: On the basis that the reliability of the user data is completely reliable, obtaining a first blood sugar health state relationship pair including the first target blood sugar health state from the plurality of blood sugar health state relationship pairs; and Based on the first blood sugar health state relationship pair, obtaining a matching blood sugar health state related to the first target blood sugar health state; The status details of the blood sugar health status include the blood sugar health status category and the intervention level; The step of obtaining a matching blood sugar health state related to the first target blood sugar health state from a preset database based on the reliability of the user data corresponding to the target user further includes: On the basis that the reliability of the user data is not completely reliable, based on the interference level of the blood sugar health status included in each of the first blood sugar health status relationship pairs, obtaining the interference level of each of the first blood sugar health status relationship pairs; Based on the status details of the target user, obtaining a target interferability level corresponding to the target user; Based on the target intervenability level and the intervenability levels of the first blood sugar health status relationship pairs, obtaining a third blood sugar health status relationship pair from the first blood sugar health status relationship pair; and Based on the third blood sugar health state relationship pair, a matching blood sugar health state related to the first target blood sugar health state is obtained.

10. A server system, characterized in that: The method comprises a server, wherein the server is used to execute the method described in any one of claims 1 to 9.

Images