A tongue diagnosis system for screening diabetes

Abstract

The application relates to the technical field of medical diagnosis, in particular to a tongue diagnosis system for screening diabetes, which comprises the following steps: a tongue sensing module collects tongue physiological data, oral environment parameters and user basic health information; a dynamic tongue twin body construction module constructs a digital twin body, simulates tongue physiological state changes through a microcirculation evolution algorithm, and obtains simulation data; an intelligent analysis hub identifies diabetes-related tongue abnormal features and evaluates a risk level based on the simulation data and in combination with an improved Transform-CNN model; a visual interaction unit constructs an interactive virtual tongue image scene, real-time renders tongue morphology, metabolite distribution and microcirculation state, and generates a screening report and intervention suggestions; and an adaptive screening regulation unit predicts a diabetes trend, dynamically adjusts screening frequency and intervention schemes according to the diabetes risk level and the diabetes screening report. Thus, the problems of low precision and poor adaptability in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of medical diagnostic technology, specifically to a tongue diagnosis system for screening diabetes. Background Technology

[0002] Diabetes, as a prevalent chronic metabolic disease, requires early and accurate screening for effective intervention and to reduce the risk of complications. This directly impacts public health and the rational allocation of social medical resources. Currently, mainstream clinical diabetes screening methods primarily involve blood glucose and glycated hemoglobin (HbA1c) testing. These methods require invasive blood sample collection, are relatively cumbersome, and have high costs, making them unsuitable for efficient initial screening in scenarios such as community health checkups and large-scale population screening. Tongue diagnosis, a characteristic diagnostic method in Traditional Chinese Medicine, uses the theory of judging the body's health status by identifying tongue color, coating, and shape. Modern research has confirmed a link between this theory and the pathogenesis of diabetes—diabetic patients commonly exhibit specific tongue characteristics such as a reddish tongue, thick and greasy coating, and a swollen tongue with teeth marks, providing a feasible pathway for non-invasive diabetes screening. However, traditional tongue diagnosis relies entirely on the subjective experience of physicians, resulting in poor consistency in diagnosis among different physicians and an inability to quantify tongue features. Existing tongue diagnosis technologies are mostly limited to basic image acquisition and simple feature extraction, without constructing a dedicated analysis system for the pathological features of diabetes. This makes it difficult to adapt to the differences in tongue features among people of different ages and physical conditions. As health management upgrades towards intelligence and precision, higher requirements are placed on the standardization, quantification, and efficiency of tongue diagnosis screening. There is an urgent need for a tongue diagnosis screening system that can achieve accurate feature extraction and in-depth correlation analysis.

[0003] However, traditional and existing tongue diagnosis screening technologies have many inherent defects: diagnostic strategies are rigid; traditional methods rely solely on physician visual observation, and existing technologies often rely on fixed thresholds for feature judgment, failing to integrate multi-dimensional information such as the examinee's age, weight, and dietary habits, as well as interfering factors like ambient lighting and shooting angle. This makes it impossible to accurately match complex individual differences in tongue feature variations, resulting in low screening accuracy and a high rate of missed and misdiagnosed cases. Furthermore, the image acquisition process lacks standardized design; most devices do not uniformly control key parameters such as light intensity, color temperature, and shooting distance, and differences in acquisition conditions under different scenarios can easily lead to image distortion. Quality fluctuations lead to increased feature extraction errors; the analysis model is crude, with existing technologies mostly using single features or simple feature combinations for judgment, failing to deeply explore the correlation between tongue features and the course of diabetes and blood glucose levels, resulting in insufficient targeting of feature analysis, which not only leads to low screening efficiency but also makes it easy to cause judgment bias due to interference from invalid features; moreover, there is a lack of a full-process quality control mechanism, resulting in a lag in response to problems such as abnormal collection and feature extraction errors, and a failure in a single link can lead to screening interruption. Overall, it faces multiple technical challenges such as serious subjectivity, low accuracy, and poor adaptability, making it difficult to meet the stringent requirements of large-scale clinical screening. Summary of the Invention

[0004] This application provides a tongue diagnosis system for screening diabetes, in order to solve the problems of low accuracy and poor adaptability in the prior art.

[0005] The first aspect of this application provides a tongue diagnosis system for screening diabetes, comprising: a tongue sensing module, a tongue twin construction module, an intelligent analysis center, a visualization interaction unit, and an adaptive screening control unit; wherein, the tongue sensing module is used to collect tongue physiological data, oral environment parameters, and basic user health information; the tongue twin construction module is used to construct a digital twin of the tongue, and based on the tongue's biological attributes and metabolic characteristics, uses a microcirculation evolution algorithm to simulate changes in the tongue's physiological state to obtain tongue simulation data; the intelligent analysis center is used to identify diabetes-related tongue abnormalities and assess the diabetes risk level by using the tongue simulation data and an improved Transformer-CNN model; the visualization interaction unit is used to construct an interactive virtual tongue scene, render tongue morphological features, metabolite distribution, and microcirculation status in real time, and generate a diabetes screening report and intervention suggestions; the adaptive screening control unit is used to predict the diabetes progression trend and dynamically adjust the screening frequency and intervention plan based on the diabetes risk level and the diabetes screening report.

[0006] Preferably, the tongue sensing module includes a hyperspectral tongue imaging unit, a tongue coating thickness detection unit, a tongue microcirculation sensor, and an oral environment monitoring group. The hyperspectral tongue imaging unit is used to capture tongue spectral features and tissue texture data; the tongue coating thickness detection unit acquires tongue coating distribution density and thickness information; the tongue microcirculation sensor collects capillary blood flow velocity and blood oxygen saturation; and the oral environment monitoring group collects oral humidity, pH value, and volatile metabolite concentration.

[0007] Preferably, the tongue twin construction module includes a tongue twin construction unit and a microcirculation evolution simulation unit. The tongue twin construction unit is used to construct a three-dimensional digital twin containing tongue morphology, mucosal texture, tongue coating distribution, and capillary network based on hyperspectral tongue image data and laser scanning point cloud. The microcirculation evolution simulation unit is used to simulate the dynamic changes of blood perfusion, tissue metabolism, and tongue coating evolution of the tongue under blood glucose fluctuations using a microcirculation evolution algorithm based on the biological attributes and metabolic characteristics of the tongue, and output multi-physics field simulation data of the tongue.

[0008] Preferably, the intelligent analysis center includes an abnormal feature recognition unit and a risk level assessment module. The abnormal feature recognition unit is used to identify diabetes-related tongue abnormalities such as tongue color, microcirculation stagnation, and abnormal metabolite concentration by using an improved Transformer-CNN model, combined with tongue physiological data, oral environment parameters, and clinical blood glucose correlation data. The risk level assessment module is used to quantitatively assess the screening risk of low, medium, and high diabetes based on the type, intensity, and distribution pattern of the abnormal features, combined with the user's age, body mass index, family medical history, and other basic health information.

[0009] Preferably, the visualization interaction unit includes a tongue image rendering engine, a screening report generator, and a multimodal interaction module. The tongue image rendering engine is used to render the three-dimensional morphology, spectral feature distribution, metabolite concentration cloud map, and microcirculation hotspot map of the tongue in real time, and simultaneously mark abnormal areas. The screening report generator is used to generate a diabetes screening report containing details of tongue abnormalities, risk quantification results, dietary adjustment plans, exercise suggestions, and medical advice based on risk level and abnormal feature analysis. The multimodal interaction module provides voice commands, gesture control, and virtual dissection functions, allowing for rotation, zooming, and focused viewing of abnormal areas in the virtual tongue scene.

[0010] Preferably, the adaptive screening and control unit includes a progression trend prediction module and a screening strategy adjustment unit. The progression trend prediction module is used to predict the incidence and progression rate of diabetes in high-risk users based on log-normal distribution and tongue physiological state degeneration data. The screening strategy adjustment unit is used to dynamically adjust the screening frequency and intervention recommendations based on the prediction results and user lifestyle data.

[0011] A second aspect of this application provides a tongue diagnosis method for screening diabetes, comprising: acquiring tongue physiological data, oral environment parameters, and basic user health information; processing the tongue physiological data, oral environment parameters, and basic health information to generate a tongue health dataset; constructing a three-dimensional digital twin of the tongue based on the tongue health dataset using hyperspectral tongue image data and laser scanning point cloud; simulating the dynamic changes of tongue blood flow, metabolism, and tongue coating using a microcirculation evolution algorithm to obtain tongue simulation data; and identifying diabetes-related tongue abnormalities by comparing the deviation between measured data and virtual simulation results and combining an improved Transformer-CNN model. The system quantifies and assesses the risk level of diabetes, obtaining risk assessment results and abnormal feature data. Based on the risk assessment results and abnormal feature data, combined with historical screening data of the tongue twin, an interactive virtual tongue scene is constructed, rendering tongue morphological features, metabolite distribution, and microcirculation status in real time, generating a diabetes screening report and personalized intervention suggestions. Based on the diabetes risk level and the diabetes screening report, adaptive screening regulation is performed, real-time correction of the biological attribute parameters of the tongue digital twin is made, and the progression trend of diabetes is predicted by combining log-normal distribution. Based on the prediction results and user lifestyle data, the screening frequency and intervention suggestion details are dynamically adjusted.

[0012] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a tongue diagnosis method for screening diabetes as described in the above embodiments.

[0013] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a tongue diagnosis method for screening diabetes as described in the above embodiments.

[0014] A fifth aspect of this application provides a computer program product, including a computer program or instructions, for implementing a tongue diagnosis method for screening diabetes as described in the above embodiments.

[0015] Therefore, this application has the following beneficial effects:

[0016] This application's embodiments utilize a tongue sensing module to accurately collect multidimensional data on tongue spectral texture, tongue coating thickness, blood flow and oxygenation, and oral temperature and humidity, providing comprehensive and reliable data support for screening. The tongue twin construction module, based on laser point cloud and microcirculation evolution algorithms, constructs a three-dimensional digital twin encompassing tongue morphology and capillary networks, dynamically simulating tongue metabolism and microcirculation changes under blood glucose fluctuations, generating realistic simulation data. The intelligent analysis center employs an improved Transformer-CNN model, integrating clinically relevant data to accurately identify features such as abnormal tongue color and abnormal metabolite concentrations, and quantitatively classifying low, medium, and high risk based on the user's basic health information, significantly improving identification accuracy. The visualization and interaction unit intuitively presents the tongue's state through 3D rendering and heatmap marking, generating detailed reports with dietary and exercise recommendations, and supporting multimodal interactive viewing. The adaptive control unit predicts disease trends based on a log-normal distribution, dynamically adjusting screening frequency and intervention plans to facilitate early screening and intervention, improving diabetes management efficiency. This solves the problems of low accuracy and poor adaptability in existing technologies.

[0017] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0019] Figure 1 This is a schematic diagram of a tongue diagnosis system for screening diabetes according to an embodiment of this application;

[0020] Figure 2 This is a schematic diagram of a tongue sensing module according to an embodiment of this application;

[0021] Figure 3 This is a schematic diagram of a tongue twin construction module provided according to an embodiment of this application;

[0022] Figure 4 This is a schematic diagram of an intelligent analysis hub provided according to an embodiment of this application;

[0023] Figure 5 This is a schematic diagram of a visual interactive unit provided according to an embodiment of this application;

[0024] Figure 6 This is a schematic diagram of an adaptive screening control unit provided according to an embodiment of this application;

[0025] Figure 7 This is a schematic diagram of a tongue diagnosis system for screening diabetes according to an embodiment of this application;

[0026] Figure 8 This is a flowchart of a tongue diagnosis method for screening diabetes according to an embodiment of this application;

[0027] Figure 9 This is a schematic diagram of a tongue diagnosis method for screening diabetes according to an embodiment of this application;

[0028] Figure 10 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] The following describes an embodiment of a tongue diagnosis system for screening diabetes, with reference to the accompanying drawings. Addressing the poor adaptability issue mentioned in the background section, this application provides a tongue diagnosis system for screening diabetes. In this system, a tongue sensing module accurately collects multi-dimensional data on tongue spectral texture, tongue coating thickness, blood flow and oxygenation, and oral temperature and humidity, providing comprehensive and reliable data support for screening. A tongue twin construction module, relying on laser point cloud and microcirculation evolution algorithms, constructs a three-dimensional digital twin encompassing tongue morphology and capillary network, dynamically simulating tongue metabolism and microcirculation changes under blood glucose fluctuations, generating realistic simulation data. The intelligent analysis center employs an improved Transformer-CNN model, integrating clinical correlation data to accurately identify features such as abnormal tongue color and abnormal metabolite concentrations, and quantitatively classifying low, medium, and high risk based on the user's basic health information, significantly improving identification accuracy. The visualization and interaction unit visually presents the tongue's condition through 3D rendering and heatmap marking, generating detailed reports with dietary and exercise recommendations, and supporting multimodal interactive viewing. The adaptive control unit predicts disease trends based on a log-normal distribution, dynamically adjusting screening frequency and intervention plans to facilitate early screening and intervention, thus improving the efficiency of diabetes management. This addresses the problems of low accuracy and poor adaptability in existing technologies.

[0031] Figure 1 This is a schematic diagram of a tongue diagnosis system for screening diabetes, provided as an embodiment of this application.

[0032] This application provides a tongue diagnosis system for screening diabetes, the system 10 comprising:

[0033] Tongue sensing module 100, tongue twin construction module 200, intelligent analysis center 300, visualization interaction unit 400, and adaptive screening and control unit 500.

[0034] The system comprises the following components: a tongue perception module 100 for collecting tongue physiological data, oral environment parameters, and basic user health information; a tongue twin construction module 200 for constructing a digital tongue twin, using a microcirculation evolution algorithm based on tongue biological attributes and metabolic characteristics to simulate changes in tongue physiological state and obtain tongue simulation data; an intelligent analysis center 300 for identifying diabetes-related tongue abnormalities and assessing diabetes risk levels using tongue simulation data combined with an improved Transformer-CNN model; a visualization interaction unit 400 for constructing an interactive virtual tongue scene, rendering tongue morphological features, metabolite distribution, and microcirculation status in real time, and generating diabetes screening reports and intervention recommendations; and an adaptive screening and control unit 500 for predicting diabetes progression trends and dynamically adjusting screening frequency and intervention plans based on diabetes risk levels and diabetes screening reports.

[0035] It is understood that in this embodiment, the tongue sensing module accurately collects multi-dimensional data on tongue spectral texture, tongue coating thickness, blood flow and oxygenation, and oral temperature and humidity, providing comprehensive and reliable data support for screening. The tongue twin construction module, relying on laser point cloud and microcirculation evolution algorithms, constructs a three-dimensional digital twin encompassing tongue morphology and capillary networks, dynamically simulating tongue metabolism and microcirculation changes under blood glucose fluctuations, generating realistic simulation data. The intelligent analysis center uses an improved Transformer-CNN model, integrating clinical correlation data to accurately identify features such as abnormal tongue color and abnormal metabolite concentrations, and quantitatively classifies low, medium, and high risk based on the user's basic health information, significantly improving identification accuracy. The visualization interaction unit intuitively presents the tongue's state through 3D rendering and heatmap marking, generating detailed reports with dietary and exercise recommendations, and supporting multimodal interactive viewing; the adaptive control unit predicts disease trends based on log-normal distribution, dynamically adjusting screening frequency and intervention plans to facilitate early screening and intervention, improving diabetes management efficiency. This solves the problems of low accuracy and poor adaptability in existing technologies.

[0036] In this embodiment of the application, the tongue sensing module 100 includes: Figure 2 As shown, the system includes a hyperspectral tongue imaging unit, a tongue coating thickness detection unit, a tongue microcirculation sensor, and an oral environment monitoring group.

[0037] Among them, the hyperspectral tongue imaging unit is used to capture the spectral characteristics and tissue texture data of the tongue; the tongue coating thickness detection unit obtains the distribution density and thickness information of the tongue coating; the tongue microcirculation sensor is used to collect the capillary blood flow velocity and blood oxygen saturation of the tongue; and the oral environment monitoring group is used to collect oral humidity, pH value and volatile metabolite concentration.

[0038] It is understood that the embodiments of this application use a hyperspectral tongue imaging unit to capture the spectral features and tissue texture data of the tongue in real time, providing a basis for the analysis of tongue tissue features; a tongue coating thickness detection unit to accurately obtain the distribution density and thickness information of the tongue coating, directly related to the physiological state of the tongue, ensuring the accuracy of tongue coating abnormality identification; a tongue microcirculation sensor to collect the capillary blood flow velocity and blood oxygen saturation of the tongue, reflecting the circulatory and metabolic status of the tongue, providing support for the judgment of microcirculation abnormalities; and an oral environment monitoring group to collect oral humidity, pH value and volatile metabolite concentration, analyze the impact of the oral environment on the state of the tongue, reduce monitoring errors caused by environmental interference, and provide comprehensive raw data for the construction, intelligent analysis and subsequent regulation of the tongue twin, improving the full range and reliability of monitoring.

[0039] For example, in the context of diabetes tongue diagnosis screening, the hyperspectral tongue imaging unit adopts a non-contact, non-invasive scanning method. Relying on multispectral detection covering the visible to near-infrared bands, it breaks through the limitations of traditional tongue image acquisition, which can only capture the visible morphology. It can not only accurately capture the differentiated spectral characteristics of different areas of the tongue (tip, middle, sides, and root)—such as the balanced spectral reflectance curve corresponding to the color of a healthy tongue, the characteristic spectral peak shift corresponding to the darker / redder tongue color caused by metabolic abnormalities in diabetic patients, and the specific spectral signals of biomarkers such as hemoglobin and glycated hemoglobin related to blood glucose metabolism, but also clearly restore the subtle details of the tongue tissue texture, including the texture density of the tongue mucosa, the distribution of the tongue papillae, the depth and direction of microcracks, and the hidden morphology of ecchymosis and petechiae. The collected raw data undergoes preprocessing steps such as spectral noise reduction, image registration, and feature extraction. On one hand, this provides high-fidelity foundational data for the 3D morphological modeling and spectral information mapping of the digital tongue twin, ensuring the twin accurately replicates the tongue's physiological characteristics. On the other hand, it provides core input to the improved Transformer-CNN model in the intelligent analysis center, helping the model accurately identify spectral abnormalities and texture variations related to diabetes by comparing massive amounts of clinical sample data, effectively avoiding the limitations of traditional tongue diagnosis which relies on subjective judgment. Simultaneously, its non-invasive and rapid acquisition advantages enhance the patient experience, complementing data from other units such as tongue coating thickness detection and microcirculation sensing, further strengthening the comprehensiveness of tongue feature acquisition and the accuracy of diabetes risk screening.

[0040] In this embodiment of the application, the tongue twin construction module 200 includes: as follows Figure 3 As shown, the tongue twin construction unit and the microcirculation evolution simulation unit are shown.

[0041] Among them, the tongue twin construction unit is used to construct a three-dimensional digital twin based on hyperspectral tongue image data and laser scanning point cloud, including tongue morphology, mucosal texture, tongue coating distribution and capillary network; the microcirculation evolution simulation unit is used to simulate the dynamic changes of blood perfusion, tissue metabolism and tongue coating evolution of the tongue under blood glucose fluctuations based on the biological attributes and metabolic characteristics of the tongue, using microcirculation evolution algorithm, and output multi-physics field simulation data of the tongue.

[0042] It is understood that, through the tongue twin construction unit, based on hyperspectral tongue image data and laser scanning point cloud, a three-dimensional digital twin containing tongue morphology, mucosal texture, tongue coating distribution, and capillary network is accurately constructed, providing a three-dimensional replication of the tongue's physiological structure and accurate physical model support for subsequent simulation analysis. The microcirculation evolution simulation unit, based on the tongue's biological attributes and metabolic characteristics, uses a microcirculation evolution algorithm to dynamically simulate the dynamic changes in blood perfusion intensity, tissue metabolic rate, and tongue coating evolution under blood glucose fluctuations, outputting multi-dimensional multi-physics field simulation data of the tongue. This breaks through the limitations of traditional static data acquisition, fully restoring the evolution process of the tongue's physiological state related to diabetes. It provides comprehensive data input combining static structure and dynamic evolution for the intelligent analysis center, and also provides a rich data foundation for scene rendering of the visualization interaction unit, improving the scientificity, dynamism, and accuracy of diabetes screening.

[0043] It should be noted that the construction of a three-dimensional digital twin of the tongue based on hyperspectral tongue image data and laser scanning point cloud data begins with the hyperspectral tongue image data as a foundation. This involves extracting physiological characteristic parameters of the tongue (such as the density, orientation, and distribution of microcracks in the mucosal texture; the distribution range, density level, and thickness differences of the tongue coating; and the distribution morphology, branching structure, and characteristic spectral correlation information of the capillary network) to form the initial physiological characteristic model of the digital twin. Simultaneously, high-precision three-dimensional scanning of the tongue is performed using laser scanning to acquire point cloud data containing the overall contour of the tongue (the curvature and contour curves of the tip, middle, sides, and root of the tongue), surface undulation details, and the spatial posture of the tongue, among other realistic geometric features. After denoising, filtering, and meshing, the data is transformed into a three-dimensional solid geometric model. Subsequently, the solid geometric model obtained from the scan is registered, compared, and fused with the initial hyperspectral physiological feature model. This corrects the differences in matching geometric morphology and physiological features caused by scanning angle deviation and natural tongue peristalsis. By integrating key information such as tongue geometric parameters, mucosal texture attributes, tongue coating distribution patterns, and capillary network structure, a three-dimensional digital twin that can accurately map the geometric morphology and physiological features of the tongue, mucosal texture, tongue coating distribution, and capillary network is finally constructed. This provides a high-precision digital carrier for microcirculation evolution simulation, diabetes-related abnormal feature analysis, visualization interactive rendering, and risk assessment.

[0044] Finite element analysis formula:

[0045]

[0046]

[0047]

[0048] in, for Tongue Location of capillary blood flow velocity, This represents the baseline blood flow velocity at that location. for The deviation of blood glucose concentration at that location from the normal threshold at that moment. The factor representing the effect of blood glucose is... for Blood oxygen saturation at that location at any given time. This is the normal blood oxygen saturation threshold. This refers to the blood oxygen regulation coefficient. for Tongue Location of tissue metabolic rate, This is the baseline value for the metabolic rate of normal tissues. This is the metabolic coupling coefficient. for The concentration of volatile metabolites at that location at that moment; for Tongue Location of tongue coating thickness This refers to the initial thickness of the tongue coating. This represents the tongue coating growth coefficient. This represents the attenuation coefficient of tongue coating shedding. For integration time; It is an integral infinitesimal element; for Tongue Location of tissue metabolic rate; e is the natural constant.

[0049] For example, in the diabetes screening process, the tongue biological attributes provided by the tongue twin construction unit are first integrated, including the microstructural parameters of the tongue mucosa (such as mucosal layer thickness, tongue papilla density, branching morphology and diameter distribution of capillary network), biomechanical attributes (such as tongue mucosal elastic modulus and vascular permeability), and metabolic characteristics (real-time blood glucose fluctuation curve, dynamic changes in blood oxygen saturation, time series data of volatile metabolite concentration, etc.) collected by the tongue sensing module. Subsequently, relying on the microcirculation evolution algorithm, the dynamic evolution of the tongue's physiological state was simulated in stages: when blood glucose concentration increased, the algorithm accurately calculated the nonlinear changes in capillary blood flow velocity in various regions of the tongue through the blood perfusion model (e.g., in the tip of the tongue, due to the high vascular density, the attenuation of blood flow velocity with increasing blood glucose was more significant); at the same time, the tissue metabolism model deeply coupled blood flow changes with metabolic activity, simulating the accumulation process of metabolites (such as ketone bodies and inflammatory factors) in areas with reduced metabolic rate, showing the diffusion pattern of metabolic disorder from local to global; while the tongue coating evolution model dynamically extrapolated the time-varying characteristics of tongue coating thickness and distribution density based on metabolic state (e.g., under continuous stimulation of hyperglycemia, the tongue coating in the middle region of the tongue accumulates keratin due to slowed metabolism, and the thickness increases in a stepwise manner). The final output of the tongue multiphysics simulation data not only includes multi-dimensional visualization results such as blood flow velocity distribution cloud maps, metabolic rate heat maps, and tongue coating thickness time series curves, but also quantifies the degradation rate of tongue physiological indicators under different blood glucose fluctuation scenarios. These data provide the intelligent analysis center with dynamic abnormal feature identification basis (such as accurately locating high-risk areas of microcirculation stagnation and abnormal accumulation of metabolites by comparing simulation data with clinical thresholds). On the other hand, they support the visualization interaction unit to construct an interactive virtual scene of tongue physiological evolution, allowing doctors to intuitively trace the dynamic development path of diabetes risk and allowing patients to understand the progression of their condition through temporal tongue changes, thereby significantly improving the foresight of diabetes screening and the accuracy of intervention plans.

[0050] In this embodiment of the application, the intelligent analysis center 300 includes: Figure 4 As shown, there are an anomaly feature identification unit and a risk level assessment module.

[0051] The abnormal feature recognition unit uses an improved Transformer-CNN model, combined with tongue physiological data, oral environment parameters, and clinical blood glucose correlation data, to identify abnormal tongue features related to diabetes, such as tongue color, microcirculation stagnation, and abnormal metabolite concentration. The risk level assessment module is used to quantitatively assess the screening risk of low, medium, and high diabetes based on the type, intensity, and distribution pattern of abnormal features, combined with the user's age, body mass index, family medical history, and other basic health information.

[0052] It is understood that the embodiments of this application, through the abnormal feature recognition unit, utilize an improved Transformer-CNN model, combined with tongue physiological data, oral environment parameters, and clinical blood glucose correlation data, to accurately identify diabetes-related tongue abnormalities such as abnormal tongue color, microcirculatory stagnation, and abnormal metabolite concentration, providing core evidence for risk assessment; the risk level assessment module, based on the type, intensity, and distribution pattern of abnormal features, combined with basic health information such as user age, body mass index, and family medical history, quantitatively assesses the screening risks of low, medium, and high diabetes, which not only breaks through the limitations of traditional tongue diagnosis relying on subjective experience, but also provides a quantifiable risk assessment standard for diabetes screening, effectively improving the accuracy of screening and clinical reference value.

[0053] It should be noted that the improved Transformer-CNN model formula is as follows:

[0054]

[0055]

[0056]

[0057]

[0058]

[0059] in, Local features of the tongue extracted by CNN; The kernel weight matrix; This refers to localized input data. This is the convolution bias term; For activation functions; For query matrix; The key matrix; It is a value matrix; The normalized activation function; The dimension of the key matrix; For the output of the attention mechanism; For layer normalization operation; Input as global features; Output features for the Transformer module; To integrate weighting coefficients Features after fusion; For activation functions; These are the weights for the fully connected layer; Bias for fully connected layers; Output the probability of anomaly detection; Multiply the query matrix by the transpose of the key matrix; is the square root of the dimension of the key matrix.

[0060] Risk level assessment formula:

[0061]

[0062]

[0063] in, For comprehensive risk scoring; For the first Weights of anomalous features; For the first Quantified values ​​of anomaly characteristics; For the first The weight of basic health information items; For the first Quantitative values ​​of basic health information; This represents the number of abnormal feature categories; Number of basic health information items; The probability corresponding to the risk level; Risk level category (low, medium, high); It is a natural constant; For the first Risk-based score mapping values; Number of risk level categories (3 categories); The normalized denominator; This represents the score mapping value for the k-th type of risk; It is a natural constant; For anomaly feature index; Used as a basic health information index.

[0064] The risk assessment module integrates abnormal tongue features related to diabetes (types of abnormal tongue color, such as dark / reddish, intensity, spectral deviation from normal tongue color, distribution patterns such as localized abnormalities at the tip or across the entire tongue surface; the extent of slowed capillary blood flow due to microcirculatory stasis, and the area of ​​stasis; the multiple by which abnormal metabolite concentrations exceed normal thresholds, and the types and quantities of abnormal metabolites) with the user's basic health information (different stages of age, degree of body mass index exceeding the standard, and positive or negative family history of diabetes). It first performs a multi-dimensional correlation analysis between these two types of information: if the tongue is predominantly reddish and there is extensive microcirculatory stasis (blood flow velocity significantly lower than the normal threshold), and the user is elderly (e.g., over 65 years old), has a severely excessive body mass index (e.g., ≥28), and has a positive family history of diabetes, they can be classified as having high-risk diabetes. Disease warning; if there is only localized ecchymosis on the tongue with mild abnormal metabolite concentration, the user is of moderate age (40-55 years old), has a normal body mass index and no family history, it is considered medium risk and requires close monitoring; if there is only a single mild abnormal feature (such as slightly high concentration of a certain type of metabolite) and the basic health information is risk-free, it is classified as low risk, and the risk level is differentiated - for high risk (such as multiple abnormalities and high risk basic information), an immediate specialist consultation and intervention strategy is generated (including dynamic blood glucose monitoring and personalized blood sugar lowering plan suggestions), for medium risk, a phased follow-up plan is formulated (regular tongue examination and diet and exercise adjustment plan), and for low risk, health management guidance is provided (daily tongue self-examination methods and annual diabetes screening reminders), and finally, a diabetes risk stratification plan that takes into account the accuracy of screening, the timeliness of intervention and the cost of health management is output.

[0065] In this embodiment of the application, the visual interaction unit 400 includes, as follows: Figure 5 As shown, the components include a tongue rendering engine, a screening report generator, and a multimodal interaction module.

[0066] The tongue rendering engine is used to render the three-dimensional morphology, spectral feature distribution, metabolite concentration cloud map, and microcirculation hotspot map of the tongue in real time, and simultaneously mark abnormal areas; the screening report generator is used to generate a diabetes screening report that includes details of tongue abnormalities, risk quantification results, dietary adjustment plan, exercise suggestions, and medical advice based on risk level and abnormal feature analysis; the multimodal interaction module is used to provide voice commands, gesture control, and virtual dissection functions, allowing users to rotate, zoom, and focus on abnormal areas in the virtual tongue scene.

[0067] It is understood that the embodiments of this application use a tongue rendering engine to render the three-dimensional morphology, spectral feature distribution, metabolite concentration cloud map, and microcirculation hotspot map of the tongue in real time and simultaneously mark abnormal areas, intuitively presenting the physiological and pathological state of the tongue; the screening report generator generates a diabetes screening report containing details of tongue abnormalities, risk quantification results, dietary adjustment plans, exercise suggestions, and medical advice based on risk level and abnormal feature analysis, transforming professional analysis into actionable health management guidance; the multimodal interaction module provides voice commands, gesture control, and virtual dissection functions, supports rotation, zooming, and focused viewing of abnormal areas in the virtual tongue scene, improves the convenience of interaction and the efficiency of detailed insight, provides accurate diagnostic assistance for medical personnel, and enhances the visualization, report practicality, and user-friendliness of diabetes screening.

[0068] It should be noted that the screening report generator integrates the diabetes screening risk level (low, medium, high), tongue abnormality feature analysis results (type of tongue color abnormality such as dark / reddish, intensity such as the magnitude of spectral deviation from the normal threshold, distribution pattern such as local / global abnormality; degree of slowed blood flow due to microcirculatory stasis, proportion of stasis area; multiple of abnormal metabolite concentration, number of abnormal types) and the user's basic health information (age, body mass index, family medical history). It first performs a multi-dimensional correlation analysis on these three types of information: If the risk level is high and there are multiple severe abnormal features (such as dark tongue color across the entire area + large-scale microcirculatory stasis + significantly elevated metabolite concentration), and the user is elderly, has a severely elevated BMI, and has a family history of diabetes, they can be identified as a high-risk group for diabetes; if the risk level is medium and there is only a single moderate abnormal feature (such as localized tongue ecchymosis + mild metabolite elevation), and the user is of moderate age, has a normal BMI, or no family history, they belong to the medium-risk monitoring group for diabetes; if the risk level is low and there are only slight abnormal features (such as slightly elevated concentration of a certain type of metabolite) or no abnormalities, Users with normal basic health information are classified as low-risk for diabetes. The system then differentiates health management needs based on risk level: For high-risk individuals, it generates detailed tongue abnormality information (including abnormal area location and quantitative data), clarifies risk quantification results (such as risk score and comparison of abnormal items), provides targeted dietary adjustment plans (such as low-sugar, low-fat diet ratios and a list of prohibited foods), offers personalized exercise suggestions (such as suitable exercise types and durations), and provides emergency medical attention reminders (recommended departments and examinations). For medium-risk individuals, it develops descriptions of key abnormal characteristics, risk quantification values, phased diet and exercise plans (such as weekly diet adjustment plans and gradual exercise arrangements), and regular check-up reminders (check-up cycle and self-check methods). For low-risk individuals, it provides minor abnormality notes (if any), risk quantification results, daily healthy diet and exercise guidance (such as balanced diet recommendations and daily activity requirements), and annual diabetes screening reminders. The final output is a diabetes screening report that is professional, targeted, and practical, providing accurate data for user health management and clinical reference for medical personnel.

[0069] In this embodiment, the adaptive screening control unit 500 includes, as follows: Figure 6 As shown, the progress trend prediction module and the screening strategy adjustment unit are included.

[0070] The progression trend prediction module is used to predict the probability of diabetes onset and progression rate of high-risk users based on log-normal distribution and tongue physiological state degeneration data; the screening strategy adjustment unit is used to dynamically adjust the screening frequency and intervention recommendations based on the prediction results and user lifestyle data.

[0071] It is understood that the progression trend prediction module in this application relies on a log-normal distribution model and tongue physiological state degeneration data to accurately quantify the probability of diabetes onset and the rate of disease progression in high-risk users; the screening strategy adjustment unit combines the prediction results with user lifestyle data to dynamically adjust the screening frequency and optimize intervention recommendations to avoid over-screening or under-intervention, thereby improving the accuracy of screening and intervention and user compliance.

[0072] It should be noted that the formula for the log-normal distribution is:

[0073]

[0074]

[0075] in, Let be the probability density function of the log-normal distribution; A random variable that follows a log-normal distribution; The location parameters of a log-normal distribution; is the scaling parameter of the log-normal distribution; is the base of the natural logarithm; Pi; for The natural logarithm; The cumulative distribution function of the log-normal distribution; The cumulative distribution function of the standard normal distribution; For integration variables; It is an integral infinitesimal element; It is negative infinity.

[0076] Based on the prediction results and user lifestyle data, the frequency of diabetes screening and intervention suggestions such as diet, exercise, and medical treatment are dynamically adjusted. The prediction results of incidence probability and progression rate generated by the progression trend prediction module are compared and analyzed in real time with the user's real-time collected lifestyle data (such as diet structure, exercise frequency, and work-rest patterns) to accurately identify the deviation between the user's actual health status and the diabetes risk control target. Based on the deviation analysis results, the strategy adjustment mechanism is automatically activated to adjust the screening frequency (shorten the screening interval to once per quarter for users with high progression rate and high sugar diet to achieve timely risk monitoring) and intervention suggestions (strengthen the aerobic exercise program three times a week for sedentary users, develop a phased low-salt diet transition plan for users with high salt diet, and give priority to the department to visit or regular self-examination reminders based on the incidence probability) to improve the adaptability to individual differences in user lifestyle, different disease progression rates, and diverse health management needs.

[0077] For example, Mr. Li, 45 years old, was included in the high-risk screening cohort due to a family history of diabetes and a BMI of 28.5 (overweight). The progression trend prediction module, combined with a log-normal distribution and 3-month data on tongue physiological deterioration, assessed that his probability of developing diabetes within 3 months was 60%, his tongue remained dark red with a 30% increase in the area of ​​microcirculatory stagnation at the tip, and the concentration of tongue metabolites associated with fasting blood glucose was 1.2 times higher than at the initial test, with a deterioration rate 40% faster than that of high-risk individuals of the same age. Simultaneously, the screening strategy adjustment unit collected data on Mr. Li's lifestyle habits: he consistently ate two high-oil, high-salt takeout meals daily, exercised only once a week (less than 30 minutes), and frequently stayed up late (averaging falling asleep after 1 AM). Based on the multi-dimensional analysis results of "high incidence rate + rapid progression rate + high-risk lifestyle habits", the adjustment unit dynamically upgraded Mr. Li's original screening frequency from once per quarter to once per month, and added dynamic monitoring of tongue microcirculation and retesting of metabolite concentration to each screening. In terms of intervention recommendations, a "three-dimensional adaptation plan" was generated: in terms of diet, "takeout alternative menus" were provided (such as replacing fried dough sticks and soy milk with multigrain porridge and boiled eggs for breakfast, and replacing fried chicken and rice with steamed fish and brown rice for lunch, with the differences in calories and carbohydrate content of each meal marked). The exercise plan was tailored to a "step-by-step adaptation plan" (15 minutes of post-meal walks daily in the first month, increasing to 20 minutes of brisk walking daily in the second month, and swimming three times a week in the third month, taking into account the patient's pre-existing knee injury; a self-examination guide for tongue appearance after exercise was also provided). Regarding sleep, a "guide to overcoming sleep deprivation" was given (going to bed 15 minutes earlier each night, combined with foot soaks before bed, screen time control, and other supplementary measures). A "risk warning list" was also added, clearly identifying six symptoms requiring immediate medical attention, such as a darkening tongue color, dry mouth, and frequent urination, and recommending an initial consultation with an endocrinologist, including tests such as fasting blood glucose and glycated hemoglobin. To ensure effective implementation, the adjustment unit also integrated with a health management app, sending weekly habit check-in reminders and fine-tuning the plan monthly based on check-in data and follow-up results.

[0078] This application proposes a tongue diagnosis system for screening diabetes. The system uses a tongue sensing module to accurately collect multidimensional data on tongue spectral texture, tongue coating thickness, blood flow and oxygenation, and oral temperature and humidity, providing comprehensive and reliable data support for screening. A tongue twin construction module, based on laser point cloud and microcirculation evolution algorithms, constructs a three-dimensional digital twin encompassing tongue morphology and capillary networks, dynamically simulating tongue metabolism and microcirculation changes under blood glucose fluctuations to generate realistic simulation data. The intelligent analysis center employs an improved Transformer-CNN model, integrating clinical correlation data to accurately identify features such as abnormal tongue color and abnormal metabolite concentrations. Combined with the user's basic health information, it quantitatively classifies low, medium, and high risk, significantly improving recognition accuracy. A visualization interaction unit intuitively presents the tongue's state through 3D rendering and heatmap marking, generating detailed reports with dietary and exercise recommendations, and supporting multimodal interactive viewing. An adaptive control unit predicts disease trends based on a log-normal distribution, dynamically adjusting screening frequency and intervention plans to facilitate early screening and intervention, improving diabetes management efficiency. This solves the problems of low accuracy and poor adaptability in existing technologies.

[0079] The following will illustrate a tongue diagnosis system for screening diabetes through a specific embodiment, such as... Figure 7 As shown, it includes:

[0080] In community health service centers' health screening areas, general practice clinics' chronic disease management rooms, and family health monitoring scenarios, high-risk groups for diabetes, such as middle-aged and elderly people over 40, those with a family history of diabetes, overweight and obese individuals, and those who are sedentary and lack exercise, have an increasingly urgent need for non-invasive, convenient, and accurate early screening. As a highly prevalent chronic disease, diabetes has insidious and nonspecific early symptoms. If it is not detected and intervened in time, it can easily lead to serious complications such as retinopathy and nephropathy. Traditional diabetes screening methods have many limitations. Currently, commonly used invasive blood glucose tests (such as fasting blood glucose and glycated hemoglobin tests) require blood sample collection, which not only causes pain for users but also has the problems of long testing cycles and high costs per test, resulting in low user compliance and difficulty in achieving long-term continuous monitoring. At the same time, traditional screening relies on doctors' subjective observation of tongue appearance and consultation, lacking standardized data collection and analysis processes. The judgment of key signs such as abnormal tongue color and changes in tongue coating varies from person to person, and the data records are fragmented and inconsistent, making it difficult to capture early weak pathological signals (such as slight stagnation in tongue microcirculation and subtle fluctuations in metabolite concentration), resulting in insufficient accuracy in risk assessment and inability to provide users with personalized intervention guidance and dynamic management plans. To overcome this predicament, a tongue diagnosis system for screening diabetes is introduced, integrating non-invasive sensing, digital twin, intelligent analysis and dynamic regulation technologies.

[0081] The tongue sensing module, as the core of the system's data acquisition, integrates four types of high-precision sensing units to ensure the comprehensiveness and accuracy of data acquisition. The hyperspectral tongue imaging unit uses a high-resolution imaging device in the 400-1000nm wavelength band, achieving a resolution of 1024×768 pixels. During acquisition, a built-in supplemental lighting system provides uniform white light illumination to avoid ambient light interference. The user needs to keep the tongue naturally extended for 3 seconds. The device automatically captures the spectral reflectance characteristics and mucosal texture details of the tongue surface, generating 16-bit deep hyperspectral image data to accurately identify abnormal spectral signals such as reddish or dark tongue color. The tongue coating thickness detection unit is based on the laser triangulation principle, with a laser wavelength set to 650nm and a measurement accuracy of 0.01mm. By scanning 200 evenly distributed sampling points on the tongue surface, it obtains the distribution density and thickness data of the tongue coating in the tip, middle, sides, and root regions of the tongue, automatically eliminating measurement errors caused by saliva reflection. The tongue microcirculation sensor employs a miniature laser Doppler flowmeter with a sampling frequency of 100Hz. The sensor probe is gently placed against the tongue mucosa to collect capillary blood flow velocity (mm / s) and blood oxygen saturation (%). Data is continuously collected for 10 seconds and averaged to ensure data stability. The oral environment monitoring group consists of a miniature humidity sensor, a pH electrode, and a gas sensor array. The humidity sensor measures 30%-95%RH, the pH electrode has a measurement accuracy of ±0.01pH, and the gas sensor array can detect the concentration (ppm) of 12 volatile metabolites related to diabetes, such as acetone and acetic acid. During data collection, the user must close their mouth for 3 minutes. The device uses a disposable sterile sampling head to extract oral gas and mucosal secretions, avoiding cross-infection. Simultaneously, the system guides the user to input basic health information such as age, gender, height, weight, family history of diabetes, and past medical history via a touchscreen. All collected data is converted from digital to digital and transmitted in a standardized format to the data preprocessing module for noise reduction and normalization, providing high-quality data support for subsequent modeling and analysis.

[0082] The tongue twin construction module builds a high-fidelity digital twin of the tongue based on collected multi-source data, enabling dynamic simulation of the tongue's physiological state. The tongue twin construction unit first integrates hyperspectral tongue image data with laser scanning point clouds, and then uses a 3D reconstruction algorithm to generate a 3D model that includes tongue morphology, mucosal texture, tongue coating distribution, and capillary network. The model's mesh density reaches 100 vertices per cubic millimeter, accurately reproducing the physiological structural details of the tongue. During model construction, the model's tissue optical properties and hemodynamic parameters are calibrated based on the user's age, gender, and other biological attributes, as well as physiological indicators such as metabolite concentration and blood oxygen saturation, ensuring consistency between the twin and the user's actual tongue state. The microcirculation evolution simulation unit employs an improved lattice Boltzmann algorithm as its core algorithm. Based on the correlation between blood glucose fluctuations and tongue microcirculation, it sets blood perfusion parameters (such as vascular permeability and blood viscosity) at different blood glucose concentrations to simulate changes in blood flow velocity, blood oxygen exchange, and tissue metabolic dynamics in the tongue under different physiological states, such as fasting and 2 hours postprandial. Simultaneously, it combines volatile metabolite concentration data to simulate the evolutionary trends of tongue coating thickness and color. For example, when blood glucose concentration increases, the model can simulate pathological changes such as microcirculatory stasis in the tongue (blood flow velocity decreases by 15%-30%), local accumulation of metabolites on the tongue surface (concentration increases by 2-3 times), and thickening and uneven distribution of the tongue coating. The final output is multi-physics simulation data of the tongue, encompassing hemodynamics, tissue metabolism, and tongue coating evolution, providing dynamic data support for subsequent abnormal feature identification.

[0083] The intelligent analysis center is the core processing unit of the system. It uses an improved Transformer-CNN model to accurately identify diabetes-related abnormal features and quantify risk levels. The abnormal feature identification unit employs an improved Transformer-CNN model that combines the local feature extraction advantages of CNNs with the global correlation capture capabilities of Transformers: First, it extracts features from local regions (such as the tip and sides of the tongue) of the 3D tongue model using three convolutional layers (kernel sizes of 3×3, 5×5, and 3×3), obtaining local features such as tongue color spectral deviations, abnormal tongue coating thickness, and microcirculation stagnation areas; then, it models the global data using a Transformer encoder (containing 6 attention heads and a hidden layer dimension of 512). The study captures the correlation between abnormal features in different regions of the tongue (such as the synergy between tongue stasis and increased metabolite concentration at the tongue tip). Finally, local and global features are weighted and fused together. A labeled dataset is constructed by combining tongue data from 10,000 diabetic patients and healthy individuals collected clinically. The Softmax classifier is used to identify three core abnormal features: abnormal tongue color (reddish, dark, purplish), microcirculation stasis (mild, moderate, severe), and abnormal metabolite concentration (acetone, acetic acid, etc. exceeding the standard multiple). The intensity value (0-10 points, the higher the score, the more significant the abnormality) and distribution coordinates of each abnormal feature are output. The risk level assessment module performs quantitative calculations based on identified abnormal features and the user's basic health information. First, weights are assigned to different types of abnormal features (e.g., severe microcirculation stagnation weight 0.35, significantly excessive metabolite concentration weight 0.3), and the total score of abnormal features is calculated based on their intensity values. Then, basic health information such as age (0.5 points for every 10 years over 40), BMI (1 point for ≥28, 0.5 points for 24-27.9), and family history of diabetes (1 point for positive cases) are converted into risk factor scores. Finally, the total score is calculated using the comprehensive risk scoring formula (S = total score of abnormal features × 0.7 + basic health information score × 0.3), and the risk level is determined based on the score range: S < 3 is low risk, 3 ≤ S < 6 is medium risk, and S ≥ 6 is high risk, thus achieving quantitative assessment of risk level.

[0084] The visualization and interaction unit provides users and medical personnel with an intuitive and convenient interface, enabling the visualization of screening results and personalized intervention guidance. The system is equipped with a 15.6-inch high-definition touchscreen as the interactive terminal. The tongue rendering engine, based on a digital twin model of the tongue, renders in real-time the tongue's three-dimensional morphology, pseudo-color images of spectral feature distribution, metabolite concentration cloud maps, and microcirculation hotspot maps (high stagnation areas are marked in red). Users can rotate and zoom the virtual tongue image 360° (up to 5x magnification) via touch, clearly viewing the location and details of abnormal areas. The multimodal interaction module supports voice commands (such as "show metabolite distribution" and "zoom in on the tongue side area"), gesture controls (slide to switch views, click to view details), and virtual cross-sectioning. Virtual cross-sections can be performed along the coronal and sagittal planes of the tongue, allowing for intuitive observation of the distribution and stagnation status of the internal capillary network. The screening report generator automatically generates a structured diabetes screening report based on the assessment results of the intelligent analysis center. The report includes basic user information, details of tongue abnormalities (including quantitative data of each abnormal indicator and comparison with normal ranges), risk level assessment results, risk quantification score, dietary adjustment plan (such as recommended daily carbohydrate intake for low-risk users, and a list of high-sugar foods to avoid for medium- and high-risk users), exercise recommendations (recommended exercise types, durations, and frequencies based on age and physical condition), and medical advice (recommended departments for high-risk users, and required examinations such as fasting blood glucose and glycated hemoglobin testing). The report supports electronic export and printing, and can also be synchronized to the user's health management APP via Bluetooth for convenient follow-up viewing.

[0085] The adaptive screening and control unit, based on screening results and user dynamic data, enables prospective management of diabetes risk and personalized adjustment of screening strategies. The progression trend prediction module adopts a log-normal distribution model, taking the user's current tongue physiological state degradation data (such as the rate of microcirculation blood flow velocity decay, the rate of metabolite concentration increase, and the rate of tongue coating thickening in the past 3 months) as input. The logarithmic mean μ and logarithmic standard deviation σ are obtained by fitting the maximum likelihood estimation method. The cumulative distribution function F(x;μ,σ) is used to calculate the probability of diabetes onset in the next 3 months, 6 months, and 12 months, and the progression rate is quantified by the σ value (σ>0.8 indicates rapid progression, 0.4≤σ≤0.8 indicates moderate progression, and σ<0.4 indicates slow progression). The screening strategy adjustment unit guides users to regularly input information such as dietary structure (e.g., daily intake of sugary foods, types of staple foods), exercise (number of exercise sessions per week, duration of each session), and sleep patterns (daily sleep time, frequency of staying up late) through the system's built-in lifestyle data collection module. Based on the predicted progression, the screening frequency is dynamically adjusted: for low-risk users with slow progression, screening is set once a year; for medium-risk users or those with moderate progression, it is adjusted to once every three months; and for high-risk users or those with rapid progression, screening is set once a month, with each screening including dynamic monitoring of microcirculation and retesting of metabolite concentrations. Simultaneously, intervention recommendations are optimized based on users' lifestyle data. For example, for high-risk users with high-sugar diets, the diet plan is updated to a "low-sugar alternative diet" (e.g., replacing white rice with whole grain rice, replacing sugary drinks with unsweetened yogurt), with the glycemic index of each food item labeled; for users lacking exercise, a "step-by-step exercise plan" is developed (gradually increasing from 15 minutes of brisk walking twice a week to 30 minutes of jogging four times a week); and for users with irregular sleep patterns, a "sleep improvement guide" is provided. Every three months, the system updates the predictive model parameters based on users' follow-up examination data and lifestyle check-in records, and readjusts the screening frequency and intervention plan to ensure the targeted and effective health management.

[0086] In summary, this application's embodiments collect non-invasive, multi-dimensional data through a tongue sensing module, providing accurate input. The construction of a tongue twin enables virtual-real synchronization, laying the foundation for analysis. The intelligent analysis center, leveraging an improved Transformer-CNN model, accurately identifies abnormal tongue features and quantifies risk levels, significantly reducing early risk omissions and misjudgments caused by subjective biases in traditional manual tongue diagnosis, thus mitigating the risk of delayed intervention for diabetes and its complications. The visualization and interaction unit enhances the efficiency of screening result interpretation and health management through intuitive rendering and convenient operation. Combined with a screening report generator, it outputs personalized dietary, exercise, and medical advice, reducing the blindness of health guidance. The adaptive control unit dynamically adjusts screening frequency and intervention plans based on disease trend prediction, balancing risk monitoring and screening costs, reducing over-screening or under-intervention, while avoiding the pain and data inconsistencies of traditional invasive testing. This achieves intelligent management of the entire diabetes screening process, improving user screening compliance, risk assessment accuracy, stable health management effects, and reducing overall health risks.

[0087] Next, referring to the accompanying drawings, a tongue diagnosis method for screening diabetes is described according to an embodiment of this application.

[0088] like Figure 8 As shown, this tongue diagnosis method for screening diabetes includes the following steps:

[0089] In step S101, tongue physiological data, oral environment parameters, and basic user health information are acquired.

[0090] It is understood that, by acquiring tongue physiological data, oral environment parameters, and user basic health information, the tongue twin can dynamically synchronize the user's tongue physiological state, oral environment influence, and basic health background, and capture potential pathological signals of the tongue, the influence of the oral environment on the tongue state, and basic factors of user health risk in real time. This provides high-quality data for subsequent accurate modeling of the tongue twin, intelligent abnormal feature identification and risk assessment, and adaptive screening strategy adjustment, enabling accurate perception and early intervention of tongue pathological state and user health risk, reducing the risk of missed or misjudged early diabetes risk and the occurrence of complications, and improving the accuracy of diabetes screening and the efficiency of health management.

[0091] In step S102, the physiological data of the tongue, oral environment parameters, and basic health information are processed to generate a tongue health dataset. Based on the tongue health dataset, a three-dimensional digital twin of the tongue is constructed by using hyperspectral tongue image data and laser scanning point cloud. The microcirculation evolution algorithm is used to simulate the dynamic changes of tongue blood flow, metabolism, and tongue coating to obtain tongue simulation data. By comparing the deviation between the measured data and the virtual simulation results, and combining the improved Transformer-CNN model, abnormal tongue features related to diabetes are identified, and the risk level of diabetes is quantitatively assessed to obtain risk assessment results and abnormal feature data.

[0092] Among them, the improved Transformer-CNN model is an intelligent analysis model that integrates the advantages of CNN in local feature extraction with the global correlation capture capability of Transformer and is used to accurately identify diabetes-related tongue abnormalities and quantify risk levels after structural optimization.

[0093] It is understood that the embodiments of this application combine the advantages of CNN local feature extraction with the global correlation capture capability of Transformer to process multi-source data such as hyperspectral tongue images and laser scanning point clouds in the tongue health dataset. By combining the deviation between measured data and virtual simulation results, it accurately identifies abnormal tongue features related to diabetes, such as abnormal tongue color, microcirculation stagnation, and excessive metabolite concentration. At the same time, it quantitatively assesses the risk level, reduces the subjective bias of traditional manual judgment and the missed or misjudged early weak pathological signals, and provides reliable data support for the formulation of subsequent personalized intervention suggestions and the adjustment of adaptive screening strategies, thereby improving the accuracy and intelligence level of diabetes screening.

[0094] It should be noted that the process of processing tongue physiological data, oral environment parameters, and basic health information to generate a tongue health dataset involves several steps. First, for the hyperspectral tongue image data in the tongue physiological data, Gaussian filtering is used to remove ambient light interference, and spectral reflectance and mucosal texture features in the 400-1000nm band are extracted. For the tongue coating thickness data, a threshold segmentation algorithm is used to exclude reflective points of saliva on the tongue, and the average thickness and distribution density of each region of the tongue tip, middle, sides, and root are calculated. For the tongue microcirculation data, fluctuations in the initial 1 second of data collection are removed, and the average blood flow velocity and blood oxygen saturation are taken over 9 consecutive seconds. For the oral environment parameters, the original readings are corrected by combining the calibration curves of the humidity sensor and pH electrode, and baseline correction is performed on the concentration data of 12 volatile metabolites detected by the gas sensor array to eliminate the influence of device zero drift. For the user's basic health information, missing items are filled through system logic verification (such as BMI value automatically calculated based on height and weight), and the completeness of key information such as family history of diabetes and past medical history is verified. Subsequently, data alignment and multi-source fusion were performed: using the timestamp of tongue perception acquisition as a benchmark, time alignment of tongue physiological data and oral environment parameters at the same screening time was achieved; using the three-dimensional coordinates of the tongue constructed by laser scanning point cloud as a reference, hyperspectral tongue image features, tongue coating thickness data, microcirculation indicators and specific areas of the tongue (such as the tip and sides of the tongue) were spatially aligned to form multi-dimensional data with "location-physiology-environment" association; at the same time, basic health information such as user age, BMI value, and family medical history were used as tags and associated with the multi-source perception data of the corresponding screening period. Next, standardization and key feature extraction were performed: Z-score standardization was used to unify the dimensions of various indicators (e.g., converting data from different units such as blood flow velocity (mm / s) and metabolite concentration (ppm) into standardized data with a mean of 0 and a standard deviation of 1). Based on clinical experience and data characteristics, core features such as tongue color spectral deviation, tongue coating thickness variation coefficient, microcirculation blood flow velocity attenuation rate, and acetone concentration exceeding the standard multiple were extracted. Simultaneously, basic health information was transformed into quantifiable risk factor data (e.g., encoding age over 40 years old with a score increasing by 0.5 points every 10 years). Finally, quality verification and dataset encapsulation were performed: outlier detection algorithms (e.g., the 3σ criterion) were used to remove abnormal data caused by poor user cooperation (e.g., blurred images due to tongue movement) and temporary equipment interference. Qualified data was organized and encapsulated according to the structure of "user unique identifier - screening time - data type - feature indicator - tag information," forming a complete tongue health dataset containing raw sensory data, preprocessed data, core feature data, and associated tags.

[0095] For example, Ms. Zhang, 48 years old, was included in the high-risk screening cohort due to a family history of diabetes and a BMI of 27.5. The tongue perception module collected data such as her hyperspectral tongue image (locally dark spectral signal on the tongue surface), tongue coating thickness (0.15mm at the root of the tongue), tongue microcirculation (capillary blood flow velocity 2.3mm / s), and oral volatile metabolites (acetone concentration 1.9ppm) to generate a tongue health dataset. After combining the deviation between the simulated data of "blood flow-metabolism-tongue coating" output by the tongue twin and the measured data, the improved Transformer-CNN model was launched for analysis: firstly, local abnormal features such as reddish spectral deviation at the tip of the tongue, thickened tongue coating at the root of the tongue, and local microcirculation stagnation at the edges of the tongue were extracted through three convolutional layers with different kernels. The Transformer encoder with six attention heads then captures the global correlation between "tongue stasis and excessive acetone concentration" and "thickened tongue coating and reduced blood flow velocity," ultimately accurately identifying three types of diabetes-related abnormalities: "mild tongue color abnormality + moderate microcirculation stasis + mild metabolite excess." The model outputs a medium-risk level and labels the distribution coordinates and intensity values ​​of each abnormality (stasis intensity 6.5 points). Compared to the limitations of traditional manual methods that can only make preliminary judgments about tongue appearance abnormalities, this model successfully captures early, weak signals of microcirculation stasis and metabolite synergistic abnormalities, avoiding the risk of missed judgments. This provides accurate abnormal feature basis for the subsequent visualization interactive unit to generate a "targeted microcirculation improvement whole grain diet + brisk walking exercise plan."

[0096] In step S103, based on the risk assessment results and abnormal feature data, combined with the historical screening data of the tongue twin, an interactive virtual tongue scene is constructed, which renders the tongue morphological features, metabolite distribution and microcirculation status in real time, and generates a diabetes screening report and personalized intervention suggestions.

[0097] Among them, the tongue twin is a three-dimensional digital model constructed based on hyperspectral tongue image data and laser scanning point cloud, which includes tongue morphology, mucosal texture, tongue coating distribution and capillary network. It can simulate the dynamic changes of tongue blood flow-metabolism-tongue coating through microcirculation evolution algorithm and output simulation data, providing support for the identification and risk assessment of abnormal tongue features related to diabetes.

[0098] It is understood that the embodiments of this application construct a three-dimensional digital model containing tongue morphology, mucosal texture, tongue coating distribution, and capillary network based on hyperspectral tongue image data and laser scanning point cloud. Combined with microcirculation evolution algorithms, it simulates the dynamic changes of tongue blood flow, metabolism, and tongue coating and outputs simulation data. This enables the interactive virtual tongue image scene to synchronize the real physiological state and historical evolution process of the tongue, presenting the tongue morphological characteristics, metabolite distribution, and microcirculation status in real time. It clearly captures the changes in early pathological signals of the tongue (such as the evolution of microcirculation stagnation and the trend of metabolite accumulation) and historical screening data, providing accurate data and model support for the subsequent construction of interactive virtual tongue image scenes, generation of diabetes screening reports, and formulation of personalized intervention suggestions. The dynamic presentation and historical tracing of the tongue state reduces the limitations of traditional static tongue image observation in interpreting pathological changes and improves the intuitiveness and interpretation efficiency of screening results. At the same time, it provides a basis for tracing the source of abnormal features and predicting the intervention effect, reducing the blindness of health guidance and improving the accuracy of diabetes screening and the pertinence of health management.

[0099] It should be noted that, based on the risk assessment results (including low / medium / high risk levels, quantitative scores, and judgment criteria), abnormal characteristic data (such as the type and distribution coordinates of abnormal tongue color, the degree and range of microcirculation stagnation, and the areas and multiples of metabolite concentration exceeding the standard), combined with multi-time-point historical screening data stored in the tongue twin (covering the three-dimensional tongue model of each screening, physiological parameter evolution curves, abnormal characteristic change trajectories, etc.), an interactive virtual tongue image scene is constructed. First, multi-source data integration and correlation calibration are carried out: using the three-dimensional digital model of the tongue twin as the basic framework, the current risk assessment quantitative score (e.g., medium risk 6.2 points) is used... The core abnormal features (such as moderate stasis at the tip of the tongue and accumulation of metabolites at the root of the tongue) are precisely bound to the spatial coordinates of the three-dimensional model, clarifying the specific location of the abnormal features in the tongue outline, mucosal texture, and capillary network. At the same time, historical screening data are sorted out according to the time axis, and key information such as morphological changes of the tongue model in each iteration (such as the tongue coating thickness increasing from 0.12mm to 0.15mm), evolution of abnormal areas (such as the stasis range expanding from the tip of the tongue to the sides of the tongue), and fluctuations in physiological parameters (such as blood flow velocity decreasing from 2.5mm / s to 2.1mm / s) are extracted and compared with the current data to form a "history-current" comparison and correlation, ensuring the continuity and accuracy of the scene data. Next, high-fidelity scene modeling and visualization rendering are performed: relying on the tongue image rendering engine, the tongue twin model is rendered in detail based on the bound data. This not only restores the basic features such as the natural shape of the tongue, details of the mucosal texture, and density of the tongue coating, but also presents core information through multi-dimensional visualization methods. The pseudo-color image is used to intuitively display the spectral distribution of tongue color (e.g., warm colors are used to mark reddish areas and neutral colors are used to distinguish normal areas). The concentration cloud map dynamically presents the aggregation state of metabolites (the darker the color, the higher the concentration). The heat map marks the degree of microcirculation stagnation (red marks severe stagnation and yellow marks mild stagnation). Numerical labels are superimposed on abnormal areas (e.g., "stagnation intensity 7.0 points" and "acetone concentration 2.3ppm"). At the same time, historical data is integrated to achieve "dynamic traceability" rendering. The tongue image status at different screening time points can be viewed by switching the time slider, clearly presenting the evolution process of abnormal features (e.g., clicking "3 months ago" can show the scene where only slight stagnation was present at the tip of the tongue at that time). Subsequently, multimodal interaction functions were embedded to enhance the usability of operation: rich interaction methods were designed to meet the different needs of medical personnel in diagnosis and users in viewing. At the basic operation level, touch / mouse is supported to realize 360° rotation of the virtual tongue image and 0-5x zoom, making it easy to focus on key areas such as the tip and sides of the tongue; at the professional interpretation level, a virtual dissection function is provided, which can dissect the model along the coronal and sagittal planes of the tongue to intuitively observe the stasis distribution of the internal capillary network and the infiltration trajectory of metabolites; at the intelligent interaction level, voice commands (such as "show the stasis changes in the past 6 months") and gesture control (such as swiping to switch between concentration cloud map / heat map views) are supported to quickly retrieve the required information.Finally, dynamic linkage between the scene and information is achieved: information pop-ups and data panels are set up in the scene. Clicking on any abnormal area will bring up related details (such as "Tongue tip stasis: current blood flow velocity 2.1mm / s, down 16% from last month, corresponding risk factor weight 0.35"). The data panel synchronizes risk assessment results, abnormal feature summaries, and historical trend curves in real time. It also supports linkage between the scene and screening report. Clicking "Generate Report" will convert the abnormal visualization information and historical comparison data in the scene into text and graphics content in the structured report. The interactive virtual tongue scene built through this series of operations completely breaks the limitations of traditional static tongue observation. It allows medical personnel to accurately locate the source of abnormality and trace the pathological evolution process, improving the efficiency of diagnosis and interpretation. It also allows users to intuitively understand their own tongue health status and risk level, enhancing their acceptance of personalized intervention suggestions, and providing an efficient visualization carrier for the practical application of diabetes screening results.

[0100] For example, in high-risk screening for diabetes, when using tongue twin technology on the tongue of Ms. Wang, who is 55 years old (with a family history of type 2 diabetes and a BMI of 28.3, and is overweight), the tongue sensing module first collects her hyperspectral tongue image, laser scanning point cloud, and physiological parameters (tongue microcirculation blood flow velocity 2.0 mm / s, oral acetone concentration 2.3 ppm) to construct a tongue health dataset containing 12,000 grid vertices. Based on biological attributes (age 55 years, metabolite concentration) and tissue parameters (mucosal tissue optical refractive index 1.37, tongue coating thermal conductivity 0.23 W / (m・K)), a tongue multiphysics coupling model is constructed. During the simulation, the blood flow field (changes in capillary blood flow velocity at the tip of the tongue), metabolic field (aggregation and distribution of acetone on the tongue surface), and tongue coating evolution field (dynamic changes in tongue coating thickness at the root of the tongue) under fasting and 2-hour postprandial blood glucose fluctuations were simultaneously simulated. The simulation results showed that the blood flow velocity at the tip of the tongue decreased by 25% after a meal, the concentration of acetone at the edge of the tongue reached 2.6 ppm, and the tongue coating at the root of the tongue thickened by 0.04 mm. The simulated blood flow stasis cloud map and metabolite concentration curve were compared with the physical layer sensing data: the 1.8 mm / s blood flow velocity captured by the tongue microcirculation sensor corresponds to the blood flow state in the stasis area at the edge of the tongue in the simulation; the 2.4 ppm acetone concentration measured by the gas sensor deviates from the simulated value by ≤4%; and the abnormal spectral area at the edge of the tongue detected by hyperspectral imaging deviates from the simulated stasis range by only 2.8%. By calibrating the model parameters in real time using a tongue twin, early pathological signals (local blood flow velocity below 1.9 mm / s) and the risk of metabolite accumulation caused by blood stasis on the tongue side can be accurately identified. This provides quantitative identification basis for the intelligent analysis center (such as marking the stasis intensity as 7.2 points), effectively avoiding the omission of early pathological signals due to static tongue observation, and realizing the three-dimensional dynamic presentation of the tongue's physiological state and accurate risk tracing.

[0101] In step S104, adaptive screening regulation is performed based on the diabetes risk level and diabetes screening report. The biological attribute parameters of the tongue digital twin are corrected in real time. The progression trend of diabetes is predicted by combining the log-normal distribution. The screening frequency and intervention recommendations are dynamically adjusted based on the prediction results and user lifestyle data.

[0102] Among them, the log-normal distribution is a continuous probability distribution in which the logarithm of a variable follows a normal distribution. In diabetes screening, it is used as a statistical model to predict the probability of developing diabetes and the rate of progression of diabetes in high-risk users based on data on the degeneration of the physiological state of the tongue.

[0103] It is understood that the embodiments of this application, by using data on the physiological degeneration of the tongue (such as the decline in microcirculation blood flow velocity, the rate of increase in metabolite concentration, etc.), combined with diabetes risk level and screening report, assist in real-time correction of the biological attribute parameters of the digital twin of the tongue, accurately predict the probability of diabetes onset and progression rate of high-risk users at different time periods in the future, provide quantitative basis for dynamically adjusting screening frequency and intervention recommendations, effectively avoid the lack of trend prediction and blind intervention caused by traditional static assessment, improve the foresight and personalization of diabetes risk management, and ensure the pertinence and effectiveness of health management.

[0104] For example, in the high-risk screening and management of diabetes, the log-normal distribution model uses real-time data on the physiological deterioration of the tongue (including the rate of decrease in microcirculatory blood flow velocity, the rate of increase in oral acetone concentration, and the rate of increase in tongue coating thickness) and basic health information stored in the tongue twins of high-risk individuals over the past three months to construct a risk progression trend prediction model. This model uses the deterioration data as input and fits the logarithmic mean μ and logarithmic standard deviation σ using the maximum likelihood estimation method to simulate the risk evolution under different lifestyle interventions. The model parameters are calibrated by comparing previous prediction results with actual follow-up data—for example, predicting two months in advance that Mr. Chen, 54 years old (with a family history of diabetes and a BMI of 28.5), has a 65% risk of developing diabetes with a "rapid" progression rate (σ=0.88>0.8). Simultaneously, suggestions for adjusting the screening strategy are pushed to the adaptive control unit, upgrading the original quarterly screening to monthly screening and adding metabolite retesting. The intervention recommendation is optimized to "low-sugar diet + swimming three times a week". This approach ensures that the risk assessment error is controlled within 5%, while also preventing early onset of disease due to lack of timely intervention, thereby increasing the intervention effectiveness rate for high-risk groups by more than 30%.

[0105] According to an embodiment of this application, a tongue diagnosis method for screening diabetes is proposed. This method uses a tongue perception module to accurately collect multidimensional data on tongue spectral texture, tongue coating thickness, blood flow and oxygenation, and oral temperature and humidity, providing comprehensive and reliable data support for screening. A tongue twin construction module, relying on laser point cloud and microcirculation evolution algorithms, constructs a three-dimensional digital twin encompassing tongue morphology and capillary networks, dynamically simulating tongue metabolism and microcirculation changes under blood glucose fluctuations, generating realistic simulation data. The intelligent analysis center employs an improved Transformer-CNN model, integrating clinical correlation data to accurately identify features such as abnormal tongue color and abnormal metabolite concentrations. Combined with the user's basic health information, it quantitatively classifies low, medium, and high risk, significantly improving identification accuracy. A visualization interaction unit intuitively presents the tongue's state through 3D rendering and heatmap marking, generating detailed reports with dietary and exercise recommendations, and supporting multimodal interactive viewing. An adaptive control unit predicts disease trends based on a log-normal distribution, dynamically adjusting screening frequency and intervention plans to facilitate early screening and intervention, improving the efficiency of diabetes management. This solves the problems of low accuracy and poor adaptability in existing technologies.

[0106] The following will illustrate a tongue diagnosis method for screening diabetes through a specific embodiment, such as... Figure 9 As shown, it includes:

[0107] Two hundred subjects were selected, including 58 males and 42 females, aged 45-72 years, with a disease duration of 1-15 years. Those with diabetic ketoacidosis, severe diabetic nephropathy (estimated glomerular filtration rate <30 ml / min / 1.73 m²), and foot lesions were excluded. Fifty individuals with prediabetes were identified through screening at a health check-up center and met the diagnostic criteria of fasting blood glucose 6.1-6.9 mmol / L, or 2-hour postprandial blood glucose 7.8-11.0 mmol / L, or glycated hemoglobin 6.0%-6.4%. These included 29 males and 21 females, aged 40-65 years. High-risk individuals with a family history of diabetes and BMI ≥30 kg / m² were excluded. Fifty healthy controls were community healthy individuals with normal blood glucose and glycated hemoglobin levels. These included 26 males and 24 females, aged 35-60 years. There were no statistically significant differences in gender and age composition between these controls and the first two groups (P>0.05). All participants signed informed consent forms approved by the ethics committee (approval number: 2024031501) and underwent special examinations to exclude those with organic lesions of the tongue (such as tongue cancer, tongue hemangioma), oral mucosal diseases (such as oral ulcers, lichen planus), and those who had taken medications that affect tongue appearance (such as iron supplements, bismuth preparations, antibiotics) within the past month. At the same time, pregnant and lactating women, patients with severe liver and kidney dysfunction, and patients with autoimmune diseases were excluded.

[0108] The data acquisition phase innovatively adopts a "multi-device collaboration + standardized process + triple verification of quality control" model to construct a comprehensive tongue health data acquisition system. The core acquisition equipment consists of two independently developed specialized devices: a hyperspectral tongue imaging system (model: HSI-T200) with full spectral coverage of 400-1000nm, including 320 spectral bands, a spatial resolution of 0.1mm, and an imaging speed of 15 frames / second, capable of simultaneously acquiring spectral information and texture features of the tongue surface; and a high-precision laser scanning device (model: LS-3D600) using blue laser scanning technology, with a scanning accuracy of ±0.02mm, a point cloud density of 1000 points / mm², and a single-frame scanning time of ≤2 seconds, enabling rapid reconstruction of the tongue's three-dimensional morphology. To eliminate environmental interference, a standardized collection room was specially constructed, with the room temperature controlled at 22-25℃ and relative humidity at 40%-60%. A standard white light source with a color temperature of 5500K (illuminance ≥800 lux) was used, and the distance between the light source and the tongue was fixed at 30cm, forming a 45° angle with the midline of the tongue. Subjects were required to fast and abstain from water for 8 hours in advance (a small amount of plain water was allowed), and were prohibited from smoking, chewing gum, and eating spicy foods for 30 minutes before collection. During collection, subjects assumed a standard sitting position, with their head fixed in a special hydraulic support (adjustable in height and angle), keeping the plane of their ears and eyes parallel to the ground, the tip of their tongue gently touching the inside of their lower front teeth, and the tongue naturally extended and kept still, avoiding tongue tremors or excessive force. The operation was performed by operators who had undergone 3 months of specialized training. Each subject underwent 3 imaging and scanning sessions, and the intermediate sample with the highest repeatability was taken as the valid data. Simultaneously, the system's built-in algorithm extracted tongue morphology parameters (length: distance from the tip of the tongue to the root of the tongue, accuracy 0.01 mm; width: distance at the widest point of the tongue; thickness: distance from the midpoint of the midline of the tongue dorsum to the ventral surface of the tongue), tongue coating characteristics (color: quantified by CIEL*a*b* color space; thickness: calculated based on spectral reflectance, unit μm; distribution: assessed by dividing the tongue into four regions: tip, middle, sides, and root of the tongue) and tongue surface microcirculation information (vascular density: number of blood vessels per unit area, unit roots / mm²; blood flow velocity: measured using Doppler effect, unit mm / s). Oral environment parameters were collected using an integrated multi-parameter sensor array (model: OMS-800). This device integrates a pH electrode, a thermocouple temperature sensor, a capacitive humidity sensor, and an enzyme electrode glucose sensor. The pH measurement range is 3.5-8.5 (accuracy ±0.01), temperature is 32-38℃ (accuracy ±0.1℃), humidity is 40%-80%RH (accuracy ±2%RH), and the detection limit for salivary glucose concentration is 0.1 mmol / L (accuracy ±0.05 mmol / L). Before collection, the sensor was disinfected by soaking in 75% medical alcohol for 10 minutes, then rinsed three times with physiological saline, dried, and placed 1 cm below the tongue near the frenulum. Data was recorded after the displayed value stabilized for 30 seconds. Measurements were repeated three times for each subject, and the average value was taken.User basic health information is acquired using a three-dimensional model of "electronic questionnaire + clinical testing + medical history verification": The electronic questionnaire contains 32 items across 5 dimensions, covering demographic characteristics (age, gender, occupation), anthropometry indicators (height, weight, waist circumference, hip circumference, BMI and waist-to-hip ratio calculation), family history (diabetes status in first-degree relatives), and lifestyle (daily staple food intake, vegetable and fruit intake, alcohol consumption frequency, smoking history, exercise type, and weekly exercise duration); Clinical testing indicators include fasting blood glucose (glucose oxidase method), 2-hour postprandial blood glucose (75g oral glucose tolerance test), glycated hemoglobin (HPLC method), four lipid profiles (total cholesterol, triglycerides, LDL cholesterol, HDL cholesterol), and liver and kidney function indicators; Medical history verification involves reviewing electronic medical records to confirm past medical history and medication history. All collected data is uploaded to Alibaba Cloud servers in real time, stored using AES-256 encryption algorithm, and a three-level access control system (researcher, attending physician, administrator) is implemented to ensure data security and privacy.

[0109] In the data processing and feature recognition stages, a full-process analysis system of "layered processing + multi-model fusion + cross-validation" is constructed to ensure dataset quality and diagnostic accuracy. For raw data preprocessing, targeted algorithms are used to eliminate various interferences: Hyperspectral tongue image data first undergoes dark current correction (acquiring and subtracting dark current images under no-light conditions) and spectral calibration (calibrating spectral values ​​using a standard reflector). Then, a three-level wavelet transform is performed using the db4 wavelet basis to decompose it into low-frequency approximate components and high-frequency detail components. High-frequency noise is removed using a thresholding method (threshold set to 0.02). Finally, based on principal component analysis, spectral values ​​of five characteristic wavelengths—420nm (yellowness of tongue coating), 550nm (redness of tongue body), 630nm (blood oxygen saturation), 760nm (hemoglobin concentration), and 940nm (moisture content)—are extracted, and reflectance, absorptivity, and second-order derivative spectral values ​​are calculated for each wavelength. For laser scanning point cloud data, a statistical filtering algorithm is first used to remove outliers (setting the number of neighboring points to 15, with a distance threshold). The value is 1.5 times the standard deviation. Then, the tongue surface model is fitted by RANSAC algorithm (1000 iterations, in-point threshold 0.03mm). The tongue volume (calculated by Monte Carlo integration), surface area (based on the sum of triangular mesh areas), curvature (Gaussian curvature and mean curvature), tongue enlargement index (tongue volume / standard volume ratio), and 18 morphological parameters such as crack number, length, and width are extracted. Oral environment parameters and basic health information are standardized by Z-score (converting the data into a standard normal distribution with a mean of 0 and a standard deviation of 1) to eliminate the influence of dimensions. Missing values ​​(missing rate <5%) are filled by K=5 K-nearest neighbor algorithm. Outliers (identified by box plot method, i.e. data exceeding 1.5 times the interquartile range) are processed by Winsorize standardization (replacing outliers with interquartile range boundary values). The tongue health dataset, comprising 128 feature dimensions, is integrated using feature fusion technology, combining hyperspectral features, morphological features, oral environment features, and basic health information. The Relief-F algorithm is used to calculate the importance score of each feature, and the top 60 key features are selected for subsequent modeling. The 3D digital twin is constructed using a "spectral-point cloud precision fusion" technique: First, coordinate calibration is performed based on three anatomical feature points—the tip of the tongue, the midpoint of the left edge of the tongue, and the midpoint of the root of the tongue—to establish a unified coordinate system between the hyperspectral image and the laser point cloud data (with the tip of the tongue as the origin, the midline of the tongue as the X-axis, and the perpendicular to the tongue surface as the Z-axis). An iterative nearest-point algorithm is used to achieve precise matching between the two (matching error ≤ 0.05 mm). Then, a 3D mesh model of the tongue (≥ 1 million triangular meshes) is constructed using MeshLab software. Preprocessed hyperspectral texture information is superimposed onto the surface of the mesh model to generate a visualized digital twin of the tongue containing morphological, spectral, and texture information, supporting rotation, scaling, and arbitrary cross-sectional cutting for viewing.The microcirculation evolution simulation is based on the improved lattice Boltzmann method (introducing a correction model for the non-Newtonian fluid properties of blood). A microcirculation network model of the tongue is constructed (including arterial end, capillary network and venous end, with blood vessel diameter ranging from 10-100 μm and blood vessel density consistent with the measured values). Physiological parameters such as blood viscosity of 3.5 mPa·s, hematocrit of 45%, and elastic coefficient of blood vessel wall of 0.05 MPa are set. Measured data such as blood flow velocity, blood oxygen saturation and hemoglobin concentration on the tongue surface are input as initial conditions. The evolution equation is solved using the finite difference method (time step 0.1 seconds, spatial step 0.01 mm). The simulation simulates the entire process of blood perfusion, glucose metabolism (based on the Michaelis-Menten equation to simulate glucose transport and consumption) and tongue coating formation (considering the dynamic balance of exfoliated epithelial cells, bacterial community and mucus secretion). The total simulation time is 60 seconds, and the data of blood flow velocity distribution, blood oxygen saturation distribution, glucose concentration distribution and tongue coating thickness change are output every 10 seconds. Feature recognition employs an improved Transformer-CNN hybrid model. The model structure consists of an input layer, a feature extraction layer, a fusion layer, and an output layer. The input layer is a deviation matrix between measured and simulated data (containing eight dimensions, including blood flow velocity deviation, blood oxygen saturation deviation, glucose concentration deviation, and tongue coating thickness deviation, with a matrix size of 64×64). In the feature extraction layer, the CNN module includes three convolutional layers (first convolutional layer: 32 3×3 kernels, stride 1, padding 1; second convolutional layer: 64 3×3 kernels, stride 1, padding 1; third convolutional layer: 128 3×3 kernels, stride 1, padding 1) and two max-pooling layers (2×2 pooling kernels, stride 2) for extracting local features. The Transformer module contains 6 encoder layers, each consisting of a multi-head self-attention mechanism (8 attention heads, 256 hidden layers) and a FeedForward network (512 hidden layers) to capture global feature associations. The fusion layer uses a concatenation method to fuse features extracted by CNN and Transformer, and reduces overfitting through a batch normalization layer. The output layer uses the Softmax function to output the diabetes risk level (high risk: glycated hemoglobin ≥6.5% or fasting blood glucose ≥7.0mmol / L; medium risk: glycated hemoglobin 6.0%-6.4% or fasting blood glucose 6.1-6.9mmol / L; low risk: all indicators are normal). Model training was implemented using the PyTorch framework. The training set consisted of data from 160 subjects (divided into training and validation sets in a 7:3 ratio), and the test set consisted of data from 40 subjects. The model parameters were optimized using 5-fold cross-validation (initial learning rate of 0.001, adjusted using cosine annealing, batch size of 32, and 100 iterations). An early stopping mechanism was introduced (training was stopped if the validation set loss did not decrease for 10 consecutive rounds) to prevent overfitting.After optimization, the model exhibits excellent performance: accuracy of 92.5% (37 out of 40 cases in the test set were correctly identified), sensitivity of 94.0% for high-risk identification (23 out of 25 high-risk cases were correctly identified), specificity of 91.0% (13 out of 15 non-high-risk cases were correctly identified), and an AUC of 0.95. This is significantly better than a single CNN model (85.0% accuracy) and traditional machine learning models (such as random forest with 82.5% accuracy and SVM with 80.0% accuracy).

[0110] The virtual scene construction and report generation phase focuses on "clinical practicality + user experience + intervention guidance" to create an interactive intelligent diagnostic platform. Based on the Unity3D 2021.3 engine, an interactive virtual tongue scene is built, importing a 3D digital twin of the tongue and dynamic simulation data. A multimodal rendering module is developed to achieve multi-dimensional information visualization: PBR (Physically Based Rendering) technology is used to render the tongue surface texture, realistically restoring the tongue's color and coating texture; pseudo-color coding technology is used to display the distribution of metabolites (saliva glucose concentration: red ≥1.0 mmol / L, orange 0.5-1.0 mmol / L, green <0.5 mmol / L) and microcirculation status (blood flow velocity: green ≥0.5 mm / s, yellow 0.3-0.5 mm / s, red <0.3 mm / s); transparent rendering is used to display the distribution of the internal vascular network of the tongue. The scene control module supports diverse interactive operations: zoom ratios from 0.1 to 10 times, rotation angles from 0 to 360°, and customizable cutting planes for cross-sections (XY, XZ, YZ planes and any tilted plane). It also supports local area magnification (up to 20 times) to observe subtle cracks or changes in tongue coating. The doctor's workstation integrates intelligent annotation tools, supporting point, line, and surface annotation modes. After annotating abnormal features, the system automatically retrieves the corresponding pathological significance from the clinical knowledge base (e.g., a dark red tongue indicates microcirculatory disorders, a thick, greasy tongue coating indicates insulin resistance, and tongue cracks indicate vitamin deficiency or poor blood sugar control), and displays relevant clinical research evidence (citing the latest meta-analysis results).The diabetes screening report adopts a three-dimensional architecture of "structured + visual + personalized" and contains approximately 2,000 words. It is divided into five core modules: The basic information section includes the subject's name, gender, age, testing time, data collection device model, and operator information, along with the original image of the subject's tongue and a 3D twin screenshot; the risk assessment results section uses a star rating (1-5 stars) to display the risk level, clearly defining core risk indicators (such as HbA1c 6.8%, fasting blood glucose 7.2 mmol / L) and deviations from the normal range (such as HbA1c 0.3% higher than the upper limit of normal), and includes an ROC curve to demonstrate the model's diagnostic efficacy; the abnormal feature analysis section displays abnormal features in order of importance. The system displays abnormal features (such as thick, greasy tongue coating, enlarged tongue, and slowed blood flow), providing detailed descriptions of the location of these features (e.g., thick, greasy tongue coating mainly distributed in the central region of the tongue), quantitative indicators (e.g., tongue coating thickness 85μm, normal range 30-60μm), and corresponding pathological significance (e.g., thick, greasy tongue coating is positively correlated with salivary glucose concentration, correlation coefficient 0.72). The twin simulation analysis section includes key screenshots of the virtual scene (e.g., heatmap of blood flow velocity distribution, glucose metabolism curve) and dynamic change videos (30-second simplified version), annotating physiological indicator changes at key time points. The intervention suggestion section generates personalized recommendations based on abnormal feature data, risk level, and historical screening data (test records from the past 6 months). The plan adopts a tiered intervention strategy: For the medium-risk group with thick, greasy tongue coating and elevated salivary glucose (0.8 mmol / L), a low-carbohydrate diet is recommended (daily carbohydrate intake ≤ 45% of total calories, specifically 200-250g / day of staple food, prioritizing whole grains and avoiding refined rice and flour; daily vegetable intake ≥ 500g, mainly leafy green vegetables; protein intake 1.0-1.2g / kg body weight). The exercise plan includes a brisk 30-minute walk one hour after meals (heart rate controlled at 100-120 beats / min) or swimming three times a week (40 minutes each time). For oral care, it is recommended to use a mildly alkaline mouthwash (pH 7.5-8) weekly. 0) Clean your mouth 2-3 times, with a follow-up examination every 3 months; for high-risk individuals with tongue fissures (length ≥5mm, number ≥3) and microcirculatory disturbances (blood flow velocity 0.25mm / s), in addition to dietary and exercise interventions, increase the frequency of blood glucose monitoring (5 times daily: fasting + 2 hours after each meal + before bedtime), and supplement with B vitamins (complex vitamin B tablets, 100mg daily, divided into 2 doses), and refer them to the endocrinology department for insulin release testing and screening for diabetic complications; for low-risk individuals, it is recommended to have a physical examination once a year, maintain a healthy lifestyle of 300-350g of staple food daily, exercise for more than 30 minutes, and perform regular oral cleaning. The report supports PDF export (including electronic signature function) and cloud push. Subjects can view the report through WeChat mini program. The system automatically sets abnormal indicator reminders (such as sending SMS reminders when blood glucose exceeds the standard) and provides an intervention plan execution check-in function.

[0111] The adaptive screening and control phase constructs a closed-loop management system of "dynamic correction - precise prediction - intelligent control" to achieve personalized dynamic optimization of screening strategies. Based on risk assessment results, the system automatically corrects the biological attribute parameters of the tongue digital twin, constructing a personalized model that matches the subject's actual pathological state: for high-risk individuals (diagnosed diabetic patients), the blood flow velocity parameter is adjusted to 90% of the measured value (simulating slowed blood flow due to microcirculatory disturbances), the salivary glucose metabolism rate is adjusted to 85% of the measured value (simulating metabolic disorders caused by insulin resistance), and the tongue coating formation rate is adjusted to 120% of the measured value (simulating excessive bacterial growth under hyperglycemic conditions); for medium-risk individuals (prediabetes), the parameters are adjusted to 95% (blood flow velocity, metabolic rate) and 110% (tongue coating formation rate) of the measured values, simulating mild pathological changes; for low-risk individuals, the measured parameters remain unchanged for dynamic monitoring of changes in health status. The prediction of diabetes progression trends employs a log-normal distribution model combined with an LSTM time-series prediction network. Inputs include corrected twin parameters (blood flow velocity, glucose metabolism rate, etc.), historical blood glucose data (mean fasting blood glucose and blood glucose variability over the past 3 months), and lifestyle data (daily staple food intake, exercise duration, smoking and alcohol consumption). Three prediction periods are set: 3 months, 6 months, and 12 months. The model uses maximum likelihood estimation to determine the location parameter (μ) and scale parameter (σ) of the log-normal distribution. The LSTM network captures long-term dependencies in the time-series data (network structure: 16 neurons in the input layer, 3 hidden layers, each with 64 neurons, and 3 neurons in the output layer corresponding to different risk levels). The output shows the probability of elevated blood glucose and the trend of risk level changes at different time points.The adaptive regulation rules are formulated based on prediction results and clinical guidelines, and intelligent regulation is achieved using a decision tree algorithm: For high-risk individuals, if the predicted probability of HbA1c ≥7.0% within 6 months is ≥60%, the screening frequency is adjusted from once a month to once every 2 weeks, and intervention recommendations include adding an insulin sensitivity improvement program (such as 20 minutes of yoga every morning or 30 minutes of Tai Chi after dinner, supplementing with alpha-lipoic acid 600mg / day), with monthly follow-up by an endocrinologist; For intermediate-risk individuals, if the predicted probability of progressing to high-risk within 12 months is ≥50%, the screening frequency is adjusted from once every 3 months to once a month, and dietary interventions are strengthened (such as limiting refined sugar intake to ≤25g / day, avoiding sugary drinks, and consuming [unspecified food] weekly). For individuals with a risk level of ≤50g of nuts, participation in diabetes prevention health courses (twice a month) is recommended. For individuals at medium risk with a predicted probability of turning low-risk within 12 months ≥40%, maintain a screening frequency of once every 3 months and appropriately relax dietary restrictions (carbohydrate intake can be increased to 50% of total calories). For individuals at low risk with a predicted probability of abnormal blood sugar within 12 months <10%, maintain a screening frequency of once every 6 months, with intervention recommendations focusing on health education (such as regular oral hygiene, regular sleep patterns, and avoiding staying up late). For individuals at low risk with a predicted probability of abnormal blood sugar within 12 months ≥20% (e.g., a family history of diabetes and a BMI ≥28kg / m²), adjust the screening frequency to once every 3 months, and add waist circumference monitoring and dietary guidance. All control parameters are synchronized to the cloud database in real time. The system automatically calls the updated twin model during the next screening to achieve dynamic adaptation of the screening process. This adaptive control mechanism can increase the early detection rate of diabetic complications in high-risk individuals by 35% and reduce the conversion rate in medium-risk individuals by 28%, significantly improving the timeliness and targeting of screening.

[0112] In summary, the embodiments of this application employ a rigorous subject screening process to eliminate confounding factors, and conduct comprehensive, multi-source data collection and standardized processing to provide a data foundation for diagnostic assessment. A hybrid model assessment system combining tongue digital twin modeling and optimization accurately distinguishes diabetes risk levels, significantly improving the reliability and accuracy of diagnostic assessment. The construction of a three-dimensional visualized virtual scene and personalized stratified intervention programs effectively improve the early detection of complications in high-risk populations and reduce the probability of intermediate-risk individuals progressing to high-risk. Structured reports and cloud-based push notifications further ensure timely delivery of intervention guidance. An adaptive closed-loop control mechanism linked to a dynamic trend prediction system integrates multi-dimensional data to accurately predict risk progression, dynamically optimizes screening frequency and intervention strategies, significantly improves screening targeting and reduces ineffective screening, shortens the risk intervention response cycle, and comprehensively enhances the accuracy, efficiency, and personalization of diabetes screening and intervention, providing strong support for diabetes prevention and control.

[0113] Figure 10A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0114] The memory 1001, the processor 1002, and the computer program stored on the memory 1001 and capable of running on the processor 1002.

[0115] When the processor 1002 executes the program, it implements a tongue diagnosis method for screening diabetes provided in the above embodiments.

[0116] Furthermore, electronic devices also include:

[0117] Communication interface 1003 is used for communication between memory 1001 and processor 1002.

[0118] The memory 1001 is used to store computer programs that can run on the processor 1002.

[0119] The memory 1001 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0120] If the memory 1001, processor 1002, and communication interface 1003 are implemented independently, then the communication interface 1003, memory 1001, and processor 1002 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 10 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0121] Optionally, in a specific implementation, if the memory 1001, processor 1002, and communication interface 1003 are integrated on a single chip, then the memory 1001, processor 1002, and communication interface 1003 can communicate with each other through an internal interface.

[0122] The processor 1002 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0123] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the tongue diagnosis method for screening diabetes as described above.

[0124] Furthermore, this application also provides a computer program product, including a computer program or instructions, which, when executed, implement the above-described tongue diagnosis method for screening diabetes.

[0125] In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0126] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0127] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0128] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0129] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0130] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A tongue diagnosis system for screening diabetes, characterized in that, include: The system includes a tongue sensing module, a tongue twin construction module, an intelligent analysis center, a visualization interaction unit, and an adaptive screening and control unit; among these components... The tongue sensing module is used to collect tongue physiological data, oral environment parameters, and basic user health information. The tongue twin construction module is used to construct a digital twin of the tongue. Based on the biological attributes and metabolic characteristics of the tongue, it uses a microcirculation evolution algorithm to simulate changes in the physiological state of the tongue and obtain tongue simulation data. The intelligent analysis center is used to identify diabetes-related tongue abnormalities and assess diabetes risk levels by using the tongue simulation data and combining it with an improved Transformer-CNN model. The visualization interaction unit is used to construct an interactive virtual tongue scene, render tongue morphological features, metabolite distribution and microcirculation status in real time, and generate diabetes screening reports and intervention suggestions. The adaptive screening control unit is used to predict the progression trend of diabetes based on the diabetes risk level and the diabetes screening report, and dynamically adjust the screening frequency and the intervention plan.

2. The tongue diagnosis system for screening diabetes according to claim 1, characterized in that, The tongue sensing module includes a hyperspectral tongue imaging unit, a tongue coating thickness detection unit, a tongue microcirculation sensor, and an oral environment monitoring group. The hyperspectral tongue imaging unit is used to capture the spectral characteristics and tissue texture data of the tongue; the tongue coating thickness detection unit acquires the distribution density and thickness information of the tongue coating; the tongue microcirculation sensor is used to collect the capillary blood flow velocity and blood oxygen saturation of the tongue; and the oral environment monitoring group is used to collect oral humidity, pH value, and volatile metabolite concentration.

3. A tongue diagnosis system for screening diabetes according to claim 1, characterized in that, The tongue twin construction module includes a tongue twin construction unit and a microcirculation evolution simulation unit. The tongue twin construction unit is used to construct a three-dimensional digital twin containing tongue morphology, mucosal texture, tongue coating distribution, and capillary network based on hyperspectral tongue image data and laser scanning point cloud. The microcirculation evolution simulation unit is used to simulate the dynamic changes of blood perfusion, tissue metabolism, and tongue coating evolution of the tongue under blood glucose fluctuations using a microcirculation evolution algorithm based on the biological attributes and metabolic characteristics of the tongue, and output multi-physics field simulation data of the tongue.

4. A tongue diagnosis system for screening diabetes according to claim 1, characterized in that, The intelligent analysis center includes an abnormal feature recognition unit and a risk level assessment module. The abnormal feature recognition unit uses an improved Transformer-CNN model, combined with tongue physiological data, oral environment parameters, and clinical blood glucose correlation data, to identify tongue color, microcirculation stagnation, abnormal metabolite concentration, and diabetes-related tongue abnormalities. The risk level assessment module is used to quantitatively assess the screening risk of low, medium, and high diabetes based on the type, intensity, and distribution pattern of abnormal features, combined with the user's age, body mass index, family medical history, and other basic health information.

5. A tongue diagnosis system for screening diabetes according to claim 1, characterized in that, The visualization interaction unit includes a tongue rendering engine, a screening report generator, and a multimodal interaction module. The tongue rendering engine is used to render the three-dimensional morphology, spectral feature distribution, metabolite concentration cloud map, and microcirculation heat map of the tongue in real time, and simultaneously mark abnormal areas. The screening report generator is used to generate a diabetes screening report that includes details of tongue abnormalities, risk quantification results, dietary adjustment plans, exercise suggestions, and medical advice based on risk level and abnormal feature analysis. The multimodal interaction module is used to provide voice commands, gesture control, and virtual dissection functions, allowing users to rotate, zoom, and focus on abnormal areas in the virtual tongue scene.

6. A tongue diagnosis system for screening diabetes according to claim 1, characterized in that, The adaptive screening and control unit includes a progression trend prediction module and a screening strategy adjustment unit. The progression trend prediction module is used to predict the probability of diabetes onset and progression rate of high-risk users based on log-normal distribution and tongue physiological state degeneration data. The screening strategy adjustment unit is used to dynamically adjust the screening frequency and intervention recommendations based on the prediction results and user lifestyle data.

7. A method for using a tongue diagnosis system for screening diabetes according to any one of claims 1-6, characterized in that, The method includes: Acquire tongue physiological data, oral environment parameters, and basic user health information; The tongue physiological data, oral environment parameters, and basic health information are processed to generate a tongue health dataset. Based on the tongue health dataset, a three-dimensional digital twin of the tongue is constructed using hyperspectral tongue image data and laser scanning point cloud. The dynamic changes of tongue blood flow, metabolism, and tongue coating are simulated using a microcirculation evolution algorithm to obtain tongue simulation data. By comparing the deviation between the measured data and the virtual simulation results, and combining the improved Transformer-CNN model, abnormal tongue features related to diabetes are identified, and the risk level of diabetes is quantitatively assessed to obtain risk assessment results and abnormal feature data. Based on the risk assessment results and the abnormal feature data, combined with the historical screening data of the tongue twin, an interactive virtual tongue scene is constructed to render the tongue morphology, metabolite distribution and microcirculation status in real time, and generate a diabetes screening report and personalized intervention suggestions. Based on the diabetes risk level and the diabetes screening report, adaptive screening regulation is performed, the biological attribute parameters of the tongue digital twin are corrected in real time, the progression trend of diabetes is predicted by combining log-normal distribution, and the screening frequency and intervention recommendations are dynamically adjusted according to the prediction results and user lifestyle data.

8. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the tongue diagnosis method for screening diabetes as described in claim 7.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When a computer program or instruction is executed, it implements the tongue diagnosis method for screening diabetes as described in claim 7.

10. A computer program product, comprising a computer program or instructions, characterized in that, When a computer program or instruction is executed, it implements the tongue diagnosis method for screening diabetes as described in claim 7.

Images