Intelligent nutrition diagnosis and treatment simulation system
By constructing a nutritional metabolism mechanism model and a physical information neural network, combined with the DeepSHAP method, the intelligent nutrition diagnosis and treatment system has achieved real-time data acquisition and personalized nutrition plan generation, solving the problems of insufficient data acquisition and inaccurate assessment in existing systems, and improving the health management effect of patients with chronic diseases.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-26
- Publication Date
- 2026-04-03
AI Technical Summary
Existing nutritional diagnosis and treatment systems struggle to achieve real-time data collection, lack clinical-grade precision and personalized assessment, and are unable to meet the personalized nutritional management needs of patients with chronic diseases.
The system uses a data monitoring module to acquire and preprocess user vital signs data, constructs a nutritional metabolism mechanism model and trains a physical information neural network, and combines the DeepSHAP method to identify key factors and generate personalized nutritional treatment plans.
It enables real-time data collection and personalized nutritional diagnosis and treatment, improves the accuracy and interpretability of health management, ensures the correlation between nutritional adjustment measures and the goal of improving vital signs, and enhances the trust of users and doctors.
Smart Images
Figure CN121789991A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent medical technology, and more specifically to an intelligent nutrition diagnosis and treatment simulation system. Background Technology
[0002] Currently, the prevalence of chronic diseases such as diabetes and hypertension continues to rise, and precision nutrition management has become a core element in the prevention and control of these diseases. However, existing nutrition diagnosis and treatment systems have significant technical shortcomings, making it difficult to meet the personalized needs of patients. The main problems are concentrated in three aspects: First, the data collection method is outdated. Existing systems generally rely on users to manually input dietary data, which is not only cumbersome but also prone to incomplete and inaccurate data due to estimation errors. At the same time, manual input cannot achieve real-time collection of dietary data, resulting in poor time synchronization with vital sign data and making it difficult to establish an immediate correlation between diet and vital signs, thus failing to provide timely support for nutritional adjustments.
[0003] Secondly, the monitoring of vital signs lacks real-time capability. The vital signs of patients with chronic diseases, such as blood sugar and blood pressure, often fluctuate dynamically with diet and medication. However, existing systems mostly rely on users to manually upload data periodically or only connect to a single static device, which cannot continuously capture changes in vital signs. This often leads to missing the best time to adjust the treatment plan, increasing health risks.
[0004] Third, the assessment logic lacks clinical-grade precision. Existing nutritional assessment systems are mostly based on general dietary guidelines or simple algorithms, failing to fully incorporate clinical information such as patient disease type and complications. They also rely heavily on human experience, with inconsistent assessment standards, making it difficult to output personalized nutritional parameters such as daily carbohydrate intake and sodium intake limits, thus failing to meet the precise needs of patients with chronic diseases.
[0005] In summary, there is an urgent need to develop an intelligent nutrition diagnosis and treatment simulation system that can achieve real-time data acquisition and clinical-level assessment. Summary of the Invention
[0006] In view of this, the present invention provides an intelligent nutrition diagnosis and treatment simulation system that can realize personalized, precise and interpretable nutrition diagnosis and treatment, effectively assist clinical nutrition decision-making and improve health management results.
[0007] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent nutrition diagnosis and treatment simulation system, comprising: a data monitoring module, a nutrition diagnosis and treatment module, a nutrition assessment module, and a suggestion generation module; The data monitoring module is used to acquire and preprocess users' historical vital signs data to obtain preliminary vital signs data. The nutrition diagnosis and treatment module constructs a nutritional metabolism mechanism model based on preliminary vital signs data and trains the nutritional metabolism mechanism model. The nutrition assessment module is used to obtain the current user's vital signs data and input the data into the nutrition metabolism mechanism model to obtain a personalized nutrition diagnosis and treatment plan; The suggestion generation module provides professional advice based on personalized nutritional treatment plans.
[0008] Preferably, the process by which the data monitoring module preprocesses the user's historical vital sign data includes: Remove noise and outliers from users' historical vital signs data, and standardize the numerical data; Filter out vital signs data that have a direct impact on the user's nutritional status; By aggregating and analyzing historical vital signs data, statistical indicators over a period of time are calculated to obtain preliminary vital signs data.
[0009] Preferably, the nutrient metabolism mechanism model includes a physical information neural network, and the process of constructing the physical information neural network includes: Set boundary conditions and initial model parameters based on preliminary vital sign data; The Sobol method was used to determine the sensitive parameters of the model, and Bayesian optimization was used to calibrate the sensitive parameters. The mean square error and Nash efficiency coefficient were used to evaluate the model accuracy. After running and calibrating, the nutrient metabolism mechanism model generated nutrient metabolism flux index and vital sign index data. A training dataset is constructed by aligning the nutrient metabolic flux indicators, vital sign indicators, and the preliminary vital sign data by time. A physical information neural network is then built based on the training dataset to provide physical constraints for the nutrient metabolic mechanism model.
[0010] Preferably, the process of training a physical information neural network includes: A physical guidance regularization term is introduced into the mean square error loss, and a self-stepping learning strategy is used to pre-train the physical information neural network. With the goal of minimizing the validation set error, Bayesian optimization is used to search for the optimal combination of learning rate, regularization coefficient, and number of hidden layer nodes to complete the hyperparameter optimization. The main structure of the fixed physical information neural network is set, and the weights of the last two fully connected layers are set as adjustable parameters. The model is fine-tuned using a fine-tuning dataset with transfer learning and weighted sampling balance. The root mean square error and Nash efficiency coefficient are used to evaluate the accuracy of the fine-tuned model.
[0011] Preferably, the boundary conditions in the nutritional metabolism mechanism model include ambient temperature, gut microbiota metabolic efficiency coefficient, and hormone regulation coefficient; the simulation effect of the nutritional metabolism mechanism model is evaluated using user historical medical data and nutritional metabolic flux measurement data.
[0012] Preferably, in the nutritional metabolism mechanism model, the introduction of a physical-guided regularization term includes: A nutrient metabolic flux conservation term is introduced, and the residual is calculated based on the mass balance condition. The residual is then added to the loss function. The physical relationship between dietary intake, exercise expenditure, and physical indicators is defined by a partial dependency graph, and the physical relationship is learned by a regularization-forced model.
[0013] Preferably, the nutritional metabolism mechanism model can also use the DeepSHAP method to calculate the contribution of each input feature to the output result of the scheme, identify key factors affecting the user's vital signs based on SHAP values, and analyze the response relationship between key factors and vital sign indicators through partial dependency graphs.
[0014] Preferably, the suggestion generation module provides professional advice, including dietary structure adjustment, exercise intensity adaptation, and lifestyle optimization, based on the personalized nutrition treatment plan and key factor analysis results.
[0015] As can be seen from the above technical solution, compared with the prior art, the present invention discloses an intelligent nutrition diagnosis and treatment simulation system, which has the following beneficial effects: By coupling a nutritional metabolism mechanism model with a physical information neural network, the model utilizes the prior knowledge of the mechanism model and fits data patterns through deep learning, thus avoiding the defects of purely data-driven models and the limitations of purely mechanism models. Through strategies such as Bayesian optimization for parameter calibration and transfer learning for fine-tuning, the model's adaptability to individual differences is improved, ensuring stable prediction accuracy across different user groups.
[0016] This invention can also generate personalized nutritional treatment plans based on users' real-time vital sign data, identify key influencing factors by combining the DeepSHAP method, and analyze the response relationship between factors and vital signs through partial dependency graphs, making the plans more targeted; it solves the problem of uninterpretable decision-making in traditional models, clarifies the correlation between adjustment measures, influencing factors, and vital sign improvement goals, and enhances the trust of users and doctors in the plans. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0018] Figure 1 The system structure diagram provided for this invention. Detailed Implementation
[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0020] like Figure 1 As shown in the figure, an embodiment of the present invention discloses an intelligent nutrition diagnosis and treatment simulation system, including: a data monitoring module, a nutrition diagnosis and treatment module, a nutrition assessment module, and a suggestion generation module; The data monitoring module is used to acquire and preprocess users' historical vital signs data to obtain preliminary vital signs data. The nutrition diagnosis and treatment module constructs a nutritional metabolism mechanism model based on preliminary vital signs data and trains the nutritional metabolism mechanism model. The nutrition assessment module is used to obtain the current user's vital signs data and input the data into the nutrition metabolism mechanism model to obtain a personalized nutrition diagnosis and treatment plan; The suggestion generation module provides professional advice based on personalized nutritional treatment plans.
[0021] Specifically, the data monitoring module's preprocessing of users' historical vital sign data includes: Remove noise and outliers from users' historical vital signs data, and standardize the numerical data; Filter out vital signs data that have a direct impact on the user's nutritional status; By aggregating and analyzing historical vital signs data, statistical indicators over a period of time are calculated to obtain preliminary vital signs data.
[0022] In a specific embodiment of the present invention, the historical vital signs data includes: physical examination data of the subject over the past year retrieved through the electronic medical record system of the community health service center, including 12 indicators such as fasting blood glucose, 2-hour postprandial blood glucose, total cholesterol, triglycerides, BMI, waist circumference, and blood pressure.
[0023] Specifically, the nutritional metabolism mechanism model includes a physical information neural network, and the process of constructing the physical information neural network includes: Set boundary conditions and initial model parameters based on preliminary vital sign data; The Sobol method was used to determine the sensitive parameters of the model, and Bayesian optimization was used to calibrate the sensitive parameters. The mean square error and Nash efficiency coefficient were used to evaluate the model accuracy. After running and calibrating, the nutrient metabolism mechanism model generated nutrient metabolism flux index and vital sign index data. A training dataset is constructed by aligning the nutrient metabolic flux indicators, vital sign indicators, and the preliminary vital sign data by time. A physical information neural network is then built based on the training dataset to provide physical constraints for the nutrient metabolic mechanism model.
[0024] In a specific embodiment of the present invention, noise and outlier processing adopts the 3σ principle to remove outlier data that exceed the mean ± 3 standard deviations, such as a single mismeasured fasting blood glucose of 15 mmol / L, and completes the missing daily data through linear interpolation. For numerical indicators, Z-score standardization is used to convert data of different dimensions such as blood glucose (unit: mmol / L) and BMI (unit: kg / m²) into standard values with a mean of 0 and a standard deviation of 1, so as to avoid the impact of differences in magnitude on model training. Statistical indicators are calculated on a weekly basis, such as the weekly average fasting blood glucose level and the total weekly exercise duration. This results in preliminary vital signs data for each subject, comprising 12 characteristics, which are then stored in the system's cloud database.
[0025] Specifically, the process of training a physical information neural network includes: A physical guidance regularization term is introduced into the mean square error loss, and a self-stepping learning strategy is used to pre-train the physical information neural network. With the goal of minimizing the validation set error, Bayesian optimization is used to search for the optimal combination of learning rate, regularization coefficient, and number of hidden layer nodes to complete the hyperparameter optimization. The main structure of the fixed physical information neural network is set, and the weights of the last two fully connected layers are set as adjustable parameters. The model is fine-tuned using a fine-tuning dataset with transfer learning and weighted sampling balance. The root mean square error and Nash efficiency coefficient are used to evaluate the accuracy of the fine-tuned model.
[0026] Specifically, the boundary conditions in the nutritional metabolism mechanism model include ambient temperature, gut microbiota metabolic efficiency coefficient, and hormone regulation coefficient; the simulation effect of the nutritional metabolism mechanism model is evaluated using user historical medical data and nutritional metabolic flux measurement data.
[0027] Specifically, in the nutritional metabolism mechanism model, the introduction of the physical-guided regularization term includes: A nutrient metabolic flux conservation term is introduced, and the residual is calculated based on the mass balance condition. The residual is then added to the loss function. The physical relationship between dietary intake, exercise expenditure, and physical indicators is defined by a partial dependency graph, and the physical relationship is learned by a regularization-forced model.
[0028] Specifically, the nutritional metabolism mechanism model can also use the DeepSHAP method to calculate the contribution of each input feature to the output result of the scheme, identify key factors affecting the user's vital signs based on SHAP values, and analyze the response relationship between key factors and vital sign indicators through partial dependency graphs.
[0029] Specifically, the suggestion generation module provides professional advice based on personalized nutritional treatment plans and key factor analysis results, including adjustments to dietary structure, appropriate exercise intensity, and optimization of lifestyle.
[0030] In one specific embodiment of the present invention, the subject uploads current vital signs data (such as fasting blood glucose and daily breakfast intake) daily via a system applet. The nutrition assessment module automatically inputs the data into a trained nutritional metabolism mechanism model and outputs a customized nutritional treatment plan. Taking this embodiment (male, 52 years old, fasting blood glucose 6.3 mmol / L, BMI 26.5 kg / m²) as an example, the system-generated plan includes: Dietary recommendations: 1800kcal total daily, with carbohydrates making up 45% of the total intake. Prioritize low-GI foods such as brown rice and oats. A daily intake of 25% protein requires one egg and 100g of lean meat. If fat content is 30%, animal fat intake should be limited; Key points of vital sign monitoring: Additional monitoring of postprandial blood glucose 2 hours every Monday, Wednesday and Friday; Metabolic goals: Reduce fasting blood glucose to below 5.9 mmol / L and BMI to below 25 kg / m² within 3 months.
[0031] Furthermore, it is suggested that the generation module combine the DeepSHAP factor analysis results (based on the fact that the subjects' postprandial blood glucose was most affected by refined carbohydrate intake) to generate multidimensional suggestions: Dietary adjustments: Replace white rice with brown rice every day, add ≥5g / 100g of dietary fiber to dinner, and avoid eating fruit after dinner; Exercise adjustments: Take a brisk 30-minute walk 1 hour after dinner every day (heart rate controlled at 100-120 beats / minute), and add 2 Tai Chi training sessions per week (40 minutes each time). Optimize your sleep schedule: Go to bed before 10:30 PM every day (to avoid staying up late, which can raise cortisol levels and affect blood sugar), and drink 1500-1700 mL of drinking water multiple times a day. All suggestions will be reviewed by a community endocrinologist and then sent to the participant's mini-program. The doctor can check the participant's progress through the system backend and conduct an online follow-up consultation once a month to make adjustments.
[0032] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.
[0033] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. An intelligent nutrition diagnosis and treatment simulation system, characterized in that, include: The module includes a data monitoring module, a nutrition diagnosis and treatment module, a nutrition assessment module, and a suggestion generation module. The data monitoring module is used to acquire and preprocess users' historical vital signs data to obtain preliminary vital signs data. The nutrition diagnosis and treatment module constructs a nutritional metabolism mechanism model based on preliminary vital signs data and trains the nutritional metabolism mechanism model. The nutrition assessment module is used to obtain the current user's vital signs data and input the data into the nutrition metabolism mechanism model to obtain a personalized nutrition diagnosis and treatment plan; The suggestion generation module provides recommendations based on personalized nutritional treatment plans.
2. The intelligent nutrition diagnosis and treatment simulation system according to claim 1, characterized in that, The data monitoring module preprocesses user historical vital sign data, including the following steps: Remove noise and outliers from users' historical vital signs data, and standardize the numerical data; Filter out vital signs data that have a direct impact on the user's nutritional status; By aggregating and analyzing historical vital signs data, statistical indicators over a period of time are calculated to obtain preliminary vital signs data.
3. The intelligent nutrition diagnosis and treatment simulation system according to claim 2, characterized in that, The nutritional metabolism mechanism model includes a physical information neural network, and the process of constructing the physical information neural network includes: Set boundary conditions and initial model parameters based on preliminary vital sign data; The Sobol method was used to determine the sensitive parameters of the model, and Bayesian optimization was used to calibrate the sensitive parameters. The mean square error and Nash efficiency coefficient were used to evaluate the model accuracy. After running and calibrating, the nutrient metabolism mechanism model generated nutrient metabolism flux index and vital sign index data. A training dataset is constructed by aligning the nutrient metabolic flux indicators, vital sign indicators, and the preliminary vital sign data by time. A physical information neural network is then built based on the training dataset to provide physical constraints for the nutrient metabolic mechanism model.
4. The intelligent nutrition diagnosis and treatment simulation system according to claim 3, characterized in that, The process of training a physical information neural network includes: A physical guidance regularization term is introduced into the mean square error loss, and a self-stepping learning strategy is used to pre-train the physical information neural network. With the goal of minimizing the validation set error, Bayesian optimization is used to search for the optimal combination of learning rate, regularization coefficient, and number of hidden layer nodes to complete the hyperparameter optimization. The main structure of the fixed physical information neural network is set, and the weights of the last two fully connected layers are set as adjustable parameters. The model is fine-tuned using a fine-tuning dataset with transfer learning and weighted sampling balance. The root mean square error and Nash efficiency coefficient are used to evaluate the accuracy of the fine-tuned model.
5. The intelligent nutrition diagnosis and treatment simulation system according to claim 4, characterized in that, The boundary conditions in the nutritional metabolism mechanism model include ambient temperature, gut microbiota metabolic efficiency coefficient, and hormone regulation coefficient; the simulation effect of the nutritional metabolism mechanism model is evaluated using user historical medical data and nutritional metabolic flux measurement data.
6. The intelligent nutrition diagnosis and treatment simulation system according to claim 4, characterized in that, In the aforementioned nutritional metabolism mechanism model, the introduction of a physical-guided regularization term includes: A nutrient metabolic flux conservation term is introduced, and the residual is calculated based on the mass balance condition. The residual is then added to the loss function. The physical relationship between dietary intake, exercise expenditure, and physical indicators is defined by a partial dependency graph, and the physical relationship is learned by a regularization-forced model.
7. The intelligent nutrition diagnosis and treatment simulation system according to claim 3, characterized in that, The nutritional metabolism mechanism model can also use the DeepSHAP method to calculate the contribution of each input feature to the output result of the scheme, identify key factors affecting the user's vital signs based on SHAP values, and analyze the response relationship between key factors and vital sign indicators through partial dependency graphs.
8. The intelligent nutrition diagnosis and treatment simulation system according to claim 7, characterized in that, Based on the personalized nutrition treatment plan and the analysis results of key factors, the suggestion generation module provides professional suggestions including dietary structure adjustment, exercise intensity adaptation, and lifestyle optimization.