Personalized diet intervention system and method for patient with coronary heart disease

Abstract

The invention relates to the field of medical informatization, in particular to a personalized diet intervention system and method for coronary heart disease patients. The invention provides a personalized diet intervention system for coronary heart disease patients, and integrates intelligent tableware weighing, image recognition, knowledge reasoning, feature embedding Transform network, electronic taste sensing, a medicine-food interaction database, a food knowledge map and an intestinal flora detection and traceability module. The system monitors food weight, types and sodium and potassium intake in real time, analyzes dietary behaviors in combination with blood sugar and blood fat data, and identifies diet-cardiovascular event association. Through knowledge graph reasoning, a personalized recipe is generated, medicine-food interaction is considered, and diet suggestions are optimized. And intestinal flora detection and block chain traceability improve the suggestion accuracy and credibility. Multiple technologies are combined, comprehensive, accurate and easy-to-use diet intervention is achieved, the effect is improved, the health of patients is improved, and a new normal form is established for intelligent development of chronic disease management.

Description

Technical Field

[0001] The present invention relates to the field of medical informatization, and particularly to a personalized dietary intervention system and method for coronary heart disease patients. Background Art

[0002] With the high incidence and mortality of cardiovascular diseases globally, the dietary management of coronary heart disease patients has become a key concern in the medical field. Traditional dietary intervention methods usually adopt unified dietary guidelines, which are difficult to meet the individual needs of each patient. In recent years, with the progress of technology, personalized dietary intervention systems have gradually become a research hotspot.

[0003] Existing personalized dietary intervention systems mainly rely on self-reported dietary records and regular physical examination data of patients. This method has problems of inaccurate data and untimely updates, and it is difficult to capture the subtle changes in patients' daily dietary behaviors. Some systems attempt to introduce smart devices for data collection, but they are often limited to single-dimensional data, such as only recording food weight or calories, lacking a comprehensive analysis of food types and nutritional components.

[0004] In terms of data analysis, existing systems mostly adopt simple statistical methods or traditional machine learning algorithms. These methods perform poorly in dealing with high-dimensional and time-series health data, and it is difficult to accurately capture the complex relationship between dietary behaviors and cardiovascular health. At the same time, when generating dietary recommendations, existing systems often ignore important factors such as drug-food interactions and gut microbiota, which may lead to inconsistent recommendation results with the actual situation of patients.

[0005] In addition, existing systems generally have problems of poor user experience and low compliance. Patients need to manually record dietary information, which is cumbersome to operate; the recommendation results are presented in text form, lacking intuitiveness and interactivity, and it is difficult to stimulate patients' interest in continuous use.

[0006] Facing these problems, there is an urgent need for a system that can comprehensively, accurately, and real-time monitor patients' dietary behaviors, deeply analyze the diet-health relationship, and provide personalized and easy-to-implement dietary intervention plans. Summary of the Invention

[0007] The personalized dietary intervention system and method for coronary heart disease patients of the present invention are precisely proposed to address the above technical problems. Through the organic combination of innovative technologies such as multi-modal data collection, deep learning analysis, and knowledge graph reasoning, the present invention realizes the all-round and real-time monitoring and analysis of patients' dietary behaviors, and greatly improves the accuracy and effectiveness of dietary intervention.

[0008] The present invention proposes a personalized dietary intervention system and method for coronary heart disease patients, including:

[0009] An intelligent tableware weighing unit, for:

[0010] Obtain food weight data in real time;

[0011] Transmit the food weight data to the server;

[0012] An image recognition module, communicatively connected to the intelligent tableware weighing unit, for:

[0013] Obtain a food image;

[0014] Classify and recognize the food image based on a pre-trained neural network to obtain food category information;

[0015] A knowledge inference unit, communicatively connected to the image recognition module, for:

[0016] Receive the food category information and the food weight data;

[0017] Generate a food intake list based on the food category information and the food weight data;

[0018] Analyze the food intake list according to a preset knowledge base to generate a personalized dietary intervention plan;

[0019] An improved feature-embedded Transformer network model, communicatively connected to the knowledge inference unit, for:

[0020] Receive the patient's dietary behavior data, blood glucose data, and blood lipid data;

[0021] Construct an association model of the patient's dietary behavior - blood lipid and blood glucose monitoring results including time dimension information;

[0022] Identify the diet - cardiovascular event association pattern;

[0023] An electronic taste sensing module, communicatively connected to the knowledge inference unit, for:

[0024] Monitor the patient's sodium and potassium intake in real time;

[0025] Send the data including the sodium and potassium intake to the knowledge inference unit;

[0026] A drug - food interaction database, communicatively connected to the knowledge inference unit, for:

[0027] Store a set of rules related to antiplatelet drugs and food ingredients;

[0028] Respond to the query request of the knowledge inference unit and provide drug - food interaction information;

[0029] The rule-based food knowledge graph construction module, which is communicatively connected to the knowledge reasoning unit and the drug-food interaction database, is configured to:

[0030] Construct a food knowledge graph based on food ingredient entities, drug ingredient entities, food-drug interaction rules, and knowledge triples;

[0031] The knowledge graph recommendation system based on graph neural networks, which is communicatively connected to the rule-based food knowledge graph construction module and the knowledge reasoning unit, is configured to:

[0032] Generate personalized recipes based on the recommendation results of the food knowledge graph;

[0033] Send the personalized recipes to the knowledge reasoning unit;

[0034] The gut microbiota detection module, which is communicatively connected to the knowledge reasoning unit, is configured to:

[0035] Extract and qualitatively analyze microbial genes through a rapid detection kit;

[0036] Send the detection results to the knowledge reasoning unit;

[0037] The traceability module, which is communicatively connected to the knowledge reasoning unit, is configured to:

[0038] Monitor and verify the traceability of the entire production process of food through blockchain technology;

[0039] Send the traceability information to the knowledge reasoning unit.

[0040] Preferably, the intelligent tableware weighing unit includes:

[0041] A pressure sensor, which is arranged at the bottom of the dinner plate and is used to record the weight information of a single piece of tableware;

[0042] An RFID chip, which is used to record the dinner plate ID and cooking information;

[0043] A data processing unit, which is communicatively connected to the pressure sensor and the RFID chip and is configured to:

[0044] Receive the weight information, dinner plate ID, and cooking information;

[0045] Package the weight information, dinner plate ID, and cooking information to generate lunch box information;

[0046] Perform an encrypted signature on the lunch box information;

[0047] Upload the encrypted and signed lunch box information to the server.

[0048] Preferably, the knowledge inference unit includes:

[0049] A neural network training module for training a neural network for food recognition;

[0050] A data preprocessing module for:

[0051] Removing outliers and noise data;

[0052] Performing normalization and binning operations on the remaining data;

[0053] Obtaining data blocks in batches;

[0054] An attention mechanism modeling module, communicatively connected to the data preprocessing module, for:

[0055] Receiving the data blocks output by the data preprocessing module;

[0056] Dividing each block of data into multiple vectors;

[0057] Concatenating the multiple vectors and inputting them into a layer of neural network to generate attention weights;

[0058] Applying the attention weights to each vector in the data block simultaneously;

[0059] Fusing the updated vectors and outputting the features with enhanced attention;

[0060] A neural network implementation module, communicatively connected to the attention mechanism modeling module, for:

[0061] Utilizing neural network technology to realize the non-linear dependence relationship between features;

[0062] Combining a convolutional neural network to extract a feature mapping function;

[0063] Realizing the expression of the data after attention mechanism modeling;

[0064] A knowledge graph analysis module based on the attention mechanism, communicatively connected to the neural network implementation module, for:

[0065] Calculating the similarity of each entity in the knowledge base;

[0066] Sorting the calculation results;

[0067] Inferring implicit information based on the relationships between entities;

[0068] Obtaining a personalized diet recommendation intervention plan for coronary heart disease patients;

[0069] A knowledge base management module, communicatively connected to the knowledge graph analysis module based on the attention mechanism, for:

[0070] Store the generated dietary recommendation intervention plan;

[0071] Carry out learning and training on the dietary recommendation intervention plan;

[0072] Use the learning result as an extension and supplement to the knowledge graph.

[0073] Preferably, it further includes:

[0074] A blood drug concentration detection module, communicatively connected to the knowledge reasoning unit, for:

[0075] Real-time monitor the interaction process between antiplatelet drugs and food through a chip based on electrochemical sensor technology;

[0076] Realize real-time dynamic monitoring of blood drug concentration;

[0077] Send the monitoring result to the knowledge reasoning unit;

[0078] An augmented reality module, communicatively connected to the knowledge reasoning unit, for:

[0079] Present the personalized dietary recommendation intervention plan in an augmented reality manner through a VR device;

[0080] Record the food intake status of the user through an AR device;

[0081] Construct a virtual tableware through three-dimensional reconstruction technology and record the food weight information;

[0082] Assist the user in dining.

[0083] Preferably, the input data of the improved feature embedding Transformer network model is mapped to a certain range through quantile transformation, and the potential features of the input data are learned using the self-attention mechanism after passing through the Embedding layer.

[0084] Preferably, the electronic taste sensing module includes:

[0085] A brush for forming an electric current on the tongue surface;

[0086] A button battery for powering the brush;

[0087] A metal bracket for fixing the button battery;

[0088] An electronic taste chip, electrically connected to the brush, for:

[0089] Record the change of current pulse;

[0090] Perceive and classify the tastes of different foods according to the current change;

[0091] A Bluetooth interface, communicatively connected to the electronic taste chip, for transmitting the sensing result to the knowledge inference unit.

[0092] Preferably, the rule-based food knowledge graph construction module is further configured to:

[0093] Mine the association relationship between ingredients and foods through a rule mining algorithm;

[0094] Add the association relationship to the food knowledge graph.

[0095] Preferably, the knowledge graph recommendation system based on graph neural network is further configured to:

[0096] Perform path search based on the triple information in the graph data;

[0097] Find the recommendation results of the maximum probability path and similar nodes.

[0098] Preferably, the intestinal flora detection module is based on the LAMP / CRISPR nucleic acid detection technology and uses LoRaWAN Internet of Things to realize real-time upload of sample information and real-time processing in the cloud.

[0099] A method for personalized dietary intervention for coronary heart disease patients, comprising the following steps:

[0100] S1. Obtain the basic human information, dietary intake information, blood lipid level and blood glucose data, and construct a knowledge base;

[0101] S2. Obtain historical data, preprocess the historical data, then obtain attention weights, apply the attention weights to each vector in the data block, generate an attention representation for the input data, and fuse the updated vectors to output the attention-enhanced features;

[0102] S3. Express the extracted features in a neural network, use neural network technology to realize the non-linear dependence relationship between features, and combine a convolutional neural network to extract a feature mapping function, and generate a dietary recommendation intervention plan according to the user's diet data and the knowledge base;

[0103] S4. Weigh the user's dinner plate through the intelligent tableware weighing detection function, obtain the real-time data of the food weight for uploading, and collect data on the food ingested by the user through image recognition technology to generate a food intake list;

[0104] S5. By real-time monitoring the state of the food ingested by the user, match the dinner plate data with the food intake list, and update the food intake list in real time;

[0105] S6. Use an electronic taste sensing module to detect sodium and potassium in real time and determine whether the intake of sodium and potassium is excessive;

[0106] S7. Through a blood drug concentration detection module, conduct real-time dynamic monitoring of the blood drug concentration by antiplatelet drugs and food;

[0107] S8. Through the intestinal flora detection module, blockchain technology and Internet of Things technology, record the intestinal flora environment and traceability, and combine with augmented reality technology to construct a virtual food plate and present a personalized diet recommendation and intervention plan;

[0108] S9. Through a neural network learning algorithm, train and generate a data set and a knowledge base, and based on the user's real-time data and the knowledge base, generate a personalized diet recommendation and intervention plan, and adjust the food intake list according to the personalized diet recommendation and intervention plan;

[0109] S10. Update and adjust the food intake list according to the real-time update of the food intake list. At the same time, through a neural network learning algorithm, combine the user's real-time data and the knowledge base, generate a personalized diet recommendation and intervention plan, and adjust the food intake list according to the personalized diet recommendation and intervention plan;

[0110] S11. Combine the intestinal flora data provided by the intestinal flora detection module, the health knowledge base data and the basic human data to generate a personalized diet recommendation and intervention plan through the knowledge inference unit;

[0111] S12. Use a rapid detection method for intestinal flora data, generate a virtual recommendation and intervention plan through augmented reality technology, use an electronic taste sensing module to detect sodium and potassium in real time, and use a blood drug concentration detection module to monitor the blood drug concentration in real time, determine whether the intake of sodium and potassium is excessive, and at the same time determine whether the intake of sodium and potassium is excessive.

[0112] Specifically, the beneficial effects of the present invention are mainly reflected in the following aspects:

[0113] First of all, the present invention constructs a multi-source and real-time data acquisition network. Through the collaborative work of multiple hardware devices such as an intelligent tableware weighing unit, an image recognition module, and an electronic taste sensing module, the system can automatically and accurately record each eating behavior of the patient, including information such as the type, weight, and nutritional components of the food. This non-intrusive data acquisition method greatly improves the accuracy and integrity of the data, and at the same time reduces the usage burden of the patient.

[0114] Secondly, the present invention adopts an improved feature-embedding Transformer network model, effectively solving the characteristics of high-dimensional and strong temporal characteristics of health data. This model can capture the potential associations between patients' long-term dietary behaviors and cardiovascular health indicators, providing a more reliable theoretical basis for dietary intervention. The self-attention mechanism of the model can also automatically identify important dietary events, such as the impact of high-fat food intake on blood lipid levels, thus realizing more accurate risk warnings.

[0115] Furthermore, the present invention introduces a rule-based food knowledge graph and a graph neural network recommendation system, organically combining professional nutritional knowledge with patients' personal characteristics. This not only improves the scientificity and personalization degree of the recommendation results, but also takes into account the impacts of factors such as drug-food interactions and gut microbiota, ensuring the safety and effectiveness of the recommended solutions. The dynamic update mechanism of the knowledge graph also enables the system to continuously learn and adapt to new medical discoveries and patient feedback.

[0116] In addition, the augmented reality module of the present invention provides intuitive and interactive dietary guidance for patients. Through the visual display and real-time feedback of virtual food plates, patients can more intuitively understand and execute dietary recommendations, greatly improving the compliance of the intervention plan. This innovative interaction method also enhances patients' sense of participation and achievement, contributing to the cultivation of long-term healthy eating habits.

[0117] Finally, the system of the present invention has strong scalability and adaptability. Through modular design, the system can be conveniently integrated with new sensing devices or analysis algorithms. At the same time, the dynamic adjustment mechanism based on online learning enables the system to continuously optimize its recommendation strategies to adapt to changes in patients' health conditions and dietary preferences.

[0118] Generally speaking, through the organic combination of innovative technologies such as multi-dimensional data collection, deep learning analysis, knowledge graph reasoning, and augmented reality interaction, the present invention constructs a comprehensive, accurate, and easy-to-use personalized dietary intervention system for coronary heart disease patients. This system can not only significantly improve the effect of dietary intervention and improve patients' cardiovascular health conditions, but also provide a new paradigm for the personalized and intelligent development in the field of chronic disease management. Brief Description of the Drawings

[0119] Figure 1 is the main module and data flow diagram of the overall system of the present invention;

[0120] Figure 2 is the internal logic diagram of the intelligent tableware weighing unit of the present invention;

[0121] Figure 3 is the internal structure diagram of the knowledge reasoning unit of the present invention;

[0122] Figure 4 This is the structural diagram of the electronic taste sensing module of the present invention; Detailed implementation manners

[0123] Please refer to the appendix Figures 1-4 The present invention provides a personalized dietary intervention system and method for coronary heart disease patients, aiming to provide accurate and personalized dietary intervention plans for coronary heart disease patients. The following will describe the detailed implementation manners of the present invention.

[0124] The personalized dietary intervention system for coronary heart disease patients of the present invention includes an intelligent tableware weighing unit 1, an image recognition module 2, a knowledge reasoning unit 3, an improved feature embedding Transformer network model 4, an electronic taste sensing module 5, a drug-food interaction database 6, a rule-based food knowledge graph construction module 7, a knowledge graph recommendation system 8 based on a graph neural network, an intestinal flora detection module 9, and a traceability module 10.

[0125] The intelligent tableware weighing unit 1 is used to obtain food weight data in real time and transmit it to the server. In a preferred embodiment of the present invention, the intelligent tableware weighing unit 1 adopts a high-precision pressure sensor, and its measurement accuracy can reach ±0.1 g. This high-precision measurement can ensure that the system obtains accurate food intake data, thereby providing a reliable basis for subsequent analysis and recommendation.

[0126] The image recognition module 2 is communicatively connected to the intelligent tableware weighing unit 1, and is used to obtain food images and classify and recognize the food images based on a pre-trained neural network to obtain food type information. The present invention adopts the convolutional neural network (CNN) technology in deep learning, especially the ResNet-50 architecture, to achieve high-accuracy food recognition. The ResNet-50 model is pre-trained on the ImageNet dataset and then fine-tuned on a self-built dataset containing more than 10,000 common foods. This method enables the image recognition module 2 to recognize most of the foods that may appear in the daily diet of coronary heart disease patients, and the recognition accuracy rate can reach more than 95%.

[0127] The knowledge reasoning unit 3 is communicatively connected to the image recognition module 2, and is used to receive food type information and food weight data, generate a food intake list, and analyze the food intake list according to a preset knowledge base to generate a personalized dietary intervention plan. The knowledge reasoning unit 3 adopts a method combining a rule-based expert system and machine learning. Among them, the expert system includes dietary guidance rules jointly formulated by cardiovascular specialists and nutrition experts, and the machine learning algorithm can continuously optimize the recommendation strategy according to the patient's historical data and feedback.

[0128] The improved feature-embedding Transformer network model 4 is communicatively connected to the knowledge inference unit 3, and is used to receive the dietary behavior data, blood glucose data, and blood lipid data of a patient, construct an association model of the patient's dietary behavior - blood lipid and blood glucose monitoring results including time dimension information, and identify the diet - cardiovascular event association pattern. The present invention improves the standard Transformer model by introducing a time embedding layer to capture the temporal characteristics of the data. The mathematical expression of the model is as follows:

[0129]

[0130] Among them, Q, K, and V respectively represent the query, key, and value matrices, and d k is the dimension of the key vector. The time embedding is added in the following way:

[0131]

[0132] Among them, pos is the position in the sequence, i is the dimension index, and d m odel is the dimension of the model.

[0133] The electronic taste sensing module 5 is communicatively connected to the knowledge inference unit 3, and is used to monitor the sodium and potassium intake of the patient in real time, and send the data including the sodium and potassium intake to the knowledge inference unit 3. The present invention adopts an innovative wearable electronic taste sensor, and its working principle is based on the electrochemical impedance spectroscopy technology. The sensor can detect the concentrations of sodium ions and potassium ions in saliva, with a measurement range of 0 - 200 mmol / L and an accuracy of ±1 mmol / L. This high-precision real-time monitoring enables the system to timely detect the possible excessive sodium and potassium intake of the patient, so as to provide timely intervention suggestions.

[0134] The drug - food interaction database 6 is communicatively connected to the knowledge inference unit 3, and is used to store the rule set related to antiplatelet drugs and food ingredients, and respond to the query request of the knowledge inference unit 3 to provide drug - food interaction information. The database contains the interaction information between more than 1000 common foods and 50 commonly used cardiovascular drugs, and the data is sourced from authoritative medical literature and clinical research results.

[0135] The rule-based food knowledge graph construction module 7 is communicatively connected to the knowledge inference unit 3 and the drug - food interaction database 6, and is used to construct a food knowledge graph based on food ingredient entities, drug ingredient entities, food - drug interaction rules, and knowledge triples. This module constructs the knowledge graph using the ontology method and defines concepts and relationships using the OWL (Web Ontology Language). The knowledge graph contains more than 1 million entities and 10 million relationships, covering knowledge in multiple fields such as food, nutritional components, drugs, and diseases.

[0136] The knowledge graph recommendation system 8 based on graph neural network is connected to the rule-based food knowledge graph construction module 7 and the knowledge reasoning unit 3 for generating personalized recipes based on the recommendation results of the food knowledge graph and sending the personalized recipes to the knowledge reasoning unit 3. This system uses the graph attention network (GAT) to learn the representation of entities and relationships, and its core formula is as follows:

[0137]

[0138] Among them, α ij is the attention coefficient of node j to node i, h i and h j is the node feature, W is the weight matrix, a is the attention vector, is the sum of neighbors of node i.

[0139] The intestinal flora detection module 9 is connected to the knowledge reasoning unit 3 for the extraction and qualitative analysis of microbial genes through a rapid detection kit, and the detection results are sent to the knowledge reasoning unit 3. This module uses the latest metagenomics technology, which can complete the processing and analysis of samples within 2 hours, and the detection accuracy reaches 95% at the genus level and 85% at the species level. This rapid and accurate intestinal flora detection provides an important basis for the formulation of personalized dietary intervention programs.

[0140] The traceability module 10 is connected to the knowledge reasoning unit 3 for monitoring and traceability verification of the entire food production process through blockchain technology, and sending the traceability information to the knowledge reasoning unit 3. This module uses the Hyperledger Fabric framework to build a private permissioned chain network to ensure data security and privacy. Each food batch has a unique blockchain identifier that records the entire process from production, processing to distribution.

[0141] The intelligent tableware weighing unit 1 includes a pressure sensor 11, an RFID chip 12 and a data processing unit 13. The pressure sensor 11 is arranged at the bottom of the dinner plate to record the weight information of a single tableware. In one embodiment of the present invention, a thin film strain gauge pressure sensor is used, the sensitivity of which can reach 0.1% FS (full scale) and the linearity is better than 0.1% FS. This high-precision pressure sensor can accurately capture the slight changes in the weight of food, providing a guarantee for accurately calculating the food intake.

[0142] The RFID chip 12 is used to record the plate ID and cooking information. The present invention uses an NFC (Near Field Communication) chip operating in the 13.56MHz frequency band, with a storage capacity of 4KB, which is sufficient to store key information such as plate ID, cooking method, main ingredients, etc. This information is essential for the system to accurately identify and analyze food.

[0143] The data processing unit 13 is connected to the pressure sensor 11 and the RFID chip 12 for receiving weight information, plate ID and cooking information, packaging these information to generate lunch box information, encrypting and signing the lunch box information, and uploading the encrypted and signed lunch box information to the server. The data processing unit 13 uses an ARM Cortex-M4 microcontroller with a main frequency of up to 80MHz, a built-in hardware encryption engine, and supports the AES-256 encryption algorithm. This ensures the security and integrity of the data during transmission.

[0144] The knowledge reasoning unit 3 includes a neural network training module 31, a data preprocessing module 32, an attention mechanism modeling module 33, a neural network implementation module 34, an attention mechanism-based knowledge graph analysis module 35 and a knowledge base management module 36.

[0145] The neural network training module 31 is used to train the neural network for food recognition. This module adopts a transfer learning strategy and fine-tunes the pre-trained ResNet-50 model on a self-built dataset containing more than 100,000 food images. The Adam optimizer is used in the training process, the learning rate is initially set to 0.001, and the cosine annealing strategy is used for dynamic adjustment. This method can obtain a highly accurate food recognition model in a shorter training time.

[0146] The data preprocessing module 32 is used to remove outliers and noise data, normalize and bin the remaining data, and obtain data blocks in batches. In outlier detection, the present invention adopts an improved local outlier factor (LOF) algorithm, the core idea of ​​which is to calculate the local density deviation of the data point. For a data point p, its LOF is defined as follows:

[0147]

[0148] Among them, N k (p) is the set of k nearest neighbors of p, lrd k (p) is the local reachable density of p. The present invention sets the LOF threshold to 1.5, that is, when the LOF value of a data point is greater than 1.5, it is regarded as an outlier.

[0149] The attention mechanism modeling module 33 is communicatively connected to the data preprocessing module 32, and is configured to receive the data blocks output by the data preprocessing module 32, divide each data block into multiple vectors, then splice them and input them into a neural network layer to generate attention weights, and apply the attention weights to each vector in the data block simultaneously. Finally, the updated vectors are fused to output the features with enhanced attention. This module adopts a multi-head self-attention mechanism, and its mathematical expression is as follows:

[0150] MultiHead(Q,K,V)=Concat(head 1 ,...,head h )W O ,

[0151] where,

[0152]

[0153] and W O are learnable parameter matrices. In the present invention, 8 attention heads are set, and the dimension of each head is 64. Such a setting can effectively capture the features of multiple aspects of the data in practice.

[0154] The neural network implementation module 34 is communicatively connected to the attention mechanism modeling module 33, and is configured to utilize neural network technology to realize the non-linear dependence relationship between features, and combine a convolutional neural network to extract a feature mapping function to realize the expression of the data after attention mechanism modeling. This module adopts a structure of residual connection and layer normalization to alleviate the problem of gradient disappearance in the training of deep networks. The mathematical expression of the residual connection is as follows:

[0155] y=F(x,W i )+x,

[0156] where, x is the input, and F(x,W i ) represents the residual mapping. The calculation formula of layer normalization is:

[0157]

[0158] where, μ and σ are the mean and standard deviation respectively, α and β are learnable scaling and bias parameters, and ∈ is a small constant for numerical stability.

[0159] The knowledge graph analysis module 35 based on the attention mechanism is communicatively connected to the neural network implementation module 34, and is used to calculate the similarity of each entity in the knowledge base, sort the calculation results, infer implicit information based on the relationships between entities, and obtain a personalized diet recommendation and intervention plan for coronary heart disease patients. This module uses the Graph Attention Network (GAT) to learn the representations of entities and relationships, and its core formula is as described above. During the reasoning process, a path-based reasoning method is used, and a path confidence score is defined as follows:

[0160]

[0161] p is a path that contains n relationships r i and n + 1 entities e i . P(r i |e i-1 ,e i ) is the probability of relationship r i given the previous and subsequent entities. This method can effectively utilize the structural information in the knowledge graph and generate more reasonable recommendation results.

[0162] The knowledge base management module 36 is communicatively connected to the knowledge graph analysis module 35 based on the attention mechanism, and is used to store the generated diet recommendation and intervention plan, perform learning and training on the diet recommendation and intervention plan, and use the learning results as an extension and supplement to the knowledge graph. This module adopts an incremental learning strategy, which can continuously integrate new information without losing the existing knowledge. Specifically, the Elastic Weight Consolidation (EWC) algorithm is used to balance the learning of new and old knowledge. The loss function of EWC is defined as follows:

[0163] L(θ) = L B (θ) + λΣ i F i (θ i - θ A,i ) 2 ,

[0164] where L B (θ) is the loss of the new task, θ A,i is the parameter of the old model, F i is the Fisher information matrix, and λ is a hyperparameter that weighs the importance of the new and old tasks. In the present invention, λ is set to 0.4, and this value shows a good balancing effect in the experiment.

[0165] Through the collaborative work of the above-mentioned various modules, the personalized dietary intervention system for coronary heart disease patients of the present invention can provide accurate and personalized dietary intervention plans for patients. The working process of the system is roughly as follows:

[0166] First, when a patient uses the intelligent tableware for dining, the intelligent tableware weighing unit 1 will record the weight change of the food in real time. Meanwhile, the image recognition module 2 will capture and identify the type of food. This information will be sent to the knowledge inference unit 3 for processing.

[0167] The knowledge inference unit 3 will combine the patient's historical data, current physiological indicators (such as blood glucose, blood lipid levels, etc.) and medication usage, and use the improved feature-embedded Transformer network model 4 to analyze the association between the patient's dietary behavior and cardiovascular health. Meanwhile, the electronic taste sensing module 5 will monitor the patient's sodium and potassium intake in real time, providing an important reference for the intervention plan.

[0168] During the process of generating the intervention plan, the system will query the drug-food interaction database 6 to avoid possible adverse reactions. The rule-based food knowledge graph construction module 7 and the knowledge graph recommendation system 8 based on the graph neural network will work together to generate personalized recipe suggestions according to the patient's specific situation.

[0169] In addition, the microbiome data provided by the gut microbiota detection module 9 will also be taken into consideration because the gut microbiota is closely related to cardiovascular health. The traceability module 10 ensures the traceability of the food ingredients, increasing the patient's confidence in the recommended plan.

[0170] The core advantage of the entire system lies in its high degree of personalization and real-time nature. By continuously monitoring and analyzing the patient's dietary behavior, physiological indicators and environmental factors, the system can timely adjust the intervention plan to achieve precise dietary management. For example, if the system detects that the patient's sodium intake is close to the daily recommended upper limit (usually 2300 mg / day), it will immediately give suggestions to reduce the intake of high-sodium foods.

[0171] Another important feature is the learning ability of the system. The knowledge base management module 36 can continuously learn from new data and update the knowledge graph, making the recommendation system become more and more intelligent and personalized as the usage time increases.

[0172] In practical applications, this system has significantly improved the dietary management effect of coronary heart disease patients. In a 6-month clinical trial, the patient group using this system had an average 15% reduction in LDL cholesterol level, a 25% increase in the blood pressure control compliance rate, and a 40% increase in patient compliance compared with the control group. These results fully demonstrate the potential of this invention in improving the quality of life and prognosis of coronary heart disease patients.

[0173] Generally speaking, the personalized dietary intervention system and method for coronary heart disease patients provided by the present invention integrate a variety of advanced technologies to achieve precise, real-time, and personalized dietary intervention, providing a new and powerful tool for the prevention and treatment of coronary heart disease.

[0174] In a preferred embodiment of the present invention, the personalized dietary intervention system for coronary heart disease patients further includes a blood drug concentration detection module 14 and an augmented reality module 15. The introduction of these two modules further enhances the functionality and user experience of the system.

[0175] The blood drug concentration detection module 14 is communicatively connected to the knowledge inference unit 3 and is used to monitor the interaction process between antiplatelet drugs and food in real time through a chip based on electrochemical sensor technology, realizing real-time dynamic monitoring of blood drug concentration and sending the monitoring results to the knowledge inference unit 3. The present invention adopts a method combining microfluidic technology and electrochemical sensing to achieve rapid analysis of trace blood samples. Specifically, the blood drug concentration detection module 14 includes a microfluidic chip and an electrochemical detection unit. The microfluidic chip is made of PDMS (polydimethylsiloxane) material, in which a microchannel network is etched for the processing and separation of blood samples. The electrochemical detection unit adopts a three-electrode system, including a working electrode, a reference electrode, and an auxiliary electrode.

[0176] The surface of the working electrode is modified with an aptamer that specifically recognizes antiplatelet drugs. When the drug molecules in the blood sample bind to the aptamer, it will cause a change in the electrochemical signal at the electrode interface. This change can be detected by cyclic voltammetry (CV) or differential pulse voltammetry (DPV). Taking aspirin as an example, the linear range of its detection is 0.1 - 100 μg / mL, and the detection limit can reach 0.05 μg / mL. This high-sensitivity real-time monitoring ability enables the system to promptly detect abnormal changes in drug concentration, providing an important basis for adjusting the medication plan.

[0177] The augmented reality module 15 is communicatively connected to the knowledge inference unit 3 and is used to present the personalized diet recommendation intervention plan in an augmented reality manner through VR devices, record the food intake status of users through AR devices, construct virtual tableware through three-dimensional reconstruction technology, record food weight information, and assist users in dining. The core of this module is to present complex dietary intervention plans to users in an intuitive and interactive way, thereby improving user compliance and satisfaction.

[0178] In practical applications, the augmented reality module 15 will superimpose virtual food models and nutritional information on the user's field of view according to the personalized dietary plan generated by the knowledge reasoning unit 3. For example, when the user looks at a plate of food, the AR device will display key nutritional information such as the calorie content, fat content, and sodium content of the food in real time. At the same time, the system will also provide visual suggestions, such as prompting the user to reduce the intake by shrinking the virtual portion size of certain high-fat foods.

[0179] The application of 3D reconstruction technology enables the system to more accurately estimate the volume and weight of food. The present invention adopts structured light 3D scanning technology. By projecting a specific pattern of grating onto the object surface, and then capturing the deformed grating pattern through a camera, a three-dimensional model of the food is reconstructed using the principle of triangulation. The volume estimation error of this method can be controlled within ±5%, greatly improving the accuracy of food intake recording.

[0180] The input data of the improved feature embedding Transformer network model 4 of the present invention is mapped to a certain range through quantile transformation. After passing through the Embedding layer, the self-attention mechanism is used to learn the latent features of the input data. This design aims to process physiological index data of different scales and capture the complex relationships between the data.

[0181] Quantile transformation is a non-parametric method that can effectively handle outliers and data of different scales. Specifically, for each feature $x$, its value $x'$ after quantile transformation is calculated as follows:

[0182]

[0183] where, Φ -1 is the inverse function of the standard normal distribution, rank(x) is the rank of x among all samples, and n is the total number of samples. This transformation maps all features to the same distribution, which helps the training and generalization of the model.

[0184] After quantile transformation, the data will pass through an Embedding layer. In the present invention, the dimension of the Embedding layer is set to 512, and this dimension shows a good balance between expressive ability and computational efficiency in practice. The embedded data will be input into the self-attention mechanism, and its core formula is as described above.

[0185] It should be noted that the present invention introduces an innovation on the basis of the standard Transformer: time-aware positional encoding. Considering the temporal characteristics of dietary data and physiological index data, traditional positional encoding may not be sufficient to capture time information. Therefore, the present invention designs a new time-aware positional encoding method:

[0186]

[0187] Among them, t is the actual timestamp of the data point, and T is a scaling factor (set to 24 hours in this embodiment). This encoding method not only considers the relative positions in the sequence but also introduces absolute time information, enabling the model to better understand the time patterns of the data.

[0188] The electronic taste sensing module 5 of the present invention includes a brush 51, a button battery 52, a metal bracket 53, an electronic taste chip 54, and a Bluetooth interface 55. This modular design makes the electronic taste sensor easy to integrate and maintain.

[0189] The brush 51 is used to form an electric current on the tongue surface. In a preferred embodiment of the present invention, the brush 51 adopts a microelectrode array made of biocompatible material. The diameter of each microelectrode is about 100 microns, and the electrode spacing is 200 microns. This design can not only provide sufficient spatial resolution but also minimize the discomfort to the user.

[0190] The button battery 52 powers the brush 51. Considering safety and battery life, the present invention selects a 3V CR2032 lithium manganese battery with a capacity of 225 mAh, which can support the device to work continuously for more than 8 hours.

[0191] The metal bracket 53 is used to fix the button battery 52. The bracket adopts a medical-grade stainless steel material, which not only ensures strength but also has good biocompatibility.

[0192] The electronic taste chip 54 is electrically connected to the brush 51 and is used to record the change of current pulses, and sense and classify the tastes of different foods according to the current change. This chip integrates signal acquisition, processing, and analysis functions. Its core is a low-power ARM Cortex-M4F microcontroller with a main frequency of 80 MHz, built-in 12-bit ADC, and a sampling rate of up to 1 MSPS. The signal processing adopts a method combining wavelet transform and support vector machine (SVM), which can effectively distinguish different taste types.

[0193] The Bluetooth interface 55 is communicatively connected to the electronic taste chip 54 and is used to transmit the sensing result to the knowledge inference unit 3. This interface adopts Bluetooth 5.0 technology, with a transmission rate of up to 2 Mbps and a transmission distance of up to 100 meters. This ensures the real-time and reliable data transmission.

[0194] The rule-based food knowledge graph construction module 7 of the present invention is also used to mine the association relationship between ingredients and foods through a rule mining algorithm and add the association relationship to the food knowledge graph. This function enables the knowledge graph to continuously self-improve and expand.

[0195] In one embodiment of the present invention, the rule mining algorithm adopts an improved Apriori algorithm. The traditional Apriori algorithm has low efficiency in processing large-scale data, so the present invention optimizes it. Specifically, a hash-based pruning strategy and a transaction compression technique are introduced.

[0196] The core idea of the hash-based pruning strategy is to use a hash function to map the k-1 itemset to the corresponding bucket. When generating the k-item candidate set, only the items with a count not less than the minimum support in the corresponding bucket will be considered. This greatly reduces the number of candidate items to be processed. The hash function is defined as follows:

[0197]

[0198] where i j is the j-th item in the itemset, p j is a predefined prime number, and m is the number of buckets. In this embodiment, m is set to 1000, which shows a good balance in practice.

[0199] The transaction compression technique compresses the transaction database by deleting the items that do not meet the minimum support. This not only reduces the size of the database but also improves the efficiency of subsequent scans.

[0200] Through these optimizations, when the rule mining algorithm of the present invention processes a dataset containing millions of food-ingredient relationships, compared with the traditional Apriori algorithm, the running time is reduced by about 70%, and the memory usage is reduced by about 50%.

[0201] The mined association rules will be represented in the form of triples, such as (food, contains, ingredient), and then added to the existing food knowledge graph. To ensure the consistency and reliability of the knowledge graph, the newly added relationships will go through a verification process. The verification process includes a consistency check with the existing knowledge and verification based on external authoritative data sources (such as the USDA Food Composition Database). Only the relationships that pass the verification will be finally added to the knowledge graph.

[0202] In this way, the food knowledge graph of the present invention can be continuously enriched and improved, providing a more comprehensive and accurate knowledge basis for personalized dietary recommendations.

[0203] In another preferred embodiment of the present invention, the knowledge graph recommendation system 8 based on the graph neural network is also used to perform path search based on the triple information in the graph data to find the maximum probability path and the recommendation results of similar nodes. This path-based reasoning method can make full use of the structural information of the knowledge graph to generate more reasonable and interpretable recommendation results.

[0204] Specifically, the present invention adopts a path ranking algorithm based on the attention mechanism. This algorithm first finds all paths with a length not exceeding L (in this embodiment, L is set to 4) in the knowledge graph through bidirectional breadth-first search. Then, the attention mechanism is used to rank these paths. The importance score of a path is calculated as follows:

[0205]

[0206] where p is a path containing n relations, r i is the i-th relation, e i -1 and e i are the entities connected to r i , f is a scoring function, and α i is the attention weight. The scoring function f adopts the idea of the TransE model and is defined as:

[0207] f(r i ,e i-1 ,e i ) = -||e i-1 +r i -e i || 2 ,

[0208] The attention weight α i is calculated through a multi-layer perceptron (MLP):

[0209] α i = softmax(MLP([e i-1 ; r i ; e i )),

[0210] where [;] represents the vector concatenation operation.

[0211] Through this method, the system can find the most relevant paths, thereby generating more accurate recommendation results. For example, when recommending foods suitable for a certain coronary heart disease patient, the system may find such a path: (patient)-[has disease]->(coronary heart disease)-[needs to control]->(blood lipid)-[can be reduced]->(omega-3 fatty acid)-[contains]->(salmon). This not only gives the recommendation result (salmon) but also provides the reason for the recommendation, enhancing the interpretability of the system.

[0212] The intestinal flora detection module 9 of the present invention is based on the LAMP / CRISPR nucleic acid detection technology and uses LoRaWAN Internet of Things to realize the real-time upload of sample information and real-time processing in the cloud. This innovative technology combination makes the detection of intestinal flora fast, accurate and easy for remote monitoring.

[0213] The LAMP (Loop-mediated Isothermal Amplification) technique is an isothermal nucleic acid amplification technique. Compared with traditional PCR methods, LAMP has higher specificity and sensitivity, and the reaction conditions are simpler. In the present invention, the primer design for the LAMP reaction targets several key bacterial genera commonly found in the intestine, such as Bifidobacterium, Lactobacillus, and Bacteroides, etc. The reaction temperature is set at 65°C, and the reaction time is 30 minutes.

[0214] The CRISPR technique is used for nucleic acid detection. The present invention employs the Cas12a protein, which has collateral cleavage activity, that is, after recognizing a specific DNA sequence, it will non-specifically cleave the surrounding single-stranded DNA. By designing a fluorescent reporter probe, highly sensitive detection of specific bacterial species can be achieved. The detection limit can reach 10 copies / μL, greatly improving the accuracy of detection.

[0215] To achieve real-time uploading and processing of sample information, the present invention adopts the LoRaWAN (Long Range Wide Area Network) technology. LoRaWAN features long-distance transmission and low power consumption, making it very suitable for Internet of Things applications. In this system, the detection device transmits the detection results to the cloud server in real time through the LoRaWAN module. The transmission protocol uses AES-128 encryption to ensure the security of data transmission.

[0216] After receiving the data, the cloud server will immediately process and analyze it. The analysis algorithm adopts a machine learning-based method, which can identify patterns related to cardiovascular health from the intestinal microbiota composition. Specifically, the present invention uses the random forest algorithm to establish an association model between the intestinal microbiota composition and cardiovascular risk factors (such as blood pressure, blood lipid levels, etc.). The accuracy of the model reaches 85% on the test set, providing an important basis for formulating personalized dietary intervention programs.

[0217] The present invention also provides a method for personalized dietary intervention for coronary heart disease patients, including multiple steps from S1 to S12. This method comprehensively covers the entire process from data collection, analysis to the generation and implementation of the intervention program, reflecting the systematicness and innovation of the present invention.

[0218] In step S1, the system acquires basic human information, dietary intake information, blood lipid levels, and blood glucose data to construct a knowledge base. This step lays the foundation for subsequent personalized analysis. The construction of the knowledge base adopts an ontology method and uses OWL (Web Ontology Language) to define concepts and relationships. The knowledge base contains professional knowledge in multiple fields such as nutrition, medicine, and food science, providing theoretical support for precise dietary intervention.

[0219] Steps S2 and S3 involve data preprocessing and feature extraction. The innovation here is the introduction of a method that combines an attention mechanism and a convolutional neural network. The attention mechanism can capture the importance differences between different features, while the convolutional neural network is good at extracting local features. The combination of the two enables the system to better understand complex physiological and dietary data.

[0220] In the feature extraction process, the present invention adopts the design of multi-scale convolutional kernels. Specifically, convolutional kernels of three sizes, 1x1, 3x3, and 5x5, are used, and the number of convolutional kernels of each size is 64. This design enables the network to capture features of different scales simultaneously, improving the expression ability of the model. After the convolutional layer, a global average pooling layer is connected to convert the feature map into a vector of fixed length for subsequent processing.

[0221] Steps S4 to S6 involve real-time data collection and monitoring. The key here lies in the fusion and real-time processing of multi-source data. The present invention designs a data fusion algorithm based on Kalman filtering, which can effectively integrate data from intelligent tableware, image recognition modules, and electronic taste sensors. The state equation and observation equation of Kalman filtering are as follows:

[0222] x k =Ax k-1 +w k ,

[0223] x k =Hx k +v k ,

[0224] Among them, x k is the state vector at time k, z k is the observation vector, A is the state transition matrix, H is the observation matrix, w k and v k are the process noise and observation noise respectively. By adjusting these parameters, the system can achieve a balance between the accuracy and real-time performance of different data sources.

[0225] Steps S7 to S9 involve multi-dimensional health monitoring and intervention plan generation. The innovation here lies in combining traditional nutritional knowledge with modern artificial intelligence technology. For example, when generating a personalized diet recommendation intervention plan, the system not only considers traditional nutritional balance but also introduces an optimization algorithm based on deep reinforcement learning.

[0226] Specifically, the present invention models the dietary recommendation problem as a Markov decision process (MDP). The state space includes the patient's physiological indicators, current dietary intake, etc.; the action space is all possible food combinations; and the reward function is designed based on factors such as nutritional balance, taste preference matching degree, and improvement degree of health indicators. The Deep Q-Network (DQN) algorithm is used to learn the optimal recommendation strategy. The loss function of DQN is defined as follows:

[0227]

[0228] where r is the immediate reward, γ is the discount factor, and θ and θ - are the parameters of the current network and the target network, respectively.

[0229] Steps S10 to S12 involve the implementation and dynamic adjustment of the plan. The focus here is to establish a closed-loop feedback system to ensure that the intervention plan can be adjusted in a timely manner according to the actual situation of the patient. The present invention adopts an online learning algorithm based on Thompson sampling, which can continuously optimize the strategy during the recommendation process.

[0230] The core idea of Thompson sampling is to sample the return distribution of each action and select the action with the highest expected return. In this system, the return distribution of each food combination is modeled as a Beta distribution. After each recommendation, the corresponding Beta distribution parameters are updated according to the patient's feedback. This method can not only explore new possibilities but also utilize known effective strategies, achieving a good balance between exploration and exploitation.

[0231] Through the above steps, the method of the present invention can provide a highly personalized and dynamically adjusted dietary intervention plan for coronary heart disease patients, significantly improving the accuracy and effectiveness of the intervention. In a 12-month clinical trial, the average LDL cholesterol level of the experimental group patients using this method decreased by 18% compared with the control group, the blood pressure compliance rate increased by 30%, and the quality of life score increased by 25%. These results fully demonstrate the effectiveness and practical value of the method of the present invention.

[0232] It should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. Personalized dietary intervention system for patients with coronary heart disease, characterized by: include: Smart tableware weighing unit for: Get food weight data in real time; Transmitting food weight data to a server; The image recognition module is connected to the intelligent tableware weighing unit for: Get food images; Classifying and identifying the food images based on a pre-trained neural network to obtain food category information; A knowledge reasoning unit, in communication with the image recognition module, is used to: receiving the food type information and the food weight data; generating a food intake list based on the food type information and the food weight data; Analyze the food intake list according to a preset knowledge base to generate a personalized dietary intervention plan; The improved feature embedding Transformer network model is communicatively connected with the knowledge reasoning unit and is used for: Receive the patient's dietary behavior data, blood sugar data, and blood lipid data; Construct a correlation model between patient dietary behavior and blood lipid and blood glucose monitoring results that includes time dimension information; Identify patterns of diet-cardiovascular event associations; The electronic taste sensor module is communicatively connected to the knowledge reasoning unit and is used for: Real-time monitoring of patients’ sodium and potassium intake; sending the data including the sodium and potassium intake to the knowledge reasoning unit; A drug-food interaction database is communicatively connected to the knowledge reasoning unit and is used for: Stores a set of rules related to antiplatelet drugs and food ingredients; providing drug-food interaction information in response to a query request from the knowledge reasoning unit; A rule-based food knowledge graph construction module is communicatively connected with the knowledge reasoning unit and the drug-food interaction database, and is used for: Construct food knowledge graph based on food ingredient entities, drug ingredient entities, food-drug interaction rules and knowledge triples; A knowledge graph recommendation system based on a graph neural network, which is in communication connection with the rule-based food knowledge graph construction module and the knowledge reasoning unit, is used to: Generate personalized recipes based on the recommendation results of the food knowledge graph; Sending the personalized recipe to the knowledge reasoning unit; The intestinal flora detection module is connected to the knowledge reasoning unit for: Extraction and qualitative analysis of microbial genes can be achieved through rapid detection kits; Sending the detection result to the knowledge reasoning unit; The tracing module is connected to the knowledge reasoning unit and is used to: The entire food production process is monitored and traceability verified through blockchain technology; The traceability information is sent to the knowledge reasoning unit.

2. The personalized dietary intervention system for patients with coronary heart disease according to claim 1, characterized in that: The intelligent tableware weighing unit comprises: A pressure sensor is provided at the bottom of the dinner plate to record the weight information of a single tableware; RFID chip, used to record plate ID and cooking information; A data processing unit is communicatively connected to the pressure sensor and the RFID chip, and is used to: receiving the weight information, the plate ID and the cooking information; Packing the weight information, the plate ID and the cooking information to generate the meal box information; Encrypting and signing the lunch box information; Upload the encrypted and signed lunch box information to the server.

3. The personalized dietary intervention system for patients with coronary heart disease according to claim 1, characterized in that: The knowledge reasoning unit comprises: A neural network training module, used to train a neural network for food recognition; Data preprocessing module, used to: Eliminate outliers and noise data; Normalize and bin the remaining data; Obtain data blocks in batches; The attention mechanism modeling module is connected to the data preprocessing module for: Receiving a data block output by the data preprocessing module; Divide each piece of data into multiple vectors; The multiple vectors are concatenated and input into a layer of neural network to generate attention weights; Applying the attention weight to each vector in the data block simultaneously; Fuse the updated vectors and output the attention-enhanced features; A neural network implementation module is communicatively connected to the attention mechanism modeling module and is used to: Use neural network technology to realize nonlinear dependencies between features; Combine convolutional neural network to extract feature mapping function; Realize the expression of data after attention mechanism modeling; The knowledge graph analysis module based on the attention mechanism is connected to the neural network implementation module for: Calculate the similarity of each entity in the knowledge base; Sort the calculation results; Infer implicit information based on the relationships between entities; Obtain personalized dietary recommendations for intervention programs for patients with coronary heart disease; The knowledge base management module is connected to the attention mechanism-based knowledge graph analysis module and is used to: storing the generated dietary recommendation intervention plan; Conducting learning and training on the dietary recommendation intervention program; Use the learning results as an extension and supplement to the knowledge graph.

4. The personalized dietary intervention system for patients with coronary heart disease according to claim 1, characterized in that: Also includes: The blood drug concentration detection module is connected to the knowledge reasoning unit for: Real-time monitoring of the interaction between antiplatelet drugs and food through a chip based on electrochemical sensor technology; Realize real-time dynamic monitoring of blood drug concentration; Sending the monitoring results to the knowledge reasoning unit; An augmented reality module, in communication with the knowledge reasoning unit, is used to: Personalized dietary recommendation intervention plans are presented in an augmented reality way through VR devices; Record the user's food intake status through AR devices; Build virtual tableware through 3D reconstruction technology and record food weight information; Assist users with dining.

5. The personalized dietary intervention system for patients with coronary heart disease according to claim 1, characterized in that: The input data of the improved feature embedding Transformer network model is mapped to a certain range through quantile transformation, and the potential features of the input data are learned by using the self-attention mechanism after passing through the Embedding layer.

6. The personalized dietary intervention system for patients with coronary heart disease according to claim 1, characterized in that: The electronic taste sensor module comprises: Brushes, used to create an electric current on the tongue; A button battery, for supplying power to the brush; A metal bracket, used for fixing the button battery; The electronic taste chip is electrically connected to the brush and is used for: Record the changes in current pulses; sensing and classifying the taste of different foods according to the changes in the electric current; A Bluetooth interface is connected to the electronic taste chip for transmitting the perception result to the knowledge reasoning unit.

7. The personalized dietary intervention system for patients with coronary heart disease according to claim 1, characterized in that: The rule-based food knowledge graph construction module is also used for: Mining the association between ingredients and food through rule mining algorithms; The association relationship is added to the food knowledge graph.

8. The personalized dietary intervention system for patients with coronary heart disease according to claim 1, characterized in that: The knowledge graph recommendation system based on graph neural network is also used for: Perform path search based on triple information in graph data; Find the maximum probability path and the recommendation results of similar nodes.

9. The personalized dietary intervention system for patients with coronary heart disease according to claim 1, characterized in that: The intestinal flora detection module is based on LAMP / CRISPR nucleic acid detection technology and uses the LoRaWAN Internet of Things to achieve real-time uploading of sample information and real-time processing in the cloud.

10. A method for personalized dietary intervention for patients with coronary heart disease, based on the system according to any one of claims 1 to 9, characterized in that: The following steps are involved: S1. Obtain basic human information, dietary intake information, blood lipid levels and blood sugar data to build a knowledge base; S2. Obtain historical data, preprocess the historical data, and then obtain attention weights, apply the attention weights to each vector in the data block, generate attention representations for the input data, fuse the updated vectors, and output attention-enhanced features; S3, expressing the extracted features in a neural network, using neural network technology to realize nonlinear dependencies between features, and combining convolutional neural network to extract feature mapping functions, and generating a dietary recommendation intervention plan based on the user's dietary data and the knowledge base; S4. Weigh the user's plate through the intelligent tableware weighing detection function, obtain real-time data of food weight and upload it, and collect data on the user's food intake through image recognition technology to generate a food intake list; S5. Matching the plate data with the food intake list by real-time monitoring of the food intake status of the user, and updating the food intake list in real time; S6, using an electronic taste sensor module to detect sodium and potassium in real time to determine whether the sodium and potassium intake is excessive; S7, real-time dynamic monitoring of the effects of antiplatelet drugs and food on blood drug concentration through a blood drug concentration detection module; S8. Record the intestinal flora environment and traceability through the intestinal flora detection module, blockchain technology and Internet of Things technology, and build a virtual plate with augmented reality technology to present personalized dietary recommendation intervention plans; S9, training and generating a data set and a knowledge base through a neural network learning algorithm, and generating a personalized personalized diet recommendation intervention plan based on the user's real-time data and the knowledge base, and adjusting the food intake list according to the personalized diet recommendation intervention plan; S10, updating and adjusting the food intake list according to the real-time update of the food intake list, and at the same time, generating a personalized diet recommendation intervention plan by combining the user's real-time data and the knowledge base through a neural network learning algorithm, and adjusting the food intake list according to the personalized diet recommendation intervention plan; S11, combining the intestinal flora data provided by the intestinal flora detection module, the health knowledge base data and the basic human body data through the knowledge reasoning unit to generate a personalized diet recommendation intervention plan; S12. Using the rapid detection method of intestinal flora data, a virtual recommended intervention plan is generated through augmented reality technology, and the electronic taste sensor module is used to detect sodium and potassium in real time, and the blood drug concentration is monitored in real time through the blood drug concentration detection module to determine whether the sodium and potassium intake is excessive.

Images