Chronic disease early warning intervention system based on artificial intelligence

Abstract

The invention discloses a chronic disease early warning intervention system based on artificial intelligence, and relates to the technical field of chronic disease early warning, and the technical scheme is characterized in that multi-modal data of a patient is collected and preprocessed, and a feature vector is generated; performing feature extraction on the feature vector to generate a fusion feature vector; predicting a risk level by using a multi-layer perceptron classifier, and estimating a chronic disease occurrence probability; checking and correcting the risk prediction result according to the expert knowledge base; constructing a chronic disease association map to calculate the joint probability of complications, and generating a personalized intervention scheme of the patient; and performing feedback optimization on the intervention scheme through a reinforcement learning algorithm. According to the system and the method, multi-modal data and expert knowledge are fused, and an artificial intelligence technology is combined, so that chronic disease early-stage accurate early warning, cross-disease complication risk assessment and personalized intervention scheme generation are realized, better health management services can be brought to chronic disease patients, and the burden of chronic diseases on personal health and society is reduced.

Description

Technical Field

[0001] The present invention relates to the field of chronic disease early warning technology, and more specifically, to an artificial intelligence-based chronic disease early warning intervention system. Background Art

[0002] With the accelerated aging of the global population and changes in lifestyle, the incidence of chronic diseases (such as diabetes, hypertension, and cardiovascular disease) continues to rise. Chronic diseases not only cause long-term pain and a decline in quality of life for patients, but also impose a significant economic burden, profoundly impacting individuals, families, and society as a whole. Early detection, early warning, and intervention of chronic diseases are crucial for controlling disease progression, reducing the risk of complications, and improving patient survival and quality of life.

[0003] In recent years, the application of artificial intelligence technology in healthcare has made significant progress, providing new opportunities for early warning and intervention of chronic diseases. By using artificial intelligence algorithms to analyze and mine large amounts of medical data, it is possible to accurately predict the risk of chronic diseases and develop personalized intervention plans. However, current chronic disease early warning and intervention technologies are still in the process of continuous development and improvement, and the following problems still exist:

[0004] (1) Existing systems mainly rely on limited physiological indicator data (such as blood sugar, blood pressure, etc.) to assess chronic disease risks. However, limited indicators make it difficult to comprehensively and accurately assess individual chronic disease risks. Many chronic diseases may not have obvious clinical symptoms in the early stages or only show slight changes in physiological indicators, which are easily overlooked, causing patients to miss the best time for intervention; (2) Existing systems perform risk assessment and early warning for a single chronic disease (such as diabetes or hypertension), ignoring the correlation and comorbidity between chronic diseases, and are unable to comprehensively assess the risk of complications because they have not established a cross-disease correlation feature learning mechanism and cannot fully consider the interactions and impacts between different diseases, thereby affecting the accuracy and completeness of risk assessment; (3) Existing systems rely entirely on data-driven artificial intelligence algorithms (such as deep learning models) for risk prediction. Although they can to a certain extent mine the potential patterns in the data, there is a problem of black-box decision logic. Clinicians and patients have low trust in their prediction results, which largely limits the widespread application and application effect of these systems in actual medical scenarios.

[0005] Therefore, the present invention aims to provide an artificial intelligence-based early warning intervention system for chronic diseases to solve the above problems. Summary of the Invention

[0006] The purpose of the present invention is to provide an artificial intelligence-based early warning and intervention system for chronic diseases. By integrating multimodal data and expert knowledge and utilizing advanced artificial intelligence technologies such as deep learning, graph neural networks, and reinforcement learning, the present invention realizes the functions of early and accurate warning of chronic diseases, risk assessment of cross-disease complications, and generation of personalized intervention plans. It effectively solves the problems of incomplete evaluation and insufficient cross-disease correlation in existing chronic disease management systems, provides an innovative solution for chronic disease management, has good application prospects and promotion value, can bring better health management services to the majority of chronic disease patients, and reduce the burden of chronic diseases on personal health and society.

[0007] The above technical objectives of the present invention are achieved through the following technical solutions: an artificial intelligence-based chronic disease early warning intervention system, comprising a data collection module, a data processing module, a feature extraction module, a risk prediction module, a knowledge fusion module, an association learning module, a program intervention module, and an evaluation and optimization module;

[0008] The data collection module is used to collect multimodal data of the patient and transmit it to the data processing module;

[0009] The data processing module is used to receive the multimodal data transmitted by the data collection module, perform preprocessing, generate feature vectors in a unified format and transmit them to the feature extraction module;

[0010] The feature extraction module is used to receive the feature vector transmitted by the data processing module, and use the convolutional neural network and the long short-term memory network to extract features, and then use the attention mechanism to fuse the extracted features to generate a fused feature vector and transmit it to the risk prediction module;

[0011] The risk prediction module is used to receive the fused feature vector transmitted by the feature extraction module, use the multi-layer perceptron classifier to predict the patient's chronic disease risk level, and estimate the probability of the patient developing chronic diseases in the future through the logistic regression model, generate an initial risk prediction result and transmit it to the knowledge fusion module;

[0012] The knowledge fusion module is used to store expert knowledge related to chronic diseases and form an expert knowledge base. By receiving the initial risk prediction results transmitted by the risk prediction module, the risk prediction results are verified and corrected according to the rules in the expert knowledge base, and the disease risk prediction results are generated and transmitted to the association learning module and the program intervention module.

[0013] The association learning module is used to receive the disease risk prediction results transmitted by the knowledge fusion module, use the graph neural network to construct a chronic disease association map, mine the association characteristics between different diseases, and calculate the joint probability of various complications in patients, generate complication risk prediction results and transmit them to the program intervention module;

[0014] The intervention program module is used to receive the disease risk prediction results and complication risk prediction results transmitted by the knowledge fusion module and the association learning module, and generate personalized intervention programs for patients at different risk levels in combination with the rules of the expert knowledge base;

[0015] The evaluation and optimization module is used to receive the personalized intervention plan transmitted by the plan intervention module, perform multi-dimensional evaluation, generate intervention effects, and use the reinforcement learning algorithm to provide feedback optimization for the intervention plan.

[0016] The present invention is further configured as follows: the data processing module includes a data cleaning unit, a data standardization unit and a data fusion unit; the data cleaning unit is used to identify and correct abnormal data points, and at the same time use an interpolation method to fill in missing data; the data standardization unit is used to convert data from different sources and different dimensions into dimensionless relative values ​​so that the data are at the same order of magnitude; the data fusion unit is used to construct a data fusion framework, aligning the data according to time series based on individual patients to form a complete chronic disease risk feature vector for the patient.

[0017] The present invention is further configured as follows: the feature extraction module includes a vector extraction unit and a feature fusion unit; the vector extraction unit is used to extract features from feature vectors using a convolutional neural network and a long short-term memory network; the feature fusion unit is used to realize multimodal feature fusion using an attention mechanism, and automatically assign weights according to the importance of each modal feature to chronic disease risk prediction, highlighting key modal information and enhancing feature expression capabilities.

[0018] The present invention is further configured as follows: the risk prediction module includes a risk level prediction unit and a concurrent probability estimation unit; the risk level prediction unit is used to receive the feature vector after multimodal fusion, use a nonlinear activation function to transform the input features and extract high-order features, and use a softmax function to map the output to a probability distribution to predict the patient's chronic disease risk level; the concurrent probability estimation unit is used to use a logistic regression model to estimate the probability of a patient developing chronic disease complications within a specific time window in the future.

[0019] The present invention is further configured as follows: the knowledge fusion module includes an expert knowledge base unit and a rule reasoning unit; the expert knowledge base unit is used to store expert knowledge data related to chronic diseases and form an expert knowledge base; the rule reasoning unit is used to embed the rules in the expert knowledge base into the rule reasoning engine to achieve rapid matching and reasoning of expert rules.

[0020] The present invention is further configured as follows: the association learning module includes a chronic disease association unit and a complication assessment unit; the chronic disease association unit is used to construct a chronic disease association map, with different diseases as nodes and the association relationships between diseases as edges, and GNN is used to embed learning of nodes and edges to capture the cross-association characteristics of diseases; the complication assessment unit is used to construct a complication risk assessment model, combining the individual characteristics of the patient with the cross-disease association characteristics, and using Bayesian networks or integrated learning methods to calculate the joint probability of various complications in the patient.

[0021] The present invention also provides an artificial intelligence-based early warning intervention method for chronic diseases, comprising the following steps:

[0022] S1. Collect patients' physiological indicator data, lifestyle data, clinical diagnosis data, imaging data, and genetic data;

[0023] S2. Preprocess the collected data to generate feature vectors in a unified format;

[0024] S3, using convolutional neural networks and long short-term memory networks to extract features, and then using the attention mechanism to fuse the extracted features to generate a fused feature vector;

[0025] S4. Based on the fused feature vector, a multi-layer perceptron classifier is used to predict the patient's chronic disease risk level, and a logistic regression model is used to estimate the patient's probability of developing chronic diseases in the future.

[0026] S5. Verify and correct the risk prediction results based on the rules in the expert knowledge base;

[0027] S6. Use GNN to construct a chronic disease association map, mine the association features between different diseases, and calculate the joint probability of various complications in patients based on this;

[0028] S7. Generate personalized intervention plans for patients at different risk levels based on risk prediction results and expert knowledge base;

[0029] S8. Conduct multi-dimensional evaluation of intervention effects and provide feedback optimization of intervention plans based on reinforcement learning algorithms.

[0030] The present invention also provides an artificial intelligence-based early warning intervention device for chronic diseases, comprising at least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement an artificial intelligence-based early warning intervention method for chronic diseases.

[0031] The present invention also provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to be executed by the computer to implement an early warning intervention method for chronic diseases based on artificial intelligence.

[0032] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements an artificial intelligence-based early warning intervention method for chronic diseases.

[0033] In summary, the present invention has the following beneficial effects:

[0034] 1. In multimodal data fusion, the present invention integrates multimodal data such as patients' physiological indicators, lifestyle, clinical diagnosis, imaging, and genes to construct a comprehensive feature vector, breaking the limitation of traditional evaluation relying only on limited physiological indicators, achieving more accurate prediction of chronic disease risks, effectively capturing the potential risks of patients with no obvious symptoms or slight indicator changes in the early stage, avoiding missed diagnosis, and buying critical time for early intervention. At the same time, in cross-disease association analysis, by using graph neural networks to construct chronic disease association maps, the association characteristics between different diseases are explored, breaking the limitations of single disease assessment. When treating patients with multiple diseases, it can comprehensively assess the risk of complications, provide a scientific basis for multi-disease management, make risk assessment more in line with the actual complex conditions, help doctors plan comprehensive treatment plans in advance, and reduce the harm of multiple diseases.

[0035] 2. The present invention can integrate deep learning with expert knowledge. By combining deep learning technologies such as convolutional neural networks and long short-term memory networks to mine deep data features, and integrating clinical experience through expert knowledge bases and rule-based reasoning engines, it not only leverages the powerful data processing advantages of artificial intelligence, but also uses human expertise to correct and optimize prediction results, solving the black box problem of pure data-driven models, improving the accuracy and interpretability of predictions, increasing clinicians' trust in the system's prediction results, and increasing the adoption rate. At the same time, it can conduct quantitative assessments of complication risks. By using Bayesian networks or integrated learning methods to calculate the joint probability of various complications in patients, it provides doctors with a basis for quantitative risk assessment, assists in formulating comprehensive and personalized treatment strategies, optimizes medical decisions from the perspective of preventing complications, improves treatment efficacy and patient prognosis quality, and reduces the cost and burden of long-term treatment of chronic diseases.

[0036] 3. The present invention can formulate targeted intervention plans based on the patient's chronic disease risk level, covering multiple aspects such as lifestyle, drug prevention and monitoring plans, to meet the personalized needs of different patient groups and improve intervention compliance and effectiveness. Low-risk patients are mainly treated with basic lifestyle intervention, while medium-risk patients are combined with drug prevention on this basis. High-risk patients receive intensive intervention, combined with drug treatment and close monitoring, to achieve precision medicine and improve the quality of chronic disease management. At the same time, it can also continuously optimize the intervention plan based on the evaluation of the intervention effect by utilizing reinforcement learning algorithms, forming a data-driven closed-loop management mechanism. As the patient's condition changes and data accumulates, the system automatically adjusts the strategy to ensure that the intervention plan is always in the best state, adapting to the long-term management needs of chronic diseases, continuously improving the patient's health status, reducing the risk of disease deterioration, and providing patients with high-quality, sustainable health management services;

[0037] 4. The present invention collects and analyzes patient data and provides timely feedback to the expert knowledge base, thereby promoting the dynamic updating of expert knowledge, promoting the inheritance of medical experience and knowledge innovation, making the system keep pace with the times, and always keeping track of and applying the latest chronic disease diagnosis and treatment technologies and management concepts, thereby improving the professional level and service quality of the entire medical team and benefiting more patients. In addition, accurate early warning and personalized intervention help to identify high-risk patients in advance, so that limited medical resources can be tilted towards patients in urgent need of intervention, improve resource utilization efficiency, alleviate medical resource constraints, reduce the occurrence of complications, reduce patients' long-term treatment costs, reduce the operating burden on the medical system, and create greater social and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a schematic diagram of the module contents of an artificial intelligence-based chronic disease early warning intervention system in Example 1 of the present invention;

[0039] Figure 2 This is a schematic diagram of the process structure of an artificial intelligence-based chronic disease early warning intervention system in Example 1 of the present invention;

[0040] Figure 3 This is a data flow diagram of an artificial intelligence-based chronic disease early warning intervention system in Example 1 of the present invention;

[0041] Figure 4 This is a method flow chart of an artificial intelligence-based early warning intervention method for chronic diseases in Example 2 of the present invention. DETAILED DESCRIPTION

[0042] The following is combined with Figure 1-Figure 4 The present invention is described in further detail.

[0043] Example 1: An artificial intelligence-based chronic disease early warning intervention system includes a data collection module, a data processing module, a feature extraction module, a risk prediction module, a knowledge fusion module, an association learning module, a program intervention module and an evaluation optimization module; the modules work together to realize the full process function of chronic disease early warning intervention.

[0044] In this embodiment, the data collection module includes a data interaction unit and a format conversion unit; the data interaction unit is used to call the data interface of the medical system to obtain the patient's multimodal data, and the format conversion unit is used to organize and transmit the obtained multimodal data according to a unified data format; the data and its source are as follows:

[0045] Patient's physiological indicator data: By connecting with medical testing equipment (such as blood glucose meters, blood pressure monitors, blood lipid analyzers, weight scales, etc.), the patient's blood glucose, blood pressure, blood lipids, body mass index, waist circumference and other physiological indicator data can be automatically collected and uploaded to the system database at a set time frequency (such as daily or weekly).

[0046] Patient lifestyle data: By developing mobile applications (APPs) or integrating with wearable devices (such as smart bracelets and smart watches), we can obtain patients' diet records (including detailed information such as food types, intake, and nutritional ingredients), exercise status (exercise type, duration, intensity, and other parameters), smoking and drinking habits (smoking amount, drinking amount, frequency, etc.), and sleep quality (monitoring data such as sleep duration, deep sleep duration, and sleep efficiency) to ensure daily data updates.

[0047] Patients' clinical diagnostic data: Interacts with the electronic medical record system (EMR) in the hospital information system (HIS) to integrate patients' previous disease diagnosis results, family medical history, medication records, allergy history and other medical information, enabling real-time synchronous data updates.

[0048] Patient imaging data: For patients who need imaging examinations, access the medical image storage and communication system (PACS) to collect medical imaging data such as fundus imaging and cardiac ultrasound images, and store them in a professional imaging database according to the prescribed examination cycle (such as every six months or every year).

[0049] Patient's genetic data: Cooperate with genetic testing institutions to obtain patients' genetic testing reports, extract gene locus information related to chronic disease susceptibility (such as the TCF7L2 gene related to type 2 diabetes, etc.), and input it into the system as basic data at one time.

[0050] The various data sources of patients are organized and transmitted in a unified data format. Specific physiological indicator data are in structured CSV format, lifestyle data are stored in JSON format, clinical diagnostic data follow the HL7 standard format, imaging data are in DICOM format, and genetic data are imported in FASTQ or VCF format. Secure encryption protocols (such as HTTPS, SSL / TLS) are used during data transmission to ensure the confidentiality and integrity of data during transmission and prevent data leakage and tampering.

[0051] In this embodiment, the data processing module includes a data cleaning unit, a data standardization unit, and a data fusion unit; the data cleaning unit is used to identify and correct abnormal data points, and at the same time, use interpolation methods to fill in missing data; the data standardization unit is used to convert data from different sources and different dimensions into dimensionless relative values ​​so that the data are at the same order of magnitude; the data fusion unit is used to construct a data fusion framework, aligning the data in time series based on individual patients to form a complete chronic disease risk feature vector for the patient.

[0052] During the data cleaning process, a method based on a combination of statistical analysis and domain knowledge is used to identify and remove obviously erroneous data points, such as blood glucose values ​​that exceed the human physiological range (such as blood glucose values ​​below 0 or above 50mmol / L) or abnormal blood pressure values ​​(such as systolic blood pressure below 50mmHg or above 250mmHg). For missing data, the missing patterns and causes are analyzed, and different interpolation methods are used to fill them in. For example, for time series physiological indicator data, linear interpolation or moving average interpolation based on adjacent time points is used. For non-time series lifestyle data or clinical diagnosis data, the mean or mode of similar patient groups is used for filling. Then, a data quality assessment indicator system is established, and the quality of the cleaned data is regularly assessed to ensure the accuracy, completeness and consistency of the data.

[0053] In the data standardization process, the physiological indicator data of different sources and dimensions are converted using the Z-score standardization method. The formula is:

[0054]

[0055] Among them, X is the original data, μ is the overall mean of the indicator, and σ is the overall standard deviation. The data are converted into dimensionless relative values ​​so that they are on the same order of magnitude to facilitate subsequent analysis and modeling. Other types of data (such as lifestyle data and some numerical indicators in clinical diagnosis data) are also standardized according to actual conditions.

[0056] During the data fusion process, a data fusion framework is constructed to align multimodal data in time series based on individual patients. Specifically, on a daily basis, the physiological indicators and lifestyle data collected on that day are associated with the corresponding clinical diagnostic information and genetic data to form a complete chronic disease risk feature vector for the patient, providing a comprehensive data foundation for subsequent analysis. By adopting data fusion algorithms (such as principal component analysis (PCA) and linear discriminant analysis (LDA), etc.), feature extraction and dimensionality reduction processing are performed on multi-source data, reducing data dimensions while retaining key information, thereby improving data processing efficiency and model performance.

[0057] In this embodiment, the feature extraction module includes a vector extraction unit and a feature fusion unit; the vector extraction unit is used to extract features from feature vectors using a convolutional neural network and a long short-term memory network; the feature fusion unit is used to use an attention mechanism to realize multimodal feature fusion, and automatically assign weights according to the importance of each modal feature to chronic disease risk prediction, highlighting key modal information and enhancing feature expression capabilities.

[0058] Convolutional neural networks can process image data and genetic data. For image data, a multi-layer convolutional neural network structure is constructed, including convolution layers, pooling layers, and activation functions, to automatically extract key features in the image, such as diabetic retinopathy features such as microaneurysms and hemorrhage spots in fundus imaging. The mathematical expression of the convolution layer is:

[0059]

[0060] Among them, Y c,h,w is the value of the output tensor at channel c and position (h,w), is the input tensor at channel c′, position h+k h -1,w+k w The value at -1, The convolution kernel is at the output channel c, input channel c′, position k h ,k w The value at b c is the bias term, the bias value at the output channel c.

[0061] For genetic data, the gene sequence is converted into a grayscale image and then input into the CNN. The convolution operation is used to extract the potential characteristic patterns in the gene fragments, capturing genetic variation information related to chronic diseases, such as single nucleotide polymorphisms (SNPs). The long short-term memory network processes time series physiological indicator data and lifestyle data. LSTM, through its unique gating structure (input gate, forget gate, output gate) and memory unit, can effectively capture long-term dependencies in the data. The memory unit state update formula is:

[0062] it =σ(W i x t +U i h t-1 +b i )

[0063] f t =σ(W f x t +U f h t-1 +b f )

[0064] o t =σ(W o x t +U o h t-1 +b o )

[0065] c t =f t ⊙c t-1 +it⊙g(W c x t +U c h t-1 +b c )

[0066] h t =o t ⊙g(c t )

[0067] Among them, x t is the current time step input, h t is the hidden state, c t is the cell state, i t 、f t 、o t They are the activation values ​​of the input gate, forget gate, and output gate respectively. σ represents the sigmoid activation function, g represents the tanh activation function, and W i 、W f 、W o 、W c etc. are network weight parameters, which are obtained through training and learning, U i h、U f h、U c u、U o h are the corresponding hidden state weight matrices, b i 、b f 、b c 、b o is the bias term.

[0068] Through long-short-term memory networks, we can analyze the correlation between the dynamic changing trends of patients' physiological indicators over a period of time (such as blood sugar fluctuations) and lifestyle factors (such as diet and exercise), providing key features in the time dimension for chronic disease risk prediction.

[0069] In the feature fusion unit, the attention mechanism is used to achieve multimodal feature fusion. The weight is automatically assigned according to the importance of each modal feature to chronic disease risk prediction, highlighting key modal information and enhancing feature expression capabilities. The fusion formula is:

[0070] F=α1F1+α2F2+…+α n F n

[0071] Among them, F is the fused feature vector, F i represents the eigenvector of the i-th mode, α i is the attention weight of the corresponding modality, and α1+α2+…+α n =1, attention weight α i The calculation of is based on the correlation between each modality feature and the risk of chronic diseases, which is determined through training and learning.

[0072] A multimodal fusion network structure is constructed, and the image and gene features extracted by CNN and the physiological and lifestyle features extracted by LSTM are input into the fusion network. After operations such as the fully connected layer and activation function, a fused feature vector is generated as the input of the subsequent risk prediction module.

[0073] In this embodiment, the risk prediction module includes a risk level prediction unit and a concurrent probability estimation unit; the risk level prediction unit is used to receive the feature vector after multimodal fusion, use a nonlinear activation function to transform the input features and extract high-order features, and use a softmax function to map the output to a probability distribution to predict the patient's chronic disease risk level; the concurrent probability estimation unit is used to use a logistic regression model to estimate the probability of a patient developing chronic disease complications within a specific time window in the future.

[0074] In the risk level prediction unit, a multi-layer perceptron classifier is constructed, including an input layer, multiple hidden layers, and an output layer. The input layer receives the feature vector after multimodal fusion. The hidden layer uses a nonlinear activation function (such as ReLU) to transform the input features and extract high-order features. The output layer uses a softmax function to map the output to a probability distribution to predict the patient's chronic disease risk level (such as low risk, medium risk, and high risk). Assuming that the number of hidden layer neurons is N, the output probability calculation formula of the MLP classifier is:

[0075]

[0076] Among them, Z k is the input of the kth neuron in the output layer, K is the number of risk level categories, and the model parameters (weights and biases) are learned through training to make the prediction results as close as possible to the actual risk level.

[0077] In the concurrent probability estimation unit, based on the risk quantification and probability estimation of the fused features, a logistic regression model is used to estimate the probability of a patient developing chronic disease complications (such as diabetic nephropathy, cardiovascular disease, etc.) within a specific time window in the future (such as 1 year, 3 years, 5 years). The output formula of the logistic regression model is:

[0078]

[0079] Among them, β0 is the intercept term, β1, β2, ..., β n are model coefficients, obtained by maximum likelihood estimation based on historical data training, X1, X2, ..., X n To identify factors that affect the risk of chronic diseases (including key indicators in multimodal fusion features), provide clinicians and patients with a more intuitive and quantitative basis for risk assessment, and assist in formulating intervention measures.

[0080] In this embodiment, the knowledge fusion module includes an expert knowledge base unit and a rule reasoning unit; the expert knowledge base unit is used to store expert knowledge data related to chronic diseases and form an expert knowledge base; the rule reasoning unit is used to embed the rules in the expert knowledge base into the rule reasoning engine to achieve rapid matching and reasoning of expert rules.

[0081] In the expert knowledge base unit, knowledge collection and organization is achieved by inviting medical experts from various fields, such as endocrinology, cardiology, and nutrition, to collect and organize clinical guidelines, expert consensus, and empirical rules related to the diagnosis, treatment, prevention, and risk assessment of chronic diseases. For example, knowledge on the diagnostic criteria for diabetes (fasting blood glucose ≥7.0mmol / L or random blood glucose ≥11.1mmol / L, etc.) and the weight of the impact of different risk factors on the occurrence of diabetic complications (such as hypertension increasing the risk of cardiovascular disease by X times, and obesity increasing the risk of diabetic nephropathy by Y times, etc.) is compiled.

[0082] The knowledge representation approach uses a combination of semantic networks and ontologies to represent expert knowledge. Semantic networks are used to describe the relationships between chronic disease-related concepts (such as diseases, symptoms, and indicators), such as the "complication association" relationship between "hypertension" and "diabetes." Ontologies further clarify the attributes and classification of concepts. For example, "blood sugar" is divided into sub-concepts such as "fasting blood sugar" and "postprandial blood sugar," and their normal ranges, units, and other attributes are defined. This makes knowledge representation more rigorous and clear, facilitating systematic understanding and application. The storage structure of the expert knowledge base is constructed, including a combination of relational and graph databases to store structured knowledge rules and semantic network relationships.

[0083] In the rule-based reasoning unit, rules from the expert knowledge base are embedded into the rule-based reasoning engine. Once the system obtains a preliminary risk assessment result for a patient based on multimodal data mining, the rule-based reasoning engine verifies and corrects the result according to the preset expert rules. For example, if the data-driven model predicts that a patient has a moderate risk of diabetes, but according to the expert rule "If the patient has a clear family history of type 2 diabetes and a BMI ≥ 30, the risk level should be increased," the system will automatically adjust the risk level to high risk based on the patient's family history and BMI data, enhancing the accuracy and clinical rationality of the risk assessment. At the same time, it uses efficient reasoning algorithms (such as forward reasoning, backward reasoning, or hybrid reasoning algorithms) to achieve rapid matching and reasoning of expert rules, optimize the performance of the rule-based reasoning engine, and ensure that inference results can be given in a timely and accurate manner in the case of large-scale knowledge bases and high concurrent queries, without affecting the overall operational efficiency of the system.

[0084] In this embodiment, the association learning module includes a chronic disease association unit and a complication assessment unit; the chronic disease association unit is used to construct a chronic disease association map, with different diseases as nodes and the association relationships between diseases as edges, and GNN is used to embed learning of nodes and edges to capture the cross-association characteristics of diseases; the complication assessment unit is used to construct a complication risk assessment model, combining the individual characteristics of patients with cross-disease association characteristics, and using Bayesian networks or integrated learning methods to calculate the joint probability of various complications in patients.

[0085] In the chronic disease association unit, a chronic disease association graph is constructed based on a graph neural network (GNN). Different diseases (such as diabetes, hypertension, obesity, cardiovascular disease, etc.) are used as nodes in the graph, and the associations between diseases (such as causal relationships, comorbidity relationships, complications, etc.) are used as edges. GNN is used to embed the nodes and edges in the graph to capture the cross-association characteristics of diseases. The information aggregation formula of GNN is:

[0086] H (i+1) =f(H (i) ,A)

[0087] Among them, H (i) Represents the feature matrix of the i-th layer node, A is the adjacency matrix of the graph, and f is the neural network transformation function. Through iterative learning of multi-layer GNN, the node features are updated so that the features of each disease node contain its association information with other diseases, thereby providing a basis for cross-disease complication risk assessment.

[0088] In this example, the specific structure and parameters of the graph neural network, including graph convolutional layers and graph attention layers, are optimized based on the correlation characteristics of different diseases and the data scale. The GNN model is trained using disease correlation information from historical data as a supervisory signal, enabling the model to accurately learn cross-correlation feature representations of diseases.

[0089] In the complication assessment unit, a comprehensive complication risk assessment model based on cross-disease association features is constructed for patients who suffer from multiple chronic diseases (such as hypertension and obesity). The individual characteristics of the patient (features obtained by multimodal data mining) are combined with cross-disease association features, and the Bayesian network or ensemble learning method (such as random forest, gradient boosting tree, etc.) is used to calculate the joint probability of the patient developing various complications (such as diabetes combined with cardiovascular disease, obesity combined with diabetic nephropathy, etc.); taking the Bayesian network as an example, its nodes represent diseases and complications, and the edges represent the conditional probability relationship between them. By learning the conditional probability distribution in historical data, the probability of complications is inferred based on the patient's current disease status and characteristics. The joint probability calculation formula of the model is:

[0090]

[0091] Among them, D1, D2, ..., D n Indicates the disease the patient currently suffers from, C1, C2,…, C m Indicates possible complications, Parents (D i ) and Parents(C j ) represent disease D i and complications C j The parent node of (i.e., other diseases or characteristics directly related to it), P(D i |Parents(D i )) and P(C j |Parents(C j )) is the corresponding conditional probability, which is obtained based on historical data statistics to achieve a comprehensive and accurate assessment of the risk of cross-disease complications.

[0092] In this embodiment, the intervention program module includes a strategy generation unit and a personalized intervention unit; the strategy generation unit is used to divide patients into different risk levels based on the patient's chronic disease risk assessment results, and recommend different intervention measures according to different risk levels; the personalized intervention unit is used to recommend the most appropriate personalized intervention program for the patient based on the patient's individual characteristics, risk assessment results and intervention cases in the expert knowledge base.

[0093] In the strategy generation unit, the intervention strategy recommendation based on risk stratification is to divide patients into different risk levels according to the results of the chronic disease risk assessment (including the risk of a single disease and the risk of complications). For low-risk patients, basic lifestyle intervention measures are recommended, such as a reasonable diet (controlling carbohydrate intake, increasing dietary fiber, etc.), moderate exercise (150 minutes of moderate-intensity aerobic exercise per week) and regular monitoring (monitoring blood sugar, blood pressure and other indicators every 3-6 months); for medium-risk patients, on the basis of lifestyle intervention, combined with expert knowledge recommendations, drug prevention may be required (for example, according to expert consensus, for patients with prediabetes and high cardiovascular risk, aspirin can be considered for primary prevention), and the monitoring cycle is shortened (monitoring every 1-3 months); for high-risk patients or patients who have shown signs of complications, an intensive intervention plan is formulated, including combined drug therapy (for example, for patients with diabetes and hypertension, a combination of drugs that can both lower blood sugar and blood pressure is recommended). , such as ACEI / ARB drugs combined with metformin), personalized diet and exercise rehabilitation plans (developed by nutritionists and rehabilitation physicians based on the patient's specific situation) and close monitoring (monthly monitoring of multiple physiological indicators and complication-related indicators) to ensure timely adjustment of treatment plans. Specifically, in the process of generating intervention plans, the experience in the expert knowledge base is fully integrated. For example, for patients with specific types of diabetes (such as elderly patients with diabetes or patients with gestational diabetes), according to the special intervention principles recommended by experts for this population (such as elderly patients need to pay attention to the risk of hypoglycemia and adjust the drug dosage; pregnant patients need to strictly control the blood sugar target range while ensuring fetal nutrition, etc.), the general intervention plan is personalized adjusted to improve the pertinence and effectiveness of the plan to meet the needs of different patient groups.

[0094] In the personalized intervention unit, a machine learning-based intervention plan recommendation algorithm is constructed to recommend the most appropriate personalized intervention plan for patients based on their individual characteristics, risk assessment results, and intervention cases in the expert knowledge base. Collaborative filtering, content-based recommendation, or hybrid recommendation algorithms are used to comprehensively consider factors such as the similarity between patients and intervention plans, historical intervention effects, etc., to improve the accuracy and satisfaction of recommendations. The recommendation algorithm is continuously optimized and updated based on newly collected patient data and intervention effect feedback, and the algorithm parameters and model structure are constantly adjusted to enable the recommended plan to better adapt to individual differences and changes in patients' conditions, thereby improving the intervention effect.

[0095] In this embodiment, the evaluation and optimization module includes an effect evaluation unit and a feedback optimization unit; the effect evaluation unit is used to establish a multi-dimensional intervention effect evaluation indicator system; the feedback optimization unit uses the intervention plan as the action of the intelligent agent and the intervention effect as the reward signal. Through continuous trial and error learning, it adjusts the parameters and recommendation logic of the intervention strategy, so that the system can automatically generate a better personalized intervention plan based on individual differences of patients and dynamic changes in their condition.

[0096] In the effect evaluation unit, a set of intervention effect evaluation indicator systems covering multiple dimensions such as improvement of physiological indicators, change of lifestyle, reduction of complication incidence and improvement of patients' quality of life is established. Physiological indicators for improvement include blood glucose control rate (such as the proportion of patients with glycated hemoglobin (HbA1c) <7%), blood pressure control rate (such as the proportion of patients with systolic blood pressure <140 mmHg and diastolic blood pressure <90 mmHg), and the extent of reduction in blood lipid levels (such as the percentage of reduction in low-density lipoprotein cholesterol (LDL-C)). Lifestyle change indicators include the degree of dietary structure optimization (such as the proportion of increased daily vegetable and fruit intake and the proportion of reduced saturated fat intake), the proportion of improved exercise compliance (such as the proportion of days per week that meet the recommended amount of exercise), and the status of smoking cessation and alcohol restriction (such as the proportion of decreased smoking and drinking rates). The complication rate reduction indicator is a statistical analysis of the changes in the incidence of chronic disease complications, such as diabetic nephropathy and cardiovascular events, in patients within a certain period of time before and after the intervention (such as 1 or 2 years). The quality of life improvement indicator is an assessment of patients using a standardized quality of life scale (such as the SF-36 scale) to calculate the changes in patients' quality of life scores in terms of physiological function, psychological function, social function, etc.

[0097] In the feedback optimization unit, a reinforcement learning algorithm is used to optimize the intervention plan based on the results of the intervention effect evaluation. By treating the intervention plan as the agent's action and the intervention effect as the reward signal, the system continuously learns through trial and error, adjusting the parameters and recommendation logic of the intervention strategy. This enables the system to automatically generate a more optimized personalized intervention plan based on individual patient differences and the dynamic changes in the patient's condition. When designing the specific structure and parameters of the reinforcement learning algorithm, the state space (the patient's current condition, physiological indicators, lifestyle, and other characteristics), the action space (different combinations of intervention measures), the reward function (calculated based on a comprehensive analysis of multi-dimensional effect evaluation indicators), and the policy network (such as the deep Q network (DQN) or the policy gradient algorithm (PG)) are included. By training the reinforcement learning model, the recommended strategy of the intervention plan is optimized and the intervention effect is improved.

[0098] The evaluation results will be fed back to the expert knowledge base to provide a basis for updating and improving expert knowledge. Medical experts will be organized to review and update the expert knowledge base on a regular basis. Based on new clinical research results, treatment guidelines, and lessons learned from intervention practices, the content of the knowledge base will be adjusted and supplemented to form a closed-loop optimization mechanism driven by data and knowledge collaboration, continuously improve system performance and intervention effects, and promote the development of chronic disease management towards precision and personalization.

[0099] The hardware environment implemented in this embodiment is as follows:

[0100] Servers: High-performance cloud computing servers are used, equipped with multi-core CPUs, large memory capacities (e.g., 128GB or higher), and high-speed storage devices (e.g., SSDs, with capacity determined by data size) to meet the needs of large-scale data storage, processing, and model training. The number of servers can be elastically expanded based on the number of system users and concurrency.

[0101] Client: Supports a variety of client devices, including desktop computers, laptops, smartphones, and tablets. Client devices access the system through a browser or a dedicated APP. They are required to have a stable network connection (such as Wi-Fi or 4G / 5G network) and basic hardware configuration (such as a smartphone processor with a main frequency of not less than 1.5GHz and a memory of not less than 2GB).

[0102] The software environment is as follows:

[0103] Operating system: The server-side operating system uses the Linux operating system (such as CentOS or Ubuntu), which has the characteristics of high stability, security, and strong scalability; the client operating system supports mainstream operating systems such as Windows, macOS, Android, and iOS.

[0104] Database management system: Use relational databases (such as MySQL and PostgreSQL) to store structured data (such as basic patient information and clinical diagnostic data), and non-relational databases (such as MongoDB) to store unstructured or semi-structured data (such as imaging data and genetic data). Build a distributed database cluster to achieve efficient data storage, query, and management.

[0105] Data analysis and mining tools: Use Python and its related data science libraries (such as NumPy, Pandas, Scikit-learn, etc.) for data preprocessing, statistical analysis, and machine learning model training; use deep learning frameworks (such as TensorFlow, PyTorch) to build and train deep learning models (such as CNN, LSTM, GNN, etc.) to realize multimodal feature mining and risk prediction functions.

[0106] Knowledge graphs and reasoning tools: Use knowledge graph construction tools (such as Neo4j and OrientDB) to build and manage semantic networks and ontologies in expert knowledge bases to realize knowledge storage, query and reasoning functions; use rule engines (such as Drools and CLIPS) to realize the reasoning and execution of expert rules, and provide support for risk assessment and intervention plan generation.

[0107] Front-end development framework: We use front-end development frameworks such as Vue.js and React to build user interfaces, enabling system visualization and user interaction. We use technologies such as HTML5, CSS3, and JavaScript to develop responsive web pages and mobile application interfaces, ensuring a smooth display and operational experience across various client devices.

[0108] Back-end development framework: Use back-end development frameworks such as Spring Boot and Django to build the system's back-end services, implement business logic processing, data interface development, and interactive functions with the front-end. Adopt a RESTful API design style to define the system's data interface specifications and ensure efficient communication and data transmission between the front-end and back-end.

[0109] The data security and privacy protection process is as follows:

[0110] Data encryption: Patient data is encrypted during storage and transmission. The Advanced Encryption Standard (AES) is used to encrypt data storage, and the Secure Sockets Layer (SSL / TLS) is used to encrypt data transmission to ensure data confidentiality and integrity.

[0111] Access control: Establish a strict user authentication and authorization mechanism, assign different access rights to different user roles (such as patients, doctors, researchers, etc.), and restrict access to and manipulation of sensitive data. Use multi-factor authentication (such as username and password + mobile phone verification code) to improve system login security.

[0112] Data anonymization and de-identification: When conducting data analysis and mining, patient data is anonymized and de-identified to remove or replace the patient's personal identity information (such as name, ID number, contact information, etc.) to protect patient privacy.

[0113] Data backup and recovery: System data is regularly backed up, using a multi-copy backup strategy and storing backup data in different storage media and geographical locations to ensure timely data recovery in the event of data loss or damage. A data backup and recovery plan and process are established, and regular backup data verification and recovery drills are conducted to ensure the effectiveness of data backup and the reliability of recovery. The overall system utilizes a cloud computing deployment architecture, deploying server-side applications and databases on a cloud platform. The elastic computing, storage, and network resources provided by the cloud platform enable efficient system operation and rapid expansion. Clients access the system via the internet, eliminating the need to install complex software environments and requiring only a basic web browser or mobile application to use system functions.

[0114] Example 2: An artificial intelligence-based early warning intervention method for chronic diseases, comprising the following steps:

[0115] S1. Collect patients' physiological indicator data, lifestyle data, clinical diagnosis data, imaging data, and genetic data;

[0116] S2. Preprocess the collected data to generate feature vectors in a unified format;

[0117] S3, using convolutional neural networks and long short-term memory networks to extract features, and then using the attention mechanism to fuse the extracted features to generate a fused feature vector;

[0118] S4. Based on the fused feature vector, a multi-layer perceptron classifier is used to predict the patient's chronic disease risk level, and a logistic regression model is used to estimate the patient's probability of developing chronic diseases in the future.

[0119] S5. Verify and correct the risk prediction results based on the rules in the expert knowledge base;

[0120] S6. Use GNN to construct a chronic disease association map, mine the association features between different diseases, and calculate the joint probability of various complications in patients based on this;

[0121] S7. Generate personalized intervention plans for patients at different risk levels based on risk prediction results and expert knowledge base;

[0122] S8. Conduct multi-dimensional evaluation of intervention effects and provide feedback optimization of intervention plans based on reinforcement learning algorithms.

[0123] Example 3: An artificial intelligence-based chronic disease early warning intervention device, comprising at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the processor, the instructions being used to be executed by the processor to implement an artificial intelligence-based chronic disease early warning intervention method, the method comprising: collecting physiological indicator data, lifestyle data, clinical diagnostic data, imaging data, and genetic data of a patient; preprocessing the collected data to generate a feature vector in a unified format; using a convolutional neural network and a long short-term memory network for feature extraction, and then using an attention mechanism to fuse the extracted features to generate a fused feature vector; based on the fused feature vector, using a multi-layer perceptron classifier to predict the patient's chronic disease risk level, and estimating the probability of the patient developing a chronic disease in the future through a logistic regression model; verifying and correcting the risk prediction results according to the rules in the expert knowledge base; using a GNN to construct a chronic disease association map, mining the association features between different diseases, and based on this, calculating the joint probability of the patient developing various complications; generating personalized intervention plans for patients at different risk levels based on the risk prediction results and the expert knowledge base; performing a multi-dimensional evaluation of the intervention effect, and performing feedback optimization on the intervention plan based on a reinforcement learning algorithm.

[0124] Example 4: A computer-readable storage medium, which stores computer instructions, and the computer instructions are used to be executed by a computer to implement an artificial intelligence-based early warning intervention method for chronic diseases, the method comprising: collecting physiological indicator data, lifestyle data, clinical diagnosis data, imaging data and genetic data of patients; preprocessing the collected data to generate a feature vector in a unified format; using a convolutional neural network and a long short-term memory network to extract features, and then using an attention mechanism to fuse the extracted features to generate a fused feature vector; based on the fused feature vector, using a multi-layer perceptron classifier to predict the patient's chronic disease risk level, and using a logistic regression model to estimate the probability of the patient developing a chronic disease in the future; verifying and correcting the risk prediction results according to the rules in the expert knowledge base; using GNN to construct a chronic disease association map, mining the association features between different diseases, and based on this, calculating the joint probability of the patient developing various complications; generating personalized intervention plans for patients at different risk levels based on the risk prediction results and the expert knowledge base; conducting a multi-dimensional evaluation of the intervention effect, and feedback optimization of the intervention plan based on a reinforcement learning algorithm.

[0125] Example 5: A computer program product, comprising a computer program, which, when executed by a processor, implements an artificial intelligence-based early warning intervention method for chronic diseases, the method comprising: collecting physiological indicator data, lifestyle data, clinical diagnostic data, imaging data, and genetic data of patients; preprocessing the collected data to generate a feature vector in a unified format; using a convolutional neural network and a long short-term memory network to extract features, and then using an attention mechanism to fuse the extracted features to generate a fused feature vector; based on the fused feature vector, using a multi-layer perceptron classifier to predict the patient's chronic disease risk level, and estimating the probability of the patient developing a chronic disease in the future through a logistic regression model; verifying and correcting the risk prediction results according to the rules in the expert knowledge base; using GNN to construct a chronic disease association map, mining the association features between different diseases, and based on this, calculating the joint probability of the patient developing various complications; generating personalized intervention plans for patients at different risk levels based on the risk prediction results and the expert knowledge base; performing a multi-dimensional evaluation of the intervention effect, and performing feedback optimization on the intervention plan based on a reinforcement learning algorithm.

[0126] This specific embodiment is merely an explanation of the present invention and is not intended to limit the present invention. After reading this specification, those skilled in the art may make non-creative modifications to this embodiment as needed. However, as long as such modifications are within the scope of the claims of the present invention, they are protected by patent law.

Claims

1. An artificial intelligence-based chronic disease early warning intervention system, characterized by: It includes data collection module, data processing module, feature extraction module, risk prediction module, knowledge fusion module, association learning module, program intervention module and evaluation optimization module; The data collection module is used to collect multimodal data of the patient and transmit it to the data processing module; The data processing module is used to receive the multimodal data transmitted by the data collection module, perform preprocessing, generate feature vectors in a unified format and transmit them to the feature extraction module; The feature extraction module is used to receive the feature vector transmitted by the data processing module, and use the convolutional neural network and the long short-term memory network to extract features, and then use the attention mechanism to fuse the extracted features to generate a fused feature vector and transmit it to the risk prediction module; The risk prediction module is used to receive the fused feature vector transmitted by the feature extraction module, use the multi-layer perceptron classifier to predict the patient's chronic disease risk level, and estimate the probability of the patient developing chronic diseases in the future through the logistic regression model, generate an initial risk prediction result and transmit it to the knowledge fusion module; The knowledge fusion module is used to store expert knowledge related to chronic diseases and form an expert knowledge base. By receiving the initial risk prediction results transmitted by the risk prediction module, the risk prediction results are verified and corrected according to the rules in the expert knowledge base, and the disease risk prediction results are generated and transmitted to the association learning module and the program intervention module. The association learning module is used to receive the disease risk prediction results transmitted by the knowledge fusion module, use the graph neural network to construct a chronic disease association map, mine the association characteristics between different diseases, and calculate the joint probability of various complications in patients, generate complication risk prediction results and transmit them to the program intervention module; The intervention program module is used to receive the disease risk prediction results and complication risk prediction results transmitted by the knowledge fusion module and the association learning module, and generate personalized intervention programs for patients at different risk levels in combination with the rules of the expert knowledge base; The evaluation and optimization module is used to receive the personalized intervention plan transmitted by the plan intervention module, perform multi-dimensional evaluation, generate intervention effects, and use the reinforcement learning algorithm to provide feedback optimization for the intervention plan.

2. The artificial intelligence-based chronic disease early warning intervention system according to claim 1, characterized in that: The data processing module includes a data cleaning unit, a data standardization unit and a data fusion unit; the data cleaning unit is used to identify and correct abnormal data points, and at the same time use interpolation methods to supplement missing data; the data standardization unit is used to convert data from different sources and different dimensions into dimensionless relative values ​​so that the data are at the same order of magnitude; the data fusion unit is used to construct a data fusion framework, aligning the data in time series based on individual patients to form a complete chronic disease risk feature vector for the patient.

3. The artificial intelligence-based chronic disease early warning intervention system according to claim 1, characterized in that: The feature extraction module includes a vector extraction unit and a feature fusion unit; the vector extraction unit is used to extract features from feature vectors using a convolutional neural network and a long short-term memory network; the feature fusion unit is used to implement multimodal feature fusion using an attention mechanism, and automatically assign weights based on the importance of each modal feature to chronic disease risk prediction, highlighting key modal information and enhancing feature expression capabilities.

4. The artificial intelligence-based chronic disease early warning intervention system according to claim 1, characterized in that: The risk prediction module includes a risk level prediction unit and a concurrent probability estimation unit; the risk level prediction unit is used to receive the feature vector after multimodal fusion, use a nonlinear activation function to transform the input features and extract high-order features, and use a softmax function to map the output to a probability distribution to predict the patient's chronic disease risk level; The concurrent probability estimation unit is used to estimate the probability of a patient developing chronic disease complications within a specific future time window using a logistic regression model.

5. The artificial intelligence-based chronic disease early warning intervention system according to claim 1, characterized in that: The knowledge fusion module includes an expert knowledge base unit and a rule reasoning unit; the expert knowledge base unit is used to store expert knowledge data related to chronic diseases and form an expert knowledge base; the rule reasoning unit is used to embed the rules in the expert knowledge base into the rule reasoning engine to achieve rapid matching and reasoning of expert rules.

6. The artificial intelligence-based chronic disease early warning intervention system according to claim 1, characterized in that: The association learning module includes a chronic disease association unit and a complication assessment unit; the chronic disease association unit is used to construct a chronic disease association map, using different diseases as nodes and the association relationships between diseases as edges, and using GNN to embed nodes and edges to capture the cross-association characteristics of diseases; The complication assessment unit is used to construct a complication risk assessment model, combining the individual characteristics of the patient with cross-disease association characteristics, and using a Bayesian network or ensemble learning method to calculate the joint probability of the patient developing various complications.

7. An artificial intelligence-based chronic disease early warning intervention method, applied to the artificial intelligence-based chronic disease early warning intervention system according to any one of claims 1 to 6, characterized in that: The following steps are involved: S1. Collect patients' physiological indicator data, lifestyle data, clinical diagnosis data, imaging data, and genetic data; S2. Preprocess the collected data to generate feature vectors in a unified format; S3, using convolutional neural networks and long short-term memory networks to extract features, and then using the attention mechanism to fuse the extracted features to generate a fused feature vector; S4. Based on the fused feature vector, a multi-layer perceptron classifier is used to predict the patient's chronic disease risk level, and a logistic regression model is used to estimate the patient's probability of developing chronic diseases in the future. S5. Verify and correct the risk prediction results based on the rules in the expert knowledge base; S6. Use GNN to construct a chronic disease association map, mine the association features between different diseases, and calculate the joint probability of various complications in patients based on this; S7. Generate personalized intervention plans for patients at different risk levels based on risk prediction results and expert knowledge base; S8. Conduct multi-dimensional evaluation of intervention effects and provide feedback optimization of intervention plans based on reinforcement learning algorithms.

8. An artificial intelligence-based chronic disease early warning intervention device, characterized by: It includes at least one processor; and a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the artificial intelligence-based early warning intervention method for chronic diseases described in claim 7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, which are used to be executed by the computer to implement the artificial intelligence-based early warning intervention method for chronic diseases described in claim 7.

10. A computer program product, characterized in that: It includes a computer program, which, when executed by a processor, implements the artificial intelligence-based early warning intervention method for chronic diseases described in claim 7.

Images