Digital chronic disease intelligent management platform based on AI model and multi-dimensional data fusion

Abstract

The invention relates to a digital chronic disease intelligent management platform based on an AI model and multi-dimensional data fusion, clinical diagnosis and treatment data, wearable equipment monitoring data, medication record data and environment monitoring data are acquired through a data acquisition module, and after standardized preprocessing is performed through a data fusion processing module, deep analysis is performed through an AI analysis module, and the data fusion processing module performs data fusion processing; in combination with medical knowledge of the knowledge base module, the intelligent decision-making module generates a personalized management scheme, and the personalized management scheme is implemented through the intervention execution module and the intelligent interaction module. Multi-dimensional health data are processed through an AI large model, a complex mode and an association relationship are automatically learned, and accurate disease prediction and risk assessment are realized; the pertinence of the scheme and the compliance of a patient are greatly improved; and real-time interaction and personalized guidance are provided, the participation degree and the self-management ability of the patient are effectively enhanced, and a benign health management cycle is formed.

Description

Technical Field

[0001] This invention relates to the field of computer science, and more specifically to a digital intelligent management platform for chronic diseases based on the fusion of AI models and multi-dimensional data. Background Technology

[0002] Existing digital chronic disease management platforms primarily rely on patients manually inputting information or using basic medical devices for simple data collection. For example, patients may need to periodically record blood glucose and blood pressure readings and input them into the platform; some platforms connect to basic devices like blood glucose meters and blood pressure monitors to automatically acquire some data. However, this data collection method has several limitations. Firstly, manual input is prone to errors and struggles to guarantee the timeliness and completeness of the data; patients may forget or fail to record data on time, leading to data gaps. Secondly, the data collected by basic medical devices is relatively limited, mostly focusing on routine physiological indicators and lacking information on other important factors that may influence the development of chronic diseases, such as patients' lifestyle habits, genetic information, and psychological state.

[0003] In the data analysis phase, existing platforms mostly employ traditional statistical methods or simple rule engines. These methods typically only provide superficial statistical analysis, such as calculating averages and variances, failing to delve into the underlying patterns and correlations within the data. For example, for patient blood glucose data, traditional statistical methods may only yield the average blood glucose level over a period of time, but cannot analyze the specific relationship between blood glucose fluctuations and the patient's diet and exercise habits. Simultaneously, simple rule engines can only make judgments based on pre-set, fixed rules. When encountering complex and ever-changing real-world situations, their accuracy and flexibility are significantly reduced, making it difficult to identify potential disease risk factors.

[0004] In terms of management plan development, existing platforms mostly offer standardized, general plans that lack precise adjustments for individual patients. These plans are often based on average data from a large number of patients or general medical guidelines, without fully considering each patient's unique circumstances, such as physical condition, lifestyle, and genetic background. For example, for two patients with the same type of diabetes, existing platforms may provide identical management plans in terms of diet, exercise, and medication. However, in reality, these two patients may have significant differences in their body's response to medication, exercise tolerance, and other aspects, and the standardized plan cannot meet their personalized needs.

[0005] Regarding improving patient interaction and engagement, existing platforms employ relatively simple and simplistic interaction mechanisms. They typically only send management suggestions and reminders to patients via SMS and email, lacking effective two-way communication and interaction. Patients often passively receive the content pushed by the platform, lacking the initiative and motivation to actively participate in health management. Furthermore, the platforms fail to fully utilize modern technologies, such as intelligent interactive interfaces and mobile applications, to enhance patient engagement and experience. Summary of the Invention

[0006] In view of the above problems, the present invention provides a digital chronic disease intelligent management platform based on the fusion of AI models and multi-dimensional data. By integrating patients' medical records, lifestyle habits, genetic information and other multi-dimensional data, the platform uses advanced AI models for in-depth analysis and prediction to provide patients with personalized health management plans and enhance their participation and self-management capabilities.

[0007] To achieve the above objectives, this invention provides a digital chronic disease intelligent management platform based on AI model and multi-dimensional data fusion, including a data acquisition module, a data fusion and processing module, an AI analysis module, a knowledge base module, an intelligent decision-making module, an intervention execution module, and an intelligent interaction module; the data acquisition module is used to collect multi-source heterogeneous health data of patients according to a preset cycle, and the multi-source heterogeneous health data includes clinical diagnosis and treatment data, wearable device monitoring data, medication record data, and environmental monitoring data;

[0008] The data fusion processing module is used to perform standardized preprocessing on multi-source heterogeneous health data. The preprocessing includes data cleaning, time alignment and feature normalization, and outputs time-aligned structured health data.

[0009] The AI ​​analysis module employs a pre-trained large model based on the Transformer architecture, which is integrated with a medical knowledge graph. It includes a physiological feature extraction unit, a behavior pattern recognition unit, and a risk prediction unit. The physiological feature extraction unit extracts metabolic function features and organ function indicators from physiological parameters. The behavior pattern recognition unit identifies lifestyle habits and abnormal behavior patterns from activity data. The risk prediction unit establishes a correlation model between multi-dimensional health indicators and disease progression. The knowledge base module stores medical knowledge including disease causes, symptoms, diagnostic criteria, treatment plans, and health recommendations, and interacts with the AI ​​analysis module through the RAG architecture.

[0010] The intelligent decision-making module generates personalized management plans based on metabolic function characteristics, lifestyle habits, and risk prediction results, including medication adjustment plans, lifestyle intervention suggestions, and follow-up visit plans. The intervention execution module transforms the personalized management plans into executable intervention instructions, including a medication reminder unit, an exercise prescription execution unit, a dietary guidance push unit, and a remote follow-up trigger unit. The intelligent interaction module includes a chatbot unit based on natural language processing, which receives patient inquiries and provides personalized answers based on the knowledge base module.

[0011] In some embodiments, the data acquisition module is used to collect multi-source heterogeneous health data of patients according to a preset period. This multi-source heterogeneous health data includes clinical diagnosis and treatment data, wearable device monitoring data, medication record data, and environmental monitoring data, including:

[0012] Acquire clinical diagnosis and treatment data, including outpatient records, laboratory reports, and imaging data, through the interface of the medical information system;

[0013] Receive real-time monitoring data from wearable devices such as smart bracelets, blood glucose meters, and blood pressure monitors via Bluetooth / Wi-Fi communication protocols;

[0014] Medication record data is collected through sensors in the smart pillbox or through information reported by the patient.

[0015] Environmental monitoring data matching the patient's activity area is obtained through multiple environmental monitoring device interfaces;

[0016] The collected clinical diagnosis and treatment data, wearable device monitoring data, medication record data, and environmental monitoring data are stored in association according to the patient ID;

[0017] When missing data is detected, a supplementary data collection process is automatically triggered to obtain it.

[0018] Generate complete multi-source heterogeneous health data corresponding to the current patient.

[0019] In some embodiments, the data fusion processing module is used to perform standardized preprocessing on multi-source heterogeneous health data. The preprocessing includes data cleaning, time alignment, and feature normalization, outputting time-aligned structured health data, including:

[0020] Outlier detection and correction are performed on clinical diagnosis and treatment data, and clinical diagnosis and treatment data that exceed the reasonable range of preset physiological parameters are removed;

[0021] Align the time series of data monitored by wearable devices with a standard time base to eliminate clock skew between different devices;

[0022] Medication record data is standardized in terms of dosage units and converted into standard units of measurement.

[0023] The spatial resolution of the environmental monitoring data was adjusted to a level of accuracy that matches the patient's activity trajectory;

[0024] Establish a time-related index between clinical diagnosis and treatment data, wearable device monitoring data, medication record data and environmental monitoring data, and perform normalization processing to make parameters from different sources have comparable dimensions;

[0025] Generate structured health data containing uniform timestamps and standardized features.

[0026] In some embodiments, adjusting the spatial resolution of environmental monitoring data to a level of accuracy that matches the patient's activity trajectory includes:

[0027] Based on the GPS location data recorded by the patient's smart terminal device, the spatial distribution range of the patient's activity trajectory is determined;

[0028] Based on the spatial distribution range, extract the original environmental monitoring data of the corresponding geographical area from the environmental monitoring database;

[0029] The original environmental monitoring data was spatially matched with the patient's activity trajectory, and supplementary environmental parameters were generated for the uncovered areas using Kriging interpolation.

[0030] Based on the duration of the patient's stay in the monitoring area corresponding to each original environmental monitoring data, the original environmental monitoring data and supplementary environmental parameters are time-weighted to generate weighted environmental parameters.

[0031] By spatiotemporally correlating weighted environmental parameters with patient activity trajectories, environmental monitoring records are established.

[0032] Based on the spatial distribution characteristics of the patient's activity trajectory, the spatial resolution of the environmental monitoring records was adjusted;

[0033] Output the adjusted environmental monitoring data.

[0034] In some embodiments, physiological parameters include blood glucose monitoring data, blood lipid monitoring data, continuous blood pressure monitoring data, electrocardiogram monitoring data, pulmonary function test data, and renal function-related laboratory test data;

[0035] The physiological feature extraction unit is used to extract metabolic function features and organ function indicators from physiological parameters, including:

[0036] Blood glucose monitoring data were extracted from structured health data, and wavelet transform was used to analyze intraday blood glucose fluctuation characteristics. Blood glucose coefficient of variation and blood glucose fluctuation amplitude were calculated as blood glucose metabolism characteristics.

[0037] Fourier transform analysis was performed on blood lipid test data to extract the periodic change pattern of blood lipids and establish blood lipid metabolism characteristics, including the trend of total cholesterol change and the fluctuation pattern of triglycerides.

[0038] Blood pressure variability indexes were calculated using continuous blood pressure monitoring data, and nonlinear dynamic analysis methods were used to obtain autonomic blood pressure regulation characteristics as blood pressure regulation features.

[0039] RR interval sequences were extracted from electrocardiogram monitoring data, and heart rate variability indicators were calculated as cardiovascular function indicators using time-domain and frequency-domain analysis methods.

[0040] Principal component analysis was performed on the pulmonary function test data to extract respiratory function indicators, including the rate of change of forced vital capacity and the characteristic value of maximum voluntary ventilation.

[0041] Based on renal function-related laboratory test data, establish renal function assessment indicators based on changes in creatinine clearance rate and urinary protein quantification;

[0042] Metabolic function characteristics are generated from blood glucose metabolism characteristics, blood lipid metabolism characteristics, and blood pressure regulation characteristics.

[0043] In addition, organ function indicators are generated from cardiovascular function indicators, respiratory function indicators, and renal function assessment indicators.

[0044] Physiological feature vectors are generated based on metabolic function characteristics and organ function indicators.

[0045] In some embodiments, activity data includes steps recorded by the wearable device, sleep duration and activity intensity data, dietary data, medication adherence data, and heart rate variability data.

[0046] The behavior pattern recognition unit is used to identify lifestyle characteristics and abnormal behavior patterns from activity data, including:

[0047] Step count, sleep duration, and activity intensity data recorded by wearable devices are extracted from structured health data, and time series clustering analysis is used to identify daily activity patterns.

[0048] Text mining and nutritional analysis were performed on patients’ dietary records to establish dietary habit characteristics, including the distribution of eating times and nutrient intake patterns.

[0049] By analyzing patient medication adherence data, we can identify deviations in medication timing and frequency of missed doses, and construct characteristics of medication adherence behavior.

[0050] Lifestyle characteristics are generated based on activity patterns, dietary habits, and medication adherence behaviors.

[0051] Furthermore, stress response patterns are extracted from heart rate variability data and combined with activity data to identify abnormal behavior patterns.

[0052] The characteristics of abnormal behavior episodes are classified and labeled to generate abnormal behavior patterns.

[0053] In some embodiments, the risk prediction unit is used to establish a correlation model between multidimensional health indicators and disease progression, including:

[0054] The amplitude of blood glucose fluctuations, the duration of dyslipidemia, and the coefficient of variation of blood pressure were extracted from metabolic function characteristics as the first type of predictive indicators.

[0055] The degree of decrease in heart rate variability, the rate of decline in lung function, and the trend of deterioration in kidney function were selected from organ function indicators as the second type of predictive indicators.

[0056] The number of days with insufficient exercise, frequency of dietary deviations, and number of medication non-adherence were obtained from lifestyle characteristics as the third type of predictive indicators;

[0057] Based on historical disease course data, a dose-response relationship model between the first type of predictive indicators and the risk of diabetic complications was established using survival analysis methods.

[0058] A nonlinear mapping relationship between the second type of predictive indicators and the rate of organ failure progression was constructed using machine learning algorithms.

[0059] The correlation between the third type of predictive indicator and the effectiveness of disease control was analyzed using time series forecasting methods.

[0060] By integrating dose-response relationship models, nonlinear mapping relationships, and correlation analysis results, a multidimensional disease progression association model is generated that includes short-term risk scores and long-term prognostic assessments.

[0061] In some embodiments, dose-response relationship models, nonlinear mapping relationships, and correlation analysis results are integrated to generate a multidimensional disease progression association model that includes short-term risk scores and long-term prognostic assessments, including:

[0062] The complication risk probability output by the dose-response model is standardized and transformed to generate a metabolic function risk score;

[0063] The rate of organ failure progression predicted by the nonlinear mapping relationship is converted into an organ function risk score.

[0064] The deviation of disease control effectiveness in the correlation analysis results was normalized to obtain the behavioral risk score;

[0065] A multimodal fusion algorithm was used to integrate metabolic function risk scores, organ function risk level scores, and behavioral risk scores, with the weights of each score dynamically adjusted based on clinical evidence.

[0066] A tiered risk early warning mechanism is established based on the integrated comprehensive risk score.

[0067] By using time-series prediction models to extrapolate multiple possible trajectories of patient disease progression, long-term prognostic assessment results are generated.

[0068] By linking and integrating the tiered risk warning mechanism with the results of long-term prognostic assessment, a multi-dimensional correlation model of disease progression can be formed.

[0069] In some embodiments, the intelligent decision-making module is used to generate a personalized management plan that includes medication adjustment plans, lifestyle intervention recommendations, and follow-up visit plans based on metabolic function characteristics, lifestyle habits, and risk prediction results.

[0070] Based on the amplitude of blood glucose fluctuations and the duration of dyslipidemia in metabolic function characteristics, combined with current medication record data, a dosage optimization algorithm is used to generate a medication adjustment plan that includes drug type, dosage and administration time.

[0071] Based on the number of days with insufficient exercise and the frequency of dietary deviations in lifestyle habits, lifestyle intervention recommendations containing specific exercise programs, exercise durations, and dietary restrictions are generated from a pre-set intervention strategy library.

[0072] By analyzing the correlation between the rate of organ failure progression and the effectiveness of disease control in the risk prediction results, a follow-up plan including necessary examinations and recommended follow-up frequencies was determined.

[0073] The medication adjustment plan was compared and verified with a drug interaction database to ensure the plan's safety.

[0074] Feasibility assessments were conducted on lifestyle intervention recommendations, and adaptive adjustments were made based on the characteristics of the patient's living environment.

[0075] Based on the examination items in the follow-up visit plan, the system automatically generates instructions on pre-examination preparations and precautions.

[0076] Integrate validated medication adjustment plans, adaptively adjusted lifestyle intervention recommendations, and follow-up appointment plans with precautions to form a complete personalized management plan;

[0077] Furthermore, reinforcement learning algorithms are used to dynamically optimize and adjust medication adjustment plans, lifestyle intervention recommendations, and follow-up visit plans based on patient feedback data.

[0078] In some embodiments, the intervention execution module is used to convert personalized management plans into executable intervention instructions, including:

[0079] The medication adjustment plan in the personalized management plan is analyzed, the drug name, dosage adjustment amount and administration time information are extracted, and a medication reminder instruction with a clear execution time is generated and output to the medication reminder unit;

[0080] Based on the exercise prescription in the lifestyle intervention recommendations, combined with the patient's real-time location and weather data, a daily exercise plan instruction containing the type, duration and intensity of exercise is generated and output to the exercise prescription execution unit;

[0081] The dietary structure adjustment suggestions are converted into specific ingredient lists and recipe recommendations, generating daily dietary guidance push instructions for three meals a day, which are then output to the dietary guidance push unit;

[0082] Based on the examination items and appointment time in the follow-up visit plan, the system automatically generates and sends a follow-up visit reminder instruction containing hospital navigation and precautions, and outputs it to the remote follow-up trigger unit.

[0083] In addition, establish an intervention instruction execution feedback mechanism to monitor medication records, exercise completion status and diet logs in real time, and generate compliance assessment reports;

[0084] When a deviation in the execution of an intervention instruction is detected, a dynamic adjustment mechanism is triggered to regenerate an optimized intervention instruction that is adapted to the current situation.

[0085] Unlike existing technologies, the above technical solution collects clinical diagnosis and treatment data, wearable device monitoring data, medication record data, and environmental monitoring data through a data acquisition module. After standardized preprocessing by a data fusion processing module, the data is deeply analyzed by an AI analysis module based on the Transformer architecture. Combined with medical knowledge from a knowledge base module, an intelligent decision-making module generates personalized management plans, which are then implemented through an intervention execution module and an intelligent interaction module. This technical solution processes multi-dimensional health data through a large AI model, automatically learning complex patterns and relationships, significantly improving data processing capabilities and analytical depth. The risk prediction unit accurately establishes a correlation model between health indicators and disease progression, achieving precise disease prediction and risk assessment. The intelligent decision-making module generates personalized management plans based on multi-dimensional analysis results, including medication adjustment plans, lifestyle intervention suggestions, and follow-up visit plans. Through reinforcement learning, it dynamically adjusts the plans, greatly improving their relevance and patient compliance. The chatbot unit of the intelligent interaction module provides real-time interaction and personalized guidance, effectively enhancing patient participation and self-management capabilities, forming a virtuous cycle of health management.

[0086] The above description of the invention is merely an overview of the technical solution of this application. In order to enable those skilled in the art to better understand the technical solution of this application and to implement it based on the description and drawings, and to make the above-mentioned objectives and other objectives, features and advantages of this application easier to understand, the following description is provided in conjunction with the specific embodiments and drawings of this application. Attached Figure Description

[0087] The accompanying drawings are only used to illustrate the principles, implementation methods, applications, features, and effects of specific embodiments of the present invention and other related contents, and should not be considered as limitations on this application.

[0088] In the accompanying drawings of the instruction manual:

[0089] Figure 1 This is a schematic diagram of the modules of the intelligent management platform described in a specific implementation method;

[0090] Figure 2 A schematic diagram of the AI ​​analysis module described in a specific implementation;

[0091] Figure 3 A flowchart illustrating the data collection process described in the specific implementation method;

[0092] Figure 4 This is a schematic diagram illustrating the construction and interaction of a knowledge base as described in a specific implementation method.

[0093] Figure 5 This is a schematic diagram of the large model described in the specific implementation method.

[0094] Figure label:

[0095] 1. Intelligent management platform;

[0096] 11. Data acquisition module;

[0097] 12. Data fusion and processing module;

[0098] 13. AI Analysis Module;

[0099] 131. Physiological Feature Extraction Unit;

[0100] 132. Behavior pattern recognition unit;

[0101] 133. Risk Prediction Unit;

[0102] 14. Knowledge Base Module;

[0103] 15. Intelligent Decision-Making Module;

[0104] 16. Intervention Execution Module;

[0105] 17. Intelligent Interaction Module. Detailed Implementation

[0106] To illustrate the possible application scenarios, technical principles, implementable specific solutions, and achievable objectives and effects of this application in detail, the following description, in conjunction with the listed specific embodiments and accompanying drawings, provides a detailed explanation. The embodiments described herein are merely illustrative of the technical solutions of this application and are therefore intended to limit the scope of protection of this application.

[0107] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0108] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit this application.

[0109] In the description of this application, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and A and B exist simultaneously. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" logical relationship.

[0110] In this application, terms such as “first” and “second” are used only to distinguish one entity or operation from another, and do not necessarily require or imply any actual quantity, hierarchy or order relationship between these entities or operations.

[0111] Without further limitations, the use of terms such as “comprising,” “including,” “having,” or other similar open-ended expressions in this application is intended to cover non-exclusive inclusion, which does not exclude the presence of additional elements in a process, method, or product that includes the stated elements, such that a process, method, or product that includes a list of elements may include not only those defined elements but also other elements not expressly listed, or elements inherent to such a process, method, or product.

[0112] Similar to the understanding in the Examination Guidelines, in this application, expressions such as "greater than," "less than," and "exceeding" are understood to exclude the stated number; expressions such as "above," "below," and "within" are understood to include the stated number. Furthermore, in the description of the embodiments in this application, "multiple" means two or more (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups" and "multiple times," unless otherwise explicitly specified.

[0113] The processor described in the embodiments of this application can be implemented by hardware, firmware, software, or a combination thereof. It can be a circuit, one or more of an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field-programmable gate array (FPGA), a central processing unit (CPU), a controller, a microcontroller, or a microprocessor. It also includes other physical, biological, or chemical structures that can implement the same or equivalent functions as the processors listed above, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute some or all of the steps in the computer program or method involved in the various embodiments of this application, or any combination of the steps mentioned therein.

[0114] The computer program involved in the embodiments can be stored in a computer device readable storage medium, which includes, but is not limited to, disks, magnetic tapes, magnetic cards, floppy disks, flash memory, optical disks, optical cards, read-only memory (ROM), random access memory (RAM), erasable programmable ROM (EPROM), and electrically erasable programmable ROM (EEPROM), etc., and also includes other biological, physical, or chemical structures that can achieve the same or equivalent functions as the storage media listed above, such as DNA, RNA, proteins, and other units with information storage capabilities. In specific embodiments, the storage medium involved can be one of the above-mentioned media types, or a combination of the above-mentioned media types. In different embodiments, the computer program involved in the embodiments can be centrally stored in a single medium, or distributed and stored in multiple media. The memory containing the computer device readable storage medium can be non-volatile memory or random access memory. These computer device readable storage media can be built into the device, or can be connected to the device involved in the embodiments as an external device or part of an external device. In some embodiments, the memory having a computer device readable storage medium is deployed locally; in other embodiments, the memory may be deployed remotely from the processor, for example, as a network-attached memory accessed via RF circuitry or an external port and a communication network, wherein the communication network may be the Internet, one or more intranets, a local area network (LAN), a wide area network (WLAN), a storage area network (SAN), or a suitable combination thereof, as long as computer device access to the memory is enabled. Furthermore, the computer program involved in the embodiments may be stored in plaintext / ciphertext form, or it may be designed as training data, integrated and recombined through model training and implicitly stored in the parameter states of a deep neural network or other machine learning model.

[0115] The existing technology has the following drawbacks:

[0116] Insufficient data processing capabilities: existing platforms suffer from problems such as single data dimensions and inconsistent data quality during data collection. Furthermore, data analysis is limited by traditional algorithms and methods, making it difficult to effectively process large-scale, high-dimensional, and complex data, and thus difficult to uncover the potential value within the data. This results in inaccurate and incomplete analysis and prediction of diseases.

[0117] The lack of personalized management and the failure to accurately grasp the individual characteristics of patients when formulating management plans reduce the effectiveness and pertinence of chronic disease management, and patients may find it difficult to adhere to the plan because it is not suitable.

[0118] Limited predictive capabilities: Due to limitations in analytical methods, existing platforms struggle to accurately predict disease progression and potential risks, making it impossible to take effective interventions in advance. Patients may miss the optimal treatment window, leading to a worsening of their condition.

[0119] Low patient participation: The limited interaction methods and lack of intelligent interaction result in low patient enthusiasm and initiative in health management, making it difficult to form long-term health management habits and affecting the overall effectiveness of chronic disease management.

[0120] To address the shortcomings of existing technologies, this embodiment proposes a digital intelligent management platform for chronic diseases based on the fusion of AI large-scale models and multi-dimensional data. By integrating multi-dimensional data such as patients' medical records, lifestyle habits, and genetic information, and utilizing advanced AI large-scale models for in-depth analysis and prediction, it provides patients with personalized health management plans and enhances patient participation and self-management capabilities.

[0121] For details, please refer to Figures 1 to 5 This embodiment provides a digital chronic disease intelligent management platform 1 based on AI model and multi-dimensional data fusion, including a data acquisition module 11, a data fusion processing module 12, an AI analysis module 13, a knowledge base module 14, an intelligent decision-making module 15, an intervention execution module 16, and an intelligent interaction module 17; the data acquisition module 11 is used to collect multi-source heterogeneous health data of patients according to a preset cycle, and the multi-source heterogeneous health data includes clinical diagnosis and treatment data, wearable device monitoring data, medication record data, and environmental monitoring data;

[0122] The data fusion processing module 12 is used to perform standardized preprocessing on multi-source heterogeneous health data. The preprocessing includes data cleaning, time alignment and feature normalization, and outputs time-aligned structured health data.

[0123] The AI ​​analysis module 13 employs a pre-trained large model based on the Transformer architecture. This large model is integrated with a medical knowledge graph and includes a physiological feature extraction unit 131, a behavior pattern recognition unit 132, and a risk prediction unit 133. The physiological feature extraction unit 131 is used to extract metabolic function features and organ function indicators from physiological parameters. The behavior pattern recognition unit 132 is used to identify lifestyle habits and abnormal behavior patterns from activity data. The risk prediction unit 133 is used to establish a correlation model between multi-dimensional health indicators and disease progression. The knowledge base module 14 stores medical knowledge containing disease etiologies, symptoms, diagnostic criteria, treatment plans, and health recommendations, and interacts with the AI ​​analysis module 13 through the RAG architecture.

[0124] The intelligent decision-making module 15 is used to generate a personalized management plan that includes medication adjustment plans, lifestyle intervention suggestions, and follow-up visit plans based on metabolic function characteristics, lifestyle characteristics, and risk prediction results. The intervention execution module 16 is used to convert the personalized management plan into executable intervention instructions. The intervention execution module 16 includes a medication reminder unit, an exercise prescription execution unit, a diet guidance push unit, and a remote follow-up trigger unit. The intelligent interaction module 17 includes a chatbot unit based on natural language processing, which is used to receive patient inquiries and provide personalized answers based on the knowledge base module 14.

[0125] In this embodiment, the data acquisition module 11 is responsible for interacting with the electronic medical record system of medical institutions, smart wearable devices, patient mobile applications, and gene testing institutions to comprehensively collect multi-dimensional patient data. This data includes information from the electronic medical record system of medical institutions, smart wearable devices (such as smartwatches and smart bracelets that can monitor heart rate, steps, sleep quality, etc. in real time), data entered voluntarily by patients through mobile applications (dietary records, symptom descriptions, etc.), and genetic information provided by gene testing institutions. This multi-dimensional data is transmitted to the platform database via a secure and encrypted communication protocol.

[0126] The data fusion processing module 12 is used to perform standardized preprocessing on multi-source heterogeneous health data, clean the collected multi-source heterogeneous data, remove noise and duplicate data; perform data conversion to unify data of different formats into a standard format; and process the data through normalization, standardization and other methods to make it suitable for the input requirements of large AI models.

[0127] Advanced AI models are employed to perform in-depth analysis of preprocessed data, enabling functions such as disease prediction, risk assessment, and etiology inference. The AI ​​analysis module 13 is divided into a physiological feature extraction unit 131, a behavioral pattern recognition unit 132, and a risk prediction unit 133 to meet different functional requirements. Specifically, pre-trained AI models (such as a model fusing a large-scale language model based on the Transformer architecture with a medical knowledge graph) are used to perform in-depth analysis of the preprocessed data. The AI ​​model can automatically learn complex patterns and relationships in the data, enabling disease prediction (such as predicting the fluctuation trends of blood sugar and blood pressure in a patient's future, and the risk of disease onset), risk assessment (assessing the likelihood of a patient developing complications), and etiology inference (analyzing potential factors leading to the development of chronic diseases, such as lifestyle habits and genetic factors).

[0128] The intelligent decision-making module 15 dynamically adjusts and optimizes the management plan based on the analysis results of the AI ​​big data model, combined with factors such as the patient's personal preferences and lifestyle habits, using reinforcement learning algorithms. The plan includes personalized dietary recommendations (recommending appropriate food types and intake based on the patient's blood sugar, blood lipids, and other indicators), exercise plans (developing personalized exercise plans based on the patient's physical condition and exercise capacity), medication reminders (setting precise medication reminder times and dosages based on the patient's condition and drug characteristics), and psychological intervention measures (providing corresponding psychological support and guidance suggestions for the patient's mental state), etc.

[0129] The intelligent interaction module 17 interacts with patients in real time through an intelligent chatbot (driven by a large AI model based on natural language processing technology). Patients can consult the chatbot at any time about health issues, provide feedback on their physical condition and the implementation status of treatment plans. The chatbot can understand the patient's natural language questions and provide accurate and personalized answers and suggestions. At the same time, the platform continuously optimizes the analytical capabilities of the large AI model and the effectiveness of management plan formulation based on patient feedback data.

[0130] In this embodiment, the knowledge base module 14 is used to construct a chronic disease management knowledge base system based on the RAG architecture. It uses vector retrieval tools and embedding models to structurally store medical knowledge such as disease etiology, symptoms, diagnostic criteria, treatment plans, and health advice, and establishes an efficient indexing mechanism, enabling the AI ​​analysis module 13 to retrieve relevant knowledge in real time to assist in analysis and decision-making. The knowledge base content covers a complete diagnostic and treatment knowledge system for common chronic diseases such as diabetes, hypertension, and coronary heart disease, and is regularly updated with clinical guidelines and the latest research findings.

[0131] In this embodiment, the intervention execution module 16 is used to achieve precise execution of the management plan. The medication reminder unit generates timed reminders based on the type, dosage, and administration time of the medication in the medication adjustment plan. The exercise prescription execution unit dynamically adjusts the exercise plan based on the patient's real-time location data and weather conditions. The diet guidance push unit generates a daily customized diet based on the patient's dietary habits. The remote follow-up trigger unit automatically schedules appointments and pushes examination precautions according to the follow-up visit plan, forming a complete intervention closed loop.

[0132] In this embodiment, the security control module is used to ensure system data security, implement end-to-end encryption of multi-source heterogeneous health data during transmission, control the access levels of different users to the system through role-based access control, adopt a sensitive information filtering mechanism to prevent privacy data leakage, and establish a complete operation audit log to record all data access behaviors.

[0133] In this embodiment, the feedback optimization module is used to continuously improve system performance. By collecting patient feedback on the execution of intervention instructions, consultation records of the intelligent interaction module 17, and clinical effect evaluation data, the module uses reinforcement learning algorithms to dynamically optimize the prediction model parameters of the AI ​​analysis module 13 and the solution generation strategy of the intelligent decision-making module 15, thereby achieving iterative upgrades of the management plan.

[0134] For details, please refer to Figure 4 The following specific application examples are provided for explanation:

[0135] Clearly define business requirements and determine the major models to be deployed (such as DEEPSEEK R1, Doubao 1.5, MOONSHOT, Baichuan, Tongyi Qianwen, Zhipu Qingyan GLM, etc.) as well as the specific functions and application scenarios of the chronic disease management knowledge base.

[0136] The scope of the planned knowledge base should cover information such as the causes, symptoms, diagnostic criteria, treatment plans, and dietary and exercise recommendations for common chronic diseases (such as diabetes, hypertension, and coronary heart disease).

[0137] Prepare the necessary hardware and software environment for model deployment and knowledge base operation, including servers, storage devices, operating systems, database management systems, dependency libraries, etc.

[0138] Data related to chronic disease management is collected from various data sources, such as the Health Road Diagnosis and Treatment Database, medical literature, clinical guidelines, and expert experience.

[0139] The collected data is cleaned, preprocessed, and structured to ensure its accuracy and consistency.

[0140] Choose a suitable knowledge base management system, such as a solution based on the RAG (Retrieval-Augmented Generation) architecture, and use vector retrieval tools (such as qdrant) and embedding models (such as bce, bge) to build the knowledge base.

[0141] Import the organized chronic disease management data into the knowledge base and create an index for quick retrieval and querying.

[0142] Conduct functional testing on the knowledge base to ensure that it can accurately store and retrieve chronic disease management knowledge.

[0143] The knowledge base is optimized based on the test results, such as adjusting the indexing strategy and optimizing the data storage structure.

[0144] Through the unified API interface of the Health Road AI Gateway, standardized access methods are provided for different applications, services and AI agents, enabling them to easily call large model services and chronic disease management knowledge bases.

[0145] Configure the AI ​​agent to interact with various large models and chronic disease management knowledge bases on behalf of different applications or services. The AI ​​agent is responsible for handling operations such as request forwarding and protocol conversion.

[0146] Subscribing on the developer platform grants access to various large models and chronic disease management knowledge bases, along with associated resources. This may involve obtaining API keys and managing quotas.

[0147] Enable security-related configurations, such as prompt word templates, to ensure that the data input into the model and knowledge base conforms to the specifications and does not contain sensitive information; set up a security degradation mechanism to promptly reduce the service level or take other security measures when security risks occur.

[0148] Access to the chronic disease management knowledge base is controlled to ensure that only authorized users can access and modify the knowledge content.

[0149] Enable the audit log function to record all requests and responses through the AI ​​gateway, as well as access and modification operations to the chronic disease management knowledge base, to facilitate subsequent auditing and troubleshooting.

[0150] Configure a retry mechanism to automatically retry when model requests or knowledge base queries fail, thereby improving service reliability.

[0151] Configure load balancing to distribute requests evenly across multiple model instances and knowledge base servers, avoiding single points of failure and performance bottlenecks.

[0152] Integrating the chronic disease management knowledge base with various large models allows the models to reference the professional knowledge in the knowledge base when generating answers. For example, using RAG technology, relevant information is retrieved from the knowledge base before the model generates an answer, and then the retrieved information is provided to the model along with the input question.

[0153] Integrate AI gateways with applications, services, and AI agents. For example, embed the AI ​​gateway's API into applications, enabling them to directly access large model services and chronic disease management knowledge bases; interface AI agents with the AI ​​gateway, allowing them to leverage the capabilities of large models and knowledge bases for intelligent decision-making and interaction.

[0154] Collect feedback from users and business systems to understand the effectiveness and problems of models and knowledge bases in actual use.

[0155] Based on feedback and business development needs, the model was updated and optimized to improve its performance and accuracy. Simultaneously, the content of the chronic disease management knowledge base was updated, adding new chronic disease management knowledge and research findings.

[0156] Update the AI ​​gateway configuration to accommodate the new model and knowledge base services.

[0157] Through the above process, we can ensure that the large-scale model service of Health Road and the chronic disease management knowledge base can operate stably, securely, and efficiently, and provide strong intelligent support for various applications and businesses related to chronic disease management.

[0158] The above example has the following advantages:

[0159] With powerful data processing capabilities, it utilizes AI big models to process large-scale, multi-dimensional, and complex health data. It can automatically learn complex patterns and relationships in the data, uncover the potential value behind the data, improve the depth and breadth of data analysis, and provide strong support for accurate disease prediction and risk assessment.

[0160] Accurate disease prediction and risk assessment, through the deep analysis capabilities of AI big data models, can accurately predict the development trend and potential risks of diseases, detect patients' health problems in advance, provide a scientific basis for doctors to formulate treatment plans and for patients to manage themselves, and help improve treatment effectiveness and prevent the occurrence of complications.

[0161] A highly personalized management plan is developed based on the patient's specific condition and the analysis results of the AI ​​big data model. The plan can be dynamically adjusted according to the patient's feedback and changes in the condition, thereby improving the effectiveness and relevance of the management plan and enhancing the patient's adherence to the plan.

[0162] Enhanced patient engagement: Real-time interaction with patients through intelligent chatbots provides personalized health guidance and consultation services, enhancing patients' sense of participation and experience, stimulating their enthusiasm and initiative in health management, helping to form long-term health management habits, and improving the overall effectiveness of chronic disease management.

[0163] This embodiment collects clinical diagnosis and treatment data, wearable device monitoring data, medication record data, and environmental monitoring data through the data acquisition module 11. After standardized preprocessing by the data fusion processing module 12, the AI ​​analysis module 13 based on the Transformer architecture performs in-depth analysis. Combined with medical knowledge from the knowledge base module 14, the intelligent decision-making module 15 generates a personalized management plan, which is then implemented through the intervention execution module 16 and the intelligent interaction module 17. By processing multi-dimensional health data through a large AI model, complex patterns and correlations are automatically learned, significantly improving data processing capabilities and analytical depth. The risk prediction unit 133 accurately establishes a correlation model between health indicators and disease progression, achieving precise disease prediction and risk assessment. The intelligent decision-making module 15 generates a personalized management plan based on the multi-dimensional analysis results, including medication adjustment plans, lifestyle intervention suggestions, and follow-up visit plans. Through reinforcement learning, it dynamically adjusts the plan, greatly improving its relevance and patient compliance. The chatbot unit of the intelligent interaction module 17 provides real-time interaction and personalized guidance, effectively enhancing patient participation and self-management capabilities, forming a virtuous cycle of health management.

[0164] In some embodiments, the data acquisition module 11 is used to collect multi-source heterogeneous health data of patients according to a preset cycle. The multi-source heterogeneous health data includes clinical diagnosis and treatment data, wearable device monitoring data, medication record data, and environmental monitoring data, including:

[0165] Acquire clinical diagnosis and treatment data, including outpatient records, laboratory reports, and imaging data, through the interface of the medical information system;

[0166] Receive real-time monitoring data from wearable devices such as smart bracelets, blood glucose meters, and blood pressure monitors via Bluetooth / Wi-Fi communication protocols;

[0167] Medication record data is collected through sensors in the smart pillbox or through information reported by the patient.

[0168] Environmental monitoring data matching the patient's activity area is obtained through multiple environmental monitoring device interfaces;

[0169] The collected clinical diagnosis and treatment data, wearable device monitoring data, medication record data, and environmental monitoring data are stored in association according to the patient ID;

[0170] When missing data is detected, a supplementary data collection process is automatically triggered to obtain it.

[0171] Generate complete multi-source heterogeneous health data corresponding to the current patient.

[0172] In this embodiment, when obtaining clinical diagnosis and treatment data through the medical information system interface, the system uses medical data standard protocols such as HL7 / FHIR to interface with the hospital's HIS system, automatically parsing diagnostic information in outpatient records, indicator data in laboratory reports, and structured reports of imaging examinations to ensure the integrity and accuracy of medical data.

[0173] In this embodiment, when receiving wearable device monitoring data via Bluetooth / Wi-Fi communication protocol, a device authentication mechanism and data verification process are established to perform real-time quality detection on the heart rate and blood oxygen data collected by the smart bracelet, the blood glucose value monitored by the blood glucose meter, and the blood pressure data measured by the blood pressure monitor, eliminating abnormal fluctuations and measurement values ​​that exceed the reasonable range.

[0174] In this embodiment, when collecting medication record data through the smart pillbox, the built-in weight sensor and RFID tag identification technology are used to accurately record the removal time and dosage of each drug. At the same time, a mobile application is provided for patients to supplement their self-administered medication information, forming a complete medication time series data.

[0175] In this embodiment, when acquiring data through the environmental monitoring equipment interface, the nearest air quality monitoring station, weather station, and noise monitoring point are dynamically selected based on the patient's GPS location information. Environmental parameters such as PM2.5, temperature, humidity, and ultraviolet intensity are collected, and a spatial correlation model with the patient's activity trajectory is established.

[0176] In this embodiment, when storing multi-source data in association, a distributed database architecture is adopted to establish an independent data space for each patient. Heterogeneous data from different systems are indexed uniformly through the patient ID to form a complete patient health record.

[0177] In this embodiment, when the supplementary data collection process is triggered, the missing key indicators are identified based on the data integrity assessment algorithm, and supplementary data collection is prioritized through the automated interface. When the automated supplementary data collection fails, a manual supplementary data collection reminder is sent to the patient through the intelligent interaction module 17 to ensure the integrity of the dataset.

[0178] In this embodiment, when generating multi-source heterogeneous health data, data consistency checks and timestamp alignment are performed to verify the logical correlation between data in each dimension, and finally a standardized health dataset that can be directly used by the AI ​​analysis module 13 is output.

[0179] This embodiment achieves efficient integration and quality control of multi-source health data. Through standardized interfaces and intelligent supplementary collection mechanisms, it ensures the integrity and timeliness of clinical diagnosis and treatment data, wearable device monitoring data, medication record data, and environmental monitoring data. The adoption of multi-level data verification and quality control processes significantly improves the accuracy and reliability of health data.

[0180] In some embodiments, the data fusion processing module 12 is used to perform standardized preprocessing on multi-source heterogeneous health data. The preprocessing includes data cleaning, time alignment, and feature normalization, outputting time-aligned structured health data, including:

[0181] Outlier detection and correction are performed on clinical diagnosis and treatment data, and clinical diagnosis and treatment data that exceed the reasonable range of preset physiological parameters are removed;

[0182] Align the time series of data monitored by wearable devices with a standard time base to eliminate clock skew between different devices;

[0183] Medication record data is standardized in terms of dosage units and converted into standard units of measurement.

[0184] The spatial resolution of the environmental monitoring data was adjusted to a level of accuracy that matches the patient's activity trajectory;

[0185] Establish a time-related index between clinical diagnosis and treatment data, wearable device monitoring data, medication record data and environmental monitoring data, and perform normalization processing to make parameters from different sources have comparable dimensions;

[0186] Generate structured health data containing uniform timestamps and standardized features.

[0187] In this embodiment, when detecting outliers in clinical diagnosis and treatment data, a dynamic threshold algorithm is used in conjunction with the patient's individual baseline data to establish a personalized reasonable range model for different physiological parameters. This not only identifies and removes obviously erroneous outliers, but also marks critically suspicious data for manual review, ensuring the reliability of clinical diagnosis and treatment data.

[0188] In this embodiment, when aligning the time series of wearable device monitoring data, a device clock calibration protocol is established, and the time of each device is synchronized through an NTP time server. Furthermore, a dynamic time warping algorithm is used to correct the time axis of historically collected data, thereby resolving the time deviation problem caused by device clock drift.

[0189] In this embodiment, when standardizing the medication record data, a built-in drug dosage conversion knowledge base is used to automatically identify the specification differences of drugs from different manufacturers, convert the dosage of various drugs into standard active ingredient dosages, and retain the original records for traceability and verification.

[0190] In this embodiment, when adjusting the spatial resolution of environmental monitoring data, an adaptive grid partitioning algorithm is adopted based on the spatial density distribution of the patient's activity trajectory. A fine grid is used in high-frequency activity areas, and a coarse-grained grid is used in low-frequency activity areas, so as to achieve the best balance between monitoring accuracy and data processing efficiency.

[0191] In this embodiment, when establishing the time-related index, a multi-level time matching mechanism is developed. Real-time monitoring data accurate to the minute, medication data recorded daily, and irregular clinical examination results are associated using different time windows to establish cross-time scale data association relationships.

[0192] In this embodiment, when performing feature normalization, the most suitable methods such as Z-score normalization and Min-Max normalization are used for different types of health parameters, while preserving the original numerical range and transformation parameters to ensure data reversibility and without affecting clinical interpretation.

[0193] In this embodiment, when generating structured health data, a hierarchical data structure design is adopted, which stores the original data, processing metadata and standardized data in layers, ensuring both the convenience of data analysis and the complete preservation of data traceability information.

[0194] This embodiment achieves deep integration and standardization of multi-source health data. Through intelligent anomaly detection and personalized data processing strategies, it significantly improves the quality and consistency of health data. The innovative spatiotemporal alignment algorithm effectively solves the key challenges in integrating multi-source heterogeneous data, enabling seamless connection and collaborative analysis of health data from different sources and dimensions.

[0195] In some embodiments, adjusting the spatial resolution of environmental monitoring data to a level of accuracy that matches the patient's activity trajectory includes:

[0196] Based on the GPS location data recorded by the patient's smart terminal device, the spatial distribution range of the patient's activity trajectory is determined;

[0197] Based on the spatial distribution range, extract the original environmental monitoring data of the corresponding geographical area from the environmental monitoring database;

[0198] The original environmental monitoring data was spatially matched with the patient's activity trajectory, and supplementary environmental parameters were generated for the uncovered areas using Kriging interpolation.

[0199] Based on the duration of the patient's stay in the monitoring area corresponding to each original environmental monitoring data, the original environmental monitoring data and supplementary environmental parameters are time-weighted to generate weighted environmental parameters.

[0200] By spatiotemporally correlating weighted environmental parameters with patient activity trajectories, environmental monitoring records are established.

[0201] Based on the spatial distribution characteristics of the patient's activity trajectory, the spatial resolution of the environmental monitoring records was adjusted;

[0202] Output the adjusted environmental monitoring data.

[0203] In this embodiment, when determining the activity trajectory based on the GPS positioning data recorded by the patient's smart terminal device, the original positioning point is processed using trajectory compression and denoising algorithms to identify the patient's permanent residence and movement path, and to construct an activity heat map accurate to 50 meters, which serves as the basis for determining the spatial distribution range.

[0204] In this embodiment, when extracting data from the environmental monitoring database, a dynamic query radius mechanism is established. The query range is automatically adjusted according to the degree of concentration of the patient's activity area. A fine query of 100 meters is used for densely populated activity areas, and an extended query of 500 meters is used for dispersed activity areas to ensure accurate correspondence between monitoring data and activity areas.

[0205] In this embodiment, when performing spatial matching and Kriging interpolation, the distribution density of monitoring stations, terrain features, and pollutant diffusion models are comprehensively considered. Variation function parameters are set, and spherical models are used for interpolation of pollutants such as PM2.5, while Gaussian models are used for interpolation of meteorological parameters such as temperature and humidity, thereby improving the accuracy of supplementary environmental parameters.

[0206] In this embodiment, when performing time-weighted processing, a segmented weighting function is established. Areas where people stay for more than 1 hour are given full weight, while areas that are only briefly passed through are weighted according to the length of stay. At the same time, the differences in environmental exposure impact at different times are considered, and parameters during the nighttime rest period are given higher weight.

[0207] In this embodiment, when establishing spatiotemporally correlated environmental monitoring records, a spatiotemporal cube data model is adopted to combine two-dimensional geographic space with the time dimension, label environmental parameter values ​​for each spatiotemporal unit, and record the data source (actual measurement or interpolation) and confidence score.

[0208] In this embodiment, when adjusting the spatial resolution, a dynamic grid optimization algorithm is implemented. A fine grid of 100×100 meters is used in the patient's main activity area, a coarse grid of 500×500 meters is used in the secondary activity area, and an ultra-high precision of 50×50 meters is maintained in key locations such as medical institutions to achieve optimal resource allocation.

[0209] This embodiment achieves accurate assessment of environmental exposure. Based on the matching of environmental monitoring data with real activity trajectories, it significantly improves the individualization and accuracy of environmental exposure assessment. The intelligent spatial resolution adjustment not only ensures the analysis accuracy of key areas, but also optimizes data processing efficiency.

[0210] In some embodiments, physiological parameters include blood glucose monitoring data, blood lipid monitoring data, continuous blood pressure monitoring data, electrocardiogram monitoring data, pulmonary function test data, and renal function-related laboratory test data;

[0211] Physiological feature extraction unit 131 is used to extract metabolic function features and organ function indicators from physiological parameters, including:

[0212] Blood glucose monitoring data were extracted from structured health data, and wavelet transform was used to analyze intraday blood glucose fluctuation characteristics. Blood glucose coefficient of variation and blood glucose fluctuation amplitude were calculated as blood glucose metabolism characteristics.

[0213] Fourier transform analysis was performed on blood lipid test data to extract the periodic change pattern of blood lipids and establish blood lipid metabolism characteristics, including the trend of total cholesterol change and the fluctuation pattern of triglycerides.

[0214] Blood pressure variability indexes were calculated using continuous blood pressure monitoring data, and nonlinear dynamic analysis methods were used to obtain autonomic blood pressure regulation characteristics as blood pressure regulation features.

[0215] RR interval sequences were extracted from electrocardiogram monitoring data, and heart rate variability indicators were calculated as cardiovascular function indicators using time-domain and frequency-domain analysis methods.

[0216] Principal component analysis was performed on the pulmonary function test data to extract respiratory function indicators, including the rate of change of forced vital capacity and the characteristic value of maximum voluntary ventilation.

[0217] Based on renal function-related laboratory test data, establish renal function assessment indicators based on changes in creatinine clearance rate and urinary protein quantification;

[0218] Metabolic function characteristics are generated from blood glucose metabolism characteristics, blood lipid metabolism characteristics, and blood pressure regulation characteristics.

[0219] In addition, organ function indicators are generated from cardiovascular function indicators, respiratory function indicators, and renal function assessment indicators.

[0220] Physiological feature vectors are generated based on metabolic function characteristics and organ function indicators.

[0221] In this embodiment, when analyzing blood glucose monitoring data using the wavelet transform method, the db4 wavelet basis function is selected for multi-scale decomposition. Characteristic blood glucose fluctuation patterns after breakfast, lunch and dinner are identified in the time and frequency domain. The coefficient of variation (CV) reflecting overall stability and the maximum amplitude (MAGE) reflecting extreme fluctuations are calculated to construct a multi-dimensional blood glucose metabolism feature vector.

[0222] In this embodiment, when performing Fourier transform analysis on blood lipid test data, the amplitude spectrum and phase spectrum of the periodic changes of indicators such as total cholesterol (TC), triglycerides (TG), high-density lipoprotein (HDL) and low-density lipoprotein (LDL) are calculated respectively to identify the blood lipid fluctuation patterns related to dietary cycles and establish a feature matrix reflecting the dynamic characteristics of blood lipid metabolism.

[0223] In this embodiment, when analyzing blood pressure variability using nonlinear dynamics, detrended fluctuation analysis (DFA) is used to calculate the long-term correlation index α, the complexity of the blood pressure sequence is assessed by sample entropy (SampEn), and traditional time-domain indicators (SDNN, RMSSD) are combined to comprehensively characterize the autonomic nervous regulation function of blood pressure.

[0224] In this embodiment, when extracting heart rate variability indicators from electrocardiogram data, SDNN and RMSSD are calculated in the time domain to reflect the overall variability, and the power of low-frequency (LF) and high-frequency (HF) components is analyzed through Lomb-Scargle periodogram in the frequency domain to construct a multidimensional cardiovascular function indicator set that reflects the autonomic nervous system balance.

[0225] In this embodiment, when performing principal component analysis on lung function data, key parameters such as forced vital capacity (FVC), FEV1 / FVC, and maximum mid-expiratory flow rate (MMEF) are selected. The feature vectors reflecting restrictive and obstructive lesions are extracted through PCA dimensionality reduction to establish a comprehensive respiratory function assessment model.

[0226] In this embodiment, when establishing renal function assessment indicators, the Cockcroft-Gault formula is used to calculate creatinine clearance rate (Ccr), and combined with the dynamic trend of urine protein / creatinine ratio (UPCR), a comprehensive evaluation index reflecting glomerular filtration function and the degree of renal tubular damage is constructed.

[0227] In this embodiment, when generating physiological feature vectors, metabolic function features and organ function indicators are standardized respectively, and a unified feature space is constructed using a feature-level fusion method to preserve the clinical interpretability of each indicator, while also marking the normal reference range and clinical significance of each feature.

[0228] This embodiment achieves a deep characterization of the patient's health status. The multimodal signal processing method fully explores the deep feature information contained in various physiological parameters. The indicator design based on clinical pathological mechanisms ensures the medical significance and practicality of the features.

[0229] In some embodiments, activity data includes steps recorded by the wearable device, sleep duration and activity intensity data, dietary data, medication adherence data, and heart rate variability data.

[0230] Behavioral pattern recognition unit 132 is used to identify lifestyle characteristics and abnormal behavioral patterns from activity data, including:

[0231] Step count, sleep duration, and activity intensity data recorded by wearable devices are extracted from structured health data, and time series clustering analysis is used to identify daily activity patterns.

[0232] Text mining and nutritional analysis were performed on patients’ dietary records to establish dietary habit characteristics, including the distribution of eating times and nutrient intake patterns.

[0233] By analyzing patient medication adherence data, we can identify deviations in medication timing and frequency of missed doses, and construct characteristics of medication adherence behavior.

[0234] Lifestyle characteristics are generated based on activity patterns, dietary habits, and medication adherence behaviors.

[0235] Furthermore, stress response patterns are extracted from heart rate variability data and combined with activity data to identify abnormal behavior patterns.

[0236] The characteristics of abnormal behavior episodes are classified and labeled to generate abnormal behavior patterns.

[0237] In this embodiment, when processing wearable device data using time series clustering analysis, sliding window segmentation and feature extraction are performed on step count, sleep duration, and activity intensity data. Hierarchical clustering is performed using DTW (Dynamic Time Warping) distance metric to identify typical activity patterns on weekdays / rest days, as well as abnormal activity periods (such as prolonged sitting or abnormal nighttime activity), and to construct a feature vector reflecting the patient's behavioral rhythm.

[0238] In this embodiment, when performing text mining on dietary record data, named entity recognition technology based on medical knowledge graph is used to accurately extract information such as food type, portion size and cooking method from the patient's freely described dietary content. Combined with the nutrition database, the energy and nutrient intake of each meal is calculated, and the changing trends of the intake ratio of carbohydrates, fats and proteins over time are analyzed to establish a quantitative dietary habit feature model.

[0239] In this embodiment, when analyzing medication adherence data, a simple missed dose rate index is calculated. The pattern characteristics of medication time series are analyzed through Markov chain model to identify different patterns of accidental missed doses and habitual missed doses. At the same time, the potential impact of adherence on drug efficacy is assessed by combining drug half-life parameters, and a multidimensional medication behavior feature vector is constructed.

[0240] In this embodiment, when identifying stress responses from heart rate variability data, a mutation detection algorithm is used to locate abnormal fluctuations in HRV indicators (such as the LF / HF ratio), and behavioral events (such as strenuous exercise, emotional fluctuations, etc.) in activity data are simultaneously correlated to distinguish between physiological stress and pathological stress responses, and a quantitative assessment model for the intensity and duration of stress responses is established.

[0241] In this embodiment, when classifying abnormal behavioral characteristics, a three-level classification system is established based on clinical guidelines: Level 1 is mild deviation (such as occasional late nights or missed doses), Level 2 is moderate abnormality (such as insufficient exercise for three consecutive days), and Level 3 is severe abnormality (such as circadian rhythm reversal or persistent drug refusal). Each level is associated with different intervention strategies and early warning mechanisms.

[0242] This embodiment enables a comprehensive assessment of patients' lifestyles. The extraction and quantification of multi-dimensional behavioral characteristics make previously subjective lifestyle habits measurable and analyzable. Early identification and graded warning of abnormal behaviors provide a basis for timely intervention, significantly improving the accuracy and effectiveness of health management.

[0243] In some embodiments, the risk prediction unit 133 is used to establish a correlation model between multidimensional health indicators and disease progression, including:

[0244] The amplitude of blood glucose fluctuations, the duration of dyslipidemia, and the coefficient of variation of blood pressure were extracted from metabolic function characteristics as the first type of predictive indicators.

[0245] The degree of decrease in heart rate variability, the rate of decline in lung function, and the trend of deterioration in kidney function were selected from organ function indicators as the second type of predictive indicators.

[0246] The number of days with insufficient exercise, frequency of dietary deviations, and number of medication non-adherence were obtained from lifestyle characteristics as the third type of predictive indicators;

[0247] Based on historical disease course data, a dose-response relationship model between the first type of predictive indicators and the risk of diabetic complications was established using survival analysis methods.

[0248] A nonlinear mapping relationship between the second type of predictive indicators and the rate of organ failure progression was constructed using machine learning algorithms.

[0249] The correlation between the third type of predictive indicator and the effectiveness of disease control was analyzed using time series forecasting methods.

[0250] By integrating dose-response relationship models, nonlinear mapping relationships, and correlation analysis results, a multidimensional disease progression association model is generated that includes short-term risk scores and long-term prognostic assessments.

[0251] In this embodiment, when establishing a dose-response relationship model using survival analysis, Cox proportional hazards models are constructed for different types of diabetic complications (such as retinopathy, nephropathy, and neuropathy). Continuous predictive indicators such as blood glucose fluctuation amplitude are converted into equivalent effect intervals. The hazard ratio (HR) and 95% confidence interval of each risk interval are calculated, and a visual risk ladder is generated to intuitively show the quantitative relationship between the degree of metabolic abnormality and the risk of complications.

[0252] In this embodiment, when constructing nonlinear mapping relationships using machine learning algorithms, the Gradient Boosting Decision Tree (GBDT) model is used to process organ function index data. The weight allocation of each organ function is determined through feature importance analysis, and the SHAP value is used to interpret the model prediction results. This not only outputs the predicted value of organ failure risk, but also provides attribution analysis of key influencing factors, helping clinicians to understand the risk formation mechanism.

[0253] In this embodiment, when analyzing the influence of behavioral factors using time series prediction methods, a VAR (vector autoregression) model is established to capture the dynamic interaction between lifestyle indicators and disease control indicators (such as HbA1c, blood pressure, etc.). The Granger causality test is used to determine the temporal causal relationship between behavioral changes and health outcomes, and to quantify the expected effect and lag period of different behavioral interventions.

[0254] In this embodiment, when integrating the prediction results of multiple models, a fusion algorithm based on evidence weighting is used to assign dynamic weights to metabolic-related risks, organ function risks, and writing-related risks. The weight coefficients are adaptively adjusted according to the patient's current disease stage and the characteristics of previous treatment responses. The resulting risk score reflects both the overall risk of disease progression and retains the risk contribution analysis of each dimension.

[0255] In this embodiment, a standardized scale of 0-100 points is used to construct the short-term risk score. The score range is divided with reference to clinical practice guidelines: 0-30 points are low risk (routine follow-up), 31-70 points are medium risk (intensive intervention), and 71-100 points are high risk (emergency treatment). The score is updated daily and the main sources of risk are marked.

[0256] In this embodiment, when generating long-term prognostic assessments, the natural disease process is simulated based on a Markov multi-state transition model, and the distribution of health status over 1 year, 3 years, and 5 years is predicted in combination with the patient's current risk characteristics, providing a comparative analysis of prognostic outcomes under different intervention scenarios.

[0257] This embodiment achieves accurate assessment of disease progression, and the synergistic analysis of multi-dimensional indicators overcomes the limitations of traditional single-indicator prediction; the dynamic risk scoring mechanism enables real-time monitoring and early warning of disease risk; the comprehensive assessment of long and short term provides a scientific basis for hierarchical diagnosis and treatment and personalized intervention, enabling chronic disease management to shift from passive response to proactive prevention, and significantly improving the foresight and effectiveness of health management.

[0258] In some embodiments, dose-response relationship models, nonlinear mapping relationships, and correlation analysis results are integrated to generate a multidimensional disease progression association model that includes short-term risk scores and long-term prognostic assessments, including:

[0259] The complication risk probability output by the dose-response model is standardized and transformed to generate a metabolic function risk score;

[0260] The rate of organ failure progression predicted by the nonlinear mapping relationship is converted into an organ function risk score.

[0261] The deviation of disease control effectiveness in the correlation analysis results was normalized to obtain the behavioral risk score;

[0262] A multimodal fusion algorithm was used to integrate metabolic function risk scores, organ function risk level scores, and behavioral risk scores, with the weights of each score dynamically adjusted based on clinical evidence.

[0263] A tiered risk early warning mechanism is established based on the integrated comprehensive risk score.

[0264] By using time-series prediction models to extrapolate multiple possible trajectories of patient disease progression, long-term prognostic assessment results are generated.

[0265] By linking and integrating the tiered risk warning mechanism with the results of long-term prognostic assessment, a multi-dimensional correlation model of disease progression can be formed.

[0266] In this embodiment, when standardizing the probability of complication risk, the logistic function is used to map the original probability value to a scoring range of 0-100 points, and differentiated conversion curves are set according to the clinical severity of different complications, so that the risk score of serious consequences such as end-stage renal disease increases more significantly, thereby improving the clinical relevance of risk warnings.

[0267] In this embodiment, the rate of progression of organ failure can be classified into five levels: Level 1 is the functional compensation stage (annual decline rate <2%), Level 2 is mild decompensation (annual decline rate 2-5%), Level 3 is moderate decompensation (5-10%), Level 4 is severe decompensation (10-20%), and Level 5 is the terminal stage (>20%). Each level corresponds to different clinical treatment strategies and follow-up frequencies.

[0268] In this embodiment, when normalizing the deviation of disease control effect, Z-score standardization combined with Tukey's fences outlier detection is used to identify abnormal periods that significantly deviate from the expected control target, and the cumulative deviation integral within a rolling time window (such as the last 30 days) is calculated to reflect the persistence and cumulative effect of behavioral risk.

[0269] In this embodiment, when using the multimodal fusion algorithm, a weighting strategy based on the evidence pyramid is developed: the weight of indicators supported by evidence from randomized controlled trials is set to 0.5, that supported by evidence from cohort studies is set to 0.3, and that supported by expert consensus is set to 0.2. Clinicians are allowed to make ±20% personalized adjustments based on individual patient characteristics to achieve a balance between standardization and personalization.

[0270] In this embodiment, when establishing a graded risk warning mechanism, a three-level response system is designed: a yellow warning (comprehensive score of 30-50 points) triggers the system to automatically push health education content; an orange warning (51-70 points) initiates remote medical team intervention; and a red warning (71-100 points) directly notifies the attending physician and suggests emergency assessment. Each warning level is associated with different treatment procedures and time requirements.

[0271] In this embodiment, when extrapolating the disease progression trajectory, the Monte Carlo simulation method is used to generate three typical development paths, including the best-case scenario, the most likely scenario, and the worst-case scenario. Each path is marked with key turning points and intervention opportunity windows, and the transition probability of each path is calculated.

[0272] In this embodiment, when linking and integrating early warning and prognosis results, a risk-prognosis matrix model is constructed, which cross-maps short-term risk levels (rows) with long-term prognosis groups (columns). Each matrix unit recommends a corresponding management strategy to achieve precise alignment between risk management and long-term goals.

[0273] This embodiment achieves systematization and precision in disease risk assessment. The scientific integration of multi-dimensional scores comprehensively reflects the complex evolution mechanism of the disease. The dynamic weight adjustment mechanism takes into account both evidence-based medicine and individual differences. The linkage design of early warning and prognosis ensures that clinical intervention focuses on both current risk control and long-term treatment goals, significantly improving the systematic nature of chronic disease management.

[0274] In some embodiments, the intelligent decision-making module 15 is used to generate a personalized management plan that includes medication adjustment plans, lifestyle intervention recommendations, and follow-up visit plans based on metabolic function characteristics, lifestyle habits, and risk prediction results.

[0275] Based on the amplitude of blood glucose fluctuations and the duration of dyslipidemia in metabolic function characteristics, combined with current medication record data, a dosage optimization algorithm is used to generate a medication adjustment plan that includes drug type, dosage and administration time.

[0276] Based on the number of days with insufficient exercise and the frequency of dietary deviations in lifestyle habits, lifestyle intervention recommendations containing specific exercise programs, exercise durations, and dietary restrictions are generated from a pre-set intervention strategy library.

[0277] By analyzing the correlation between the rate of organ failure progression and the effectiveness of disease control in the risk prediction results, a follow-up plan including necessary examinations and recommended follow-up frequencies was determined.

[0278] The medication adjustment plan was compared and verified with a drug interaction database to ensure the plan's safety.

[0279] Feasibility assessments were conducted on lifestyle intervention recommendations, and adaptive adjustments were made based on the characteristics of the patient's living environment.

[0280] Based on the examination items in the follow-up visit plan, the system automatically generates instructions on pre-examination preparations and precautions.

[0281] Integrate validated medication adjustment plans, adaptively adjusted lifestyle intervention recommendations, and follow-up appointment plans with precautions to form a complete personalized management plan;

[0282] Furthermore, reinforcement learning algorithms are used to dynamically optimize and adjust medication adjustment plans, lifestyle intervention recommendations, and follow-up visit plans based on patient feedback data.

[0283] In this embodiment, when generating a medication adjustment plan using a dose optimization algorithm, a drug response surface is established based on a pharmacokinetic-pharmacodynamic (PK-PD) model. Combining individual patient pharmacodynamic parameters (such as insulin sensitivity coefficient) and historical medication response data, a multi-objective optimization algorithm is used to balance efficacy and side effect risk, outputting a personalized dosing plan that includes the starting dose, titration regimen, and target monitoring indicators, and annotating the clinical judgment basis for each decision point.

[0284] In this embodiment, when matching lifestyle recommendations from the intervention strategy library, a three-level matching mechanism is implemented: Level 1 matching is based on standard recommendations from clinical guidelines; Level 2 matching considers data on the patient's previous intervention effects; and Level 3 matching integrates real-time environmental factors (such as seasonal changes and air quality) to generate dynamically adjusted exercise intensity formulas and personalized diets, along with alternative options to adapt to temporary changes in circumstances.

[0285] In this embodiment, when formulating the follow-up visit plan, a risk-based dynamic scheduling algorithm is adopted to convert the predicted value of organ failure risk into the optimal monitoring interval. High-risk organs are scheduled for intensive follow-up (such as kidney function tests every 2 weeks), while the interval for stable systems is extended (such as fundus examinations every 3 months). At the same time, the algorithm intelligently avoids peak working hours and times when transportation is inconvenient, thereby improving the feasibility of the plan.

[0286] In this embodiment, a multi-level review system is constructed when verifying drug interactions: the first level of review detects absolutely contraindicated combinations, the second level of review assesses clinically significant interactions, the third level of review identifies risks in special populations (such as metabolic effects in patients with hepatic or renal insufficiency), and automatically generates alternative drug recommendations and enhanced monitoring requirements for risky regimens, forming a safe medication decision tree.

[0287] In this embodiment, when assessing the feasibility of lifestyle recommendations, GIS data of the patient's living environment, family support assessment, and economic affordability analysis are integrated. For exercise recommendations, community-accessible facilities are given priority, and for dietary recommendations, local food supply and cooking habits are taken into account, ensuring that each recommendation is accompanied by three specific feasible implementation plans.

[0288] In this embodiment, when generating examination precautions, knowledge graph reasoning technology is used to automatically associate specimen collection requirements (such as fasting duration), medication adjustment requirements (such as suspending diuretics) and symptom observation points, starting from the examination item ontology, to generate two versions: a patient-specific concise list and a clinical-specific detailed description.

[0289] In this embodiment, when implementing reinforcement learning optimization, a dual-model collaborative mechanism is established: the strategy model generates intervention plans, the value model evaluates the effectiveness of the plans, and the model parameters are continuously updated through patient execution feedback data (such as medication adherence and improvement in physiological indicators) to achieve monthly iterative optimization of the plans, while retaining the clinical confirmation process for major adjustments to ensure safety.

[0290] This embodiment achieves precision and adaptability in the management plan. The personalized plan design based on multi-dimensional features significantly improves the treatment targeting and patient compliance. The strict safety verification system and feasibility assessment mechanism ensure the clinical applicability of the plan. The reinforcement learning-driven dynamic optimization enables the management system to have continuous improvement capabilities, adapt to changes in patient conditions and environmental changes, and achieve truly personalized health management.

[0291] In some embodiments, the intervention execution module 16 is used to convert the personalized management plan into executable intervention instructions, including:

[0292] The medication adjustment plan in the personalized management plan is analyzed, the drug name, dosage adjustment amount and administration time information are extracted, and a medication reminder instruction with a clear execution time is generated and output to the medication reminder unit;

[0293] Based on the exercise prescription in the lifestyle intervention recommendations, combined with the patient's real-time location and weather data, a daily exercise plan instruction containing the type, duration and intensity of exercise is generated and output to the exercise prescription execution unit;

[0294] The dietary structure adjustment suggestions are converted into specific ingredient lists and recipe recommendations, generating daily dietary guidance push instructions for three meals a day, which are then output to the dietary guidance push unit;

[0295] Based on the examination items and appointment time in the follow-up visit plan, the system automatically generates and sends a follow-up visit reminder instruction containing hospital navigation and precautions, and outputs it to the remote follow-up trigger unit.

[0296] In addition, establish an intervention instruction execution feedback mechanism to monitor medication records, exercise completion status and diet logs in real time, and generate compliance assessment reports;

[0297] When a deviation in the execution of an intervention instruction is detected, a dynamic adjustment mechanism is triggered to regenerate an optimized intervention instruction that is adapted to the current situation.

[0298] In this embodiment, when generating medication reminder instructions, an intelligent time-sharing reminder strategy is adopted. For key drugs (such as insulin), a main reminder is set 15 minutes before medication and a confirmation reminder is set 15 minutes after medication. For routine drugs, a progressive reminder is adopted (if no confirmation is made after the first reminder, a stronger reminder is given every 5 minutes, up to a maximum of 3 times). The reminder content is embedded with information such as drug images, administration methods and possible side effects to improve medication accuracy and safety.

[0299] In this embodiment, when formulating the daily exercise plan, an environmental adaptation adjustment algorithm is implemented. When the air quality index (AQI) is detected to be >100, outdoor exercise is automatically replaced with an indoor alternative. When the temperature exceeds 32°C, the exercise intensity level is reduced. The exercise duration is also dynamically adjusted in combination with the patient's real-time heart rate data to ensure a balance between exercise safety and effectiveness.

[0300] In this embodiment, when generating dietary guidance instructions, a scenario-based recommendation technology is adopted. Breakfast recommendations focus on convenience (such as recipes that can be completed in 10 minutes), lunch considers work scenario limitations (such as meals suitable for takeout), and dinner emphasizes family sharing. Based on the food inventory data of the smart refrigerator, available food combinations are recommended first to reduce execution obstacles.

[0301] In this embodiment, a three-level reminder system is established when processing follow-up visit reminders: a reminder of preparation matters (such as fasting requirements) is sent 3 days before the appointment; a reminder of transportation routes and required documents is sent 1 day before the appointment; and a final confirmation reminder is sent 2 hours in advance on the day of the appointment. The system also integrates real-time visitor flow information from the hospital navigation system to recommend the best time for the visit.

[0302] In this embodiment, when constructing the execution feedback mechanism, a multi-source data cross-validation method is adopted to perform spatiotemporal correlation analysis on the opening records of the smart pillbox, the motion monitoring of the wearable device, and the food photography records to identify the difference between the actual execution status and apparent compliance (such as opening the pillbox but not taking the medicine) and generate a compliance assessment report with confidence scores.

[0303] In this embodiment, when dynamic adjustment is triggered, a tiered intervention strategy is implemented: a gentle reminder is sent for the first deviation, alternative solutions are suggested for two consecutive deviations, and intervention by a human health manager is triggered for three or more deviations. At the same time, abnormal behavioral patterns that may require clinical attention are automatically marked to provide a basis for subsequent plan adjustments.

[0304] This embodiment realizes closed-loop control of health management. Context-aware instruction generation transforms abstract management plans into operable daily behaviors, greatly improving the feasibility of execution. Multi-level feedback mechanisms ensure timely identification of execution obstacles and clinical risks. Intelligent dynamic adjustment enables the management system to have adaptive capabilities, flexibly responding to changes in actual conditions while maintaining treatment continuity, forming a continuously optimized closed-loop health management system.

[0305] Unlike existing technologies, the above technical solution collects clinical diagnosis and treatment data, wearable device monitoring data, medication record data, and environmental monitoring data through the data acquisition module 11. After standardized preprocessing by the data fusion processing module 12, the data is deeply analyzed by the AI ​​analysis module 13 based on the Transformer architecture. Combined with the medical knowledge of the knowledge base module 14, the intelligent decision-making module 15 generates a personalized management plan, which is then implemented through the intervention execution module 16 and the intelligent interaction module 17. This technical solution processes multi-dimensional health data through a large AI model, automatically learning complex patterns and relationships, significantly improving data processing capabilities and analytical depth. The risk prediction unit 133 accurately establishes a correlation model between health indicators and disease progression, achieving precise disease prediction and risk assessment. The intelligent decision-making module 15 generates a personalized management plan based on the multi-dimensional analysis results, including medication adjustment plans, lifestyle intervention suggestions, and follow-up visit plans. Through reinforcement learning, it dynamically adjusts the plan, greatly improving its relevance and patient compliance. The chatbot unit of the intelligent interaction module 17 provides real-time interaction and personalized guidance, effectively enhancing patient participation and self-management capabilities, forming a virtuous cycle of health management.

[0306] Finally, it should be noted that although the above embodiments have been described in the text and drawings of this application, this should not limit the scope of patent protection of this application. Any technical solutions that are based on the essential concept of this application and utilize the content described in the text and drawings of this application, resulting in equivalent structural or procedural substitutions or modifications, as well as the direct or indirect application of the technical solutions of the above embodiments to other related technical fields, are all included within the scope of patent protection of this application.

Claims

1. A digital chronic disease intelligent management platform based on the fusion of AI models and multi-dimensional data, characterized in that, include: The data acquisition module is used to collect multi-source heterogeneous health data of patients according to a preset cycle. The multi-source heterogeneous health data includes clinical diagnosis and treatment data, wearable device monitoring data, medication record data and environmental monitoring data. The data fusion processing module is used to perform standardized preprocessing on the multi-source heterogeneous health data. The preprocessing includes data cleaning, time alignment and feature normalization, and outputs time-aligned structured health data. The AI ​​analysis module employs a pre-trained large model based on the Transformer architecture, which is integrated with a medical knowledge graph, including: The physiological feature extraction unit is used to extract metabolic function features and organ function indicators from physiological parameters. The behavior pattern recognition unit is used to identify lifestyle characteristics and abnormal behavior patterns from activity data; The risk prediction unit is used to establish a correlation model between multidimensional health indicators and disease progression; The knowledge base module stores medical knowledge including disease causes, symptoms, diagnostic criteria, treatment plans, and health advice, and interacts with the AI ​​analysis module through the RAG architecture. The intelligent decision-making module is used to generate a personalized management plan that includes medication adjustment plans, lifestyle intervention suggestions, and follow-up visit plans based on the metabolic function characteristics, lifestyle habits, and risk prediction results. The intervention execution module is used to convert the personalized management plan into executable intervention instructions. The intervention execution module includes a medication reminder unit, an exercise prescription execution unit, a diet guidance push unit, and a remote follow-up trigger unit. The intelligent interaction module includes a chatbot unit based on natural language processing, which receives patient inquiries and provides personalized answers based on a knowledge base module.

2. The digital chronic disease intelligent management platform based on AI model and multi-dimensional data fusion as described in claim 1, characterized in that, The data acquisition module is used to collect multi-source heterogeneous health data from patients according to a preset cycle. This multi-source heterogeneous health data includes clinical diagnosis and treatment data, wearable device monitoring data, medication record data, and environmental monitoring data, including: Acquire clinical diagnosis and treatment data, including outpatient records, laboratory reports, and imaging data, through the interface of the medical information system; Receive real-time monitoring data from wearable devices such as smart bracelets, blood glucose meters, and blood pressure monitors via Bluetooth / Wi-Fi communication protocols; Medication record data is collected through sensors in the smart pillbox or through information reported by the patient. Environmental monitoring data matching the patient's activity area is obtained through multiple environmental monitoring device interfaces; The collected clinical diagnosis and treatment data, wearable device monitoring data, medication record data, and environmental monitoring data are stored in association according to the patient ID; When missing data is detected, a supplementary data collection process is automatically triggered to obtain it. Generate complete multi-source heterogeneous health data corresponding to the current patient.

3. The digital chronic disease intelligent management platform based on AI model and multi-dimensional data fusion as described in claim 1, characterized in that, The data fusion processing module is used to perform standardized preprocessing on the multi-source heterogeneous health data. The preprocessing includes data cleaning, time alignment, and feature normalization, outputting time-aligned structured health data, including: The clinical diagnosis and treatment data are subjected to outlier detection and correction, and clinical diagnosis and treatment data that exceed the reasonable range of preset physiological parameters are removed; Align the time series of the wearable device monitoring data with a standard time reference to eliminate clock deviations between different devices; The medication record data is processed to standardize the dosage units and converted into standard units of measurement. The spatial resolution of the environmental monitoring data is adjusted to a level of accuracy that matches the patient's activity trajectory; Establish a time correlation index between the clinical diagnosis and treatment data, wearable device monitoring data, medication record data and environmental monitoring data, and perform normalization processing to make the parameters from different sources have comparable dimensions; Generate structured health data containing uniform timestamps and standardized features.

4. The digital chronic disease intelligent management platform based on AI model and multi-dimensional data fusion as described in claim 3, characterized in that, Adjusting the spatial resolution of the environmental monitoring data to a level of accuracy that matches the patient's activity trajectory includes: Based on the GPS location data recorded by the patient's smart terminal device, the spatial distribution range of the patient's activity trajectory is determined; Based on the spatial distribution range, extract the original environmental monitoring data of the corresponding geographical area from the environmental monitoring database; The original environmental monitoring data was spatially matched with the patient's activity trajectory, and supplementary environmental parameters were generated for uncovered areas using Kriging interpolation. Based on the duration of the patient's stay in the monitoring area corresponding to each original environmental monitoring data, the original environmental monitoring data and supplementary environmental parameters are time-weighted to generate weighted environmental parameters. The weighted environmental parameters are spatiotemporally correlated with the patient's activity trajectory to establish an environmental monitoring record; Based on the spatial distribution characteristics of the patient's activity trajectory, the spatial resolution of the environmental monitoring records is adjusted; Output the adjusted environmental monitoring data.

5. The digital chronic disease intelligent management platform based on AI model and multi-dimensional data fusion as described in claim 1, characterized in that, The physiological parameters include blood glucose monitoring data, blood lipid monitoring data, continuous blood pressure monitoring data, electrocardiogram monitoring data, pulmonary function test data, and renal function-related laboratory test data; The physiological feature extraction unit is used to extract metabolic function features and organ function indicators from physiological parameters, including: Blood glucose monitoring data is extracted from the structured health data, and the intraday blood glucose fluctuation characteristics are analyzed using wavelet transform method. The blood glucose coefficient of variation and blood glucose fluctuation amplitude are calculated as blood glucose metabolism characteristics. Fourier transform analysis was performed on blood lipid test data to extract the periodic change pattern of blood lipids and establish blood lipid metabolism characteristics, including the trend of total cholesterol change and the fluctuation pattern of triglycerides. Blood pressure variability indexes were calculated using continuous blood pressure monitoring data, and nonlinear dynamic analysis methods were used to obtain autonomic blood pressure regulation characteristics as blood pressure regulation features. RR interval sequences were extracted from electrocardiogram monitoring data, and heart rate variability indicators were calculated as cardiovascular function indicators using time-domain and frequency-domain analysis methods. Principal component analysis was performed on the pulmonary function test data to extract respiratory function indicators, including the rate of change of forced vital capacity and the characteristic value of maximum voluntary ventilation. Based on renal function-related laboratory test data, establish renal function assessment indicators based on changes in creatinine clearance rate and urinary protein quantification; The blood glucose metabolism characteristics, blood lipid metabolism characteristics, and blood pressure regulation characteristics are used to generate metabolic function characteristics. Furthermore, the cardiovascular function indicators, respiratory function indicators, and renal function assessment indicators are used to generate organ function indicators; Physiological feature vectors are generated based on the metabolic function characteristics and organ function indicators.

6. The digital chronic disease intelligent management platform based on AI model and multi-dimensional data fusion as described in claim 1, characterized in that, The activity data includes steps recorded by wearable devices, sleep duration and activity intensity data, dietary records, medication adherence data, and heart rate variability data; The behavior pattern recognition unit is used to identify lifestyle characteristics and abnormal behavior patterns from activity data, including: The number of steps, sleep duration and activity intensity data recorded by wearable devices are extracted from the structured health data, and time series clustering analysis is used to identify daily activity patterns. Text mining and nutritional analysis were performed on patients’ dietary records to establish dietary habit characteristics, including the distribution of eating time and nutrient intake patterns. By analyzing patient medication adherence data, we can identify medication adherence behavior characteristics by analyzing medication time deviations and missed dose frequencies. Lifestyle characteristics are generated based on the activity patterns, dietary habits, and medication adherence behaviors described above. Furthermore, stress response patterns are extracted from heart rate variability data and combined with activity data to identify abnormal behavior patterns. The characteristics of abnormal behavior episodes are classified and labeled to generate abnormal behavior patterns.

7. The digital chronic disease intelligent management platform based on AI model and multi-dimensional data fusion as described in claim 1, characterized in that, The risk prediction unit is used to establish a correlation model between multidimensional health indicators and disease progression, including: The amplitude of blood glucose fluctuations, the duration of dyslipidemia, and the coefficient of variation of blood pressure were extracted from the metabolic function characteristics as the first type of predictive indicators. The degree of decrease in heart rate variability, the rate of decline in lung function, and the trend of deterioration in kidney function were selected from the organ function indicators as the second type of predictive indicators. The number of days with insufficient exercise, frequency of dietary deviations, and number of medication non-compliances were obtained from the aforementioned lifestyle characteristics as the third type of predictive indicators; Based on historical disease course data, a dose-response relationship model between the first type of predictive index and the risk of diabetic complications was established using survival analysis methods. A nonlinear mapping relationship between the second type of predictive indicators and the rate of organ failure progression is constructed using machine learning algorithms; The correlation between the third type of predictive indicator and the disease control effect was analyzed using time series forecasting methods; The dose-response relationship model, nonlinear mapping relationship, and correlation analysis results are integrated to generate a multidimensional disease progression association model that includes short-term risk scores and long-term prognostic assessments.

8. The digital chronic disease intelligent management platform based on AI model and multi-dimensional data fusion as described in claim 7, characterized in that, The dose-response relationship model, nonlinear mapping relationship, and correlation analysis results are integrated to generate a multi-dimensional disease progression association model that includes short-term risk scores and long-term prognostic assessments, including: The complication risk probability output by the dose-response model is standardized and transformed to generate a metabolic function risk score; The rate of organ failure progression predicted by the nonlinear mapping relationship is converted into an organ function risk level score; The deviation in disease control effectiveness in the correlation analysis results was normalized to obtain a behavioral risk score; A multimodal fusion algorithm was used to integrate the metabolic function risk score, organ function risk level score, and behavioral risk score, with the weight of each score being dynamically adjusted based on clinical evidence. A tiered risk early warning mechanism is established based on the integrated comprehensive risk score. By using time-series prediction models, multiple possible trajectories of patient disease progression can be extrapolated to generate long-term prognostic assessment results; By linking and integrating the aforementioned graded risk early warning mechanism and long-term prognostic assessment results, a multi-dimensional correlation model of disease progression is formed.

9. The digital chronic disease intelligent management platform based on AI model and multi-dimensional data fusion as described in claim 1, characterized in that, The intelligent decision-making module is used to generate a personalized management plan based on the metabolic function characteristics, lifestyle habits, and risk prediction results. This plan includes medication adjustment schemes, lifestyle intervention recommendations, and follow-up appointment plans. Based on the blood glucose fluctuation range and the duration of dyslipidemia in the metabolic function characteristics, combined with the current medication record data, a dosage optimization algorithm is used to generate a medication adjustment plan that includes drug type, dosage and administration time. Based on the number of days with insufficient exercise and the frequency of dietary deviations in the aforementioned lifestyle characteristics, lifestyle intervention recommendations containing specific exercise programs, exercise durations, and dietary restrictions are generated from a pre-set intervention strategy library. By analyzing the correlation between the rate of organ failure progression and the effectiveness of disease control in the risk prediction results, a follow-up plan including necessary examinations and recommended follow-up frequencies is determined. The medication adjustment plan was compared and verified with a drug interaction database to ensure its safety. The feasibility of the proposed lifestyle interventions was assessed, and adaptive adjustments were made based on the characteristics of the patient's living environment. Based on the examination items in the follow-up visit plan, automatically generate instructions on pre-examination preparations and precautions; Integrate the validated medication adjustment plan, the adaptively adjusted lifestyle intervention recommendations, and the follow-up visit plan with precautions to form a complete personalized management plan; Furthermore, reinforcement learning algorithms are used to dynamically optimize and adjust the medication adjustment plan, lifestyle intervention suggestions, and follow-up visit plan based on patient feedback data.

10. The digital chronic disease intelligent management platform based on AI model and multi-dimensional data fusion as described in claim 1, characterized in that, The intervention execution module is used to convert the personalized management plan into executable intervention instructions, including: The medication adjustment plan in the personalized management plan is analyzed, the drug name, dosage adjustment amount and administration time information are extracted, a medication reminder instruction with a clear execution time is generated, and it is output to the medication reminder unit; Based on the exercise prescription in the lifestyle intervention recommendations, combined with the patient's real-time location and weather data, a daily exercise plan instruction containing the type, duration, and intensity of exercise is generated and output to the exercise prescription execution unit; The dietary structure adjustment suggestions are converted into specific ingredient lists and recipe recommendations, generating daily dietary guidance push instructions for three meals a day, and outputting them to the dietary guidance push unit; Based on the examination items and appointment time in the follow-up visit plan, the system automatically generates and sends a follow-up visit reminder instruction containing hospital navigation and precautions, and outputs it to the remote follow-up trigger unit. In addition, establish an intervention instruction execution feedback mechanism to monitor medication records, exercise completion status and diet logs in real time, and generate compliance assessment reports; When a deviation in the execution of an intervention instruction is detected, a dynamic adjustment mechanism is triggered to regenerate an optimized intervention instruction that is adapted to the current situation.

Images