Intelligent chronic disease management system based on multi-modal data fusion

Abstract

The invention discloses a chronic disease intelligent management system based on multi-modal data fusion, and relates to the technical field of intelligent management, and the system comprises a data collection and preprocessing module which is used for collecting electronic medical record data, historical medication data and living habit data, and constructing a patient database; the cross-domain knowledge fusion module is used for fusing multi-domain related medical data and a knowledge representation technology and constructing a cross-domain medical knowledge graph; the medication plan generation module is used for analyzing the physiological monitoring data, collected in real time, of the patient and generating a personalized medication plan based on the medical knowledge graph; and the dynamic optimization module is used for optimizing the personalized medication scheme by combining the medical knowledge graph according to the medication feedback data and the real-time physiological monitoring data of the personalized medication scheme. According to the method, by adopting ontology mapping and semantic similarity calculation, deep alignment of traditional Chinese medicine knowledge and modern medical knowledge is realized, and the risk of treatment scheme conflict caused by knowledge splitting is reduced.

Description

Technical Field

[0001] The present invention relates to the field of intelligent management technology, and more particularly to an intelligent chronic disease management system based on multimodal data fusion. Background Art

[0002] Currently, the management model for chronic diseases primarily relies on a management system that combines offline follow-up visits with paper health records. Doctors periodically assess patients' physiological indicators through outpatient follow-up visits, while patients rely on paper prescriptions to record their medication regimens. Although mobile medical platforms have gradually introduced medication reminder functions, existing systems generally use static rule engines based on electronic medical records, which can only implement standard dosage reminders at fixed time points. However, this single-dimensional information processing method has significant limitations: first, it ignores the correlation analysis between patients' medication compliance behavior patterns and drug blood concentration curves, and cannot implement risk warnings for behaviors such as missed doses and accidental ingestion; second, the lack of cross-domain knowledge integration leads to limited accuracy and personalization of medication reminders. Therefore, how to provide personalized and accurate medication management solutions is an urgent problem that those skilled in the art need to solve. Summary of the Invention

[0003] In view of this, the present invention provides an intelligent chronic disease management system based on multimodal data fusion, which overcomes the above-mentioned defects.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] An intelligent chronic disease management system based on multimodal data fusion, including:

[0006] The data collection and preprocessing module is used to collect electronic medical record data, historical medication data, and lifestyle data, and perform preprocessing to build a patient database;

[0007] A cross-domain knowledge fusion module is used to integrate relevant medical data and knowledge representation technologies from multiple fields, build a cross-domain medical knowledge graph, and extract associations using a large language model;

[0008] A medication plan generation module is used to analyze the patient's physiological monitoring data collected in real time and generate a personalized medication plan based on the medical knowledge graph;

[0009] A dynamic optimization module is used to optimize the personalized medication regimen through the medication feedback data and real-time physiological monitoring data of the personalized medication regimen in combination with the medical knowledge graph to form a closed-loop optimization mechanism.

[0010] Optionally, the data acquisition and preprocessing module includes:

[0011] an electronic medical record parsing unit, configured to extract diagnostic information, examination result data, and treatment record data based on the electronic medical record data, and to identify and mark abnormal values;

[0012] A medication time series extraction unit is used to parse the drug name, drug dosage and administration time data according to the historical medication data and establish a structured medication time series table;

[0013] a life habit quantification unit, configured to quantify the life habit data to generate life habit quantification data;

[0014] The preprocessing unit is used to process the data by filling missing values, correcting outliers and normalizing features to generate a structured patient database.

[0015] Optionally, the lifestyle data is health-related information, including age, eating habits, amount of exercise, sleep duration, smoking frequency, and drinking frequency.

[0016] Optionally, the cross-domain knowledge fusion module includes:

[0017] Entity recognition unit, used to perform entity recognition on fused data using the BiLSTM-CRF architecture to extract disease entities, drug categories, and behavioral features;

[0018] Data quantification unit, used to mine drug synergy and behavioral feature association rules based on disease entities, drug categories, and behavioral features through the attention mechanism;

[0019] A semantic alignment unit is used to perform knowledge semantic alignment based on the medical ontology library, the disease entity, the drug category and the behavioral characteristics using RDF triples to construct the medical knowledge graph.

[0020] Optionally, the medication plan generation module includes:

[0021] An initial prescription generating unit, used to generate an initial prescription based on the patient's basic data and historical experience data;

[0022] A prescription optimization unit, configured to call the medical knowledge graph to perform personalized optimization on the initial prescription based on the patient's physiological monitoring data to generate an optimized prescription;

[0023] The interactive display unit is used to present the optimized prescription through a visual interface and set gradient intelligent reminder trigger conditions.

[0024] Optionally, the medication plan generation module further includes:

[0025] The timing adjustment unit optimizes the medication interval of the initial prescription in combination with regional seasonal characteristics.

[0026] Optionally, the dynamic optimization module includes:

[0027] A multi-source feedback receiving unit, configured to receive medication feedback data of the optimized prescription and physiological detection data within a preset time;

[0028] a semantic association analysis unit, configured to perform cross-modal semantic analysis on the medication feedback data and the physiological detection data, associate them with the medical knowledge graph, and generate an initial optimization plan;

[0029] The dynamic optimization decision-making unit uses a reinforcement learning algorithm to build a personalized early warning model, automatically triggers the initial optimization plan adjustment strategy, and generates a personalized medication optimization plan.

[0030] Optionally, the loss function of closed-loop optimization is expressed as:

[0031] L=α||Y_pred-Y_real||_2^2+β·KL(P_optim||P_init)+γγ·Entrop(S_side);

[0032] Where α, β, and γ are weight parameters, Y is the predicted value of the physiological index, P is the probability distribution of the medication regimen, and S is the incidence rate of side effects.

[0033] It can be seen from the above technical solutions that compared with the existing technology, the present invention provides an intelligent chronic disease management system based on multimodal data fusion. By adopting ontology mapping (OWL) and semantic similarity calculation, it realizes the deep alignment of traditional Chinese medicine and modern medical knowledge, constructs a dynamic knowledge graph covering diseases, drugs, and living habits, and reduces the risk of treatment plan conflicts caused by knowledge fragmentation; through multi-task collaborative optimization, by dynamically adjusting the weight of the joint loss function based on shared underlying feature extraction and task branching architecture, it reduces the efficacy prediction error in data sparse scenarios; intelligent reminder and feedback optimization, pushes real-time medication reminders through SMS and APP, and automatically optimizes medication plans based on patient feedback to improve patient compliance. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0035] Figure 1 This is a schematic diagram of the system structure provided by the present invention. DETAILED DESCRIPTION

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0037] The embodiment of the present invention discloses an intelligent management system for chronic diseases based on multimodal data fusion, such as Figure 1 Shown, including:

[0038] The data collection and preprocessing module is used to collect electronic medical record data, historical medication data, and lifestyle data, and perform preprocessing to build a patient database;

[0039] A cross-domain knowledge fusion module is used to integrate relevant medical data and knowledge representation technologies from multiple fields, build a cross-domain medical knowledge graph, and extract associations using a large language model;

[0040] The medication plan generation module is used to analyze the real-time collected patient physiological monitoring data and generate personalized medication plans based on the medical knowledge graph;

[0041] The dynamic optimization module is used to optimize personalized medication plans through medication feedback data and real-time physiological monitoring data, combined with medical knowledge graphs, to form a closed-loop optimization mechanism.

[0042] In one embodiment, the data acquisition and preprocessing module includes:

[0043] An electronic medical record parsing unit, which is used to extract diagnostic information, examination result data, and treatment record data based on electronic medical record data, and to identify and mark abnormal values;

[0044] Medication time series extraction unit, used to parse drug names, drug dosages, and administration time data based on historical medication data and establish a structured medication time series table;

[0045] A life habit quantification unit, used to quantify life habit data and generate life habit quantification data;

[0046] The preprocessing unit is used to process the data by filling missing values, correcting outliers and normalizing features to generate a structured patient database.

[0047] In one embodiment, the lifestyle data is health-related information, including age, eating habits, amount of exercise, sleep duration, and frequency of smoking and drinking.

[0048] Furthermore, in the data acquisition and preprocessing module, data collection involves real-time acquisition of patients' electronic medical records, including disease diagnoses, examination results, and treatment records. Medication history data, including drug name, dosage, duration of use, and frequency, is extracted from prescription and purchase records. Furthermore, through wearable devices or questionnaires, patient lifestyle data is collected, including health-related information such as age, dietary habits, exercise levels, sleep duration, and smoking and drinking frequency.

[0049] Furthermore, the data collection and preprocessing module provides structured data support for chronic disease management through multi-dimensional information integration. During the data collection phase, the system connects to medical institution information systems to access patients' electronic medical records in real time, capturing key information on chronic diseases such as hypertension, diabetes, and cardiovascular disease. This includes disease diagnoses (e.g., stage 2 hypertension, type 2 diabetes, coronary artery disease), test results (systolic blood pressure 150 mmHg / diastolic blood pressure 95 mmHg, fasting blood glucose 8.2 mmol / L, ST-segment depression on electrocardiogram), and treatment records (amlodipine antihypertensive therapy, metformin glucose control regimen). Outliers are also flagged (e.g., systolic blood pressure >180 mmHg or fasting blood glucose >10.0 mmol / L). Medication history data extracts detailed medication information from prescription records and drug purchase data, including regimens such as amlodipine 5 mg taken daily at 8:00 AM and metformin 500 mg once daily. It also records the use of cardiovascular medications such as atorvastatin 20 mg. Lifestyle data were collected through wearable devices (smart blood pressure monitors, blood glucose meters) or questionnaires, such as a 55-year-old male patient's daily salt intake of 10 grams, 5,000 steps of exercise, a resting heart rate of 70 beats / minute, and high-risk behaviors such as smoking 10 cigarettes a day and drinking three times a week.

[0050] During the data preprocessing phase, a systematic cleaning strategy was implemented, including removing duplicate records, imputing missing physiological indicators with mean values ​​(e.g., defaulting to systolic blood pressure of 140 mmHg and fasting blood glucose of 7.5 mmol / L), and correcting anomalous medication data (e.g., adjusting an erroneously recorded amlodipine 20 mg three times daily to the standard 5 mg once daily). Key features were focused on, retaining core data such as a stage 2 hypertension diagnosis, amlodipine 5 mg medication history, and a daily salt intake of 10 grams. Normalization was used to convert values ​​such as age (55 years → 0.55) and medication dosage (5 mg → 0.5) into a standardized format. Ultimately, this data was integrated to form a structured patient database. A typical example is patient data package ID:001, which includes basic information about age 55, blood pressure of 150 / 95 mmHg, blood glucose of 8.2 mmol / L, and amlodipine 5 mg medication regimen, providing a high-quality data foundation for subsequent modeling.

[0051] In one embodiment, the cross-domain knowledge fusion module includes:

[0052] Entity recognition unit, used to perform entity recognition on fused data using the BiLSTM-CRF architecture to extract disease entities, drug categories, and behavioral features;

[0053] Data quantification unit, which uses attention mechanism to mine drug synergy and behavioral characteristics association rules;

[0054] The semantic alignment unit is used to perform knowledge semantic alignment based on the medical ontology library, disease entities, drug categories and behavioral characteristics using RDF triples to construct a medical knowledge graph.

[0055] Furthermore, in the cross-domain knowledge integration stage, the system integrates cross-domain knowledge, including Traditional Chinese Medicine theory, modern medical standards, and knowledge representation technology, in order to build an accurate chronic disease management model. First, Traditional Chinese Medicine theory is incorporated into the model, such as the knowledge of the correlation between "liver yang rising" and hypertension, to provide a more comprehensive health management perspective. Secondly, modern medical knowledge is integrated, such as the World Health Organization (WHO) hypertension diagnostic criteria and diabetes blood sugar control targets, to ensure that the model complies with international medical standards. Finally, through knowledge representation and alignment technology, OWL ontology language and RDF triples are used to achieve semantic alignment, resolve conflicts between terms in different fields, and ensure the accuracy and consistency of knowledge. This integration of cross-domain knowledge provides a scientific and comprehensive theoretical basis for the chronic disease management model, supporting more accurate health management decisions.

[0056] Furthermore, during the cross-domain knowledge integration phase, the system constructs a dynamic cognitive framework through multimodal data fusion and deep semantic analysis. Specifically, heterogeneous information, such as the patient's electronic medical record (e.g., a diagnosis of stage 2 hypertension and taking 5mg of amlodipine), medication schedule (daily medication record at 8:00 a.m.), and lifestyle habits (quantified data on daily salt intake of 10 grams), is first input into a large language model (LLM). This is then structured through text vectorization and feature alignment. Secondly, fine-grained entity recognition is performed based on the BiLSTM-CRF architecture, accurately extracting disease entities (e.g., associating "ST-segment depression" with coronary heart disease), drug categories (classifying "amlodipine" as a calcium channel blocker and "atorvastatin" as a statin lipid-lowering drug), and behavioral characteristics (identifying "10 cigarettes per day" as a cardiovascular risk factor). This significantly improves entity recognition accuracy (based on a test set of 100,000 medical texts).

[0057] At the relationship extraction level, the system combines a medical ontology library with an attention mechanism to construct a cross-domain association network. This not only establishes basic drug-indication relationships (e.g., "amlodipine - treatment - hypertension"), but also identifies drug synergies and antagonisms (detecting improved glycemic control in 15.3% of cases when metformin is combined with ACE inhibitors), and explores the impact of lifestyle habits on treatment efficacy. A knowledge graph constructed based on RDF triples integrates medical entities and relationship edges, supporting multidimensional reasoning (e.g., when a patient presents with the Traditional Chinese Medicine syndrome of "liver yang rising" and a systolic blood pressure >160 mmHg), a combined Chinese and Western medicine approach is preferred).

[0058] Using the hierarchical attention mechanism of the Transformer architecture, the system deeply analyzes complex patterns in chronic disease management. It identifies morning blood pressure peaks and optimizes medication timing accordingly (recommending amlodipine to be taken earlier than the default of 8:00 AM to 6:00 AM). It also discovers a nonlinear relationship between postprandial blood sugar and exercise levels in patients with type 2 diabetes (for every 1,000 additional steps of daily exercise, peak blood sugar levels decrease by 0.8 mmol / L), generating dynamic medication recommendations. Clinically validated, this model improves drug conflict detection and enhances adherence to personalized medication regimens, achieving an intelligent transition from data to decision-making.

[0059] In one embodiment, the medication plan generation module includes:

[0060] An initial prescription generating unit, used to generate an initial prescription based on the patient's basic data and historical experience data;

[0061] The prescription optimization unit is used to call the medical knowledge graph to perform personalized optimization on the initial prescription based on the patient's physiological monitoring data and generate an optimized prescription;

[0062] The interactive display unit is used to present the optimized prescription through a visual interface and set gradient intelligent reminder trigger conditions.

[0063] In one embodiment, the medication plan generation module further includes: a timing adjustment unit that optimizes the medication interval of the initial prescription in combination with regional seasonal characteristics.

[0064] Furthermore, during the medication plan generation phase, the system constructs a large-scale chronic disease medication reminder system based on automatic prompts and cross-domain information fusion. Medication plan generation is a core component, providing patients with scientific and personalized medication management plans through multi-dimensional data integration and intelligent analysis. The system first integrates the medical institution's medical conditions with the pharmacist's professional advice to form a basic medication framework. Based on this, the cross-domain knowledge fusion module utilizes a large language model (LLM) to combine medical guidelines, Traditional Chinese Medicine syndrome differentiation knowledge, and real-time patient monitoring data to conduct in-depth analysis of drug interactions and lifestyle compatibility. The system further dynamically adjusts the medication plan based on individual patient characteristics. For example, for a 55-year-old patient with hypertension and diabetes, the system automatically optimizes the medication plan based on liver function indicators, 6 hours of sleep duration, and a daily salt intake of 10 grams. The generated medication plan presents key information through a visual interface and utilizes RDF triple technology for semantic standardization to avoid terminology ambiguity. During the push and optimization phase, the system uses multi-channel intelligent reminders through app pop-ups and text messages to improve medication compliance, while dynamically adjusting the plan based on patient feedback and symptom information. Through long-term tracking data and periodic analysis of physiological indicators, iterative optimization of medication plans can be achieved, ultimately forming a closed-loop management chain of "monitoring-analysis-execution-feedback", providing chronic disease patients with full-cycle precise medication support.

[0065] Furthermore, within the chronic disease medication plan generation and optimization system, the system uses an intelligent decision-making engine to precisely design treatment plans. For example, for a 55-year-old male patient with hypertension and diabetes, the initial prescription includes amlodipine 5mg taken daily at 8:00 a.m. and metformin 500mg twice daily. The system connects to the EMR to obtain key patient indicators (systolic blood pressure 150 mmHg, fasting blood glucose 8.2 mmol / L, BMI 28.3), while also integrating pharmacist recommendations (adjusting the amlodipine dose if blood pressure is not within target) with wearable device monitoring data (daily salt intake 12 grams, resting heart rate 75 beats / minute).

[0066] The cross-domain knowledge fusion module calls the medical knowledge graph (containing 320,000 drug relationship nodes) to identify the synergistic effect of amlodipine and metformin (combined use can reduce systolic blood pressure by an additional 5.2 mmHg), but warns patients that their current salt intake may weaken the effect of the drug (the efficacy is reduced by 18% when salt intake is >10g / day). Through LLM in-depth analysis of the patient's morning peak blood pressure characteristics (average systolic blood pressure of 160 mmHg from 7 to 9 am), combined with Chinese medicine time pharmacology, the amlodipine dosing time is optimized to 6:30 and increased to 7.5 mg. At the same time, a low-sodium diet is recommended (target <6g / day). In response to the continuous excess of fasting blood glucose values ​​(average of 7.8 mmol / L for 3 months), the system recommends adjusting metformin to a sustained-release dosage form of 1000 mg to be taken at dinner based on the reinforcement learning model, which is expected to reduce glycated hemoglobin by 0.9%.

[0067] The generated medication plan is presented through a visual interface (e.g., the amlodipine dosage adjustment curve and the expected blood pressure reduction trend), and a gradient of intelligent reminders is set: a medication reminder is pushed at 6:30 a.m. (with an animation explaining the medication principle), and a blood glucose monitoring prompt is triggered 15 minutes before meals. When a patient reports ankle edema (with an incidence of 5.7%), the system immediately accesses the drug side effect knowledge base and generates an alternative plan (valsartan 80mg + hydrochlorothiazide 12.5mg) within 24 hours. The system then connects with the attending physician via a video consultation channel to confirm the adjustment. Long-term tracking data shows that after six months of dynamic optimization, the patient blood pressure target rate has increased, medication compliance has reached 92%, and the incidence of adverse drug reactions has decreased, forming an intelligent management closed loop of "monitoring-decision-execution-feedback."

[0068] In one embodiment, the dynamic optimization module includes:

[0069] A multi-source feedback receiving unit is used to receive medication feedback data for optimizing prescriptions and physiological detection data within a preset time;

[0070] Semantic association analysis unit, used to perform cross-modal semantic analysis on medication feedback data and physiological test data, associate them with medical knowledge graphs, and generate initial optimization plans;

[0071] Dynamically optimize the decision-making unit and use reinforcement learning algorithms to build a personalized early warning model, automatically trigger the initial optimization plan adjustment strategy, and generate a personalized medication optimization plan.

[0072] In one embodiment, the loss function of closed-loop optimization is expressed as:

[0073] L=α||Y_pred-Y_real||_2^2+β·KL(P_optim||P_init)+γγ·Entrop(S_side);

[0074] Where α, β, and γ are weight parameters, Y is the predicted value of the physiological indicator, P is the probability distribution of the medication regimen, and S is the incidence of side effects. Through iterative optimization using the gradient descent method, the blood pressure / blood sugar composite compliance rate was increased to ≥76%, and the incidence of adverse drug reactions was reduced by ≥38%.

[0075] Furthermore, during the user feedback and optimization phase, within the medication feedback and optimization mechanism, the system forms a dynamic closed-loop management link through multimodal data collection and intelligent analysis. Patients can submit medication feedback in real time through smartphone applications and smart home terminals. After pre-processing, the feedback information is analyzed through a large language model to correlate the semantic features of the patient's historical medication records and TCM constitution identification results to identify potential risk patterns. Based on the analysis results, using reinforcement learning algorithms, the system identifies regional differences in medication characteristics and establishes a personalized early warning model to automatically trigger medication plan adjustment strategies, forming a complete intelligent closed loop from data collection to decision optimization.

[0076] Furthermore, in the dynamic optimization mechanism driven by medication feedback, the system continuously improves treatment plans through multi-source data fusion and intelligent decision-making engines. When a 55-year-old hypertensive patient submits feedback through a smart medicine box that "after taking 10mg of amlodipine, the morning blood pressure is stable at 130 / 85mmHg", the system immediately collects the blood pressure fluctuation curve of the patient for the past 30 days (the morning peak average value dropped from 160mmHg to 142mmHg), and simultaneously integrates dynamic electrocardiogram data (ST segment depression improved by 0.1mV) and sleep quality monitored by wearable devices (deep sleep increased to 1.5 hours). The large language model (LLM) performs multi-dimensional analysis based on the medical knowledge graph: the BiLSTM-CRF model is used to identify "ankle edema" as a typical side effect of amlodipine (probability of occurrence 8.2%), combined with the patient's eGFR of 65mL / min / 1.73m 2 renal function indicators, triggering the drug replacement mechanism. The system generated an optimization plan within 2 hours: amlodipine was adjusted to 80 mg of valsartan, and the medication time was advanced from 8:00 to 6:30 based on the principle of chronopharmacology to cope with the morning blood pressure peak (data analysis showed that the morning blood pressure compliance rate increased by 27% after adjustment). In view of the patient's combined diabetes characteristics, the model found through reinforcement learning algorithms that taking 1000 mg of metformin at dinner can lower glycated hemoglobin compared to taking it for breakfast, and simultaneously pushed personalized dietary recommendations (daily salt intake is limited from 12 grams to 6 grams). The optimization plan is semantically verified through RDF triples (such as "valsartan-contraindications-bilateral renal artery stenosis") to ensure medical safety.

[0077] A continuous iteration mechanism enables the system to adapt and evolve: when monitoring increased blood pressure fluctuations among hypertensive patients in a certain region during winter, the model automatically adjusted dosage calculation parameters, significantly improving the accuracy of medication adjustments for this group within three months. After 12 months of dynamic optimization, the incidence of adverse drug reactions among patients continued to decline, the rate of achieving combined blood pressure and blood sugar targets increased, and medication compliance significantly improved, forming a closed loop of precision medicine: "data collection, intelligent decision-making, and effect verification."

[0078] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0079] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent chronic disease management system based on multimodal data fusion, characterized by: include: The data collection and preprocessing module is used to collect electronic medical record data, historical medication data, and lifestyle data, and perform preprocessing to build a patient database; A cross-domain knowledge fusion module is used to integrate relevant medical data and knowledge representation technologies from multiple fields, build a cross-domain medical knowledge graph, and extract associations using a large language model; A medication plan generation module is used to analyze the patient's physiological monitoring data collected in real time and generate a personalized medication plan based on the medical knowledge graph; A dynamic optimization module is used to optimize the personalized medication regimen through the medication feedback data and real-time physiological monitoring data of the personalized medication regimen in combination with the medical knowledge graph to form a closed-loop optimization mechanism.

2. The intelligent chronic disease management system based on multimodal data fusion according to claim 1 is characterized in that: The data acquisition and preprocessing module includes: an electronic medical record parsing unit, configured to extract diagnostic information, examination result data, and treatment record data based on the electronic medical record data, and to identify and mark abnormal values; A medication time series extraction unit is used to parse the drug name, drug dosage and administration time data according to the historical medication data and establish a structured medication time series table; a life habit quantification unit, configured to quantify the life habit data to generate life habit quantification data; The preprocessing unit is used to process the data by filling missing values, correcting outliers and normalizing features to generate a structured patient database.

3. The intelligent chronic disease management system based on multimodal data fusion according to claim 2 is characterized in that: The lifestyle data are health-related information, including age, eating habits, amount of exercise, sleep duration, smoking frequency, and drinking frequency.

4. The intelligent chronic disease management system based on multimodal data fusion according to claim 1 is characterized in that: The cross-domain knowledge fusion module includes: Entity recognition unit, used to perform entity recognition on fused data using the BiLSTM-CRF architecture to extract disease entities, drug categories, and behavioral features; Data quantification unit, used to mine drug synergy and behavioral feature association rules based on disease entities, drug categories, and behavioral features through the attention mechanism; A semantic alignment unit is used to perform knowledge semantic alignment based on the medical ontology library, the disease entity, the drug category and the behavioral characteristics using RDF triples to construct the medical knowledge graph.

5. The intelligent chronic disease management system based on multimodal data fusion according to claim 1 is characterized in that: The medication plan generation module includes: An initial prescription generating unit, used to generate an initial prescription based on the patient's basic data and historical experience data; A prescription optimization unit, configured to call the medical knowledge graph to perform personalized optimization on the initial prescription based on the patient's physiological monitoring data to generate an optimized prescription; The interactive display unit is used to present the optimized prescription through a visual interface and set gradient intelligent reminder trigger conditions.

6. The intelligent chronic disease management system based on multimodal data fusion according to claim 5 is characterized in that: The medication plan generation module also includes: The timing adjustment unit optimizes the medication interval of the initial prescription in combination with regional seasonal characteristics.

7. The intelligent chronic disease management system based on multimodal data fusion according to claim 5 is characterized in that: The dynamic optimization module includes: A multi-source feedback receiving unit, configured to receive medication feedback data of the optimized prescription and physiological detection data within a preset time; a semantic association analysis unit, configured to perform cross-modal semantic analysis on the medication feedback data and the physiological detection data, associate them with the medical knowledge graph, and generate an initial optimization plan; The dynamic optimization decision-making unit uses a reinforcement learning algorithm to build a personalized early warning model, automatically triggers the initial optimization plan adjustment strategy, and generates a personalized medication optimization plan.

8. The intelligent chronic disease management system based on multimodal data fusion according to claim 7 is characterized in that: The loss function of closed-loop optimization is expressed as: L=α||Y_pred-Y_real||_2^2+β·KL(P_optim||P_init)+γγ·Entrop(S_side); Where α, β, and γ are weight parameters, Y is the predicted value of the physiological index, P is the probability distribution of the medication regimen, and S is the incidence rate of side effects.

Images