Intelligent recommendation method and system for individualized physical examination items based on multi-modal data fusion
By using multimodal data fusion and intelligent recommendation algorithms, the technical challenges of personalized recommendations in health checkup services have been solved. This has enabled efficient screening of high-risk individuals and reduced the examination rate for low-risk individuals, promoting the transformation of health checkup services towards personalization and improving the efficiency of medical resource utilization.
Patent Information
- Application Number
- CN202511794260.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-02
- Publication Date
- 2026-03-03
AI Technical Summary
Current health check services fail to adequately consider health heterogeneity factors such as individual genetic susceptibility, lifestyle, and environmental exposure history, resulting in insufficient screening of high-risk individuals and delays in early disease diagnosis. Meanwhile, low-risk individuals receive unnecessary examinations, leading to a waste of medical resources. Furthermore, existing medical information systems lack systematic methods for integrating multi-source data, making it impossible to comprehensively characterize an individual's health status.
A multimodal data fusion approach is adopted, including collecting heterogeneous physical examination data from multiple sources, standardizing and integrating the data, constructing a physical examination knowledge graph, and combining the expert Delphi method and deep reinforcement learning algorithms to generate personalized physical examination recommendation schemes. The recommendation strategy is optimized through collaborative filtering and knowledge graph reasoning.
It has improved the screening efficiency for high-risk groups, reduced the over-examination rate for low-risk groups, realized the personalization of physical examination services, and improved the efficiency of medical resource utilization.
Smart Images
Figure CN121601137A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of data fusion processing technology, and specifically relates to a method and system for intelligent recommendation of individualized physical examination items based on multimodal data fusion. Background Technology
[0002] Current health checkup services mostly adopt a coarse-grained package model based on age and gender demographic characteristics, which fails to fully consider health heterogeneity factors such as individual genetic susceptibility, behavior and lifestyle, and environmental exposure history. This results in insufficient screening for high-risk individuals, which delays the early diagnosis of diseases, while low-risk individuals receive unnecessary examinations, leading to a waste of medical resources.
[0003] Existing medical information systems have accumulated large-scale heterogeneous health data, including structured electronic medical records, medical imaging data, genomic data, and physiological parameters from wearable devices. However, they lack systematic methods for integrating multi-source data and have failed to form an effective decision support knowledge network. Current health checkup recommendation systems suffer from technical limitations: over-reliance on a single data modality prevents them from comprehensively representing the multidimensional characteristics of an individual's health status; the application of multimodal machine learning methods in the medical field faces challenges such as difficulties in cross-modal feature alignment and insufficient model interpretability, severely restricting their clinical translational value.
[0004] The innovative application of artificial intelligence technology in the field of health management has been clearly identified as a key development direction. Therefore, developing personalized intelligent recommendation technology for physical examinations based on multimodal data fusion, and realizing the paradigm shift of physical examination services from experience-driven to evidence-driven, has important scientific significance and application value. Summary of the Invention
[0005] This invention provides a method and system for intelligent recommendation of individualized physical examination items based on multimodal data fusion, which solves the technical problems of poor accurate matching and redundancy control of physical examination items. This invention improves the screening efficiency of high-risk groups and reduces the over-examination rate of low-risk groups.
[0006] The intelligent recommendation method for individualized physical examination items based on multimodal data fusion includes the following steps:
[0007] Collect multi-source heterogeneous physical examination data from individuals; among which, multi-source heterogeneous physical examination data includes structured data, unstructured data, time-series data, and high-dimensional data; structured data comes from electronic health examination records and laboratory indicators, unstructured data comes from imaging reports and free text, time-series data comes from wearable devices, and high-dimensional data comes from the genome / proteome;
[0008] Standardize, integrate, and manage multi-source heterogeneous physical examination data; this includes cross-modal data alignment, building a unified index, and handling missing data.
[0009] Optionally, the Delphi method can be used to quantify the contribution of each data modality to the risk of a specific disease, thereby completing the initial fusion of multimodal data;
[0010] The data modalities include image data, questionnaire text data, and gene mutation data. The questionnaire text data covers information on smoking history and drinking history.
[0011] Optionally, a physical examination knowledge graph can be constructed to achieve advanced fusion of multimodal data;
[0012] The health checkup knowledge graph includes a personal basic information module, a questionnaire module, a health checkup item library module, a health checkup item combination recommendation rule library module, and a disease library module. The health checkup item library module is divided into a basic layer, a risk layer, and a customization layer.
[0013] Optionally, the construction process of the health checkup item combination recommendation rule base module is as follows: refer to the latest guidelines, standards and consensus on disease screening to formulate a basic rule base;
[0014] Medical experts regularly review, revise, and update the recommended rules in the basic rule base to ensure the scientific validity and timeliness of the rules.
[0015] Optionally, during the cold start phase, a collaborative filtering algorithm is used for user similarity matching, combined with medical rule reasoning from the physical examination knowledge graph, to generate an initial physical examination recommendation scheme; the collaborative filtering algorithm is used to solve the problem of recommendation for new users without historical data.
[0016] Optionally, during the long-term iteration phase, the user's re-examination results are used as a reward signal, and a deep reinforcement learning algorithm is used to dynamically optimize the recommendation strategy and update the physical examination recommendation scheme; the re-examination results are used to provide feedback on the screening effectiveness of the recommended items.
[0017] Optionally, it also includes: generating multi-module output results based on the optimized recommendation strategy, including risk assessment results, recommended physical examination items and health education content; the risk assessment results are presented through risk heatmaps, risk levels and disease risk interpretation details.
[0018] Optional, recommended physical examination items include a core package and flexible supplementary items. The core package consists of basic mandatory items, while the flexible supplementary items are individualized additions based on individual disease risk assessments. The flexible supplementary items include low-dose CT scans for high-risk individuals of lung cancer.
[0019] Optionally, health education content is generated using natural language processing technology and combined with patient profiles to achieve personalized matching; the health education content includes personalized health improvement suggestions, pre-examination preparation requirements and precautions.
[0020] A personalized health checkup item intelligent recommendation system based on multimodal data fusion includes: a system integration module, which has a front-end and a back-end. The front-end supports access from multiple platforms, including web and mobile devices, while the back-end adopts a microservice architecture, deploying data governance, model inference, and knowledge graph functional modules respectively. The system integration module includes:
[0021] The data acquisition module is used to collect multi-source heterogeneous physical examination data of individuals, including structured data, unstructured data, time-series data, and high-dimensional data.
[0022] The data governance module, connected to the data acquisition module, is used to standardize, integrate, and govern multi-source heterogeneous physical examination data, including cross-modal data alignment, building a unified index, and handling missing data.
[0023] The hierarchical fusion module, connected to the data governance module, includes a primary fusion unit and an advanced fusion unit. The primary fusion unit uses the expert Delphi method to quantify the contribution of each data modality, while the advanced fusion unit is used to construct a physical examination knowledge graph.
[0024] The algorithm recommendation module, connected to the hierarchical fusion module, includes a cold start recommendation unit and an iterative optimization unit. The cold start recommendation unit uses a collaborative filtering + knowledge graph reasoning algorithm, while the iterative optimization unit uses a deep reinforcement learning algorithm.
[0025] The output display module, connected to the algorithm recommendation module, is used to present risk assessment results, recommended physical examination items, and health education content.
[0026] The beneficial effects of this invention are:
[0027] This initiative promotes a shift in health checkup services from a "one-size-fits-all" approach to personalized care, improving screening efficiency for high-risk groups and reducing over-testing rates among low-risk individuals. It also aligns with medical equipment development plans, facilitating the effective implementation of AI in health management. Attached Figure Description
[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0029] Figure 1 This is a schematic diagram of the system structure of the present invention;
[0030] Figure 2 This is a flowchart of the process of the present invention;
[0031] Figure 3This is a schematic diagram of the internal structure of the data governance module of the present invention;
[0032] Figure 4 A knowledge graph architecture diagram is recommended for the personalized health check-up program of this invention;
[0033] Figure 5 This is a diagram illustrating the precise generation and dynamic optimization architecture of the personalized health checkup plan of the present invention. Detailed Implementation
[0034] The embodiments of this application will now be described in detail with reference to the accompanying drawings.
[0035] Example 1;
[0036] like Figure 1 As shown, the intelligent recommendation system for personalized physical examination items based on multimodal data fusion includes: a system integration module, which has a front-end and a back-end. The front-end supports access from multiple platforms, including web and mobile devices, while the back-end adopts a microservice architecture, deploying data governance, model reasoning, and knowledge graph functional modules respectively. The system integration module includes:
[0037] The data acquisition module is used to collect multi-source heterogeneous physical examination data of individuals, including structured data, unstructured data, time-series data, and high-dimensional data.
[0038] The data governance module, connected to the data acquisition module, is used to standardize, integrate, and govern multi-source heterogeneous physical examination data, including cross-modal data alignment, building a unified index, and handling missing data.
[0039] The hierarchical fusion module, connected to the data governance module, includes a primary fusion unit and an advanced fusion unit. The primary fusion unit uses the expert Delphi method to quantify the contribution of each data modality, while the advanced fusion unit is used to construct a physical examination knowledge graph.
[0040] The algorithm recommendation module, connected to the hierarchical fusion module, includes a cold start recommendation unit and an iterative optimization unit. The cold start recommendation unit uses a collaborative filtering + knowledge graph reasoning algorithm, while the iterative optimization unit uses a deep reinforcement learning algorithm.
[0041] The output display module, connected to the algorithm recommendation module, is used to present risk assessment results, recommended physical examination items, and health education content.
[0042] Example 2;
[0043] like Figure 2 As shown, this embodiment provides an intelligent recommendation method for individualized physical examination items based on multimodal data fusion, including the following steps:
[0044] Collect multi-source heterogeneous physical examination data from individuals; among which, multi-source heterogeneous physical examination data includes structured data, unstructured data, time-series data, and high-dimensional data; structured data comes from electronic health examination records and laboratory indicators, unstructured data comes from imaging reports and free text, time-series data comes from wearable devices, and high-dimensional data comes from the genome / proteome;
[0045] Standardize, integrate, and manage multi-source heterogeneous physical examination data; this includes cross-modal data alignment, building a unified index, and handling missing data.
[0046] like Figure 3 As shown, the data governance module includes a hospital information system (HIS), an electronic medical record system (EMR), a laboratory information system (LIS), a medical imaging system (PACS / RIS), a pathology information system (PIS), and a physical examination system. The physical examination system has data interface with wearable devices and uses HL7FHIR or API protocols to achieve real-time / batch data transmission.
[0047] The data governance module provides cross-modal data alignment, unified index building, and missing data processing. It offers web and mobile questionnaire interfaces, supports voice input (extracting keywords based on voice denoising and voice recognition technology), and image upload (e.g., photos of previous medical examination reports). It also collects users' subjective health needs and supplementary health information.
[0048] For the construction of a multimodal physical examination data governance system:
[0049] Data from the multi-source heterogeneous dataset is passed to the data governance module, which governs and integrates the multi-source data: ① Integration of different types of data: structured data (electronic health check-up records, laboratory indicators), unstructured data (image reports, free text), time-series data (wearable devices), and high-dimensional data (genomics / proteomics); ② Integration of data from different sources: questionnaire data, physical examination data, and outpatient / hospitalization data.
[0050] Layered fusion and dynamic modeling:
[0051] Primary fusion: Modal weighting based on expert Delphi method (measuring the contribution of data from different sources to disease screening).
[0052] Advanced Integration: Construct a health check knowledge graph covering multiple modules such as health questionnaires, main examination conclusion terms, diseases, health check indicators (items), and recommendation rules, supporting interpretable reasoning to build a health check knowledge base (e.g., node = disease / examination item / sign, edge = association rules supported by medical guidelines).
[0053] Recommendation algorithm / system development:
[0054] Algorithm design: ① Short-term requirements: Collaborative filtering + knowledge graph reasoning (to solve the cold start problem); ② Long-term iteration: Deep reinforcement learning (dynamically optimized based on user feedback).
[0055] Output format: Personalized physical examination recommendation plan with multiple modules. Module 1: Risk assessment (risk heat map (visualized by organ system), risk level (low / medium / high / very high risk), disease risk interpretation details, etc.); Module 2: Recommended physical examination items (core package "1" (basic items); flexible additional items (personalized additional items based on risk assessment); interpretation of examination items, pre-examination preparation and precautions, etc.); Module 3: Health improvement education and popular science.
[0056] Example 3;
[0057] Based on Example 2, such as Figure 4 As shown, the specific hierarchical fusion and dynamic modeling are as follows: An interpretable multimodal fusion model is established to support personalized risk assessment. The specific steps are as follows:
[0058] (1) Primary fusion: Modal weight allocation;
[0059] The Delphi method was used to quantify the contribution of each modality to the risk of specific diseases. For example, the weight of imaging data for lung cancer screening was 40%, the weight of smoking history (text questionnaire) was 30%, the weight of gene mutation (such as EGFR) was 20%, and others were 10%.
[0060] (2) Advanced integration: Constructing a physical examination knowledge graph.
[0061] ① Knowledge graph architecture for personalized health check-up recommendations.
[0062] The knowledge graph architecture for personalized health check-up item recommendations includes five major modules: personal basic information module, questionnaire module, health check-up item database module, health check-up item combination recommendation rule database, and disease database.
[0063] Module 1: Personal Basic Information Module;
[0064] Information includes name, gender, age, ethnicity, education level, marital status, occupation, medical card number, and ID card number.
[0065] Module Two: Questionnaire Module;
[0066] The "Health Checkup Self-Assessment Questionnaire" includes the following information: basic personal information, fertility plan, family history, personal medical history, surgical history, history of taking medications or nutritional supplements, history of physical symptoms, smoking history, history of passive smoking, history of drinking alcohol, dietary habits, depression screening scale (PHQ-9), generalized anxiety disorder scale (GAD-7), sleep quality, sedentary behavior, physical activity and exercise, long-term residence and history of residence in high-altitude areas, history of exposure to occupational hazards, female reproductive history and menstrual status, etc.
[0067] Module 3: Physical Examination Item Library Module;
[0068] Basic Tier - Basic Health Checkup Items Database: It is recommended that all examinees undergo basic health checkups for their first health checkup. The results of these basic checkups can reflect the examinee's basic health status (reference: industry standards / guidelines / expert consensus, etc.).
[0069] Risk Layer - Personalized Health Checkup Item Library: Personalized health checkup items are designed to identify individuals at high risk of disease based on multimodal information and recommend customized health checkup items for these individuals.
[0070] Customized Layer - Special Needs Health Checkup Item Library: The Special Needs Health Checkup Item Library recommends customized, in-depth health checkup items based on the examinee's subjective needs or clinical disease screening needs.
[0071] Module 4: Rule base for recommending combinations of physical examination items;
[0072] The rule base for recommending combinations of physical examination items is developed with reference to the latest guidelines, standards, norms, or consensus documents on disease screening. Based on the basic rule base, experts regularly review, revise, and update the rules for recommending individualized health examination items to ensure the scientific validity and timeliness of the recommendations.
[0073] Module 5: Disease Knowledge Base (covering ≥50 common diseases).
[0074] Example 4;
[0075] Based on Example 2, the recommendation algorithm / system development aims to achieve accurate generation and dynamic optimization of personalized health checkup plans. The specific process is as follows:
[0076] (1) Algorithm design: Cold start stage: collaborative filtering (user similarity matching) + knowledge graph reasoning (based on medical rules); long-term iteration: deep reinforcement learning (DRL), using user re-examination results as reward signals to optimize the recommendation strategy.
[0077] (2) The system output is shown in Table 1 below:
[0078] Table 1 System Output Content and Technical Implementation Methods
[0079]
[0080] (3) System integration:
[0081] Frontend: Web version + mobile version (supports access from multiple platforms);
[0082] Backend: Microservice architecture (data governance, model reasoning, and knowledge graph are deployed separately).
[0083] The operation process is as follows Figure 5 As shown;
[0084] (5) Expected outputs: a deployable intelligent recommendation system prototype; user manual and API documentation.
[0085] Example 5;
[0086] Based on all the above embodiments, the present invention has the following effects:
[0087] Multimodal data governance system: Construct a standardized solution for integrating multi-source heterogeneous physical examination data and form data governance standards.
[0088] Layered fusion model: Propose a multimodal dynamic risk perception algorithm based on expert weight allocation (primary fusion) and knowledge graph (advanced fusion), and publish 1-2 SCI / EI papers.
[0089] Algorithm innovation: Develop a hybrid recommendation algorithm that combines collaborative filtering, knowledge graph reasoning, and deep reinforcement learning, and apply for 1-2 invention patents.
[0090] Health checkup knowledge graph: covers ≥50 common diseases, includes a disease-item-rule association library, and supports interpretable recommendations.
[0091] Intelligent recommendation system prototype: Complete the development of web and mobile systems, and conduct clinical validation through a top-tier hospital / health checkup center (provide test reports).
[0092] API documentation and user manual: Supports subsequent commercial promotion or cooperative applications.
[0093] This initiative promotes a shift in health checkup services from a "one-size-fits-all" approach to personalized care, improving screening efficiency for high-risk groups and reducing over-testing rates among low-risk individuals. It also aligns with medical equipment development plans, facilitating the effective implementation of AI in health management.
[0094] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope described in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A method for intelligent recommendation of individualized physical examination items based on multimodal data fusion, characterized in that: Includes the following steps: Collect multi-source heterogeneous physical examination data from individuals; among which, multi-source heterogeneous physical examination data includes structured data, unstructured data, time-series data, and high-dimensional data; structured data comes from electronic health examination records and laboratory indicators, unstructured data comes from imaging reports and free text, time-series data comes from wearable devices, and high-dimensional data comes from the genome / proteome; Standardize, integrate, and manage multi-source heterogeneous physical examination data; this includes cross-modal data alignment, building a unified index, and handling missing data.
2. The intelligent recommendation method for individualized physical examination items based on multimodal data fusion according to claim 1, characterized in that, The Delphi method was used to quantify the contribution of each data modality to the risk of a specific disease, and the initial fusion of multimodal data was completed. The data modalities include image data, questionnaire text data, and gene mutation data, with the questionnaire text data covering information on smoking history and drinking history.
3. The intelligent recommendation method for individualized physical examination items based on multimodal data fusion according to claim 2, characterized in that, Construct a knowledge graph for physical examinations to achieve advanced fusion of multimodal data; The physical examination knowledge graph includes a personal basic information module, a questionnaire module, a physical examination item library module, a physical examination item combination recommendation rule library module, and a disease library module. The physical examination item library module is divided into a basic layer, a risk layer, and a customization layer.
4. The intelligent recommendation method for individualized physical examination items based on multimodal data fusion according to claim 3, characterized in that, The construction process of the recommended rule base module for physical examination item combinations is as follows: a basic rule base is developed by referring to the latest guidelines, standards and consensus on disease screening; Medical experts regularly review, revise, and update the recommended rules in the basic rule base to ensure the scientific validity and timeliness of the rules.
5. The intelligent recommendation method for individualized physical examination items based on multimodal data fusion according to claim 3, characterized in that, During the cold start phase, a collaborative filtering algorithm is used to match user similarity, and combined with medical rule reasoning from the physical examination knowledge graph, an initial physical examination recommendation scheme is generated. The collaborative filtering algorithm is used to solve the problem of recommending new users who have no historical data.
6. The intelligent recommendation method for individualized physical examination items based on multimodal data fusion according to claim 5, characterized in that, During the long-term iteration phase, the user's re-examination results are used as a reward signal, and a deep reinforcement learning algorithm is used to dynamically optimize the recommendation strategy and update the physical examination recommendation scheme; the re-examination results are used to provide feedback on the screening effectiveness of the recommended items.
7. The intelligent recommendation method for individualized physical examination items based on multimodal data fusion according to claim 6, characterized in that, Also includes: Based on the optimized recommendation strategy, multi-module output results are generated, including risk assessment results, recommended physical examination items, and health education content. The risk assessment results are presented through risk heatmaps, risk levels, and detailed interpretations of disease risks.
8. The intelligent recommendation method for individualized physical examination items based on multimodal data fusion according to claim 7, characterized in that, The recommended physical examination items include a core package and flexible supplementary items. The core package consists of basic mandatory items, while the flexible supplementary items are individualized additions based on individual disease risk assessments. The flexible supplementary items include low-dose CT scans for high-risk individuals of lung cancer.
9. The intelligent recommendation method for individualized physical examination items based on multimodal data fusion according to claim 7, characterized in that, The health education content is generated using natural language processing technology and combined with patient profiles to achieve personalized matching; the health education content includes personalized health improvement suggestions, pre-examination preparation requirements and precautions.
10. A personalized health checkup item intelligent recommendation system based on multimodal data fusion, used to execute the personalized health checkup item intelligent recommendation method based on multimodal data fusion according to any one of claims 1-9, comprising: The system integration module has a front-end and a back-end. The front-end supports access from multiple platforms, including web and mobile devices, while the back-end adopts a microservice architecture, deploying data governance, model reasoning, and knowledge graph functional modules respectively. The system integration module is characterized by including: The data acquisition module is used to collect multi-source heterogeneous physical examination data of individuals, including structured data, unstructured data, time-series data, and high-dimensional data. The data governance module, connected to the data acquisition module, is used to standardize, integrate, and govern the multi-source heterogeneous physical examination data, including cross-modal data alignment, construction of a unified index, and handling of missing data. The hierarchical fusion module, connected to the data governance module, includes a primary fusion unit and an advanced fusion unit. The primary fusion unit uses the expert Delphi method to quantify the contribution of each data modality, while the advanced fusion unit is used to construct a physical examination knowledge graph. The algorithm recommendation module, connected to the hierarchical fusion module, includes a cold start recommendation unit and an iterative optimization unit. The cold start recommendation unit uses a collaborative filtering + knowledge graph reasoning algorithm, while the iterative optimization unit uses a deep reinforcement learning algorithm. The output display module, connected to the algorithm recommendation module, is used to present risk assessment results, recommended physical examination items, and health education content.