Community cognitive weakness risk intelligent intervention system based on multi-source data fusion

Abstract

The invention discloses a community cognitive weakness risk intelligent intervention system based on multi-source data fusion. The system comprises an intelligent data gateway layer which collects multi-source data through a household patient end APP, community screening equipment and a hospital medical system; the ontology and standardization mapping layer is used for mapping the intermediate data into a standardization concept entity defined in the domain ontology; the knowledge graph construction and fusion storage layer is used for constructing and storing a panoramic knowledge graph of the cognitive decline patient; the risk early warning module is used for calculating a risk level based on the standardized feature vector and triggering an alarm when the risk level reaches an early warning threshold value; the intelligent intervention module is used for carrying out unsupervised clustering analysis and outputting a crowd classification label, and a scheme generation engine carries out reasoning, matching and combination of standardized intervention components in a classification-intervention rule knowledge graph according to the crowd classification label and a risk level to generate a personalized intervention scheme; and the three-level joint control platform comprises a hospital end, a community end and a household patient end and is used for data sharing, feedback and early warning.

Description

Technical Field

[0001] This invention relates to the field of cognitive decline risk intervention technology, and in particular to a community-based intelligent intervention system for cognitive decline risk based on multi-source data fusion. Background Technology

[0002] The number of dementia patients worldwide is increasing rapidly, and the enormous economic and social burden it brings (approximately 55 million patients globally in 2022, projected to reach 78 million by 2030). This research aims to construct a predictive model for community-based dementia patients by integrating risk and protective factors of cognitive impairment (CF) through machine learning algorithms. This will create an intelligent early warning tool and a management pathway based on key factors, enabling early intervention for community-based CF patients to reverse or delay the progression to dementia. Its significance lies in providing a scientific basis for early identification and intervention from a theoretical perspective, and in reducing the incidence of dementia to alleviate the burden on patients' families, society, and healthcare systems from a practical perspective. Simultaneously, it promotes secondary prevention, improves the quality of life for the elderly, and delays cognitive decline. Based on the characteristics of CF as a reversible pre-dementia stage (with both physical frailty and cognitive impairment), the research emphasizes that timely screening and intervention can effectively block the progression of most elderly people to dementia, ultimately achieving the strategic goal of reducing the global burden of dementia and optimizing community health management.

[0003] In existing technologies, most interventions are based on general guidelines and lack detailed differentiation of patients’ specific clinical subtypes, risk levels and individual characteristics, resulting in insufficient targeting of the program and low patient participation and compliance. Therefore, we propose an intelligent intervention system for community cognitive decline risk based on multi-source data fusion. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an intelligent intervention system for the risk of community cognitive decline based on multi-source data fusion.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A community-based intelligent intervention system for the risk of cognitive decline, based on multi-source data fusion, includes: Intelligent data gateway layer: Collects multi-source data through home patient terminal APP, community screening equipment and hospital medical system, parses the incoming non-standardized data and converts it into an internally defined standard intermediate data model, and completes the collection and standardization of multi-source data; Ontology and Standardization Mapping Layer: A medical ontology is pre-built, and by executing a terminology standardization mapping process, the medical terms, assessment items and intervention items contained in the intermediate data are mapped to standardized conceptual entities defined in the domain ontology, and the semantic relationships between entities are extracted. Knowledge graph construction and fusion storage layer: used to receive standardized concept entities and semantic relationships between entities output by the ontology and standardization mapping layer, construct and store a panoramic knowledge graph of patients with cognitive impairment, wherein the panoramic knowledge graph of patients with cognitive impairment uses standardized concept entities as nodes and semantic relationships as edges, and dynamically fuses full-dimensional information of patients. Risk warning module: A cognitive decline risk prediction model is constructed based on machine learning algorithms. The input features of the prediction model are standardized feature vectors queried and extracted from the panoramic knowledge graph of cognitive decline patients. The risk level is calculated based on the standardized feature vectors, and an alarm is triggered when the warning threshold is reached. The intelligent intervention module includes a three-dimensional subtyping model and a treatment plan generation engine. The three-dimensional subtyping model performs unsupervised clustering analysis based on standardized multimodal patient data obtained from the panoramic knowledge graph of cognitive impairment patients, and outputs population subtyping labels (clinical subgroups). The treatment plan generation engine has a built-in subtyping-patient-intervention mapping knowledge graph. The nodes and relationships of the subtyping-patient-intervention mapping knowledge graph are all based on the ontology definition of the cognitive impairment domain. The treatment plan generation engine performs reasoning in the subtyping-intervention rule knowledge graph according to the population subtyping labels and the risk level, matches and combines standardized intervention components, and generates personalized intervention plans. The three-tiered joint control platform includes hospital, community, and home-based patient terminals, enabling data sharing, feedback, and early warning. Feedback and Adjustment Module: Personalized intervention plans are dynamically adjusted through regular evaluations.

[0006] The above technical solution further includes: Furthermore, the intelligent data gateway layer includes: The protocol adapter unit has built-in drivers for multiple interface protocols, supporting HL7 FHIR, HTTP / HTTPS, and MQTT protocols, and can be used to connect to hospital HIS systems, community systems, and home devices. The data extraction and format conversion unit is used to parse non-standardized data and convert it into a standard intermediate data model defined internally by the system.

[0007] Furthermore, the terminology standardization mapping process utilizes a clustering method based on enhanced semantic representation to generate semantically adapted templates, thereby achieving accurate and efficient mapping of new terms, including: Perform medical entity recognition on the input historical text to be mapped, and extract source terms; The source terms are initially mapped to standard concept nodes in a preset medical knowledge base, and an enhanced semantic representation is generated for each successfully mapped concept node. This representation integrates its general semantic vector, hierarchical encoding in the ontology, and associated context vector. Based on the enhanced semantic representation, all concept nodes are subjected to relation-aware clustering to form high-frequency concept clusters and determine the central concept node. Based on historical mapping results, a semantic adaptation template is constructed for each high-frequency concept cluster, with the mapping text of the central concept node as the benchmark, and the adaptable units in the template are marked. For a new term to be mapped, calculate its mixed similarity with each central concept node, select the target cluster, and generate the final standardized mapping result by adapting the adaptable units in the semantic adaptation template.

[0008] Furthermore, the panoramic knowledge graph of patients with cognitive decline is centered on the individual patient and connects the entity and relationship network of the patient's diagnosis, symptoms, assessment results, physiological parameters, and intervention records.

[0009] Furthermore, the three-dimensional subtyping model employs hierarchical clustering or autoencoder algorithms to subgroup patients based on multi-dimensional data extracted from the panoramic knowledge graph of cognitive impairment patients and standardized by ontology, and outputs subtyping labels.

[0010] Furthermore, the classification-patient-intervention mapping knowledge graph includes: Core entities: individual patients, population subtypes, intervention components, and intervention protocol templates; Core relationships: belonging relationship, corresponding relationship, containing relationship, adapting relationship, taboo relationship, strengthening or weakening relationship.

[0011] Furthermore, the intervention components include a movement component, a cognitive component, a nutrition component, and an education component.

[0012] Furthermore, the risk prediction model of the risk warning module is constructed using machine learning algorithms. The features used in its training and inference processes are all standardized features that have been unified at the ontology layer and have clear semantics by querying the panoramic knowledge graph of cognitive impairment patients. The risk warning module uses support vector machine, decision tree or random forest algorithms to construct the prediction model and outputs the risk level according to the patient's classification label. When abnormal data is detected, an early warning is automatically triggered and the three-level joint control platform is notified.

[0013] Furthermore, the workflow of the intelligent intervention module includes: Patient subtyping and risk assessment: Based on a three-dimensional subtyping model and a risk prediction model, subtyping labels and risk levels are output; Protocol template matching: Matching pre-defined intervention protocol templates using a classification-patient-intervention mapping knowledge graph; Personalized component optimization: Dynamically adjust the intensity, frequency, and content of intervention components based on adaptation rules, taboo rules, and risk levels; The final personalized intervention plan is generated and pushed to the three-level joint control platform.

[0014] Furthermore, the feedback and optimization module utilizes graph learning algorithms to analyze the subgraph patterns of the high-response patient group in the panoramic knowledge graph of cognitive decline patients, discover their potential associations with specific standardized intervention components in the graph, and update the associations to the classification-patient-intervention mapping knowledge graph in the form of new or enhanced semantic relationships.

[0015] The present invention has the following beneficial effects: In this invention, a knowledge graph mapping between subtype, individual patient, and intervention is used to generate personalized intervention plans. This makes the intervention plans highly consistent with the patient's current subtype, risk level, and individual characteristics, providing personalized health guidance for different types of cognitive impairment populations, which greatly improves the pertinence and expected effect of the intervention. Attached Figure Description

[0016] Figure 1 This is a system block diagram of the intelligent intervention system for community cognitive decline risk based on multi-source data fusion proposed in this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 As shown, this invention is an intelligent intervention system for the risk of community cognitive decline based on multi-source data fusion, comprising: Intelligent data gateway layer: Collects multi-source data through home patient terminal APP, community screening equipment and hospital medical system, parses the incoming non-standard data and converts it into an internally defined standard intermediate data model, and completes the collection and standardization of multi-source data. More specifically, the multi-source data includes patient basic information registration data, assessment and screening scale scores and measurements, medical data related records, and training compliance and effect data generated during the intervention implementation process. Ontology and Standardization Mapping Layer: A medical ontology is pre-built, and by executing a terminology standardization mapping process, medical terms, assessment items and intervention items contained in the intermediate data (e.g., education level, occupation, etc. in patient basic information registration, AD8 score, grip strength value, MMSE score, etc. in assessment and screening, and diagnosis, test results, etc. in medical data related records) are mapped to standardized conceptual entities defined in the domain ontology, and semantic relationships between entities are extracted; Knowledge graph construction and fusion storage layer: used to receive standardized concept entities and semantic relationships between entities output by the ontology and standardization mapping layer, construct and store a panoramic knowledge graph of patients with cognitive impairment, wherein the panoramic knowledge graph of patients with cognitive impairment uses standardized concept entities as nodes and semantic relationships as edges, and dynamically fuses full-dimensional information of patients. Risk warning module: A cognitive decline risk prediction model is constructed based on machine learning algorithms. The input features of the prediction model are standardized feature vectors queried and extracted from the panoramic knowledge graph of cognitive decline patients. The risk level is calculated based on the standardized feature vectors, and an alarm is triggered and pushed to the corresponding port when the warning threshold is reached. The intelligent intervention module includes a three-dimensional subtyping model and a treatment plan generation engine. The three-dimensional subtyping model performs unsupervised clustering analysis based on standardized multimodal patient data obtained from the panoramic knowledge graph of cognitive impairment patients, and outputs population subtyping labels (clinical subgroups). The treatment plan generation engine has a built-in subtyping-patient-intervention mapping knowledge graph. The nodes and relationships of the subtyping-patient-intervention mapping knowledge graph are all based on the ontology definition of the cognitive impairment domain. The treatment plan generation engine performs reasoning in the subtyping-intervention rule knowledge graph according to the population subtyping labels and the risk level, matches and combines standardized intervention components, and generates personalized intervention plans. The three-tiered joint control platform includes hospital, community, and home-based patient terminals, enabling data sharing, feedback, and early warning. More specifically, the hospital terminal provides support for treatment decision-making, plan review, critical value warning, remote guidance, and data statistical analysis; the community terminal handles screening implementation, intervention plans, risk grading, referral management, and family follow-up; and the home-based patient terminal provides education and knowledge learning, self-training, chronic disease risk data monitoring, emergency calls, medication reminders, and nutritional monitoring. Feedback and Adjustment Module: Personalized intervention plans are dynamically adjusted through regular evaluations.

[0019] In one embodiment, the intelligent data gateway layer includes: The protocol adapter unit has built-in drivers for multiple interface protocols, supporting HL7 FHIR, HTTP / HTTPS, and MQTT protocols, and can be used to connect to hospital HIS systems, community systems, and home devices. The data extraction and format conversion unit is used to parse non-standardized data and convert it into a standard intermediate data model defined internally by the system.

[0020] In one embodiment, the terminology standardization mapping process utilizes a clustering method based on enhanced semantic representation to generate semantically adapted templates, thereby achieving accurate and efficient mapping of new terms, including: Perform medical entity recognition on the input historical text to be mapped, and extract source terms; The source terms are initially mapped to standard concept nodes in a preset medical knowledge base, and an enhanced semantic representation is generated for each successfully mapped concept node. This representation integrates its general semantic vector, hierarchical encoding in the ontology, and associated context vector. Based on the enhanced semantic representation, all concept nodes are subjected to relation-aware clustering to form high-frequency concept clusters and determine the central concept node. Based on historical mapping results, a semantic adaptation template is constructed for each high-frequency concept cluster, with the mapping text of the central concept node as the benchmark, and the adaptable units in the template are marked. For a new term to be mapped, calculate its mixed similarity with each central concept node. More specifically, the formula for calculating the mixed similarity is: S ,in, For general semantic similarity, Based on the similarity of the ontology hierarchy, Similarity is based on the associated context, and the weight coefficients satisfy... To enhance the role of ontology structure and medical logic in mapping decisions, after selecting the target cluster, the final standardized mapping result is generated by adapting the adaptable units in the semantic adaptation template.

[0021] In one embodiment, the panoramic knowledge graph of cognitive impairment patients is centered on the individual patient and connects the patient's diagnosis, symptoms, assessment results (MMSE, MoCA, grip strength, etc.), physiological parameters, and intervention records into an entity and relationship network. More specifically, the nodes include the patient's basic information, physiological parameters, cognitive function assessment results, medical records, and intervention records, while the edges include the relationships between the patient and the classification, the classification and the intervention, and the intervention and the effect.

[0022] In one embodiment, the three-dimensional classification model employs hierarchical clustering or autoencoder algorithms to subgroup patients based on ontology-standardized multi-dimensional data (including pathological biomarkers, frailty phenotypes, cognitive domain scores, and daily functional status) extracted from the panoramic knowledge graph of the cognitively impaired patients, and outputs clinically meaningful classification labels.

[0023] It should be noted that the specific analysis process for the clinically significant typing labels output by the three-dimensional typing model is as follows: Multi-dimensional feature data acquisition and feature matrix construction: From the panoramic knowledge graph of patients with cognitive impairment, multi-dimensional feature data of the target patient group are extracted according to the preset graph pattern.

[0024] The core dimensions and specific features extracted include: Pathological biomarker dimension: Extracting standardized laboratory test results from medical records; Weak phenotype dimension: Extract indicators standardized based on the Fried phenotype from the evaluation results; Cognitive domain scoring dimensions: Standardized scores for multiple cognitive domains are extracted from the assessment results; Daily functional status dimension: Extract scores from the instrumental activities of daily living scale, basic activities of daily living scale, and regular activity pattern data uploaded from the community or home devices from the assessment results and patient attributes; All extracted numerical features are normalized, and categorical features are one-hot encoded to construct a unified high-dimensional feature vector for each patient, which is then integrated into a feature matrix for all patients.

[0025] High-dimensional feature reduction: PCA is used to reduce the dimensionality of the feature matrix; Unsupervised clustering and subgroup identification Based on the dimensionality-reduced feature matrix, unsupervised machine learning algorithms are used for cluster analysis to identify intrinsic patient subgroups (i.e., classifications). Specifically, one of the following two algorithms is implemented: Hierarchical clustering algorithm: Agglomerative hierarchical clustering is adopted. First, each patient is regarded as an independent cluster. The similarity or distance between all clusters is calculated. The two closest clusters are iteratively merged to form a new cluster. This process uses the Ward connection method to minimize the increase in intra-cluster variance after merging. By observing the dendrogram and combining the rationality of clinical interpretation, the dendrogram is cut at an appropriate level to determine the final number of clusters and the cluster label of each patient (i.e., the preliminary "population typing"). Autoencoder algorithm: Construct a stacked autoencoder, whose encoder part compresses the high-dimensional input features into a low-dimensional latent space representation. On the latent space representation output by the encoder, apply the K-Means clustering algorithm. Simultaneously train the autoencoder (aiming to minimize the loss between the input high-dimensional features and the decoder reconstructed output) and K-Means clustering (aiming to minimize the intra-cluster distance). Through joint optimization, the autoencoder learns features that are both conducive to data reconstruction and enable the latent space representation to have a good clustering structure. After training, the latent space representation of all patients is obtained through the encoder, and the final cluster label of each patient is output by the K-Means algorithm. Clinical interpretation and validation of subtyping labels: For each cluster (subgroup) generated by the clustering algorithm, multi-dimensional feature profiling is performed. The mean or distribution of the subgroup on all original features (as listed in the multi-dimensional feature data acquisition and feature matrix construction) is calculated and compared with the overall mean or other subgroups. Based on the feature profiling results, each subgroup is assigned a clinically meaningful subtyping label. The effectiveness and stability of the subtyping are confirmed through internal validation (profile coefficient to assess clustering quality) and external validation (comparing differences among different subgroups in known clinical outcomes such as future dementia conversion rate and rate of functional decline). Knowledge graph integration and dynamic updating of typing results: The finalized typing labels are used as new population typing entity nodes, and are imported or updated in batches to the panoramic knowledge graph of cognitive impairment patients, establishing a periodic or triggered retraining mechanism for the typing model. When the new patient data accumulated in the knowledge graph reaches a certain scale, or when clinical experts propose significant revisions to the existing typing, the process from step one is automatically or manually triggered to update the typing information in the typing model and the graph using the full amount of data, ensuring the synchronous evolution of typing and clinical practice.

[0026] In one embodiment, the typing-patient-intervention mapping knowledge graph includes: Core entities: individual patients, population subtypes, intervention components, and intervention protocol templates; Core relationships: Belonging relationship (connecting individual patients with population subtypes), Corresponding relationship (connecting population subtypes with intervention program templates), Containing relationship (connecting intervention program templates with intervention components), Adapting relationship (connecting intervention components with patient characteristics), Contraindication relationship (connecting intervention components with patient characteristics), Strengthening or weakening relationship (connecting risk warning level with intervention intensity or frequency).

[0027] In one embodiment, the intervention components include a movement component, a cognitive component, a nutrition component, and an education component; More specifically, including: Exercise components: Baduanjin (Eight Pieces of Brocade), rhythmic stepping, resistance band training; Cognitive components: number repetition games, supermarket shopping games; Nutritional components: Mediterranean diet plan, high-protein soft food program; Educational components: Videos on the dangers of cognitive decline, and illustrated guides on adhering to antihypertensive medication.

[0028] In one embodiment, the risk prediction model of the risk warning module is constructed using machine learning algorithms. The features used in its training and inference processes are all standardized features that have been unified at the ontology layer and have clear semantics by querying the panoramic knowledge graph of the cognitive decline patients. The risk warning module uses support vector machine, decision tree or random forest algorithms to construct the prediction model and outputs the risk level according to the patient's classification label. When abnormal data is detected, an early warning is automatically triggered and the three-level joint control platform is notified.

[0029] In this embodiment: Abnormal data includes, but is not limited to: If the patient's AD8 self-assessment score is ≥2, contact the community. A patient's FRAIL self-assessment score of ≥3 points triggers a contact with the community. A score of ≤2 on the Community Assessment Scale Mini-Cog will trigger a contact with the hospital. If the MMSE or MoCA assessment score is ≤26, a regular contact reminder will be triggered. If a blood pressure reading greater than 180 / 110 mmHg or lower than 90 / 60 mmHg is detected, it will trigger a connection between the community and hospital. If a blood glucose level is detected to be higher than 22.2 mmol / L or lower than 2.5 mmol / L, it will trigger communication with the community and hospital. The system detected a weight loss of more than 10% over the past three months, triggering communication with the community and hospital.

[0030] In one embodiment, the workflow of the intelligent intervention module includes: Patient subtyping and risk assessment: Based on a three-dimensional subtyping model and a risk prediction model, subtyping labels and risk levels are output; Protocol template matching: Matching pre-defined intervention protocol templates using a classification-patient-intervention mapping knowledge graph; Personalized component optimization: Dynamically adjust the intensity, frequency, and content of intervention components based on adaptation rules, taboo rules, and risk levels; The final personalized intervention plan is generated and broken down into tasks, which are then pushed to the three-level joint control platform.

[0031] In one embodiment, the feedback and optimization module uses a graph learning algorithm to analyze the subgraph patterns of the high-response patient group in the panoramic knowledge graph of cognitive impairment patients, discover their potential associations with specific standardized intervention components in the graph, and update the associations to the classification-patient-intervention mapping knowledge graph in the form of new or enhanced semantic relationships.

[0032] It should be noted that the specific analysis process for the clinically significant typing labels output by the three-dimensional typing model is as follows: Constructing and labeling the intervention effect subgraph: From the panoramic knowledge graph of patients with cognitive decline, extract patient nodes and their associated data that have completed the intervention and received regular assessments within the preset intervention period to form an intervention effect analysis subgraph. Add effect labels to the patient nodes in the intervention effect analysis subgraph. The effect labels include at least two categories: high response and low response. The determination is based on the magnitude of change in cognitive function assessment scores (such as MMSE, MoCA) and the degree of improvement in decline indicators (such as grip strength, walking speed). Pattern discovery based on graph neural networks: The subgraphs of intervention effect analysis with effect labels are input into the graph neural network model for training. The graph neural network model learns to extract embedding representations from the attribute features of patient nodes (such as pathological indicators and functional status), the intervention component nodes connected to them, and the association paths between them. Through model analysis, subgraph patterns that are significantly related to high response labels are identified. The subgraph patterns represent the co-occurrence and association structure of specific combinations of patient features and specific sets of intervention components in the graph. Generate and validate candidate association rules: Based on the high-response subgraph patterns identified by pattern discovery based on graph neural networks, candidate association rules are generated. The candidate association rules are expressed as follows: when a patient has a feature set {P} and belongs to a population subtype {T}, using the intervention component set {I} or a specific combination thereof is expected to achieve better intervention results (expected response improvement). The candidate association rules are validated in historical data or new validation patient subgraph backtracking data, and their confidence and support are calculated. Update the intervention mapping knowledge graph: Transform the validated candidate association rules into new semantic relationships or strengthen the weights of existing relationships, and update them in the classification-patient-intervention mapping knowledge graph; The specific update operations include: Establish or strengthen an efficient correspondence between specific population subtyping nodes and intervention component nodes; Add new adaptive feature attributes to the intervention component nodes, pointing to the patient feature set {P} that has a high response to them; Adjust the association weights so that when generating treatment plans for patients with feature {P}, the system prioritizes or enhances the configuration of the intervention component set {I}. Program iteration and system evolution: By utilizing the updated subtyping-patient-intervention mapping knowledge graph, a new generation of personalized intervention plans is generated for newly enrolled patients. New rounds of intervention effect data are continuously collected, and personalized intervention plans are dynamically adjusted through repeated evaluation.

[0033] All data used in this invention has been authorized by the user and complies with legal regulations.

[0034] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart intervention system for community cognitive decline risk based on multi-source data fusion, characterized in that, include: Intelligent data gateway layer: Collects multi-source data through home patient terminal APP, community screening equipment and hospital medical system, and parses the incoming non-standard data and converts it into an internally defined standard intermediate data model; Ontology and Standardization Mapping Layer: A medical ontology is pre-built, and by executing a terminology standardization mapping process, the medical terms, assessment items and intervention items contained in the intermediate data are mapped to standardized conceptual entities defined in the domain ontology, and the semantic relationships between entities are extracted. Knowledge Graph Construction and Fusion Storage Layer: Constructs and stores a panoramic knowledge graph of patients with cognitive impairment. The panoramic knowledge graph of patients with cognitive impairment uses standardized conceptual entities as nodes and semantic relationships as edges, and dynamically fuses the full-dimensional information of patients. Risk warning module: Constructs a cognitive decline risk prediction model. The input features of the prediction model are standardized feature vectors queried and extracted from the panoramic knowledge graph of cognitive decline patients. The risk level is calculated based on the standardized feature vectors, and an alarm is triggered when the warning threshold is reached. The intelligent intervention module includes a three-dimensional subtyping model and a treatment plan generation engine. The three-dimensional subtyping model performs unsupervised clustering analysis based on multimodal patient data obtained from the panoramic knowledge graph of cognitive impairment patients, and outputs population subtyping labels. The treatment plan generation engine has a built-in subtyping-patient-intervention mapping knowledge graph. The nodes and relationships of the subtyping-patient-intervention mapping knowledge graph are all based on the ontology definition of the cognitive impairment domain. The treatment plan generation engine performs reasoning in the subtyping-intervention rule knowledge graph according to the population subtyping labels and the risk level, matches and combines standardized intervention components, and generates personalized intervention plans. The three-tiered joint control platform includes hospital, community, and home-based patient terminals, enabling data sharing, feedback, and early warning. Feedback and Adjustment Module: Personalized intervention plans are dynamically adjusted through regular evaluations.

2. The intelligent intervention system for community cognitive decline risk based on multi-source data fusion according to claim 1, characterized in that: The intelligent data gateway layer includes: The protocol adapter unit has built-in drivers for multiple interface protocols, supporting HL7 FHIR, HTTP / HTTPS, and MQTT protocols, and can be used to connect to hospital HIS systems, community systems, and home devices. The data extraction and format conversion unit is used to parse non-standardized data and convert it into a standard intermediate data model defined internally by the system.

3. The intelligent intervention system for community cognitive decline risk based on multi-source data fusion according to claim 1, characterized in that: The terminology standardization mapping process utilizes a clustering method based on enhanced semantic representation to generate semantically adapted templates, including: Perform medical entity recognition on the input historical text to be mapped, and extract source terms; The source terms are initially mapped to standard concept nodes in a preset medical knowledge base, and an enhanced semantic representation is generated for each successfully mapped concept node. Based on the enhanced semantic representation, all concept nodes are subjected to relation-aware clustering to form high-frequency concept clusters and determine the central concept node. Based on historical mapping results, a semantic adaptation template is constructed for each high-frequency concept cluster, with the mapping text of the central concept node as the benchmark, and the adaptable units in the template are marked. For a new term to be mapped, calculate its mixed similarity with each central concept node, select the target cluster, and generate the final standardized mapping result by adapting the adaptable units in the semantic adaptation template.

4. The intelligent intervention system for community cognitive decline risk based on multi-source data fusion according to claim 1, characterized in that: The panoramic knowledge graph for patients with cognitive impairment is centered on the individual patient and connects entities and relationships related to the patient's diagnosis, symptoms, assessment results, physiological parameters, and intervention records.

5. The intelligent intervention system for community cognitive decline risk based on multi-source data fusion according to claim 1, characterized in that: The three-dimensional classification model uses hierarchical clustering or autoencoder algorithms to subgroup patients based on multi-dimensional data extracted from the panoramic knowledge graph of cognitive impairment patients and standardized by ontology, and outputs classification labels.

6. The intelligent intervention system for community cognitive decline risk based on multi-source data fusion according to claim 1, characterized in that: The classification-patient-intervention mapping knowledge graph includes: Core entities: individual patients, population subtypes, intervention components, and intervention protocol templates; Core relationships: belonging relationship, corresponding relationship, containing relationship, adapting relationship, taboo relationship, strengthening or weakening relationship.

7. The intelligent intervention system for community cognitive decline risk based on multi-source data fusion according to claim 6, characterized in that: The intervention components include a movement component, a cognitive component, a nutrition component, and an education component.

8. The intelligent intervention system for community cognitive decline risk based on multi-source data fusion according to claim 1, characterized in that: The risk prediction model of the risk warning module is constructed using machine learning algorithms. The features used in its training and inference processes are all standardized features that have been unified at the ontology layer and have clear semantics by querying the panoramic knowledge graph of cognitive impairment patients. The risk warning module uses support vector machine, decision tree or random forest algorithms to construct the prediction model and outputs the risk level according to the patient's classification label. When abnormal data is detected, an early warning is automatically triggered and the three-level joint control platform is notified.

9. The intelligent intervention system for community cognitive decline risk based on multi-source data fusion according to claim 1, characterized in that: The workflow of the intelligent intervention module includes: Patient subtyping and risk assessment: Based on a three-dimensional subtyping model and a risk prediction model, subtyping labels and risk levels are output; Protocol template matching: Matching pre-defined intervention protocol templates using a classification-patient-intervention mapping knowledge graph; Personalized component optimization: Dynamically adjust the intensity, frequency, and content of intervention components based on adaptation rules, taboo rules, and risk levels; The final personalized intervention plan is generated and pushed to the three-level joint control platform.

10. The intelligent intervention system for community cognitive decline risk based on multi-source data fusion according to claim 1, characterized in that: The feedback and optimization module uses graph learning algorithms to analyze the subgraph patterns of high-response patient groups in the panoramic knowledge graph of cognitive impairment patients, discover their potential associations with specific standardized intervention components in the graph, and update the associations to the classification-patient-intervention mapping knowledge graph in the form of new or enhanced semantic relationships.

Images