A knowledge graph-based cross-institutional patient follow-up data integration method and system

Abstract

The application discloses a kind of based on knowledge graph's cross-institutional patient follow-up data integration method and system, belong to medical data processing field.The method first obtains the multiple types of follow-up data such as electronic medical record, examination report, follow-up record of cross-institutional, then constructs the multi-source heterogeneous data integration model of five-layer architecture containing heterogeneous data adaptation layer, entity analysis layer etc., data is standardized after training processing;Subsequently, the entity, relationship and feature information in the data are extracted, the direct and indirect correlation between entities are established, and the unified patient health record is formed by fusion;Finally, receive patient identification information, match and present corresponding data according to preset logic arrangement.The scheme effectively eliminates the differences in format, field and semantics of cross-institutional data, realizes data centralized acquisition, improves the accuracy and integrity of data integration, and provides reliable support for medical decision-making and cross-institutional medical cooperation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical data processing, in particular to a cross-institutional patient follow-up data integration method and system based on a knowledge graph. BACKGROUND

[0002] In the medical and health field, cross-institutional diagnosis and treatment and long-term follow-up have become an important mode to improve the quality of medical services. More and more patients will transfer between different medical institutions to receive diagnosis, treatment and follow-up services due to their medical needs. In this process, a large amount of data related to patient follow-up will be generated, including electronic medical records, examination reports, follow-up records, rehabilitation training trajectory data, patient self-management log data, and cross-institutional referral transfer certificate data, etc. At present, each medical institution will collect and store these data according to its own business processes and management needs, forming independent data management systems. The industry has gradually realized the importance of cross-institutional data integration. Some institutions have tried to promote data sharing by establishing data transmission interfaces and developing basic data collection specifications. Some preliminary data integration technologies have also been applied to local scenarios, but overall, the data is still in a state of scattered storage and separate operation. The systematicness and completeness of data integration need to be improved.

[0003] With the further promotion of cross-institutional medical cooperation, the existing data management and integration mode gradually exposes obvious technical bottlenecks. Due to the different information system construction standards of different medical institutions, the data storage format, field definition and semantic expression lack unified specifications, resulting in significant heterogeneity of cross-institutional patient follow-up data. Data from different sources cannot be directly interchanged and cooperatively applied. At the same time, existing technologies fail to effectively extract entity information, relationship information and feature information from the data, and cannot establish direct and indirect association relationships between entities, making patient data scattered in different institutions isolated and difficult to form a unified and complete patient health record. In addition, the lack of a specialized integration model for cross-institutional multi-source heterogeneous follow-up data results in a lack of standardized processes for data processing. When users need to retrieve cross-institutional follow-up data, they cannot quickly and centrally obtain the required information, which not only affects the continuity and efficiency of medical services, but also hinders the scientificity of medical decision-making, making it difficult to meet the needs of cross-institutional medical cooperation for complete and unified patient data. SUMMARY

[0004] The present application aims to overcome the shortcomings of the prior art and provide a cross-institutional patient follow-up data integration method and system based on a knowledge graph.

[0005] The purpose of the present application is achieved by the following technical solutions:

[0006] A knowledge graph-based cross-institutional patient follow-up data integration method is provided, which comprises the following steps:

[0007] S1. Obtain cross-institutional patient follow-up related data, including electronic medical records, examination reports, follow-up records, rehabilitation training trajectory data, patient self-management log data, and cross-institutional referral handover credential data;

[0008] S2. Construct a multi-source heterogeneous data integration model, which sets up a five-layer architecture including a heterogeneous data adaptation layer, an entity resolution layer, a semantic enhancement layer, a dynamic correlation layer, and an attribute mapping layer. After training the model, the obtained patient follow-up related data is standardized based on the model, and the data format, field definition, and semantic expression are unified;

[0009] S3. Extract entity information, relationship information, and feature information from the standardized data by using semantic mapping technology combined with model dynamic correlation layer rules, establish direct and indirect correlation relationships between entities, and integrate and form a unified patient health record;

[0010] S4. Receive patient identification information, match corresponding patient follow-up related data based on the unified patient health record and model attribute mapping layer rules, and present them according to the preset logic, realizing centralized acquisition of cross-institutional data.

[0011] Further, step S1 comprises:

[0012] S1.1. Determine the source institution and data transmission protocol of the cross-institutional patient follow-up related data;

[0013] S1.2. Collect patient follow-up related data from the source institution according to the determined data transmission protocol, including real-time interface docking, timed batch import, and offline file transmission;

[0014] S1.3. Perform integrity check and validity check on the collected patient follow-up related data, including core field missing check and data record integrity verification, and validity check including data format compliance check and logical consistency verification;

[0015] S1.4. Retain the patient follow-up related data that passes the check, and temporarily store the patient follow-up related data that fails the check after labeling the abnormal type, for subsequent supplement or correction processing.

[0016] Further, step S2 comprises:

[0017] S2.1. Construct a multi-source heterogeneous data integration model, configure each layer to realize connection through an adaptive component, an encoding module, and a feature enhancement component, specify that the heterogeneous data adaptation layer processes different types of follow-up data, configure the entity analysis layer to include entity feature extraction and classification submodules, set the semantic enhancement layer to enhance semantic features through a multi-layer neural network, configure the dynamic association layer to include association rules, derivation, and strength calculation units, and specify that the attribute mapping layer is responsible for entity attribute definition and matching;

[0018] S2.2. Construct a training set and a validation set using the labeled multi-source heterogeneous medical data across institutions, construct a joint loss function composed of a heterogeneous data adaptation loss function, an entity matching loss function, and a semantic association loss function, set an initial learning rate, an iteration batch, and a model convergence threshold, alternately adjust the parameters of each layer through a backpropagation algorithm, and sequentially optimize the adaptation accuracy, analysis accuracy, and association matching degree until the comprehensive adaptation accuracy of the validation set is stable.

[0019] S2.3. Develop standardized rules based on the model structure, and complete data standardization through format adaptation, entity encoding, semantic enhancement, and attribute matching.

[0020] Further, step S3 includes:

[0021] S3.1. Extract entity information, relationship information, and feature information from the standardized patient follow-up related data, entity information includes unified encoding and attribute values of each entity, relationship information includes direct association types and indirect association clues between entities, and feature information includes time sequence features, text features, and structured features.

[0022] S3.2. Based on the rules of the dynamic association layer of the multi-source heterogeneous data integration model, use an improved cosine similarity algorithm to calculate the semantic similarity of entities in different institutional patient follow-up related data, combine the dynamic weights of the association strength calculation unit, set a similarity threshold as the association determination condition, and establish direct association relationships between entities.

[0023] S3.3. Through the indirect association derivation unit of the dynamic association layer, derive indirect association relationships between entities based on attribute association clues, time sequence association clues, and event association clues, calculate indirect association strength and compare it with a preset strength threshold, and retain valid indirect association relationships.

[0024] S3.4. Based on the established direct association relationships and valid indirect association relationships, associate and fuse different institutional, different type patient follow-up related data corresponding to the same patient, maintain the consistency of entity unified encoding, attribute information, and association relationships during the fusion process, and organize them into a unified patient health record according to the classification of the entity analysis layer and the structure of the attribute mapping layer.

[0025] Further, step S4 includes:

[0026] S4.1. receiving patient identification information in the data retrieval request, the patient identification information including a patient unique number, identity information, and entity uniform coding;

[0027] S4.2. traversing and matching corresponding patient follow-up related data in the unified patient health record according to the patient identification information and the dynamic association layer rule of the multi-source heterogeneous data integration model, first matching direct association data corresponding to the core patient entity, and then extending and matching all associated data based on indirect association relationships;

[0028] S4.3. classifying and arranging the matched patient follow-up related data in chronological order, data type, and association strength, the data type including electronic medical record type, examination report type, follow-up record type, rehabilitation training track type, self-management log type, and referral handover voucher type;

[0029] S4.4. presenting the patient follow-up related data in the order of classification and arrangement, synchronously displaying direct association relationships and indirect association relationships among the data, and supporting centralized acquisition and associated viewing of cross-institutional data.

[0030] Further, in step S2.1, each layer of the multi-source heterogeneous data integration model is constructed using modular packaging design, the heterogeneous data adaptation layer and the entity analysis layer are configured to realize bidirectional data transmission through a data interface adaptation component, the entity analysis layer and the semantic enhancement layer are configured to be connected through an entity vector coding module, the semantic enhancement layer and the dynamic association layer are configured to establish a mapping through a semantic feature strengthening component, the dynamic association layer and the attribute mapping layer are configured to realize association through a dynamic weight distribution module, and the components of each layer are configured to communicate through standardized data interfaces.

[0031] Further, in step S2.3, when processing the format adaptation of the cross-institutional referral handover voucher data, the structured data adaptation component is used to analyze the fixed fields and variable fields in the voucher, complete the field mapping relationship according to the requirements of the entity attribute definition unit, and convert the unstructured voucher note information into a standardized text format; the rehabilitation training track data is processed through the sliding window submodule of the time series data adaptation component to extract time series features, and the patient self-management log data is processed through the word segmentation and stop word filtering submodule of the text data adaptation component to process text information.

[0032] Further, in step S3.2, the cosine similarity algorithm is improved by introducing an entity attribute weight factor and a time sequence decay factor, the entity attribute weight factor is determined by the attribute type matching unit of the attribute mapping layer, the time sequence decay factor is dynamically adjusted according to the time stamp difference of the rehabilitation training track data and the follow-up record, and the association matching accuracy of data with different time spans is improved.

[0033] Further, in step S3.3, the indirect association derivation unit adopts a multi-step reasoning mechanism to first establish an initial indirect association clue based on the referral event in the cross-institution referral handover credential data, and then gradually derive the multi-level indirect association relationship between entities by combining the behavior characteristics in the patient self-management log data and the progress characteristics of the rehabilitation training track data, and the rationality of each step of reasoning result is verified by the semantic feature fusion module of the semantic enhancement layer.

[0034] A cross-institutional patient follow-up data integration system based on a knowledge graph is provided, and the system includes a data acquisition module, a data processing module, a data association module, and a data retrieval module.

[0035] The data acquisition module is configured to acquire cross-institutional patient follow-up related data, including electronic medical records, examination reports, follow-up records, rehabilitation training track data, patient self-management log data, and cross-institutional referral handover credential data. The data processing module is configured to construct a multi-source heterogeneous data integration model and perform standardized processing on the patient follow-up related data. The model is set to a five-layer architecture and connected through adaptive components, etc. The model training unit completes the training based on the training set, the validation set, and the joint loss function. The data association module is configured to extract data information and establish an association relationship by combining model rules through semantic mapping technology, and fuse to form a unified patient health record. The data retrieval module is configured to receive patient identification information, match and organize the corresponding patient follow-up related data based on the health record and model rules.

[0036] The beneficial effects of the present application are:

[0037] (1) Relying on the five-layer architecture of multi-source heterogeneous data integration and standardized processing, the differences in format, field, and semantics of cross-institutional follow-up data are eliminated, a unified patient health record is formed, and cross-institutional data is centrally acquired.

[0038] (2) With the data integrity and validity verification mechanism and entity association rules, the accuracy and completeness of data integration are improved, and reliable data support is provided for medical decision-making and follow-up management;

[0039] (3) The entity association design and structured presentation method of the knowledge graph break down the barriers of cross-institutional data, simplify the data retrieval process, and support efficient cross-institutional medical collaboration. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 A cross-institutional patient follow-up data integration method and system based on a knowledge graph step flowchart is provided.

[0041] Figure 2 A cross-institutional patient follow-up data integration method based on a knowledge graph step implementation flowchart is provided.

[0042] Figure 3 The knowledge graph entity and attribute architecture schematic diagram provided for the embodiment;

[0043] Figure 4 The multi-source heterogeneous data integration model architecture schematic diagram provided for the embodiment. DETAILED DESCRIPTION

[0044] The technical solutions of the present application will be described below in conjunction with the embodiments, obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0045] Embodiment 1

[0046] Referring to Figure 1 , a knowledge graph-based cross-institutional patient follow-up data integration method is provided, which comprises the following steps:

[0047] S1. Obtain cross-institutional patient follow-up related data, including electronic medical records, examination reports, follow-up records, rehabilitation training trajectory data, patient self-management log data and cross-institutional referral handover voucher data;

[0048] S2. Construct a multi-source heterogeneous data integration model, which sets up a five-layer architecture including a heterogeneous data adaptation layer, an entity analysis layer, a semantic enhancement layer, a dynamic association layer and an attribute mapping layer. After training the model, the obtained patient follow-up related data is standardized based on the model, and the data format, field definition and semantic expression are unified;

[0049] S3. Through semantic mapping technology combined with model dynamic association layer rules, extract entity information, relationship information and feature information from standardized data, establish direct and indirect association relationships between entities, and form a unified patient health record through association and fusion;

[0050] S4. Receive patient identification information, based on the unified patient health record and the model attribute mapping layer rules, match the corresponding patient follow-up related data and present according to the preset logic, realize the centralized acquisition of cross-institutional data.

[0051] Step S1 includes:

[0052] S1.1. Determine the source institution and data transmission protocol of the cross-institutional patient follow-up related data;

[0053] S1.2. Collecting patient follow-up related data from the source agency according to the determined data transmission protocol, including real-time interface docking, regular batch import and offline file transmission;

[0054] S1.3. Integrity check and validity check are performed on the collected patient follow-up related data, the integrity check includes core field missing check and data record integrity verification, and the validity check includes data format compliance check and logical consistency verification;

[0055] S1.4. Retain the patient follow-up related data that passes the check, and mark the patient follow-up related data that fails the check with an exception type and store it temporarily for subsequent supplement or correction processing.

[0056] Step S2 includes:

[0057] S2.1. Construct a multi-source heterogeneous data integration model, configure each layer to realize connection through an adaptive component, an encoding module and a feature enhancement component, specify the heterogeneous data adaptation layer to process different types of follow-up data, configure the entity analysis layer to include entity feature extraction and classification submodules, set the semantic enhancement layer to enhance semantic features through a multi-layer neural network, configure the dynamic association layer to include association rules, derivation and strength calculation unit, and specify the attribute mapping layer to be responsible for entity attribute definition and matching;

[0058] S2.2. Construct a training set and a validation set using the labeled multi-source heterogeneous medical data across agencies, construct a joint loss function composed of a heterogeneous data adaptation loss function, an entity matching loss function and a semantic association loss function, set the initial learning rate, iteration batch and model convergence threshold, adjust the parameters of each layer alternately through the back propagation algorithm, optimize the adaptation accuracy, analysis accuracy and association matching degree in turn, until the comprehensive adaptation accuracy of the validation set is stable;

[0059] S2.3. Formulate standardized rules based on the model structure, complete data standardization through format adaptation, entity coding, semantic enhancement and attribute matching.

[0060] Step S3 includes:

[0061] S3.1. Extract entity information, relationship information and feature information from the standardized patient follow-up related data, entity information includes the unified coding and attribute values of each entity, relationship information includes the direct association type and indirect association clues between entities, and feature information includes time sequence features, text features and structured features;

[0062] S3.2. Based on the dynamic association layer rule of the multi-source heterogeneous data integration model, the semantic similarity of entities in the patient follow-up related data of different institutions is calculated by using the improved cosine similarity algorithm, the dynamic weight of the association strength calculation unit is combined, the similarity threshold is set as the association judgment condition, and the direct association relationship between entities is established;

[0063] S3.3. Through the indirect association derivation unit of the dynamic association layer, the indirect association relationship between entities is derived according to the attribute association clue, the time sequence association clue and the event association clue between entities, the indirect association strength is calculated and compared with the preset strength threshold, and the effective indirect association relationship is reserved;

[0064] S3.4. According to the established direct association relationship and effective indirect association relationship, the patient follow-up related data of different institutions and different types corresponding to the same patient are associated and fused, the fusion process keeps the consistency of entity unified coding, attribute information and association relationship, and is organized into a unified patient health record according to the classification of the entity analysis layer and the structure of the attribute mapping layer.

[0065] Step S4 includes:

[0066] S4.1. Receive the patient identification information in the data retrieval request, which includes the patient unique number, the identity recognition information and the entity unified coding;

[0067] S4.2. According to the patient identification information and the dynamic association layer rule of the multi-source heterogeneous data integration model, the corresponding patient follow-up related data in the unified patient health record is traversed and matched, the direct association data corresponding to the core patient entity is matched first, and then all the data associated based on the indirect association relationship is extended and matched;

[0068] S4.3. The matched patient follow-up related data is classified and arranged according to the time sequence, the data type and the association strength, and the data type includes electronic medical record, examination report, follow-up record, rehabilitation training track, self-management log and referral transfer voucher;

[0069] S4.4. The patient follow-up related data is presented in the order of classification and arrangement, the direct association relationship and the indirect association relationship between the data are synchronously displayed, and the centralized acquisition and associated viewing of cross-institutional data are supported.

[0070] In step S2.1, the layers of the multi-source heterogeneous data integration model are constructed using a modular packaging design. The heterogeneous data adaptation layer and the entity resolution layer are configured to realize bidirectional data transmission through a data interface adaptation component. The entity resolution layer and the semantic enhancement layer are connected through an entity vector encoding module. The semantic enhancement layer and the dynamic association layer are mapped through a semantic feature strengthening component. The dynamic association layer and the attribute mapping layer are associated through a dynamic weight distribution module. The components of each layer communicate through standardized data interfaces.

[0071] In step S2.3, when performing format adaptation processing on the cross-institution referral handover certificate data, the structured data adaptation component is used to analyze the fixed fields and variable fields in the certificate, complete the field mapping relationship according to the requirements of the entity attribute definition unit, and convert the unstructured certificate note information into a standardized text format. The time series data adaptation component's sliding window submodule extracts time series features from rehabilitation training trajectory data. The text data adaptation component's word segmentation and stop word filtering submodule processes text information from patient self-management log data.

[0072] In step S3.2, the cosine similarity algorithm is improved by introducing an entity attribute weight factor and a time series decay factor. The entity attribute weight factor is determined by the attribute type matching unit of the attribute mapping layer. The time series decay factor is dynamically adjusted according to the timestamp difference between the rehabilitation training trajectory data and the follow-up record, improving the accuracy of matching and associating data with different time spans.

[0073] In step S3.3, the indirect association derivation unit uses a multi-step reasoning mechanism. First, it establishes an initial indirect association clue based on the referral event in the cross-institution referral handover certificate data. Then, it combines the behavior characteristics in the patient self-management log data and the progress characteristics in the rehabilitation training trajectory data to gradually derive the multi-level indirect association relationships between entities. The reasoning result of each step is verified for rationality by the semantic feature fusion module of the semantic enhancement layer.

[0074] In some embodiments, a knowledge graph-based cross-institution patient follow-up data integration system includes a data acquisition module, a data processing module, a data association module, and a data retrieval module.

[0075] The data acquisition module is used to acquire cross-institutional patient follow-up related data, and the patient follow-up related data includes electronic medical records, examination reports, follow-up records, rehabilitation training track data, patient self-management log data and cross-institutional referral handover voucher data; the data processing module is used to construct a multi-source heterogeneous data integration model and standardize the patient follow-up related data, set the model to a five-layer architecture and connect each layer through an adaptive component, and the model training unit completes the training based on the training set, the validation set and the joint loss function; the data correlation module is used to extract data information and establish a correlation relationship through semantic mapping technology combined with model rules, and fuse to form a unified patient health record; the data retrieval module is used to receive patient identification information, match and organize the corresponding patient follow-up related data based on the health record and model rules.

[0076] Embodiment 2

[0077] The embodiment discloses a cross-institutional patient follow-up data integration method and a corresponding system based on a knowledge graph, which realizes efficient integration and centralized acquisition of cross-institutional patient follow-up data through a series of steps such as standardization processing and correlation fusion, as shown in Figure 2 The specific steps are as follows:

[0078] S1. Data acquisition:

[0079] Data acquisition is the basic link of cross-institutional patient follow-up data integration, and complete and effective related data is collected from different sources to provide data support for subsequent processing.

[0080] S1.1 Determine the data source and transmission protocol:

[0081] Firstly, the scope of the source institutions of the cross-institutional patient follow-up related data needs to be determined, and these source institutions cover different types of medical service providers. The data storage and management modes of each institution are different. On this basis, for each source institution, the corresponding data transmission protocol is determined. The transmission protocol needs to meet the safety, stability and compatibility requirements of cross-institutional data transmission, so as to ensure that the data can be smoothly transmitted between the systems of different institutions and avoid data loss or transmission failure.

[0082] S1.2 Collect cross-institutional patient follow-up related data:

[0083] According to the determined data transmission protocol, the patient follow-up related data is collected from each source institution. The collection methods include real-time interface docking, timed batch import and offline file transmission. According to the data volume, real-time demand and system bearing capacity of different institutions, the appropriate collection method can be selected.

[0084] Real-time interface connection is suitable for scenarios that require timely data acquisition. By establishing a dedicated interface between the source agency system and the data integration system, real-time synchronization of data transmission is achieved, ensuring that data can be quickly collected into the integration system after being generated. Batch import at regular intervals is suitable for cases where data volume is large and real-time requirements are not high. According to the pre-set time interval, data files are periodically exported from the source agency and then imported into the integration system, avoiding excessive pressure on the system caused by real-time transmission of large amounts of data. Offline file transmission is suitable for agencies that cannot implement real-time interface connection or regular import. Data files are obtained through mobile storage devices or other means and then manually imported into the integration system.

[0085] S1.3 Data verification processing:

[0086] The collected patient follow-up related data is subjected to integrity verification and validity verification to ensure that the data meets the requirements of subsequent processing.

[0087] Integrity verification includes core field missing check and data record integrity verification. Core field missing check refers to checking whether the key core fields in each piece of data are missing. These core fields are essential elements of patient follow-up information and cannot be missing. Data record integrity verification checks whether the data record is complete and free of defects, ensuring that each piece of data can fully reflect the information related to a follow-up.

[0088] Validity verification includes data format compliance check and logical consistency verification. Data format compliance check verifies whether the data format meets the pre-set standard format requirements, such as date format, coding format, and numerical format. Logical consistency verification checks whether there are logical contradictions between data, such as whether the follow-up time sequence of the same patient is reasonable and whether the test results are consistent with the diagnosis conclusion.

[0089] S1.4 Processing of verification results:

[0090] Patient follow-up related data that passes the verification is retained and included in the subsequent data processing flow. For patient follow-up related data that fails the verification, the type of abnormality is marked in detail, including core field missing, data format error, logical contradiction, and record defect. The data with marked abnormalities is then temporarily stored.

[0091] The temporarily stored abnormal data will be used for subsequent supplement or correction processing, and corresponding processing measures are taken for different types of abnormalities: if the core field is missing, a supplement request can be initiated to the source agency to obtain the missing field information; if the data format is incorrect or logically inconsistent, the source agency can be notified to correct the data, or automatic correction can be performed in the integration system according to the preset correction rules (under the premise of ensuring data accuracy); if the record is incomplete and cannot be supplemented or corrected, the data will be archived separately and will not be included in the subsequent integration process.

[0092] S2. Model construction and data standardization:

[0093] Model construction and data standardization are the implementation of cross-agency data integration, which standardizes the effective data collected through the construction of a multi-source heterogeneous data integration model, eliminates data differences, and lays the foundation for data correlation and fusion.

[0094] S2.1 Construction of multi-source heterogeneous data integration model:

[0095] A multi-source heterogeneous data integration model is constructed, which sets up a five-layer architecture including a heterogeneous data adaptation layer, an entity analysis layer, a semantic enhancement layer, a dynamic correlation layer, and an attribute mapping layer. Each layer is designed with modular encapsulation to ensure the independence and scalability of each layer's functions.

[0096] The layers are connected and interact through specific components: the heterogeneous data adaptation layer and the entity analysis layer implement bidirectional data transmission through a data interface adaptation component; the entity analysis layer and the semantic enhancement layer are connected through an entity vector encoding module; the semantic enhancement layer and the dynamic correlation layer establish a mapping through a semantic feature strengthening component; the dynamic correlation layer and the attribute mapping layer implement correlation through a dynamic weight distribution module; and the components of each layer communicate through standardized data interfaces to ensure data transmission consistency and compatibility.

[0097] The heterogeneous data adaptation layer processes different types of follow-up data, setting up three sub-components: a text data adaptation component, a time series data adaptation component, and a structured data adaptation component. The text data adaptation component includes a word segmentation and stop word filtering sub-module for processing patient self-management log data; the time series data adaptation component includes a sliding window sub-module for processing rehabilitation training trajectory data; and the structured data adaptation component includes a fixed field and variable field analysis sub-module for processing structured or semi-structured data such as cross-agency referral handover credentials, electronic medical records, examination reports, and follow-up records.

[0098] The entity analysis layer includes an entity feature extraction and classification submodule, and sets a patient entity analysis unit, a medical institution entity analysis unit, a medical data entity analysis unit and a medical event entity analysis unit, which are used to extract and classify patient entities, medical institution entities, medical data entities and medical event entities respectively. Each entity analysis unit extracts the key features of the corresponding entity through the feature extraction submodule, and then classifies the entity through the classification submodule, and finally assigns a unique unified code to each entity.

[0099] The semantic enhancement layer strengthens semantic features through a multi-layer neural network, and sets a context semantic encoding module, a cross-domain semantic migration module and a semantic feature fusion module. The context semantic encoding module is used to convert text type data into vector form for subsequent processing. The cross-domain semantic migration module is used to realize semantic adaptation between different institutions and different data types, and eliminate semantic differences. The semantic feature fusion module is used to fuse multi-dimensional semantic features and verify the rationality of subsequent inference results.

[0100] The dynamic association layer includes an association rule, derivation and strength calculation unit, and sets an association rule unit, an indirect association derivation unit and an association strength calculation unit. The association rule unit is used to formulate the judgment rules of direct association between entities. The indirect association derivation unit is used to derive the indirect association relationship between entities through multi-step reasoning. The association strength calculation unit includes a dynamic weight adjustment submodule, which is used to calculate the strength of the association relationship between entities.

[0101] The attribute mapping layer is responsible for entity attribute definition and matching, and sets an entity attribute definition unit, an attribute type matching unit and an attribute value standardization unit. The entity attribute definition unit is used to define the attribute information of various entities in the form of key-value pairs. The attribute type matching unit is used to determine the matching relationship between different entity attributes and clarify the entity attribute weight factor. The attribute value standardization unit is used to unify the format of various entity attribute values and ensure the consistency of attribute information.

[0102] S2.2 Training of multi-source heterogeneous data integration model:

[0103] The labeled cross-institution multi-source heterogeneous medical data is used to construct the training set and the verification set. The annotation process needs to ensure the accuracy and consistency of data annotation. The annotation content includes entity category, entity attribute, entity relationship and other information. The training set is used for model parameter learning, and the verification set is used for verifying the training effect of the model. The division ratio of the training set and the verification set is determined according to the total amount of data and the data distribution, so as to ensure that the two can cover different types and different institutions of medical data.

[0104] A joint loss function is constructed, which is composed of a heterogeneous data adaptation loss function, an entity matching loss function and a semantic association loss function. The heterogeneous data adaptation loss function is used to measure the error of the model in the data adaptation process. The entity matching loss function is used to measure the error of the model in the entity recognition and matching process. The semantic association loss function is used to measure the error of the model in the semantic association establishment process. Through the joint loss function, the training effect of each link of the model can be considered comprehensively, and the optimization of the overall performance of the model can be ensured.

[0105] Training parameters such as initial learning rate, iteration batch and model convergence threshold are set. The initial learning rate is used to control the step size of model parameter update. The iteration batch is used to determine the number of model training. The model convergence threshold is used to determine whether the model training is completed. Through the back propagation algorithm, the parameters of each layer are adjusted alternately. In each iteration process, the weight parameters of each layer of the model are adjusted in reverse according to the error calculated by the joint loss function, and the adaptation accuracy, analysis accuracy and association matching degree are optimized in turn.

[0106] In the training process, after completing each iteration batch, the model is evaluated for performance using the validation set, and the comprehensive adaptation accuracy of the validation set is calculated. If the comprehensive adaptation accuracy does not reach the model convergence threshold, the next round of iteration training is continued. If the comprehensive adaptation accuracy reaches or exceeds the model convergence threshold and remains stable in consecutive multiple iteration batches without significant improvement, it is determined that the model training is completed, and the training of the model is stopped and the trained model parameters are saved.

[0107] S2.3 Data standardization processing:

[0108] Based on the structure of the multi-source heterogeneous data integration model trained, standardization rules are developed, which cover data format, field definition, semantic expression and other aspects, to ensure that the data after standardization processing can meet the requirements of subsequent association and fusion.

[0109] According to the standardization rules, the patient follow-up related data that passes the verification is standardized. The processing process includes four steps: format adaptation, entity coding, semantic enhancement and attribute matching.

[0110] The format adaptation link is completed by a heterogeneous data adaptation layer, and corresponding adaptation modes are adopted for different types of data: for cross-institutional referral handover voucher data, the structured data adaptation component is used to parse the fixed fields and variable fields in the voucher, complete the field mapping relationship according to the requirements of the entity attribute definition unit, and convert the unstructured voucher note information into a standardized text format; for rehabilitation training trajectory data, the time series data adaptation component is used to extract time series features through the sliding window submodule, and the data is arranged into a unified time series data format; for patient self-management log data, the text data adaptation component is used to process text information through the word segmentation and stop word filtering submodule, remove invalid information, and unify the text format; for electronic medical records, examination reports, follow-up records and other data, the structured data adaptation component is used to format the data to ensure that the field arrangement, data type, etc. meet the standardization requirements.

[0111] The entity coding link is completed by an entity analysis layer, which performs entity recognition and extraction on the formatted data of various types. The entity features are extracted through the feature extraction submodule of each entity analysis unit, and then classified through the classification submodule. Finally, a unique unified code is assigned to each entity, which includes an institution identification segment and an internal number segment, ensuring the uniqueness and traceability of the entity.

[0112] The semantic enhancement link is completed by a semantic enhancement layer, which converts the text data in the entity coded data into a vector form through the context semantic coding module, eliminates the semantic differences between data from different institutions using the cross-domain semantic transfer module, and then fuses the multi-dimensional semantic features through the semantic feature fusion module to strengthen the semantic expression of the data and improve the recognition of the data.

[0113] The attribute matching link is completed by an attribute mapping layer, which establishes the matching relationship between the entity attributes in the data and the standard attributes according to the attribute information defined by the entity attribute definition unit, and determines the attribute weight factor. Then, the attribute value standardization unit is used to unify the attribute values of different formats into a standard format, ensuring the consistency of the attribute information.

[0114] S3. Data association fusion:

[0115] Data association fusion is a key link for cross-institutional data integration, which extracts key information from the data, establishes the association relationship between entities, integrates patient follow-up data scattered in different institutions, and forms a unified patient health record.

[0116] S3.1 Extract key information from data:

[0117] Information extraction is performed on the standardized patient follow-up related data, and the extracted information includes entity information, relationship information and feature information.

[0118] The entity information includes unified codes and attribute values of each entity. The unified code is a unique code assigned by the entity analysis layer, and the attribute value is the attribute information standardized by the attribute mapping layer. The attribute values of different types of entities are different. The attribute values of the patient entity include basic information such as age, gender, and medical history. The attribute values of the medical institution entity include information such as institution type, region (not involving specific place names), and service range. The attribute values of the medical data entity include information such as data generation time, data content, and data type. The attribute values of the medical event entity include information such as event occurrence time, event type, and event participating entity.

[0119] The relationship information includes direct association types and indirect association clues between entities. The direct association type refers to the explicit association relationship between entities, such as the "ownership" relationship between the patient and the electronic medical record, and the "generation" relationship between the medical institution and the medical data. The indirect association clue refers to information that can indirectly reflect the association relationship between entities, such as the association clue between the behavior characteristics in the patient's self-management log and the progress characteristics of the rehabilitation training track data, and the association clue between the transfer event in the cross-institution transfer transfer voucher and different institutions.

[0120] The feature information includes time sequence features, text features, and structured features. The time sequence feature refers to the time-related features contained in the data, such as the time sequence features of the rehabilitation training track data and the time interval features of the follow-up records. The text feature refers to the semantic features and keyword features contained in the text type data. The structured feature refers to the field features and data type features contained in the structured data.

[0121] S3.2 Establish direct association relationship between entities:

[0122] Based on the dynamic association layer rule of the multi-source heterogeneous data integration model, the improved cosine similarity algorithm is used to calculate the semantic similarity of entities in the patient follow-up related data of different institutions. The improved cosine similarity algorithm introduces an entity attribute weight factor and a time sequence decay factor. The entity attribute weight factor is determined by the attribute type matching unit of the attribute mapping layer, and different weights are assigned according to the importance of different attributes to the entity association. The time sequence decay factor is dynamically adjusted according to the timestamp difference between the rehabilitation training track data and the follow-up records. The larger the timestamp difference, the smaller the time sequence decay factor, so as to reduce the influence of data with large time span on association matching and improve the association matching accuracy of data with different time span.

[0123] Combined with the dynamic weight of the association strength calculation unit, the similarity threshold is set as the association judgment condition. The dynamic weight is adjusted according to factors such as the type of data and the credibility of the source institution. The similarity threshold is determined according to the verification results in the model training process, to ensure that effective association and invalid association can be accurately distinguished.

[0124] The calculated entity semantic similarity is compared with a similarity threshold. If the semantic similarity is greater than or equal to the similarity threshold, it is determined that there is a direct association relationship between the two entities, and the association type and association strength are recorded. If the semantic similarity is less than the similarity threshold, it is determined that there is no direct association relationship between the two entities, and no association record is made.

[0125] S3.3 Establishing indirect association relationship between entities:

[0126] Through the indirect association derivation unit of the dynamic association layer, a multi-step reasoning mechanism is used to derive the indirect association relationship between entities. First, an initial indirect association clue is established based on the transfer events in the cross-institution transfer handover voucher data. The transfer event involves multiple entities such as patients, transferring-out institutions, transferring-in institutions, etc. Through the association relationship between these entities, a basic framework for indirect association is initially constructed.

[0127] In combination with the behavior characteristics in the patient self-management log data and the progress characteristics of the rehabilitation training track data, the multi-level indirect association relationship between entities is gradually derived. For example, according to the rehabilitation training behavior recorded in the patient self-management log, the corresponding training progress in the rehabilitation training track data is associated, and then the training progress is associated to the corresponding medical advice, and further associated to the medical institutions or medical personnel that give the advice, etc. entities; or according to the time sequence of the follow-up records of patients in different institutions, the patient's disease trend is associated, and then the disease trend is associated to the corresponding examination report, diagnosis conclusion, etc.

[0128] In each step of reasoning, the semantic feature fusion module of the semantic enhancement layer verifies the rationality of the reasoning result. If the reasoning result conforms to the semantic logic and business rules, the next step of reasoning is continued. If the reasoning result has logical contradictions or does not conform to the business rules, the derivation of the indirect association clue is terminated, and the reasoning abnormal information is recorded.

[0129] The association strength of the indirect association relationship is calculated. The calculation of the association strength combines factors such as the association level between entities and the degree of feature matching. The calculated indirect association strength is compared with a preset strength threshold. If the indirect association strength is greater than or equal to the preset strength threshold, the indirect association relationship is retained, and the association type, association strength and reasoning path are recorded. If the indirect association strength is less than the preset strength threshold, the indirect association relationship is discarded, and no association record is made.

[0130] S3.4 Association fusion to form a unified patient health record:

[0131] According to the established direct association relationship and effective indirect association relationship, different institutions and different types of patient follow-up related data corresponding to the same patient are associated and fused. During the fusion process, the consistency of entity unified coding, attribute information and association relationship is strictly maintained to ensure that there is no conflict or contradiction in the fused data.

[0132] According to the classification of the entity analysis layer and the structure of the attribute mapping layer, the fused data is organized and arranged to form a unified patient health record. The patient health record takes the patient entity as the core and is structured stored according to the data type, time sequence and other dimensions. The data type includes electronic medical record, examination report, follow-up record, rehabilitation training track, self-management log and referral transfer voucher. Each data type contains corresponding specific data content, associated entity information and association relationship information.

[0133] During the fusion process, if differences are found in the same type of data from different institutions, priority determination needs to be made in combination with factors such as data generation time, credibility of the source institution, data integrity and accuracy, and the data with higher priority is selected as the main data, while the other difference data is retained as supplementary information, to ensure the comprehensiveness and accuracy of the patient health record.

[0134] S4. Data retrieval and presentation:

[0135] Data retrieval and presentation is the final link of cross-institutional patient follow-up data integration, aiming to provide users with convenient and efficient data acquisition methods, and support centralized viewing and associated analysis of cross-institutional data.

[0136] S4.1 Receive patient identification information:

[0137] Receive the data retrieval request initiated by the user, which contains patient identification information including patient unique number, identity information and entity unified coding. These identification information is the key basis for matching patient related data, ensuring accurate positioning of the target patient.

[0138] Carry out legality verification on the received patient identification information, including whether the format of the identification information meets the requirements, whether there are invalid characters, etc. If the identification information verification is passed, proceed to the subsequent data matching link; if the identification information verification is not passed, return an identification information error prompt to the user and require the user to provide legal patient identification information again.

[0139] S4.2 Match corresponding patient follow-up related data:

[0140] According to the dynamic association layer rule of the patient identification information and the multi-source heterogeneous data integration model, the corresponding patient follow-up related data is traversed in the unified patient health record. The matching process adopts a hierarchical matching strategy, which first matches the directly associated data corresponding to the core patient entity. The directly associated data includes electronic medical records, examination reports, follow-up records and other data directly associated with the patient entity, which can be quickly located through entity unified coding.

[0141] Based on the indirect association relationship, all associated data is expanded and matched. According to the established indirect association relationship, other entities corresponding to the core patient entity are traversed and deduced, such as medical institutions related to the patient, medical events, rehabilitation training plans and other data, to ensure that the matched data comprehensively covers the patient's follow-up related information in various institutions.

[0142] In the matching process, the matching path and association strength of the data are recorded, which facilitates subsequent data sorting and presentation. If the corresponding patient follow-up related data is not matched, the user is returned a prompt message indicating that the data does not exist.

[0143] S4.3 Classification and arrangement of matched data:

[0144] The matched patient follow-up related data is classified and arranged according to time sequence, data type and association strength. When arranging by time sequence, the data is sorted from early to late based on the timestamp of data generation, which facilitates the user to view the time evolution process of patient follow-up data; when arranging by data type, the data is divided into electronic medical records, examination reports, follow-up records, rehabilitation training trajectories, self-management logs and referral transfer credentials, and the data under each category is arranged in chronological order; when arranging by association strength, the data is sorted according to association strength from high to low, the higher the association strength, the stronger the relevance to the patient's core needs, and the data is presented to the user first.

[0145] In the classification and arrangement process, duplicate data and invalid data are removed. Duplicate data refers to data with the same content or consistent information, and invalid data refers to data that is found to have errors or no actual meaning after verification, to ensure that the arranged data is concise and effective.

[0146] S4.4 Presentation of patient follow-up related data:

[0147] The patient follow-up related data is presented in the order of classification and arrangement, and the presentation method adopts structured display to clearly display the basic information, core content and association information of the data. The basic information includes the data generation institution, generation time, data type, etc.; the core content includes the specific field information and numerical value of the data (without specific privacy numerical value); the association information includes the direct and indirect association relationships between the data and other data, and the association path and association strength are displayed in a visual manner.

[0148] The user can view other associated data by clicking the associated information, support cross-institutional data association viewing, realize data tracing and linkage analysis. At the same time, the system provides data export function, the user can export the presented data to standard format file according to the demand, convenient for subsequent use.

[0149] In some embodiments, to realize the above-mentioned knowledge graph-based cross-institutional patient follow-up data integration method, a corresponding cross-institutional patient follow-up data integration system is deployed, which includes a data acquisition module, a data processing module, a data association module and a data retrieval module. Each module cooperates with each other to complete the integration and retrieval of cross-institutional patient follow-up data.

[0150] The data acquisition module is used to acquire cross-institutional patient follow-up related data, including electronic medical records, examination reports, follow-up records, rehabilitation training trajectory data, patient self-management log data and cross-institutional referral handover voucher data. This module includes a source institution interfacing unit, a data acquisition unit, a data verification unit and an abnormal data processing unit. The source institution interfacing unit is used to determine the data source institution and the data transmission protocol, and to establish a communication connection with the source institution. The data acquisition unit is used to collect data according to the determined transmission protocol and acquisition method. The data verification unit is used to perform integrity and validity verification on the collected data. The abnormal data processing unit is used to process abnormal data that fails to pass the verification.

[0151] The data processing module is used to construct a multi-source heterogeneous data integration model and to standardize the patient follow-up related data. The model is set to a five-layer architecture and each layer is connected through an adaptive component. This module includes a model construction unit, a model training unit and a data standardization unit. The model construction unit is used to construct the five-layer architecture of the multi-source heterogeneous data integration model and each layer subcomponent, and to define the connection relationship and data interaction mode of each layer. The model training unit is used to complete model training based on the training set, the validation set and the joint loss function, and to optimize the model parameters. The data standardization unit is used to develop standardization rules based on the model structure, and to perform format adaptation, entity coding, semantic enhancement and attribute matching processing on the data that passes the verification.

[0152] The data association module is used to extract data information and establish an association relationship by combining model rules through semantic mapping technology, and to form a unified patient health record. The module includes an information extraction unit, a direct association establishment unit, an indirect association establishment unit, and a data fusion unit. The information extraction unit is used to extract entity information, relationship information, and feature information from standardized data. The direct association establishment unit is used to establish direct association relationships between entities using an improved cosine similarity algorithm. The indirect association establishment unit is used to establish indirect association relationships between entities through a multi-step reasoning mechanism. The data fusion unit is used to associate and fuse related data of the same patient based on the established association relationships, forming a unified patient health record.

[0153] The data retrieval module is used to receive patient identification information, match and organize corresponding patient follow-up related data based on the health record and model rules. The module includes an identification information receiving unit, a data matching unit, a data organizing unit, and a data presenting unit. The identification information receiving unit is used to receive patient identification information input by the user and perform legality verification. The data matching unit is used to match corresponding patient follow-up related data in the patient health record. The data organizing unit is used to classify and organize the matched data according to pre-set rules. The data presenting unit is used to present the data in the order of organization, supporting associated viewing and exporting of the data.

[0154] The knowledge graph-based cross-institutional patient follow-up data integration method and system provided in this embodiment realizes standardized processing of patient follow-up data of different institutions and different types by constructing a multi-source heterogeneous data integration model, eliminates differences in data format, field definition, and semantic expression, and ensures data consistency and compatibility. By establishing direct and indirect association relationships between entities, patient follow-up data scattered in different institutions is effectively associated and fused to form a unified patient health record, realizing centralized acquisition of cross-institutional data.

[0155] In the data processing process, through multi-level verification and optimization mechanisms, the integrity, validity, and accuracy of the data are guaranteed, providing reliable data support for subsequent medical decision-making and follow-up management. At the same time, the system adopts modular design, with clear module functions and collaborative work, improving the efficiency and flexibility of data processing, supporting fast retrieval and associated viewing of data, and providing users with a convenient data usage experience. In addition, the scheme enhances the stability and scalability of the scheme through clear model architecture, standardized training process, and standardized processing steps, and can meet the cross-institutional data integration needs in different scenarios. Among them, the centralized acquisition of cross-institutional data is directly realized through standardized processing, association fusion, etc., breaking down the data barriers between different institutions, allowing users to conveniently obtain complete follow-up data of patients in various institutions, and providing strong support for cross-institutional medical collaboration.

[0156] Example 3

[0157] In some implementations, referring to Figure 3 The knowledge graph architecture of the present embodiment includes an entity layer and an attribute layer, which support each other, combine multi-source heterogeneous data integration processes, realize systematic integration and efficient use of cross-institutional patient follow-up data, and the entire implementation process is carried out around the functions of the two-layer architecture and the technical connection.

[0158] The entity layer covers four types of entities, namely patient entities, medical institution entities, medical data entities and medical event entities, and the construction and extraction of each type of entity is completed by the entity analysis layer of the multi-source heterogeneous data integration model. In the data acquisition stage, through real-time interface connection, timed batch import or offline file transmission, etc., electronic medical records, examination reports, follow-up records, rehabilitation training trajectory data, patient self-management log data and cross-institutional referral transfer certificate data are collected from different institutions, and after integrity and validity verification, qualified data is retained for entity construction. The entity analysis layer extracts the key features of each type of entity through the feature extraction submodule, and then completes entity classification through the classification submodule, assigns a unique unified code to each entity, and ensures the uniqueness and traceability of the entity. Among them, the medical data entity corresponds to the six types of follow-up related data collected, and each data entry is taken as an instance of the medical data entity, and the corresponding generating institution (medical institution entity) and the associated patient (patient entity) are associated; the medical event entity covers specific medical behaviors such as referral, examination, follow-up, etc., and is associated with the patient entity, the medical institution entity and the corresponding medical data entity involved.

[0159] The attribute layer contains direct relationships, indirect associations, relationship attributes and entity attributes, and covers key attribute information such as data types and generation times. The construction and improvement of the attribute layer run through the whole process of data standardization and association fusion. The entity attributes are defined and matched by the attribute mapping layer. The attributes of patient entities include basic information such as age, gender and medical history. The attributes of medical institution entities include information such as institution type and service range. The attributes of medical data entities include information such as data type, generation time and data content. The attributes of medical event entities include information such as event type, occurrence time and participating entities. All attributes are processed by the attribute value standardization unit to unify the format and expression standard. The establishment of direct relationships and indirect associations is dominated by the dynamic association layer. The direct relationship is determined based on the calculation of entity semantic similarity by improving the cosine similarity algorithm. The algorithm introduces an entity attribute weight factor and a time sequence decay factor. The entity attribute weight factor is determined by the attribute type matching unit of the attribute mapping layer. The time sequence decay factor is dynamically adjusted according to the data timestamp difference to ensure the accuracy of entity association with different time spans and different attribute importance. The indirect association is derived through a multi-step reasoning mechanism. Taking the referral event in the cross-institution referral handover voucher data as the initial clue, combining the behavior characteristics of patient self-management logs and the progress characteristics of rehabilitation training tracks, a multi-level indirect association is gradually constructed. The reasoning result of each step is verified for rationality by the semantic feature fusion module of the semantic enhancement layer. Finally, the effective indirect association with a qualified association strength is retained. The relationship attributes are used to describe the characteristics of direct relationships and indirect associations, including association strength, association type, reasoning path and other information, which provides a basis for the credibility evaluation of data association.

[0160] The landing of the entity layer and the attribute layer relies on the cooperative work of the five-layer architecture of the multi-source heterogeneous data integration model. In the model construction stage, modular encapsulation design is adopted. The heterogeneous data adaptation layer processes different types of data through special components to provide a unified data source for the entity layer. The entity analysis layer directly supports the entity extraction and classification of the entity layer. The semantic enhancement layer strengthens the semantic expression of entities and features through multi-layer neural networks to provide semantic support for the establishment of the association relationship of the attribute layer. The dynamic association layer focuses on the construction and strength calculation of the direct relationship and indirect association of the attribute layer. The attribute mapping layer is responsible for the definition, matching and standardization of entity attributes and relationship attributes. During model training, the labeled multi-source heterogeneous medical data across institutions is used to construct the training set and the verification set. The joint loss function is used to optimize the adaptation accuracy, analysis accuracy and association matching degree to ensure that the model can accurately support the construction of the entity layer and the attribute layer.

[0161] In the data association and fusion stage, based on the unified coding of the entity layer and the association relationship of the attribute layer, data from different institutions and of different types corresponding to the same patient are integrated. Following the classification of the entity parsing layer and the structure of the attribute mapping layer, a unified patient health record is formed. Within the health record, various entities at the entity layer form a network of connections through direct and indirect relationships at the attribute layer. Attributes such as data type and generation time serve as criteria for retrieval and sorting, supporting efficient data retrieval. Upon receiving patient identification information (including the patient's unique number, identity information, and unified entity coding), the health record is traversed based on the association rules of the attribute layer. First, directly related data corresponding to the patient entity is matched. Then, the matching range is expanded through indirect associations. The results are categorized and presented according to time order, data type, and association strength, simultaneously displaying the direct and indirect relationships between data, supporting centralized acquisition and associated viewing of cross-institutional data.

[0162] This architecture clarifies the data subjects through the entity layer and standardizes entity characteristics and relationships through the attribute layer. Combined with the model's standardized processing and association fusion capabilities, it effectively solves the problems of heterogeneous and loosely related data across institutions. The structured design of entities and attributes makes data integration more logical, and the clear division of various entities and attributes improves the efficiency of data retrieval and utilization, providing reliable data support for cross-institutional medical collaboration while ensuring the standardization of the data integration process and the accuracy of the results.

[0163] Example 4

[0164] In some specific implementations, refer to Figure 4 This embodiment constructs a multi-source heterogeneous data integration model with a five-layer architecture, and combines standardized processing and correlation fusion processes to achieve systematic integration and centralized retrieval of patient follow-up data across institutions.

[0165] During the data acquisition phase, the source institutions and data transmission protocols for cross-institutional patient follow-up data were first identified. Following these protocols, methods such as real-time interface integration, scheduled batch import, or offline file transfer were used to collect electronic medical records, examination reports, follow-up records, rehabilitation training trajectory data, patient self-management logs, and cross-institutional referral handover documents. After collection, the data underwent integrity and validity verification. Integrity verification included checking for missing core fields and verifying the completeness of data records. Validity verification included checking for data format compliance and logical consistency. Data that passed verification was retained for subsequent processing, while data that failed was marked as abnormal and temporarily stored for later supplementary collection or correction.

[0166] In the model building phase, a modular encapsulation design is adopted to build a five-layer architecture. Each layer is connected through adapter components, coding modules, and feature enhancement components, and all components are required to communicate through standardized data interfaces.

[0167] The heterogeneous data adaptation layer sets three special components. The text data adaptation component includes a word segmentation and stop word filtering submodule, which is specially designed to process patient self-management log data. The time series data adaptation component includes a sliding window submodule, which is responsible for processing rehabilitation training trajectory data. The structured data adaptation component includes a fixed field and variable field parsing submodule, which is used to process cross-institution referral handover voucher data, electronic medical records, examination reports, and follow-up records. The overall data format adaptation is achieved, and the text, sequence, and structured formats are unified.

[0168] The entity resolution layer includes four analysis units. The patient entity analysis unit, the medical institution entity analysis unit, the medical data entity analysis unit, and the medical event entity analysis unit are equipped with entity feature extraction and classification submodules. They output patient entities, medical institution entities, medical data entities, and referral / examination / follow-up event entities, respectively, and assign unique uniform codes to each entity. The entity recognition and coding work is completed.

[0169] The semantic enhancement layer sets three core modules. The context semantic coding module realizes the conversion of text to vector. The cross-domain semantic migration module completes the semantic adaptation of data from different institutions. The semantic feature fusion module is responsible for multi-dimensional feature fusion and subsequent reasoning result verification. Through multiple layers of neural network, the semantic features are strengthened, and the feature recognition degree is improved.

[0170] The dynamic association layer includes three functional units. The association rule unit uses an improved cosine similarity algorithm for direct association judgment. The indirect association derivation unit carries out multi-step reasoning based on attributes, time series, and event clues. The association strength calculation unit is responsible for calculating the association strength and performing threshold screening, realizing the establishment of effective direct and indirect association relationships.

[0171] The attribute mapping layer sets three units. The entity attribute definition unit defines entity attributes in the form of key-value pairs. The attribute type matching unit determines the entity attribute weight factor. The attribute value standardization unit unifies the attribute value format, completing the binding of entity and relationship attributes.

[0172] During model training, the labeled cross-institutional multi-source heterogeneous medical data is used to construct the training set and the verification set. The annotation content includes entity category, attribute and entity relationship. A joint loss function is constructed by heterogeneous data adaptation loss function, entity matching loss function and semantic association loss function. The initial learning rate, iteration batch and model convergence threshold are set. The parameters of each layer are adjusted alternately by the back propagation algorithm. The adaptation accuracy, analysis accuracy and association matching degree are optimized in turn until the comprehensive adaptation accuracy of the verification set is stable. The model training is completed and the parameters are saved. Based on the model structure, standardization rules are formulated for the data that passes the verification. The format adaptation is completed by the heterogeneous data adaptation layer, and the data is allocated to the corresponding component for processing according to the data type. The entity extraction and coding are performed on the adapted data by the entity analysis layer, and the unique unified coding is allocated. The text data is converted into vectors by the semantic enhancement layer to eliminate the semantic differences between different institutions and fuse multi-dimensional features. The attribute mapping layer completes the definition, matching and standardization of entity attributes, and finally unifies the data format, field definition and semantic expression.

[0173] In the data association and fusion stage, entity information, relationship information and feature information are first extracted from the standardized data. The entity information includes the unified coding and attribute values of each entity. The relationship information includes the direct association type and indirect association clues between entities. The feature information includes time sequence characteristics, text characteristics and structured characteristics. Based on the rules of the dynamic association layer, the association rule unit uses an improved cosine similarity algorithm that introduces entity attribute weight factors and time sequence decay factors to calculate the semantic similarity of entities in different institution data. Combined with the dynamic weight of the association strength calculation unit and the preset similarity threshold, the direct association relationship between entities is established. The indirect association derivation unit takes the referral event in the cross-institutional referral handover voucher data as the initial clue, combines the behavior characteristics of the patient self-management log and the progress characteristics of the rehabilitation training track, and carries out multi-step reasoning to derive indirect association relationships. The reasoning result of each step is verified for rationality by the semantic feature fusion module of the semantic enhancement layer, the indirect association strength is calculated and compared with the preset threshold, and the effective indirect association relationship is retained. According to the established direct and effective indirect association relationship, the different institution and different type data corresponding to the same patient are associated and fused. The fusion process maintains the consistency of entity unified coding, attribute information and association relationship. According to the classification of the entity analysis layer and the structure of the attribute mapping layer, a unified patient health record is formed.

[0174] In the data retrieval presentation stage, the patient identification information in the user data retrieval request is received, including the patient unique number, identity information and entity unified code. After the legality of the identification information is verified, the matching data in the unified patient health record is searched based on the model attribute mapping layer rule and the dynamic association layer rule. The direct association data corresponding to the core patient entity is matched first, and then all related data is expanded and matched based on the indirect association relationship. The matching results are classified and arranged in chronological order, data type and association strength. The data type is divided into electronic medical record type, examination report type, follow-up record type, rehabilitation training track type, self-management log type and referral transfer voucher type. The data is presented in a structured manner according to the arrangement order, and the direct and indirect association relationships between the data are synchronously displayed to support user association viewing and realize centralized acquisition of cross-institutional data.

[0175] The implementation process effectively solves the adaptation and semantic unification of multi-source heterogeneous data through the cooperative work of the five-layer model architecture. The standardized processing eliminates the differences in format and field definition of data from different institutions, and the unified patient health record formed by association and fusion breaks through the cross-institutional data barrier. The modular design of each layer component improves the extensibility and maintainability of the model, and the application of entity coding and association rules guarantees the accuracy of data integration. The structured retrieval and presentation method provides convenience for data use, and is suitable for various cross-institutional medical data integration scenarios.

[0176] The above only describes the preferred embodiments of the present application, and it should be understood that the present application is not limited to the forms disclosed herein, and should not be considered as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the concepts described herein by the above-mentioned teaching or related technical or knowledge. The modifications and changes made by those skilled in the art without departing from the spirit and scope of the present application shall be within the scope of protection of the appended claims of the present application.

Claims

1. A method for integrating cross-institutional patient follow-up data based on knowledge graphs, characterized in that, Includes the following steps: S1. Obtain patient follow-up data across institutions, including electronic medical records, examination reports, follow-up records, rehabilitation training trajectory data, patient self-management log data, and cross-institutional referral handover voucher data; S2. Construct a multi-source heterogeneous data integration model. This model has a five-layer architecture, including a heterogeneous data adaptation layer, an entity parsing layer, a semantic enhancement layer, a dynamic association layer, and an attribute mapping layer. After training the model, the acquired patient follow-up data is standardized based on the model to unify the data format, field definition, and semantic expression. S3. By combining semantic mapping technology with the dynamic association layer rules of the model, entity information, relationship information and feature information are extracted from the standardized data, and direct and indirect association relationships between entities are established. The associations are then integrated to form a unified patient health record. S4. Receive patient identification information, match the corresponding patient follow-up data based on the unified patient health record and model attribute mapping layer rules, and organize and present the data according to the preset logic to achieve centralized acquisition of cross-institutional data; Step S2 includes: S2.

1. Construct a multi-source heterogeneous data integration model, configure each layer to be connected through adaptation components, encoding modules and feature enhancement components, specify the heterogeneous data adaptation layer to process different types of follow-up data, configure the entity parsing layer to include entity feature extraction and classification sub-modules, set the semantic enhancement layer to enhance semantic features through multi-layer neural networks, configure the dynamic association layer to include association rules, derivation and strength calculation units, and specify the attribute mapping layer to be responsible for entity attribute definition and matching; S2.

2. Annotated cross-institutional multi-source heterogeneous medical data is used to construct training and validation sets. A joint loss function is constructed, which consists of heterogeneous data adaptation loss function, entity matching loss function and semantic association loss function. The initial learning rate, iteration batch and model convergence threshold are set. The parameters of each layer are adjusted alternately through the backpropagation algorithm to optimize the adaptation accuracy, parsing accuracy and association matching degree in turn until the overall adaptation accuracy of the validation set is stable. S2.

3. Based on the model structure, standardization rules are formulated, and data standardization is completed through format adaptation, entity encoding, semantic enhancement and attribute matching.

2. The method according to claim 1, characterized in that, Step S1 includes: S1.

1. Determine the source institutions and data transmission protocols for cross-institutional patient follow-up data; S1.

2. Collect patient follow-up data from the source institution according to the established data transmission protocol. The collection methods include real-time interface connection, timed batch import, and offline file transfer. S1.

3. Perform integrity and validity checks on the collected patient follow-up data. Integrity checks include checking for missing core fields and verifying the integrity of data records. Validity checks include checking for data format compliance and verifying logical consistency. S1.

4. Retain the follow-up data of patients who pass the verification, and temporarily store the follow-up data of patients who fail the verification after marking the abnormality type, for subsequent supplementary collection or correction processing.

3. The method according to claim 1, characterized in that, Step S3 includes: S3.

1. Extract entity information, relation information and feature information from the standardized patient follow-up data. Entity information includes the unified code and attribute value of each entity. Relationship information includes the direct association type and indirect association clue between entities. Feature information includes time-series features, text features and structured features. S3.

2. Based on the dynamic association layer rules of the multi-source heterogeneous data integration model, the improved cosine similarity algorithm is used to calculate the semantic similarity of entities in the patient follow-up data of different institutions. Combined with the dynamic weight of the association strength calculation unit, a similarity threshold is set as the association judgment condition to establish the direct association relationship between entities. S3.

3. Through the indirect association derivation unit of the dynamic association layer, based on the attribute association clues, temporal association clues and event association clues between entities, the indirect association relationship between entities is deduced, the indirect association strength is calculated and compared with the preset strength threshold, and the valid indirect association relationship is retained. S3.

4. Based on the established direct and effective indirect relationships, the follow-up data of patients from different institutions and of different types corresponding to the same patient are linked and integrated. The integration process maintains the consistency of entity unified coding, attribute information and relationship. According to the classification of the entity parsing layer and the structure of the attribute mapping layer, the data is organized into a unified patient health record.

4. The method according to claim 1, characterized in that, Step S4 includes: S4.

1. Receive patient identification information from the data retrieval request. The patient identification information includes the patient's unique number, identity information, and entity unique code. S4.

2. Based on the dynamic association layer rules of the patient identification information and the multi-source heterogeneous data integration model, the corresponding patient follow-up related data are traversed and matched in the unified patient health record. First, the directly related data corresponding to the core patient entity is matched, and then all related data are matched based on the indirect association relationship. S4.

3. Classify and organize the matched patient follow-up data according to time order, data type and correlation strength. Data types include electronic medical records, examination reports, follow-up records, rehabilitation training tracks, self-management logs and referral handover vouchers; S4.

4. Present patient follow-up data in the sorted order, simultaneously display the direct and indirect relationships between data, and support centralized acquisition and linked viewing of data across institutions.

5. The method according to claim 1, characterized in that, In step S2.1, a modular encapsulation design is used to construct each layer of the multi-source heterogeneous data integration model. The heterogeneous data adaptation layer and the entity parsing layer are configured to achieve bidirectional data transmission through a data interface adaptation component. The entity parsing layer and the semantic enhancement layer are configured to be connected through an entity vector encoding module. The semantic enhancement layer and the dynamic association layer are configured to establish a mapping through a semantic feature enhancement component. The dynamic association layer and the attribute mapping layer are configured to be associated through a dynamic weight allocation module. It is stipulated that the components of each layer communicate through a standardized data interface.

6. The method according to claim 1, characterized in that, In step S2.3, when performing format adaptation processing on the cross-institutional referral handover voucher data, the fixed and variable fields in the voucher are parsed through the structured data adaptation component, and the field mapping relationship is completed according to the requirements of the entity attribute definition unit. The unstructured voucher remarks information is converted into a standardized text format. The temporal features of the rehabilitation training trajectory data are extracted through the sliding window submodule of the temporal data adaptation component, and the text information of the patient self-management log data is processed through the word segmentation and stop word filtering submodule of the text data adaptation component.

7. The method according to claim 3, characterized in that, In step S3.2, the improved cosine similarity algorithm introduces an entity attribute weight factor and a time decay factor. The entity attribute weight factor is determined by the attribute type matching unit of the attribute mapping layer, and the time decay factor is dynamically adjusted according to the timestamp differences between rehabilitation training trajectory data and follow-up records, thereby improving the accuracy of association matching of data across different time spans.

8. The method according to claim 3, characterized in that, In step S3.3, the indirect association derivation unit adopts a multi-step reasoning mechanism. First, it establishes initial indirect association clues based on the referral events in the cross-institutional referral handover voucher data. Then, it combines the behavioral characteristics in the patient self-management log data and the progress characteristics of the rehabilitation training trajectory data to gradually deduce the multi-level indirect association relationships between entities. The rationality of each step of the reasoning result is verified by the semantic feature fusion module of the semantic enhancement layer.

9. A cross-institutional patient follow-up data integration system based on knowledge graphs, characterized in that, The system includes a data acquisition module, a data processing module, a data association module, and a data retrieval module; The data acquisition module is used to acquire patient follow-up data across institutions, including electronic medical records, examination reports, follow-up records, rehabilitation training trajectory data, patient self-management log data, and cross-institutional referral handover voucher data. The data processing module is used to construct a multi-source heterogeneous data integration model and standardize the patient follow-up data. The model is set as a five-layer architecture, and each layer is connected through an adapter component. The model training unit completes training based on the training set, validation set, and joint loss function. The multi-source heterogeneous data integration model includes a five-layer architecture: heterogeneous data adaptation layer, entity parsing layer, semantic enhancement layer, dynamic association layer, and attribute mapping layer. Each layer is configured to be connected through adaptation components, encoding modules, and feature enhancement components. The heterogeneous data adaptation layer is specified to handle different types of follow-up data. The entity parsing layer is configured to include entity feature extraction and classification sub-modules. The semantic enhancement layer is set to enhance semantic features through multi-layer neural networks. The dynamic association layer is configured to include association rules, derivation, and strength calculation units. The attribute mapping layer is specified to be responsible for entity attribute definition and matching. A training set and a validation set were constructed using labeled cross-institutional multi-source heterogeneous medical data. A joint loss function was constructed, which consists of a heterogeneous data adaptation loss function, an entity matching loss function, and a semantic association loss function. An initial learning rate, iteration batch, and model convergence threshold were set. The parameters of each layer were adjusted alternately through the backpropagation algorithm to optimize the adaptation accuracy, parsing accuracy, and association matching degree in turn until the overall adaptation accuracy of the validation set stabilized. Standardization rules are formulated based on the model structure, and data standardization is completed through format adaptation, entity encoding, semantic enhancement and attribute matching. The data association module is used to extract data information and establish association relationships by combining semantic mapping technology with model rules, and integrate them to form a unified patient health record; the data retrieval module is used to receive patient identification information, match and organize the corresponding patient follow-up related data based on health records and model rules.

Images