Hospital comprehensive management system and method based on multi-source heterogeneous data fusion

Abstract

The present application relates to the technical field of hospital management, in particular to a hospital comprehensive management system and method based on multi-source heterogeneous data fusion; the method comprises the following steps: constructing a collection terminal, collecting heterogeneous data of various business systems of the hospital, forming an original data set, and performing pre-checking of hierarchical permission control, assigning different data access and collection permissions based on roles, and recording data collection timestamps and data source identifiers; constructing a unified system, mapping, converting and double-checking multi-source heterogeneous data based on a medical data mapping rule library, and outputting standardized data; performing multi-source data deep fusion based on the standardized data, and outputting a classified unified data mart; the system comprises a multi-source heterogeneous data collection module, a data mapping and conversion module, and a multi-source heterogeneous data fusion and hierarchical control module; through the above-mentioned method, multi-source heterogeneous data fusion, data mapping and conversion are realized, thereby meeting the needs of smart hospital management.

Description

Technical Field

[0001] This invention relates to the field of hospital management technology, and in particular to a comprehensive hospital management system and method based on the fusion of multi-source heterogeneous data. Background Technology

[0002] With the rapid development of medical informatization, various hospital business systems (HIS, LIS, PACS, EMR, etc.) are constantly being improved. At the same time, emerging data sources such as wearable devices and follow-up systems are continuously increasing, resulting in massive amounts of multi-source heterogeneous data. This data exhibits significant heterogeneity in format (CSV, HL7, DICOM, JSON, etc.), semantics (different encoding standards such as ICD-9 and ICD-10), storage methods, and time series. In terms of time series, it covers differences between second-level collections (such as electrocardiograms) and monthly records (such as outpatient statistics), making it difficult to effectively integrate, share, and utilize the data.

[0003] In current hospital comprehensive management, traditional management methods and existing related technologies have the following shortcomings: First, the accuracy of multi-source heterogeneous data fusion is low. Existing fusion methods mostly use simple splicing or static rule matching, which do not fully consider the semantic heterogeneity and temporal sequence characteristics of medical data. This easily leads to data conflicts and redundancy, and makes it impossible to achieve deep data fusion. Second, the data mapping and transformation mechanism is imperfect. There is a lack of unified mapping rules and standardized transformation processes. Data with different formats and semantics cannot be accurately mapped and efficiently transformed, resulting in poor data consistency. The data consistency verification pass rate of traditional methods is only 72%, and the diagnostic coding accuracy rate is about 88%, which is far from meeting the needs of smart hospital management.

[0004] Therefore, it is essential to propose a comprehensive hospital management system and method that integrates, maps, and transforms multi-source heterogeneous data to meet the needs of smart hospital management. Summary of the Invention

[0005] The purpose of this invention is to provide a hospital integrated management system and method based on the fusion of multi-source heterogeneous data, which aims to achieve the fusion, mapping and transformation of multi-source heterogeneous data to meet the needs of smart hospital management.

[0006] To achieve the above objectives, the present invention employs a hospital integrated management method based on multi-source heterogeneous data fusion, comprising the following steps: A data acquisition terminal was built, which used three methods: interface call, direct database connection, and real-time capture to collect heterogeneous data from various business systems of the hospital, forming a raw data set. Prior verification of hierarchical permission control was performed, and different data access and acquisition permissions were assigned based on roles. At the same time, the data acquisition timestamp and data source identifier were recorded. A unified system is constructed to map, transform, and perform dual verification of multi-source heterogeneous data based on a medical data mapping rule base, and output standardized data. A two-layer fusion approach, consisting of a feature layer and a decision layer, is adopted. Based on standardized data, multi-source data is deeply fused to output a unified data mart and associated with hierarchical permission control rules to set corresponding storage permissions for different types of fused data.

[0007] The process of building the data acquisition terminal involves using three methods—interface calls, direct database connections, and real-time capture—to collect heterogeneous data from various hospital business systems, forming a raw data set. Priority verification with tiered access control is then performed, assigning different data access and acquisition permissions based on roles, and recording data acquisition timestamps and data source identifiers. A multi-source data acquisition terminal was built, employing three acquisition methods: interface call, direct database connection, and real-time capture. Structured data, semi-structured data, and unstructured data from various business systems of the hospital were collected and aggregated to form a raw data set. The structured data included basic patient information and medical records, the semi-structured data included medical process forms, and the unstructured data included medical images and medical records. For each piece of raw data collected, the collection timestamp and data source identifier are recorded synchronously.

[0008] Before the steps of building a multi-source data acquisition terminal and using three acquisition methods—interface calls, direct database connections, and real-time capture—to collect structured, semi-structured, and unstructured data from various hospital business systems and aggregate them into a raw dataset: Assign corresponding data access permissions and collection permissions to job roles.

[0009] In the process of building a multi-source data acquisition terminal, using three acquisition methods—interface calls, direct database connections, and real-time capture—to collect structured, semi-structured, and unstructured data from various hospital business systems and aggregate them into a raw data set: In the interface call data collection method, a data collection request is initiated through a standardized interface that is pre-connected to various business systems of the hospital. After the interface permission is verified, the target data of the corresponding system is acquired in real time at a preset frequency. In the direct database access acquisition method, for business systems with no open interfaces and large data volumes, a dedicated database access account is configured to connect to the backend database of the business system and extract the required heterogeneous data according to preset data filtering rules. In the real-time data capture method, for semi-structured and unstructured data that cannot be obtained through direct connection to the interface and database, a lightweight data capture program is deployed to monitor the specified data display interface and storage path of the target business system in real time, capture data according to rules, and perform preliminary processing.

[0010] Among the steps involved in building a unified system, mapping, transforming, and performing dual verification of multi-source heterogeneous data based on a medical data mapping rule base, and outputting standardized data: Construct a unified data mapping and transformation system, and build a medical data mapping rule base to classify various mapping relationships between different data sources and different data formats; Data mapping operations are performed based on a medical data mapping rule base, using a dynamic mapping method that adjusts the mapping rules according to real-time changes in the data. Perform data transformation operations to convert semi-structured data into structured data, and use natural language processing technology to extract key features from unstructured data and convert them into structured feature data; Perform a double verification operation to check the rationality of the mapping relationship, the standardization of the conversion process, and the accuracy of the conversion result.

[0011] In the step of performing data mapping operations based on a medical data mapping rule base, a dynamic mapping method is adopted, and the mapping rules are adjusted according to real-time changes in the data: The system calls upon preset mapping relationships in the medical data mapping rule base to match each field in the semi-structured data with the corresponding field in the structured data for both structured and semi-structured data. For unstructured data, key information is initially extracted through natural language processing. The extracted key information is then mapped to the feature fields of structured data to establish the association between unstructured and structured data. For different coding systems, semantic mapping is performed based on the semantic correspondence in the medical ontology database. Real-time monitoring of changes in data format and content; dynamic adjustment of mapping rules when new fields are added or encodings are updated.

[0012] In the data transformation process, which involves converting semi-structured data into structured data and extracting key features from unstructured data using natural language processing techniques, and then converting these features into structured feature data: For semi-structured data, according to the preset field splitting rules, the information in the semi-structured data is split into the corresponding preset fields, the field attributes are supplemented and the field format is standardized, and the data is converted into structured data. For unstructured data, word segmentation, entity recognition, and keyword extraction are performed on text-based unstructured data to extract key features of patient symptoms, treatment measures, and follow-up results, which are then converted into structured feature data. For non-text-based unstructured data, visual features such as image size and lesion location are extracted and converted into structured feature parameters. Unify the encoding format of all data, identify and remove invalid data with format errors or exceeding reasonable limits during the conversion process, identify and delete duplicate data through data fingerprint comparison, and identify and remove abnormal data through anomaly detection algorithms.

[0013] Among them, the steps of performing dual verification operations to verify the rationality of the mapping relationship, the standardization of the conversion process, and the accuracy of the conversion result are as follows: The first layer of verification is the mapping relationship verification, which reviews the mapping relationships in the medical data mapping rule base one by one to verify whether the field mapping, feature mapping, and semantic mapping comply with the hospital's medical business specifications and data standards, and whether there are any mapping errors or omissions. The second layer of verification involves verifying the conversion process and results. During the conversion process, it verifies whether the steps and rules for data conversion are standardized, whether the conversion operation is carried out strictly according to the preset process, and whether the conversion tool is running normally. In the conversion results verification, a three-level consistency verification strategy is adopted. The structural consistency verification focuses on checking whether the field types, field lengths, and formats of the converted data are uniform and standardized, and whether they are consistent with the preset standards. The semantic consistency verification, combined with the medical ontology, verifies whether the encoding and terminology of the converted data are uniform, whether the semantics are coherent, and whether there is any ambiguity. The logical consistency verification is based on medical business logic and verifies whether the logical relationships between the data are reasonable. Data that fails verification is returned to the corresponding conversion stage for reprocessing until it passes verification.

[0014] Among them, the steps of using a two-layer fusion of feature layer and decision layer, performing deep fusion of multi-source data based on standardized data, outputting a unified data mart, and associating hierarchical permission control rules to set corresponding storage permissions for different types of fused data are as follows: Feature layer fusion is performed by extracting features from standardized data from different data sources, dynamically allocating fusion weights, and performing weighted fusion of features to obtain a unified feature vector. To achieve decision-level fusion, feature vectors will be analyzed in multiple dimensions based on feature vectors, medical business rules, and data mapping relationships, and deep fusion of multi-source data will be carried out. Based on the hospital management scenario, the merged multi-source data is classified and organized to form a unified data mart; In the process of data fusion, hierarchical permission control is associated, and corresponding storage permissions are set for different types of fused data according to the type and sensitivity level of the fused data.

[0015] This invention also provides a hospital integrated management system based on multi-source heterogeneous data fusion, including a multi-source heterogeneous data acquisition module, a data mapping and conversion module, and a multi-source heterogeneous data fusion and hierarchical control module; wherein: The multi-source heterogeneous data acquisition module is used to build an acquisition terminal and collect heterogeneous data from various business systems of the hospital in three ways: interface call, direct database connection and real-time capture. It forms a raw data set and performs pre-verification of hierarchical permission control. Different data access and acquisition permissions are assigned based on roles, and the data acquisition timestamp and data source identifier are recorded. The data mapping and conversion module is used to build a unified system, perform mapping, conversion and dual verification of multi-source heterogeneous data based on the medical data mapping rule base, and output standardized data. The multi-source heterogeneous data fusion and hierarchical management module is used to perform deep fusion of multi-source data based on standardized data using a two-layer fusion of feature layer and decision layer, output a unified data mart, and associate hierarchical permission management rules to set corresponding storage permissions for different types of fused data.

[0016] This invention discloses a hospital integrated management system and method based on multi-source heterogeneous data fusion. The system comprises a multi-source heterogeneous data acquisition module, a data mapping and conversion module, and a multi-source heterogeneous data fusion and hierarchical control module, comprising the following steps: Constructing an acquisition terminal, using interface calls, direct database connections, and real-time capture to collect heterogeneous data from various hospital business systems, forming a raw data set, and performing pre-verification for hierarchical permission control, assigning different data access and acquisition permissions based on roles, and recording data acquisition timestamps and data source identifiers; Constructing a unified system, mapping, converting, and performing dual verification of multi-source heterogeneous data based on a medical data mapping rule base, and outputting standardized data; Employing a two-layer fusion approach (feature layer and decision layer), deeply fusion of multi-source data based on standardized data, outputting a unified data mart, and associating it with hierarchical permission control rules to set corresponding storage permissions for different types of fused data; Through these methods, multi-source heterogeneous data fusion, data mapping, and conversion are achieved, thereby meeting the needs of smart hospital management. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1This is a flowchart of the steps of the hospital comprehensive management method based on multi-source heterogeneous data fusion of the present invention.

[0019] Figure 2 This is a flowchart of steps S100 of the present invention.

[0020] Figure 3 This is a flowchart of steps S200 of the present invention.

[0021] Figure 4 This is a flowchart of steps S300 of the present invention.

[0022] Figure 5 This is a schematic diagram of the hospital integrated management system based on multi-source heterogeneous data fusion according to the present invention.

[0023] Figure 6 This is a schematic diagram of the electronic device of the present invention.

[0024] 401 - Multi-source heterogeneous data acquisition module, 402 - Data mapping and conversion module, 403 - Multi-source heterogeneous data fusion and hierarchical management module. Detailed Implementation

[0025] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0026] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0027] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0028] Please see Figures 1-4 This invention provides a hospital comprehensive management method based on multi-source heterogeneous data fusion, comprising the following steps: S100: Construct a data acquisition terminal, employing three methods—interface calls, direct database connections, and real-time capture—to collect heterogeneous data from various hospital business systems, forming a raw data set. Pre-verification with hierarchical access control is then performed, assigning different data access and acquisition permissions based on roles, while simultaneously recording data acquisition timestamps and data source identifiers. The specific process is as follows: S101: Assign corresponding data access permissions and data collection permissions to job roles; S102: Build a multi-source data acquisition terminal, using three acquisition methods: interface call, direct database connection, and real-time capture, to collect structured data, semi-structured data, and unstructured data from various business systems in the hospital, and summarize them to form a raw data set; among them, structured data includes basic patient information and medical records, semi-structured data includes medical process forms, and unstructured data includes medical images and medical records. In the interface call data collection method, a data collection request is initiated through a standardized interface that is pre-connected to various business systems of the hospital. After the interface permission is verified, the target data of the corresponding system is acquired in real time at a preset frequency. In the direct database access acquisition method, for business systems with no open interfaces and large data volumes, a dedicated database access account is configured to connect to the backend database of the business system and extract the required heterogeneous data according to preset data filtering rules. In the real-time capture and collection method, for semi-structured and unstructured data that cannot be obtained through direct connection to the interface and database, a lightweight capture program is deployed to monitor the specified data display interface and storage path of the target business system in real time, capture data according to rules and perform preliminary processing. S103: For each piece of raw data collected, the collection timestamp and data source identifier are recorded synchronously.

[0029] In the above process, corresponding data access permissions and collection permissions are assigned to job roles. Specifically, the responsibilities of each position in the hospital are reviewed, the access and collection needs of different positions for medical data are clarified, and a role-based access control (RBAC) model is adopted to divide different roles such as core management positions, business operation positions, and medical and nursing positions. A precise permission scope is assigned to each role: core management positions can access and collect all types of data, business operation positions can only access and collect non-sensitive data within their corresponding business scope, and medical and nursing positions can only access and collect relevant data of the patients they are responsible for. After the permissions are assigned, they are entered into the permission management system and the personnel information is synchronized to ensure that the permissions and roles are accurately matched. At the same time, the effective time of the permissions is set. After the permission assignment is completed, an initial verification is performed to confirm that there are no errors or omissions in permission assignment before it is put into use.

[0030] A multi-source data acquisition terminal was built, employing three acquisition methods: interface calls, direct database connections, and real-time capture. Structured, semi-structured, and unstructured data were collected from various hospital business systems, and aggregated to form a raw data set. The structured data included patient basic information and medical records; the semi-structured data included medical process forms; and the unstructured data included medical images and medical records. The specific methods were as follows: Building a multi-source data acquisition terminal: The acquisition terminal is built using industrial-grade hardware equipment, equipped with multi-interface compatible modules, and adapted to mainstream business systems such as hospital HIS (Hospital Information System), LIS (Laboratory Information System), PACS (Picture Archiving and Communication System), and EMR (Electronic Medical Record System). Dedicated data acquisition software is installed, and a data caching module is configured to avoid data loss during the acquisition process. At the same time, an access control plugin is embedded to link with the access control system to ensure that the acquisition operation complies with the access control requirements. After the construction is completed, debugging is carried out to verify the compatibility and stability of the acquisition terminal and ensure that it can collect data from various business systems normally.

[0031] Interface call data collection method: Beforehand, the system interfaces are integrated with the management of each hospital's business system to obtain standardized open interfaces for each system. Interface call specifications, data transmission formats, and access permissions are clarified, and this interface information is entered into the interface management module of the data collection terminal. During data collection, the data collection terminal initiates a data collection request to the target business system through the standardized interface according to the preset collection requirements. The request carries the current data collection personnel's permission identifier. After passing the business system's interface permission verification (verifying the validity of the permission identifier and the reasonableness of the collection scope), the target data from the corresponding system is acquired at a preset frequency (real-time collection, initiating a collection request every 10 seconds). The collected data is transmitted in real-time to the data collection terminal's cache module to ensure the timeliness, standardization, and compatibility of data collection.

[0032] Direct database connection data acquisition method: For business systems with no open interfaces, low data transmission efficiency, and large data volumes (such as EMR systems and financial data systems in large hospitals), a dedicated database access account is configured in collaboration with the hospital's information department. This account is only assigned data read permissions and has no data modification or deletion permissions, ensuring the data security of the business system. The acquisition terminal establishes a stable connection with the backend database of the business system by entering the account and password through a database connection tool. According to preset data filtering rules (such as filtering relevant data for a specified time period or a specified department), the required heterogeneous data is extracted through SQL query statements. The extraction process adopts a batch extraction mode to avoid the impact of extracting a large amount of data at once on the normal operation of the business system. After preliminary deduplication, the extracted data is stored in the acquisition terminal cache and then aggregated into the original dataset.

[0033] Real-time data capture method: For semi-structured and unstructured data (such as medical records, medical image preview files, follow-up recordings, etc.) that cannot be obtained through direct interface and database connection, a lightweight capture program is deployed on the collection terminal, and capture rules are configured (specifying the URL of the target business system's specified data display interface, data storage path, and capture frequency). The capture program monitors data changes on the target interface or storage path in real time. When new data is generated or updated, it captures data according to preset rules. Among them, text-based semi-structured data (such as screenshots of diagnosis and treatment process forms) uses OCR recognition technology to extract text information, while unstructured data (such as medical images and follow-up recordings) directly captures the original files and labels the data type. After the capture is completed, the data is initially sorted (labeling the data source and capture time) and stored in the cache module to ensure the comprehensiveness of data collection.

[0034] Data aggregation: The structured, semi-structured, and unstructured data obtained from the three acquisition methods are uniformly imported into the data aggregation module of the acquisition terminal. The module automatically identifies the data types and classifies and marks them, removes obviously invalid data (such as blank data and garbled data), and aggregates them into a complete set of raw data. At the same time, an acquisition log is generated to record the quantity, type, and acquisition time of the acquired data.

[0035] For each piece of raw data collected, the collection timestamp and data source identifier are recorded synchronously. Specifically, the collection terminal is configured with a time synchronization module that interfaces with the hospital's unified time server to ensure the accuracy of the collection timestamp. For each piece of raw data collected, the system automatically generates a corresponding collection timestamp (accurate to the second). At the same time, based on the data source, a unique data source identifier is assigned to each piece of data (the identifier format is "business system abbreviation-collection method-data type", such as "HIS-interface call-structured"). The timestamp and data source identifier are bound and stored with the raw data to form a data traceability tag, ensuring that the collection time and source of each piece of data can be quickly queried during subsequent data processing, verification, and tracing, avoiding confusion in data traceability.

[0036] S200: Construct a unified system to map, transform, and double-verify multi-source heterogeneous data based on a medical data mapping rule base, and output standardized data; the specific process is as follows: S201: Construct a unified data mapping and transformation system, and build a medical data mapping rule base to classify various mapping relationships between different data sources and different formats; S202: Data mapping operations are performed based on a medical data mapping rule base, using a dynamic mapping method that adjusts the mapping rules according to real-time data changes; the process is as follows: The system calls upon preset mapping relationships in the medical data mapping rule base to match each field in the semi-structured data with the corresponding field in the structured data for both structured and semi-structured data. For unstructured data, key information is initially extracted through natural language processing. The extracted key information is then mapped to the feature fields of structured data to establish the association between unstructured and structured data. For different coding systems, semantic mapping is performed based on the semantic correspondence in the medical ontology database. Real-time monitoring of changes in data format and content; dynamic adjustment of mapping rules when new fields are added or encodings are updated. S203: Perform data transformation operations, converting semi-structured data into structured data, and extracting key features from unstructured data using natural language processing techniques and converting them into structured feature data; the process is as follows: For semi-structured data, according to the preset field splitting rules, the information in the semi-structured data is split into the corresponding preset fields, the field attributes are supplemented and the field format is standardized, and the data is converted into structured data. For unstructured data, word segmentation, entity recognition, and keyword extraction are performed on text-based unstructured data to extract key features of patient symptoms, treatment measures, and follow-up results, which are then converted into structured feature data. For non-text-based unstructured data, visual features such as image size and lesion location are extracted and converted into structured feature parameters. Unify the encoding format of all data, identify and remove invalid data with format errors or exceeding reasonable range during the conversion process, identify and delete duplicate data through data fingerprint comparison, and identify and remove abnormal data through anomaly detection algorithm; S204: Perform a double verification operation to check the rationality of the mapping relationship, the standardization of the conversion process, and the accuracy of the conversion result; the process is as follows: The first layer of verification is the mapping relationship verification, which reviews the mapping relationships in the medical data mapping rule base one by one to verify whether the field mapping, feature mapping, and semantic mapping comply with the hospital's medical business specifications and data standards, and whether there are any mapping errors or omissions. The second layer of verification involves verifying the conversion process and results. During the conversion process, it verifies whether the steps and rules for data conversion are standardized, whether the conversion operation is carried out strictly according to the preset process, and whether the conversion tool is running normally. In the conversion results verification, a three-level consistency verification strategy is adopted. The structural consistency verification focuses on checking whether the field types, field lengths, and formats of the converted data are uniform and standardized, and whether they are consistent with the preset standards. The semantic consistency verification, combined with the medical ontology, verifies whether the encoding and terminology of the converted data are uniform, whether the semantics are coherent, and whether there is any ambiguity. The logical consistency verification is based on medical business logic and verifies whether the logical relationships between the data are reasonable. Data that fails verification is returned to the corresponding conversion stage for reprocessing until it passes verification.

[0037] In the above process, a unified data mapping and transformation system is constructed, and a medical data mapping rule base is built to classify various mapping relationships between different data sources and different data formats; the specific method is as follows: Construct a unified data mapping and transformation system: Based on hospital medical business specifications and national medical data standards, build a unified system consisting of a data mapping module, a data transformation module, and a verification module, clearly defining the division of responsibilities for each module: the data mapping module is responsible for mapping various types of data, the data transformation module is responsible for data format and encoding conversion and data cleaning, and the verification module is responsible for verifying the mapping, transformation process and results. Establish a data linkage mechanism between the modules to ensure a smooth data flow from mapping to transformation and verification. At the same time, configure a system management interface to facilitate technical personnel to maintain the system and adjust parameters.

[0038] A medical data mapping rule base was established: Based on SNOMEDCT (Systematic Medical Terminology) and LOINC (Logical Naming and Coding System for Observational Indicators), a medical ontology was constructed. The data sources and formats of various hospital business systems were analyzed, and three types of mapping relationships were defined and entered into the rule base: First, field mapping between structured and semi-structured data, clarifying the correspondence between each field in semi-structured data (such as treatment process forms) and the corresponding field in structured data (such as patient treatment records) (e.g., "Date of Visit" in a form corresponds to "Time of Visit" in the record); second, feature mapping between unstructured and structured data, clarifying the correspondence between key information in unstructured data and feature fields in structured data (e.g., "Fever" symptom in a medical record corresponds to "Symptom Code" in structured data); third, semantic mapping between different coding systems, analyzing the correspondence between different coding systems such as ICD-9 and ICD-10, and establishing coding mapping rules based on the semantic interpretation of the medical ontology. The rule base adopts an extensible architecture, supporting the addition of mapping rules for new data sources and new data formats. After entry, the rules are jointly reviewed by medical business experts and technical personnel to ensure that the mapping rules conform to medical business logic.

[0039] Data mapping operations are performed based on a medical data mapping rule base, using a dynamic mapping approach that adjusts the mapping rules according to real-time data changes. The process is as follows: Calling mapping rules: The data mapping module automatically calls the preset mapping relationships in the medical data mapping rule library. Based on the type and data source of the original data, it matches the corresponding mapping rules to ensure that the mapping rules are accurately matched with the data type and avoid mapping deviations.

[0040] Structured and semi-structured data mapping: For semi-structured data (such as treatment process forms and medical order templates), the mapping module accurately matches each field in the semi-structured data with the corresponding field in the structured data according to preset field mapping rules, clarifying the field correspondence and data conversion rules (such as matching the "patient name" field in the form with the "name" field in the treatment record to ensure that the field content is consistent). After the matching is completed, a field mapping reference table is generated for subsequent data conversion.

[0041] Mapping Unstructured and Structured Data: For unstructured data, key information is first extracted using Natural Language Processing (NLP) technology. A dedicated NLP program is deployed to perform word segmentation, part-of-speech tagging, and entity recognition on text-based unstructured data (such as medical records and transcripts of follow-up recordings) to extract key information such as patient symptoms, treatment measures, and follow-up results. For non-textual unstructured data (such as medical images and pathological slides), visual features such as image size, lesion location, and lesion size are extracted using image recognition technology. After extraction, the key information and visual features are mapped to the feature fields of structured data to establish a relationship between unstructured and structured data (e.g., mapping the "lesion location" feature in medical images to the "lesion site" field in structured data), ensuring that unstructured data can be used collaboratively with structured data.

[0042] Semantic mapping for different coding systems: For different coding systems (such as ICD-9 and ICD-10, hospital internal coding and national unified coding), the mapping module automatically performs semantic mapping for different codes based on the semantic correspondence in the medical ontology database, converting non-unified codes into standard codes specified by the hospital (such as converting ICD-9 codes into ICD-10 codes). During the mapping process, cases of ambiguous codes or unclear semantics are automatically marked and reminded to technical personnel for manual review to ensure the accuracy of the coding mapping.

[0043] Dynamically adjust mapping rules: The mapping module monitors changes in data format and content in real time. By comparing data features, it identifies situations such as newly added fields, updated codes, and changed field names. When data changes are detected, it automatically calls the semantic information of the medical ontology library and combines it with historical mapping data to dynamically adjust the corresponding mapping rules. At the same time, it records the rule adjustment log (including adjustment time, adjustment content, and adjustment reason). After the adjustment is completed, it is reviewed and confirmed by technical personnel to ensure that the adjusted mapping rules comply with medical business specifications and avoid mapping errors caused by data changes.

[0044] The data transformation process involves converting semi-structured data into structured data, and using natural language processing techniques to extract key features from unstructured data and convert them into structured feature data. The process is as follows: Semi-structured data conversion: For semi-structured data (such as treatment process forms and medical order templates), the data conversion module splits the information in the semi-structured data into corresponding preset fields according to preset field splitting rules, supplements field attributes (such as field type, field length, and required field identifiers), standardizes field formats (such as date format being uniformly "YYYY-MM-DD", and numeric format being uniformly retained to two decimal places), and automatically marks and temporarily stores any missing fields or incomplete information that occurs during the splitting process. These are then supplemented and improved in conjunction with medical business logic, and finally converted into structured data that conforms to hospital data standards.

[0045] Unstructured data transformation: Unstructured data is processed in two categories: First, for text-based unstructured data (such as medical records and transcripts of follow-up audio recordings), a dedicated natural language processing program is deployed to preprocess the data (removing garbled characters, redundant characters, and invalid spaces), followed by word segmentation, entity recognition, and keyword extraction to extract key features such as patient symptoms, treatment measures, follow-up results, and medication information. The extracted key features are then entered into the corresponding structured feature fields according to a preset format, converting the data into structured feature data. Second, for non-text-based unstructured data (such as medical images and pathological slides), visual features such as image size, lesion location, lesion size, and lesion morphology are extracted using image recognition algorithms. These visual features are then converted into standardized structured feature parameters (e.g., lesion size is converted to "XXmm×XXmm"), linked to the patient's basic information, and form structured feature data.

[0046] Data encoding and format unification: The conversion module performs encoding and format unification processing on all converted structured data and structured feature data—the encoding is uniformly converted to ICD-10 encoding (disease encoding) and SNOMEDCT encoding (terminology encoding) to ensure uniform encoding standards; the data file format is uniformly converted to the JSON standard format to facilitate subsequent data fusion, storage and access; during the unification process, data with non-standard encoding or incorrect format is automatically marked and returned to the corresponding conversion stage for reprocessing.

[0047] Data cleaning and processing: Data cleaning is carried out simultaneously during the conversion process to remove invalid, duplicate, and abnormal data: Invalid data (such as patient age greater than 120 years old or blood pressure values ​​outside the normal range) is identified and deleted using preset rules (such as incorrect formatting, values ​​outside the reasonable range, and blank data); duplicate data is identified and deleted by calculating data fingerprints (hash value calculation for each data entry) to ensure data uniqueness; abnormal data (such as data inconsistent with patient medical history and treatment logic) is identified and removed by analyzing data characteristics using an anomaly detection algorithm (isolated forest algorithm). The specific process is as follows: First, the preliminarily standardized structured data and structured feature data after conversion are processed. As the input dataset for the Isolation Forest algorithm, the input data undergoes feature standardization, transforming all feature parameters to the same order of magnitude (using the min-max standardization method to map the data to the [0,1] interval), eliminating the impact of differences in feature dimension magnitudes on the algorithm's detection results. Secondly, the Isolation Forest model is initialized, setting the number of decision trees (100-200 trees depending on the data size) and the sample size for each decision tree (80% of the total input dataset). Samples are randomly drawn from the input dataset to construct multiple isolated trees forming the Isolation Forest. Then, each piece of data to be detected is input into the Isolation Forest, and the results are processed by each isolated tree. The data is path-divided, and the path length of the data in each isolated tree is calculated (i.e., the number of edges the data traverses from the root to a leaf node). The average of the path lengths in all isolated trees is taken as the anomaly score for that data. Next, an anomaly score threshold is set (based on the characteristics of medical data and business logic, the threshold is calibrated through multiple tests, usually set as the mean of all anomaly scores plus 1.5 times the standard deviation). When the anomaly score of data is higher than this threshold, it is judged as abnormal data. Data that is inconsistent with the patient's medical history and treatment logic is prioritized for screening (such as data where the patient has a history of diabetes but has normal blood glucose levels and no record of hypoglycemic treatment, or data with test results that contradict the surgical indications shortly after surgery, etc.). Finally, data deemed abnormal undergoes manual review. Medical business experts confirm that the data indeed contains logical anomalies and lacks a reasonable medical explanation before removing it from the dataset. Simultaneously, the anomaly identifier, anomaly type, and removal time are recorded, forming an anomaly data processing log to ensure the data cleanup process is traceable. For missing data, priority is given to combining medical business logic with interpolation based on patient medical history, relevant examination results, and data from similar patients. For data that cannot be supplemented through business logic, supplementation methods based on medical data patterns are used (such as referencing the average data of patients in the same department and with the same symptoms) to ensure the rationality and accuracy of the supplemented data and guarantee the integrity of the transformed data.

[0048] Perform a double verification operation to check the rationality of the mapping relationship, the standardization of the conversion process, and the accuracy of the conversion result; the process is as follows: The first layer of verification (mapping relationship verification): A verification team composed of medical business experts and technicians is formed to review the mapping relationships in the medical data mapping rule base one by one, using a combination of manual review and automatic system verification. The automatic system verification mainly checks the completeness and non-duplication of mapping relationships and identifies mapping errors and omissions. The manual review focuses on verifying whether the field mapping, feature mapping, and semantic mapping comply with the hospital's medical business specifications and data standards, whether they are consistent with the actual diagnosis and treatment logic, and whether there are problems such as mapping deviations and semantic ambiguities. For unreasonable mapping relationships found in the review, the verification team puts forward adjustment suggestions, and the technicians optimize the mapping rules according to the suggestions. After the adjustment is completed, the review is carried out again until all mapping relationships meet the requirements.

[0049] The second level of verification (conversion process and result verification): This consists of two parts: conversion process verification and conversion result verification, and is carried out using a combination of automatic system verification and manual spot checks. Conversion process verification: The verification module monitors the entire data conversion process in real time, checking whether the data conversion steps and rules are strictly carried out according to the preset process, whether the conversion tool is running normally, whether the data flow is smooth, and whether there are any problems such as data loss or conversion interruption. For any abnormal situations that occur during the conversion process, the module automatically records and alarms, and technical personnel promptly investigate and handle them to ensure that the conversion process is standardized and stable.

[0050] Transformation result verification: A three-level consistency verification strategy is adopted to ensure the accuracy and availability of the transformed data. Structure consistency check: The system automatically checks whether the field types, field lengths, and formats of the converted data are uniform and standardized, and whether they are consistent with the preset data standards (such as whether the date format, numeric format, and encoding format meet the requirements). It also checks whether the fields are complete and without missing data. Data with inconsistent structures is automatically marked and returned to the data conversion stage for reprocessing.

[0051] Semantic consistency verification: In conjunction with the medical ontology database, the system automatically verifies whether the encoding and terminology of the converted data are consistent, and whether the semantics are coherent and unambiguous (such as whether the disease code corresponds to the disease name, and whether the terminology description conforms to medical standards). Data with inconsistent semantics or incorrect encoding is automatically marked, and then manually reviewed and confirmed by medical business experts. After adjustment, it is re-verified.

[0052] Logical consistency verification: Based on the hospital's medical business logic, the system automatically checks whether the logical relationships between data are reasonable (such as whether the patient's age matches the treatment measures, whether the treatment time conforms to the chronological logic, whether the examination results are consistent with the diagnosis conclusion, etc.). At the same time, some data are manually sampled to further verify the logical rationality. For data with inconsistent logic, it is returned to the corresponding transformation stage and reprocessed in combination with the original data and mapping rules.

[0053] Verification result processing: For all data that fails verification, the system automatically records the reason for failure and the data identifier, and returns it to the corresponding conversion stage (e.g., data with inconsistent structure is returned to the format unification stage, and data with inconsistent semantics is returned to the encoding conversion stage). The conversion process is repeated, and the data is submitted for verification again after processing, until all data passes verification, ensuring that the converted data meets the needs of subsequent multi-source data fusion and hospital management.

[0054] S300: Employs a two-layer fusion approach, combining feature and decision layers. It performs deep fusion of multi-source data based on standardized data, outputting a unified data mart and associating it with hierarchical access control rules to set corresponding storage permissions for different types of fused data. The specific process is as follows: S301: Perform feature layer fusion, extract features from standardized data from different data sources, dynamically allocate fusion weights and perform weighted feature fusion to obtain a unified feature vector; S302: Conduct decision-level fusion, which will perform multi-dimensional analysis of feature vectors based on feature vectors, medical business rules and data mapping relationships, and conduct deep fusion of multi-source data; S303: Based on the hospital management scenario, classify and organize the merged multi-source data to form a unified data mart; S304: In the process of data fusion, hierarchical permission control is associated, and corresponding storage permissions are set for different types of fused data according to the type and sensitivity level of the fused data.

[0055] In the above process, feature layer fusion is performed, which involves extracting features from standardized data from different data sources, dynamically assigning fusion weights, and performing weighted feature fusion to obtain a unified feature vector; the specific method is as follows: Feature extraction: Feature extraction algorithms (such as PCA principal component analysis algorithm) are used to extract features from standardized data from different data sources (HIS, LIS, PACS, EMR, etc.). The core features of each type of data are extracted (such as disease features in patient diagnosis and treatment data, indicator features in examination and testing data, and utilization features in resource data). Redundant features are removed, and core features that are valuable for hospital management are retained to form feature sets for each data source.

[0056] Integration Weight Allocation: An improved weighted integration method is adopted to dynamically allocate integration weights for each data source. First, medical business experts determine the initial business weights based on the business importance of each data source (e.g., EMR system data business weight 0.3, LIS system data business weight 0.2). Then, a data quality assessment algorithm is used to score the standardized data from each data source (based on data completeness, accuracy, and real-time performance, with a maximum score of 10). The specific process is as follows: First, the three core dimensions of data quality assessment (completeness, accuracy, and real-time performance) are defined, and corresponding weights are assigned to each dimension. Completeness accounts for 40% (maximum score 4 points), accuracy accounts for 40% (maximum score 4 points), and real-time performance accounts for 20% (maximum score 2 points). The weight allocation is combined with the medical data usage scenarios to ensure that the scoring aligns with hospital management needs. Second, specific scoring standards are developed for each assessment dimension, and data quality testing and scoring are carried out one by one. 1. Completeness Score (4 points): The missing rate statistical method is used to detect missing fields in the standardized data of each data source. The missing field of each data point is counted as the proportion of missing fields to the total number of fields in the data source. Missing rate = (number of missing fields / total number of fields) × 100%. The scoring rules are as follows: missing rate ≤ 5%, 4 points; 5% < missing rate ≤ 10%, 3 points; 10% < missing rate ≤ 20%, 2 points; 20% < missing rate ≤ 30%, 1 point; missing rate > 30%, 0 points. At the same time, the missing status of core fields (such as patient ID, treatment time, disease code, etc.) is checked. If a core field is missing, 2 points are deducted from the completeness score of the data source to ensure the completeness of core data.

[0057] 2. Accuracy Score (4 points): A combination of automatic verification and manual sampling is used. First, the system automatically compares the data with medical business specifications and standard codes (ICD-10, SNOMEDCT) to identify data errors (such as coding errors, numerical anomalies, and non-standard terminology). The percentage of erroneous data in the total data volume of the data source is calculated as follows: Error Rate = (Number of erroneous data / Total data volume) × 100%. Then, 10% of the sample data is randomly selected for review and verification by medical business experts to confirm the data logic. Logical rationality (such as consistency between medical data and medical history, and matching of examination results with diagnostic conclusions); scoring rules are as follows: 4 points for error rate ≤ 3% and no logical errors found in random checks; 3 points for error rate ≤ 5% or 1-2 minor logical errors found in random checks; 2 points for error rate ≤ 10% or 3-5 minor logical errors found in random checks; 1 point for error rate ≤ 15% or 1 serious logical error found in random checks; 0 points for error rate > 15% or 2 or more serious logical errors found in random checks.

[0058] 3. Real-time performance score (2 points): The update delay time of standardized data from each data source is calculated as follows: Delay time = data generation time - time to complete data standardization and synchronization to the fusion system. The scoring rules are as follows: ≤5 minutes, 2 points; 5 minutes < ≤10 minutes, 1.5 points; 10 minutes < ≤30 minutes, 1 point; 30 minutes < ≤60 minutes, 0.5 points; >60 minutes, 0 points. For data sources with high real-time requirements (such as emergency room data and intensive care unit data), the delay time scoring standard is increased: ≤3 minutes, 2 points; >30 minutes, 0 points. This adapts to the real-time requirements of different data sources.

[0059] Finally, a comprehensive quality score is calculated for each data source. The comprehensive score is calculated as follows: Completeness Score + Accuracy Score + Real-time Score. The final score is rounded to one decimal place, with a maximum score of 10. After scoring, a data quality assessment report is generated, recording the scores for each dimension of each data source, the reasons for deductions, and improvement suggestions. This provides accurate data support for the subsequent dynamic calculation of fusion weights. Combining business weights and data quality scores, the fusion weights of each data source are dynamically calculated to ensure that data sources with high data quality and high business importance have higher fusion weights, avoiding the impact of biases from a single data source on the fusion results.

[0060] Feature weighted fusion: The core features extracted from each data source are weighted and fused according to the calculated fusion weights. By linearly weighting and summing, the features from different data sources are integrated into a unified feature vector, eliminating the differences between features from different data sources. This ensures that the feature vector can comprehensively and accurately reflect the core information of multi-source data, laying the foundation for subsequent decision-making fusion.

[0061] To achieve decision-level fusion, multi-dimensional analysis of feature vectors will be conducted based on feature vectors, medical business rules, and data mapping relationships, resulting in deep fusion of multi-source data. Specifically, the method is as follows: The process of generating the fusion decision model is as follows: A Transformer-based fusion decision model is constructed. Data preparation: Collect historical multi-source heterogeneous data from the hospital over the past 3 years. After S100 collection, S200 mapping transformation and verification, a standardized training dataset (80%) and a test dataset (20%) are formed. The data categories are labeled (such as medical quality, resource scheduling, and patient management). Model initialization: Build the Transformer model architecture, set the number of encoder and decoder layers, configure the attention mechanism, and initialize model parameters (such as learning rate and number of iterations). Model training: Input the training dataset into the model, combine medical business rules (such as diagnosis and treatment guidelines and management standards) and data mapping relationships, set the cross-entropy loss function, and use the gradient descent method to train the model. The learning rate is set to 0.001, the number of iterations is set to 100 rounds, and the model performance is verified through the test dataset after each round of training. Model optimization: Based on the validation results of the test dataset, adjust the model parameters (such as attention weights and number of iterations), introduce low-rank adaptation (LoRA) fine-tuning technology to improve model training efficiency and adaptability until the model's fusion accuracy reaches over 98%, ensuring that the model can accurately achieve deep fusion of multi-source data; 5. Model deployment: Deploy the trained fusion decision model to the data fusion system, linking it with the feature layer fusion module to ensure that feature vectors can be input into the model for processing in real time.

[0062] Deep fusion of multi-source data: The unified feature vector obtained by feature layer fusion is input into the deployed fusion decision model. The model combines the preset medical business rules (such as diagnosis and treatment process specifications and resource scheduling standards) and data mapping relationships to perform multi-dimensional analysis on the feature vector (such as correlation analysis, trend analysis, and cluster analysis), identify the inherent relationship between data from different data sources, eliminate data redundancy and conflicts, and deeply integrate multi-source data into unified fused data, ensuring the integrity, accuracy and consistency of the fused data.

[0063] Based on hospital management scenarios, the integrated multi-source data is categorized and organized to form a unified data mart. Specifically, the core hospital management scenarios are identified and divided into four main categories: medical quality, resource allocation, cost accounting, and patient management, each with specific data requirements. The integrated multi-source data is then categorized and organized according to these management scenarios, extracting the core data required for each scenario, removing irrelevant data, and establishing categorized data subsets. Each data subset is then standardized, with unified data naming and formatting, and a data index is established, forming medical quality, resource allocation, cost accounting, and patient management data marts. Relationships are established between these data marts to support cross-scenario data querying and analysis, breaking down data silos and providing comprehensive and accurate data support for hospital integrated management.

[0064] During the data fusion process, hierarchical access control is implemented. Based on the type and sensitivity level of the fused data, corresponding storage permissions are set for different types of fused data. Specifically, in conjunction with the hospital's access control rules, the fused data is categorized by sensitivity level into three types: core sensitive data (such as patient privacy information and core financial data), general sensitive data (such as performance evaluation data for medical staff), and ordinary data (such as routine medical statistics). For fused data of different sensitivity levels, corresponding storage access permissions are set based on job role permissions: core sensitive data is only accessible to core management positions, general sensitive data is accessible to corresponding management and operational positions, and ordinary data is accessible to all authorized positions. These permission rules are embedded in the data fusion system. After the fused data is generated, it is automatically associated with corresponding permission identifiers. Storage areas are divided according to permission levels, and users with different permission levels can only access the fused data in their corresponding storage areas. Simultaneously, data access logs are recorded to ensure the security of data storage and access, preventing unauthorized access and leakage.

[0065] Corresponding to the aforementioned embodiments of the hospital integrated management method based on multi-source heterogeneous data fusion, this application also provides embodiments of a hospital integrated management system based on multi-source heterogeneous data fusion.

[0066] Figure 5 This is a schematic diagram illustrating the structural principle of a hospital integrated management system based on multi-source heterogeneous data fusion, according to an exemplary embodiment. (Refer to...) Figure 5 The system may include: a multi-source heterogeneous data acquisition module 401, a data mapping and conversion module 402, and a multi-source heterogeneous data fusion and hierarchical management module 403; wherein: The multi-source heterogeneous data acquisition module 401 is used to build an acquisition terminal and use three methods, namely interface call, direct database connection and real-time capture, to collect heterogeneous data from various business systems of the hospital, form a raw data set, and perform pre-verification of hierarchical permission control, assign different data access and acquisition permissions based on roles, and record the data acquisition timestamp and data source identifier. The data mapping and conversion module 402 is used to construct a unified system, perform mapping, conversion and dual verification of multi-source heterogeneous data based on the medical data mapping rule base, and output standardized data. The multi-source heterogeneous data fusion and hierarchical management module 403 is used to perform deep fusion of multi-source data based on standardized data by adopting a two-layer fusion of feature layer and decision layer, outputting a unified data mart, and associating hierarchical permission management rules to set corresponding storage permissions for different types of fused data.

[0067] In this embodiment, the multi-source heterogeneous data acquisition module 401 constructs an acquisition terminal, employing three methods—interface calls, direct database connections, and real-time capture—to collect heterogeneous data from various hospital business systems, forming a raw data set. Pre-verification with hierarchical permission control is performed, assigning different data access and acquisition permissions based on roles, and recording data acquisition timestamps and data source identifiers. The data mapping and conversion module 402 constructs a unified system, mapping, converting, and performing dual verification of multi-source heterogeneous data based on a medical data mapping rule base, and outputting standardized data. The multi-source heterogeneous data fusion and hierarchical management module 403 employs a two-layer fusion approach—feature layer and decision layer—to perform deep fusion of multi-source data based on standardized data, outputting a unified data mart, and associating it with hierarchical permission control rules to set corresponding storage permissions for different types of fused data. Through these methods, multi-source heterogeneous data fusion, mapping, and conversion are achieved, thereby meeting the needs of smart hospital management.

[0068] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0069] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0070] Accordingly, this application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; and, when the one or more programs are executed by the one or more processors, causing the one or more processors to implement the hospital integrated management method based on multi-source heterogeneous data fusion as described above. Figure 6 The diagram shown is a hardware structure diagram of any device with data processing capabilities within a hospital integrated management system based on multi-source heterogeneous data fusion, as provided in an embodiment of the present invention. (Except for...) Figure 6 In addition to the processor, memory, and network interface shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0071] Accordingly, this application also provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the hospital integrated management method based on multi-source heterogeneous data fusion as described above. The computer-readable storage medium can be an internal storage unit of any data processing device as described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0072] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0073] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A comprehensive hospital management method based on multi-source heterogeneous data fusion, characterized in that, Includes the following steps: A data acquisition terminal was built, which used three methods: interface call, direct database connection, and real-time capture to collect heterogeneous data from various business systems of the hospital, forming a raw data set. Prior verification of hierarchical permission control was performed, and different data access and acquisition permissions were assigned based on roles. At the same time, the data acquisition timestamp and data source identifier were recorded. A unified system is constructed to map, transform, and perform dual verification of multi-source heterogeneous data based on a medical data mapping rule base, and output standardized data. A two-layer fusion approach, consisting of a feature layer and a decision layer, is adopted. Based on standardized data, multi-source data is deeply fused to output a unified data mart and associated with hierarchical permission control rules to set corresponding storage permissions for different types of fused data.

2. The hospital comprehensive management method based on multi-source heterogeneous data fusion as described in claim 1, characterized in that, In building the data acquisition terminal, three methods were used: interface calls, direct database connections, and real-time capture. This process collected heterogeneous data from various hospital business systems to form a raw data set. Priority verification with tiered access control was performed, assigning different data access and acquisition permissions based on roles, and recording data acquisition timestamps and data source identifiers. A multi-source data acquisition terminal was built, employing three acquisition methods: interface call, direct database connection, and real-time capture. Structured data, semi-structured data, and unstructured data from various business systems of the hospital were collected and aggregated to form a raw data set. The structured data included basic patient information and medical records, the semi-structured data included medical process forms, and the unstructured data included medical images and medical records. For each piece of raw data collected, the collection timestamp and data source identifier are recorded synchronously.

3. The hospital comprehensive management method based on multi-source heterogeneous data fusion as described in claim 2, characterized in that, Before setting up a multi-source data acquisition terminal and using three acquisition methods—interface calls, direct database connections, and real-time capture—to collect structured, semi-structured, and unstructured data from various hospital business systems and aggregate them into a raw dataset: Assign corresponding data access permissions and collection permissions to job roles.

4. The hospital comprehensive management method based on multi-source heterogeneous data fusion as described in claim 2, characterized in that, In the process of building a multi-source data acquisition terminal, using three acquisition methods—interface calls, direct database connections, and real-time capture—to collect structured, semi-structured, and unstructured data from various hospital business systems and aggregate them into a raw data set: In the interface call data collection method, a data collection request is initiated through a standardized interface that is pre-connected to various business systems of the hospital. After the interface permission is verified, the target data of the corresponding system is acquired in real time at a preset frequency. In the direct database access acquisition method, for business systems with no open interfaces and large data volumes, a dedicated database access account is configured to connect to the backend database of the business system and extract the required heterogeneous data according to preset data filtering rules. In the real-time data capture method, for semi-structured and unstructured data that cannot be obtained through direct connection to the interface and database, a lightweight data capture program is deployed to monitor the specified data display interface and storage path of the target business system in real time, capture data according to rules, and perform preliminary processing.

5. The hospital comprehensive management method based on multi-source heterogeneous data fusion as described in claim 1, characterized in that, In the process of building a unified system, mapping, transforming, and double-validating multi-source heterogeneous data based on a medical data mapping rule base, and outputting standardized data: Construct a unified data mapping and transformation system, and build a medical data mapping rule base to classify various mapping relationships between different data sources and different data formats; Data mapping operations are performed based on a medical data mapping rule base, using a dynamic mapping method that adjusts the mapping rules according to real-time changes in the data. Perform data transformation operations to convert semi-structured data into structured data, and use natural language processing technology to extract key features from unstructured data and convert them into structured feature data; Perform a double verification operation to check the rationality of the mapping relationship, the standardization of the conversion process, and the accuracy of the conversion result.

6. The hospital comprehensive management method based on multi-source heterogeneous data fusion as described in claim 5, characterized in that, In the data mapping operation based on a medical data mapping rule base, a dynamic mapping method is used to adjust the mapping rules according to real-time changes in the data: The system calls upon preset mapping relationships in the medical data mapping rule base to match each field in the semi-structured data with the corresponding field in the structured data for both structured and semi-structured data. For unstructured data, key information is initially extracted through natural language processing. The extracted key information is then mapped to the feature fields of structured data to establish the association between unstructured and structured data. For different coding systems, semantic mapping is performed based on the semantic correspondence in the medical ontology database. Real-time monitoring of changes in data format and content; dynamic adjustment of mapping rules when new fields are added or encodings are updated.

7. The hospital comprehensive management method based on multi-source heterogeneous data fusion as described in claim 5, characterized in that, In the data transformation process, which involves converting semi-structured data into structured data and extracting key features from unstructured data using natural language processing techniques, and then converting these features into structured data: For semi-structured data, according to the preset field splitting rules, the information in the semi-structured data is split into the corresponding preset fields, the field attributes are supplemented and the field format is standardized, and the data is converted into structured data. For unstructured data, word segmentation, entity recognition, and keyword extraction are performed on text-based unstructured data to extract key features of patient symptoms, treatment measures, and follow-up results, which are then converted into structured feature data. For non-text-based unstructured data, visual features such as image size and lesion location are extracted and converted into structured feature parameters. Unify the encoding format of all data, identify and remove invalid data with format errors or exceeding reasonable limits during the conversion process, identify and delete duplicate data through data fingerprint comparison, and identify and remove abnormal data through anomaly detection algorithms.

8. The hospital comprehensive management method based on multi-source heterogeneous data fusion as described in claim 5, characterized in that, In the process of performing dual verification operations to check the rationality of the mapping relationship, the standardization of the conversion process, and the accuracy of the conversion result: The first layer of verification is the mapping relationship verification, which reviews the mapping relationships in the medical data mapping rule base one by one to verify whether the field mapping, feature mapping, and semantic mapping comply with the hospital's medical business specifications and data standards, and whether there are any mapping errors or omissions. The second layer of verification is the verification of the conversion process and results. During the conversion process, it verifies whether the steps and rules for data conversion are standardized, whether the conversion operation is carried out strictly in accordance with the preset process, and whether the conversion tool is running normally. In the verification of the conversion results, a three-level consistency verification strategy is adopted. The structural consistency verification focuses on checking whether the field types, field lengths, and formats of the converted data are uniform and standardized, and whether they are consistent with the preset standards. Semantic consistency verification, combined with a medical ontology, verifies whether the encoding and terminology of the converted data are consistent, and whether the semantics are coherent and unambiguous. Logical consistency verification is based on medical business logic, verifying whether the logical relationships between data are reasonable; Data that fails verification is returned to the corresponding conversion stage for reprocessing until it passes verification.

9. The hospital integrated management method based on multi-source heterogeneous data fusion as described in claim 1, characterized in that, In the process of employing a two-layer fusion approach involving feature layer and decision layer, performing deep fusion of multi-source data based on standardized data, outputting a unified data mart, and associating hierarchical permission control rules to set corresponding storage permissions for different types of fused data: Feature layer fusion is performed by extracting features from standardized data from different data sources, dynamically allocating fusion weights, and performing weighted fusion of features to obtain a unified feature vector. To achieve decision-level fusion, feature vectors will be analyzed in multiple dimensions based on feature vectors, medical business rules, and data mapping relationships, and deep fusion of multi-source data will be carried out. Based on the hospital management scenario, the merged multi-source data is classified and organized to form a unified data mart; In the process of data fusion, hierarchical permission control is associated, and corresponding storage permissions are set for different types of fused data according to the type and sensitivity level of the fused data.

10. A hospital integrated management system based on multi-source heterogeneous data fusion, employing the hospital integrated management method based on multi-source heterogeneous data fusion as described in claim 1, characterized in that, It includes a multi-source heterogeneous data acquisition module, a data mapping and conversion module, and a multi-source heterogeneous data fusion and hierarchical management module; among which: The multi-source heterogeneous data acquisition module is used to build an acquisition terminal and collect heterogeneous data from various business systems of the hospital in three ways: interface call, direct database connection and real-time capture. It forms a raw data set and performs pre-verification of hierarchical permission control. Different data access and acquisition permissions are assigned based on roles, and the data acquisition timestamp and data source identifier are recorded. The data mapping and transformation module is used to build a unified system, perform mapping, transformation and dual verification of multi-source heterogeneous data based on the medical data mapping rule base, and output standardized data. The multi-source heterogeneous data fusion and hierarchical management module is used to perform deep fusion of multi-source data based on standardized data using a two-layer fusion of feature layer and decision layer, output a unified data mart, and associate hierarchical permission management rules to set corresponding storage permissions for different types of fused data.

Images