A medical data multi-source fusion verification and correction method and system

Abstract

The present invention belongs to the technical field of medical data processing, and discloses a method and system for multi-source fusion verification and correction of medical data, the method comprising: acquiring multi-source data; constructing a dynamic baseline model of the emergency department, ICU and general ward, responding to and verifying abnormalities in patient data and issuing warnings; establishing a terminology library, integrating multi-source patient data acquired from multiple sources, and performing data corrections; generating a multi-source data quality evaluation table for different departments based on full-process data, and updating the terminology library and fusion correction weights based on feedback from the quality evaluation table. The present invention can timely discover and correct quality problems in data through real-time monitoring and intelligent early warning mechanisms, reduce human errors and significantly reduce data management costs. Through multi-source data fusion and consistency verification, the system can ensure seamless connection of data from different data sources, achieve data consistency and collaborative work across systems and platforms, thereby improving the accuracy of medical decision-making and the reliability of data.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical data processing, and in particular relates to a method and system for multi-source fusion verification and correction of medical data. Background Art

[0002] With the development of informatization in the medical industry, various medical data are being widely collected and stored, and the types and sources of data are constantly increasing. The accuracy, consistency, and completeness of medical data directly affect patients' diagnosis and treatment decisions and health management. However, current medical data management faces the following challenges:

[0003] 1. Diverse data sources and complex formats: Medical data comes from multiple systems such as HIS, EMR, LIMS, etc., and the data formats, structures, and storage methods are different, making data integration difficult.

[0004] 2. Data quality issues are prominent: Due to the complexity of medical data collection and processing, data are often missing, duplicated, erroneous, or inconsistent in format, affecting data availability and the accuracy of analysis results.

[0005] 3. Lack of automated data correction mechanisms: Existing systems mostly rely on manual intervention to handle data issues, which is inefficient, error-prone, and difficult to cope with the real-time processing needs of large-scale data.

[0006] 4. Difficulty in cross-system data consistency verification: The lack of unified data standards and coordination mechanisms among multiple data sources leads to data conflicts and inconsistencies when exchanging data between different systems.

[0007] Therefore, there is an urgent need for a system that can monitor data quality in real time in a multi-source data environment and improve data quality and integration efficiency through intelligent automated correction and consistency verification methods. Summary of the Invention

[0008] In view of this, the present invention provides a medical data multi-source fusion verification and correction method and system to solve the above problems.

[0009] To solve the above technical problems, the present invention provides a medical data multi-source fusion verification and correction method, comprising:

[0010] Obtain multi-source data, standardize and pre-process it, and assign trust weights to each data point based on its source;

[0011] Build dynamic baseline models for the emergency department, ICU, and general wards, establish correlations between patient data and model prediction data, and verify abnormalities in patient data through data stream responses and issue warnings.

[0012] Establish a terminology database of international standard terms and regional commonly used terms, match patient data obtained from multiple sources to generate standard terms, then perform multi-source fusion, and perform data correction based on the fusion results;

[0013] Based on the full-process data, a multi-source data quality evaluation table is generated for different departments, including completeness rate, standardization rate, conflict resolution rate and clinical recognition rate. The terminology library and fusion correction weight are updated based on the feedback from the quality evaluation table.

[0014] As an optional method, during standardization, the raw data is given a unified label encoding through a pre-loaded lightweight medical knowledge graph.

[0015] As an optional method,

[0016] Building dynamic baseline models for the emergency department, ICU, and general wards involves setting baselines for each type of patient data that match the data source, which serve as standard ranges for abnormal alerts.

[0017] Establishing the association between patient data and model prediction data includes: inputting the timestamp, department, personal information and data type of the patient data, and predicting the reasonable fluctuation range of the corresponding data through the LSTM neural network.

[0018] As an optional method, abnormalities in patient data validation through data stream responses include:

[0019] After dividing the data sources by department, the data flow platform is used to visualize the input patient data information;

[0020] Connect the data flow platform to the dynamic baselines of different departments, monitor whether there are indicators in the patient data exceeding the preset threshold, and call the medical knowledge base API to verify the data and logical consistency in real time to generate early warning strategies.

[0021] As an optional method, early warning includes:

[0022] The first warning strategy is set to provide audio and visual prompts for data anomalies that directly endanger the safety of diagnosis and treatment. The second warning strategy is set to generate to-do tasks and send them to the quality control port for data anomalies that affect data quality but are not urgent. The third warning strategy is set to mark semantic risks and generate logs.

[0023] As an optional approach, trust weights are assigned to each piece of data to add a source authority score. Patient data obtained from multiple sources are matched to generate standard terms for unified term expression, which include:

[0024] Extract text fields from patient data and use pre-trained models for matching and recognition; term matching includes exact matching and fuzzy matching; exact matching is used to directly extract text information of the matching term library from the data; fuzzy matching maps incorrect and / or abbreviated regional common terms to international standard terms; among them, if it is detected that any regional common term has accumulated a preset number of occurrences and has not been mapped, the term will be dynamically adjusted.

[0025] As an optional method, before outputting the matching recognition results, semantic conflict recognition and dynamic priority adjustment are also included; semantic conflict recognition is used to record conflicts in text data;

[0026] Dynamic priority adjustment is performed using the following method:

[0027] Set weight dimensions, including source authority, time decay, and operator qualifications;

[0028] Set conflict resolution rules, including: if the difference between the total weight score of any data source and other sources is greater than the preset value, it will be adopted directly; if the difference between the total weight score of any data source and other sources is less than the preset value, manual review will be conducted and the conflict type and cause will be recorded.

[0029] As an optional approach, the amendments include:

[0030] Fill in missing data based on similar records, convert non-compliant data into a standard format, identify patients using unique identifiers, delete duplicate records, and then set up a tiered correction strategy for low-risk, medium-risk, and high-risk conflicts;

[0031] Low-risk conflict correction is used to correct differences in terminology between different systems, but conflicts with consistent semantics are based on correction;

[0032] Medium-risk conflict correction is used to correct conflicts that are numerically contradictory but logically inferable; its correction weight is determined based on the weight dimensions of source authority, time decay, and operator qualifications;

[0033] High-risk conflict correction is used to correct contradictory data that directly affects diagnosis or treatment safety, immediately issue an early warning, and freeze the internal process of the current medical order;

[0034] After completing any level of correction, the original conflict data snapshot and the basis for the correction decision are recorded in the database.

[0035] As an optional method, the completeness rate is used to calculate the proportion of missing data, which is used to indicate the completeness of the data; the standardization rate is used to calculate the proportion of actual terms that conform to international standard terminology; the conflict resolution rate is used to indicate the proportion of automatic conflict resolution without human intervention in the total conflicts; and the clinical acceptance rate is used to sample the proportion of doctors adopting the revised results.

[0036] On the other hand, the present invention also provides a medical data multi-source fusion verification and correction system, comprising:

[0037] The data preprocessing module is used to obtain multi-source data, standardize preprocessing, and assign trust weights to each data based on different sources;

[0038] The data monitoring module is used to build dynamic baseline models for the emergency department, ICU, and general wards, establish the relationship between patient data and model prediction data, and verify abnormalities in patient data through data flow response;

[0039] The knowledge base and fusion correction module is used to establish a terminology database of international standard terms and regional common terms, match patient data obtained from multiple sources to generate standard terms, then perform multi-source fusion and perform data correction based on the fusion results;

[0040] The feedback optimization module is used to generate multi-source data quality evaluation tables for different departments based on full-process data, including completeness rate, standardization rate, conflict resolution rate and clinical approval rate, and update the terminology library and fusion correction system based on feedback from the quality evaluation table.

[0041] The beneficial effects of the present invention are:

[0042] Through real-time monitoring and intelligent early warning mechanisms, this invention can promptly detect and correct quality issues in data, reducing human error and significantly lowering data management costs. Furthermore, automated verification and correction functions improve data processing efficiency, making data management more efficient. Through multi-source data fusion and consistency verification, the system ensures seamless integration of data from different data sources, achieving cross-system and cross-platform data consistency and collaboration, thereby improving the accuracy of medical decision-making and the reliability of data. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A flow chart of a multi-source fusion verification and correction method for medical data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with specific implementation methods.

[0045] See also Figure 1This embodiment provides a method for multi-source fusion verification and correction of medical data, including: obtaining multi-source data, assigning trust weights to each data based on different sources after standardized preprocessing; constructing dynamic baseline models for the emergency department, ICU and general ward, establishing the correlation between patient data and model prediction data, verifying abnormalities in patient data through data stream response and issuing warnings; establishing a terminology library of international standard terms and regional commonly used terms, matching patient data obtained from multiple sources to generate standard terms, and then performing multi-source fusion, and correcting data based on the fusion results; generating a multi-source data quality evaluation table for different departments based on the full-process data, including completeness rate, standardization rate, conflict resolution rate and clinical approval rate, and updating the terminology library and fusion correction weight based on feedback from the quality evaluation table.

[0046] In this embodiment, data is first obtained from multiple medical data sources, such as hospital information systems, electronic medical record systems, and laboratory information management systems. Data from each data source is typically stored in different formats and structures. For example, a hospital information system may store basic patient information, from which this embodiment obtains patient registration and basic file information; an electronic medical record system stores medical records, from which this embodiment obtains diagnostic records, medical orders, and other information; and a laboratory information management system stores laboratory test data, from which this embodiment obtains test reports, imaging data, and other data. Furthermore, real-time blood pressure and blood oxygen levels can also be obtained from IoT-connected devices, such as vital signs monitors.

[0047] After collecting the data, data preprocessing is first performed to unify the data format, clean the data and remove invalid information. Afterwards, in an optional scenario, a medical knowledge graph constructed by the International Classification of Diseases standard code ICD-10 and the human and veterinary medical system terminology is preloaded to give the original data a unified label code, such as automatically mapping myocardial infarction to the code of myocardial infarction, so that the data initially maintains the same standard. Weights are then assigned based on the source. This embodiment is set from three dimensions: source authority, time decay and operator qualifications. Optionally, in source authority, the weight of tertiary hospitals is greater than that of community hospitals; in time decay, the weight of recent calibration is greater than that of non-recent calibration; in operator qualifications, the weight of doctors is greater than that of nurses. The specific settings are not limited in this embodiment.

[0048] Next, this embodiment begins to establish a dynamic quality assessment standard to solve the problem that traditional fixed thresholds cannot adapt to the variability of medical scenarios.

[0049] Baseline alarm is one of the most important functions of most database monitoring software. The implementation of dynamic baseline is to solve the problem of the combined fluctuation range of certain indicators under different business loads during the day and night. Taking into account the specificity of different departments, this embodiment constructs a dynamic baseline model based on the differences between the emergency department, ICU and general wards, establishes the correlation between patient data and model prediction data, and verifies the abnormal situation of patient data through data stream response and issues early warning. This is because, generally speaking, the reasonable range of blood pressure in the ICU is: systolic pressure 70-160mmHg, while the reasonable range of blood pressure in the general ward is: 90-140mmHg. Conflicts are prone to occur when performing data fusion.

[0050] After that, you can input historical data timestamps, departments, patient ages, test items and other data, and use the LSTM neural network to capture diurnal / seasonal patterns to output the predicted reasonable value range. This embodiment does not limit specific data types and data flow software. You can use any common data flow software on the market for monitoring, and verify numerical anomalies and logical conflicts by responding to the data flow. Optionally, the monitoring type is divided by department, and the dynamic baseline generated above is connected to monitor whether the indicator exceeds the threshold. Optionally, considering that in some cases, blood sugar data may be inconsistent with the history of diabetes due to fluctuations, the existing knowledge graph API can also be called to verify the logical consistency of the data in real time.

[0051] Among them, this embodiment sets up a conflict warning classification mechanism, including setting a first warning strategy to provide audio and visual prompts for data anomalies that directly endanger medical safety, such as blood potassium >6.5mmol / L; setting a second warning strategy to generate to-do tasks and send them to the quality control port for data anomalies that affect data quality but are not urgent, such as missing fields such as gender, medical history, and medical records; and setting a third warning strategy to mark and generate logs for terms that are not strictly unified but can be logically inferred and have semantic risks.

[0052] The fusion process of this embodiment includes:

[0053] Assign trust weights to add source authority scores to each piece of data, and perform term matching on patient data obtained from multiple sources to generate standard terms for unified term expression. It includes: extracting text fields from patient data and using a pre-trained model for both exact matching and fuzzy matching. Exact matching is used to directly extract text information that matches the term library from the data; fuzzy matching maps common terms in incorrect and / or abbreviated areas to international standard terms (e.g., "esophagus" written as "Si Dao", handling spelling mistakes or dialects such as "old chronic bronchitis" mapped to "chronic obstructive pulmonary disease J44.9", "liver cancer" mapped to "hepatic malignant tumor C22.0", "chemotherapy" mapped to "oncological chemotherapy Z51.1"). Among them, if it is detected that any common term in an area accumulates a preset number of times and is not mapped, the term is dynamically adjusted. This can overcome the problem that conventional text matching cannot solve the expression differences.

[0054] Before outputting the matching recognition results, it also includes semantic conflict recognition and dynamic priority adjustment; semantic conflict recognition is used to record the conflicts in text data; dynamic priority adjustment is adjusted using the following method:

[0055] Set weight dimensions, including source authority, time decay, and operator qualification; set conflict resolution rules, including: if the total weight score of any data source differs from that of other sources by more than a preset value, it is directly adopted; if the total weight score of any data source differs from that of other sources by less than a preset value, manual review is performed, and the conflict type and conflict reason are recorded.

[0056] After completing data fusion, this embodiment starts to correct:

[0057] First, fill in missing data based on similar records, convert data that does not conform to the format into the standard format, identify patients through unique identifiers, and delete duplicate records to ensure that the data is complete and in the same dimension. Then set a hierarchical correction strategy for low-risk conflicts, medium-risk conflicts, and high-risk conflicts:

[0058] Based on the term standardization conflict, set low-risk conflict correction to correct conflicts with different term expressions between different systems but the same semantics; [[ID=!6]]

[0059] Based on the problem that there are differences in records between different systems but the rationality can be inferred through associated data, set medium-risk conflict correction to correct conflicts with numerical contradictions but logical inferability; its correction weight is determined based on the weight dimensions of source authority, time decay, and operator qualification;

[0060] Based on the conflict of key indicators, high-risk conflict corrections are set to correct the contradictory data that directly affects the safety of diagnosis or treatment, and an early warning is issued immediately and the internal process of the current medical order is frozen. The high-risk conflict is because the patient's allergy history recorded in different systems is inconsistent with the allergen due to expression or doctor's record problems, which can easily lead to unexpected situations in the later treatment. After completing the correction at any level, the original conflict data snapshot and the basis for the correction decision are recorded in the database. Through the above-mentioned corrections, this embodiment significantly improves the clinical usability of the solution through strategy stratification and transparent traceability while reducing technical complexity.

[0061] Finally, this embodiment also sets up a multi-source data quality evaluation table for different departments, including completeness rate, standardization rate, conflict resolution rate and clinical recognition rate. The completeness rate is used to calculate the proportion of missing data, which is used to indicate the completeness of the data; the standardization rate is used to calculate the proportion of actual terms that conform to international standard terms; the conflict resolution rate is used to indicate the proportion of automatic conflict resolution without human participation in the total conflict, and the clinical recognition rate is used to sample and obtain the proportion of doctors adopting the correction results. It is convenient to generate reports for reference by the unit, and it can also provide feedback to drive the iterative update of the terminology library. For example, when a local term such as chronic bronchitis is manually revised by a doctor for ≥5 times, it is automatically upgraded to a standard term alias. In this way, through feedback and new data, the machine learning model is continuously optimized to improve the accuracy of data quality monitoring, correction and consistency verification. For example, through historical data, the outlier detection model is optimized to improve the system's ability to recognize new data.

[0062] On the other hand, this embodiment also provides a medical data multi-source fusion verification and correction system, including:

[0063] The data preprocessing module is used to obtain multi-source data, standardize preprocessing, and assign trust weights to each data based on different sources;

[0064] The data monitoring module is used to build dynamic baseline models for the emergency department, ICU, and general wards, establish the relationship between patient data and model prediction data, and verify abnormalities in patient data through data flow response;

[0065] The knowledge base and fusion correction module is used to establish a terminology database of international standard terms and regional common terms, match patient data obtained from multiple sources to generate standard terms, then perform multi-source fusion and perform data correction based on the fusion results;

[0066] The feedback optimization module is used to generate multi-source data quality evaluation tables for different departments based on full-process data, including completeness rate, standardization rate, conflict resolution rate and clinical approval rate, and update the terminology library and fusion correction system based on feedback from the quality evaluation table.

[0067] Through the above scheme, this embodiment can timely discover and correct quality problems in the data through real-time monitoring and intelligent early warning mechanisms, reduce human errors and significantly reduce data management costs. At the same time, the automated verification and correction functions improve data processing efficiency, making data management more efficient. Through multi-source data fusion and consistency verification, it can ensure that data from different data sources are seamlessly connected, achieve cross-system and cross-platform data consistency and collaboration, thereby improving the accuracy of medical decision-making and the reliability of data. It also continuously optimizes the data quality management process through machine learning algorithms, so that the system can adapt to complex medical data management needs and gradually improve its intelligence level. With the continued use of the system, it can continuously optimize and adjust the data quality monitoring and correction rules to improve the processing capabilities of new data.

[0068] The above are merely preferred embodiments of the present invention. It should be noted that the above preferred embodiments should not be construed as limiting the present invention, and the scope of protection of the present invention should be determined by the scope defined in the claims. Persons skilled in the art will appreciate that improvements and modifications may be made without departing from the spirit and scope of the present invention, and such improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for multi-source fusion verification and correction of medical data, characterized in that: include: Obtain multi-source data, standardize and pre-process it, and assign trust weights to each data point based on its source; Build dynamic baseline models for the emergency department, ICU, and general wards, inputting patient data timestamps, departments, personal information, and data types. Use the LSTM neural network to predict the reasonable fluctuation range of the corresponding data. Verify abnormalities in patient data through data stream responses. The abnormal situations of verifying patient data by data stream response include: After dividing the data sources by department, the data flow platform is used to visualize the input patient data information; Connect the data flow platform to the dynamic baselines of different departments to monitor whether any indicators in patient data exceed the preset thresholds, call the medical knowledge base API to verify data and logic consistency in real time, and generate early warning strategies; Establish a terminology database of international standard terms and regional commonly used terms, perform term matching on patient data obtained from multiple sources to generate standard terms, then perform multi-source fusion, and perform data correction based on the fusion results. Before outputting the term matching results, it also includes: Record conflicts in text data and set weight dimensions, including source authority, time decay, and operator qualifications; Set conflict resolution rules. The rules include: if the difference between the total weight score of any data source and other sources is greater than the preset value, it will be directly adopted; if the difference between the total weight score of any data source and other sources is less than the preset value, manual review will be carried out and the conflict type and cause will be recorded; Based on the full-process data, a multi-source data quality evaluation table including completeness rate, standardization rate, conflict resolution rate and clinical approval rate is generated for different departments. The terminology library and fusion correction weight are updated based on the feedback of the quality evaluation table. Among them, the completeness rate is used to calculate the proportion of missing data, which is used to indicate the completeness of the data; the standardization rate is used to calculate the proportion of actual terms that conform to international standard terminology; the conflict resolution rate is used to indicate the proportion of automatic conflict resolution without human intervention in the total conflicts; the clinical approval rate is used to sample and obtain the proportion of doctors adopting the correction results.

2. A medical data multi-source fusion verification and correction method according to claim 1, characterized in that: During the standardization process, the raw data is assigned a unified label encoding through a pre-loaded lightweight medical knowledge graph.

3. A medical data multi-source fusion verification and correction method according to claim 1, characterized in that: The construction of the dynamic baseline model for the emergency department, ICU and general ward includes setting a baseline that conforms to the data source for each type of patient data, which is used as a standard range for abnormal alarms.

4. A medical data multi-source fusion verification and correction method according to claim 1, characterized in that: The warnings include: The first warning strategy is set to provide audio and visual prompts for data anomalies that directly endanger the safety of diagnosis and treatment. The second warning strategy is set to generate to-do tasks and send them to the quality control port for data anomalies that affect data quality but are not urgent. The third warning strategy is set to mark semantic risks and generate logs.

5. A medical data multi-source fusion verification and correction method according to claim 2, characterized in that: The assigning of trust weights is used to add a source authority score to each piece of data. The term matching of patient data obtained from multiple sources is performed to generate standard terms for unified term expression, which includes: Extract text fields from patient data and use a pre-trained model for matching and identification; the term matching includes exact matching and fuzzy matching; the exact matching is used to directly extract text information that matches the term library from the data; the fuzzy matching maps erroneous and / or abbreviated regional common terms to international standard terms; wherein, if it is detected that any regional common term has accumulated a preset number of occurrences and has not been mapped, the term is dynamically adjusted.

6. A medical data multi-source fusion verification and correction method according to claim 1, characterized in that: The amendments include: Fill in missing data based on similar records, convert non-compliant data into a standard format, identify patients using unique identifiers, delete duplicate records, and then set up a tiered correction strategy for low-risk, medium-risk, and high-risk conflicts; The low-risk conflict correction is used to correct conflicts between different systems that have different terminology expressions but consistent semantics; The medium-risk conflict correction is used to correct conflicts that are numerically contradictory but logically inferable; its correction weight is determined based on the weight dimensions of source authority, time decay, and operator qualifications; The high-risk conflict correction is used to correct contradictory data that directly affects the safety of diagnosis or treatment, immediately issue an early warning, and freeze the internal process of the current medical order; After completing any level of correction, the original conflict data snapshot and the basis for the correction decision are recorded in the database.

7. A medical data multi-source fusion verification and correction system, characterized by: include: The data preprocessing module is used to obtain multi-source data, standardize preprocessing, and assign trust weights to each data based on different sources; The data monitoring module is used to build dynamic baseline models for the emergency department, ICU, and general wards. It inputs the timestamp, department, personal information, and data type of patient data, and uses an LSTM neural network to predict the reasonable fluctuation range of the corresponding data. It also verifies abnormalities in patient data through data stream responses. The abnormal situations of verifying patient data by data stream response include: After dividing the data sources by department, the data flow platform is used to visualize the input patient data information; Connect the data flow platform to the dynamic baselines of different departments to monitor whether any indicators in patient data exceed the preset thresholds, call the medical knowledge base API to verify data and logic consistency in real time, and generate early warning strategies; The knowledge base and fusion correction module is used to establish a terminology library of international standard terms and regional common terms. It matches patient data obtained from multiple sources to generate standard terms, then performs multi-source fusion and data correction based on the fusion results. Before outputting the term matching results, it also includes: Record conflicts in text data and set weight dimensions, including source authority, time decay, and operator qualifications; Set conflict resolution rules. The rules include: if the difference between the total weight score of any data source and other sources is greater than the preset value, it will be directly adopted; if the difference between the total weight score of any data source and other sources is less than the preset value, manual review will be carried out and the conflict type and cause will be recorded; The feedback optimization module is used to generate a multi-source data quality evaluation table for different departments based on the full-process data, including the completeness rate, standardization rate, conflict resolution rate and clinical recognition rate, and update the terminology library and fusion correction weight based on the feedback of the quality evaluation table; wherein, the completeness rate is used to calculate the proportion of missing data, which is used to indicate the degree of data completion; the standardization rate is used to calculate the proportion of actual terms that conform to international standard terminology; the conflict resolution rate is used to indicate the proportion of automatic conflict resolution without human participation in the total conflict; and the clinical recognition rate is used to sample and obtain the proportion of doctors adopting the correction results.

Images