Clinical test result review method and system based on artificial intelligence and big data

By building a knowledge graph and using blockchain technology, combining multimodal sensor data, high-precision clinical test results review is achieved, solving the problems of traditional audit efficiency and unstable quality, improving audit efficiency and accuracy, and reducing labor costs and error risks.

CN118629571BActive Publication Date: 2024-11-29SUZHOU KAIRUIYUAN INFORMATION TECH CO LTD

Patent Information

Application Number
CN202410676222.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-29
Publication Date
2024-11-29
Estimated Expiration
2044-05-29

AI Technical Summary

Technical Problem

Traditional clinical test results review efficiency is low, unstable, lack data mining capabilities, and it is difficult to identify abnormal results from massive data, which increases the work burden of medical staff.

Method used

Using an artificial intelligence and big data method, by obtaining the correspondence between drugs and disease indicators, building a knowledge graph, combining multimodal sensor data and blockchain technology, a spatiotemporal consistency verification mechanism is built, and a high-precision automatic audit model is generated. Conduct intelligent audits and perform semantic inferences, output intelligent audit reports, and ensure the credibility of reports through blockchain sharing and expert review.

Benefits of technology

It improves the efficiency and accuracy of inspection results review, reduces the cost and error risks of manual review, realizes highly reliable intelligent review and human-machine collaboration, and supports medical quality management and auxiliary diagnosis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118629571B_ABST
    Figure CN118629571B_ABST
Patent Text Reader

Abstract

The present invention provides a clinical test result review method and system based on artificial intelligence and big data, which relates to the field of medical information processing technology, including: constructing a first knowledge graph, acquiring physiological disease characteristics and quality control information, historical test data and medication conditions, generating heterogeneous data sets, performing intelligent fusion, extracting semantic features and expanding them, and obtaining high-quality sample data; acquiring real-time message data of test equipment, performing security authentication, comparing adjacent sample data, generating sample association verification results, constructing an initial automatic review model and outputting automatic review results, comparing the automatic review results with manual approval results, and obtaining a high-precision automatic review model; acquiring sample data and inputting it into the high-precision automatic review model, performing semantic reasoning, judging abnormal risks and outputting an intelligent review report, calculating confidence and comparing it with a confidence threshold, and outputting an intelligent review report with a confidence higher than the confidence threshold as a credible review result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of medical information processing technology, and in particular to a clinical test result review method and system based on artificial intelligence and big data. Background Art

[0002] With the rapid development of modern medicine and the continuous deepening of medical informatization, the test data generated by medical institutions every day has exploded. The massive test result data contains valuable medical information and laws. However, how to identify abnormal results from the vast amount of data, ensure the quality of the test, and reduce the review workload of medical staff has become a difficult problem that needs to be solved in medical quality management.

[0003] Traditional test result review mainly relies on manual review by medical staff. Medical staff need to review each test report one by one based on their own professional knowledge and experience, determine whether the result is abnormal, and analyze the possible causes of the abnormality. However, traditional test result review has limitations such as low review efficiency, unstable review quality, and lack of data mining.

[0004] In recent years, artificial intelligence technology represented by machine learning has made great progress and demonstrated strong performance in the fields of computer vision, natural language understanding, and medical image analysis. It has attracted widespread attention from academia and industry, and has begun to introduce machine learning methods into the field of automatic review of test results.

[0005] Therefore, a solution is urgently needed to solve the deficiencies in the prior art. Summary of the invention

[0006] The embodiments of the present invention provide a clinical test result review method and system based on artificial intelligence and big data, which can at least solve some of the problems existing in the prior art.

[0007] A first aspect of an embodiment of the present invention provides a clinical test result review method based on artificial intelligence and big data, comprising:

[0008] Obtain the correspondence between drugs and disease indicators and drug properties, construct a first knowledge graph, obtain physiological disease characteristics and quality control information collected by multimodal sensors, obtain the historical test data and medication status of the current patient in combination with the first knowledge graph link, generate a heterogeneous data set, intelligently fuse the heterogeneous data set, extract semantic features, and expand the heterogeneous data set through data enhancement technology to obtain high-quality sample data;

[0009] Acquire real-time message data of the inspection equipment, combine blockchain technology for security authentication, build a spatiotemporal consistency verification mechanism in combination with the high-quality sample data, compare adjacent sample data located in the same test tube rack, generate sample association verification results, use the high-quality sample data as input and the sample association verification results as constraints, build an initial automatic audit model and output automatic audit results, compare the automatic audit results with the manual approval results through blockchain verification, adjust the automatic audit rules based on the verification results, and obtain a high-precision automatic audit model;

[0010] The sample data of the current patient is obtained and input into the high-precision automatic audit model, the first knowledge graph is integrated and semantic reasoning is performed, the abnormal risk of the sample is comprehensively judged and an intelligent audit report is output, the intelligent audit report is securely shared through blockchain technology, the confidence of the current intelligent audit report is calculated based on multi-dimensional evidence and compared with a preset confidence threshold, and the intelligent audit report with a confidence higher than the confidence threshold is output as a credible audit result, and for the intelligent audit report that has not been output, it is reviewed through remote collaboration of experts, and the review result is output as a credible audit result.

[0011] In an optional embodiment,

[0012] Obtaining the correspondence between drugs and disease indicators and drug attributes, and constructing the first knowledge graph includes:

[0013] Collecting information corresponding to drugs and disease indicators from multiple medical databases, combining and generating original heterogeneous data, preprocessing the original heterogeneous data, standardizing the drug names and disease names through existing medical dictionaries, segmenting and tagging the text through natural language processing tools, and generating structured data through data format conversion operations and adding them to the structured database;

[0014] Random sampling is performed based on the structured data, and the sampled structured data is manually labeled through interactive feedback to determine the association relationship of the current structured data. Based on the association relationship, unlabeled data is predicted through a preset relationship extraction model, and the prediction result is compared with the true label of the unlabeled data, and the relationship extraction model is optimized based on the comparison result;

[0015] Extracting triplets corresponding to drugs, diseases, and disease indicators from the structured database based on the association relationship and the relationship extraction model; extracting and merging triplets through named entity recognition for unstructured text data; calculating the corresponding triple credibility for each triple based on the data source and the degree of matching between the entity word properties; and retaining triplets with triple credibility higher than a preset credibility threshold;

[0016] Based on the retained triples, the text labels in the retained triples are mapped to the ontology concepts through the semantic similarity calculation technology, the semantic relationship types in the retained triples are mapped to the relationship types in the ontology, the attribute information of the drugs, diseases and disease indicators are extracted from the structured data and the attributes are assigned to the ontology, and the ontology concepts are mapped to the pre-built medical knowledge base based on the semantic relationship between the concepts through the ontology reasoning mechanism and the concept instances in the medical knowledge base are linked to obtain a complete medical knowledge base;

[0017] The first knowledge graph is constructed using the medical knowledge base as ontology concept nodes and the relationship types as connecting edges.

[0018] In an optional embodiment,

[0019] Acquire the physiological symptoms and quality control information collected by the multimodal sensor, obtain the historical test data and medication status of the current patient in combination with the first knowledge graph link, generate a heterogeneous data set, perform intelligent fusion on the heterogeneous data set, extract semantic features, and expand the heterogeneous data set through data enhancement technology to obtain high-quality sample data including:

[0020] The physiological characteristic data of the patient is collected in real time through pre-deployed wearable devices and IoT sensors, the imaging characteristics of the patient are obtained through medical imaging equipment and are aligned in time and space with the physiological characteristic data to obtain physiological disease characteristics, and the keywords and semantic information related to quality control are extracted from text information such as electronic medical records and nursing records through natural language processing technology as quality control information, and the historical test data and medication status of the current patient are obtained through semantic links based on the first knowledge graph, and the quality control information, the physiological disease characteristics and the historical test data are integrated to generate a heterogeneous data set;

[0021] For the physiological feature data, the long-distance dependency of the corresponding time series data is determined by a long short-term memory network, the local features and global features corresponding to the imaging features are extracted by a convolutional neural network, and the long-distance dependency, local features and global features are mapped to a low-dimensional semantic space by a fully connected layer to obtain multimodal features;

[0022] The quality control information is encoded through a language model and a word-level distributed representation is obtained, which is aggregated in combination with an attention mechanism to obtain a comprehensive semantic embedding vector corresponding to the quality control information, the historical inspection data is modeled through a gated recurrent unit to obtain an inspection embedding vector, and the inspection embedding vector, the semantic embedding vector and the mapped multimodal features are fused through a cross-modal interaction and attention fusion mechanism to obtain an intelligent fusion result;

[0023] Based on the intelligent fusion result, synthetic sample data is generated by generating an adversarial network in combination with patient attribute conditions, and the synthetic sample data is added to the intelligent fusion result to obtain a supplemented heterogeneous data set. The supplemented heterogeneous data set is mapped to the corresponding conceptual entities in the first knowledge graph, and a patient knowledge subgraph is constructed. The neighbor information of the nodes is aggregated through a graph convolutional neural network and the embedded representation of the nodes is learned to obtain the high-quality sample data.

[0024] In an optional embodiment,

[0025] Obtain real-time message data from the inspection equipment, combine blockchain technology for security authentication, build a spatiotemporal consistency verification mechanism based on the high-quality sample data, compare adjacent sample data located in the same test tube rack, and generate sample association verification results including:

[0026] The real-time message data generated is obtained through the communication interface of the inspection equipment, and the collected real-time message data is parsed and extracted to be converted into structured message data;

[0027] Organizing the structured message data into blocks in chronological order, wherein each block contains multiple message data, a timestamp and a hash value of the previous block, verifying each block based on a consensus mechanism, associating each message data with the blockchain transaction ID of the current block, recording the data generation time and processing process, and completing security authentication;

[0028] A spatiotemporal consistency verification mechanism is constructed for the high-quality sample data. Adjacent sample data located on the same test tube rack are paired according to the sample number and timestamp of the structured message data. The difference between the test results of adjacent samples is calculated to determine whether the spatiotemporal consistency is met. If not, the current sample is considered a suspicious sample, and the verification is repeated until all samples are verified. For each sample, the corresponding associated verification result is generated based on the verification results of adjacent samples and manual review feedback.

[0029] In an optional embodiment,

[0030] Taking the high-quality sample data as input and the sample association verification results as constraints, an initial automatic audit model is constructed and the automatic audit results are output. The automatic audit results are compared with the manual approval results for blockchain verification. The automatic audit rules are adjusted based on the verification results. The high-precision automatic audit model is obtained, including:

[0031] The high-quality sample data is used as input and randomly sampled, 80% of the elements in the high-quality sample data are extracted as a training set, and the unselected 20% elements are used as a training set, automatic audit rules are predefined, and the association verification results are used as constraints to build an initial automatic audit model;

[0032] Add the elements in the training set to the automatic audit model, audit each element according to the automatic audit rules, add additional audit rules if the current element is pre-marked as a suspicious sample, generate an automatic audit result, record the automatic audit result and generate a unique identifier, perform consensus verification with the manual approval result pre-generated and stored in the blockchain network, extract the hash values ​​corresponding to the automatic audit result and the manual approval result for verification and comparison, if the hash values ​​are the same, the automatic audit result is the same as the manual approval result, otherwise they are different;

[0033] The automatic audit rules are adjusted based on the verification results, and the updates are repeated until a preset maximum number of iterations is reached to obtain the high-precision automatic audit model.

[0034] In an optional embodiment,

[0035] The sample data of the current patient is obtained and input into the high-precision automatic audit model, the first knowledge graph is integrated and semantic reasoning is performed, the abnormal risk of the sample is comprehensively judged and an intelligent audit report is output, the intelligent audit report is securely shared through blockchain technology, the confidence of the current intelligent audit report is calculated based on multi-dimensional evidence and compared with a preset confidence threshold, and the intelligent audit report with a confidence higher than the confidence threshold is output as a credible audit result. For the intelligent audit report that has not been output, it is reviewed through remote collaboration of experts, and the review result is output as a credible audit result, including:

[0036] Preprocess the original sample data of the current patient and add it to the high-precision automatic audit model, calculate the abnormal risk score of the sample through model reasoning, integrate the first knowledge graph for semantic reasoning, map the test indicators in the original sample data to the corresponding nodes of the first knowledge graph, determine the association relationship between the test indicators through semantic association, calculate the correlation between the abnormal indicators and the disease in combination with the path-based reasoning algorithm, generate the abnormal cause and add it to the intelligent audit report;

[0037] Add the smart audit report to the blockchain network and add access rights to medical institutions and medical staff, bind each smart audit report to the blockchain transaction ID and record the generation time;

[0038] For the intelligent audit report, an initial confidence threshold is set based on the accuracy of historical intelligent audit reports and expert review results, and multidimensional evidence that affects the confidence of the intelligent audit report is collected, including sample quality, patient medical history and expert feedback. The confidence of the current intelligent audit report is calculated based on the multidimensional evidence and compared with the pre-set confidence threshold. The intelligent audit report with a confidence higher than the confidence threshold is output as a credible audit result. For the intelligent audit report that has not been output, it is reviewed through remote expert collaboration, and the review result is output as a credible audit result.

[0039] In an optional embodiment,

[0040] The confidence of the current intelligent audit report is calculated based on the multi-dimensional evidence as shown in the following formula:

[0041]

[0042] Where C represents the confidence level, α represents the posterior probability control factor, P(R|H) represents the posterior probability that the intelligent audit report is credible given the multidimensional evidence set H, β represents the evidence quality control factor, m(A) represents the evidence quality, γ represents the information entropy adjustment factor, P(E i ) represents the marginal probability of the i-th dimension of evidence, and n represents the total number of evidence dimensions.

[0043] A second aspect of an embodiment of the present invention provides a clinical test result review system based on artificial intelligence and big data, comprising:

[0044] The first unit is used to obtain the corresponding relationship between drugs and disease indicators and drug properties, construct a first knowledge graph, obtain physiological disease characteristics and quality control information collected by multimodal sensors, obtain the historical test data and medication status of the current patient in combination with the first knowledge graph link, generate a heterogeneous data set, perform intelligent fusion on the heterogeneous data set, extract semantic features, and expand the heterogeneous data set through data enhancement technology to obtain high-quality sample data;

[0045] The second unit is used to obtain real-time message data of the inspection equipment, combine blockchain technology for security authentication, build a spatiotemporal consistency verification mechanism in combination with the high-quality sample data, compare the adjacent sample data located in the same test tube rack, generate sample association verification results, use the high-quality sample data as input and the sample association verification results as constraints, build an initial automatic review model and output automatic review results, compare the automatic review results with the manual approval results for blockchain verification, adjust the automatic review rules based on the verification results, and obtain a high-precision automatic review model;

[0046] The third unit is used to obtain the sample data of the current patient and input it into the high-precision automatic audit model, integrate the first knowledge graph and perform semantic reasoning, comprehensively judge the sample abnormality risk and output an intelligent audit report, securely share the intelligent audit report through blockchain technology, calculate the confidence of the current intelligent audit report based on multi-dimensional evidence and compare it with a preset confidence threshold, output the intelligent audit report with a confidence higher than the confidence threshold as a credible audit result, and for the intelligent audit report that has not been output, review it through remote collaboration of experts, and output the review result as a credible audit result.

[0047] According to a third aspect of the embodiments of the present invention,

[0048] An electronic device is provided, comprising:

[0049] processor;

[0050] a memory for storing processor-executable instructions;

[0051] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0052] A fourth aspect of the embodiments of the present invention is:

[0053] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0054] In the present invention, by acquiring the correspondence between drugs and disease indicators and drug properties, a first knowledge graph is constructed, which provides rich domain knowledge and semantic information for subsequent data fusion, feature extraction, abnormality judgment and other tasks, enhances the system's ability to understand medical concepts and relationships, improves the accuracy and interpretability of intelligent auditing, and performs intelligent fusion, semantic feature extraction and data enhancement on heterogeneous data to obtain high-quality sample data, providing high-quality and diversified data support for subsequent model training and optimization. When acquiring real-time message data of the inspection equipment, blockchain technology is combined for security authentication to ensure the authenticity and non-tamperability of the data source. By introducing spatiotemporal consistency verification, the risk of sample confusion and erroneous audit is reduced, and the accuracy of automatic audit is improved. By introducing manual feedback and verification, the automatic audit model is continuously optimized and improved, the generalization ability and robustness of the model are improved, and an intelligent audit process of human-machine collaboration and continuous learning is realized. In summary, the present invention realizes intelligent auditing, high-trust sharing and human-machine collaborative optimization of medical inspection sample data, greatly improves audit efficiency, accuracy and credibility, reduces labor costs and error risks, and provides strong technical support for medical quality management and auxiliary diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flow chart of a clinical test result review method based on artificial intelligence and big data according to an embodiment of the present invention;

[0056] Figure 2 It is a structural diagram of a clinical test result review system based on artificial intelligence and big data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0059] Figure 1 FIG. 1 is a flow chart of a clinical test result review method based on artificial intelligence and big data according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0060] S1. Obtain the correspondence between drugs and disease indicators and drug properties, construct a first knowledge graph, obtain physiological disease characteristics and quality control information collected by multimodal sensors, obtain the historical test data and medication status of the current patient in combination with the first knowledge graph link, generate a heterogeneous data set, intelligently fuse the heterogeneous data set, extract semantic features, and expand the heterogeneous data set through data enhancement technology to obtain high-quality sample data;

[0061] The disease indicators are indicators used to evaluate and monitor disease status or risks. The physiological disease characteristics are signs and symptoms related to diseases or pathological conditions, reflecting the body's physiological state and functional abnormalities. The quality control information refers to relevant information for controlling and evaluating the quality of medical data, and is an important basis for ensuring the accuracy and reliability of intelligent audit reports. The data enhancement technology refers to increasing the diversity and scale of data sets by transforming existing data or generating new data to improve the generalization ability of the model. The drug properties refer to the physicochemical properties, biological effects and pharmacokinetic characteristics of the drug, which determine the efficacy and usage of the drug.

[0062] In an optional embodiment,

[0063] Obtaining the correspondence between drugs and disease indicators and drug attributes, and constructing the first knowledge graph includes:

[0064] Collecting information corresponding to drugs and disease indicators from multiple medical databases, combining and generating original heterogeneous data, preprocessing the original heterogeneous data, standardizing the drug names and disease names through existing medical dictionaries, segmenting and tagging the text through natural language processing tools, and generating structured data through data format conversion operations and adding them to the structured database;

[0065] Random sampling is performed based on the structured data, and the sampled structured data is manually labeled through interactive feedback to determine the association relationship of the current structured data. Based on the association relationship, unlabeled data is predicted through a preset relationship extraction model, and the prediction result is compared with the true label of the unlabeled data, and the relationship extraction model is optimized based on the comparison result;

[0066] Extracting triplets corresponding to drugs, diseases, and disease indicators from the structured database based on the association relationship and the relationship extraction model; extracting and merging triplets through named entity recognition for unstructured text data; calculating the corresponding triple credibility for each triple based on the data source and the degree of matching between the entity word properties; and retaining triplets with triple credibility higher than a preset credibility threshold;

[0067] Based on the retained triples, the text labels in the retained triples are mapped to the ontology concepts through the semantic similarity calculation technology, the semantic relationship types in the retained triples are mapped to the relationship types in the ontology, the attribute information of the drugs, diseases and disease indicators are extracted from the structured data and the attributes are assigned to the ontology, and the ontology concepts are mapped to the pre-built medical knowledge base based on the semantic relationship between the concepts through the ontology reasoning mechanism and the concept instances in the medical knowledge base are linked to obtain a complete medical knowledge base;

[0068] The first knowledge graph is constructed using the medical knowledge base as ontology concept nodes and the relationship types as connecting edges.

[0069] The original heterogeneous data refers to a collection of data from different sources, different formats, different industries or different structures. The structured data refers to data with a fixed format and organization method, which is usually stored in a relational database and represented in the form of rows and columns. The triple credibility is a core unit used to describe and define entities and their relationships in a certain field in ontology. The attribute assignment refers to the assignment of specific attribute values ​​to entities in the ontology concept to describe the characteristics and status of the entity. The ontology reasoning mechanism refers to a method of automatically deriving new knowledge or verifying existing knowledge in an ontology through logical reasoning technology.

[0070] Collect information related to drugs, diseases, and disease indicators from multiple medical databases, including structured data (such as drug attributes, disease attributes, etc.) and unstructured text data (such as literature abstracts, case reports, etc.), combine them to generate original heterogeneous data sets, use existing medical dictionaries to standardize the representation of drug names and disease names, unify vocabulary, use word segmentation tools to segment text, use part-of-speech tagging tools to tag the word segmentation results, convert the preprocessed structured data and unstructured text data into a unified structured format through data format conversion operations, and store the preprocessed structured data in a structured database;

[0071] Based on the preprocessed structured data, a portion of the data is randomly sampled as a training set. Through an interactive interface, experts determine the type of association relationship of each training data according to the semantic relationship between drugs, diseases and disease indicators, store the annotation results to form a labeled training set, divide the training set data into a training set and a validation set, use the training set data to train the model, and use the validation set data to evaluate and tune the model performance, input unlabeled data into the model, predict its association relationship type, compare the prediction results with the true labels of the unlabeled data, calculate the model's accuracy, recall rate, F1 value and other evaluation indicators, optimize the relationship extraction model according to the evaluation indicators (such as adjusting the model structure, hyperparameters, etc.), and repeat the optimization until the preset maximum number of iterations is reached;

[0072] Based on the relation extraction model and structured data, the triplets corresponding to drugs, diseases and disease indicators are extracted from the structured database. The named entity recognition tool is used to identify and extract the drug, disease and indicator entities in the text. According to the co-occurrence relationship of the entities in the text, candidate triplets are generated. The triplets obtained from the structured data and unstructured text data are merged to obtain a complete set of triplets. Different weights are assigned according to the data sources of the drug, disease and indicator entities in the triplet (such as authoritative databases, core journals, etc.). The degree of word nature matching between the entity in the triplet and the text in which it is located (such as whether the drug entity is a noun) is calculated. Combining the data source weight and the degree of word nature matching, the credibility score of each triplet is calculated, and a credibility threshold is set. Triples with credibility higher than the threshold are retained, and noise triplets with lower credibility are filtered out.

[0073] Use semantic similarity calculation method to map text labels in triples to ontology concepts. Define ontology relationship types (such as "treatment", "cause", etc.) according to semantic relationship types in triples. Map triple relationships to ontology relationships. Extract attribute information of drugs, diseases and indicators (such as chemical structure of drugs, symptoms of diseases, etc.) from structured databases, and assign attributes to corresponding ontology concepts. Link ontology concepts to corresponding concept nodes in the knowledge base according to hierarchical relationships between concepts (such as "drug" is a subordinate concept of "chemical substance"). Establish corresponding relationship edges between concept nodes in the knowledge base according to ontology relationship types. Use entity linking technology (such as dictionary-based matching, context-based disambiguation) to match and link ontology concept instances with existing instances in the knowledge base. For unmatched ontology concept instances, create new instance nodes in the knowledge base. Integrate drug-disease-indicator ontology with medical knowledge base to obtain a complete medical knowledge base.

[0074] The concept nodes in the medical knowledge base are used as the entity nodes of the knowledge graph, including drugs, diseases, indicators, etc. The ontology relationship types are used as the relationship edges of the knowledge graph to connect the corresponding entity nodes. For each entity node, its attribute information is extracted from the medical knowledge base as the attribute label of the entity node to construct a medical knowledge graph with drugs, diseases and indicators as the core, that is, the first knowledge graph.

[0075] In this embodiment, information corresponding to drugs and disease indicators is collected from multiple medical databases, and the original heterogeneous data is combined to achieve the integration of data from multiple sources and multiple formats, thereby increasing the diversity and comprehensiveness of the data. The text is segmented and POS tagged by natural language processing tools, and structured data is generated by data format conversion operations and added to the structured database to ensure the consistency of drug and disease names and reduce the confusion caused by multiple representations of the same entity. Based on the association relationship and relationship extraction model, the triples corresponding to drugs, diseases and disease indicators are extracted from the structured database. Through calculation and screening, it is ensured that the retained triples have a high degree of credibility and reduce erroneous information. Through the ontology reasoning mechanism, the ontology concepts are mapped to the pre-built medical knowledge base based on the semantic relationship between concepts, and the concept instances in the knowledge base are linked, thereby achieving the semantic association between concepts, expanding the content of the knowledge base, and enhancing the network structure of the knowledge base through instance links, thereby ensuring the systematicity and interconnectedness of knowledge. In summary, the medical knowledge graph generated by this embodiment has high quality, comprehensiveness, accuracy and availability, and provides powerful data support and intelligent tools for medical research, clinical application and decision support.

[0076] In an optional embodiment,

[0077] Acquire the physiological symptoms and quality control information collected by the multimodal sensor, obtain the historical test data and medication status of the current patient in combination with the first knowledge graph link, generate a heterogeneous data set, perform intelligent fusion on the heterogeneous data set, extract semantic features, and expand the heterogeneous data set through data enhancement technology to obtain high-quality sample data including:

[0078] The physiological characteristic data of the patient is collected in real time through pre-deployed wearable devices and IoT sensors, the imaging characteristics of the patient are obtained through medical imaging equipment and are aligned in time and space with the physiological characteristic data to obtain physiological disease characteristics, and the keywords and semantic information related to quality control are extracted from text information such as electronic medical records and nursing records through natural language processing technology as quality control information, and the historical test data and medication status of the current patient are obtained through semantic links based on the first knowledge graph, and the quality control information, the physiological disease characteristics and the historical test data are integrated to generate a heterogeneous data set;

[0079] For the physiological feature data, the long-distance dependency of the corresponding time series data is determined by a long short-term memory network, the local features and global features corresponding to the imaging features are extracted by a convolutional neural network, and the long-distance dependency, local features and global features are mapped to a low-dimensional semantic space by a fully connected layer to obtain multimodal features;

[0080] The quality control information is encoded through a language model and a word-level distributed representation is obtained, which is aggregated in combination with an attention mechanism to obtain a comprehensive semantic embedding vector corresponding to the quality control information, the historical inspection data is modeled through a gated recurrent unit to obtain an inspection embedding vector, and the inspection embedding vector, the semantic embedding vector and the mapped multimodal features are fused through a cross-modal interaction and attention fusion mechanism to obtain an intelligent fusion result;

[0081] Based on the intelligent fusion result, synthetic sample data is generated by generating an adversarial network in combination with patient attribute conditions, and the synthetic sample data is added to the intelligent fusion result to obtain a supplemented heterogeneous data set. The supplemented heterogeneous data set is mapped to the corresponding conceptual entities in the first knowledge graph, and a patient knowledge subgraph is constructed. The neighbor information of the nodes is aggregated through a graph convolutional neural network and the embedded representation of the nodes is learned to obtain the high-quality sample data.

[0082] The imaging features are information extracted from medical images for auxiliary diagnosis and research. The physiological feature data refers to various types of data reflecting the physiological state of an individual. The long-distance dependency refers to the dependency between long-distance elements in sequence data. The multimodal features refer to features from different modalities (such as text, image, audio, video, etc.). The word-level distributed representation is a technology for embedding words into a continuous vector space, and each word is represented by a multidimensional vector. The comprehensive semantic embedding vector integrates semantic information from different sources or different modalities into a unified embedding vector. The cross-modal interaction refers to the interaction and information exchange between data of different modalities. The adversarial network is a model that generates data by competing with two neural networks. The patient knowledge subgraph is a subgraph in the knowledge graph specifically for individual patient information, including information such as the patient's medical history, diagnosis, treatment, drugs, test results, etc. The neighbor information in the knowledge graph refers to other nodes directly connected to a node and their relationships.

[0083] The patient's physiological characteristic data (such as heart rate, blood pressure, body temperature, respiratory rate, and sleep status) are collected in real time through pre-deployed wearable devices (such as smart bracelets and smart clothes) and IoT sensors (such as mattress pressure sensors and indoor environment sensors). The patient's imaging features (such as organ morphology, lesion location, and image density) are obtained through medical imaging equipment (such as CT, MRI, and X-ray). The collected physiological characteristic data and imaging features are aligned according to timestamps and spatial positions to form physiological disease characteristic data that is consistent in time and space.

[0084] Obtain the patient's medical record text and nursing record text from the electronic medical record system and nursing record system, use natural language processing technology (such as word segmentation, part-of-speech tagging, named entity recognition, syntactic analysis) to process the text, and extract information related to quality control (such as adverse events, complications, adverse drug reactions) through keyword matching and semantic analysis to form quality control information data;

[0085] Based on the first knowledge graph (medical knowledge graph), the current patient is matched with the patient entity in the knowledge graph through semantic linking technology (such as entity linking, ontology mapping, etc.), and the patient's historical test data (such as blood routine, biochemical full items, immune markers) and medication status (such as drug name, dosage, frequency, and course of treatment) are obtained through the relationship edges in the knowledge graph. The obtained quality control information, physiological disease characteristics and historical test data are integrated to generate a heterogeneous data set containing multiple data types and modalities, and the necessary data cleaning, standardization and normalization processing are performed on the heterogeneous data set. , ensure data quality and consistency, use long short-term memory network to model time series data, capture long-distance dependencies in time series through the gating mechanism and memory units of long short-term memory network, use convolutional neural network (CNN) to extract features from medical images, extract local features (such as edges, textures) in images through convolutional layers and pooling layers, obtain global feature representation of images through global average pooling or fully connected layers, map time series representation and image representation to the same low-dimensional semantic space through fully connected layers, and obtain multimodal representation of physiological features and image features;

[0086] Use the pre-trained medical language model to encode each word in the quality control information to obtain a distributed representation at the word level. Use the attention mechanism to perform weighted aggregation on the word vectors in the quality control information to obtain a comprehensive semantic embedding vector for the entire quality control information. Use the gated recurrent unit to model the patient's historical test data sequence, use the hidden state as the embedding representation of the historical test data, and capture the temporal variation of the test value.

[0087] The semantic embedding vector of quality control information, the embedding vector of historical inspection data and multimodal features are spliced ​​to form a comprehensive feature vector. A cross-modal interaction mechanism is designed to strengthen the information interaction between different modal features. Through the attention fusion mechanism (such as self-attention and multi-head attention), weights are adaptively assigned to different modal features to obtain the final intelligent fusion result. A generative adversarial network is used to take the intelligent fusion result as input and generate synthetic sample data in combination with patient attribute conditions. The generated synthetic sample data is added to the intelligent fusion result to obtain a heterogeneous data set with increased data volume.

[0088] The patient information, physiological characteristics, imaging characteristics, test results, etc. in the supplemented heterogeneous data set are mapped and linked to the corresponding conceptual entities in the first knowledge graph. According to the mapping results, the patient entity is taken as the central node to extract the entities such as diseases, symptoms, examinations, treatments and their relationships related to it, and a knowledge subgraph with the patient as the core is constructed. The graph convolutional neural network is used to perform representation learning on the patient knowledge subgraph, aggregate the node's neighbor information, update the node's feature representation, capture the high-order relationships and long-range dependencies in the knowledge graph, learn the low-dimensional embedding representation of the node, and fuse the learned node embedding representation with the original heterogeneous data set to obtain high-quality sample data with richer information and more comprehensive information.

[0089] In this embodiment, key information related to quality control is automatically extracted from unstructured medical texts, reducing the workload of manual review and improving the efficiency and accuracy of quality control. The medical knowledge graph is used to quickly retrieve and associate patients' historical medical data, providing comprehensive information support for intelligent diagnosis and quality control and improving the accuracy of clinical decision-making. Through the language model and attention mechanism, the quality control information is converted into a semantically rich vector representation, effectively capturing the key information and semantic relationships therein, providing strong support for intelligent quality control. Through cross-modal interaction and attention fusion, the deep fusion of quality control information, historical test data and physiological imaging features is achieved, and a comprehensive and accurate representation of the patient's status is obtained, which greatly improves the performance of intelligent diagnosis and quality control. In summary, this embodiment realizes the intelligent collection, fusion, representation and enhancement of medical data, generates high-quality high-quality sample data, helps to improve the accuracy of intelligent diagnosis and treatment, supports doctors' clinical decision-making, and improves the quality of medical services.

[0090] S2. Acquire real-time message data of the inspection equipment, combine blockchain technology for security authentication, build a spatiotemporal consistency verification mechanism in combination with the high-quality sample data, compare the adjacent sample data located in the same test tube rack, generate sample association verification results, use the high-quality sample data as input and the sample association verification results as constraints, build an initial automatic audit model and output the automatic audit results, compare the automatic audit results with the manual approval results through blockchain verification, adjust the automatic audit rules based on the verification results, and obtain a high-precision automatic audit model;

[0091] The blockchain is a distributed ledger technology with the following characteristics: decentralization, immutability, traceability, security and transparency. Through cryptographic principles and consensus mechanisms, the blockchain forms a chained data block structure in a distributed network. Each data block contains transaction data and status information within a certain period of time. The real-time message data refers to structured data generated and transmitted in real time during the medical process. The sample association verification refers to the process of associating and verifying medical data from different sources and in different formats, with the aim of ensuring the consistency and accuracy of the data. The automatic audit rules refer to a mechanism for automatically auditing and verifying medical data using pre-set rule conditions.

[0092] In an optional embodiment,

[0093] Obtain real-time message data from the inspection equipment, combine blockchain technology for security authentication, build a spatiotemporal consistency verification mechanism based on the high-quality sample data, compare adjacent sample data located in the same test tube rack, and generate sample association verification results including:

[0094] The real-time message data generated is obtained through the communication interface of the inspection equipment, and the collected real-time message data is parsed and extracted to be converted into structured message data;

[0095] Organizing the structured message data into blocks in chronological order, wherein each block contains multiple message data, a timestamp and a hash value of the previous block, verifying each block based on a consensus mechanism, associating each message data with the blockchain transaction ID of the current block, recording the data generation time and processing process, and completing security authentication;

[0096] A spatiotemporal consistency verification mechanism is constructed for the high-quality sample data. Adjacent sample data located on the same test tube rack are paired according to the sample number and timestamp of the structured message data. The difference between the test results of adjacent samples is calculated to determine whether the spatiotemporal consistency is met. If not, the current sample is considered a suspicious sample, and the verification is repeated until all samples are verified. For each sample, the corresponding associated verification result is generated based on the verification results of adjacent samples and manual review feedback.

[0097] The hash value refers to the hash value calculated by a hash function, usually represented by a hexadecimal string. The consensus mechanism refers to a mechanism in a distributed system whereby participating nodes reach a consensus on a proposal or state through a specific algorithm. The spatiotemporal consistency verification refers to a mechanism in a distributed system that verifies whether the time and space states of different nodes are consistent.

[0098] Through the communication interface of the inspection equipment, the message data generated by the inspection equipment is obtained in real time, and the collected real-time message data is parsed and extracted. According to the message format, the message content is converted into a structured data format, and key fields in the message, such as sample number, inspection item, inspection result, instrument number, operator, etc., are extracted to form structured message data;

[0099] The structured message data is organized into blocks in chronological order. Each block contains multiple message data within a certain time range (such as 10 minutes). A unique block ID is generated for each block, the generation timestamp of the block is recorded, and the hash value of the previous block is calculated to form a block header. The block header and message data are packaged together to form a complete block structure. Each block is verified based on the consensus mechanism to ensure the legitimacy and consistency of the block. Each message data is associated with the block ID of the current block, and the generation time and processing process of the data are recorded to complete the security authentication and non-tamperability guarantee of the data.

[0100] For the generated high-quality sample data, a spatiotemporal consistency verification mechanism is constructed. According to the sample number and timestamp in the structured message data, the adjacent sample data located on the same test tube rack are paired, and the difference between the adjacent sample test results is calculated to determine whether it meets the preset spatiotemporal consistency threshold. If the difference between adjacent samples exceeds the threshold, the current sample is considered a suspicious sample and needs to be verified repeatedly. The verification is repeated until all samples have completed the spatiotemporal consistency verification. For each sample, the corresponding associated verification result is generated based on the verification results of adjacent samples and manual review feedback. The associated verification result includes information such as whether the sample has passed the spatiotemporal consistency verification, whether repeated testing is required, and the manual review results. The associated verification result is associated with the corresponding sample data to form a complete sample verification record;

[0101] High-quality sample data and associated verification results that have been verified by blockchain and time-space consistency are stored in a secure and reliable database, and necessary encryption and access control are performed on the stored data to ensure data confidentiality and privacy protection.

[0102] In this embodiment, by organizing the message data into a blockchain structure and utilizing the consensus mechanism and hash link of the blockchain, the data is securely stored, tamper-proof and traceable, the integrity and consistency of the data are ensured, the data is prevented from being maliciously modified or deleted, and the credibility of the data is improved. By constructing a spatiotemporal consistency verification mechanism, adjacent sample data are paired and the difference is calculated, and automatic verification of sample quality is achieved. Suspicious samples can be discovered in a timely manner, the workload of manual verification is reduced, the efficiency and accuracy of quality control are improved, and the reliability of sample data is ensured. By generating associated verification results, sample quality verification is combined with manual review and feedback to form a complete quality control closed loop, which provides an objective basis for sample quality and can serve as an important reference for quality traceability and improvement, promoting the normalization and standardization of the inspection process. In summary, this embodiment improves the efficiency, accuracy and intelligence level of inspection data management, and provides strong support for the continuous improvement of inspection quality and the optimization of medical decision-making.

[0103] In an optional embodiment,

[0104] Taking the high-quality sample data as input and the sample association verification results as constraints, an initial automatic audit model is constructed and the automatic audit results are output. The automatic audit results are compared with the manual approval results for blockchain verification. The automatic audit rules are adjusted based on the verification results. The high-precision automatic audit model is obtained, including:

[0105] The high-quality sample data is used as input and randomly sampled, 80% of the elements in the high-quality sample data are extracted as a training set, and the unselected 20% elements are used as a training set, automatic audit rules are predefined, and the association verification results are used as constraints to build an initial automatic audit model;

[0106] Add the elements in the training set to the automatic audit model, audit each element according to the automatic audit rules, add additional audit rules if the current element is pre-marked as a suspicious sample, generate an automatic audit result, record the automatic audit result and generate a unique identifier, perform consensus verification with the manual approval result pre-generated and stored in the blockchain network, extract the hash values ​​corresponding to the automatic audit result and the manual approval result for verification and comparison, if the hash values ​​are the same, the automatic audit result is the same as the manual approval result, otherwise they are different;

[0107] The automatic audit rules are adjusted based on the verification results, and the updates are repeated until a preset maximum number of iterations is reached to obtain the high-precision automatic audit model.

[0108] The automatic audit model refers to a model that uses artificial intelligence technology, especially machine learning and deep learning algorithms, to automatically audit and verify data. The unique identifier refers to an identifier used to uniquely identify and mark an object or entity in a specific context or system. The consensus verification refers to the process in which multiple nodes in a distributed system reach a consensus on a proposal or state through consultation and voting.

[0109] The high-quality sample data that has passed blockchain verification and spatiotemporal consistency verification is used as the input data set of the automatic audit model. The high-quality sample data is randomly sampled, and 80% of the elements are extracted as the training set, and the remaining 20% ​​of the elements are used as the test set to ensure that the distribution of the training set and the test set is balanced, covering sample data of different types and quality levels;

[0110] Predefine a set of basic automatic audit rules, such as sample information integrity, result range rationality, report format standardization, etc., use the associated verification results of sample data as constraints to guide the setting and optimization of automatic audit rules, and build an initial automatic audit model based on the predefined audit rules and constraints;

[0111] Add the elements in the training set to the automatic review model one by one, conduct automatic review on each element, judge and classify the current element according to the predefined review rules, generate automatic review results (such as pass, warning, rejection, etc.), and add additional review rules for elements pre-marked as suspicious samples for key review, such as data consistency verification, clinical relevance verification, etc., record the automatic review results of each element, and generate a unique identifier (such as a hash value) to associate with the element, conduct consensus verification on the automatic review results and the manual approval results pre-generated and stored in the blockchain network, extract the hash values ​​corresponding to the automatic review results and the manual approval results respectively, and verify the consistency of the two results by comparing the hash values. If the hash values ​​are the same, the automatic review results are considered to be consistent with the manual approval results; otherwise, it is considered that there are differences that need further analysis, and the results of the consensus verification are recorded as the basis for evaluating the performance of the automatic review model;

[0112] Based on the results of consensus verification, the automatic audit rules are adjusted and optimized. For samples with inconsistent verification, their characteristics and causes are analyzed, and the audit rules are modified or added in a targeted manner. The optimized audit rules are applied to the automatic audit model, and the training set is re-audited and verified. The automatic audit model is iteratively optimized until the preset maximum number of iterations or performance indicators are reached to obtain the high-precision automatic audit model.

[0113] In this embodiment, by randomly sampling high-quality sample data and dividing the data set into a training set and a test set, a high-quality data foundation is provided for the construction of an automatic audit model. The introduction of constraints ensures the consistency of the model output and the associated verification results, improves the credibility of the audit results, and adds additional audit rules to suspicious samples to improve the model's ability to identify and handle abnormal situations. The results of consensus verification provide an objective evaluation basis for model optimization and ensure the reliability and credibility of the model output. In summary, this embodiment effectively avoids the subjectivity and instability of manual audits and lays a solid foundation for the trusted sharing and application of medical data.

[0114] S3. Obtain the sample data of the current patient and input it into the high-precision automatic audit model, integrate the first knowledge graph and perform semantic reasoning, comprehensively judge the risk of sample abnormality and output an intelligent audit report, securely share the intelligent audit report through blockchain technology, calculate the confidence of the current intelligent audit report based on multi-dimensional evidence and compare it with a preset confidence threshold, output the intelligent audit report with a confidence higher than the confidence threshold as a credible audit result, and for the intelligent audit report that has not been output, conduct a review through remote collaboration of experts, and output the review result as a credible audit result.

[0115] The semantic reasoning is a technology in artificial intelligence, which aims to obtain high-level information and insights implicit in the data through semantic understanding and knowledge reasoning of the data. The intelligent audit report is an output form of the automatic audit model, which aims to present the audit results and quality analysis of the sample data in a structured and visual manner. The secure sharing refers to ensuring the confidentiality, integrity and availability of data when sharing and exchanging data between different entities, and preventing the data from being illegally accessed, tampered with or leaked. The multi-dimensional evidence refers to the comprehensive consideration of data and evidence in multiple dimensions during the automatic audit and quality management process to conduct a comprehensive and objective assessment of sample quality.

[0116] In an optional embodiment,

[0117] The sample data of the current patient is obtained and input into the high-precision automatic audit model, the first knowledge graph is integrated and semantic reasoning is performed, the abnormal risk of the sample is comprehensively judged and an intelligent audit report is output, the intelligent audit report is securely shared through blockchain technology, the confidence of the current intelligent audit report is calculated based on multi-dimensional evidence and compared with a preset confidence threshold, and the intelligent audit report with a confidence higher than the confidence threshold is output as a credible audit result. For the intelligent audit report that has not been output, it is reviewed through remote collaboration of experts, and the review result is output as a credible audit result, including:

[0118] Preprocess the original sample data of the current patient and add it to the high-precision automatic audit model, calculate the abnormal risk score of the sample through model reasoning, integrate the first knowledge graph for semantic reasoning, map the test indicators in the original sample data to the corresponding nodes of the first knowledge graph, determine the association relationship between the test indicators through semantic association, calculate the correlation between the abnormal indicators and the disease in combination with the path-based reasoning algorithm, generate the abnormal cause and add it to the intelligent audit report;

[0119] Add the smart audit report to the blockchain network and add access rights to medical institutions and medical staff, bind each smart audit report to the blockchain transaction ID and record the generation time;

[0120] For the intelligent audit report, an initial confidence threshold is set based on the accuracy of historical intelligent audit reports and expert review results, and multidimensional evidence that affects the confidence of the intelligent audit report is collected, including sample quality, patient medical history and expert feedback. The confidence of the current intelligent audit report is calculated based on the multidimensional evidence and compared with the pre-set confidence threshold. The intelligent audit report with a confidence higher than the confidence threshold is output as a credible audit result. For the intelligent audit report that has not been output, it is reviewed through remote expert collaboration, and the review result is output as a credible audit result.

[0121] The anomaly risk score is a quantitative indicator used to measure the risk level of anomalies or quality problems in sample data. The association relationship refers to the correlation and dependency between different entities or attributes in the sample data, reflecting the inherent connection and influence mechanism between the data. The path-based reasoning algorithm is a method of using association relationships for knowledge reasoning and decision support. The anomaly cause refers to the root cause and influencing factors that lead to anomalies or quality problems in sample data.

[0122] Obtain the original sample data of the current patient, including sample number, collection time, test items, test results and other information, clean and convert the original sample data, remove invalid data, unify data format and unit, extract and select features of sample data, construct sample feature vectors, and add the pre-processed sample data to the high-precision automatic review model;

[0123] The high-precision automatic audit model evaluates and scores the abnormal risk of the sample through a trained classifier or regressor, and outputs the abnormal risk score of the sample, reflecting the possibility and severity of the sample abnormality;

[0124] Mapping the test indicators in the original sample data to the corresponding nodes of the first knowledge graph, using the ontological relationship and semantic rules in the first knowledge graph, performing semantic reasoning on the test indicators, determining the association relationship between the test indicators, such as causal relationship, correlation, etc., through semantic association, marking the test indicators with higher abnormal risk scores as abnormal indicators, searching the association path from the abnormal indicators to the disease nodes in the first knowledge graph, calculating the correlation between the abnormal indicators and the disease using a path-based reasoning algorithm, such as the shortest path algorithm, the random walk algorithm, etc., and inferring the potential causal relationship between the abnormal indicators and the disease based on the correlation strength and path characteristics;

[0125] Based on the abnormal risk score, semantic reasoning results and correlation analysis results, the possible reasons for the sample abnormality are inferred, and an intelligent audit report is generated based on the abnormal reasons. The report content includes sample information, abnormal indicators, abnormal risk score, abnormal reasons, etc. The generated intelligent audit report is added to the intelligent audit report library;

[0126] Upload the generated smart audit report to the blockchain network, add access rights to the smart audit report for relevant medical institutions and medical staff, bind each smart audit report to a unique blockchain transaction ID, and record the generation time of the smart audit report;

[0127] Based on the accuracy of historical intelligent audit reports and expert review results, an initial confidence threshold is set, and multi-dimensional evidence that affects the confidence of intelligent audit reports is collected, including sample quality, patient medical history, and expert feedback. The confidence of the current intelligent audit report is calculated based on the collected multi-dimensional evidence, and the calculated confidence is compared with the preset confidence threshold. For intelligent audit reports with confidence levels higher than the threshold, they are directly output as credible audit results. For intelligent audit reports whose confidence levels do not reach the threshold, experts conduct remote collaborative review. Experts access the intelligent audit reports through the remote system, and review and revise the report content based on their own experience and judgment. The intelligent audit reports reviewed by experts are output as credible audit results and updated to the blockchain network.

[0128] In this embodiment, by preprocessing and feature extraction of the original sample data, and using a high-precision automatic audit model to perform abnormal risk assessment and scoring, the efficiency and accuracy of sample auditing are greatly improved, the workload and error rate of manual auditing are reduced, and semantic reasoning is performed by integrating the medical knowledge graph, using ontological relationships and semantic rules to reveal the intrinsic correlation between test indicators in the sample data, discover potential abnormal patterns and laws, and provide rich background knowledge and reasoning paths for abnormal cause analysis. A confidence assessment mechanism is introduced to comprehensively consider multi-dimensional evidence such as sample quality, patient medical history, and expert feedback, and dynamically calculate the confidence of the intelligent audit report. By comparing with the confidence threshold, the credibility of the report is automatically judged, and suspicious reports are reviewed by experts to ensure high credibility of the output results. In summary, this embodiment realizes intelligent, highly reliable, and full-process closed-loop management of patient sample data auditing, which greatly improves the efficiency and level of medical quality management.

[0129] In an optional embodiment,

[0130] The confidence of the current intelligent audit report is calculated based on the multi-dimensional evidence as shown in the following formula:

[0131]

[0132] Where C represents the confidence level, α represents the posterior probability control factor, P(R|H) represents the posterior probability that the intelligent audit report is credible given the multidimensional evidence set H, β represents the evidence quality control factor, m(A) represents the evidence quality, γ represents the information entropy adjustment factor, P(E i ) represents the marginal probability of the i-th dimension of evidence, and n represents the total number of evidence dimensions.

[0133] In this embodiment, evidence from multiple dimensions such as sample quality, patient medical history, and expert feedback is comprehensively considered, and the posterior probability is calculated through Bayesian reasoning, which reflects the comprehensive impact of different evidence on the credibility of the report. By introducing the evidence quality control factor, the contribution of evidence quality to the confidence calculation can be adjusted, and high-quality evidence can be given a greater weight to improve the accuracy of confidence estimation. By introducing the information entropy adjustment factor, the impact of the diversity and consistency of evidence on the confidence can be balanced. In summary, this embodiment realizes the quantitative evaluation of the confidence of intelligent audit reports by integrating multi-dimensional evidence, considering evidence quality, introducing information entropy, etc. It has the advantages of high accuracy, strong adaptability, efficient calculation, and good interpretability. It provides an important basis for the credibility judgment and expert review of intelligent audit reports, and improves the automation and credibility of the intelligent audit process.

[0134] Figure 2 FIG. 1 is a schematic diagram of the structure of a clinical test result review system based on artificial intelligence and big data according to an embodiment of the present invention. Figure 2As shown, the system comprises:

[0135] The first unit is used to obtain the corresponding relationship between drugs and disease indicators and drug properties, construct a first knowledge graph, obtain physiological disease characteristics and quality control information collected by multimodal sensors, obtain the historical test data and medication status of the current patient in combination with the first knowledge graph link, generate a heterogeneous data set, perform intelligent fusion on the heterogeneous data set, extract semantic features, and expand the heterogeneous data set through data enhancement technology to obtain high-quality sample data;

[0136] The second unit is used to obtain real-time message data of the inspection equipment, combine blockchain technology for security authentication, build a spatiotemporal consistency verification mechanism in combination with the high-quality sample data, compare the adjacent sample data located in the same test tube rack, generate sample association verification results, use the high-quality sample data as input and the sample association verification results as constraints, build an initial automatic review model and output automatic review results, compare the automatic review results with the manual approval results for blockchain verification, adjust the automatic review rules based on the verification results, and obtain a high-precision automatic review model;

[0137] The third unit is used to obtain the sample data of the current patient and input it into the high-precision automatic audit model, integrate the first knowledge graph and perform semantic reasoning, comprehensively judge the sample abnormality risk and output an intelligent audit report, securely share the intelligent audit report through blockchain technology, calculate the confidence of the current intelligent audit report based on multi-dimensional evidence and compare it with a preset confidence threshold, output the intelligent audit report with a confidence higher than the confidence threshold as a credible audit result, and for the intelligent audit report that has not been output, review it through remote collaboration of experts, and output the review result as a credible audit result.

[0138] According to a third aspect of the embodiments of the present invention,

[0139] An electronic device is provided, comprising:

[0140] processor;

[0141] a memory for storing processor-executable instructions;

[0142] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0143] A fourth aspect of the embodiments of the present invention is:

[0144] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0145] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0146] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A clinical test result review method based on artificial intelligence and big data, characterized in that: include: Obtain the correspondence between drugs and disease indicators and drug properties, construct a first knowledge graph, obtain physiological disease characteristics and quality control information collected by multimodal sensors, obtain the historical test data and medication status of the current patient in combination with the first knowledge graph link, generate a heterogeneous data set, intelligently fuse the heterogeneous data set, extract semantic features, and expand the heterogeneous data set through data enhancement technology to obtain high-quality sample data; Acquire real-time message data of the inspection equipment, combine blockchain technology for security authentication, build a spatiotemporal consistency verification mechanism in combination with the high-quality sample data, compare adjacent sample data located in the same test tube rack, generate sample association verification results, use the high-quality sample data as input and the sample association verification results as constraints, build an initial automatic audit model and output automatic audit results, compare the automatic audit results with the manual approval results through blockchain verification, adjust the automatic audit rules based on the verification results, and obtain a high-precision automatic audit model; Obtain the sample data of the current patient and input it into the high-precision automatic audit model, integrate the first knowledge graph and perform semantic reasoning, comprehensively judge the abnormal risk of the sample and output an intelligent audit report, securely share the intelligent audit report through blockchain technology, calculate the confidence of the current intelligent audit report based on multi-dimensional evidence and compare it with a preset confidence threshold, output the intelligent audit report with a confidence higher than the confidence threshold as a credible audit result, and for the intelligent audit report that has not been output, review it through remote collaboration of experts, and output the review result as a credible audit result; Obtaining the correspondence between drugs and disease indicators and drug attributes, and constructing the first knowledge graph includes: Collecting information corresponding to drugs and disease indicators from multiple medical databases, combining and generating original heterogeneous data, preprocessing the original heterogeneous data, standardizing the drug names and disease names through existing medical dictionaries, segmenting and tagging the text through natural language processing tools, and generating structured data through data format conversion operations and adding them to the structured database; Random sampling is performed based on the structured data, and the sampled structured data is manually labeled through interactive feedback to determine the association relationship of the current structured data. Based on the association relationship, unlabeled data is predicted through a preset relationship extraction model, and the prediction result is compared with the true label of the unlabeled data, and the relationship extraction model is optimized based on the comparison result; Extracting triplets corresponding to drugs, diseases, and disease indicators from the structured database based on the association relationship and the relationship extraction model; extracting and merging triplets through named entity recognition for unstructured text data; calculating the corresponding triple credibility for each triple based on the data source and the degree of matching between the entity word properties; and retaining triplets with triple credibility higher than a preset credibility threshold; Based on the retained triples, the text labels in the retained triples are mapped to the ontology concepts through the semantic similarity calculation technology, the semantic relationship types in the retained triples are mapped to the relationship types in the ontology, the attribute information of the drugs, diseases and disease indicators are extracted from the structured data and the attributes are assigned to the ontology, and the ontology concepts are mapped to the pre-built medical knowledge base based on the semantic relationship between the concepts through the ontology reasoning mechanism and the concept instances in the medical knowledge base are linked to obtain a complete medical knowledge base; Using the medical knowledge base as ontology concept nodes and the relationship types as connecting edges to construct the first knowledge graph; Acquire the physiological symptoms and quality control information collected by the multimodal sensor, obtain the historical test data and medication status of the current patient in combination with the first knowledge graph link, generate a heterogeneous data set, perform intelligent fusion on the heterogeneous data set, extract semantic features, and expand the heterogeneous data set through data enhancement technology to obtain high-quality sample data including: The physiological characteristic data of the patient is collected in real time through pre-deployed wearable devices and IoT sensors, the imaging characteristics of the patient are obtained through medical imaging equipment and are aligned in time and space with the physiological characteristic data to obtain physiological disease characteristics, and the keywords and semantic information related to quality control are extracted from text information such as electronic medical records and nursing records through natural language processing technology as quality control information, and the historical test data and medication status of the current patient are obtained through semantic links based on the first knowledge graph, and the quality control information, the physiological disease characteristics and the historical test data are integrated to generate a heterogeneous data set; For the physiological feature data, the long-distance dependency of the corresponding time series data is determined by a long short-term memory network, the local features and global features corresponding to the imaging features are extracted by a convolutional neural network, and the long-distance dependency, local features and global features are mapped to a low-dimensional semantic space by a fully connected layer to obtain multimodal features; The quality control information is encoded through a language model and a word-level distributed representation is obtained, which is aggregated in combination with an attention mechanism to obtain a comprehensive semantic embedding vector corresponding to the quality control information, the historical inspection data is modeled through a gated recurrent unit to obtain an inspection embedding vector, and the inspection embedding vector, the semantic embedding vector and the mapped multimodal features are fused through a cross-modal interaction and attention fusion mechanism to obtain an intelligent fusion result; Based on the intelligent fusion result, synthetic sample data is generated by generating an adversarial network in combination with patient attribute conditions, the synthetic sample data is added to the intelligent fusion result to obtain a supplemented heterogeneous data set, the supplemented heterogeneous data set is mapped to the corresponding concept entity in the first knowledge graph, a patient knowledge subgraph is constructed, and the neighbor information of the node is aggregated by a graph convolutional neural network and the embedded representation of the node is learned to obtain the high-quality sample data; Obtain real-time message data from the inspection equipment, combine blockchain technology for security authentication, build a spatiotemporal consistency verification mechanism based on the high-quality sample data, compare adjacent sample data located in the same test tube rack, and generate sample association verification results including: The real-time message data generated is obtained through the communication interface of the inspection equipment, and the collected real-time message data is parsed and extracted to be converted into structured message data; Organizing the structured message data into blocks in chronological order, wherein each block contains multiple message data, a timestamp and a hash value of the previous block, verifying each block based on a consensus mechanism, associating each message data with the blockchain transaction ID of the current block, recording the data generation time and processing process, and completing security authentication; A spatiotemporal consistency verification mechanism is constructed for the high-quality sample data. Adjacent sample data located on the same test tube rack are paired according to the sample number and timestamp of the structured message data. The difference between the test results of adjacent samples is calculated to determine whether the spatiotemporal consistency is met. If not, the current sample is considered a suspicious sample, and the verification is repeated until all samples are verified. For each sample, the corresponding associated verification result is generated based on the verification results of adjacent samples and manual review feedback.

2. The method according to claim 1, characterized in that: Taking the high-quality sample data as input and the sample association verification results as constraints, an initial automatic audit model is constructed and the automatic audit results are output. The automatic audit results are compared with the manual approval results for blockchain verification. The automatic audit rules are adjusted based on the verification results. The high-precision automatic audit model is obtained, including: The high-quality sample data is used as input and randomly sampled, 80% of the elements in the high-quality sample data are extracted as a training set, and the unselected 20% elements are used as a training set, automatic audit rules are predefined, and the association verification results are used as constraints to build an initial automatic audit model; Add the elements in the training set to the automatic audit model, audit each element according to the automatic audit rules, add additional audit rules if the current element is pre-marked as a suspicious sample, generate an automatic audit result, record the automatic audit result and generate a unique identifier, perform consensus verification with the manual approval result pre-generated and stored in the blockchain network, extract the hash values ​​corresponding to the automatic audit result and the manual approval result for verification and comparison, if the hash values ​​are the same, the automatic audit result is the same as the manual approval result, otherwise they are different; The automatic audit rules are adjusted based on the verification results, and the updates are repeated until a preset maximum number of iterations is reached to obtain the high-precision automatic audit model.

3. The method according to claim 1, characterized in that: The sample data of the current patient is obtained and input into the high-precision automatic audit model, the first knowledge graph is integrated and semantic reasoning is performed, the abnormal risk of the sample is comprehensively judged and an intelligent audit report is output, the intelligent audit report is securely shared through blockchain technology, the confidence of the current intelligent audit report is calculated based on multi-dimensional evidence and compared with a preset confidence threshold, and the intelligent audit report with a confidence higher than the confidence threshold is output as a credible audit result. For the intelligent audit report that has not been output, it is reviewed through remote collaboration of experts, and the review result is output as a credible audit result, including: Preprocess the original sample data of the current patient and add it to the high-precision automatic audit model, calculate the abnormal risk score of the sample through model reasoning, integrate the first knowledge graph for semantic reasoning, map the test indicators in the original sample data to the corresponding nodes of the first knowledge graph, determine the association relationship between the test indicators through semantic association, calculate the correlation between the abnormal indicators and the disease in combination with the path-based reasoning algorithm, generate the abnormal cause and add it to the intelligent audit report; Add the smart audit report to the blockchain network and add access rights to medical institutions and medical staff, bind each smart audit report to the blockchain transaction ID and record the generation time; For the intelligent audit report, an initial confidence threshold is set based on the accuracy of historical intelligent audit reports and expert review results, and multidimensional evidence that affects the confidence of the intelligent audit report is collected, including sample quality, patient medical history and expert feedback. The confidence of the current intelligent audit report is calculated based on the multidimensional evidence and compared with the pre-set confidence threshold. The intelligent audit report with a confidence higher than the confidence threshold is output as a credible audit result. For the intelligent audit report that has not been output, it is reviewed through remote expert collaboration, and the review result is output as a credible audit result.

4. The method according to claim 3, characterized in that The confidence of the current intelligent audit report is calculated based on the multi-dimensional evidence as shown in the following formula: Where C represents the confidence level, α represents the posterior probability control factor, P(R|H) represents the posterior probability that the intelligent audit report is credible given the multidimensional evidence set H, β represents the evidence quality control factor, m(A) represents the evidence quality, γ represents the information entropy adjustment factor, P(E i ) represents the marginal probability of the i-th dimension of evidence, and n represents the total number of evidence dimensions.

5. A clinical test result review system based on artificial intelligence and big data, used to implement the method described in any one of claims 1 to 4, characterized in that: include: The first unit is used to obtain the corresponding relationship between drugs and disease indicators and drug properties, construct a first knowledge graph, obtain physiological disease characteristics and quality control information collected by multimodal sensors, obtain the historical test data and medication status of the current patient in combination with the first knowledge graph link, generate a heterogeneous data set, perform intelligent fusion on the heterogeneous data set, extract semantic features, and expand the heterogeneous data set through data enhancement technology to obtain high-quality sample data; The second unit is used to obtain real-time message data of the inspection equipment, combine blockchain technology for security authentication, build a spatiotemporal consistency verification mechanism in combination with the high-quality sample data, compare the adjacent sample data located in the same test tube rack, generate sample association verification results, use the high-quality sample data as input and the sample association verification results as constraints, build an initial automatic review model and output automatic review results, compare the automatic review results with the manual approval results for blockchain verification, adjust the automatic review rules based on the verification results, and obtain a high-precision automatic review model; The third unit is used to obtain the sample data of the current patient and input it into the high-precision automatic audit model, integrate the first knowledge graph and perform semantic reasoning, comprehensively judge the sample abnormality risk and output an intelligent audit report, securely share the intelligent audit report through blockchain technology, calculate the confidence of the current intelligent audit report based on multi-dimensional evidence and compare it with a preset confidence threshold, output the intelligent audit report with a confidence higher than the confidence threshold as a credible audit result, and for the intelligent audit report that has not been output, review it through remote collaboration of experts, and output the review result as a credible audit result.

6. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 4.

7. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Citation Information

Patent Citations

  • Medical insurance medical document auditing method and system based on knowledge graph

    CN113360671A

  • Method for analyzing endoscopic image and automatically generating diagnostic report

    CN117524402A

Cited By

  • Patient relation association and information matching method and device based on knowledge graph

    CN120067157A

  • Patient relationship association and information matching method and device based on knowledge graph

    CN120067157B

  • High-name-duplication user identification and retrieval method and device based on multi-feature fusion, medium and program product

    CN121958354A

  • Examination data auditing method and system based on knowledge graph

    CN122266598A

  • A knowledge graph-based inspection data auditing method and system

    CN122266598B