Clinical medical examination data processing method and system based on big data

Abstract

This invention discloses a method and system for processing clinical medical laboratory data based on big data, belonging to the field of clinical data analysis technology. First, clinical laboratory data and medical record data are collected from a hospital database. The clinical laboratory data undergoes structured preprocessing to construct a word co-occurrence matrix. Based on the entity relationships between the clinical laboratory data, clinical feature vectors are generated. A knowledge graph is established based on the medical record data. Based on the knowledge graph, the clinical feature vectors are classified using an improved ConvE model to obtain laboratory test results data. Finally, a disease probability processing engine is constructed to characterize the laboratory test results data and output the final diagnostic medical record. This invention solves the problem that the model's diagnostic logic does not conform to medical common sense, making the AI's diagnostic reasoning process closer to that of human experts.

Description

Technical Field

[0001] This invention relates to the field of clinical data analysis technology, and in particular to a method and system for processing clinical medical laboratory data based on big data. Background Technology

[0002] Clinical laboratory testing is the cornerstone of modern medicine. Through the analysis of samples such as blood, body fluids, secretions, and medical images, it provides objective and quantitative scientific evidence for disease prevention, diagnosis, treatment, and prognosis. With advancements in medical technology and the widespread adoption of healthcare information technology, the speed and scale of clinical laboratory data generation are growing at an unprecedented rate. Traditional clinical laboratory data processing logic primarily relies on manual interpretation and experience, isolated database management, and simple statistical analysis. Therefore, there is an urgent need to utilize big data technology to analyze massive amounts of multi-source clinical laboratory data.

[0003] Currently, Chinese invention patent application number 202510691563.8 discloses a clinical medical laboratory data classification system and method based on big data analysis. This invention mainly includes: a human-computer interaction module that extracts and displays user data classification results via a web / mobile interface, supporting data upload, classification result visualization, and report export; a data processing module that performs data cleaning (denoising, missing value imputation), standardization (LOINC encoding mapping), and feature extraction; an intelligent classification module that integrates random forest and SVM algorithm models to support automatic classification of laboratory items; and a database module for storing user information, user clinical medical laboratory data, historical clinical medical laboratory data, a clinical medical laboratory knowledge graph, clinical medical laboratory keywords, model metadata, and data classification results. The intelligent classification system for clinical medical laboratory data integrates standardized data with AI analysis.

[0004] The aforementioned technologies struggle to address the inconsistency in data standards between clinical laboratory data and medical record data. The diagnostic models ignore medical knowledge logic and fail to analyze disease risk test results and the probability of potential disease risks based on clinical data. Summary of the Invention

[0005] The technical problem solved by this invention is: the difficulty in handling the inconsistency of data standards between clinical test data and medical record data, the diagnostic model ignoring medical knowledge logic, and the failure to analyze the disease risk test results and the probability of disease risks based on clinical data.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] The first aspect is the clinical medical laboratory data processing method based on big data, which includes the following steps:

[0008] Step S1: Collect clinical laboratory data and medical record data based on the hospital database;

[0009] Step S2: Perform structured preprocessing on the clinical test data, construct a word co-occurrence matrix, and generate clinical feature vectors based on the entity relationships between the clinical test data.

[0010] Step S3: Build a knowledge graph based on medical record data, and classify clinical feature vectors using an improved ConvE model based on the knowledge graph to obtain test result data;

[0011] Step S4: Construct a disease probability processing engine to characterize the test results data and output the final diagnostic medical record.

[0012] Preferably, step S1 includes the following sub-steps:

[0013] Step S11: Link the hospital information system and the clinical database to establish an intermediate information database;

[0014] Step S12: Deploy API interfaces in the information middleware library, call the API to perform data acquisition and processing, and obtain clinical laboratory data and medical record data;

[0015] Step S13: Anonymize the medical record data to generate a patient number, and match the clinical test data with the medical record data one by one according to the patient number;

[0016] The clinical laboratory data includes patient ID, sample ID, sample type, test name, test result data, and reference range data;

[0017] The medical record data includes patient number, age, diagnosis information, disease progress information, and medical orders.

[0018] Preferably, step S2 includes the following sub-steps:

[0019] Step S21: Perform character cleaning and stop word filtering on the clinical test data to obtain the first data;

[0020] Step S22: Use jieba word segmentation to extract medical text words from the first data and medical record data, set up a thesaurus, normalize the medical text words, generate text sequences, and map the text sequences one by one to the first data and medical record data.

[0021] Step S23: Construct a word co-occurrence matrix and generate clinical feature vectors based on the entity relationships between the first data.

[0022] Preferably, the processing logic for constructing the word co-occurrence matrix is ​​as follows:

[0023] Set up a sliding window, traverse the text sequence of the first data, and count the number of times medical entities appear in the text sequence of each sliding window. The medical entities include test items, test results, symptoms and signs, disease diagnoses, drug treatments, and organ locations. When medical entity i and medical entity j appear simultaneously in the same sliding window, the matrix element values ​​are... Add one to obtain the word co-occurrence matrix;

[0024] The processing logic for generating clinical feature vectors based on the entity relationships between the first set of data is as follows:

[0025] The singular value decomposition method is used to reduce the dimensionality of the word co-occurrence matrix, and the floating-point number between entity i and entity j is calculated using cosine similarity to generate the embedding vector;

[0026] The standardized text sequence identified from the medical record data is averaged and aggregated with all medical entities in the embedded vector to generate a clinical feature vector, which is used to represent the clinical status and semantic information of the current patient's medical record.

[0027] Preferably, step S3 includes the following sub-steps:

[0028] Step S31: Construct a medical entity relationship knowledge graph based on medical record data, map different terms of the same concept to a unified standard entity, set entity relationship edges, the entity relationship edges are used to connect head entities and tail entities, the head entities include test items, disease diagnosis and organ location, the tail entities include test results, symptoms and signs and drug treatment, use entity relationship edges to connect head entities and tail entities according to causal relationship to generate triple entities, and save all triple entities as a medical entity relationship knowledge graph;

[0029] Step S32: For the medical entity relationship knowledge graph, an improved ConvE model is used to process the medical entity relationship knowledge graph and output the first clinical feature vector.

[0030] Step S33: Input the first clinical feature vector into the classification diagnostic model to obtain test result data, which includes disease category and disease probability.

[0031] Preferably, the processing logic for classifying clinical feature vectors using the improved ConvE model is as follows:

[0032] From clinical feature vectors Extract the corresponding medical concepts with non-zero dimensions. These medical concepts are used to represent the causal relationship in clinical diagnosis. Extract m medical concepts and save them as a set. , For each medical concept The process involves finding the corresponding embedding vectors, generating concept vector groups, identifying the corresponding concept symptoms in the knowledge graph based on these concept vector groups, extracting multiple jump domains to form a subgraph, and then using a graph attention network to generate context embedding vectors. The graph attention network uses the LeakyReLU activation function, and the knowledge graph is set as... The head entity h and relation edges r are embedded into a two-dimensional matrix, convolution is performed using kernels of different sizes, the output feature maps are concatenated, and then dot product is performed with the tail entity t through residual connections and fully connected layers to obtain the association score. Based on dual attention weights, the context embedding vector and association score are fused to find the attention of related concepts in the knowledge graph for each concept in the clinical feature vector x, thus obtaining the first context vector. The corresponding concept disease found in the knowledge graph is then fused with the attention in the clinical feature vector to obtain the second context vector. The first context vector, the second context vector, and the clinical feature vector are then fused to calculate and generate the first clinical feature vector. The expression for calculating the first clinical feature vector is:

[0033] ;

[0034] in, For related scores, This is the first context vector. For the second context vector, For clinical feature vectors, This is the first clinical feature vector.

[0035] The logic for processing the test results data by inputting the first clinical feature vector into the classification and diagnostic model is as follows:

[0036] First clinical features The latent vector z is projected into the same joint latent space as the knowledge graph embedding to obtain a latent vector z. The latent vector has k-dimensional entities, and the number of k-dimensional entities is the same as the number of disease types. The latent vector is input into the classification diagnosis model, and the probability of each disease label is output to obtain the test result data. The classification diagnosis model is an ML-KNN model.

[0037] Preferably, step S4 specifically includes:

[0038] The test results data are stratified by diagnostic confidence level, a disease probability processing engine is set up, and the test results data are classified and extracted for potential diseases to obtain the final diagnosis record. The final diagnosis record includes disease category, disease probability and potential risk label.

[0039] Preferably, a disease probability processing engine is set up, and the processing logic for classifying and extracting potential diseases from the test result data is as follows:

[0040] The test result data is subjected to a first judgment, a second judgment, and a third judgment. A diagnostic probability threshold is set. The diagnostic probability threshold includes a first threshold and a second threshold. The first judgment includes: marking the corresponding disease categories with a disease probability greater than the first threshold as specific diseases and sorting them in descending order according to the disease probability.

[0041] The second determination includes:

[0042] Disease categories with a prevalence between the first and second thresholds are labeled as potential diseases and sorted in descending order of prevalence.

[0043] The third determination includes: removing the corresponding disease categories that are less than the second threshold.

[0044] Secondly, a big data-based clinical medical laboratory data processing system includes a data acquisition module, a feature construction module, a test prediction module, and a result classification module.

[0045] The data acquisition module is used to collect clinical laboratory data and medical record data from the hospital database;

[0046] The feature building module is used to construct a word co-occurrence matrix, perform structured preprocessing on clinical test data, and obtain clinical feature vectors.

[0047] The test prediction module is used to build a knowledge graph using the ConvE model, classify clinical feature vectors, fuse context embedding vectors and association scores to obtain a first clinical feature vector, and then perform multi-label classification on the first clinical feature vector to obtain test result data.

[0048] The result classification module is used to characterize and output the probability distribution of clinical medical risks in patients by testing result data.

[0049] The beneficial effects of this invention are as follows: By combining traditional natural language processing techniques, word co-occurrence matrices, and ConvE models, the structured information of knowledge graphs is deeply embedded into the representation of clinical feature vectors using ConvE models. This achieves a re-execution classification upgrade from data features to semantic features. Through the semantic constraints of the knowledge graph, the model not only relies on statistical patterns of data during classification but also follows medical logic, effectively identifying diseases with similar clinical manifestations and reducing misdiagnosis. Traditional machine learning models are prone to learning false associations in the data. This embodiment, by introducing ConvE knowledge graphs, injects the entity relationships of human experts' medical knowledge as priors into the model, solving the problem that the model's diagnostic logic does not conform to medical common sense. This makes the AI's diagnostic reasoning process closer to that of human experts, transforming the original diagnostic codes and probabilities output by the model into a final diagnostic medical record that is easy for doctors to read and use, thus solving the problem of obscure and difficult-to-understand diagnostic results. Attached Figure Description

[0050] Figure 1 A flowchart of steps for a clinical medical laboratory data processing method based on big data is provided for one embodiment of the present invention;

[0051] Figure 2 This invention provides a basic flowchart of a big data-based clinical medical laboratory data processing system, which is provided as an embodiment of the present invention. Detailed Implementation

[0052] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0053] Example 1, referring to Figure 1 It provides a big data-based method for processing clinical medical laboratory data, including the following steps:

[0054] Step S1: Collect clinical laboratory data and medical record data based on the hospital database;

[0055] Step S2: Perform structured preprocessing on the clinical laboratory data, construct a word co-occurrence matrix, and generate clinical feature vectors based on the entity relationships between the clinical laboratory data.

[0056] Step S3: Build a knowledge graph based on medical record data, and classify clinical feature vectors using an improved ConvE model based on the knowledge graph to obtain test result data;

[0057] Step S4: Construct a disease probability processing engine to characterize the test results data and output the final diagnostic medical record.

[0058] This embodiment combines traditional natural language processing techniques, word co-occurrence matrices, with the ConvE model. By deeply embedding structured information from the knowledge graph into the representation of clinical feature vectors using the ConvE model, it achieves a re-classification upgrade from data features to semantic features. Through the semantic constraints of the knowledge graph, the model not only relies on statistical patterns in the data but also follows medical logic during classification, effectively identifying diseases with similar clinical manifestations and reducing misdiagnosis. Traditional machine learning models are prone to learning spurious associations in the data. This embodiment, by introducing the ConvE knowledge graph, injects the entity relationships of human experts' medical knowledge as priors into the model, solving the problem that the model's diagnostic logic does not conform to common medical sense. This makes the AI's diagnostic reasoning process closer to that of human experts, transforming the original diagnostic codes and probabilities output by the model into a final diagnostic medical record that is easy for doctors to read and use, thus solving the problem of obscure and difficult-to-use diagnostic results.

[0059] Step S1 includes the following sub-steps:

[0060] Step S11: Link the hospital information system and the clinical database to establish an intermediate information database;

[0061] Step S12: Deploy API interfaces in the information middleware library, call the API to perform data acquisition and processing, and obtain clinical laboratory data and medical record data;

[0062] Step S13: Anonymize the medical record data, generate patient IDs, and match the clinical test data with the medical record data one by one according to the patient IDs;

[0063] Clinical laboratory data includes patient ID, sample ID, sample type, test name, test result data, and reference range data;

[0064] Medical record data includes patient number, age, diagnosis information, disease progress information, and medical orders.

[0065] An information intermediary database was established, and data was automatically collected via API interfaces, replacing traditional manual export or reporting methods. This ensured the real-time nature and stability of the data stream. Before data processing, anonymization was performed to remove personally identifiable information, strictly adhering to data privacy regulations. Simultaneously, the generated patient IDs precisely linked clinical laboratory data and medical record data from different systems, laying the foundation for building a unified data view for each patient.

[0066] Step S2 includes the following sub-steps:

[0067] Step S21: Perform character cleaning and stop word filtering on the clinical test data to obtain the first data;

[0068] Step S22: Use jieba for word segmentation, extract medical text words from clinical test data and medical record data, set up a synonym expression library, normalize the medical text words, generate a standardized text sequence, and map the standardized text sequence to the first data and medical record data one by one;

[0069] Step S23: Construct a word co-occurrence matrix and generate clinical feature vectors for the entity relationships between the first data.

[0070] In this embodiment, first, character cleaning and stop word filtering are performed on the clinical test data to remove irrelevant characters in the text, such as special symbols (@, #, $), garbled characters, meaningless punctuation, numbers (except those related to medical values, such as "body temperature 38.5°C" which need to be retained), etc. Use a stop word list to filter out common words that contribute little to semantic analysis, such as "of", "already", "patient", "admission", "examination", and retain entities and keywords with actual medical significance;

[0071] Use the jieba word segmentation tool and enhance it by loading a medical dictionary to ensure that medical proper nouns (such as "type II diabetes", "coronary atherosclerosis") can be correctly segmented. After that, the system pre-sets a synonym expression library, mapping "fever", "high fever" to the standard term "fever"; mapping "white blood cell count", "WBC", "white blood cell" to "white blood cell count"; mapping "CA", "Ca" to "cancer". After word segmentation and normalization, the original text is transformed into a sequence composed of standard medical terms, providing a unified input column for subsequent model learning.

[0072] The processing logic for constructing the word co-occurrence matrix is as follows:

[0073] Set a sliding window, traverse the text sequence of the first data, and count the number of occurrences of medical entities in the text sequence of each sliding window. Medical entities include test items, test results, symptoms and signs, disease diagnoses, drug treatments, and organ parts. When medical entity i and medical entity j appear in the same sliding window at the same time, the matrix element value is incremented by one to obtain the word co-occurrence matrix;

[0074] The processing logic for generating clinical feature vectors for the entity relationships between the first data is as follows:

[0075] Use the singular value decomposition method to reduce the dimension of the word co-occurrence matrix, and use the cosine similarity to calculate the floating-point number between entity i and entity j to generate an embedding vector;

[0076] Average and aggregate the standardized text sequences identified from the medical record data with all medical entities in the embedding vector to generate clinical feature vectors, which are used to represent the clinical conditions and semantic information of the current patient's medical record.

[0077] Select a window of appropriate size (e.g., 5 words) and slide it over the text sequence. For each pair of medical entities (i and j) that appear simultaneously within the window, add 1 to the corresponding position (i,j) in the word co-occurrence matrix. Using the matrix obtained by SVD, the k-dimensional embedding vector of each entity can be obtained. For a medical record, identify all the canonical text sequences (i.e., medical entities), then find the embedding vectors corresponding to these entities, and average these vectors to obtain the clinical feature vector of the medical record. In this embodiment, through the co-occurrence matrix and embedding vectors, the semantic relationships of medical entities (such as symptoms and diseases that often appear simultaneously) are preserved in the vector representation. Perform SVD decomposition on the word co-occurrence matrix, select the k largest singular values ​​and their corresponding singular vectors, and reduce the original high-dimensional matrix to k dimensions. SVD dimensionality reduction reduces the dimensionality of the data while retaining the most important information, alleviating the curse of dimensionality.

[0078] Step S3 includes the following sub-steps:

[0079] Step S31: Construct a medical entity relationship knowledge graph based on medical record data, map different terms of the same concept to a unified standard entity, set entity relationship edges, and use entity relationship edges to connect head entities and tail entities. Head entities include test items, disease diagnoses and organ sites, and tail entities include test results, symptoms and signs and drug treatment. Use entity relationship edges to connect head entities and tail entities according to causal relationships to generate triple entities, and save all triple entities as a medical entity relationship knowledge graph.

[0080] Step S32: For the medical entity relationship knowledge graph, an improved ConvE model is used to process the medical entity relationship knowledge graph and output the first clinical feature vector.

[0081] Step S33: Input the first clinical feature vector into the classification and diagnosis model to obtain test result data, which includes disease category and probability of disease.

[0082] Based on medical knowledge, causal relationships between head and tail entities are defined and connected by relation edges. For example, a diagnosis of "pneumonia" may lead to the symptom "cough" (causal relationship), and a test item "white blood cell count" may produce a test result of "elevated". This knowledge is represented in the form of (head entity, relation, tail entity), such as (pneumonia, leads to, cough) or (white blood cell count, result is, elevated). All triples are stored to form a medical entity relation knowledge graph. By connecting these causal relationships, a knowledge graph network that conforms to clinical reasoning logic is constructed. The construction of this entity knowledge graph not only provides a reliable knowledge foundation for subsequent intelligent diagnostic algorithms, but also greatly enhances the interpretability of diagnostic results through the explicit expression of causal relationships.

[0083] The method for processing clinical feature vectors using the improved ConvE model is as follows:

[0084] From clinical feature vectors Extract the corresponding medical concepts with non-zero dimensions. These medical concepts represent the causal relationships in clinical diagnosis. Extract m medical concepts and store them as a set. , For each medical concept The process involves finding the corresponding embedding vectors, generating concept vector sets, identifying the corresponding conceptual symptoms in the knowledge graph based on these concept vector sets, extracting multiple jump domains to form a subgraph, and then using a graph attention network to generate context embedding vectors. The graph attention network uses LeakyReLU as the activation function in the nonlinear transformation of node feature aggregation, and sets the knowledge graph as... The head entity h and relation edges r are embedded into a two-dimensional matrix, convolution is performed using kernels of different sizes, the output feature maps are concatenated, and then dot product is performed with the tail entity t through residual connections and fully connected layers to obtain the association score. Based on dual attention weights, the context embedding vector and association score are fused to find the attention of related concepts in the knowledge graph for each concept in the clinical feature vector x, thus obtaining the first context vector. The corresponding concept disease found in the knowledge graph is then fused with the attention in the clinical feature vector to obtain the second context vector. The first context vector, the second context vector, and the clinical feature vector are then fused to calculate and generate the first clinical feature vector. The expression for calculating the first clinical feature vector is:

[0085] ;

[0086] in, For related scores, This is the first context vector. For the second context vector, For clinical feature vectors, This is the first clinical feature vector;

[0087] Among them, the correlation score The result is obtained through an improved ConvE model: the embedding vectors of the head entity h and relation r are concatenated and reshaped into a two-dimensional matrix. Multiple convolutional kernels of different sizes are used for convolution operations. The output feature maps are concatenated and mapped to a vector through a fully connected layer. This vector is then used to perform a dot product operation with the embedding vector of the tail entity t. Finally, the dot product result is normalized to the (0, 1) interval using the Sigmoid function. , The closer the value is to 1, the stronger the semantic association between the current clinical features and the pathological relationship represented by the triple (h, r, t) in the knowledge graph.

[0088] Extract medical concepts corresponding to non-zero dimensions from the clinical feature vector x. Each dimension of the clinical feature vector x corresponds to a concept in a predefined medical concept library. Select dimensions in the clinical feature vector x whose absolute value is greater than a preset significance threshold (the preset significance threshold can be set to 0.05). The medical concepts corresponding to the current dimension constitute the set C of the current medical record. Assuming there are m concepts, denoted as set C, for each concept c_i, find the corresponding node in the knowledge graph and obtain its embedding vector e_i (entity embedding obtained through a pre-trained ConvE model or other methods). For each concept c_i, extract its multi-hop neighborhood (e.g., 2-hop) in the knowledge graph to form a subgraph, and then use a graph attention network (GAT) to generate the context embedding vector e_i_ctx of the concept.

[0089] The improved ConvE model is used to calculate the association score s. The improved ConvE model concatenates the embeddings of the head entity h and relation r into a two-dimensional matrix, then performs convolution using multi-scale convolution kernels. The output feature maps of different convolution kernels are concatenated, and the result is passed through residual connections and fully connected layers. Finally, the score is obtained by dot product with the embedding of the tail entity t. This embodiment employs a dual attention mechanism: the first context vector c1 calculates the attention of each concept in the clinical feature vector x to related concepts in the knowledge graph (i.e., clinical-to-knowledge attention), and then weighted and aggregated to obtain c1; the second context vector c2 calculates the attention of concepts in the knowledge graph to the clinical feature vector (i.e., knowledge-to-clinical attention), and then weighted and aggregated to obtain c2. Finally, the association score s, the first context vector c1, the second context vector c2, and the original clinical feature vector x are fused according to the formula to obtain x'.

[0090] By integrating knowledge graph information, the semantic richness and diagnostic value of clinical feature vectors are significantly improved. By using convolutional kernels of different sizes, it is possible to simultaneously capture the close local associations and long-range dependencies between medical concepts. Through bidirectional attention from clinical to knowledge and from knowledge to clinical, deep interaction between data-driven and knowledge-driven approaches is achieved. By using graph attention networks to generate contextual embeddings, the model can dynamically adjust the importance of concepts according to specific cases.

[0091] The logic for processing the test results data by inputting the first clinical feature vector into the classification and diagnostic model is as follows:

[0092] First clinical features The latent vector z is projected into the same joint latent space as the knowledge graph embedding to obtain a latent vector z. The latent vector has k-dimensional entities, and the number of k-dimensional entities is the same as the number of disease types. The latent vector is input into the classification and diagnosis model, and the probability of each disease label is output to obtain the test result data. The classification and diagnosis model is the ML-KNN model.

[0093] In this embodiment, ML-KNN is used to naturally handle multi-label problems, and multi-label disease diagnosis is performed in the knowledge-aligned potential space. This realizes an intelligent diagnostic system that is data-efficient and meets the needs of clinical practice, and captures disease co-occurrence patterns.

[0094] Step S4 specifically includes:

[0095] The test results data are stratified by diagnostic confidence level, a disease probability processing engine is set up, and the test results data are classified and extracted for potential diseases to obtain the final diagnosis record. The final diagnosis record includes disease category, disease probability and potential risk label.

[0096] The raw probability output is transformed into clinically operable hierarchical diagnostic information. By constructing a confidence stratification mechanism and setting potential risk labels, a unified medical record format that conforms to clinical standards is generated.

[0097] The disease probability processing engine is configured to perform potential disease classification and extraction on the test results data. The processing logic is as follows:

[0098] The test results data are judged in three stages: first, second and third. A diagnostic probability threshold is set. The diagnostic probability threshold includes a first threshold and a second threshold. The first judgment includes: marking the corresponding disease categories with a disease probability greater than the first threshold as specific diseases and sorting them in descending order of disease probability.

[0099] The second determination includes:

[0100] Disease categories with a prevalence between the first and second thresholds are labeled as potential diseases and sorted in descending order of prevalence.

[0101] The third criterion includes removing the corresponding disease categories that are below the second threshold.

[0102] The first threshold is set to 0.7 and the second threshold is set to 0.3. The first and second thresholds are the result of analyzing and optimizing the disease probability output by the model based on historical labeled data (confirmed medical records). This method divides diseases into three categories by setting two thresholds (the first threshold and the second threshold): clear diseases, potential diseases, and diseases that need to be excluded. This stratification method helps doctors focus on high-probability diseases while not ignoring medium-probability diseases, thus improving the comprehensiveness and accuracy of diagnosis.

[0103] Example 2, refer to Figure 2 It provides a big data-based clinical medical laboratory data processing system, including a data acquisition module, a feature construction module, a test prediction module, and a result classification module;

[0104] The data acquisition module is used to collect clinical laboratory data and medical record data from the hospital database;

[0105] The feature building module is used to construct a word co-occurrence matrix, perform structured preprocessing on clinical laboratory data, and obtain clinical feature vectors.

[0106] The test prediction module is used to build a knowledge graph using the ConvE model, classify clinical feature vectors, fuse context embedding vectors and association scores to obtain the first clinical feature vector, and then perform multi-label classification on the first clinical feature vector to obtain test result data.

[0107] The results classification module is used to characterize and output the probability distribution of clinical medical risks in patients by testing results data.

[0108] This embodiment constructs an intelligent and reliable clinical laboratory data processing system through four closely cooperating modules. This system not only realizes intelligent analysis of laboratory data, but more importantly, it organically integrates medical knowledge, clinical data and artificial intelligence technology, providing doctors with comprehensive decision support from current diagnosis to future risks, reflecting the core value of big data and AI in the medical field.

[0109] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0110] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for processing clinical medical laboratory data based on big data, characterized in that, Includes the following steps: Step S1: Collect clinical laboratory data and medical record data based on the hospital database; Step S2: Perform structured preprocessing on the clinical test data, construct a word co-occurrence matrix, and generate clinical feature vectors based on the entity relationships between the clinical test data. Step S3: Build a knowledge graph based on medical record data, and classify clinical feature vectors using an improved ConvE model based on the knowledge graph to obtain test result data; Step S4: Construct a disease probability processing engine to characterize the test results data and output the final diagnostic medical record; The processing logic for classifying clinical feature vectors using the improved ConvE model is as follows: From clinical feature vectors Extract the corresponding medical concepts with non-zero dimensions. These medical concepts are used to represent the causal relationship in clinical diagnosis. Extract m medical concepts and save them as a set. , For each medical concept The process involves finding the corresponding embedding vectors, generating concept vector groups, identifying the corresponding concept symptoms in the knowledge graph based on these concept vector groups, extracting multiple jump domains to form a subgraph, and then using a graph attention network to generate context embedding vectors. The graph attention network uses the LeakyReLU activation function, and the knowledge graph is set as... The head entity h and relation edges r are embedded into a two-dimensional matrix, convolution is performed using kernels of different sizes, the output feature maps are concatenated, and then dot product is performed with the tail entity t through residual connections and fully connected layers to obtain the association score. Based on dual attention weights, the context embedding vector and association score are fused to find the attention of related concepts in the knowledge graph for each concept in the clinical feature vector x, thus obtaining the first context vector. The corresponding concept disease found in the knowledge graph is then fused with the attention in the clinical feature vector to obtain the second context vector. The first context vector, the second context vector, and the clinical feature vector are then fused to calculate and generate the first clinical feature vector. The expression for calculating the first clinical feature vector is: ； in, For related scores, This is the first context vector. For the second context vector, For clinical feature vectors, This is the first clinical feature vector.

2. The clinical medical laboratory data processing method based on big data as described in claim 1, characterized in that, Step S1 includes the following sub-steps: Step S11: Link the hospital information system and the clinical database to establish an intermediate information database; Step S12: Deploy API interfaces in the information middleware library, call the API to perform data acquisition and processing, and obtain clinical laboratory data and medical record data; Step S13: Anonymize the medical record data to generate a patient number, and match the clinical test data with the medical record data one by one according to the patient number; The clinical laboratory data includes patient ID, sample ID, sample type, test name, test result data, and reference range data; The medical record data includes patient number, age, diagnosis information, disease progress information, and medical orders.

3. The clinical medical laboratory data processing method based on big data as described in claim 2, characterized in that, Step S2 includes the following sub-steps: Step S21: Perform character cleaning and stop word filtering on the clinical test data to obtain the first data; Step S22: Use jieba word segmentation to extract medical text words from the first data and medical record data, set up a thesaurus, normalize the medical text words, generate text sequences, and map the text sequences one by one to the first data and medical record data. Step S23: Construct a word co-occurrence matrix and generate clinical feature vectors based on the entity relationships between the first data.

4. The clinical medical laboratory data processing method based on big data as described in claim 3, characterized in that, The processing logic for constructing the word co-occurrence matrix is ​​as follows: Set up a sliding window, traverse the text sequence of the first data, and count the number of times medical entities appear in the text sequence of each sliding window. The medical entities include test items, test results, symptoms and signs, disease diagnoses, drug treatments, and organ locations. When medical entity i and medical entity j appear simultaneously in the same sliding window, the matrix element values ​​are... Add one to obtain the word co-occurrence matrix; The processing logic for generating clinical feature vectors based on the entity relationships between the first set of data is as follows: The singular value decomposition method is used to reduce the dimensionality of the word co-occurrence matrix, and the floating-point number between entity i and entity j is calculated using cosine similarity to generate the embedding vector; The standardized text sequence identified from the medical record data is averaged and aggregated with all medical entities in the embedded vector to generate a clinical feature vector, which is used to represent the clinical status and semantic information of the current patient's medical record.

5. The clinical medical laboratory data processing method based on big data as described in claim 4, characterized in that, Step S3 includes the following sub-steps: Step S31: Construct a medical entity relationship knowledge graph based on medical record data, map different terms of the same concept to a unified standard entity, set entity relationship edges, the entity relationship edges are used to connect head entities and tail entities, the head entities include test items, disease diagnosis and organ location, the tail entities include test results, symptoms and signs and drug treatment, use entity relationship edges to connect head entities and tail entities according to causal relationship to generate triple entities, and save all triple entities as a medical entity relationship knowledge graph; Step S32: For the medical entity relationship knowledge graph, an improved ConvE model is used to process the medical entity relationship knowledge graph and output the first clinical feature vector. Step S33: Input the first clinical feature vector into the classification diagnostic model to obtain test result data, which includes disease category and disease probability.

6. The clinical medical laboratory data processing method based on big data as described in claim 5, characterized in that, The logic for processing the test results data by inputting the first clinical feature vector into the classification and diagnostic model is as follows: First clinical features The latent vector z is projected into the same joint latent space as the knowledge graph embedding to obtain a latent vector z. The latent vector has k-dimensional entities, and the number of k-dimensional entities is the same as the number of disease types. The latent vector is input into the classification diagnosis model, and the probability of each disease label is output to obtain the test result data. The classification diagnosis model is an ML-KNN model.

7. The clinical medical laboratory data processing method based on big data as described in claim 6, characterized in that, Step S4 specifically includes: The test results data are stratified by diagnostic confidence level, a disease probability processing engine is constructed, and the test results data are classified and extracted for potential diseases to obtain the final diagnosis record. The final diagnosis record includes disease category, disease probability and potential risk label.

8. The clinical medical laboratory data processing method based on big data as described in claim 7, characterized in that, The disease probability processing engine is configured to perform potential disease classification and extraction on the test result data. The processing logic is as follows: The test result data is subjected to a first judgment, a second judgment, and a third judgment. A diagnostic probability threshold is set. The diagnostic probability threshold includes a first threshold and a second threshold. The first judgment includes: marking the corresponding disease categories with a disease probability greater than the first threshold as specific diseases and sorting them in descending order according to the disease probability. The second determination includes: Disease categories with a prevalence between the first and second thresholds are labeled as potential diseases and sorted in descending order of prevalence. The third determination includes: removing the corresponding disease categories that are less than the second threshold.

9. A big data-based clinical medical laboratory data processing system, applied in any one of the big data-based clinical medical laboratory data processing methods as described in claims 1-8, characterized in that, It includes a data acquisition module, a feature building module, a test and prediction module, and a result classification module; The data acquisition module is used to collect clinical laboratory data and medical record data from the hospital database; The feature building module is used to construct a word co-occurrence matrix, perform structured preprocessing on clinical test data, and obtain clinical feature vectors. The test prediction module is used to build a knowledge graph using the ConvE model, classify clinical feature vectors, fuse context embedding vectors and association scores to obtain a first clinical feature vector, and then perform multi-label classification on the first clinical feature vector to obtain test result data. The result classification module is used to characterize and output the probability distribution of clinical medical risks in patients by testing result data.

Images