Hospital infection risk analysis system based on big data analysis

Abstract

The invention discloses a hospital infection risk analysis system based on big data analysis, and relates to the technical field of medical information analysis, and the system comprises a data collection module which comprises a data reading unit and a data preprocessing unit, and is used for carrying out the data extraction of a system database of a target hospital, and obtaining a standard data set through the standardization processing; the graph construction module comprises a word segmentation unit, an entity extraction unit and a graph generation unit, and is used for performing word segmentation processing on the standard data set, and constructing a first knowledge graph and a second knowledge graph through feature extraction and entity attribute label labeling; the infection risk prediction module comprises a graph-to-digital conversion unit and a risk analysis unit, and is used for carrying out graph-feature value conversion on the two knowledge graphs, carrying out analysis and evaluation through a multi-layer perceptron model and outputting a risk prediction result; the interaction terminal module comprises a display unit and an interaction unit and is used for providing an operation interface for a user, receiving and executing an input instruction of the user and feeding back an operation result to the user.

Description

Technical Field

[0001] The present invention relates to the technical field of medical information analysis, and in particular to a hospital infection risk analysis system based on big data analysis. Background Art

[0002] In recent years, with the continuous advancement of medical technology, various new operations and the application of antibiotics have gradually become popular. Although advanced invasive operations provide necessary means for the treatment of patients, they also destroy the body's complete immune barrier and increase the probability of infection. At the same time, with the increase in the use of antibiotics in clinical treatment, multidrug-resistant bacteria are becoming increasingly common in hospital environments. These bacteria are often resistant to a variety of commonly used antibiotics and can be transmitted through medical devices, personnel contact and other channels. In addition, as my country gradually enters an aging society, patients with underlying diseases continue to suffer from hospital infections due to insufficient immunity. Therefore, research on technologies for analyzing and predicting the risk of hospital infections has good prospects.

[0003] At present, the Chinese invention patent with application number CN202210157173.9 discloses a hospital infection risk assessment method, equipment and storage medium. The invention specifically includes: obtaining risk factors and risk data of risk factors that cause patient infection during hospitalization, and the risk data includes experimental group data and comparison group data; grouping the risk data, and using the chi-square test to determine whether the preset null hypothesis of the risk factor is established under the preset probability based on the experimental group data and comparison group data of each group; if established, the odds ratio of the risk factor is calculated, and a risk assessment model is established based on the odds ratio and the preset risk coefficient in the preset verification cycle; using the risk assessment model to conduct risk assessment on patients after admission to obtain the assessment weight, and if the assessment weight exceeds the preset warning value, the preset intervention measures are used to intervene in the patient. The invention only collects risk factors and risk data during the patient's hospitalization for assessment, relying on traditional data acquisition methods, and is inefficient when processing a large amount of complex medical record data. The data source is relatively limited, and biochemical means of actively monitoring bacteria are required, which increases the operational complexity of risk assessment. Risk assessment is performed through chi-square detection and odds ratio, and the calculation method is relatively mechanical, which is not conducive to controlling assessment errors. Summary of the invention

[0004] The technical problem solved by the present invention is that when conducting hospital infection risk assessment, the traditional method only obtains risk data directly related to the infection, the data source is not comprehensive enough, and the indirectly related risk data is not fully considered. It is inefficient when processing a large amount of complex hospital database data, and often relies on biochemical means to obtain real-time test data, which increases the complexity and error of data acquisition. In addition, the traditional method focuses on analyzing the patient and the infection occurrence, lacks analysis from the perspective of the hospital as a whole and the infection occurrence, the calculation feature dimension is limited, and the output evaluation result is usually the current infection risk status, which has the problem of insufficient predictability.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] The hospital infection risk analysis system based on big data analysis is characterized by comprising: a data acquisition module, a graph construction module, an infection risk prediction module and an interactive terminal module;

[0007] The data acquisition module includes a data reading unit and a data preprocessing unit, which are used to extract data from the system database of the target hospital to obtain an initial data set, and to normalize the initial data set to obtain a normalized data set;

[0008] The graph construction module includes a word segmentation unit, an entity extraction unit and a graph generation unit, which are used to perform word segmentation processing on the standard data set to obtain a word segmentation data set, perform feature extraction and entity attribute labeling on the word segmentation data set to obtain an entity data set, and construct a first knowledge graph and a second knowledge graph;

[0009] The infection risk prediction module includes a graph-to-feature value conversion unit and a risk analysis unit, which are used to perform graph-to-feature value conversion on the first knowledge graph and the second knowledge graph to obtain a first feature vector set and a second feature vector set, analyze and evaluate the infection risk of the target hospital through a multi-layer perceptron model, and output a risk prediction result;

[0010] The interactive terminal module includes a display unit and an interactive unit, which are used to provide an operation interface and visual results to the user, receive and execute the user's input instructions and feed back the operation results to the user.

[0011] As a preferred solution of the hospital infection risk analysis system based on big data analysis described in the present invention, the data acquisition module includes a data reading unit and a data preprocessing unit.

[0012] The data reading unit is used to extract data from the system database of the target hospital to obtain an initial data set, wherein the initial data set includes patient identity data, patient test data, financial system data, nursing records, hospitalization medication records, symptom scoring system records and historical hospital infection report data;

[0013] The data preprocessing unit is used to perform normalization processing on the initial data set to obtain a normalized data set, and the normalization processing step includes: classifying the character type of the initial data set to obtain text data and digital data, deleting HTML tags, stop words and non-alphanumeric characters in the text data, and performing uppercase and lowercase conversion and stem extraction, separating thousands of digital data, unifying decimal points and time formats, and completing normalization processing to obtain a normalized data set;

[0014] The text data includes Chinese characters, alphabetic characters, non-alphanumeric characters and punctuation marks in the initial data set;

[0015] The digital data includes integer characters, floating point characters, fixed point characters and Boolean characters.

[0016] As a preferred solution of the hospital infection risk analysis system based on big data analysis described in the present invention, wherein: the graph construction module includes a word segmentation unit, an entity extraction unit and a graph generation unit;

[0017] The word segmentation unit is used to perform word segmentation processing on the standard data set to obtain a word segmentation data set, and the word segmentation processing step includes: performing word segmentation on the standard data set based on a dictionary, performing character matching through a preset dictionary library, and extracting the matching characters into the word segmentation data set;

[0018] Based on statistics, the standard data set is segmented twice, the combination probabilities of the remaining characters in the standard data set after the first word segmentation are counted, the combination probabilities are sorted to obtain the maximum combination probability, and characters are extracted from the standard data set based on the maximum combination probability, and the extracted characters are extracted into the word segmentation data set.

[0019] As a preferred solution of the hospital infection risk analysis system based on big data analysis described in the present invention, wherein: the entity extraction unit is used to perform feature extraction and entity attribute label annotation on the word segmentation data set to obtain the entity data set, including: vectorizing the word segmentation data set, vectorizing the word segmentation data set through the character encoding of the bag of words model to obtain the word segmentation vector set, weighted calculation on the word segmentation vector set based on the parallel subspace of multi-head attention to obtain the weighted vector set, reducing the dimension and linear combination processing of the weighted vector set through the principal component analysis method to obtain the reduced dimension vector set, extracting the forward dependency features and reverse dependency features from the reduced dimension vector set through Bi-LSTM, adding the forward dependency features and reverse dependency features to obtain the word segmentation feature data set, using the hidden Markov model to calculate the state transition probability and label the entity attribute labels on the word segmentation feature data set, and completing entity classification and decoding through the maximum likelihood estimation algorithm to obtain the entity data set;

[0020] The entity attribute labels include infection index labels, patient identity labels, infection site labels, patient medical history labels, disease score labels, infection department labels, treatment method labels, hospitalization drug labels, ward status labels, and historical pathogen labels;

[0021] The infection indicator labels include albumin, prealbumin, anemia, antibiotic use, and antibiotic use level;

[0022] The patient identification tag includes the patient's name, gender, and age group;

[0023] The infection site labels include lower respiratory tract, upper respiratory tract, gastrointestinal tract, intra-abdominal tissue, urinary tract infection, respiratory system, blood system, urinary system, digestive system, skin and soft tissue, reproductive system, ear, nose and throat, and oral cavity;

[0024] The patient's medical history labels include: history of diabetes, hypertension, hyperlipidemia, atrial fibrillation, cancer, smoking, drinking, blood transfusion, and stroke;

[0025] The disease scoring labels include ADL score, mRS score, GCS score, NIHSS score, APACHEⅡ score and NRS2002 score;

[0026] The infection department labels include critical care medicine, neurosurgery, orthopedic surgery, general surgery, cardiothoracic surgery, urology, otolaryngology, infectious diseases, oncology, respiratory medicine, cardiology, neurology, gastroenterology, endocrinology, pediatrics, emergency department, and traditional Chinese medicine;

[0027] The treatment labels include antimicrobial therapy, quality control circle therapy, thrombolytic therapy, deep venous catheterization, urinary catheterization, central venous catheterization, use of ventilator, and routine care.

[0028] The inpatient medication label includes dehydration drugs, antimicrobial drugs, anticoagulants, antiplatelet drugs, and glucocorticoids;

[0029] The ward status labels include occupancy rate, number of patients receiving antibiotics, antibiotic use rate, number of patients sent for inspection before antibiotic treatment, inspection rate before antibiotic treatment, ward closure status, and frequency of hospital infection;

[0030] Historical pathogen labels include Klebsiella pneumoniae, Escherichia coli, Staphylococcus aureus, Acinetobacter baumannii, Pseudomonas aeruginosa, Enterobacter cloacae, and fungi.

[0031] As a preferred solution of the hospital infection risk analysis system based on big data analysis described in the present invention, wherein: the graph generation unit is used to construct a first knowledge graph and a second knowledge graph, generate a patient-infection attribute field based on entities containing infection indicator labels, patient identity labels, infection site labels, patient medical history labels, and disease score labels in the entity data set, and extract the relationship of the patient-infection attribute field through the SemRep model to obtain a patient-infection entity relationship, and the patient-infection entity relationship includes: a first hierarchical relationship, a first part-whole relationship, a first synonym relationship, a first near-synonymous relationship, a first antonym relationship, a first spelling variant relationship, a first abbreviation and full name relationship, a first etiology relationship, a first clinical manifestation relationship, and a first treatment relationship;

[0032] Generate hospital-infection attribute fields based on entities containing infection indicator labels, infection department labels, treatment method labels, hospitalization drug labels, ward status labels, and historical pathogen labels in the entity data set, and extract relationships from the hospital-infection attribute fields through the SemRep model to obtain hospital-infection entity relationships, wherein the hospital-infection entity relationships include: second-level relationships, second-part-whole relationships, second synonym relationships, second near-synonymous relationships, second antonymous relationships, second spelling variant relationships, second abbreviation and full name relationships, second etiology relationships, second clinical manifestation relationships, and second treatment relationships;

[0033] Importing the patient-infection entity relationship, the entity data set and the corresponding entity attribute labels into the Neo4j database, mapping each entity in the entity data set into a first graph node set, generating a first edge relationship set through the entity attribute labels and the patient-infection entity relationship, and establishing a first knowledge graph based on the first graph nodes and the first edge relationships;

[0034] The hospital-infection entity relationship, entity data set and corresponding entity attribute labels are imported into the Neo4j database, each entity in the entity data set is mapped into a second graph node set, a second edge relationship set is generated through the entity attribute labels and the hospital-infection entity relationship, and a second knowledge graph is established based on the second graph node set and the second edge relationship set.

[0035] As a preferred solution of the hospital infection risk analysis system based on big data analysis described in the present invention, wherein: the infection risk prediction module includes a graph-to-digital conversion unit and a risk analysis unit, the graph-to-digital conversion unit is used to perform graph-to-feature value conversion on the first knowledge graph and the second knowledge graph to obtain a first feature vector set and a second feature vector set, and the logic of the graph-to-feature value conversion includes: performing graph embedding processing on the first knowledge graph and the second knowledge graph respectively through the TransE algorithm, embedding the corresponding graph nodes and edge relationships into the low-dimensional vector space to obtain the first feature vector set and the second feature vector set respectively, and inputting the first feature vector set and the second feature vector set into the risk analysis unit.

[0036] As a preferred solution of the hospital infection risk analysis system based on big data analysis described in the present invention, wherein: the risk analysis unit is used to analyze and evaluate the infection risk of the target hospital and output the risk prediction result;

[0037] The processing logic of the analysis and evaluation includes: completing the infection risk analysis through a pre-trained multi-layer perceptron model, the structure of the multi-layer perceptron model includes two parallel branches and an output layer trunk, the first branch includes a left input layer and a left hidden layer, and the second branch includes a right input layer and a right hidden layer;

[0038] The number of the left input layer is one, the number of the left hidden layer is M, the number of the right input layer is one, and the number of the right hidden layer is N;

[0039] The left input layer data set is obtained by receiving the data in the first feature vector set by each input neuron in the left input layer respectively, and the left input layer data set is subjected to a first iterative calculation by the left hidden layer to obtain a first deep feature value, wherein the logic of the first iterative calculation includes: calculating the input data by the weight vector and the bias term corresponding to the layer of left hidden layer to obtain a layer of left weighted feature values, performing a nonlinear transformation on the layer of left weighted feature values ​​by the first ReLU activation function to obtain a layer of left hidden layer output, using the layer of left hidden layer output as the input of the second layer of left hidden layer and repeatedly calculating to obtain the second layer of left hidden layer output, and iteratively calculating until the first deep feature value of the M layers of left hidden layer output is obtained;

[0040] The right input layer data set is obtained by receiving the data in the second eigenvector set through each input neuron in the right input layer respectively, and the second deep eigenvalue is obtained by performing a second iterative calculation on the right input layer data set through the right hidden layer, wherein the logic of the second iterative calculation includes: calculating the input data through the weight vector and the bias term corresponding to the first right hidden layer to obtain a layer of right weighted eigenvalues, performing a nonlinear transformation on the first right weighted eigenvalue through the second ReLU activation function to obtain a layer of right hidden layer output, using the first right hidden layer output as the input of the second right hidden layer and repeatedly calculating to obtain the second right hidden layer output, and iteratively calculating until the second deep eigenvalue of the N-layer right hidden layer output is obtained;

[0041] The first deep feature value and the second deep feature value are linearly transformed through the attention mechanism in the output layer to obtain the fused feature result, and the fused feature result is mapped to the output layer classification space. The mapping value in the output layer classification space is normalized by the softmax function to obtain the classification probability corresponding to each entity, and the classification probability corresponding to each entity is sorted from high to low, and the maximum classification probability is selected as the risk prediction result corresponding to each entity;

[0042] The classification probability includes low risk probability, medium risk probability and high risk probability;

[0043] The risk prediction results include low risk, medium risk and high risk.

[0044] As a preferred solution of the hospital infection risk analysis system based on big data analysis described in the present invention, wherein: the interactive terminal module includes a display unit and an interactive unit;

[0045] The interactive unit is used to receive and execute user input instructions and feed back operation results to the user, wherein the input instructions include data import and export, system parameter setting, graph query and risk prediction result query;

[0046] The display unit is used to provide the user with an operation interface and visualization results, and the visualization results include: data display, system status display, graph display and risk prediction result display.

[0047] The beneficial effects of the present invention are as follows: all data directly and indirectly related to infection are collected from the target hospital database, which is conducive to a comprehensive analysis of potential risk relationships. By establishing a knowledge graph, a large amount of complex hospital database data can be efficiently processed, and real-time inspection data can be obtained without relying on biochemical means, which reduces the complexity and error of data acquisition. The output result of the present invention is an infection risk prediction result, which is forward-looking and conducive to discovering potential infection risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1A schematic diagram of the basic framework of a hospital infection risk analysis system based on big data analysis provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0050] Example 1, reference Figure 1 , is an embodiment of the present invention, and provides a hospital infection risk analysis system based on big data analysis, including: a data acquisition module, a graph construction module, an infection risk prediction module and an interactive terminal module;

[0051] The data acquisition module includes a data reading unit and a data preprocessing unit, which are used to extract data from the system database of the target hospital to obtain an initial data set, and to normalize the initial data set to obtain a normalized data set;

[0052] The graph construction module includes a word segmentation unit, an entity extraction unit and a graph generation unit, which are used to perform word segmentation processing on the standard data set to obtain a word segmentation data set, perform feature extraction and entity attribute labeling on the word segmentation data set to obtain an entity data set, and construct a first knowledge graph and a second knowledge graph;

[0053] The infection risk prediction module includes a graph-to-feature value conversion unit and a risk analysis unit, which are used to perform graph-to-feature value conversion on the first knowledge graph and the second knowledge graph to obtain a first feature vector set and a second feature vector set, analyze and evaluate the infection risk of the target hospital through a multi-layer perceptron model, and output a risk prediction result;

[0054] The interactive terminal module includes a display unit and an interactive unit, which are used to provide an operation interface and visual results to the user, receive and execute the user's input instructions and feed back the operation results to the user.

[0055] In this embodiment, the data acquisition module includes a data reading unit and a data preprocessing unit.

[0056] The data reading unit is used to extract data from the system database of the target hospital to obtain an initial data set, wherein the initial data set includes patient identity data, patient test data, financial system data, nursing records, hospitalization medication records, symptom scoring system records and historical hospital infection report data;

[0057] The data preprocessing unit is used to perform normalization processing on the initial data set to obtain a normalized data set, and the normalization processing step includes: classifying the character type of the initial data set to obtain text data and digital data, deleting HTML tags, stop words and non-alphanumeric characters in the text data, and performing uppercase and lowercase conversion and stem extraction, separating thousands of digital data, unifying decimal points and time formats, and completing normalization processing to obtain a normalized data set;

[0058] The text data includes Chinese characters, alphabetic characters, non-alphanumeric characters and punctuation marks in the initial data set;

[0059] The digital data includes integer characters, floating point characters, fixed point characters and Boolean characters.

[0060] In this embodiment, the graph construction module includes a word segmentation unit, an entity extraction unit and a graph generation unit;

[0061] The word segmentation unit is used to perform word segmentation processing on the standard data set to obtain a word segmentation data set, and the word segmentation processing step includes: performing word segmentation on the standard data set based on a dictionary, performing character matching through a preset dictionary library, and extracting the matching characters into the word segmentation data set;

[0062] Based on statistics, the standard data set is segmented twice, the combination probabilities of the remaining characters in the standard data set after the first word segmentation are counted, the combination probabilities are sorted to obtain the maximum combination probability, and characters are extracted from the standard data set based on the maximum combination probability, and the extracted characters are extracted into the word segmentation data set.

[0063] In this embodiment, the entity extraction unit is used to perform feature extraction and entity attribute labeling on a word segmentation data set to obtain an entity data set, including: vectorizing the word segmentation data set, vectorizing the word segmentation data set through character encoding of a bag of words model to obtain a word segmentation vector set, weighted calculation of the word segmentation vector set based on a parallel subspace of multi-head attention to obtain a weighted vector set, reducing the dimension and linearly combining the weighted vector set through a principal component analysis method to obtain a reduced dimension vector set, extracting features from the reduced dimension vector set through Bi-LSTM to obtain positive dependency features and reverse dependency features, adding the positive dependency features and the reverse dependency features to obtain a word segmentation feature data set, using a hidden Markov model to calculate state transition probabilities and label entity attribute labels on the word segmentation feature data set, and completing entity classification and decoding through a maximum likelihood estimation algorithm to obtain an entity data set;

[0064] The entity attribute labels include infection index labels, patient identity labels, infection site labels, patient medical history labels, disease score labels, infection department labels, treatment method labels, hospitalization drug labels, ward status labels, and historical pathogen labels;

[0065] The infection indicator labels include albumin, prealbumin, anemia, antibiotic use, and antibiotic use level;

[0066] The patient identification tag includes the patient's name, gender, and age group;

[0067] The infection site labels include lower respiratory tract, upper respiratory tract, gastrointestinal tract, intra-abdominal tissue, urinary tract infection, respiratory system, blood system, urinary system, digestive system, skin and soft tissue, reproductive system, ear, nose and throat, and oral cavity;

[0068] The patient's medical history labels include: history of diabetes, hypertension, hyperlipidemia, atrial fibrillation, cancer, smoking, drinking, blood transfusion, and stroke;

[0069] The disease scoring labels include ADL score, mRS score, GCS score, NIHSS score, APACHEⅡ score and NRS2002 score;

[0070] The infection department labels include critical care medicine, neurosurgery, orthopedic surgery, general surgery, cardiothoracic surgery, urology, otolaryngology, infectious diseases, oncology, respiratory medicine, cardiology, neurology, gastroenterology, endocrinology, pediatrics, emergency department, and traditional Chinese medicine;

[0071] The treatment labels include antimicrobial therapy, quality control circle therapy, thrombolytic therapy, deep venous catheterization, urinary catheterization, central venous catheterization, use of ventilator, and routine care.

[0072] The inpatient medication label includes dehydration drugs, antimicrobial drugs, anticoagulants, antiplatelet drugs, and glucocorticoids;

[0073] The ward status labels include occupancy rate, number of patients receiving antibiotics, antibiotic use rate, number of patients sent for inspection before antibiotic treatment, inspection rate before antibiotic treatment, ward closure status, and frequency of hospital infection;

[0074] Historical pathogen labels include Klebsiella pneumoniae, Escherichia coli, Staphylococcus aureus, Acinetobacter baumannii, Pseudomonas aeruginosa, Enterobacter cloacae, and fungi.

[0075] In this embodiment, the graph generation unit is used to construct a first knowledge graph and a second knowledge graph, generate a patient-infection attribute field based on an entity containing an infection indicator label, a patient identity label, an infection site label, a patient medical history label, and a disease score label in an entity data set, and extract a relationship from the patient-infection attribute field through a SemRep model to obtain a patient-infection entity relationship, wherein the patient-infection entity relationship includes: a first hierarchical relationship, a first part-whole relationship, a first synonym relationship, a first near-synonymous relationship, a first antonym relationship, a first spelling variant relationship, a first abbreviation and full name relationship, a first etiology relationship, a first clinical manifestation relationship, and a first treatment relationship;

[0076] Generate hospital-infection attribute fields based on entities containing infection indicator labels, infection department labels, treatment method labels, hospitalization drug labels, ward status labels, and historical pathogen labels in the entity data set, and extract relationships from the hospital-infection attribute fields through the SemRep model to obtain hospital-infection entity relationships, wherein the hospital-infection entity relationships include: second-level relationships, second-part-whole relationships, second synonym relationships, second near-synonymous relationships, second antonymous relationships, second spelling variant relationships, second abbreviation and full name relationships, second etiology relationships, second clinical manifestation relationships, and second treatment relationships;

[0077] Importing the patient-infection entity relationship, the entity data set and the corresponding entity attribute labels into the Neo4j database, mapping each entity in the entity data set into a first graph node set, generating a first edge relationship set through the entity attribute labels and the patient-infection entity relationship, and establishing a first knowledge graph based on the first graph nodes and the first edge relationships;

[0078] The hospital-infection entity relationship, entity data set and corresponding entity attribute labels are imported into the Neo4j database, each entity in the entity data set is mapped into a second graph node set, a second edge relationship set is generated through the entity attribute labels and the hospital-infection entity relationship, and a second knowledge graph is established based on the second graph node set and the second edge relationship set.

[0079] In this embodiment, the infection risk prediction module includes a graph-to-feature value conversion unit and a risk analysis unit. The graph-to-feature value conversion unit is used to perform graph-to-feature value conversion on the first knowledge graph and the second knowledge graph to obtain a first feature vector set and a second feature vector set. The logic of the graph-to-feature value conversion includes: performing graph embedding processing on the first knowledge graph and the second knowledge graph respectively through the TransE algorithm, embedding the corresponding graph nodes and edge relationships into the low-dimensional vector space to obtain the first feature vector set and the second feature vector set respectively, and inputting the first feature vector set and the second feature vector set into the risk analysis unit.

[0080] In this embodiment, the risk analysis unit is used to analyze and evaluate the infection risk of the target hospital and output the risk prediction result;

[0081] The processing logic of the analysis and evaluation includes: completing the infection risk analysis through a pre-trained multi-layer perceptron model, the structure of the multi-layer perceptron model includes two parallel branches and an output layer trunk, the first branch includes a left input layer and a left hidden layer, and the second branch includes a right input layer and a right hidden layer;

[0082] The number of the left input layer is one, the number of the left hidden layer is M, the number of the right input layer is one, and the number of the right hidden layer is N;

[0083] The left input layer data set is obtained by receiving the data in the first feature vector set by each input neuron in the left input layer respectively, and the left input layer data set is subjected to a first iterative calculation by the left hidden layer to obtain a first deep feature value, wherein the logic of the first iterative calculation includes: calculating the input data by the weight vector and the bias term corresponding to the layer of left hidden layer to obtain a layer of left weighted feature values, performing a nonlinear transformation on the layer of left weighted feature values ​​by the first ReLU activation function to obtain a layer of left hidden layer output, using the layer of left hidden layer output as the input of the second layer of left hidden layer and repeatedly calculating to obtain the second layer of left hidden layer output, and iteratively calculating until the first deep feature value of the M layers of left hidden layer output is obtained;

[0084] The right input layer data set is obtained by receiving the data in the second eigenvector set through each input neuron in the right input layer respectively, and the second deep eigenvalue is obtained by performing a second iterative calculation on the right input layer data set through the right hidden layer, wherein the logic of the second iterative calculation includes: calculating the input data through the weight vector and the bias term corresponding to the first right hidden layer to obtain a layer of right weighted eigenvalues, performing a nonlinear transformation on the first right weighted eigenvalue through the second ReLU activation function to obtain a layer of right hidden layer output, using the first right hidden layer output as the input of the second right hidden layer and repeatedly calculating to obtain the second right hidden layer output, and iteratively calculating until the second deep eigenvalue of the N-layer right hidden layer output is obtained;

[0085] The first deep feature value and the second deep feature value are linearly transformed through the attention mechanism in the output layer to obtain the fused feature result, and the fused feature result is mapped to the output layer classification space. The mapping value in the output layer classification space is normalized by the softmax function to obtain the classification probability corresponding to each entity, and the classification probability corresponding to each entity is sorted from high to low, and the maximum classification probability is selected as the risk prediction result corresponding to each entity;

[0086] The classification probability includes low risk probability, medium risk probability and high risk probability;

[0087] The risk prediction results include low risk, medium risk and high risk.

[0088] In this embodiment, the interactive terminal module includes a display unit and an interactive unit;

[0089] The interactive unit is used to receive and execute user input instructions and feed back operation results to the user, wherein the input instructions include data import and export, system parameter setting, graph query and risk prediction result query;

[0090] The display unit is used to provide the user with an operation interface and visualization results, and the visualization results include: data display, system status display, graph display and risk prediction result display.

[0091] Among them, the entity attribute labels are divided into two categories according to the patient and infection situation, and the hospital as a whole and the infection situation, and then the first knowledge graph and the second knowledge graph are established respectively. The multi-layer perceptron model is conducive to calculating the deep entity relationship from the perspectives of patient-infection and hospital-infection, which is conducive to efficient processing of big data and comprehensive analysis of potential infection risk relationships. It does not rely on biochemical means to obtain real-time test data, which reduces the complexity of obtaining data and improves the accuracy of infection analysis. The output result is the infection risk prediction result, which is forward-looking.

[0092] Among them, the training logic of the multi-layer perceptron model is: according to the actual infection situation, the knowledge graph sample data is labeled with low-risk labels, medium-risk labels and high-risk labels, and the knowledge graph sample data is converted from the graph to the feature value as described in the present embodiment to obtain the feature vector set as the input data of the model, and the input data is divided into a training set and a test set according to a ratio of 7:3, and the test set data is input into the initial multi-layer perceptron model, and the weights and bias items of the hidden layer are continuously adjusted through the back propagation algorithm. The number of neurons in the input layer, hidden layer and output layer, the number of hidden layers and the weights and bias items of the attention mechanism of the output layer are adjusted according to cross-validation through the test set data, and the accuracy, recall rate and F1 value of the prediction results of the evaluation model based on different training set subsets are evaluated. The model parameters are continuously adjusted to minimize the prediction error to obtain the final trained multi-layer perceptron model.

[0093] It should be understood by those skilled in the art that embodiments of the present invention may provide methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program codes. Among them, the storage medium may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, disk or optical disk. These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0094] 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 preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A hospital infection risk analysis system based on big data analysis, characterized in that: include: Data collection module, graph construction module, infection risk prediction module and interactive terminal module; The data acquisition module includes a data reading unit and a data preprocessing unit, which are used to extract data from the system database of the target hospital to obtain an initial data set, and to normalize the initial data set to obtain a normalized data set; The graph construction module includes a word segmentation unit, an entity extraction unit and a graph generation unit, which are used to perform word segmentation processing on the standard data set to obtain a word segmentation data set, perform feature extraction and entity attribute labeling on the word segmentation data set to obtain an entity data set, and construct a first knowledge graph and a second knowledge graph; The infection risk prediction module includes a graph-to-feature value conversion unit and a risk analysis unit, which are used to perform graph-to-feature value conversion on the first knowledge graph and the second knowledge graph to obtain a first feature vector set and a second feature vector set, analyze and evaluate the infection risk of the target hospital through a multi-layer perceptron model, and output a risk prediction result; The interactive terminal module includes a display unit and an interactive unit, which are used to provide an operation interface and visual results to the user, receive and execute the user's input instructions and feed back the operation results to the user.

2. The hospital infection risk analysis system based on big data analysis according to claim 1, characterized in that: The data acquisition module includes a data reading unit and a data preprocessing unit. The data reading unit is used to extract data from the system database of the target hospital to obtain an initial data set, wherein the initial data set includes patient identity data, patient test data, financial system data, nursing records, hospitalization medication records, symptom scoring system records and historical hospital infection report data; The data preprocessing unit is used to perform normalization processing on the initial data set to obtain a normalized data set, and the normalization processing step includes: classifying the character type of the initial data set to obtain text data and digital data, deleting HTML tags, stop words and non-alphanumeric characters in the text data, and performing uppercase and lowercase conversion and stem extraction, separating thousands of digital data, unifying decimal points and time formats, and completing normalization processing to obtain a normalized data set; The text data includes Chinese characters, alphabetic characters, non-alphanumeric characters and punctuation marks in the initial data set; The digital data includes integer characters, floating point characters, fixed point characters and Boolean characters.

3. The hospital infection risk analysis system based on big data analysis according to claim 1, characterized in that: The graph construction module includes a word segmentation unit, an entity extraction unit and a graph generation unit; The word segmentation unit is used to perform word segmentation processing on the standard data set to obtain a word segmentation data set, and the word segmentation processing step includes: performing word segmentation on the standard data set based on a dictionary, performing character matching through a preset dictionary library, and extracting the matching characters into the word segmentation data set; Based on statistics, the standard data set is segmented twice, the combination probabilities of the remaining characters in the standard data set after the first word segmentation are counted, the combination probabilities are sorted to obtain the maximum combination probability, and characters are extracted from the standard data set based on the maximum combination probability, and the extracted characters are extracted into the word segmentation data set.

4. The hospital infection risk analysis system based on big data analysis according to claim 1, characterized in that: The entity extraction unit is used to perform feature extraction and entity attribute labeling on a word segmentation data set to obtain an entity data set, including: vectorizing the word segmentation data set, vectorizing the word segmentation data set through character encoding of a bag-of-words model to obtain a word segmentation vector set, performing weighted calculation on the word segmentation vector set based on a parallel subspace of multi-head attention to obtain a weighted vector set, performing dimensionality reduction and linear combination processing on the weighted vector set through a principal component analysis method to obtain a reduced dimensionality vector set, performing feature extraction on the reduced dimensionality vector set through a Bi-LSTM to obtain positive dependency features and reverse dependency features, performing addition operation on the positive dependency features and reverse dependency features to obtain a word segmentation feature data set, performing state transition probability calculation and entity attribute labeling on the word segmentation feature data set using a hidden Markov model, and completing entity classification and decoding through a maximum likelihood estimation algorithm to obtain an entity data set; The entity attribute labels include infection index labels, patient identity labels, infection site labels, patient medical history labels, disease score labels, infection department labels, treatment method labels, hospitalization drug labels, ward status labels, and historical pathogen labels; The infection indicator labels include albumin, prealbumin, anemia, antibiotic use, and antibiotic use level; The patient identification tag includes the patient's name, gender, and age group; The infection site labels include lower respiratory tract, upper respiratory tract, gastrointestinal tract, intra-abdominal tissue, urinary tract infection, respiratory system, blood system, urinary system, digestive system, skin and soft tissue, reproductive system, ear, nose and throat, and oral cavity; The patient's medical history labels include: history of diabetes, hypertension, hyperlipidemia, atrial fibrillation, cancer, smoking, drinking, blood transfusion, and stroke; The disease scoring labels include ADL score, mRS score, GCS score, NIHSS score, APACHEⅡ score and NRS2002 score; The infection department labels include critical care medicine, neurosurgery, orthopedic surgery, general surgery, cardiothoracic surgery, urology, otolaryngology, infectious diseases, oncology, respiratory medicine, cardiology, neurology, gastroenterology, endocrinology, pediatrics, emergency department, and traditional Chinese medicine; The treatment labels include antimicrobial therapy, quality control circle therapy, thrombolytic therapy, deep venous catheterization, urinary catheterization, central venous catheterization, use of ventilator and routine care The inpatient medication label includes dehydration drugs, antimicrobial drugs, anticoagulants, antiplatelet drugs, and glucocorticoids; The ward status labels include occupancy rate, number of patients receiving antibiotics, antibiotic use rate, number of patients sent for inspection before antibiotic treatment, inspection rate before antibiotic treatment, ward closure status, and frequency of hospital infection; Historical pathogen labels include Klebsiella pneumoniae, Escherichia coli, Staphylococcus aureus, Acinetobacter baumannii, Pseudomonas aeruginosa, Enterobacter cloacae, and fungi.

5. The hospital infection risk analysis system based on big data analysis according to claim 1, characterized in that: The graph generation unit is used to construct a first knowledge graph and a second knowledge graph, generate a patient-infection attribute field based on entities containing infection indicator labels, patient identity labels, infection site labels, patient medical history labels, and disease score labels in the entity data set, and extract the relationship of the patient-infection attribute field through the SemRep model to obtain a patient-infection entity relationship, wherein the patient-infection entity relationship includes: a first hierarchical relationship, a first part-whole relationship, a first synonym relationship, a first near-synonymous relationship, a first antonym relationship, a first spelling variant relationship, a first abbreviation and full name relationship, a first etiology relationship, a first clinical manifestation relationship, and a first treatment relationship; Generate hospital-infection attribute fields based on entities containing infection indicator labels, infection department labels, treatment method labels, hospitalization drug labels, ward status labels, and historical pathogen labels in the entity data set, and extract relationships from the hospital-infection attribute fields through the SemRep model to obtain hospital-infection entity relationships, wherein the hospital-infection entity relationships include: second-level relationships, second-part-whole relationships, second synonym relationships, second near-synonymous relationships, second antonymous relationships, second spelling variant relationships, second abbreviation and full name relationships, second etiology relationships, second clinical manifestation relationships, and second treatment relationships; Importing the patient-infection entity relationship, the entity data set and the corresponding entity attribute labels into the Neo4j database, mapping each entity in the entity data set into a first graph node set, generating a first edge relationship set through the entity attribute labels and the patient-infection entity relationship, and establishing a first knowledge graph based on the first graph nodes and the first edge relationships; The hospital-infection entity relationship, entity data set and corresponding entity attribute labels are imported into the Neo4j database, each entity in the entity data set is mapped into a second graph node set, a second edge relationship set is generated through the entity attribute labels and the hospital-infection entity relationship, and a second knowledge graph is established based on the second graph node set and the second edge relationship set.

6. The hospital infection risk analysis system based on big data analysis according to claim 1, characterized in that: The infection risk prediction module includes a graph-to-feature value conversion unit and a risk analysis unit. The graph-to-feature value conversion unit is used to perform graph-to-feature value conversion on the first knowledge graph and the second knowledge graph to obtain a first feature vector set and a second feature vector set. The logic of the graph-to-feature value conversion includes: performing graph embedding processing on the first knowledge graph and the second knowledge graph respectively through the TransE algorithm, embedding the corresponding graph nodes and edge relationships into the low-dimensional vector space to obtain the first feature vector set and the second feature vector set respectively, and inputting the first feature vector set and the second feature vector set into the risk analysis unit.

7. The hospital infection risk analysis system based on big data analysis according to claim 1, characterized in that: The risk analysis unit is used to analyze and evaluate the infection risk of the target hospital and output the risk prediction result; The processing logic of the analysis and evaluation includes: completing the infection risk analysis through a pre-trained multi-layer perceptron model, the structure of the multi-layer perceptron model includes two parallel branches and an output layer trunk, the first branch includes a left input layer and a left hidden layer, and the second branch includes a right input layer and a right hidden layer; The number of the left input layer is one, the number of the left hidden layer is M, the number of the right input layer is one, and the number of the right hidden layer is N; The left input layer data set is obtained by receiving the data in the first feature vector set by each input neuron in the left input layer respectively, and the left input layer data set is subjected to a first iterative calculation by the left hidden layer to obtain a first deep feature value, wherein the logic of the first iterative calculation includes: calculating the input data by the weight vector and the bias term corresponding to the layer of left hidden layer to obtain a layer of left weighted feature values, performing a nonlinear transformation on the layer of left weighted feature values ​​by the first ReLU activation function to obtain a layer of left hidden layer output, using the layer of left hidden layer output as the input of the second layer of left hidden layer and repeatedly calculating to obtain the second layer of left hidden layer output, and iteratively calculating until the first deep feature value of the M layers of left hidden layer output is obtained; The right input layer data set is obtained by receiving the data in the second eigenvector set through each input neuron in the right input layer respectively, and the second deep eigenvalue is obtained by performing a second iterative calculation on the right input layer data set through the right hidden layer, wherein the logic of the second iterative calculation includes: calculating the input data through the weight vector and the bias term corresponding to the first right hidden layer to obtain a layer of right weighted eigenvalues, performing a nonlinear transformation on the first right weighted eigenvalue through the second ReLU activation function to obtain a layer of right hidden layer output, using the first right hidden layer output as the input of the second right hidden layer and repeatedly calculating to obtain the second right hidden layer output, and iteratively calculating until the second deep eigenvalue of the N-layer right hidden layer output is obtained; The first deep feature value and the second deep feature value are linearly transformed through the attention mechanism in the output layer to obtain the fused feature result, and the fused feature result is mapped to the output layer classification space. The mapping value in the output layer classification space is normalized by the softmax function to obtain the classification probability corresponding to each entity, and the classification probability corresponding to each entity is sorted from high to low, and the maximum classification probability is selected as the risk prediction result corresponding to each entity; The classification probability includes low risk probability, medium risk probability and high risk probability; The risk prediction results include low risk, medium risk and high risk.

8. The hospital infection risk analysis system based on big data analysis according to claim 1, characterized in that: The interactive terminal module includes a display unit and an interactive unit; The interactive unit is used to receive and execute user input instructions and feed back operation results to the user, wherein the input instructions include data import and export, system parameter setting, graph query and risk prediction result query; The display unit is used to provide the user with an operation interface and visualization results, and the visualization results include: data display, system status display, graph display and risk prediction result display.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the hospital infection risk analysis system based on big data analysis as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the hospital infection risk analysis system based on big data analysis according to any one of claims 1 to 8 is implemented.

Images