Medical consumable audit-oriented monitoring evaluation method and system and storage medium

By building a basic consumables database and using a Bi-LSTM model with deep learning and adversarial training, identifying the abnormal usage points and patterns of medical consumables, the problem of insufficient accuracy and robustness of data processing of traditional Chinese medicine consumables in the existing technology is solved, and the problem of insufficient data processing of traditional Chinese medicine consumables is realized. Comprehensive monitoring, evaluation and audit of medical consumables.

CN119004318APending Publication Date: 2024-11-22ANHUI PROVINCIAL HOSPITAL

Patent Information

Application Number
CN202411066619.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-05
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The BERT model of the existing Simhash-BERT network is insufficient in its accuracy and robustness when processing data such as procurement, inbound, outbound, use, collection, settlement and payment of medical consumables, mainly due to its dependence on high-quality training data and Faced with the complex situation where information is scattered in different departments and systems, the data quality is not high and complicated.

Method used

By obtaining medical consumables collection information based on the hospital medical information system, standardizing processing and building a basic consumables database, pre-constructing a self-learning early warning model based on deep learning and adversarial training, using the Bi-LSTM model of the conditional adversarial network to identify abnormal usage points and patterns. Generate identification detection results, and judge whether the preset risk threshold is exceeded through the audit knowledge graph, and trigger the early warning mechanism to generate an audit report.

Benefits of technology

It significantly improves the accuracy and completeness of medical consumables' full-cycle data, overcomes the problem of insufficient accuracy and robustness when processing complex sequence data, and is suitable for medical consumables auditing requirements, realizing the medical Comprehensive monitoring, evaluation and audit of consumables.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119004318A_ABST
    Figure CN119004318A_ABST
Patent Text Reader

Abstract

The invention discloses a medical consumable audit-oriented monitoring and evaluation method and system and a storage medium, and solves the problems that the performance of a BERT model of a Simhash-BERT network in an existing method depends on a large amount of high-quality training data to a great extent, and consumable data of which information is scattered in different departments and systems are poor in reliability and the like. The method comprises the steps of performing standardization processing on medical consumable collection information to obtain consumable standardized data, constructing a consumable basic database based on the consumable standardized data, and pre-constructing a self-learning early warning model in combination with deep learning and adversarial training; the self-learning early warning model identifies abnormal use points and abnormal use modes of the medical consumables and generates an identification detection result; according to the method, the self-learning early warning model is pre-constructed in combination with deep learning and adversarial training, the self-learning early warning model is a Bi-LSTM model based on the conditional adversarial network in combination with deep learning and adversarial training, and the accuracy and robustness of the Bi-LSTM network in processing complex sequence data are effectively improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of medical consumables management, and in particular relates to a monitoring and evaluation method, system and storage medium for medical consumables auditing. Background Art

[0002] Medical consumables are an important material basis for medical institutions to carry out medical business, and are also auxiliary tools and important carriers for clinical diagnosis and treatment of new technologies. Medical consumables are widely used in diagnosis, treatment, nursing and rehabilitation, and play a very important role in the diagnosis and treatment of patients' diseases and improving medical work efficiency. The internal audit of medical consumables in hospitals is a systematic process, which aims to ensure that the procurement, use, inventory management and charging of medical consumables comply with regulations and standards, so as to improve the economic benefits and service quality of hospitals.

[0003] Among them, Chinese patent CN114416966B discloses a method for analyzing the reasonable use of medical consumables based on the Simhash-BERT network. This method designs a word segmentation network for weight optimization. The word segmentation network mainly uses a triplet network to perform a difficult example mining strategy on the text in the document, and constructs a method for selecting positive or negative negative example pairs from the minimum training batch, and further optimizes the BERT network, thereby realizing the analysis of the reasonable use of medical consumables.

[0004] However, the performance of the BERT model of the Simhash-BERT network in the above-mentioned existing methods depends to a large extent on a large amount of high-quality training data. Faced with low-quality and complex data such as consumables procurement, warehousing, outbound delivery, usage, collection, settlement and payment, which are scattered in different departments and systems, the accuracy and robustness of the model are insufficient. To address the above problems, we proposed a monitoring and evaluation method, system and storage medium for medical consumables auditing. Summary of the invention

[0005] The purpose of the present invention is to address the shortcomings of the prior art and provide a monitoring and evaluation method, system and storage medium for medical consumables auditing, which solves the problem that the performance of the BERT model of the Simhash-BERT network in the existing method depends largely on a large amount of high-quality training data, and the model lacks accuracy and robustness in the face of low-quality and complex data such as consumables procurement, warehousing, delivery, use, collection, settlement and payment scattered in different departments and systems.

[0006] The present invention is implemented in this way: a monitoring and evaluation method for medical consumables audit, the monitoring and evaluation method for medical consumables audit comprising:

[0007] Obtaining medical consumables collection information based on the hospital medical information system, standardizing the medical consumables collection information to obtain consumables standardized data, and building a consumables basic database based on the consumables standardized data;

[0008] Capture the standardized data of consumables in the consumables basic database, divide the standardized data into training set and test set in a ratio of 3:1, assign abnormal rules and heuristic strategies to the training set and test set, and pre-build a self-learning early warning model by combining deep learning and adversarial training;

[0009] Obtain standardized data on consumables and bring the standardized data into the self-learning early warning model. The self-learning early warning model identifies abnormal usage points and abnormal usage patterns of medical consumables and generates identification and detection results.

[0010] Construct an audit knowledge graph based on risk warning and audit rules, obtain identification and detection results, identify the identification and detection results with the audit knowledge graph, and determine whether the identification and detection results exceed a preset risk threshold;

[0011] If the preset risk threshold is exceeded, the audit knowledge graph warning mechanism is triggered, and the audit knowledge graph automatically generates warning information, and based on the warning information, conducts a first audit on the standardized data of consumables and generates a first audit report;

[0012] If the preset risk threshold is not exceeded, a second audit is conducted on the standardized data of consumables and a second audit report is generated.

[0013] Obtain identification test results, the first audit report, and the second audit report, and use visualization tools to visualize the identification test results, the first audit report, and the second audit report.

[0014] Preferably, the standardized processing of the medical consumables collection information specifically includes:

[0015] The hospital medical information system responds to the data capture instruction and uploads the collected medical consumables information, wherein the hospital medical information system includes the hospital information system, the electronic health record system, the laboratory information system, the hospital logistics management system, and the hospital efficient operation management system, and the medical consumables collection information includes the name, specification, quantity, price, supplier, procurement, warehousing, outbound, use, and payment data information of the medical consumables;

[0016] Loading the collected medical consumables collection information, and performing data cleaning processing on the medical consumables collection information, wherein missing value processing data is used when performing data cleaning processing on the medical consumables collection information;

[0017] The medical consumables collection information is normalized using Z-Score Normalization, and then the open source tool Playment is used to label the normalized medical consumables collection information;

[0018] A Fourier function is preset, and continuous Fourier transform is used to convert the medical consumables collection information into consumables standardized data of visualized two-dimensional time series, wherein the consumables standardized data includes visualized two-dimensional time series marks and attribute marks;

[0019] Among them, the preset Fourier function during continuous Fourier transform is expressed as:

[0020]

[0021] Among them, δ is the scale parameter of medical consumables collection information, a is the data scaling parameter, n is the sample size of consumables standardized data, and f(x) is the mother wavelet function;

[0022]

[0023] Perform inner product between the preset Fourier function and the collected information of medical consumables to obtain the visualized two-dimensional time series of consumables standardized data;

[0024]

[0025] Among them, x τ To visualize the two-dimensional time series of standardized consumable data, ψ(x) is the Fourier transform coefficient, and b is the attribute labeling weight corresponding to the standardized consumable data.

[0026] Preferably, the construction of a consumables basic database based on the standardized consumables data specifically includes:

[0027] Acquire standardized data of consumables, identify data features of the standardized data of consumables, and extract feature vectors based on the data features of the standardized data of consumables;

[0028] Analyze the dependencies and importance between the data features of the standardized data of consumables, determine the correlation between the data features, and design the data table and architecture tree of the consumables basic database based on the correlation, feature vectors, timing tags, and attribute tags;

[0029] Use SQL statements to fill the standardized data of consumables into the data table, and associate the data table with the architecture tree through index structure based on time series tags and attribute tags to generate a basic database of consumables.

[0030] Preferably, the method of pre-building a self-learning early warning model by combining deep learning and adversarial training specifically includes:

[0031] Obtain training sets and test sets, and build a Bi-LSTM model based on a conditional adversarial network. The Bi-LSTM model of the conditional adversarial network includes a condition generator, a data discriminator, a forward LSTM layer, a backward LSTM layer, and a self-attention recognition layer.

[0032] The training set is input into the Bi-LSTM model based on the conditional adversarial network, expressed as:

[0033]

[0034] Among them, the input training set is represented by x1,x2,...x i-1 ,x i , C(x) is the predicted output result, and i is the number of samples;

[0035] The conditional generator introduces abnormal rules and heuristic strategies to form a conditional adversarial network. The data discriminator is used to improve the conditional adversarial network. The predicted output results are used as input to train the conditional adversarial network. The objective function form of the conditional adversarial network is:

[0036]

[0037] Among them, H(C) is the conditional generator function of the conditional adversarial network, L(C) is the data discriminator function of the conditional adversarial network, and C data (x) is the independent probability distribution of the predicted output, CC data (x) means that the data C(x) follows an independent probability distribution;

[0038] The self-attention recognition layer obtains the output result of the conditional adversarial network. The self-attention recognition result is expressed as:

[0039]

[0040] Among them, F(C) is the self-attention recognition result, n is the number of convolutional layers of the self-attention recognition layer, g is the self-attention expansion factor of the self-attention recognition layer, and x i-g Dilated convolution based on adversarial network for input training set;

[0041]

[0042] The backward LSTM layer extracts the hidden recognition results from the self-attention recognition results based on the swi sh function, and combines the hidden recognition results with the self-attention recognition results to generate the training set detection results, and judges whether the training set detection results meet the preset training accuracy. If they meet the preset training accuracy, the self-learning early warning model is output. If they do not meet the preset training accuracy, the first-order and second-order moments of the model are calculated based on the ADMA optimizer, and the model learning rate is adjusted. Return to the step to input the training set into the Bi-LSTM model based on the conditional adversarial network;

[0043] Obtain a test set, verify the recognition accuracy of the self-learning early warning model through the test set, and determine whether the recognition accuracy meets the preset recognition threshold. If it meets the preset recognition threshold, output the self-learning early warning model, and use the self-learning early warning model to identify abnormal usage points and abnormal usage patterns of medical consumables, and generate recognition detection results.

[0044] Preferably, the audit knowledge graph includes:

[0045] At least one set of audit trigger engines, the audit trigger engines are used to obtain identification detection results, and the trigger engines include an audit rule base, a risk threshold assessor, a command trigger, and an audit recorder;

[0046] Among them, the audit rule base is connected to the consumables basic database for communication, the audit rule base is used to traverse the audit rules and text data in the consumables basic database, the risk threshold assessor calculates and identifies whether the detection result exceeds the preset risk threshold based on the error back propagation algorithm, the instruction trigger is used to obtain the judgment result of the risk threshold assessor, and trigger the first audit instruction or the second audit instruction based on the judgment result, and the audit recorder is used to store the audit rule base, the risk threshold assessor, and the instruction trigger operation log;

[0047] An audit search engine is used to perform a first audit or a second audit on the standardized data of consumables, and the audit search engine selects audit rules and audit parameters applicable to the first audit and the second audit in the consumables basic database, writes audit operations and audit reports based on the CRDT algorithm, and verifies audit authority and audit credibility.

[0048] Preferably, the audit knowledge graph identifies the identification detection result, and the method for judging whether the identification detection result exceeds a preset risk threshold specifically includes:

[0049] Obtain identification and detection results, and calculate the risk value of the identification and detection results based on the error back propagation algorithm;

[0050] M(C)=w i f(C)+h c (8)

[0051] Among them, M(C) is the risk value of the identification test result, w i To identify the weight normalization coefficient of the detection result, h c is the offset corresponding to the recognition detection result, and f(C) is the input representation of the recognition detection result;

[0052]

[0053] The risk threshold is defined based on the audit rules and audit parameters. The risk threshold is expressed as:

[0054]

[0055] Among them, Y is the risk threshold, x τ To visualize the standardized data of consumables in a two-dimensional time series, λ is the risk coefficient of the audit parameter, and t is the manually defined threshold parameter.

[0056] The present invention also provides a monitoring and evaluation system for medical consumables auditing, and the monitoring and evaluation system for medical consumables auditing includes:

[0057] The information system integration module is composed of a data normalization unit and a basic database unit. The information system integration module obtains medical consumables collection information based on the hospital medical information system, standardizes the medical consumables collection information, obtains consumables standardized data, and builds a consumables basic database based on the consumables standardized data;

[0058] Early warning model training module: The early warning model training module is used to capture the standardized data of consumables in the consumables basic database, divide the standardized data into training set and test set in a ratio of 3:1, and assign abnormal rules and heuristic strategies to the training set and test set, and pre-build a self-learning early warning model by combining deep learning and adversarial training;

[0059] A test result generation module, which is used to bring the standardized data of consumables into the self-learning early warning model. The self-learning early warning model identifies abnormal use points and abnormal use patterns of medical consumables and generates recognition test results;

[0060] A report generation module, which constructs an audit knowledge graph based on risk warning and audit rules, obtains identification and detection results, identifies the identification and detection results with the audit knowledge graph, and determines whether the identification and detection results exceed a preset risk threshold. If the preset risk threshold is exceeded, the audit knowledge graph warning mechanism is triggered, and the audit knowledge graph automatically generates warning information, and performs a first audit on the standardized data of consumables based on the warning information to generate a first audit report; if the preset risk threshold is not exceeded, a second audit is performed on the standardized data of consumables to generate a second audit report;

[0061] The visualization presentation module is used to obtain the recognition detection results, the first audit report, and the second audit report, and use visualization tools to visually present the recognition detection results, the first audit report, and the second audit report.

[0062] The data normalization unit comprises:

[0063] An information system index module, the information system index module is used to generate a data capture instruction, and the hospital medical information system responds to the data capture instruction and uploads the collected medical consumables collection information;

[0064] A missing value processing module, which is used to load the collected medical consumables collection information and perform data cleaning on the medical consumables collection information, wherein the missing value processing data is used when performing data cleaning on the medical consumables collection information;

[0065] Data labeling module: The data labeling module is used to normalize the medical consumables collection information using Z-ScoreNormalization, and then use the open source tool Playment to attribute label the normalized medical consumables collection information;

[0066] The data visualization module is used to preset the Fourier function and use continuous Fourier transform to convert the medical consumables collection information into the consumables standardized data with visualized two-dimensional time series. The consumables standardized data includes visualized two-dimensional time series marks and attribute marks.

[0067] On the other hand, the present invention also provides a computer-readable storage medium, and when the computer program instructions are executed, the monitoring and evaluation method for medical consumables auditing is implemented.

[0068] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0069] In an embodiment of the present invention, by constructing a basic database of consumables, comprehensive collection and integration of full-cycle data of medical consumables can be achieved, which significantly improves the accuracy and completeness of the data, providing a solid foundation for subsequent self-learning early warning model analysis. At the same time, a self-learning early warning model is pre-constructed in combination with deep learning and adversarial training. The self-learning early warning model is a Bi-LSTM model based on a conditional adversarial network that combines deep learning and adversarial training. It overcomes the problem of insufficient accuracy and robustness of the prior art when processing complex sequence data, effectively improves the accuracy and robustness of the Bi-LSTM network when processing complex sequence data, and is suitable for the requirements of medical consumables audits for data integrity and accuracy.

[0070] In an embodiment of the present invention, when the medical consumables collection information is standardized, the medical consumables collection information is converted into visualized two-dimensional time series consumables standardized data by using a preset Fourier function and a continuous Fourier transform. On the one hand, it can integrate the multi-source heterogeneous data information of each independent link to achieve information transparency and resource sharing, and ensure that the data between each link can be seamlessly connected and updated in real time. On the other hand, the consumables standardized data includes visualized two-dimensional time series marks and attribute marks, which facilitates the construction of a basic database for consumables, and is also conducive to the visual presentation of identification detection results, the first audit report, and the second audit report.

[0071] In the embodiment of the present invention, the self-learning early warning model uses the Bi-LSTM model based on the conditional adversarial network as the initial model, so that the trained self-learning early warning model has the advantages of Bi-LSTM in being able to simultaneously capture the forward and backward information of the data, which helps to fully understand the contextual relationship in the data. At the same time, it overcomes the problem of insufficient accuracy and robustness of the prior art in processing complex sequence data, and effectively improves the accuracy and robustness of the Bi-LSTM network in processing complex sequence data. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a schematic diagram of the implementation process of the monitoring and evaluation method for medical consumables audit provided by the present invention.

[0073] Figure 2 It is a schematic diagram of the implementation process of the standardized processing method for medical consumables collection information provided by the present invention.

[0074] Figure 3 It is a schematic diagram of the implementation process of the method for constructing a consumable basic database based on consumable standardized data provided by the present invention.

[0075] Figure 4 It is a schematic diagram of the implementation flow of the method for pre-building a self-learning early warning model by combining deep learning and adversarial training provided by the present invention.

[0076] Figure 5 It is a schematic diagram of the implementation process of the method for using the audit knowledge graph provided by the present invention to identify the identification and detection results and determine whether the identification and detection results exceed a preset risk threshold.

[0077] Figure 6 It is a structural block diagram of the monitoring and evaluation system for medical consumables audit provided by the present invention. DETAILED DESCRIPTION

[0078] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of this application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0079] The performance of the BERT model of the Simhash-BERT network in the existing methods depends largely on a large amount of high-quality training data. However, in the face of consumables procurement, warehousing, outbound, use, collection, settlement and payment data scattered in different departments and systems, the accuracy and robustness of the model are insufficient. In response to the above problems, we proposed a monitoring and evaluation method, system and storage medium for medical consumables auditing. The monitoring and evaluation method for medical consumables auditing includes obtaining medical consumables collection information based on the hospital medical information system, standardizing the medical consumables collection information to obtain consumables standardized data, building a consumables basic database based on the consumables standardized data, combining deep learning and adversarial training to pre-build a self-learning early warning model, bringing the consumables standardized data into the self-learning early warning model, the self-learning early warning model identifies abnormal use points and abnormal use patterns of medical consumables, generates recognition detection results, constructs an audit knowledge graph based on risk warning and audit rules, obtains recognition detection results, the audit knowledge graph identifies the recognition detection results, and determines whether the recognition detection results exceed the preset risk threshold. In an embodiment of the present invention, by constructing a basic database of consumables, comprehensive collection and integration of full-cycle data of medical consumables can be achieved, which significantly improves the accuracy and completeness of the data, providing a solid foundation for subsequent self-learning early warning model analysis. At the same time, a self-learning early warning model is pre-constructed in combination with deep learning and adversarial training. The self-learning early warning model is a Bi-LSTM model based on a conditional adversarial network that combines deep learning and adversarial training. It overcomes the problem of insufficient accuracy and robustness of the prior art when processing complex sequence data, effectively improves the accuracy and robustness of the Bi-LSTM network when processing complex sequence data, and is suitable for the requirements of medical consumables audits for data integrity and accuracy.

[0080] Example 1

[0081] The embodiment of the present invention provides a monitoring and evaluation method for auditing medical consumables. Figure 1 The following is a schematic diagram of the implementation process of the monitoring and evaluation method for medical consumables auditing, wherein the monitoring and evaluation method for medical consumables auditing specifically includes:

[0082] Step S10, obtaining medical consumables collection information based on the hospital medical information system, standardizing the medical consumables collection information to obtain consumables standardized data, and building a consumables basic database based on the consumables standardized data;

[0083] Among them, in step S10 of the embodiment of the present invention, the hospital medical information system includes but is not limited to the hospital information system (HIA), the electronic health record system (EHR), the laboratory information system (LIS), the hospital logistics management system (SPD), and the hospital efficient operation management system (OES). When establishing a connection with the hospital medical information system, a connection index including the database name, port name, user name, password, and authority is created, and then the connect method is called using the created connection index to connect to the database of the hospital medical information system, and operations such as addition, deletion, modification, and query are performed through the query method.

[0084] Step S20, grabbing the standardized data of consumables in the consumables basic database, dividing the standardized data of consumables into a training set and a test set at a ratio of 3:1, and assigning abnormal rules and heuristic strategies to the training set and the test set, combining deep learning and adversarial training to pre-build a self-learning early warning model;

[0085] Step S30, obtaining standardized consumables data, bringing the standardized consumables data into the self-learning early warning model, the self-learning early warning model identifies abnormal use points and abnormal use patterns of medical consumables, and generates recognition detection results;

[0086] It should be noted that the identification and detection results include information on abnormal usage points of medical consumables and abnormal usage trend curves of medical consumables, and the information on abnormal usage points of medical consumables includes but is not limited to consumable waste points, excess inventory points, price fluctuation points, and abnormal payment points.

[0087] Step S40, constructing an audit knowledge graph based on risk warning and audit rules, obtaining identification detection results, and the audit knowledge graph identifies the identification detection results;

[0088] Step S50, determining whether the identification detection result exceeds a preset risk threshold;

[0089] Step S60: If the preset risk threshold is exceeded, the audit knowledge graph warning mechanism is triggered, the audit knowledge graph automatically generates warning information, and performs a first audit on the standardized data of consumables based on the warning information to generate a first audit report;

[0090] It should be noted that when the audit knowledge graph automatically generates warning information, such as insufficient inventory or abnormal usage of consumables, it will immediately automatically generate warning information and automatically send it to relevant departments and personnel through email, SMS or system notification, etc. In this way, relevant departments can respond quickly and take necessary measures, such as replenishing inventory or investigating abnormal usage.

[0091] Step S70: If the preset risk threshold is not exceeded, a second audit is performed on the standardized consumables data to generate a second audit report.

[0092] In this embodiment, the first audit is set as a routine audit, and the contents of the routine audit include compliance of medical consumables procurement process, price rationality, and record integrity audit, compliance of consumables use process standards, effect evaluation, and record accuracy audit, rationality of consumables inventory levels, consistency of inventory records, as well as safety and custody audit, and compliance and integrity audit of consumables scrapping process. The second audit generates a comprehensive audit report and specific improvement suggestions based on the early warning information. The second audit also includes the audit content of the first audit, thereby strengthening the data basis for management decision-making.

[0093] Step S80, obtaining the recognition detection results, the first audit report, and the second audit report, and visually presenting the recognition detection results, the first audit report, and the second audit report using a visualization tool.

[0094] In this embodiment, the visualization tools may be Gephi, Echarts, or Wolfram|Alpha. The visualization tools can transform complex analysis results and audit intelligence into intuitive and easy-to-understand visual information, ensuring that management can quickly grasp key data and make wise decisions.

[0095] In an embodiment of the present invention, by constructing a basic database of consumables, comprehensive collection and integration of full-cycle data of medical consumables can be achieved, which significantly improves the accuracy and completeness of the data, providing a solid foundation for subsequent self-learning early warning model analysis. At the same time, a self-learning early warning model is pre-constructed in combination with deep learning and adversarial training. The self-learning early warning model is a Bi-LSTM model based on a conditional adversarial network that combines deep learning and adversarial training. It overcomes the problem of insufficient accuracy and robustness of the prior art when processing complex sequence data, effectively improves the accuracy and robustness of the Bi-LSTM network when processing complex sequence data, and is suitable for the requirements of medical consumables audits for data integrity and accuracy.

[0096] The embodiment of the present invention provides a method for standardizing the collection information of medical consumables. Figure 2 The figure shows a schematic diagram of the implementation process of the standardized processing method for the medical consumables collection information, and the standardized processing method for the medical consumables collection information specifically includes:

[0097] Step S101, the hospital medical information system responds to the data capture instruction and uploads the collected medical consumables collection information, wherein the hospital medical information system includes the hospital information system, the electronic health record system, the laboratory information system, the hospital logistics management system, and the hospital efficient operation management system, and the medical consumables collection information includes the name, specification, quantity, price, supplier, procurement, warehousing, outbound, use, and payment data information of the medical consumables.

[0098] Step S102, loading the collected medical consumables collection information, and performing data cleaning processing on the medical consumables collection information, wherein missing value processing data is used when performing data cleaning processing on the medical consumables collection information;

[0099] In this embodiment, data cleaning processing of medical consumables collection information includes but is not limited to simply removing records containing missing values, using mean, median, mode or other statistical values ​​to fill missing values, and determining the cleaning processing method of data that deviates from the mean range by calculating the mean and standard deviation of the data.

[0100] Step S103, Z-Score Normalization is used to normalize the medical consumables collection information, and then the open source tool Platform is used to attribute-mark the normalized medical consumables collection information;

[0101] In this embodiment, Z-Score Normalization is used to normalize the medical consumables collection information, which converts the data into a distribution with a mean of 0 and a standard deviation of 1. This method is suitable for the case where the data obeys a normal distribution, and has little impact on outliers, so it will not affect outlier detection.

[0102] Step S104, presetting a Fourier function, and using continuous Fourier transform to convert the medical consumables collection information into visualized two-dimensional time series consumables standardized data, wherein the consumables standardized data includes a visualized two-dimensional time series mark and an attribute mark;

[0103] Among them, the preset Fourier function during continuous Fourier transform is expressed as:

[0104]

[0105] Wherein, δ is the information scale parameter for collecting medical consumables. In this embodiment, the information scale parameter is set to 1 or 2, a is the data scaling parameter, the data scaling parameter is an integer between 1 and 3, n is the sample size of the consumables standardized data, and f(x) is the mother wavelet function;

[0106]

[0107] Among them, δ is the scale parameter of medical consumables collection information, a is the data scaling parameter, and n is the sample size of consumables standardized data. The preset Fourier function and the medical consumables collection information are inner-producted to obtain the consumables standardized data of visualized two-dimensional time series;

[0108]

[0109] Among them, x τ To visualize the two-dimensional time series of standardized consumable data, ψ(x) is the Fourier transform coefficient. In this embodiment, the Fourier transform coefficient is 0.1-0.5, and b is the attribute labeling weight corresponding to the standardized consumable data. In this embodiment, the attribute labeling weight is set to 0.1-0.8.

[0110] In an embodiment of the present invention, when the medical consumables collection information is standardized, the medical consumables collection information is converted into visualized two-dimensional time series consumables standardized data by using a preset Fourier function and a continuous Fourier transform. On the one hand, it can integrate the multi-source heterogeneous data information of each independent link to achieve information transparency and resource sharing, and ensure that the data between each link can be seamlessly connected and updated in real time. On the other hand, the consumables standardized data includes visualized two-dimensional time series marks and attribute marks, which facilitates the construction of a basic database for consumables, and is also conducive to the visual presentation of identification detection results, the first audit report, and the second audit report.

[0111] The embodiment of the present invention provides a method for constructing a consumables basic database based on consumables standardized data. Figure 3 The schematic diagram of the implementation process of the method for constructing a basic database of consumables based on standardized consumables data is shown. The method for constructing a basic database of consumables based on standardized consumables data specifically includes:

[0112] Step S201, acquiring standardized data of consumables, identifying data features of the standardized data of consumables, and extracting feature vectors based on the data features of the standardized data of consumables;

[0113] Step S202, analyzing the dependency and importance between the data features of the standardized data of consumables, determining the correlation between the data features, and designing the data table and architecture tree of the consumables basic database based on the correlation, feature vectors, time series tags, and attribute tags;

[0114] Step S203, using SQL statements to fill the consumables standardized data into the data table, and based on the time series mark and the attribute mark, the data table is associated with the architecture tree by index structure to generate a consumables basic database.

[0115] In an embodiment of the present invention, when constructing a basic database for consumables, SQL statements are used to fill the standardized data of consumables into a data table, and the data table is indexed and structurally associated with the architecture tree based on time series tags and attribute tags, thereby facilitating the self-learning early warning model to conduct in-depth analysis of information such as the procurement, use, flow and financial status of various types of consumables, providing scientific and accurate data support, thereby assisting managers to make more reasonable and efficient decisions.

[0116] The embodiment of the present invention provides a method for pre-building a self-learning early warning model by combining deep learning and adversarial training. Figure 4 A schematic diagram of the implementation process of a method for pre-building a self-learning early warning model by combining deep learning and adversarial training is shown. The method for pre-building a self-learning early warning model by combining deep learning and adversarial training specifically includes:

[0117] Step S301, obtaining a training set and a test set, and constructing a Bi-LSTM model based on a conditional adversarial network, wherein the Bi-LSTM model of the conditional adversarial network includes a condition generator, a data discriminator, a forward LSTM layer, a backward LSTM layer, and a self-attention recognition layer;

[0118] When training the Bi-LSTM model based on the conditional adversarial network, the training set and test set adopt the method of generating adversarial samples, crawling, cleaning and integrating data from the consumables basic database on a quarterly basis, generating a feature space through a multi-head attention mechanism network, and then performing a random walk on this feature space to generate adversarial samples. Each iteration adds a tiny, random perturbation to the current sample until a modification rate of more than 50% is reached as the termination condition of the random walk.

[0119] It should be noted that the self-learning early warning model in the embodiment of the present invention uses the Bi-LSTM model based on the conditional adversarial network as the initial model, so that the trained self-learning early warning model has the advantages of Bi-LSTM in being able to simultaneously capture the forward and backward information of the data, which helps to fully understand the contextual relationship in the data. At the same time, it overcomes the problem of insufficient accuracy and robustness of the prior art in processing complex sequence data, and effectively improves the accuracy and robustness of the Bi-LSTM network in processing complex sequence data.

[0120] Among them, the Bi-LSTM model of the conditional adversarial network is the initial model of the self-learning early warning model, the condition generator is connected to the data discriminator, and the data discriminator is connected to the forward LSTM layer, the forward LSTM layer is connected to the self-attention recognition layer, and the backward LSTM layer is connected to the self-attention recognition layer.

[0121] Step S302, input the training set into the Bi-LSTM model based on the conditional adversarial network, expressed as:

[0122]

[0123] Among them, the input training set is represented by x1,x2,...x i-1 ,x i , C(x) is the predicted output result, and i is the number of samples. In this embodiment, the number of samples can be 100-1000;

[0124] Step S303: The condition generator introduces abnormal rules and heuristic strategies to form a conditional adversarial network. The data discriminator is used to improve the conditional adversarial network. The predicted output result is used as input to train the conditional adversarial network. The objective function form of the conditional adversarial network is:

[0125]

[0126] Among them, H(C) is the conditional generator function of the conditional adversarial network, L(C) is the data discriminator function of the conditional adversarial network, and C data (x) is the independent probability distribution of the predicted output, CC data (x) means that the data C(x) follows an independent probability distribution;

[0127] In this embodiment, when the condition generator introduces exception rules and heuristic strategies, auditors with this authority formulate a series of exception rules and heuristic strategies based on historical data, compliance requirements and industry rules, such as setting rules for abnormal payments of funds for dental implant medical consumables, and setting a corresponding series of key indicators.

[0128] Step S304, the self-attention recognition layer obtains the output result of the conditional adversarial network, and the self-attention recognition result is expressed as:

[0129]

[0130] Among them, F(C) is the self-attention recognition result, n is the number of convolutional layers of the self-attention recognition layer, g is the self-attention expansion factor of the self-attention recognition layer, and x i-g For inputting the dilated convolution of the training set based on the adversarial network, in the embodiment of the present invention, the number of convolution layers of the self-attention recognition layer is 1-5 layers, and the self-attention dilation factor of the self-attention recognition layer is set to 0.2-0.5;

[0131]

[0132] Step S305, the backward LSTM layer extracts the hidden recognition result from the self-attention recognition result based on the swi sh function, and combines the hidden recognition result with the self-attention recognition result to generate a training set detection result;

[0133] Step S306, determining whether the training set detection result meets the preset training accuracy. In the embodiment of the present invention, the preset training accuracy can be set to 0.85-1;

[0134] Step S307: if the preset training accuracy is met, output the self-learning early warning model;

[0135] Step S308: If the preset training accuracy is not met, the first-order and second-order moments of the model are calculated based on the ADMA optimizer, and the model learning rate is adjusted, and the process returns to step S402 to input the training set into the Bi-LSTM model based on the conditional adversarial network.

[0136] Step S309, obtaining a test set, verifying the recognition accuracy of the self-learning early warning model through the test set, and judging whether the recognition accuracy meets a preset recognition threshold; in this embodiment, the preset recognition threshold is set to 0.6-0.7.

[0137] Step S310: If the preset recognition threshold is met, the self-learning early warning model is output, and the abnormal use points and abnormal use patterns of medical consumables are identified through the self-learning early warning model to generate recognition detection results.

[0138] Step S311: If it does not meet the preset recognition threshold, return to step S402 and input the training set into the Bi-LSTM model based on the conditional adversarial network.

[0139] In this embodiment, the self-learning early warning model convergence judgment is performed according to the set recognition threshold and preset training accuracy. If the expected performance is not achieved, the potential errors in the rule setting will be intelligently identified, and reminders will be issued to the auditors to guide the auditors to perform targeted optimization and adjustment of the rules.

[0140] In an embodiment of the present invention, the audit knowledge graph includes:

[0141] At least one set of audit trigger engines, the audit trigger engines are used to obtain identification detection results, and the trigger engines include an audit rule base, a risk threshold assessor, a command trigger, and an audit recorder;

[0142] In the embodiment of the present invention, the risk threshold assessor sets reasonable risk warning thresholds. These thresholds include but are not limited to inventory levels below a certain level, abnormal increase in the usage of a certain consumable, etc. Once these thresholds are reached or exceeded, the risk threshold assessor will automatically trigger the first audit instruction or the second audit instruction.

[0143] Among them, the audit rule base is connected to the consumables basic database for communication, the audit rule base is used to traverse the audit rules and text data in the consumables basic database, the risk threshold assessor calculates and identifies whether the detection result exceeds the preset risk threshold based on the error back propagation algorithm, the instruction trigger is used to obtain the judgment result of the risk threshold assessor, and trigger the first audit instruction or the second audit instruction based on the judgment result, and the audit recorder is used to store the audit rule base, the risk threshold assessor, and the instruction trigger operation log;

[0144] An audit search engine is used to perform a first audit or a second audit on the standardized data of consumables, and the audit search engine selects audit rules and audit parameters applicable to the first audit and the second audit in the consumables basic database, writes audit operations and audit reports based on the CRDT algorithm, and verifies audit authority and audit credibility.

[0145] In an embodiment of the present invention, an audit knowledge graph is provided and constructed. The audit knowledge graph can cooperate with the self-learning early warning model, and the audit knowledge graph integrates big data analysis and identification detection results to automatically complete the audit process and generate comprehensive audit reports and specific improvement suggestions, thereby strengthening the data basis for management decision-making.

[0146] The embodiment of the present invention provides a method for identifying the identification detection result by an audit knowledge graph and determining whether the identification detection result exceeds a preset risk threshold. Figure 5 The present invention shows a schematic diagram of the implementation process of the method for identifying the recognition detection result by the audit knowledge graph and judging whether the recognition detection result exceeds the preset risk threshold. The method for identifying the recognition detection result by the audit knowledge graph and judging whether the recognition detection result exceeds the preset risk threshold specifically includes:

[0147] Step S401, obtaining the recognition detection result, and calculating the risk value of the recognition detection result based on the error back propagation algorithm;

[0148] M(C)=w i f(C)+h c (8)

[0149] Among them, M(C) is the risk value of the identification test result, w i To identify the weight normalization coefficient of the detection result, h c is the offset corresponding to the recognition detection result, and f(C) is the input representation of the recognition detection result;

[0150]

[0151] Step S402: define a risk threshold based on the audit rules and audit parameters. The risk threshold is expressed as:

[0152]

[0153] Among them, Y is the risk threshold, x τ To visualize the standardized data of consumables in a two-dimensional time series, λ is the risk coefficient of the audit parameter, and t is the manually defined threshold parameter.

[0154] In this embodiment, the risk coefficient of the audit parameter is 1 or 2, and the manually defined threshold parameter can be 2-10.

[0155] Step S403: Obtain a defined risk threshold, and determine whether the risk value of the identification detection result exceeds the defined risk threshold.

[0156] In an embodiment of the present invention, the identification test results are identified through an audit knowledge graph, and it is determined whether the identification test results exceed a preset risk threshold, thereby providing an audit model for medical consumables audits, automatically generating audit reports and recommendations, and providing strong support for management's decision-making.

[0157] Example 2

[0158] The embodiment of the present invention provides a monitoring and evaluation system for auditing medical consumables. Figure 6 The structural block diagram of the monitoring and evaluation system for medical consumables audit is shown, and the monitoring and evaluation system for medical consumables audit specifically includes:

[0159] The information system integration module 100 is composed of a data normalization unit 110 and a basic database unit 120. The information system integration module 100 obtains medical consumables collection information based on the hospital medical information system, standardizes the medical consumables collection information, obtains consumables standardized data, and constructs a consumables basic database based on the consumables standardized data;

[0160] The early warning model training module 200 is used to capture the standardized data of consumables in the consumables basic database, divide the standardized data of consumables into a training set and a test set in a ratio of 3:1, and assign abnormal rules and heuristic strategies to the training set and the test set, and pre-build a self-learning early warning model by combining deep learning and adversarial training;

[0161] The detection result generation module 300 is used to bring the standardized data of consumables into the self-learning early warning model, and the self-learning early warning model identifies the abnormal use points and abnormal use patterns of medical consumables and generates identification detection results;

[0162] The report generation module 400 constructs an audit knowledge graph based on risk warning and audit rules, obtains identification detection results, and the audit knowledge graph identifies the identification detection results, determines whether the identification detection results exceed a preset risk threshold, and if the preset risk threshold is exceeded, triggers the audit knowledge graph warning mechanism, and the audit knowledge graph automatically generates warning information, and performs a first audit on the consumables standardized data based on the warning information to generate a first audit report; if the preset risk threshold is not exceeded, a second audit is performed on the consumables standardized data to generate a second audit report;

[0163] The visualization presentation module 500 is used to obtain the recognition detection results, the first audit report, and the second audit report, and to visually present the recognition detection results, the first audit report, and the second audit report using visualization tools.

[0164] The embodiment of the present invention provides a monitoring and evaluation system for medical consumables auditing. The monitoring and evaluation system for medical consumables auditing is composed of an information system integration module 100, an early warning model training module 200, a test result generation module 300, a report generation module 400, and a visualization presentation module 500. The system introduces big data processing algorithms and artificial intelligence technologies to realize the automation and intelligence of the audit process, improve audit efficiency and quality, and enable the audit to dig deeper into potential problems and risk points. By using the results of big data analysis, an effective self-learning early warning model and early warning mechanism are established to timely identify potential abnormal behaviors or irregular processes, ensure the financial security of medical institutions and reduce operational risks. The system can also ensure that the tracking records of medical consumables from procurement to use are complete, meet regulatory requirements, and improve the compliance and credibility of the entire medical system.

[0165] In this embodiment, the data normalization unit 110 includes:

[0166] An information system index module 111, wherein the information system index module 111 is used to generate a data capture instruction, and the hospital medical information system responds to the data capture instruction and uploads the collected medical consumables collection information;

[0167] A missing value processing module 112, the missing value processing module 112 is used to load the collected medical consumables collection information, and perform data cleaning processing on the medical consumables collection information, wherein the missing value processing data is used when performing data cleaning processing on the medical consumables collection information;

[0168] The data marking module 113 is used to perform Z-Score Normalization normalization on the medical consumables collection information, and then use the open source tool Playment to perform attribute marking on the normalized medical consumables collection information;

[0169] The data visualization module 114 is used to preset the Fourier function and use continuous Fourier transform to convert the medical consumables collection information into visualized two-dimensional time series consumables standardized data, wherein the consumables standardized data includes visualized two-dimensional time series marks and attribute marks.

[0170] It is understandable that the monitoring and evaluation system for medical consumables auditing provided by the embodiment of the present invention corresponds to the above-mentioned monitoring and evaluation method for medical consumables auditing. The explanations, examples, beneficial effects and other parts of the relevant contents can refer to the corresponding contents in the monitoring and evaluation method for medical consumables auditing, and will not be repeated here.

[0171] Example 3

[0172] Another aspect of the embodiments of the present invention further provides a computer-readable storage medium, which stores computer program instructions, and the computer program instructions can be executed by a processor. When the computer program instructions are executed, the method of any one of the above embodiments is implemented.

[0173] The processor executes various functional applications and data processing of the server by running non-volatile software programs, instructions and modules stored in the memory, that is, realizes the monitoring and evaluation method for medical consumables auditing of the above method embodiment.

[0174] Finally, it should be noted that the computer-readable storage medium (e.g., memory) herein may be a volatile memory or a nonvolatile memory, or may include both a volatile memory and a nonvolatile memory. As an example and not by way of limitation, a nonvolatile memory may include a read-only memory (ROM), a programmable ROM (PROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), or a flash memory. A volatile memory may include a random access memory (RAM), which may act as an external cache memory. As an example and not by way of limitation, RAM may be obtained in a variety of forms, such as synchronous RAM (DRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The storage devices of the disclosed aspects are intended to include, but are not limited to, these and other suitable types of memory.

[0175] In summary, the present invention provides a monitoring and evaluation method, system and storage medium for medical consumables auditing. In the embodiments of the present invention, by constructing a basic consumables database, it is possible to comprehensively collect and integrate data of the entire cycle of medical consumables, significantly improve the accuracy and completeness of the data, and provide a solid foundation for subsequent self-learning early warning model analysis. At the same time, a self-learning early warning model is pre-constructed in combination with deep learning and adversarial training. The self-learning early warning model is a Bi-LSTM model based on a conditional adversarial network that combines deep learning and adversarial training. It overcomes the problem of insufficient accuracy and robustness of the prior art when processing complex sequence data, effectively improves the accuracy and robustness of the Bi-LSTM network when processing complex sequence data, and is suitable for the requirements of medical consumables auditing for data integrity and accuracy.

[0176] In an embodiment of the present invention, when the medical consumables collection information is standardized, the medical consumables collection information is converted into visualized two-dimensional time series consumables standardized data by using a preset Fourier function and a continuous Fourier transform. On the one hand, it can integrate the multi-source heterogeneous data information of each independent link to achieve information transparency and resource sharing, and ensure that the data between each link can be seamlessly connected and updated in real time. On the other hand, the consumables standardized data includes visualized two-dimensional time series marks and attribute marks, which facilitates the construction of a basic database for consumables, and is also conducive to the visual presentation of identification detection results, the first audit report, and the second audit report.

[0177] In the embodiment of the present invention, the self-learning early warning model uses the Bi-LSTM model based on the conditional adversarial network as the initial model, so that the trained self-learning early warning model has the advantages of Bi-LSTM in being able to simultaneously capture the forward and backward information of the data, which helps to fully understand the contextual relationship in the data. At the same time, it overcomes the problem of insufficient accuracy and robustness of the prior art in processing complex sequence data, and effectively improves the accuracy and robustness of the Bi-LSTM network in processing complex sequence data.

[0178] It should be noted that, for the above-mentioned embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described order of actions, because according to the present invention, some steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0179] In the several embodiments provided in the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of the above-mentioned units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be in the form of telecommunication or other forms.

[0180] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the embodiments of the present invention according to the circumstances without conflict and without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also belong to the scope of protection of the present invention.

Claims

1. A monitoring and evaluation method for auditing medical consumables, characterized in that: The monitoring and evaluation method for medical consumables audit includes: Obtaining medical consumables collection information based on the hospital medical information system, standardizing the medical consumables collection information to obtain consumables standardized data, and building a consumables basic database based on the consumables standardized data; Capture the standardized data of consumables in the consumables basic database, divide the standardized data into training set and test set in a ratio of 3:1, assign abnormal rules and heuristic strategies to the training set and test set, and pre-build a self-learning early warning model by combining deep learning and adversarial training; Obtain standardized data on consumables and bring the standardized data into the self-learning early warning model. The self-learning early warning model identifies abnormal usage points and abnormal usage patterns of medical consumables and generates identification and detection results. Construct an audit knowledge graph based on risk warning and audit rules, obtain identification and detection results, identify the identification and detection results with the audit knowledge graph, and determine whether the identification and detection results exceed a preset risk threshold; If the preset risk threshold is exceeded, the audit knowledge graph warning mechanism is triggered, and the audit knowledge graph automatically generates warning information, and based on the warning information, conducts a first audit on the standardized data of consumables and generates a first audit report; If the preset risk threshold is not exceeded, a second audit is conducted on the standardized data of consumables and a second audit report is generated.

2. The monitoring and evaluation method for medical consumables auditing according to claim 1, characterized in that: The monitoring and evaluation method for medical consumables audit also includes: Obtain identification test results, the first audit report, and the second audit report, and use visualization tools to visualize the identification test results, the first audit report, and the second audit report.

3. The monitoring and evaluation method for medical consumables auditing according to claim 1, characterized in that: The standardized processing of the medical consumables collection information specifically includes: The hospital medical information system responds to the data capture instruction and uploads the collected medical consumables information, wherein the hospital medical information system includes the hospital information system, the electronic health record system, the laboratory information system, the hospital logistics management system, and the hospital efficient operation management system, and the medical consumables collection information includes the name, specification, quantity, price, supplier, procurement, warehousing, outbound, use, and payment data information of the medical consumables; Loading the collected medical consumables collection information, and performing data cleaning processing on the medical consumables collection information, wherein missing value processing data is used when performing data cleaning processing on the medical consumables collection information; The medical consumables collection information was normalized using Z-Score Normalization, and then the open source tool Playment was used to label the normalized medical consumables collection information; A Fourier function is preset, and continuous Fourier transform is used to convert the medical consumables collection information into consumables standardized data of visualized two-dimensional time series, wherein the consumables standardized data includes visualized two-dimensional time series marks and attribute marks; Among them, the preset Fourier function during continuous Fourier transform is expressed as: Among them, δ is the scale parameter of medical consumables collection information, a is the data scaling parameter, n is the sample size of consumables standardized data, and f(x) is the mother wavelet function; Among them, δ is the scale parameter of medical consumables collection information, a is the data scaling parameter, and n is the sample size of consumables standardized data. The preset Fourier function and the medical consumables collection information are inner-producted to obtain the consumables standardized data of visualized two-dimensional time series; Among them, x τ To visualize the two-dimensional time series of consumables standardized data, ψ(x) is the Fourier transform coefficient, and b is the attribute labeling weight corresponding to the consumables standardized data.

4. The monitoring and evaluation method for medical consumables auditing according to claim 3 is characterized in that: The construction of a consumables basic database based on the standardized consumables data specifically includes: Acquire standardized data of consumables, identify data features of the standardized data of consumables, and extract feature vectors based on the data features of the standardized data of consumables; Analyze the dependencies and importance between the data features of the standardized data of consumables, determine the correlation between the data features, and design the data table and architecture tree of the consumables basic database based on the correlation, feature vectors, timing tags, and attribute tags; Use SQL statements to fill the standardized data of consumables into the data table, and associate the data table with the architecture tree through index structure based on time series tags and attribute tags to generate a basic database of consumables.

5. The monitoring and evaluation method for medical consumables auditing according to claim 1, characterized in that: The method of pre-building a self-learning early warning model by combining deep learning and adversarial training specifically includes: Obtain training sets and test sets, and build a Bi-LSTM model based on a conditional adversarial network. The Bi-LSTM model of the conditional adversarial network includes a condition generator, a data discriminator, a forward LSTM layer, a backward LSTM layer, and a self-attention recognition layer. The training set is input into the Bi-LSTM model based on the conditional adversarial network, expressed as: Among them, the input training set is represented by x1,x2,...x i-1 ,x i , C(x) is the predicted output result, and i is the number of samples; The conditional generator introduces abnormal rules and heuristic strategies to form a conditional adversarial network. The data discriminator is used to improve the conditional adversarial network. The predicted output results are used as input to train the conditional adversarial network. The objective function form of the conditional adversarial network is: Among them, H(C) is the conditional generator function of the conditional adversarial network, L(C) is the data discriminator function of the conditional adversarial network, and C data (x) is the independent probability distribution of the predicted output, CC data (x) means that the data C(x) follows an independent probability distribution; The self-attention recognition layer obtains the output result of the conditional adversarial network. The self-attention recognition result is expressed as: Among them, F(C) is the self-attention recognition result, n is the number of convolutional layers of the self-attention recognition layer, g is the self-attention expansion factor of the self-attention recognition layer, and x i-g Dilated convolution based on adversarial network for input training set; The backward LSTM layer extracts the hidden recognition results from the self-attention recognition results based on the swish function, and combines the hidden recognition results with the self-attention recognition results to generate the training set detection results, and judges whether the training set detection results meet the preset training accuracy. If they meet the preset training accuracy, the self-learning early warning model is output. If they do not meet the preset training accuracy, the first-order and second-order moments of the model are calculated based on the ADMA optimizer, and the model learning rate is adjusted. Return to the step to input the training set into the Bi-LSTM model based on the conditional adversarial network; Obtain a test set, verify the recognition accuracy of the self-learning early warning model through the test set, and determine whether the recognition accuracy meets the preset recognition threshold. If it meets the preset recognition threshold, output the self-learning early warning model, and use the self-learning early warning model to identify abnormal usage points and abnormal usage patterns of medical consumables, and generate recognition detection results.

6. The monitoring and evaluation method for medical consumables auditing according to claim 2, characterized in that: The audit knowledge graph includes: At least one set of audit trigger engines, the audit trigger engines are used to obtain identification detection results, and the trigger engines include an audit rule base, a risk threshold assessor, a command trigger, and an audit recorder; Among them, the audit rule base is connected to the consumables basic database for communication, the audit rule base is used to traverse the audit rules and text data in the consumables basic database, the risk threshold assessor calculates and identifies whether the detection result exceeds the preset risk threshold based on the error back propagation algorithm, the instruction trigger is used to obtain the judgment result of the risk threshold assessor, and trigger the first audit instruction or the second audit instruction based on the judgment result, and the audit recorder is used to store the audit rule base, the risk threshold assessor, and the instruction trigger operation log; An audit search engine is used to perform a first audit or a second audit on the standardized data of consumables, and the audit search engine selects audit rules and audit parameters applicable to the first audit and the second audit in the consumables basic database, writes audit operations and audit reports based on the CRDT algorithm, and verifies audit authority and audit credibility.

7. The monitoring and evaluation method for medical consumables auditing according to claim 6, characterized in that: The audit knowledge graph identifies the identification detection result, and the method for judging whether the identification detection result exceeds a preset risk threshold specifically includes: Obtain identification and detection results, and calculate the risk value of the identification and detection results based on the error back propagation algorithm; M(C)=w i f(C)+h c (8) Among them, M(C) is the risk value of the identification test result, w i To identify the weight normalization coefficient of the detection result, h c is the offset corresponding to the recognition detection result, and f(C) is the input representation of the recognition detection result; The risk threshold is defined based on the audit rules and audit parameters. The risk threshold is expressed as: Among them, Y is the risk threshold, x τ To visualize the standardized data of consumables in a two-dimensional time series, λ is the risk coefficient of the audit parameter, and t is the manually defined threshold parameter.

8. A monitoring and evaluation system for medical consumables audit, used for the monitoring and evaluation method for medical consumables audit as claimed in any one of claims 1 to 7, characterized in that: The monitoring and evaluation system for medical consumables auditing includes: The information system integration module is composed of a data normalization unit and a basic database unit. The information system integration module obtains medical consumables collection information based on the hospital medical information system, standardizes the medical consumables collection information, obtains consumables standardized data, and builds a consumables basic database based on the consumables standardized data; Early warning model training module: The early warning model training module is used to capture the standardized data of consumables in the consumables basic database, divide the standardized data into training set and test set in a ratio of 3:1, and assign abnormal rules and heuristic strategies to the training set and test set, and pre-build a self-learning early warning model by combining deep learning and adversarial training; A test result generation module, which is used to bring the standardized data of consumables into the self-learning early warning model. The self-learning early warning model identifies abnormal use points and abnormal use patterns of medical consumables and generates recognition test results; A report generation module, which constructs an audit knowledge graph based on risk warning and audit rules, obtains identification and detection results, identifies the identification and detection results with the audit knowledge graph, and determines whether the identification and detection results exceed a preset risk threshold. If the preset risk threshold is exceeded, the audit knowledge graph warning mechanism is triggered, and the audit knowledge graph automatically generates warning information, and performs a first audit on the standardized data of consumables based on the warning information to generate a first audit report; if the preset risk threshold is not exceeded, a second audit is performed on the standardized data of consumables to generate a second audit report; The visualization presentation module is used to obtain the recognition detection results, the first audit report, and the second audit report, and use visualization tools to visually present the recognition detection results, the first audit report, and the second audit report.

9. The monitoring and evaluation system for medical consumables auditing according to claim 8, characterized in that: The data normalization unit comprises: An information system index module, the information system index module is used to generate a data capture instruction, and the hospital medical information system responds to the data capture instruction and uploads the collected medical consumables collection information; A missing value processing module, which is used to load the collected medical consumables collection information and perform data cleaning on the medical consumables collection information, wherein the missing value processing data is used when performing data cleaning on the medical consumables collection information; Data labeling module: The data labeling module is used to perform Z-Score Normalization on the medical consumables collection information, and then use the open source tool Playment to perform attribute labeling on the normalized medical consumables collection information; The data visualization module is used to preset the Fourier function and use continuous Fourier transform to convert the medical consumables collection information into the consumables standardized data with visualized two-dimensional time series. The consumables standardized data includes visualized two-dimensional time series marks and attribute marks.

10. A computer-readable storage medium, characterized in that: Computer program instructions are stored, and when the computer program instructions are executed, the monitoring and evaluation method for medical consumables auditing as described in any one of claims 1 to 7 is implemented.

Citation Information

Patent Citations

  • Method for identifying patient with insurance fraud behavior based on improved generative adversarial network

    CN113628057A

  • Power system data auditing early warning method and system

    CN116562789A

  • Sequence recommendation model training method and device and media asset recommendation method and device

    CN117271878A

  • Quality control method, device and equipment for medical consumables and storage medium

    CN117594208A

  • Gravity wave instrument data completion model and method based on conditional generative adversarial network

    CN118260540A

Cited By

  • Medical consumable abnormal data type-2 fuzzy screening method based on cross-domain collaboration and isomerism map

    CN120974390A

  • A medical consumable abnormal data two-type fuzzy screening method based on cross-domain cooperation and heterogeneous atlas

    CN120974390B