Data classification quality control index calculation method and device based on deep learning and medium

Abstract

The invention discloses a data classification quality control index calculation method and device based on deep learning and a medium, and relates to the technical field of data quality control evaluation, and the method comprises the following steps: formulating a medical record home page quality control rule, and establishing a coding mapping rule; collecting medical record home page data, and establishing a medical record standard data set; performing preliminary detection and evaluation on the medical record home page based on the medical record home page quality control rule to obtain medical record preliminary detection information, and detecting the medical record home page diagnosis items by using the medical record diagnosis detection model to obtain medical record diagnosis detection information; obtaining a data classification quality control index of the medical record based on the medical record preliminary detection information and the medical record diagnosis detection information; the method is used for solving the problem that an existing data quality control evaluation technology cannot further detect outpatient emergency diagnosis and main discharge diagnosis of a home page of a medical record through a deep learning technology while quality control evaluation is performed on data of the home page of the medical record based on a quality control rule, and then a quality control evaluation model cannot be established for quality control evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of data quality control evaluation, and specifically to a method, device and medium for calculating data classification quality control index based on deep learning. Background Art

[0002] Data quality control and assessment technology is the process of applying a series of methods and algorithms to evaluate, control and improve data quality. Data quality control and assessment technology aims to ensure the accuracy, completeness, consistency and validity of data, thereby providing a reliable basis for data analysis and decision-making. In practical applications, data quality control assessment of the medical record homepage is a typical application scenario.

[0003] The medical record front page is the most important content in the medical record of medical activities generated during the hospital's diagnosis and treatment process. It is the main source of medical, hospital statistics, clinical research and national health statistics information. Therefore, the quality control evaluation of the medical record front page data is an important link to ensure the quality of medical data and improve the level of medical services. The existing medical record front page data quality control evaluation technology is mainly based on data analysis, using the medical record front page filling specifications to establish quality control rules for the medical record front page, and perform quality control evaluation on the medical record front page data through the quality control rules; or combine the quality control rules with fixed algorithms to perform quality control evaluation on the medical record front page data; for example, the patent application with publication number CN116108021A discloses a method, device and equipment for multi-dimensional data quality control based on a standard data set. The solution is to establish a data quality control related table, then determine the corresponding quality control rules, and then perform data quality control according to the quality control rules; however, there are many types of errors in the medical record front page data, and it is not enough to rely solely on establishing relevant quality control rules for quality control. The quality control evaluation is difficult and cannot flexibly handle complex situations: for example, the selection and coding of patients' outpatient and emergency diagnoses and major discharge diagnoses are generally complex. Moreover, as important links in the medical process, the accuracy of outpatient and emergency diagnoses and major discharge diagnoses is not only related to the treatment effect of patients, but also to whether they can provide reliable data for medical research. Their records in the medical record homepage play a vital role in the quality and efficiency of the entire medical service. Therefore, it is very necessary to further test the outpatient and emergency diagnoses and major discharge diagnoses. In this case, simple rules may not be able to effectively identify errors or provide accurate evaluations. Moreover, it is difficult to comprehensively reflect the actual clinical situation and lacks dynamic adjustment and optimization capabilities. Therefore, the existing data quality control evaluation technology cannot perform quality control evaluation on the medical record homepage data based on quality control rules, and at the same time, further test the outpatient and emergency diagnoses and major discharge diagnoses on the medical record homepage through deep learning technology, and then establish a quality control evaluation model for quality control evaluation. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems in the prior art to a certain extent, by establishing quality control rules for the medical record homepage and a medical record diagnosis detection model, performing quality control evaluation on the medical record homepage based on the quality control rules for the medical record homepage and utilizing the medical record diagnosis detection model, and analyzing and processing to obtain a data classification quality control index for the medical record homepage; so as to solve the problem that the existing data quality control evaluation technology is unable to perform quality control evaluation on the medical record homepage data based on the quality control rules, while further detecting the outpatient and emergency diagnoses and major discharge diagnoses of the medical record homepage through deep learning technology, and then establishing a quality control evaluation model for quality control evaluation.

[0005] To achieve the above objectives, in a first aspect, the present application provides a method for calculating a data classification quality control index based on deep learning, comprising the following steps:

[0006] Based on the items that need to be filled in the medical record front page, formulate the quality control rules for the medical record front page, and establish the coding mapping rules to map the codes for the items that need to be filled in;

[0007] Collect medical record front page data, standardize the medical record front page data, and establish a standard medical record data set;

[0008] Based on the quality control rules of the medical record homepage, a preliminary test and evaluation is performed on the medical record homepage to obtain preliminary test information of the medical record. If the diagnosis items on the medical record homepage meet the quality control rules of the medical record homepage, the medical record diagnosis test model is used to test the diagnosis items on the medical record homepage to obtain the medical record diagnosis test information;

[0009] Based on the analysis and processing of the preliminary test information and the diagnostic test information of the medical records, the data classification quality control index of the medical record homepage is obtained.

[0010] Furthermore, based on the items that need to be filled in the medical record front page, the quality control rules for the medical record front page are formulated, including the following sub-steps:

[0011] Formulate basic quality control rules: Based on the basic filling specifications of the items to be filled in on the front page of the medical record and the disease diagnosis coding rules, formulate quality control rules for all items that need to be filled in and mark them as basic quality control rules;

[0012] Formulate logical quality control rules: Based on the logical relationship between the items to be filled in the medical record homepage and the relationship between the disease diagnosis codes, formulate the quality control rules between the items to be filled in the medical record homepage, marked as logical quality control rules;

[0013] Mark the basic quality control rules and logical quality control rules as medical record homepage quality control rules.

[0014] Furthermore, establishing a coding mapping rule, and coding mapping the items to be filled in includes the following sub-steps:

[0015] Establish coding mapping rules: Use one-hot coding to encode the items that need to be filled in on the first page of the medical record, and encode them in order from left to right and from top to bottom, and ensure that the codes of any two different items that need to be filled in are not repeated.

[0016] Furthermore, collecting the medical record front page data, standardizing the medical record front page data, and establishing a standard medical record data set includes the following sub-steps:

[0017] Collect the medical record homepage data of historical patients based on the hospital database. The medical record homepage data includes hospitalization information, diagnosis and treatment information, patient information and cost information;

[0018] According to the basic filling specifications of the medical record homepage, the hospitalization information, diagnosis and treatment information, patient information and expense information contained in the medical record homepage are standardized, and the standardized medical record homepage data is marked as the first medical record data;

[0019] Based on the coding mapping rules, the first medical record data is coded and mapped respectively; ensuring that any two codes on the same medical record front page are not repeated, and that two codes of the same item on different medical record front pages are the same;

[0020] The first medical record data after coding mapping is marked as a standard medical record data set.

[0021] Furthermore, based on the quality control rules of the medical record homepage and using the medical record diagnosis detection model, a preliminary detection and evaluation of the medical record homepage includes the following sub-steps:

[0022] Testing based on the quality control rules for the medical record homepage: Use the quality control rules for the medical record homepage to test the medical record homepage to determine whether each item to be filled in on the medical record homepage complies with the basic quality control rules and logical quality control rules for that item; record the item names and the number of items that comply with the basic quality control rules and logical quality control rules, and mark them as the correct medical record situation of the medical record; traverse all the medical record homepages to be tested, obtain the correct medical record situation of all the medical record homepages to be tested, and mark them as the correct medical record homepage information; record the item names and the number of items that do not comply with the basic quality control rules and logical quality control rules, and mark them as the medical record error situation of the medical record; traverse all the medical record homepages to be tested, obtain the medical record error situation of all the medical record homepages to be tested, and mark them as the medical record homepage error information;

[0023] Mark the correct information and the wrong information of the medical record as preliminary detection information of the medical record;

[0024] Based on the correct information in the medical record, the diagnostic items on the medical record front page that meet the basic quality control rules and logical quality control rules are screened. The diagnostic items on the medical record front page include outpatient and emergency diagnoses or major discharge diagnoses. The medical record diagnosis detection model is used to detect the diagnostic items on the medical record front page that meet the quality control rules of the medical record front page to obtain the medical record diagnosis detection information.

[0025] Furthermore, the medical record diagnosis detection model is used to detect the diagnosis items on the medical record homepage to obtain the medical record diagnosis detection information, including the following sub-steps:

[0026] Based on the standard medical record data set, outpatient and emergency diagnosis information, main discharge diagnosis information, other discharge diagnosis information, admission department information, discharge department information, and western medicine cost information from the front page of each medical record were extracted and marked as the model training data set;

[0027] Mark outpatient and emergency diagnosis, main discharge diagnosis, other discharge diagnosis, admission department, discharge department, and western medicine expenses as target items;

[0028] After completion, the diagnostic items and the corresponding related items are marked as the first training data set; the first training data set is evenly divided into 10 parts, and each part is divided into a training set and a test set in a ratio of 7:3; 10 training sets and corresponding 10 test sets are obtained;

[0029] The TabNet model is trained and optimized using the training set and test set.

[0030] Furthermore, using the training set and test set to train and optimize the TabNet model includes the following sub-steps:

[0031] Set the training hyperparameters of the TabNet model and the parameter range set of the model parameters;

[0032] The parameter range set includes the value range of the training hyperparameter and the value range of the model parameter; the number of value steps of the training hyperparameter and the value step of the model parameter are set based on the value range;

[0033] Use the relevant items as the input of the TabNet model and the diagnostic items as the output of the TabNet model; initialize the training hyperparameters and model parameters;

[0034] Use 10 training sets to train the TabNet model for 10 rounds. After each round of training, use the test set to test it to obtain the model prediction accuracy of this round of training, save the model prediction accuracy of this round, and after completing 10 rounds of training and testing, use the 10 obtained model prediction accuracies to calculate the average value, and save the training hyperparameters and model parameters of these 10 rounds of training; then use grid search to change the training hyperparameters and model parameters, and use 10 training sets to train the TabNet model for 10 rounds again. After each round of training, use the corresponding test set to test it to obtain the model prediction accuracy of 10 rounds, calculate the model average accuracy, and save the training hyperparameters and model parameters of these 10 rounds of training; until all combinations of training hyperparameters and model parameters are traversed using grid search, the training hyperparameters and model parameters corresponding to the maximum model average accuracy are marked as the optimal parameters;

[0035] Based on the optimal parameters, the TabNet model was trained using 10 training sets, and after completion, the medical record diagnosis and detection model was obtained;

[0036] Detection based on the medical record diagnosis detection model: Based on the medical record diagnosis detection model, set a diagnostic item that needs to be tested, including outpatient and emergency diagnosis and major discharge diagnosis, use the medical record diagnosis detection model to make predictions, obtain the predicted content, compare the predicted content with the content of the diagnostic item in the corresponding position in the original medical record homepage, determine whether the content is consistent, record the name and number of items that need to be filled in for inconsistent content, mark them as diagnostic abnormalities of the medical record, traverse all medical record homepages to be tested, obtain the diagnostic abnormalities of all medical record homepages to be tested, and mark them as diagnostic abnormality information.

[0037] Furthermore, based on the analysis and processing of the detection information of the medical record homepage, obtaining the data classification quality control index of the medical record homepage includes the following sub-steps:

[0038] Based on the medical record homepage detection information, the number of medical record homepages with only correct medical records is counted, marked as X, the number of medical record homepages with only incorrect medical records is counted, marked as Y, and the number of medical record homepages with only abnormal diagnosis is counted, marked as Z;

[0039] Set accuracy, completeness and consistency as data classification quality control indicators, among which accuracy

[0040] = [X / (X+Y+Z)]; Completeness = [(X+Z) / (X+Y+Z)]; Consistency = 1-[Z / (X+Y+Z)];

[0041] Establish a quality control index evaluation model M, M = {D, I, R, W, S}, where D represents the set of medical record homepages that need to be evaluated for quality control index; I represents the data classification quality control index corresponding to the quality control index evaluation in D, R represents the quantitative calculation rule of I; W represents the weight of the data classification quality control index in I; S represents the final result corresponding to R and W, S = q1*accuracy+q2*completeness+q3*consistency, where q1, q2 and q3 represent weights, the value range is [0, 1], and q1+q2+q3=1;

[0042] Based on the detection information on the medical record homepage, the quality control index evaluation model output M is used to obtain the data classification quality control index S.

[0043] In a second aspect, the present application provides an electronic device, including a processor and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps in the above method are performed.

[0044] In a third aspect, the present application provides a storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps in the above method are performed.

[0045] Beneficial effects of the present invention: The present invention first formulates quality control rules for the medical record homepage based on the items to be filled in the medical record homepage, and establishes coding mapping rules to perform coding mapping on the items to be filled in; then collects medical record homepage data, standardizes the medical record homepage data, and establishes a standard medical record data set; then, based on the medical record homepage quality control rules, the medical record homepage is preliminarily tested and evaluated to obtain preliminary medical record test information; if the diagnosis items on the medical record homepage meet the medical record homepage quality control rules, the medical record diagnosis test model is used to test the diagnosis items on the medical record homepage to obtain medical record diagnosis test information; then, based on the analysis and processing of the preliminary medical record test information and the medical record diagnosis test information, a data classification quality control index for the medical record homepage is obtained; while the medical record homepage data is quality-controlled and evaluated based on the medical record homepage quality control rules, the outpatient and emergency diagnoses and the main discharge diagnoses on the medical record homepage are further tested through deep learning technology, thereby improving the efficiency and accuracy of the quality control evaluation of the medical record homepage and diagnosis-related information;

[0046] The present invention uses the medical record homepage quality control rules and the medical record diagnosis detection model to perform quality control evaluation on the medical record homepage, which can significantly improve the accuracy and completeness of the medical record homepage data. The use of the medical record homepage quality control rules is helpful to achieve the standardization and normalization of medical record homepage management; the deep learning technology can be used to automatically analyze the medical record homepage data, reduce the need for manual operation, thereby greatly improving work efficiency, and analyze and process the medical record homepage data in real time, provide feedback in time, and help doctors and managers make adjustments quickly; a medical record diagnosis detection model is established based on the TabNet model. TabNet combines the advantages of deep neural networks and decision tree models, and can deeply analyze the potential connection between the outpatient and emergency diagnoses and the main discharge diagnoses on the medical record homepage and other related items, so as to judge the abnormal conditions of the outpatient and emergency diagnoses and the main discharge diagnoses on the medical record homepage, and ensure the accuracy and efficiency of the quality control of the outpatient and emergency diagnoses and the main discharge diagnoses. A quality control index evaluation model is established to perform quality control evaluation on the medical record homepage data, which can help to monitor the data quality status of the medical record homepage in real time, timely discover and solve the data quality risks of the medical record homepage, provide a reference for the optimization of the data quality of the medical record homepage, and help maintain the high quality status of the medical record homepage data. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a flow chart of the steps of the method of the present invention;

[0048] Figure 2 It is a schematic diagram of the TabNet model structure of the present invention;

[0049] Figure 3 It is a schematic diagram of the grid search principle of the present invention;

[0050] Figure 4 It is a schematic structural diagram of the electronic device of the present invention. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] Example 1, please refer to Figure 1 As shown, in the first aspect, the present application provides a method for calculating a data classification quality control index based on deep learning, comprising the following steps:

[0053] Step S1, based on the items to be filled in the medical record front page, formulate the quality control rules for the medical record front page, and establish the coding mapping rules to perform coding mapping on the items to be filled in; Step S1 includes the following sub-steps:

[0054] Step S101, formulate basic quality control rules: based on the basic filling specifications of the items to be filled in on the front page of the medical record and the disease diagnosis coding rules, formulate quality control rules for all items to be filled in, and mark them as basic quality control rules; the quality control rules for the items to be filled in refer to the basic filling specifications of the regular items on the front page of the medical record, for example, the basic filling specifications for the name item are Chinese characters, and cannot be empty, and special characters cannot be filled in; the basic filling specifications for the ID card number follow the coding rules for the ID card number; the disease diagnosis coding rules are used to classify and number the types of diseases, referring to the existing ICD10 disease coding;

[0055] Step S102, formulate logical quality control rules: based on the logical relationship between the items to be filled in the medical record front page and the relationship between the disease diagnosis codes, formulate the quality control rules between the items to be filled in the medical record front page, marked as logical quality control rules; logical quality control rules refer to rules that can verify the logical relationship between the items to be filled in the medical record front page, for example, the date of birth item must match the 8-digit date of birth code in the ID card item, and must also match the age item; any other expense item cannot be greater than the total expense item;

[0056] Step S103, marking the basic quality control rules and the logical quality control rules as the medical record homepage quality control rules;

[0057] Step S104, establish coding mapping rules: use one-hot coding to encode the required items on the front page of the medical record, and encode them in order from left to right and from top to bottom, and ensure that the codes of any two different required items are not repeated; one-hot coding is a common coding method in machine learning and data preprocessing, mainly used to process categorical data or discrete features and disordered features; the basic principle of one-hot coding is to use an N-bit state register to encode N states, each state is represented by its independent register bit, and only one bit is valid at any time, that is, it is set to 1. For each feature, if it has m possible values, it becomes m binary features after one-hot coding; and these features are mutually exclusive, and only one is activated at a time;

[0058] In the specific implementation process, the types of data information contained in the medical record homepage are extremely rich, so the quality control rules are also numerous and complex, and when the items to be filled in the medical record homepage are adjusted, the quality control rules of the medical record homepage also need to be updated in time to ensure the accuracy and effectiveness of the quality control of the medical record homepage; the establishment of the quality control rules of the medical record homepage can refer to the "Quality Specifications for Filling in Data on the Inpatient Medical Record Homepage (Interim)"; when encoding the items to be filled in the medical record homepage, the number of encoding bits of the one-hot encoding can be set according to actual needs. In this embodiment, the number of encoding bits of the one-hot encoding is 8 bits. The one-hot encoding converts discrete classification features into a binary format that is easy for machine learning algorithms to process, improves the algorithm's processing ability for discrete features, and solves the problem of classification data processing; and by assigning an independent binary vector to each category, the one-hot encoding avoids the model from making inaccurate predictions based on erroneous numerical relationships; and to a certain extent, the one-hot encoding expands the feature attributes, enabling some machine learning algorithms to better identify and process these features.

[0059] Step S2, collecting the medical record homepage data, standardizing the medical record homepage data, and establishing a standard medical record data set; Step S2 includes the following sub-steps:

[0060] Step S201, collecting medical record homepage data of historical patients based on the hospital database, the medical record homepage data includes hospitalization information, diagnosis and treatment information, patient information and cost information; hospitalization information includes medical institution, admission time, admission department and attending physician, etc., diagnosis and treatment information includes ABO blood type, Rh blood type, (main surgery) name and (main surgery) anesthesia method, etc., patient information includes medical record number, name, gender, age and date of birth, etc., cost information includes total hospitalization cost and out-of-pocket expenses, etc.;

[0061] Step S202, based on the basic filling specifications of the medical record front page, the hospitalization information, diagnosis and treatment information, patient information and cost information contained in the medical record front page are standardized, so that the various name terms contained in the hospitalization information, diagnosis and treatment information, patient information and cost information are standardized, and the name terms of the same thing are guaranteed to be consistent in different medical record front pages; the medical record front page contains a large amount of text description information, such as hospitalization information such as admission department, admission time, discharge department and discharge time, and outpatient and emergency diagnosis, pathological diagnosis and main discharge diagnosis. Because some doctors do not pay attention to the filling of medical records, data entry is not timely or information is wrong, which affects the authenticity and reliability of the data; for example, the anesthesia method item (major surgery) will be Contents such as "QM, YM, topical anesthesia, needle anesthesia" appeared, and it was difficult to determine the specific anesthesia method. There were also assignments such as "brachial plexus, prepare for general anesthesia", "brachial plexus (prepare for general anesthesia)" and "brachial plexus prepare for general anesthesia". "Brachial plexus, prepare for general anesthesia", "brachial plexus (prepare for general anesthesia)" and "brachial plexus prepare for general anesthesia" all refer to "brachial plexus anesthesia, prepare for general anesthesia". Therefore, this type of text information should be standardized to make it a standardized hospital department, ward name and medical terminology; ICD-10 should be used uniformly for disease diagnosis codes, and ICD-9-CM-3 should be used uniformly for surgery and operation codes; regions that use Diagnosis Related Groups (DRGs) to conduct hospital performance evaluation should use the clinical version of ICD-10 and the clinical version of ICD-9-CM-3;

[0062] Step S203, marking the medical record homepage data after standardization as the first medical record data;

[0063] Step S204, code mapping is performed on the first medical record data based on the code mapping rule to ensure that any two codes on the same medical record front page are not repeated, and that two codes of the same item on different medical record front pages are the same;

[0064] Step S205, marking the first medical record data after the coding mapping is completed as a standard medical record data set;

[0065] In the specific implementation process, the medical record homepage is a summary of case data formed by medical staff using text, symbols, codes, numbers, etc. to succinctly summarize the relevant information of the patient during hospitalization in a specific table; the medical record homepage usually includes four main parts: hospitalization information, diagnosis and treatment information, patient information and cost information; patient information includes the patient's name, gender, age, ethnicity, marital status, occupation, patient address and other information; hospitalization information includes admission time, admission route and the number of hospitalization and treatment in this medical institution, as well as discharge time and discharge method and other information; diagnosis information includes the main diagnosis, other diagnoses and diagnostic basis, as well as surgical operation information and treatment, mainly filling in the corresponding surgery and operation information, as well as non-surgical treatment, nursing measures and rehabilitation treatment and other information; cost information includes the use of various costs during the treatment process; for the standardization of medical record homepage data, you can refer to the official standard documents, use computers to screen, compare and modify and replace, in this implementation, for the standardization of medical record homepage data, refer to the "Quality Specifications for Filling in Inpatient Medical Record Homepage Data (Interim)" and "New Version of Health System 4-2 Interface Specifications" and "New Version of Health System 4-2 Code Dictionary".

[0066] Step S3, based on the quality control rules of the medical record homepage, a preliminary test and evaluation is performed on the medical record homepage to obtain preliminary test information of the medical record. If the diagnosis items on the medical record homepage meet the quality control rules of the medical record homepage, the medical record diagnosis test model is used to test the diagnosis items on the medical record homepage to obtain the medical record diagnosis test information. Step S3 includes the following sub-steps:

[0067] Step S301, performing detection based on the quality control rules of the medical record homepage; Step S301 includes the following sub-steps:

[0068] Step S3011, using the medical record homepage quality control rules to test the medical record homepage, and judging whether each item to be filled in the medical record homepage complies with the basic quality control rules and logical quality control rules of the item;

[0069] Step S3012, record the item names and the number of items that need to be filled in that meet the basic quality control rules and logical quality control rules, and mark them as the correctness of the medical record of the medical record; traverse all the medical record front pages to be tested, obtain the correctness of all the medical record front pages to be tested, and mark them as the correct information of the medical record front page; for example, in a medical record front page, the ethnicity item is filled in as Han nationality, which meets the basic quality control rules, then record the item name and the number of items that meet plus one;

[0070] Step S3013, record the item names and the number of non-compliance items that do not conform to the basic quality control rules and logical quality control rules, and mark them as the medical record error situation of the medical record; traverse all the medical record front pages to be tested, obtain the medical record error situations of all the medical record front pages to be tested, and mark them as medical record front page error information; for example, in a medical record front page, the Rh blood type item is filled in as type A, which does not conform to the basic quality control rules, then record the item name and the number of non-compliance plus one;

[0071] Step S302, marking the correct medical record information and the wrong medical record information as preliminary medical record detection information;

[0072] Step S303, based on the correct information in the medical record, screen the medical record front page diagnosis items that meet the basic quality control rules and logical quality control rules. The medical record front page diagnosis items include outpatient and emergency diagnoses or major discharge diagnoses. Use the medical record diagnosis detection model to detect the medical record front page diagnosis items that meet the medical record front page quality control rules to obtain medical record diagnosis detection information.

[0073] When the outpatient diagnosis or the main discharge diagnosis is initially correct, that is, the outpatient diagnosis or the main discharge diagnosis meets the basic quality control rules and the logical quality control rules, the model diagnosis is enabled; for example, Staphylococcal shoulder arthritis is filled in as Staphylococcal elbow arthritis, and the output result of Staphylococcal elbow arthritis is correct when it is detected using the quality control rules on the medical record homepage, but the actual situation of the patient itself does not fit, and the correct result should be Staphylococcal shoulder arthritis. This error can be predicted and screened through other related items, so the model is used for secondary detection to improve the accuracy of data quality control;

[0074] Step S304, using the medical record diagnosis detection model to detect the diagnosis items on the medical record homepage to obtain medical record diagnosis detection information; Step S303 includes the following sub-steps:

[0075] Step S3041, based on the second medical record data, extract the outpatient and emergency diagnosis information, main discharge diagnosis information, other discharge diagnosis information, admission department information, discharge department information and western medicine cost information from the front page of each medical record, and mark them as the model training data set;

[0076] Step S3042, mark the outpatient diagnosis, main discharge diagnosis, other discharge diagnosis, admission department, discharge department and western medicine cost as target items; the outpatient diagnosis and the main discharge diagnosis are related items. For example, if the diagnosis item is the outpatient diagnosis, the related items of the outpatient diagnosis are the main discharge diagnosis, other discharge diagnosis information, admission department information, discharge department information and western medicine cost information.

[0077] Step S3043, after completion, the diagnostic items and the corresponding related items are marked as a first training data set;

[0078] Step S3044, the first training data set is evenly divided into 10 parts, and each part is divided into a training set and a test set in a ratio of 7:3; 10 training sets and corresponding 10 test sets are obtained; the first training data set is evenly divided into 10 parts in order to use a 10-fold cross-validation method to evaluate subsequent models, which can effectively avoid the situation where the data distribution on the medical record homepage is relatively concentrated and affects the performance of subsequent models. In specific applications, the number of evenly divided parts can also be adjusted according to actual scenarios;

[0079] Step S3045, set the training hyperparameters of the TabNet model and the parameter range set of the model parameters; see Figure 2 As shown in the figure, Tabnet is a deep learning model. Tabnet simulates the decision-making process of decision trees through the basic neural network structure. On the basis of the basic neural network structure, the basic network structure of Tabnet is constructed, which mainly includes feature converters and attention converters. Tabnet is specially used to process tabular data and pays attention to the features of tabular data and the ability to learn data relationships. The model uses neural networks to simulate tree models and adopts a serialized attention mechanism to select the features that need to be inferred in each decision step, so that the network can focus on the most important features in the table.

[0080] The parameter range set includes the value range of the training hyperparameter and the value range of the model parameter; the number of value steps of the training hyperparameter and the value step of the model parameter are set based on the value range; in this embodiment, the training hyperparameters include: learning rate ∈ [0.0001, 0.001], with a value step of 0.0001; batch size ∈ [32, 128], with a value step of 16; model parameters include: N_a ∈ [8, 10], with a value step of 1; N_step ∈ [3, 6], with a value step of 1; gamma ∈ [1.2, 1.6], with a value step of 0.1;

[0081] Step S3046, using the diagnostic item as the input of the TabNet model and the target item as the output of the TabNet model; initializing the training hyperparameters and model parameters;

[0082] Step S3047, use 10 training sets to perform 10 rounds of training on the TabNet model, use the test set to test after each round of training, obtain the model prediction accuracy of this round of training, save the model prediction accuracy of this round, and after completing 10 rounds of training and testing, use the obtained 10 model prediction accuracies to calculate the average value to obtain the model average accuracy, and save the training hyperparameters and model parameters of these 10 rounds of training;

[0083] Step S3048, then use grid search to change the training hyperparameters and model parameters, and use 10 training sets to train the TabNet model for 10 rounds again. After each round of training, use the corresponding test set to test, get the model prediction accuracy of 10 rounds, and then calculate the average model accuracy, and save the training hyperparameters and model parameters of these 10 rounds of training; until all combinations of training hyperparameters and model parameters are traversed by grid search, the training hyperparameters and model parameters corresponding to the maximum average model accuracy are marked as the optimal parameters; please refer to Figure 3 As shown, grid search is a parameter adjustment method that optimizes model performance by traversing given parameter combinations. The basic steps of grid search include defining parameter space, creating models and evaluation criteria, exhaustively searching all parameter combinations, and evaluating performance in cross-validation, and finally selecting the best performing parameter combination as the model configuration. In this embodiment, according to the learning rate ∈ [0.0001, 0.0005], the step size is 0.0001; the batch size ∈ [32, 128], the step size is 32; N_a ∈ [8, 10], the step size is 1; N_step ∈ [3, 6], the step size is 1; there are 5*4*4*3=240 parameter combinations in total;

[0084] Step S3049, based on the optimal parameters, the TabNet model is trained using 10 training sets, and upon completion, a medical record diagnosis and detection model is obtained.

[0085] Step S305, setting a diagnostic item to be tested based on the medical record diagnosis detection model, including outpatient and emergency diagnosis and main discharge diagnosis;

[0086] Step S306, using the medical record diagnosis detection model to make predictions, obtain prediction content, compare the prediction content with the content of the diagnosis item at the corresponding position in the original medical record homepage, and determine whether the content is consistent;

[0087] Step S307, record the names and numbers of the items that need to be filled in with inconsistent contents, mark them as the abnormal diagnosis of the medical record, traverse all the medical record front pages to be tested, obtain the abnormal diagnosis of all the medical record front pages to be tested, and mark them as abnormal diagnosis information;

[0088] In the specific implementation process, first determine whether the information on the medical record homepage complies with the medical record homepage quality control rules, and record the medical record homepage information that does not comply with and complies with the medical record homepage quality control rules respectively; because data quality control evaluation based on the medical record homepage quality control rules cannot cope with all data quality control scenarios, and outpatient diagnosis and discharge diagnosis are important links in the medical process, their records in the medical record homepage play a vital role in the quality and efficiency of the entire medical service. Therefore, it is necessary to further detect outpatient diagnosis and discharge diagnosis based on the Tabnet model. The medical record diagnosis detection model of the Tabnet model is needed to find out the medical record homepage data that may have quality problems and feedback the analysis information, so that relevant personnel can improve the quality of the medical record homepage data in the future; and compared with the ordinary deep learning model, the Tabnet model The sparsity of features is fully utilized, and continuous and discrete features are processed respectively through two branches: dense features and sparse features. In the processing process, a feature reconstruction mechanism is introduced to improve the expressiveness of the model by learning the relationship between features. The TabNet model can directly input the original tabular data for training without any preprocessing, and uses gradient descent-based optimization for end-to-end learning, which improves the flexibility and integration of the model. The TabNet model uses a sequential attention mechanism to select which feature reasoning to use in each decision step, and selects different features according to different samples to achieve interpretability and better learning. The TabNet model has shown excellent performance in classification and regression problems in different fields, allowing two interpretabilities: global interpretability and support for local interpretability.

[0089] Step S4, based on the analysis and processing of the detection information of the medical record homepage, obtain the data classification quality control index of the medical record homepage; Step S4 includes the following sub-steps:

[0090] Step S401, based on the medical record homepage detection information, the number of medical record homepages with only correct medical records is counted, marked as X, the number of medical record homepages with only incorrect medical records is counted, marked as Y, and the number of medical record homepages with only abnormal diagnosis is counted, marked as Z;

[0091] Step S402, setting accuracy, completeness and consistency as data classification quality control indicators;

[0092] Among them, accuracy = [X / (X+Y+Z)]; completeness = [(X+Z) / (X+Y+Z)]; consistency

[0093] = 1-[Z / (X+Y+Z)];

[0094] For example, X=69524, Y=10665, Z=7954;

[0095] Accuracy = [69524 / (69524+10665+7954)] = 0.789;

[0096] Completeness = [(69524+7954) / (69524+10665+7954)] = 0.879;

[0097] Consistency = 1 - [7954 / (69524 + 10665 + 7954)] = 0.910;

[0098] Step S403, establish a quality control index evaluation model M, M = {D, I, R, W, S}, where D represents the set of medical record homepages that need to be evaluated for quality control index; I represents the data classification quality control index corresponding to the quality control index evaluation in D, in this embodiment, accuracy, completeness and consistency. R represents the quantitative calculation rule of I, in this embodiment, accuracy = [X / (X+Y+Z)]; completeness = [(X+Z) / (X+Y+Z)]; consistency = 1-[Z / (X+Y+Z)];

[0099] W represents the weight of the data classification quality control index in I; the weight value range is (0, 1], and the weight is 1. For example, in this embodiment, the accuracy weight is 0.30, the integrity weight is 0.35, and the consistency weight is 0.25; S represents the final result corresponding to R and W; in this embodiment, S = q1*accuracy+q2*integrity+q3*consistency, where q1, q2 and q3 represent weights, the value range is [0, 1], and q1+q2+q3=1;

[0100] For example, the accuracy weight is 0.30, the completeness weight is 0.35, and the consistency weight is 0.25; accuracy = 0.789; completeness = 0.879; consistency = 0.910, then

[0101] S=0.30*0.789+0.35*0.879+0.25*0.910=0.772;

[0102] Step S404, using the quality control index evaluation model M based on the detection information of the medical record homepage, obtain the data classification quality control index S.

[0103] In the specific implementation process; integrity refers to the proportion of data that exists and is error-free; consistency refers to the proportion of data with consistent descriptions, no anomalies and logical conformity; accuracy refers to the proportion of data that is accurate, without anomalies and errors; establishing a quality control index evaluation model to conduct quality control evaluation on the medical record homepage data can help monitor the data quality status of the medical record homepage in real time, promptly discover and resolve data quality risks on the medical record homepage, provide a reference for the optimization of medical record homepage data governance, and help maintain the high quality of medical record homepage data.

[0104] Example 2, please refer to Figure 4 As shown, Figure 4 The schematic diagram of the structure of an electronic device is illustrated, and the electronic device may include: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus. The memory stores computer-readable instructions, and the processor can call the instructions in the memory. When the computer-readable instructions are executed by the processor, the steps in a method for calculating a data classification quality control index based on deep learning are executed to achieve the following functions: based on the items to be filled in the medical record homepage, formulate the quality control rules for the medical record homepage, and establish the coding mapping rules to code and map the items to be filled in; collect the data of the medical record homepage, standardize the data of the medical record homepage, and establish a standard medical record data set; perform preliminary detection and evaluation on the medical record homepage based on the quality control rules of the medical record homepage to obtain preliminary detection information of the medical record; if the diagnosis items of the medical record homepage meet the quality control rules of the medical record homepage, use the medical record diagnosis detection model to detect the diagnosis items of the medical record homepage to obtain the medical record diagnosis detection information; based on the analysis and processing of the preliminary detection information of the medical record and the medical record diagnosis detection information, obtain the data classification quality control index of the medical record homepage.

[0105] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art, and the computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk.

[0106] Embodiment 3. The present application also provides a computer program product, which includes a computer program stored on a computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute a data classification quality control index calculation method based on deep learning provided by the above methods, the method including: based on the items to be filled in the medical record homepage, formulate medical record homepage quality control rules, and establish coding mapping rules to perform coding mapping on the items to be filled in; collect medical record homepage data, standardize the medical record homepage data, and establish a standard medical record data set; based on the medical record homepage quality control rules, perform preliminary detection and evaluation on the medical record homepage to obtain preliminary detection information of the medical record, if the diagnosis items on the medical record homepage meet the medical record homepage quality control rules, then use the medical record diagnosis detection model to detect the diagnosis items on the medical record homepage to obtain medical record diagnosis detection information; based on the analysis and processing of the preliminary detection information and the medical record diagnosis detection information, obtain the data classification quality control index of the medical record homepage.

[0107] Example 4. The present application also provides a computer-readable storage medium. The present application provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method for calculating data classification quality control index based on deep learning are executed to achieve the following functions: based on the items to be filled in on the medical record homepage, medical record homepage quality control rules are formulated, and coding mapping rules are established to perform coding mapping on the items to be filled in; medical record homepage data is collected, the medical record homepage data is standardized, and a standard medical record data set is established; based on the medical record homepage quality control rules, a preliminary detection and evaluation of the medical record homepage is performed to obtain preliminary medical record detection information. If the diagnosis items on the medical record homepage meet the medical record homepage quality control rules, the medical record diagnosis detection model is used to detect the diagnosis items on the medical record homepage to obtain medical record diagnosis detection information; based on the analysis and processing of the preliminary medical record detection information and the medical record diagnosis detection information, the data classification quality control index of the medical record homepage is obtained.

[0108] Through the description of the above implementation methods, the embodiments of the present invention can be provided as methods, systems or computer program products. Based on such an understanding, the above technical solutions can be essentially or partly contributed to the prior art in the form of software products, which can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and include several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0109] In the embodiments provided in the present application, it should be understood that the disclosed system or method can be implemented in other ways. The embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules or units 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 direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of systems, modules and units can be electrical, mechanical or other forms.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for calculating data classification quality control index based on deep learning, characterized in that: The steps include: Based on the items that need to be filled in the medical record front page, formulate the quality control rules for the medical record front page, and establish the coding mapping rules to map the codes for the items that need to be filled in; Collect medical record front page data, standardize the medical record front page data, and establish a standard medical record data set; Based on the quality control rules of the medical record homepage, a preliminary test and evaluation is performed on the medical record homepage to obtain preliminary test information of the medical record. If the diagnosis items on the medical record homepage meet the quality control rules of the medical record homepage, the medical record diagnosis test model is used to test the diagnosis items on the medical record homepage to obtain the medical record diagnosis test information; Based on the analysis and processing of the preliminary test information and the diagnostic test information of the medical records, the data classification quality control index of the medical record homepage is obtained.

2. The method for calculating data classification quality control index based on deep learning according to claim 1, characterized in that: Based on the items that need to be filled in the medical record homepage, the quality control rules for the medical record homepage are formulated, including the following sub-steps: Formulate basic quality control rules: Based on the basic filling specifications of the items to be filled in on the front page of the medical record and the disease diagnosis coding rules, formulate quality control rules for all items that need to be filled in and mark them as basic quality control rules; Formulate logical quality control rules: Based on the logical relationship between the items to be filled in the medical record homepage and the relationship between the disease diagnosis codes, formulate quality control rules between the items to be filled in the medical record homepage, marked as logical quality control rules; Mark the basic quality control rules and logical quality control rules as medical record homepage quality control rules.

3. The method for calculating data classification quality control index based on deep learning according to claim 2, characterized in that: Establishing code mapping rules and performing code mapping on the items to be filled in includes the following sub-steps: Establish coding mapping rules: Use one-hot coding to encode the items that need to be filled in on the first page of the medical record, and encode them in order from left to right and from top to bottom, and ensure that the codes of any two different items that need to be filled in are not repeated.

4. The method for calculating data classification quality control index based on deep learning according to claim 3, characterized in that: Collecting medical record front page data, standardizing the medical record front page data, and establishing a standard medical record data set includes the following sub-steps: Collect the medical record homepage data of historical patients based on the hospital database. The medical record homepage data includes hospitalization information, diagnosis and treatment information, patient information and cost information; According to the basic filling specifications of the medical record homepage, the hospitalization information, diagnosis and treatment information, patient information and expense information contained in the medical record homepage are standardized, and the standardized medical record homepage data is marked as the first medical record data; Based on the coding mapping rules, the first medical record data is coded and mapped respectively; ensuring that any two codes on the same medical record front page are not repeated, and that two codes of the same item on different medical record front pages are the same; The first medical record data after coding mapping is marked as a standard medical record data set.

5. The method for calculating data classification quality control index based on deep learning according to claim 4, characterized in that: Based on the quality control rules of the medical record homepage and using the medical record diagnosis detection model, the preliminary detection and evaluation of the medical record homepage includes the following sub-steps: Testing based on the quality control rules for the medical record homepage: Use the quality control rules for the medical record homepage to test the medical record homepage to determine whether each item to be filled in on the medical record homepage complies with the basic quality control rules and logical quality control rules for that item; record the item names and the number of items that comply with the basic quality control rules and logical quality control rules, and mark them as the correct medical record situation of the medical record; traverse all the medical record homepages to be tested, obtain the correct medical record situation of all the medical record homepages to be tested, and mark them as the correct medical record homepage information; record the item names and the number of items that do not comply with the basic quality control rules and logical quality control rules, and mark them as the medical record error situation of the medical record; traverse all the medical record homepages to be tested, obtain the medical record error situation of all the medical record homepages to be tested, and mark them as the medical record homepage error information; Mark the correct information and the wrong information of the medical record as preliminary detection information of the medical record; Based on the correct information in the medical record, the diagnostic items on the medical record front page that meet the basic quality control rules and logical quality control rules are screened. The diagnostic items on the medical record front page include outpatient and emergency diagnoses or major discharge diagnoses. The medical record diagnosis detection model is used to detect the diagnostic items on the medical record front page that meet the quality control rules of the medical record front page to obtain the medical record diagnosis detection information.

6. The method for calculating data classification quality control index based on deep learning according to claim 5, characterized in that: Using the medical record diagnosis detection model to detect the diagnosis items on the medical record homepage, obtaining the medical record diagnosis detection information includes the following sub-steps: Based on the standard medical record data set, outpatient and emergency diagnosis information, main discharge diagnosis information, other discharge diagnosis information, admission department information, discharge department information, and western medicine cost information from the front page of each medical record were extracted and marked as the model training data set; Mark outpatient and emergency diagnosis, main discharge diagnosis, other discharge diagnosis, admission department, discharge department, and western medicine expenses as target items; After completion, the diagnostic items and the corresponding related items are marked as the first training data set; the first training data set is divided into 10 parts on average, and each part is divided into a training set and a test set in a ratio of 7:3; Get 10 training sets and corresponding 10 test sets; The TabNet model is trained and optimized using the training set and test set.

7. The method for calculating data classification quality control index based on deep learning according to claim 6, characterized in that: Using the training set and test set to train and optimize the TabNet model includes the following sub-steps: Set the training hyperparameters of the TabNet model and the parameter range set of the model parameters; The parameter range set includes the value range of the training hyperparameter and the value range of the model parameter; the number of value steps of the training hyperparameter and the value step of the model parameter are set based on the value range; Use the relevant items as the input of the TabNet model and the diagnostic items as the output of the TabNet model; initialize the training hyperparameters and model parameters; Use 10 training sets to train the TabNet model for 10 rounds. After each round of training, use the test set to test it to obtain the model prediction accuracy of this round of training, save the model prediction accuracy of this round, and after completing 10 rounds of training and testing, use the 10 obtained model prediction accuracies to calculate the average value, and save the training hyperparameters and model parameters of these 10 rounds of training; then use grid search to change the training hyperparameters and model parameters, and use 10 training sets to train the TabNet model for 10 rounds again. After each round of training, use the corresponding test set to test it to obtain the model prediction accuracy of 10 rounds, calculate the model average accuracy, and save the training hyperparameters and model parameters of these 10 rounds of training; until all combinations of training hyperparameters and model parameters are traversed using grid search, the training hyperparameters and model parameters corresponding to the maximum model average accuracy are marked as the optimal parameters; Based on the optimal parameters, the TabNet model was trained using 10 training sets, and after completion, the medical record diagnosis and detection model was obtained; Detection based on the medical record diagnosis detection model: Based on the medical record diagnosis detection model, set a diagnostic item that needs to be tested, including outpatient and emergency diagnosis and major discharge diagnosis, use the medical record diagnosis detection model to make predictions, obtain the predicted content, compare the predicted content with the content of the diagnostic item in the corresponding position in the original medical record homepage, determine whether the content is consistent, record the name and number of items that need to be filled in for inconsistent content, mark them as diagnostic abnormalities of the medical record, traverse all medical record homepages to be tested, obtain the diagnostic abnormalities of all medical record homepages to be tested, and mark them as diagnostic abnormality information.

8. The method for calculating data classification quality control index based on deep learning according to claim 7, characterized in that: Based on the analysis and processing of the detection information on the medical record homepage, obtaining the data classification quality control index of the medical record homepage includes the following sub-steps: Based on the medical record homepage detection information, the number of medical record homepages with only correct medical records is counted, marked as X, the number of medical record homepages with only incorrect medical records is counted, marked as Y, and the number of medical record homepages with only abnormal diagnosis is counted, marked as Z; Set accuracy, completeness and consistency as data classification quality control indicators, where accuracy = [X / (X+Y+Z)]; completeness = [(X+Z) / (X+Y+Z)]; consistency = 1-[Z / (X+Y+Z)]; Establish a quality control index evaluation model M, M = {D, I, R, W, S}, where D represents the set of medical record homepages that need to be evaluated for quality control index; I represents the data classification quality control index corresponding to the quality control index evaluation in D, R represents the quantitative calculation rule of I; W represents the weight of the data classification quality control index in I; S represents the final result corresponding to R and W, S = q1*accuracy+q2*completeness+q3*consistency, where q1, q2 and q3 represent weights, the value range is [0, 1], and q1+q2+q3=1; Based on the detection information on the medical record homepage, the quality control index evaluation model output M is used to obtain the data classification quality control index S.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the steps in the method according to any one of claims 1 to 8 are executed.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are executed.

Images