Cervical cancer structured electronic medical record system
By using deep learning models and QR code recognition technology in the structured electronic medical record system of cervical cancer, the problem of unreliability of traditional electronic medical record data is solved, the immediacy and accuracy of data are achieved, and the reliability of medical data is improved. Normative.
Patent Information
- Application Number
- CN202211685195.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-27
- Publication Date
- 2024-12-13
AI Technical Summary
Traditional electronic medical records have shortcomings in the immediacy, accuracy, rigor and confidentiality of data, and lack of terminology standards, resulting in inefficiency and unreliability of data.
A structured electronic medical record system for cervical cancer is designed, including data acquisition and preprocessing modules, model building modules and medical record output modules, and the diagnostic data is cleaned using deep learning models (sequentially connected attention layer and bidirectional long and short-term neural network). and standardized output, verify data accuracy through QR code identification and TF-IDF algorithm to ensure data immediacy and accuracy.
It realizes the immediacy and rigor of data, ensures the accuracy and standardization of electronic medical records, improves the reliability and security of medical data, and enhances the efficiency of clinical and scientific research.
Smart Images
Figure CN119132483A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of medical management, and in particular to a structured electronic medical record system for cervical cancer. Background Art
[0002] The structured electronic medical record of cervical cancer is an important part of the long-term sustainable development of electronic medical records. The national structured electronic medical record platform for cervical cancer is under construction. We will continue to promote the construction of various forms of tumor medical alliances and medical communities, carry out tumor telemedicine services, and enhance the tumor diagnosis and treatment capabilities at the grassroots level.
[0003] With the increasing number of medical visits, more and more problems have emerged in traditional electronic medical records, such as low work efficiency and lack of good writing standards, resulting in different hospitals having no professional terminology standards for describing the same condition. Therefore, the research on structured electronic medical records is of great significance.
[0004] Traditional electronic medical records are written based on spot checks of patient medical records, and such data cannot be timely, effective, accurate, safe and reliable. For medical treatment, the immediacy, accuracy, rigor and confidentiality of data are very important.
[0005] At the same time, structured electronic medical records need to have strong standardization and accuracy. Therefore, standard cleaned electronic medical records as data will lay a good foundation for the realization of structured electronic medical records. Summary of the invention
[0006] The purpose of the present invention is to provide an accurate structured electronic medical record system for cervical cancer, which solves the above problems and improves the shortcomings of previous quality control, and is more conducive to clinical and scientific research.
[0007] The technical solution adopted by the present invention to achieve the above-mentioned purpose is:
[0008] Cervical cancer structured electronic medical record system, including:
[0009] The data acquisition and preprocessing module is used to acquire and clean the diagnostic data and load the cleaned data into the database;
[0010] Model building module, used to build deep learning models and call data in the database to train the models;
[0011] The medical record output module is used to output the input diagnostic data into standardized electronic medical records through a deep learning model.
[0012] The deep learning model includes sequentially connected attention layers and a bidirectional long short-term neural network.
[0013] The method for constructing a structured electronic medical record system for cervical cancer includes the following steps:
[0014] The data acquisition and preprocessing module acquires and cleans the diagnostic data, and loads the cleaned diagnostic data into the database;
[0015] The model building module builds a deep learning model and calls the data in the database to train the model;
[0016] The medical record output module outputs the input diagnostic data into standardized electronic medical records through a deep learning model.
[0017] The obtaining of diagnostic data specifically includes:
[0018] By scanning the QR code information marked on the paper medical record, the electronic medical record is obtained, and the QR code information is converted into text information;
[0019] Determine whether the parsed text information is a link. If so, jump to the link. Otherwise, pop up a prompt window to show that the current QR code is incorrect.
[0020] Use image recognition software to identify paper medical records and extract text information;
[0021] The TF-IDF algorithm was used to compare the text information converted from the QR code with the text information extracted by image recognition to verify the accuracy of the electronic medical records.
[0022] The TF-IDF algorithm is specifically:
[0023]
[0024] Cleaning of diagnostic data is as follows:
[0025] Delete spaces and error symbols in the diagnostic data, and compare the processed diagnostic data with the standardized diagnostic data to determine whether the diagnostic data is qualified.
[0026] Load the diagnostic data into the database using SQL statements or bulk data loading methods.
[0027] The data acquisition and preprocessing module needs to modify the diagnostic data and delete erroneous data before cleaning it.
[0028] A structured electronic medical record system for cervical cancer includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the method for constructing a structured electronic medical record system for cervical cancer when executing the computer program.
[0029] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for constructing a structured electronic medical record system for cervical cancer is implemented.
[0030] The present invention has the following beneficial effects and advantages:
[0031] 1. The present invention adopts the basic information and diagnostic information of patient data; the information entry is simple, convenient and fast, reflecting the effectiveness of the data; the cleaning work of electronic medical records reflects the timeliness and rigor of the data; the terminology standards are provided by diagnosis and treatment experts, reflecting the accuracy of the data.
[0032] 2. The present invention can provide data results instantly and accurately by reading the diagnostic information in the patient information in the database, and archives massive information from the database, and historical data can be checked at any time.
[0033] 3. Use the electronic medical records that have undergone the above two cleansing steps as the basic data for structured electronic medical records, and use deep learning models to enhance the persuasiveness of the model and train the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 This is the technical framework diagram of the structured electronic medical record system for cervical cancer. DETAILED DESCRIPTION
[0035] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.
[0036] like Figure 1 As shown, the structured electronic medical record system for cervical cancer includes the following steps:
[0037] During the electronic medical record acquisition phase, users make professional modifications based on the entered diagnostic data based on the patient information already available in the hospital and send it to the database for storage;
[0038] Use the QR code information marked on the medical record to obtain electronic medical records by scanning each time. In most cases, the QR code is a website. After the QR code is obtained, it can be converted into text by scanning the QR code through software, and then judge whether it is a link based on the parsed text result. If it is, jump to this link. At the same time, compare the text information obtained by image recognition software, compare the difference between the two in extracting text features, and use the TF-IDF algorithm to verify the effect of the two methods on feature extraction, mainly to observe the effect of extracting feature words. TF-IDF is used to evaluate the importance of words to a document in a document collection. The calculation formula of TF-IDF is:
[0039]
[0040] Since currently few hospitals have conducted corresponding research on the input part of electronic medical records, and this patent is mainly aimed at electronic medical records in the field of cervical cancer, these medical records are significantly different from those of other departments in terms of disease description, use of drugs, etc., and are therefore also the main innovation of the present invention.
[0041] The cleaning phase of the electronic medical record is used to read the diagnostic information in the patient information in the database, delete the special symbols in it using the Python program, and compare and analyze the existing data with the standardized data to determine whether the data is qualified as the input of the model;
[0042] This application uses a deep learning model to train the cervical tumor medical record symptom feature library to implement a semantic recognition engine, mainly for given data, to learn the correlation between data, and to lay the foundation for the subsequent structured electronic medical record system. Bidirectional long short-term neural network (BiLSTM) is a model proposed to solve the problem of gradient disappearance or gradient explosion in recurrent neural network (RNN). It better solves the defects of RNN in the long-dependency training process. Long short-term neural network (LSTM) is a unidirectional network that mainly uses a gate mechanism to solve the long-term dependency problem in RNN. LSTM proposes the concept of three gates in hidden units: input gate, output gate, and forget gate.
[0043] The detailed explanation of the LSTM model is as follows. The improved LSTM model adopts w H c t-1 Mapping to high-dimensional space, D t-1 Initialize with Δt, C t-1 and D t-1 Make the Hadamard product so that C t-1 Different dimensions have different discount rates. The calculation process is shown in formulas (1) to (4):
[0044] C t-1 =tanh(c t-1 w H +B H ) (Formula 1)
[0045]
[0046]
[0047]
[0048] Next is the standard gating architecture of LSTM, whose calculation process is shown in equations (5) to (8):
[0049] f t =σ(W f x t+U f h t-1 +b f ) (Formula 5)
[0050] i y =σ(W i x t +U i h t-1 +b i ) (Formula 6)
[0051] o t =σ(W o x t +U o h t-1 +b o )(Formula 7)
[0052]
[0053] Among them, f t 、i t , o t They are forget gate, input gate and output gate respectively, x t Indicates the current text data, h t-1 is the previous hidden state, is the candidate state, W, U, and b represent the text weight matrix, hidden layer weight matrix, and offset vector of the corresponding gate, respectively.
[0054] Finally, calculate the current memory value C t and the final hidden state h t As shown in formulas (9) to (10):
[0055]
[0056] h t =O t *tanh(C t )(Formula 10)
[0057] After introducing the model, let's talk about the data acquisition process. After the data is acquired and further cleaned, it needs to be saved to the database, that is, the converted and cleaned data is loaded into the target database. The best method for data loading generally depends on the amount of data to be loaded and the type of operation performed. If the target database is a relational database, there are two common data loading methods:
[0058] (1) Use SQL statements, namely insert, delete, and update operations to implement it.
[0059] (2) Use batch data loading methods, such as bulk, bcp, and batch loading tools unique to relational databases.
[0060] When data is imported into the database, it needs to comply with the table specifications of the database and be converted into the form required by the target data system. The erroneous and inconsistent data in it needs to be further cleaned and processed. It can be composed of data conversion, such as data merging, filtering, data reformatting, data recalculation, data reconstruction and data summarization. In the actual data conversion process, there are often various grammatical and semantic differences between the target data source and the source data source. These differences can be divided into: different data formats, the same data type in different databases may have different representation methods and language differences because there is no unified standard; data name conflicts, the name of the data in the data source may be a reserved word of the target data source, or the field already exists in the target data source, but the meaning is different; data type conflicts, the same data on the data source and the target data source but its data type is different, or the data type has differences in data accuracy; data structure conflicts, the data definition models between the data source and the target data source are different, so it is necessary to pay attention to the attributes and relationships of the table during conversion to avoid the loss of attributes and contact information. Data conversion requires that the data required by the target data source must be obtained from the data source and converted to the target data source, and at the same time requires that this conversion cannot contain erroneous and redundant associated information.
[0061] For the data in the saved database, you can enter the attention layer for further processing. Since the information contained in a single layer of attention may not be enough to support many downstream tasks, the Transformer launched by Google in 2017 stacks it into Multi-Head Attention (hereinafter referred to as MHA), which is essentially multiple Self-Attention calculations. The Attention mechanism will make the network pay more attention to the relevant information contained in the input during the training process, and simplify irrelevant information, thereby improving the accuracy of training. Self-Attention creates three new vectors for each word vector input: Query, Key, and Value. These three vectors are the product of the word vector and the three matrices Q, K, and V. The three matrices Q, K, and V are parameters that need to be learned. The implementation formula of the Attention mechanism is shown in Formula 11.
[0062]
[0063] The present invention uses the above-mentioned cleaned data as raw data, which can be directly obtained from the database, and uses the electronic medical record as the basis to train the structured electronic medical record. The BiLSTM model is used for training. The model will eventually generate a more standardized electronic medical record that meets the doctor's writing specifications according to the standard, which is used as the basis for the structured electronic medical record. Bidirectional long short-term neural network (BiLSTM) is a model proposed to solve the problem of gradient disappearance or gradient explosion in recurrent neural network (RNN), which better solves the defects of RNN in the long-dependency training process. Long short-term neural network (LSTM) is a unidirectional network that mainly uses a gate mechanism to solve the long-term dependency problem in RNN. LSTM proposes the concept of three gates in the hidden unit: input gate, output gate, and forget gate. BiLSTM extracts features from the input and obtains the probability of each word on all labels.
[0064] The structured electronic medical record system for cervical cancer includes a section for entering patient information, including basic information registration for patient name, ID number, contact person, telephone number, date of birth, place of residence, marital status, pregnancy and delivery history, cervical disease history, previous cancer screening history, etc.
[0065] The diagnosis registration is used to register the patient's first diagnosed hospital, discovery method, diagnosis method, pathology, pathological type, histological grade, auxiliary examinations, tumor marker information, other auxiliary examinations, etc.
[0066] Define a database to store user resource information and patient diagnosis and treatment information, establish an MVC underlying framework to build a database link, and assign permissions to registered user information based on the hospital level to which the user belongs. Users log in to enter patient information, first enter basic information: name, age, place of origin, etc., and then enter diagnosis and treatment information. Most of the information entered is drop-down options that users only need to select to enter, and a small part is an input box that users can fill in according to actual conditions. After entering, click the submit button at the bottom to save the patient data to the database table based on the link established with the database. The page prompts that the save is successful, and quality control information is automatically generated after the save is successful. Users can enter the query function to view the entered information.
[0067] After training the model, the model has learned the association information between the data. For subsequent testing, a new electronic medical record will be input. During the model learning process, the model will complete the output results of the corresponding entity based on what it has learned and the information in the current electronic medical record. It can be output on the console or page for reference.
[0068] The structured electronic medical record system for cervical tumors is used to identify the QR code information from the QR code scanning area, and verify the QR code information by comparing it with the paper image information, so as to ensure the authenticity of the structured electronic medical record in the software system. The upload module is used to upload the file to the storage module. The main purpose of doing this is to find out in the training model that if there is a problem with the original data set, the incorrect input text will have a greater impact on the subsequent training model, and have a certain interference effect on the output accuracy of the model. In order to solve this problem, a method of directly scanning the QR code to enter the electronic medical record text is proposed to ensure the accuracy of the text.
[0069] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of embodiments in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.
[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. The structured electronic medical record system for cervical cancer is characterized by: include: The data acquisition and preprocessing module is used to acquire and clean the diagnostic data and load the cleaned data into the database; Model building module, used to build deep learning models and call data in the database to train the models; The medical record output module is used to output the input diagnostic data into standardized electronic medical records through a deep learning model.
2. The structured electronic medical record system for cervical cancer according to claim 1, characterized in that: The deep learning model includes sequentially connected attention layers and a bidirectional long short-term neural network.
3. A method for constructing a structured electronic medical record system for cervical cancer, characterized in that: The following steps are involved: The data acquisition and preprocessing module acquires and cleans the diagnostic data, and loads the cleaned diagnostic data into the database; The model building module builds a deep learning model and calls the data in the database to train the model; The medical record output module outputs the input diagnostic data into standardized electronic medical records through a deep learning model.
4. The method for constructing a structured electronic medical record system for cervical cancer according to claim 3, characterized in that: The obtaining of diagnostic data specifically includes: By scanning the QR code information marked on the paper medical record, the electronic medical record is obtained, and the QR code information is converted into text information; Determine whether the parsed text information is a link. If so, jump to the link. Otherwise, pop up a prompt window to show that the current QR code is incorrect. Use image recognition software to identify paper medical records and extract text information; The TF-IDF algorithm was used to compare the text information converted from the QR code with the text information extracted by image recognition to verify the accuracy of the electronic medical records.
5. The method for constructing a structured electronic medical record system for cervical cancer according to claim 4, characterized in that: The TF-IDF algorithm is specifically:
6. The method for constructing a structured electronic medical record system for cervical cancer according to claim 3, characterized in that: The specific steps for cleaning diagnostic data are as follows: Delete spaces and error symbols in the diagnostic data, and compare the processed diagnostic data with the standardized diagnostic data to determine whether the diagnostic data is qualified.
7. The method for constructing a structured electronic medical record system for cervical cancer according to claim 3, characterized in that: Load the diagnostic data into the database using SQL statements or bulk data loading methods.
8. The method for constructing a structured electronic medical record system for cervical cancer according to claim 3, characterized in that: The data acquisition and preprocessing module needs to modify the diagnostic data and delete erroneous data before cleaning it.
9. The structured electronic medical record system for cervical cancer is characterized by: It comprises a memory and a processor; the memory is used to store a computer program; the processor is used to implement the method for constructing a structured electronic medical record system for cervical cancer as described in any one of claims 3-8 when executing the computer program.
10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the method for constructing a structured electronic medical record system for cervical cancer as described in any one of claims 3-8 is implemented.