A medical literature text extraction system for hypertension

Abstract

A medical literature text extraction system for hypertension includes a literature knowledge extraction unit, which includes: an external knowledge base collection unit, a crawler download document unit, a book database scan to PDF unit, a PDF document aggregation unit, a PDF file conversion to OCR image data unit, an OCR image data conversion to text unit, a non-text content cleaning unit, a document content splitting unit, a pronoun replacement unit, a paragraph splitting unit, a compound sentence splitting unit, a word splitting unit, and a named entity recognition unit, which is used to identify all the word segments obtained by the word splitting unit as corresponding entities, wherein an entity is composed of one or more word segments, and each entity represents its own independent meaning. The present disclosure can overcome the problem that the popularization of hypertension medication knowledge itself faces obstacles in practice, and can also intelligently organize the specialized knowledge of hypertension and improve the diagnosis and treatment level of hypertension.

Description

Technical Field

[0001] The present disclosure belongs to the technical field of medical document processing, and in particular relates to a medical document text extraction system for hypertension. Background Art

[0002] Hundreds of medical papers are published daily regarding hypertension medications, making it difficult for doctors to keep up with the latest research. Furthermore, the update cycle for specific medical guidelines is long, typically lasting ten years. Therefore, the dissemination of knowledge about hypertension medications faces practical obstacles.

[0003] Furthermore, there is a lack of disease-specific knowledge on hypertension. Currently, the issue of medical knowledge organization is widely recognized, and the industry consensus is to organize medical knowledge through knowledge graph database technology. The problem is that public knowledge graph projects are generally based on a wide range of diseases, with shallow knowledge connections, and rarely delve into specific disease-specific knowledge graphs. In reality, there is no knowledge graph specifically designed for hypertension.

[0004] In view of this, how to use the new generation of information technology to realize a medical literature text extraction system for hypertension has become a technical problem that needs to be solved urgently in this field. Summary of the Invention

[0005] In order to solve the above technical problems, the present disclosure provides a medical literature text extraction system for hypertension, which is characterized by:

[0006] The system includes a literature knowledge extraction unit for acquiring medical literature related to hypertension, and using machine learning technology to extract medication methods of hypertension specialists from the medical literature and discover new medication knowledge therefrom.

[0007] Preferably, in the medical literature text extraction system for hypertension, the literature knowledge extraction unit includes:

[0008] Collect external knowledge base units to collect literature sources related to the medication treatment of hypertension in online databases and knowledge bases of Chinese and foreign medical literature.

[0009] Preferably, in the medical literature text extraction system for hypertension, the literature knowledge extraction unit includes:

[0010] The crawler download document unit is used to download documents related to the medication treatment of hypertension based on the document sources collected from the external knowledge base unit.

[0011] Preferably, in the medical literature text extraction system for hypertension, the literature knowledge extraction unit includes:

[0012] The book database scan-to-PDF unit is used to convert scanned books and materials related to hypertension medication and treatment into PDF files.

[0013] Preferably, in the medical literature text extraction system for hypertension, the literature knowledge extraction unit includes:

[0014] The PDF document aggregation unit is used to aggregate the documents downloaded by the crawler download document unit and the PDF files converted by the book database scan to PDF unit in one folder.

[0015] Preferably, in the medical literature text extraction system for hypertension, the literature knowledge extraction unit includes:

[0016] The PDF file is converted into an OCR image data unit, which is used to convert all the PDF files summarized by the summary PDF document unit into one page of image data.

[0017] Preferably, in the medical literature text extraction system for hypertension, the literature knowledge extraction unit includes:

[0018] OCR image data is converted into text units, which is used to convert PDF files into OCR image data units. The image data obtained contains text, and the text is recognized through OCR recognition technology.

[0019] Preferably, in the medical literature text extraction system for hypertension, the literature knowledge extraction unit includes:

[0020] Clean up non-text content units, which are used to convert OCR image data into text units, reorganize the text recognized, sort out the recognized text into sentences, and exclude non-text characters, where non-text characters include text in images or tables.

[0021] Preferably, in the medical literature text extraction system for hypertension, the literature knowledge extraction unit includes:

[0022] The document content splitting unit is used to split the final text obtained by cleaning the non-text content units into different sections according to the article structure of the original PDF file, and exclude content other than text paragraphs, including equation text, etc., wherein the different sections include the text body, possible abstracts, and possible reference sections.

[0023] Preferably, in the medical literature text extraction system for hypertension, the literature knowledge extraction unit includes:

[0024] A pronoun replacement unit, used to replace the pronouns in the final text obtained by splitting the document content into units with the specific nouns they refer to;

[0025] A paragraph splitting unit is used to split each paragraph in the final text obtained from the pronoun replacement unit into different sentences using punctuation marks such as periods and exclamation points, and to identify compound sentences and single sentences therein;

[0026] A compound sentence disassembling unit is used to disassemble each compound sentence in the final text obtained by the paragraph disassembling unit into separate sentences;

[0027] A word segmentation unit is used to mark the segmentations of the individual sentences obtained by the compound sentence segmentation unit, wherein the segmentations include individual Chinese and foreign characters;

[0028] The named entity recognition unit is used to identify all the segmented words obtained by the segmentation unit as corresponding entities, where an entity consists of one or more segmented words, and each entity represents its own independent meaning.

[0029] Compared with the existing technology, the present disclosure can overcome the practical obstacles faced by the popularization of knowledge on hypertension medication, and can also intelligently organize specialized knowledge on hypertension and improve the diagnosis and treatment level of hypertension.

[0030] In addition, this disclosure discloses the following technical contributions:

[0031] 1. System and method for extracting knowledge text from medical literature;

[0032] 2. A system and method for discovering new medication knowledge from medication knowledge;

[0033] 3. A system and method for converting hypertension clinical data into graph data;

[0034] 4. Systems and methods for integrating medical knowledge graphs with clinical data. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings illustrate exemplary embodiments of the present disclosure and together with the description serve to explain the principles of the present disclosure. These drawings are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification.

[0036] Figure 1 is a schematic diagram of a medication recommendation system for hypertension in one embodiment of the present disclosure;

[0037] Figure 2 is a schematic diagram of document knowledge extraction in one embodiment of the present disclosure;

[0038] Figure 3AThis is a schematic diagram of a graph database storage structure in one embodiment of the present disclosure;

[0039] Figure 3B is a corresponding schematic diagram of document processing in one embodiment of the present disclosure;

[0040] Figure 4A is a schematic diagram of patient data collection in one embodiment of the present disclosure;

[0041] Figure 4B is a schematic diagram of storing patient data in a database in a graphical form in one embodiment of the present disclosure;

[0042] Figure 5A is a schematic diagram of knowledge graph data integration in one embodiment of the present disclosure;

[0043] Figure 5B is a schematic diagram of knowledge decomposition in one embodiment of the present disclosure;

[0044] Figure 5C is a schematic diagram of simultaneously storing knowledge graph data and patient data in one embodiment of the present disclosure;

[0045] Figure 6A This is a schematic diagram of generating medication recommendation rules in one embodiment of the present disclosure;

[0046] Figure 6B is a schematic diagram of a medication recommendation rule suggesting additional administration of a CCB drug in one embodiment of the present disclosure;

[0047] Figure 7 is a schematic diagram of the knowledge relationship between a certain drug and hypertension in one embodiment of the present disclosure;

[0048] Figure 8 This is a schematic diagram of generating a personalized medication recommendation model in one embodiment of the present disclosure;

[0049] Figure 9 is a schematic diagram of a medication recommendation decision-making system in one embodiment of the present disclosure;

[0050] Figure 10 It is a schematic diagram of continuous optimization of a medication recommendation decision system in one embodiment of the present disclosure. DETAILED DESCRIPTION

[0051] The present disclosure will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only intended to explain the relevant content and are not intended to limit the present disclosure. It should also be noted that, for ease of description, only the portions relevant to the present disclosure are shown in the accompanying drawings.

[0052] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in the present disclosure can be combined with each other. The technical solution of the present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0053] Unless otherwise stated, the exemplary embodiments / examples shown are to be understood as providing exemplary features of various details of some ways in which the technical concepts of the present disclosure can be implemented in practice. Therefore, unless otherwise stated, the features of the various embodiments / examples may be further combined, separated, interchanged, and / or rearranged without departing from the technical concepts of the present disclosure.

[0054] Cross hatching and / or shading may be used in the drawings to generally clarify the boundaries between adjacent components. As such, unless otherwise indicated, the presence or absence of cross hatching or shading does not convey or indicate any preference or requirement for the specific materials, material properties, dimensions, proportions, commonalities between the components shown, and / or any other characteristics, attributes, properties, etc. of the components. In addition, in the drawings, the sizes and relative sizes of the components may be exaggerated for clarity and / or descriptive purposes. When the exemplary embodiments can be implemented differently, the specific process sequence can be performed in a different order than the described step sequence. For example, two successively described processes can be performed substantially simultaneously or in an order opposite to the described order. In addition, like reference numerals represent like components.

[0055] When a component is referred to as being “on,” “over,” “connected to,” or “coupled to” another component, the component may be directly on, directly connected to, or directly coupled to the other component, or intervening components may be present. However, when a component is referred to as being “directly on,” “directly connected to,” or “directly coupled to” another component, there are no intervening components present. For this purpose, the term “connected” may refer to a physical connection, an electrical connection, etc., with or without intervening components.

[0056] For descriptive purposes, the present disclosure may use spatially relative terms such as "below," "beneath," "under," "down," "above," "upper," "above," "higher," and "side (e.g., as in "sidewall")," to describe the relationship of one component to another (other) component as shown in the accompanying drawings. The spatially relative terms are intended to encompass different orientations of the device in use, operation, and / or manufacture in addition to the orientation depicted in the accompanying drawings. For example, if the device in the drawings is turned over, a component described as "below" or "beneath" another component or feature would then be positioned "above" the other component or feature. Thus, the exemplary term "below" can encompass both the "above" and "below" orientations. Furthermore, the device may be otherwise oriented (e.g., rotated 90 degrees or at other orientations), and as such, the spatially relative descriptors used herein should be interpreted accordingly.

[0057] The terms used herein are for the purpose of describing specific embodiments and are not intended to be restrictive. As used herein, unless the context clearly indicates otherwise, the singular forms "one (kind, person)" and "said (the)" are also intended to include plural forms. In addition, when the terms "comprise" and / or "include" and their variations are used in this specification, the features, integral bodies, steps, operations, parts, assemblies and / or their groups stated are explained, but the presence or addition of one or more other features, integral bodies, steps, operations, parts, assemblies and / or their groups is not excluded. It should also be noted that, as used herein, the terms "substantially", "approximately" and other similar terms are used as approximate terms and not as degree terms, so that they are used to explain the inherent deviations of the measured values, calculated values ​​and / or the values ​​provided that will be recognized by those of ordinary skill in the art.

[0058] See also Figure 1 In one embodiment, the present disclosure discloses a medication recommendation system for hypertension, wherein:

[0059] 100. The system can access medical literature related to hypertension from online medical literature repositories, thereby learning from the medication experience of hypertension specialists at leading hospitals. This experience is distributed across a vast amount of literature. This medical literature includes authoritative guidelines, expert consensus, clinical research and authoritative consensus, drug instructions, and physician experience. Their summaries of treatment methods and the sources of the literature will become the extracted knowledge.

[0060] Text processing utilizes a wide range of typical natural language processing (NLP) processes, built using the Python Spacy library and various HuggingFace model libraries. Open-source tools are initially used for execution, and if omissions or errors are discovered during manual proofreading, manual rules are added for precise calibration.

[0061] 200. The system connects to an external online patient management system via an Application Programming Interface (API). This system collects patient data, physician information, prescriptions, medication records, and more. A data structure conversion program is then developed for each data source, converting the data from that source into a database of hypertension patient medications. This system can connect to multiple data sources.

[0062] 300. The data organization and structure (schema) between the knowledge database and the clinical database must be consistent to ensure that knowledge concepts are consistent with related data definitions. Synonymous concepts should be expressed using the same conceptual terms. After data attribute correction, hypertension-related knowledge and patient data are imported into the hypertension atlas database, and the corresponding clinical data of the knowledge can be clearly identified.

[0063] 400. The results of medication recommendations need to have a basis, that is, they need knowledge traceability and explainability. This is because the design subject of the medication recommendation decision-making system is the rules of the relationship between patient attributes such as symptoms and drug use. The system can accept manual input of rules compiled by doctor experts, or discover the correlation between patient attributes and medication effects through knowledge graphs. If these rules have corresponding patient treatment data, relevant medication efficacy data can be collected to help experts verify the formality of the rules. Plans that are recognized by experts and have a source of evidence can be imported into the medication recommendation decision-making system for use.

[0064] 500. In addition to the manually verified rules in step 500, the medication recommendation decision system can also evaluate medication recommendations for a specific patient through graph-based machine learning. This machine learning is based on the medication history and efficacy data of other patients. The calculation goal is to estimate the most suitable medication for a particular patient. This can effectively identify new potential patient usage scenarios for drugs with a certain amount of medication data. This type of machine learning-based medication recommendation has personalized value.

[0065] 600. The Hypertension Medication Recommendation Decision System provides comprehensive assistance to physicians in assisting with hypertension medication decisions. It first eliminates secondary hypertension, referring patients with hypertension symptoms caused by other medical conditions to other departments. The system then focuses on providing medication recommendations for patients with primary hypertension. The system prioritizes knowledge-based rule-based medication recommendations, while machine learning-based medication recommendations serve as supplementary recommendations.

[0066] 700. The adoption of medication recommendations is recorded and stored in the hypertension-specific knowledge graph database, allowing us to adjust the weight of rule rankings, giving approved rules a greater chance of being recommended, especially for patients with limited clinical data. Collecting more data will also help improve the quality of machine learning medication recommendations.

[0067] In another embodiment, see Figure 2 , describe in more detail the execution process of the aforementioned 100 document knowledge extraction:

[0068] 101. Find online databases of medical literature related to the treatment of hypertension, such as PubMed, Embase, etc. You can also import Chinese knowledge bases, such as SinoMed, CBM, etc. You can enter options such as keywords above to download

[0069] 102. Since the number of relevant documents is large, you will need to use a computer program to download the documents. You can use a Python script to simulate an automated search. If the database uses the BeautifulSoup web parsing library, then use the requests link download library to download the PDF documents of the documents.

[0070] 103. In addition, there may be medical textbooks whose relevant pages can be manually scanned to form PDF documents.

[0071] 104. Gather all the PDF documents that need to be imported into a folder.

[0072] 105. To extract text from a PDF document, you need to convert the PDF into a page and a picture through a program.

[0073] 106. Use optical character recognition (OCR) to identify text in images.

[0074] 107. The text scanned by OCR needs to be reorganized, including the recognized text being put back into complete sentences. Non-text characters, such as text information in images or charts, should be excluded.

[0075] 108. Break down the text of a PDF document into sections. Generally, the first section of a document is the abstract, the middle section is the main text, and the last section is the source references. Extract the abstract and the main text. Exclude all content outside the text paragraphs, including equations and text. The remaining text will undergo in-depth text analysis.

[0076] 109. The reference section should be separated out separately as a reference for tracing the source of knowledge, and then merged with the main content of the article to remove duplication in step 116. The main content can be divided into a paragraph.

[0077] 110. Pronouns like "he" and "that" in the text need to be replaced with the nouns they refer to. This is coreference resolution in natural language processing, achieved using existing open-source tools followed by manual correction.

[0078] 111. Break text paragraphs into separate sentences using punctuation marks such as periods and exclamation points.

[0079] 112. If a sentence is composed of separate sentences, separate the separate sentences.

[0080] 113. Mark each individual sentence with tokens, which are individual words in Chinese and English.

[0081] 114. An entity can be composed of one or more participles, each representing a complete and independent word meaning. Both English and Chinese have NLP grammar analysis tools that can effectively perform this step.

[0082] The results of this grammatical analysis step may not be perfect, and you need to manage the analysis rules yourself to improve accuracy. In step 303, you will get more language processing rules defined in the knowledge graph correction, which are introduced here for execution.

[0083] 115. Knowledge is stored in the form of a graph database. In one embodiment, the storage format of the graph database is as follows: Figure 3A In this example, A through E represent nodes in a graph database, R1 through R7 represent relationships between entities, and P1 through P10 represent properties of entities. In a graph database, there are two main components: node sets and relationships connecting nodes.

[0084] Within the same sentence, the entities of primary interest are the subject, verb, and object (subject-verb-object). Adjectives become entities, establishing a relationship with the subject. Verbs and auxiliary verbs together form the relationship type. The subject and object become nodes, respectively, with the verb representing the relationship from the subject to the object. Reference link information annotated in each sentence is also included and generated as attribute information for this relationship.

[0085] For example, the following statement is found in the literature: Patients with blood pressure ≥140 / 90 mmHg are directly classified as high cardiovascular risk. By breaking down SVO, the following structure can be made: S is the patient, and blood pressure ≥140 / 90 mmHg is a description of the patient. Cardiovascular risk has been annotated as a complete entity (not just risk) in the previous step. High risk is a description of cardiovascular risk. For more details, see Figure 3B shown.

[0086] 116. All the node and relationship pairs under the entire document are collected, deduplicated and merged first, and the attribute information of the node and relationship pairs are merged together. Then the graph database is queried to see if they already exist.

[0087] 117. Only newly paired information will be prepared for import into the graph database.

[0088] In another embodiment, see Figure 4A , describes in more detail the execution process of the aforementioned 200 clinical data collection:

[0089] 201. External clinical management and health management behavior data for hypertensive patients is accessed through an online API interface. The finer the data granularity, the better. Required data categories include individual blood pressure measurements, medication records, and patient baseline and follow-up medical records.

[0090] 202. If the attribute fields of the patient data already exist in the knowledge graph database (introduced from step 305), the same corresponding fields should be used to ensure unified data definition.

[0091] 203. Patient data will also have other new fields, which the system supports adding to the schema dictionary of the entire system. The storage format of the graph database can flexibly support the storage of any data structure.

[0092] 204. After the field definitions are clear, extract the data by extracting, transforming, and loading data that is adapted to the data source and organize it according to the (structured) field definitions.

[0093] 205. If the above data is incomplete, it can be excluded or processed through data cleaning or transformation. The established hypertension medication knowledge graph can be used to guide the data cleaning, data labeling, and structuring of hypertension patient data.

[0094] 206. Then the data can be directly put into the graph database. The same process is used for other data sources.

[0095] For example, we can import the following data records from the 201 external data system:

[0096] {

[0097] "patient number":"HZ0001",

[0098] "Type":"History",

[0099] "High Blood Pressure": 1,

[0100] "diabetes":0,

[0101] "Hyperlipidemia":0,

[0102] "Coronary heart disease": 1,

[0103] "Other system diseases": 1,

[0104] "Other systemic diseases": "liver cirrhosis, chronic obstructive pulmonary disease",

[0105] "Cardiovascular disease history": ["angina pectoris", "myocardial infarction", "peripheral arteriosclerosis"],

[0106] "Respiratory system history": ["asthma", "chronic obstructive pulmonary disease"],

[0107] "Hyperuricemia":0

[0108] }

[0109] The "Type" field indicates that this data is related to medical history. The disease name fields appearing here describe the medical history of patients with hypertension. If coronary heart disease or peripheral atherosclerosis are not already recorded as medical history fields for hypertension diagnoses, they will be added to the naming dictionary at step 203.

[0110] Patient data are stored in the database in the form of graphs, e.g. Figure 4B The records shown.

[0111] In another embodiment, see Figure 5A , describes in more detail the execution process of the aforementioned 300 knowledge graph data integration:

[0112] 301. Corresponding definitions are needed for the knowledge graph and patient data fields input from steps 117 and 206, respectively, maintaining a single data structure (schema) for comprehensive management. Field unification is an iterative process, requiring continuous comparison of new literature knowledge fields with new patient data to identify field definitions that require unification. Data processing requirements will guide modifications in steps 100 and 200.

[0113] 302. If you find that the knowledge graph structure is not complete enough, you need to add or modify fields in the knowledge graph.

[0114] 303. Convert the modified field information into new document knowledge extraction rules and pass it to step 114 for processing.

[0115] 304. If it is found that the patient data structure is not complete, it is necessary to add or modify fields during the patient data extraction process.

[0116] 305. Pass the patient attribute modification field to step 202 for processing.

[0117] 306. Medical knowledge is extensive, and not all literature is necessarily related to hypertension medication. Therefore, to address the application issues of this system, only knowledge data related to hypertension and medication treatment plans is imported and stored in the hypertension disease knowledge graph database.

[0118] 307. Patient data should be imported into the hypertension knowledge graph database along with the modified (structure) fields. This process can be done in real time or periodically, for example, once a day.

[0119] For example, there is a sentence in the literature: Hypertensive patients who have coronary heart disease, ischemic stroke, or peripheral atherosclerosis should take statins for a long time and, if necessary, other lipid-lowering drugs to reduce low-density lipoprotein cholesterol (LDL-C) to below 1.8mmol / L (70mg / dl). Then, this knowledge will be broken down into Figure 5B The situation shown. It should be noted that this is a temporary result of the decomposition. New terms will be encountered in the new sentences, such as coronary heart disease and peripheral atherosclerosis. The sentence itself is related to the knowledge node "hypertensive patients," but the definition of this connection is not yet clear. New discrimination rules will be introduced in 303 to clarify that coronary heart disease and peripheral atherosclerosis are also past histories of hypertensive patients.

[0120] Regarding the 301 field correspondence, the record mentions diabetes and hyperlipidemia, which may not have appeared in the knowledge graph yet.

[0121] If the literature can extract the relationship between hypertension and diabetes, hyperlipidemia, etc., mark these relationships as past history. For example, see Figure 4B As shown in the graph dictionary, new terms are also encountered, such as statins, which are drugs related to hypertension. If external data contains information about "statin" and "drug", this can be understood as referring to statins.

[0122] It should be noted that graph data can also record more complex knowledge systems. For example, the definition of "high-risk" cardiovascular risk is as follows:

[0123] (1) Patients with blood pressure ≥140 / 90 mmHg are directly classified as high-risk cardiovascular risk. (2) Patients with systolic blood pressure of 130-139 mmHg and / or diastolic blood pressure of 80-89 mmHg, accompanied by clinical comorbidities, have a significantly increased risk of cardiovascular events or death. Active secondary prevention, including comprehensive treatment including active blood pressure reduction, can reduce the risk of cardiovascular outcome events and death, and are therefore classified as high-risk cardiovascular risk. (3) Patients with systolic blood pressure of 130-139 mmHg and / or diastolic blood pressure of 80-89 mmHg and target organ damage, active blood pressure reduction is beneficial in delaying the progression of target organ damage, and are therefore also classified as high-risk cardiovascular risk. (4) Patients with systolic blood pressure of 130-139 mmHg and / or diastolic blood pressure of 80-89 mmHg and ≥3 cardiovascular risk factors are classified as high-risk. The main reference is to the relevant recommendations of the 2020 Chinese Primary Prevention Guidelines for Cardiovascular Disease, as shown in Table 1:

[0124] Table 1. Important factors affecting cardiovascular risk stratification in hypertensive patients *

[0125]

[0126]

[0127] *See the full guide for detailed instructions

[0128] For the above table, the data is expressed as follows Figure 5B The graph structure shown in Figure 2 shows the relationship between the two graphs. Different rules can be used to process queries based on the different relationship attributes, thus implementing this rule.

[0129] like Figure 5C The figure below is an example of fusion. In a specialized knowledge graph database for hypertension, both knowledge graph data and patient data are stored. This example is a patient's personal health record with a history of stroke. Since the "stroke" node is also included in the knowledge data for hypertension assessment and examination recommendations, fusion can generate knowledge-driven application scenarios for patient data.

[0130] In another embodiment, see Figure 6A , describes in more detail the execution process of generating the aforementioned 400 medication recommendation rules:

[0131] 401. Derive knowledge relationships from the hypertension disease knowledge graph database and display them on a visual interface, allowing doctors and experts to manually determine the relationship path between medication, condition data, and hypertension treatment. Since there are many potential relationship data, the present disclosure can use a certain drug as a starting point to draw a potential association path with hypertension, for example Figure 7 As shown, the defined relationship paths are displayed in a different color than other paths, allowing medical experts to browse and select potential new medication paths. Medical experts can decide whether to test the validity of the relationship paths using patient medication data. If not, they can skip to step 405 to generate rules.

[0132] 402. Alternatively, doctors can develop medication rules based on their personal experience and knowledge. However, these rules must be described within the existing field definition structure. Doctors can decide whether to verify the effectiveness of these rules using patient medication data. If not, they can proceed directly to step 405 to generate rules.

[0133] Figure 6B It shows an example of a medication recommendation rule, which means that if the patient has a history of myocardial infarction, has been taking ACEI, and has high diastolic blood pressure, it is recommended to take CCB drugs in addition.

[0134] 403. Rules that can verify clinical effectiveness are based on data comparing the use of one or more medications, other patient attributes, and historical changes in blood pressure measurements. As long as this data exists and meets the minimum statistical sampling requirements, the effectiveness of the relevant medication rules can be verified.

[0135] 404. The expert group of the subject will verify and supplement the rules for hypertension medication. If the data lacks positive evidence, the rules can be adjusted and verified repeatedly.

[0136] 405. General medication rules are based on the knowledge that taking a particular drug or combination of drugs produces a positive therapeutic effect. There are also rules called contraindications, which describe medications that pose a greater risk of side effects if the patient meets certain conditions. These contraindications need to be added.

[0137] 406. All medication rule combinations verified by experts can be output to the medication recommendation decision system for use.

[0138] In another embodiment, see Figure 8 The execution process of generating the aforementioned 500 personalized medication recommendations is described in more detail:

[0139] 501. Extract all graph relationship data from patient nodes to drug nodes from the hypertension disease knowledge graph database.

[0140] 502. Use machine learning tools to train and generate graph embedding vector functions for each node, using models and algorithms from the knowledge graph field. For example, TransE, RotatE, DistMult, ComplEx, ConvE, ConvKB, CompGCN, NodePiece, and other models are all suitable.

[0141] Graph embedding data can be used to estimate previously unestablished relationships between patients and medications. These relationships lead to new medication recommendations. The algorithm can select the top n most compatible medications for any given patient.

[0142] 504. By default, the system selects the top n recommended drugs for a patient. The specific top number can be customized according to the doctor's needs, for example, n can be 2 to 5. The recommendation results will be used in the medication recommendation decision system.

[0143] It should be noted that the specific steps of steps 502 to 503 are as follows:

[0144] Step 1. Prepare basic data:

[0145] For each patient in the disease-specific knowledge graph database, medication recommendation results (i.e., drugs) are executed according to the medication rules, and the medication recommendation results are marked to form basic data; for example, the medication recommendation results are marked as the "has_drug_recommendation" relationship data of the corresponding patient; of course, there are other types of relationships that need to be marked.

[0146] Preferably, before executing the medication recommendation results according to the medication rules, the medication rules with a low level of evidence are first filtered out;

[0147] For example, the following Table 3 shows the basic data obtained after filtering out medication rules with a low level of evidence:

[0148] Table 3

[0149] source target type 2915 375 has_prescription 2915 374 has_physical_exam 2915 379 has_physical exam 2915 380 has physical exam 2915 384 has_physical_exam 2915 378 has_physical_exam 2915 376 has_physical_exam 2915 377 has_physical exam 2915 2916 has blood_pressure 2915 148 has_blood_pressure_range 2915 154 has_blood_pressure_range 2915 155 has_blood_pressure_range 2915 147 has_blood_pressure_range 2915 414 has_family_history 2915 413 has_family_history 2915 417 has_family_history 2915 418 has_drug_recommendation 2915 270 has_present_medical_history 2915 168 haspast_medical history

[0150] Step 2: Organize the data:

[0151] Extract the patients and drugs from the labeled basic data in the form of digital IDs, and create corresponding patient nodes and drug nodes to form a data set;

[0152] Step 3: Classify the extracted data set:

[0153] Classifying the data set to obtain a training set, a test set, and a validation set, wherein the training set, the test set, and the validation set are randomly divided from the data set in a ratio of 80%, 10%, and 10%;

[0154] It should be noted that the training set, test set, and validation set here all belong to the graph database described in this disclosure; all information in the graph database is a subset extracted from the disease-specific knowledge graph database;

[0155] Step 4: Training

[0156] Generate a training model using any suitable algorithm from the various algorithms available in the knowledge graph field, such as RotatE;

[0157] The total number of sample training times is not less than 20, and the graph embedding size is not less than 512;

[0158] In the case of non-convergence, one or more parameters are modified according to experience to improve the performance of a certain evaluation metric, so that the training process can continue until the training is completed.

[0159] Step 5: Model Reasoning

[0160] For the training model, attach any patient, or patient information that will be added to the graph database in the future, and the training model can infer the nodes with the "has_drug_recommendation" relationship coefficient;

[0161] Step 6: Save the recommended results:

[0162] The relationship coefficients are sorted from high to low to keep the top two nodes with inference relationships as the inference results, and are stored in the graph database as personalized recommendation results through relationship tags that represent personalized recommended drugs, such as the "has_predicted_drug_recommendation" relationship tag.

[0163] In another embodiment, see Figure 9 , describes in more detail the principles of the aforementioned 600 medication recommendation decision system:

[0164] 601. The use case of the medication recommendation decision system is: Before a doctor prescribes a prescription for a patient during a face-to-face consultation (either in person or virtually), he or she queries the system, which can be accessed through a web browser or mobile device. First, all the patient's physical examination data is retrieved from the disease-specific knowledge graph database.

[0165] 602. The medication recommendation decision system internally records a list of standard hypertension physical examination fields. It recommends that patients complete all of these examination items for a complete hypertension diagnosis. However, in reality, patients often do not have comprehensive physical examination records. Missing examination items are displayed here to the doctor, allowing the doctor to know which examination items the patient should complete first.

[0166] 603. Hypertension is divided into two types: primary and secondary. Primary hypertension is a condition caused by an individual's long-term, unexplained history. Secondary hypertension occurs when the patient has another medical condition and presents as a complication. Therefore, treatment of secondary hypertension requires referral rather than prioritizing the underlying hypertension. The medication recommendation system only makes medication recommendations for primary hypertension.

[0167] 604. The patient's hypertension is determined to be secondary hypertension, and the system provides referral recommendations and reasons.

[0168] 605. Step 406 obtains the medication rules that match the patient's personal data. Matching rules is performed against the patient's personal data and sorted in reverse order by rule weight. Recommended medications are output in the same order after deduplication, but the rule or rules corresponding to each medication are retained for future reference. The patient's personalized top five recommended medications are retrieved from step 504. Any duplicates in these medications are excluded from the recommended rules.

[0169] The following is an example of how to match medication rules, including the following steps:

[0170] S1. Find all the recommended rules for medication;

[0171] S2. Find the corresponding medication conditions for each rule;

[0172] S3. According to each rule, check whether the patient meets all the medication conditions;

[0173] S4. If yes, please send back the medication rules and list the corresponding recommended drugs.

[0174] The recommended results are similar to the following:

[0175]

[0176] 606. Input the patient information from step 405 and query the drug contraindication rules related to the patient. The drugs found should be removed from the drug results of step 605.

[0177] 607. Drugs in the medication rules and additional personalized recommendations are displayed on the interface by priority. Doctors can click on a drug recommended by the medication rules to view the details of the medication rules, including the literature source information. Personalized medication does not display the source information.

[0178] 608. The doctor selects the drugs to be prescribed and prescribes them to the patient according to the general standard dosage of these drugs and his own judgment.

[0179] In another embodiment, see Figure 10 , describes in more detail the continuous optimization process used by the aforementioned 700 medication recommendation decision system:

[0180] 701. Doctors use the drug recommendation and selection records of the medication recommendation decision-making system and compare them with the patient's prescription information and medication information.

[0181] 702. Doctors may not prescribe medications based on recommendations or fail to input medication selection information. Therefore, by comparing the hypertension-related medications ultimately prescribed with the recommended medication records for the same patient, we can calculate the recommended medication usage rate, which serves as the primary indicator for measuring the value of the medication recommendation decision-making system.

[0182] 703. If the hypertension drugs prescribed by the patient match the recommended results, check the medication rules behind the drug recommendations, increase the weight score of the medication rules, and store the corresponding ranking scores in the hypertension disease knowledge graph database to indicate that the medication rules are actually used, which means they are more worthy of recommendation and should be ranked higher in the subsequent use of the medication recommendation decision-making system.

[0183] In summary, it can be found that the present disclosure also systematically reveals:

[0184] A medication recommendation system for hypertension, comprising:

[0185] Literature knowledge extraction unit, patient data collection unit, knowledge graph data integration unit, disease-specific knowledge graph database, knowledge-driven medication rule unit, medication recommendation decision system,

[0186] in,

[0187] A literature knowledge extraction unit is used to obtain medical literature related to the disease, and use machine learning technology to extract the medication methods of specialists for the disease from the medical literature and discover new medication knowledge from it;

[0188] A patient data collection unit, configured to collect at least the patient data of current patients with the disease, including: patient data, information about doctors associated with the patient, the patient's prescription, the patient's medication history, and efficacy data;

[0189] The knowledge graph data integration unit is used to obtain the data obtained by the literature knowledge extraction unit and the data obtained by the patient data collection unit to integrate the knowledge and patient data related to this type of disease;

[0190] A disease-specific knowledge graph database for a certain type of disease, used to import the integrated knowledge and patient data related to the disease from the knowledge graph data integration unit to establish a disease-specific knowledge graph;

[0191] The knowledge-driven medication rule unit is used to analyze the correlation between the current patient and the medication effect of this type of disease through the disease-specific knowledge graph established in the disease-specific knowledge graph database, and provide medication rules;

[0192] The medication recommendation decision system is used to recommend medication suggestions based on the medication rules provided by the knowledge-driven medication rule unit.

[0193] Preferably, the medication recommendation system for hypertension is characterized in that:

[0194] The knowledge-driven medication rule unit is also used to accept manual input of rules compiled by doctor experts as medication rules.

[0195] Preferably, the medication recommendation system for hypertension is characterized in that the system further comprises:

[0196] The system uses a recording unit to record the adoption of medication recommendations, and on this basis, stores the medication recommendations back into the disease-specific knowledge graph database for this type of disease, and maintains the rules in the knowledge-driven medication rule unit, including possible updates to the weights of medication rules, so that recognized medication rules have a greater chance of being used for subsequent recommendations.

[0197] Preferably, the medication recommendation system for hypertension is characterized by:

[0198] The patient data collection unit is also used to collect the patient's personal data, patient-related doctor information, patient's prescriptions, patient's medication history and efficacy data of other patients with such diseases.

[0199] Preferably, the medication recommendation system for hypertension is characterized in that the system further comprises:

[0200] The clinical data-driven personalized medication recommendation unit is used to estimate the most suitable medication for the current patient based on the medication history and efficacy data of other patients and the patient data of the current patient.

[0201] Preferably, the medication recommendation system for hypertension is characterized in that the literature knowledge extraction unit includes:

[0202] Collect external knowledge base units, which are used to collect literature sources related to the medication treatment of such diseases from online databases and knowledge bases of Chinese and foreign medical literature;

[0203] The crawler download document unit is used to download documents related to the medication treatment of this type of disease based on the document sources collected from the external knowledge base unit;

[0204] The book database scan-to-PDF unit is used to convert scanned books and materials related to the medication and treatment of this type of disease into PDF files;

[0205] The PDF document aggregation unit is used to aggregate the documents downloaded by the crawler download document unit and the PDF files converted by the book database scan to PDF unit into a folder;

[0206] Convert PDF files into OCR image data units, which are used to convert all PDF files summarized by the PDF document summary unit into one page of image data;

[0207] Convert OCR image data into text units, which is used to convert PDF files into OCR image data units. The image data obtained contains text, and the text is recognized through OCR recognition technology;

[0208] Cleaning non-text content units, which is used to convert OCR image data into text units, reorganize the text recognized, sort out the recognized text into sentences, and exclude non-text characters, where non-text characters include text in images or tables;

[0209] A document content splitting unit is used to split the final text obtained by cleaning the non-text content units into different sections according to the article structure of the original PDF file, and exclude content other than text paragraphs, including equation text, etc., wherein the different sections include the text body, possible abstract, and possible reference section;

[0210] A pronoun replacement unit, used to replace the pronouns in the final text obtained by splitting the document content into units with the specific nouns they refer to;

[0211] A paragraph splitting unit is used to split each paragraph in the final text obtained from the pronoun replacement unit into different sentences using punctuation marks such as periods and exclamation points, and to identify compound sentences and single sentences therein;

[0212] A compound sentence disassembling unit is used to disassemble each compound sentence in the final text obtained by the paragraph disassembling unit into separate sentences;

[0213] A word segmentation unit is used to mark the segmentations of the individual sentences obtained by the compound sentence segmentation unit, wherein the segmentations include individual Chinese and foreign characters;

[0214] The named entity recognition unit is used to identify all the segmented words obtained by the segmentation unit as corresponding entities, where an entity consists of one or more segmented words, and each entity represents its own independent meaning;

[0215] The SVO relationship generation unit is used to generate SVO relationships (i.e., subject-verb-object) between the subject, verb, and object of all entities detected by the named entity recognition unit within the same sentence. The subject and object are used as nodes, and the verb represents the relationship from the subject to the object.

[0216] The information merging and deduplication unit is used to collect all node and relationship pairs of the entire document from the SVO relationship generation unit, and dedupe and merge them to merge the attribute information under the same node relationship together;

[0217] The import unit is used to compare the attribute information of all node relationship pairs with the information in the graph database. When the attribute information of a node relationship pair does not exist in the graph database, it is considered that there is new pairing information, and the new pairing information is imported into the graph database.

[0218] Preferably, the medication recommendation system for hypertension is characterized in that the system further comprises:

[0219] The Reference decomposition unit is used to separately decompose the reference part obtained from the document content decomposition unit so as to separate the Reference list and record it as a knowledge traceability reference.

[0220] Preferably, the medication recommendation system for hypertension is characterized by:

[0221] The SVO relationship generation unit is also used to attach the corresponding traceability reference obtained by the Reference decomposition unit to any individual sentence with annotated reference link information in the SVO relationship of the individual sentence, and generate it as the attribute information of the SVO relationship of the individual sentence.

[0222] Preferably, the medication recommendation system for hypertension is characterized in that the system further comprises:

[0223] The named entity recognition unit is also used to identify corresponding entities based on other language processing rules defined in the disease-specific knowledge graph of this type of disease.

[0224] Preferably, the medication recommendation system for hypertension is characterized by:

[0225] The knowledge graph data integration unit, during the integration process:

[0226] Make the data organization and structure of the knowledge database obtained by the literature knowledge extraction unit consistent with the clinical database obtained by the patient data collection unit;

[0227] Ensure that knowledge concepts are consistent with relevant data definitions;

[0228] Synonymous concepts are expressed using the same concept word;

[0229] Otherwise, the data attributes are corrected;

[0230] After data attribute correction, the knowledge and patient data related to this type of disease are imported into the knowledge graph database of this type of disease, so that the clinical data corresponding to the knowledge is clear.

[0231] Preferably, the medication recommendation system for hypertension is characterized by:

[0232] The knowledge graph data integration unit also includes a field correspondence unit, which is used to obtain the knowledge database obtained by the literature knowledge extraction unit and the clinical database obtained by the patient data collection unit, and establish a correspondence between the knowledge graph and the patient data fields with the relevant definitions in the knowledge graph, so as to maintain a unified set of data structures for comprehensive management and implementation of field unification; wherein,

[0233] The fields of new literature knowledge and new patient data need to be constantly compared to find the field definitions that need to be unified, and,

[0234] When necessary, the system is directed to the literature knowledge extraction unit and the patient data collection unit to unify the iterative implementation fields.

[0235] Preferably, the medication recommendation system for hypertension is characterized by:

[0236] The knowledge graph data integration unit further includes a knowledge graph structure definition correction unit, which is used to:

[0237] When the knowledge graph structure is incomplete relative to the data structure of the patient data, fields are added or modified in the knowledge graph to make the knowledge graph structure complete.

[0238] Preferably, the medication recommendation system for hypertension is characterized by:

[0239] The knowledge graph data integration unit also includes a knowledge extraction rule modification unit;

[0240] When the knowledge graph structure is incomplete relative to the data structure of the patient data, the knowledge graph data integration unit adds fields or modifies fields in the knowledge graph and then changes the knowledge extraction rule unit to:

[0241] Convert the added or modified field information into new document knowledge extraction rules.

[0242] Preferably, the medication recommendation system for hypertension is characterized by:

[0243] The named entity recognition unit is also used to identify all the segmented words obtained by the segmentation unit as corresponding entities according to the new document knowledge extraction rules.

[0244] Preferably, the medication recommendation system for hypertension is characterized by:

[0245] The knowledge graph data integration unit further includes a patient data attribute correction unit, which is used to:

[0246] When the data structure of the patient data is incomplete relative to the knowledge graph structure, fields are added to or modified in the patient data to make the data structure of the patient data complete.

[0247] Preferably, the medication recommendation system for hypertension is characterized by:

[0248] The knowledge graph data integration unit also includes a database additional information demand unit;

[0249] When the data structure of patient data is incomplete relative to the knowledge graph structure, after adding or modifying fields to the patient data, since this indicates that the attribute fields of the patient data already exist in the knowledge database, the database additional information requirement unit is used to:

[0250] The added or modified field information is passed to the patient data collection unit so that the patient data collection unit introduces patient management data from the external data source of the system, and during the patient data extraction process, the patient data collection unit collects patient data according to the same corresponding fields in the knowledge database to ensure consistent data definitions.

[0251] Preferably, the medication recommendation system for hypertension is characterized by:

[0252] The knowledge graph data integration unit also includes a patient data import and update unit;

[0253] When the data structure of patient data is incomplete relative to the knowledge graph structure, after adding or modifying fields in the patient data, the patient data import and update unit is used to:

[0254] Import patient data into the disease-specific knowledge graph database in real time or periodically along with added or modified fields.

[0255] Preferably, the medication recommendation system for hypertension is characterized by:

[0256] The knowledge graph data integration unit also includes a unit for excluding medications for diseases other than the above mentioned disease category;

[0257] When the knowledge database is obtained from the literature knowledge extraction unit, the exclusion unit for medications not used for such diseases is used to exclude information related to medications not used for such diseases from the knowledge database.

[0258] Preferably, the medication recommendation system for hypertension is characterized in that the patient data collection unit further includes:

[0259] A third-party data introduction unit is used to introduce patient management data on the clinical management and health management behavior of patients with such diseases from external data sources of the system, wherein the patient management data includes data such as each post-drug test / test / measurement index, medication records, patient baseline and follow-up medical records.

[0260] Preferably, the medication recommendation system for hypertension is characterized in that the patient data collection unit further includes:

[0261] The attribute extraction unit is used to extract relevant attribute fields of the patient data from the patient management data introduced by the third-party data introduction unit.

[0262] Preferably, the medication recommendation system for hypertension is characterized in that the patient data collection unit further includes:

[0263] Added Baseline and Follow-up Fields elements, which are used to:

[0264] When the data structure of patient data is incomplete relative to the knowledge graph structure, and

[0265] When the attribute fields of the patient data already exist in the knowledge database and the attribute fields of the patient data extracted by the attribute extraction unit from the patient management data introduced by the third-party data introduction unit are inconsistent with the attribute fields of the patient data in the knowledge database,

[0266] Baseline and follow-up fields are added to the relevant attribute fields of the patient data extracted by the attribute extraction unit.

[0267] Preferably, the medication recommendation system for hypertension is characterized in that the patient data collection unit further includes:

[0268] The arranging unit is used to add corresponding new fields to the system's structure dictionary and arrange the system's structure dictionary according to the new fields of patient data in the patient management data introduced by the third-party data introduction unit.

[0269] Preferably, the medication recommendation system for hypertension is characterized in that the patient data collection unit further includes:

[0270] The data extraction unit is used to extract data based on the fields of the patient management data obtained by the third-party data introduction unit, the fields extracted by the attribute extraction unit, and the fields in the system's structure dictionary, according to the field definition, through extraction (extract), transformation (transform), and loading adapted to the external data source of the system, and organize the data according to the field definition and data structure.

[0271] Preferably, the medication recommendation system for hypertension is characterized in that the patient data collection unit further includes:

[0272] Data cleaning unit;

[0273] When the data extracted by the data extraction unit is incomplete, the data cleaning unit is used to exclude certain data and perform data cleaning, wherein the data cleaning is used to clean out the associated data of the incomplete data.

[0274] Preferably, the medication recommendation system for hypertension is characterized in that the patient data collection unit further includes:

[0275] The import unit is used to directly import the data processed by the data cleaning unit into the graph database.

[0276] Preferably, the medication recommendation system for hypertension is characterized in that the knowledge-driven medication rule unit further includes:

[0277] Graph structure visualization unit, which is used to:

[0278] The knowledge relationships derived from the disease-specific knowledge graph database are visualized to help doctors manually judge the relationship between medication, condition data, and treatment of this type of disease.

[0279] Taking a certain drug as the starting point, the graph structure visualization unit draws the potential association paths with this type of disease. Among them, the defined relationship paths are displayed in a different color from other paths, allowing doctors and experts to browse and select potential new drug paths.

[0280] Preferably, the medication recommendation system for hypertension is characterized in that the knowledge-driven medication rule unit further includes:

[0281] The manual rule arrangement unit is used to interact with doctors and, based on existing field definitions and data structures, arrange medication rules based on doctors' experience and cognition into the system.

[0282] Preferably, the medication recommendation system for hypertension is characterized in that the knowledge-driven medication rule unit further includes:

[0283] Clinical Assessment Unit;

[0284] When a doctor decides to use patient medication data to verify the effectiveness of a medication pathway / medication rule, the clinical evaluation unit is used to evaluate and verify whether the corresponding medication pathway / medication rule is effective based on the attribute descriptions of the patient's use of the corresponding drug and other patient data, as well as the historical changes in post-medication testing / testing / measurement indicators.

[0285] Preferably, the medication recommendation system for hypertension is characterized in that the knowledge-driven medication rule unit further includes:

[0286] The review unit is used to interact with the expert group and review the new medication rules based on the expert group's verification and supplement of the medication rules for this type of disease, and,

[0287] The audit unit supports the audit of medication rules based on repeated verification of the adjusted rule scheme.

[0288] Preferably, the medication recommendation system for hypertension is characterized in that the knowledge-driven medication rule unit further includes:

[0289] Supplementary units are used to supplement the rules prohibiting the use of certain drugs in specific cases or contraindications;

[0290] and,

[0291] After the graph structure visualization unit has completed the interaction with the doctor, the doctor can directly jump to the supplementary unit through the operation of the doctor;

[0292] After the manual arrangement rule unit has completed the interaction with the doctor, it can directly jump to the supplementary unit through the doctor's operation.

[0293] Preferably, the medication recommendation system for hypertension is characterized in that the knowledge-driven medication rule unit further includes:

[0294] The output unit is used to output all medication rule combinations verified or approved by doctors to the medication recommendation decision system.

[0295] Preferably, the medication recommendation system for hypertension is characterized in that the clinical data-driven personalized medication recommendation unit further comprises:

[0296] A clinical database extraction unit is used to extract all graph relationship data from patient nodes to drug nodes from the disease-specific knowledge graph database of this type of disease, wherein the patients include the current patient and other patients.

[0297] Preferably, the medication recommendation system for hypertension is characterized in that the clinical data-driven personalized medication recommendation unit further comprises:

[0298] The generation unit is used to: based on the clinical database extraction unit, use machine learning tools to train and generate graph embedding vector functions for each node, and use model algorithms in the knowledge graph field to model and generate.

[0299] Preferably, the medication recommendation system for hypertension is characterized in that the clinical data-driven personalized medication recommendation unit further comprises:

[0300] A new medication prediction unit is configured to: estimate the unestablished relationship between the patient and the medication based on the graph embedded data in the generation unit, wherein the relationship is a discovered new medication recommendation; and

[0301] The new drug prediction unit supports recommending the top N most suitable drugs for any patient;

[0302] The new medication prediction unit outputs new medication recommendations to the medication recommendation decision system.

[0303] Preferably, the medication recommendation system for hypertension is characterized in that the system further comprises:

[0304] The physical examination data collection unit is used to find all the physical examination data of the patient from the disease-specific knowledge graph database.

[0305] Preferably, the medication recommendation system for hypertension is characterized in that the system further comprises:

[0306] The physical examination suggestion unit is used to interact with the doctor based on the standard physical examination field list of this type of disease and the patient's physical examination data collected by the physical examination data collection unit, and display the physical examination items that the patient is missing.

[0307] Preferably, the medication recommendation system for hypertension is characterized in that the system further comprises:

[0308] a disease type determination unit, which determines which disease the patient belongs to based on the patient's physical examination data collected by the physical examination data collection unit, and implements the referral function of the system when the patient belongs to a certain disease of the same type that requires referral; and

[0309] When the patient does not fall into the category of diseases requiring referral, the patient data is provided to the patient data collection unit for medication recommendation.

[0310] Preferably, the medication recommendation system for hypertension is characterized in that the system further comprises:

[0311] The history reading unit is used to read the patient's medication history and efficacy data, and compare the drug recommendation record and selection record with the patient's prescription information, medication history and efficacy data.

[0312] Preferably, the medication recommendation system for hypertension is characterized in that the system further comprises:

[0313] A statistical unit is used to compare the disease-related drugs prescribed by patients with the recommended drug records for the same patients to calculate an indicator of the recommended drug usage rate and use it as the main indicator for measuring the value of the system.

[0314] Preferably, the medication recommendation system for hypertension is characterized in that the system further comprises:

[0315] Adjust the weight unit, which is used to:

[0316] When the drug prescribed by the patient matches the recommended drug, the medication rule corresponding to the drug is checked, the weight score of the medication rule is increased, and the corresponding ranking score of the disease-specific knowledge graph database is stored back to indicate that the medication rule is actually used and is more worthy of recommendation.

[0317] In subsequent use of the system, the recommendations are arranged according to the ranking scores, with higher scores being placed at the top.

[0318] Preferably, the medication recommendation system for hypertension is characterized in that the system further comprises:

[0319] A knowledge processing unit, which is used to establish a graph database, wherein

[0320] The storage format of the graph database can flexibly support the storage of any data structure;

[0321] The graph database includes:

[0322] (1) Node corresponding to the entity;

[0323] (2) Relationships between entities;

[0324] (3) Attributes of the entity.

[0325] Preferably, the medication recommendation system for hypertension is characterized by:

[0326] Node A corresponding to the first entity and node B corresponding to the second entity are connected by a relationship R1 between the first entity and the second entity;

[0327] Node A of the first entity, with attribute P1;

[0328] The node B of the second entity has attribute P3.

[0329] It should be noted that, in the description of this specification, the description with reference to the terms "one embodiment / method", "some embodiments / methods", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment / method or example are included in at least one embodiment / method or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment / method or example. Moreover, the specific features, structures, materials or characteristics described may be combined in a suitable manner in any one or more embodiments / methods or examples. In addition, those skilled in the art may combine and combine the different embodiments / methods or examples described in this specification and the features of the different embodiments / methods or examples, unless they are contradictory.

[0330] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. Throughout the description of this application, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0331] Those skilled in the art will appreciate that the above embodiments are merely intended to clearly illustrate the present disclosure and are not intended to limit the scope of the present disclosure. Other changes or modifications may be made based on the above disclosure, and such changes or modifications are still within the scope of the present disclosure.

Claims

1. A medical literature text extraction system for hypertension, comprising: The literature knowledge extraction unit is used to obtain medical literature related to hypertension, and uses machine learning technology to extract the medication methods of hypertension specialists from the medical literature and discover new medication knowledge from it, including: The book database scan-to-PDF unit is used to convert scanned books and materials related to hypertension medication and treatment into PDF files; Aggregate PDF document unit, used to convert the documents downloaded by the crawler download document unit, the book database scan to PDF file converted by the PDF unit; Convert PDF files into OCR image data units, which are used to convert all PDF files summarized by the PDF document summary unit into one page of image data; Convert OCR image data into text units, which is used to convert PDF files into OCR image data units. The image data obtained contains text, and the text is recognized through OCR recognition technology; Cleaning non-text content units, which is used to convert OCR image data into text units, reorganize the text recognized, sort out the recognized text into sentences, and exclude non-text characters, where non-text characters include text in images or tables; A document content splitting unit is used to split the final text obtained by cleaning the non-text content unit into different parts according to the article structure of the original PDF file, and exclude the content other than the text paragraphs, including equation text, wherein the different parts include the text body, the abstract, and the reference part; A pronoun replacement unit, used to replace the pronouns in the final text obtained by splitting the document content into units with the specific nouns they refer to; A paragraph splitting unit is used to split each paragraph in the final text obtained from the pronoun replacement unit into different sentences using punctuation marks such as periods and exclamation points, and to identify compound sentences and single sentences therein; A compound sentence disassembling unit is used to disassemble each compound sentence in the final text obtained by the paragraph disassembling unit into separate sentences; A word segmentation unit is used to mark the individual words in the individual sentences obtained by the compound sentence segmentation unit, wherein the word segmentation includes individual Chinese and foreign characters; The named entity recognition unit is used to identify all the segmented words obtained by the segmentation unit as corresponding entities, where an entity consists of one or more segmented words, and each entity represents its own independent meaning; Among the entities within the same sentence, focus on the subject, verb, and object. Adjectives become entities and establish a relationship with the subject. The reference part is separated out separately as a reference for tracing the source of knowledge.

2. A medical literature text extraction system for hypertension according to claim 1, characterized in that: The literature knowledge extraction unit includes: Collect external knowledge base units to collect literature sources related to the medication treatment of hypertension in online databases and knowledge bases of Chinese and foreign medical literature.

3. A medical literature text extraction system for hypertension according to claim 1, characterized in that: The literature knowledge extraction unit includes: The crawler download document unit is used to download documents related to the medication treatment of hypertension based on the document sources collected from the external knowledge base unit.

Images