Drug use decision-making method and system based on deep learning

Abstract

The invention discloses a drug use decision method and system based on deep learning, and relates to the technical field of artificial intelligence, and the method comprises the steps: firstly obtaining the basic information of a target symptom, and inputting a large language model to obtain the symptom and the description information of a to-be-matched drug; and then determining a target drug matched with the target symptom through a preset first expert model. And then integrating the target drug related information and converting the information into a standard format to obtain standardized drug recommendation information. And finally, the information is pushed to a preset medication decision interface through a preset interface, so that intelligent, efficient and accurate medication decision assistance is realized. The main purpose of the invention is to provide an intelligent medication decision-making auxiliary method, which matches symptoms and drug information through a large language model, accurately recommends a target drug in combination with an expert model, standardizes recommendation information and pushes the standardized recommendation information to a medication decision-making interface so as to improve medical decision-making efficiency and accuracy.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a medication decision-making method and system based on deep learning. Background Art

[0002] In the medical field, accurate medication decisions are crucial to the treatment effect of patients. Traditional medication decisions rely on doctors' experience, which is subjective, time-consuming and laborious. With the development of information technology, the use of deep learning to assist medication decisions has become a trend. However, existing methods still need to be improved in terms of accurate matching of symptoms and drugs, information integration and efficient push. Summary of the invention

[0003] The purpose of the present invention is to provide a medication decision-making method and system based on deep learning.

[0004] In a first aspect, the present invention provides a medication decision-making method based on deep learning, comprising: Obtain basic information on target symptoms; Inputting the basic information into a large language model for processing to obtain description information of the target symptom and description information of each to-be-matched drug in the to-be-matched drug group for the target symptom; According to the description information of the target symptom and the description information of each to-be-matched drug in the to-be-matched drug group, a target drug to be matched and associated with the target symptom is determined by presetting a first expert model; Integrate the relevant information of the target drug and convert it into a standard format to obtain standardized drug recommendation information, wherein the relevant information includes drug name, ingredients, efficacy, applicable symptoms, usage and dosage, adverse reactions, and interactions with other drugs; The standardized drug recommendation information is pushed to the preset medication decision interface through a preset interface.

[0005] In a second aspect, the present invention provides a server system, comprising a server, wherein the server is used to execute the method described in the first aspect.

[0006] Compared with the prior art, the beneficial effects provided by the present invention include: adopting a medication decision-making method and system based on deep learning disclosed by the present invention, by obtaining the basic information of the target symptoms, inputting the large language model to obtain the description information of the symptoms and the drugs to be matched. Then, the target drug matching the target symptom is determined by presetting the first expert model. Then, the relevant information of the target drug is integrated and converted into a standard format to obtain standardized drug recommendation information. Finally, it is pushed to the preset medication decision interface through the preset interface to realize intelligent, efficient and accurate medication decision assistance. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative work.

[0008] Figure 1 A schematic diagram of the steps of a medication decision-making method based on deep learning provided in an embodiment of the present invention; Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0009] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0010] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0011] In order to solve the technical problems in the aforementioned background technology, Figure 1 A flowchart of a deep learning-based medication decision-making method provided in an embodiment of the present disclosure is provided below. The deep learning-based medication decision-making method is introduced in detail.

[0012] Step S201, obtaining basic information of the target symptom; Step S202, inputting the basic information into a large language model for processing to obtain description information of the target symptom and description information of each drug to be matched in the drug group to be matched for the target symptom; Step S203, determining a target drug to be matched and associated with the target symptom by presetting a first expert model according to the description information of the target symptom and the description information of each drug to be matched in the drug group to be matched; Step S204, integrating the relevant information of the target drug and converting it into a standard format to obtain standardized drug recommendation information, wherein the relevant information includes drug name, ingredients, efficacy, applicable diseases, usage and dosage, adverse reactions, and interaction with other drugs; Step S205: Push the standardized drug recommendation information to a preset medication decision interface through a preset interface.

[0013] In an embodiment of the present invention, it is assumed that the server is an intelligent medication decision server within the hospital, which is connected to various diagnosis and treatment systems of the hospital. A patient, Xiao Li, came to see a doctor because of physical discomfort. After preliminary questioning and examination, the doctor entered Xiao Li's symptom information into the hospital's electronic medical record system. These symptom information include fever, cough, fatigue, etc., and also record the time and severity of the symptoms. For example, the fever has lasted for two days, the body temperature has reached 39°C, and the cough is dry and frequent. The server obtains these basic information about Xiao Li's target symptoms from the electronic medical record system to prepare for the subsequent medication decision process. The server inputs the basic information of Xiao Li's symptoms, such as fever, cough, fatigue and related details, into the pre-trained large language model. This large language model has been trained with a large amount of medical text data and can understand and process symptom information described in natural language. After analyzing and processing this information, the large language model outputs more detailed and accurate description information of the target symptoms. For example, it combines symptoms such as fever, cough, and fatigue, and describes them as "fever that may be caused by upper respiratory tract infection, accompanied by persistent dry cough and general fatigue, intermittent fever, and body temperature fluctuations between 38.5℃-39℃". At the same time, the large language model also provides description information of each drug to be matched in the drug group to be matched for these target symptoms of Xiao Li based on its understanding of medical knowledge. For example, the drug group to be matched may include ibuprofen and acetaminophen for fever reduction, and dextromethorphan for cough relief. For ibuprofen, its description information may be "ibuprofen is a commonly used non-steroidal anti-inflammatory drug with antipyretic, analgesic, and anti-inflammatory effects. It can be used to relieve mild to moderate pain and fever symptoms caused by colds, etc. Common adverse reactions include gastrointestinal discomfort, etc." Obtain relevant description information: The server obtains the description information of the target symptoms output by the large language model (such as the description of the relevant symptoms caused by the upper respiratory tract infection mentioned above) and the description information of each drug to be matched (such as ibuprofen, acetaminophen, dextromethorphan, etc.). Descriptive information enhancement operation: Target symptoms: The server performs descriptive information enhancement operations through the preset first expert model based on the descriptive information of the target symptoms and the preset first guide words. For example, the preset first guide words include "pay attention to the characteristic tendency of upper respiratory tract infection symptoms, combine past similar symptom analysis records, and require the output of more accurate medication-related symptom description information" and other content. The server integrates the descriptive information of the target symptoms into the preset first guide words, obtains the optimized first guide words, and then loads them into the preset first expert model. Through deep learning reasoning, it determines the enhanced descriptive information of the target symptoms, such as "fever caused by upper respiratory tract infection, body temperature fluctuations that have not subsided for two days, frequent dry coughs that affect sleep, general fatigue that leads to mobility difficulties, and precise medication is required to relieve symptoms and prevent worsening of the disease."Drugs to be matched: For each drug to be matched, taking ibuprofen as an example, based on its description information and the preset second guide words (such as "combining the characteristics of the drug, outputting enhanced description information of the drug that better meets the current symptom needs"), the description information is enhanced through the first expert model. The description information of ibuprofen is integrated into the preset second guide words to obtain the optimized second guide words, which are loaded into the first expert model to determine the enhanced description information of ibuprofen, such as "ibuprofen is an antipyretic and analgesic. In the current situation where the patient has a fever caused by upper respiratory tract infection and the body temperature is high, and the patient is weak due to fever and fatigue, it can quickly and effectively reduce the body temperature, relieve the physical discomfort caused by fever, and reduce the symptoms of fatigue, but attention should be paid to the monitoring of gastrointestinal adverse reactions." Determine the original description feature vector: Target symptom: The server determines the description information features corresponding to each element in the enhanced description information of the target symptom based on the enhanced description information of the target symptom and the preset description information feature dictionary (where the description information features are represented by dummy variables), and then determines the original description feature vector of the target symptom. For example, for the "fever" element, the corresponding code in the feature dictionary is [1,0,0] (assuming that fever, cough, and other symptoms are encoded separately in the dictionary), and so on to determine the encoding of other elements to form the original description feature vector. Drugs to be matched: Similarly, according to the enhanced description information of each drug to be matched and the preset description information feature dictionary, the description information features corresponding to each element in the enhanced description information of each drug to be matched are determined, and the original description feature vector of each drug to be matched is determined. Taking ibuprofen as an example, the encoding corresponding to the elements such as antipyretic, analgesic, and gastrointestinal adverse reactions is determined to form the original description feature vector. Feature mapping and integrated processing: Target symptoms: The server loads the original description feature vector of the target symptom into the preset second expert model, and determines the multiple feature mapping description features corresponding to the original description feature vector of the target symptom through feature mapping processing. Then, based on these feature mapping description features, the description features corresponding to the enhanced description information of the target symptom are determined through feature integration processing. Drugs to be matched: The original description feature vectors of each drug to be matched are loaded into the second expert model, and multiple feature mapping description features corresponding to the original description feature vectors of each drug to be matched are determined through feature mapping processing, and then the description features corresponding to the enhanced description information of each drug to be matched are determined through feature integration processing. Determine the fitness coefficient: The server loads the description features corresponding to the enhanced description information of the target symptom and the description features corresponding to the enhanced description information of each drug to be matched into the preset drug decision model, and determines the fitness coefficient of the target symptom for each drug to be matched through associated adaptation operations. For example, for ibuprofen, its fitness coefficient may be determined to be 0.8 through an array of adaptation indicators (including adaptation indicators such as symptom relief degree and probability of adverse reactions).Determine the target drug: Calculate the fitness confidence: The server determines the fitness confidence corresponding to each fitness coefficient according to the fitness indicator array corresponding to each fitness coefficient and the importance factor of at least two fitness indicators in the fitness indicator array corresponding to each fitness coefficient. Assuming that in the fitness indicator array corresponding to the fitness coefficient of ibuprofen, the importance factor of symptom relief is 0.6, and the importance factor of the probability of adverse reaction is 0.4, combined with its fitness coefficient of 0.8, the fitness confidence is calculated. Screening target drugs: The server arranges each fitness confidence in descending order from high to low, and determines multiple fitness confidences before the preset ranking. For example, if the preset ranking is the top two, if the fitness confidence of ibuprofen is in the top two, then ibuprofen is used as one of the target drugs matched and associated with Xiao Li's target symptoms (there may be other eligible drugs). After the server determines the target drugs (such as ibuprofen, etc.), it integrates the relevant information of these target drugs. Taking ibuprofen as an example, its drug name (ibuprofen), ingredients (the main ingredient is ibuprofen), efficacy (antipyretic, analgesic, anti-inflammatory), applicable symptoms (fever caused by colds, etc., mild to moderate pain), usage and dosage (such as 0.3-0.6 grams per adult, 3-4 times a day, etc.), adverse reactions (gastrointestinal discomfort, etc.) and interactions with other drugs (such as increased bleeding risk when used in combination with anticoagulants, etc.). Then the integrated information is converted into a standard format to conform to the preset standardized drug recommendation information format. For example, it is converted into a structured data format, containing various information fields, to form standardized drug recommendation information, which is convenient for subsequent accurate display to doctors or patients in the medication decision interface. The server pushes the generated standardized drug recommendation information to the preset medication decision interface within the hospital through a pre-set interface. This interface may be the computer interface used by doctors during diagnosis and treatment, or the mobile interface accessed by patients through the hospital APP. When the doctor is treating Xiao Li, he can clearly see the standardized recommendation information about target drugs such as ibuprofen on the medication decision-making interface on the computer, including detailed information about the drugs, so that he can make better medication decisions; if the patient accesses the mobile interface through the hospital APP, he can also learn about the recommended drugs and increase his understanding of medication.

[0014] In an embodiment of the present invention, the target drug to be matched and associated with the target symptom is determined by presetting a first expert model based on the description information of the target symptom and the description information of each drug to be matched in the drug group to be matched, which can be implemented through the following example.

[0015] Obtaining description information of the target symptom and description information of each to-be-matched drug in the to-be-matched drug group; According to the description information of the target symptom and the description information of each to-be-matched drug, a description information enhancement operation is performed by presetting a first expert model to determine enhanced description information of the target symptom and enhanced description information of each to-be-matched drug; Performing feature extraction based on the enhanced description information of the target symptom and the enhanced description information of each to-be-matched drug, and determining description features corresponding to the enhanced description information of the target symptom and description features corresponding to the enhanced description information of each to-be-matched drug; According to the description features corresponding to the enhanced description information of the target symptom and the description features corresponding to the enhanced description information of each to-be-matched drug, determining the fitness coefficient of the target symptom for each to-be-matched drug through an associated adaptation operation; According to each fitness coefficient, a target drug that is matched and associated with the target symptom is determined, and the group of drugs to be matched includes the target drug.

[0016] In the embodiment of the present invention, for example, it is assumed that the server is the core server of the intelligent medication decision-making support system of a large medical institution, which is responsible for processing medication-related information of many patients. A patient named Mr. Zhang came to see a doctor. He reported that he was unwell. The main symptoms were chest pain and chest tightness, and the symptoms would worsen after fatigue, and occasionally accompanied by mild dyspnea. After the doctor conducted a preliminary physical examination and some necessary auxiliary examinations (such as electrocardiogram, etc.) on Mr. Zhang, the symptom information and related examination results were recorded in detail in the electronic medical record system. The server accurately obtained the description information of Mr. Zhang's target symptoms from the electronic medical record system, specifically "the patient Mr. Zhang had chest pain and chest tightness symptoms, the chest pain was intermittent, the degree was moderate, and it was aggravated after fatigue, accompanied by occasional mild dyspnea, and the electrocardiogram showed abnormalities in some bands." At the same time, the server quickly generated a drug group to be matched for these target symptoms of Mr. Zhang based on its built-in medical knowledge base and the ability to analyze a large amount of case data. This group of drugs to be matched may include nitroglycerin (used to relieve angina), aspirin (anti-platelet aggregation, prevent further development of cardiovascular disease), atorvastatin (regulate blood lipids, stabilize plaques) and other drugs. For each drug to be matched, the server also obtains the corresponding description information. Taking nitroglycerin as an example, its description information is "Nitroglycerin is a commonly used vasodilator, mainly used to treat and prevent angina attacks. It can quickly dilate coronary arteries and increase myocardial blood supply, thereby relieving angina symptoms. Common adverse reactions include headaches, hypotension, etc., and when using it, it is necessary to avoid taking it at the same time with certain antihypertensive drugs to avoid low blood pressure." Similarly, the description information of aspirin covers its anti-platelet aggregation mechanism, scope of application, common adverse reactions (such as gastrointestinal bleeding risk), etc. The description information of atorvastatin focuses on its efficacy in regulating blood lipids, possible adverse reactions such as liver damage, and interactions with other drugs. After obtaining the relevant description information of the target symptoms and the drugs to be matched, the server begins to perform description information enhancement operations through the preset first expert model. For the target symptoms: there is a preset first guide word in the server, and its content is roughly "for cardiovascular system-related symptoms, combined with the patient's age, medical history and the specific manifestations of the current symptoms, a comprehensive and in-depth analysis of the symptom characteristics is carried out to provide more detailed symptom description information for subsequent precise medication." The server integrates the description information of Mr. Zhang's target symptoms into this preset first guide word, such as "The patient Mr. Zhang has chest pain and chest tightness symptoms. The chest pain is intermittent and moderate in degree. It worsens after fatigue, accompanied by occasional mild dyspnea. The electrocardiogram shows that some bands are abnormal." This forms the optimized first guide word. Then, this optimized first guide word is loaded into the preset first expert model.This first expert model is a deep learning model trained with a large amount of cardiovascular disease-related case data and professional medical knowledge. It can perform deep learning reasoning based on the input guide words. After the model's reasoning operation, the server obtained the enhanced description information of Mr. Zhang's target symptoms, specifically "Mr. Zhang, a 65-year-old patient with a 5-year history of hypertension, has symptoms of chest pain and chest tightness that are suspected to be angina attacks. The chest pain is intermittent and the frequency of attacks gradually increases. The degree is moderate and it is significantly aggravated after fatigue. Occasional mild dyspnea has also worsened. The electrocardiogram shows abnormalities such as ST segment depression. Comprehensive consideration may be that myocardial ischemia caused by coronary atherosclerosis has worsened. The medication plan needs to be clarified as soon as possible to relieve symptoms and prevent the disease from worsening." For the drugs to be matched: Taking nitroglycerin as an example, the server also operates according to the preset second guide words. The preset second guide words are "Combined with the characteristics of the drug itself and the current symptoms of the patient, deepen the drug description information, and highlight the advantages and potential risks of the drug under the current condition." The description information of nitroglycerin "Nitroglycerin is a commonly used vasodilator, mainly used to treat and prevent angina attacks. It can quickly dilate coronary arteries, increase myocardial blood supply, and thus relieve angina symptoms. Common adverse reactions include headaches, hypotension, etc., and care should be taken to avoid taking it with certain antihypertensive drugs to prevent low blood pressure" is integrated into the preset second guide word to obtain the optimized second guide word. Then it is loaded into the preset first expert model, and the enhanced description information of nitroglycerin is determined through deep learning reasoning, such as "Nitroglycerin is a classic drug for the treatment of angina pectoris. For patients like Mr. Zhang who are suspected of having angina attacks and whose myocardial ischemia is aggravated by coronary atherosclerosis, it can quickly dilate coronary arteries, quickly increase myocardial blood supply, and effectively relieve chest pain, chest tightness and other angina symptoms. However, given that the patient has a history of hypertension, special attention should be paid to the occurrence of adverse reactions of hypotension when using it, and it should be ensured that it is reasonably combined with other antihypertensive drugs being taken to avoid further increase in the risk of hypotension." For other drugs to be matched, such as aspirin and atorvastatin, the server also follows the same process to fuse their description information with the corresponding preset guide words, load them into the preset first expert model for reasoning, and obtain their respective enhanced description information. The enhanced description information of aspirin may emphasize its importance in preventing the further development of cardiovascular disease, especially for Mr. Zhang, who has a history of hypertension and suspected angina attacks. It can effectively inhibit platelet aggregation and reduce the risk of thrombosis, but attention should be paid to the monitoring of gastrointestinal bleeding risks. The enhanced description information of atorvastatin will highlight its role in regulating blood lipids and stabilizing plaques, which is crucial to improving Mr. Zhang's coronary atherosclerosis. At the same time, it reminds to pay attention to the monitoring of adverse reactions such as liver function damage. After obtaining the enhanced description information of the target symptoms and the enhanced description information of each drug to be matched, the server will then perform feature extraction operations.The server has a preset descriptive information feature dictionary, which contains detailed coding representations of various cardiovascular disease-related symptoms, drug properties, etc., using dummy variables. For the target symptom: Take Mr. Zhang's enhanced descriptive information as an example: "Mr. Zhang, 65 years old, has a history of hypertension for 5 years. The chest pain and chest tightness he experienced this time are suspected to be angina attacks. The chest pain is intermittent and the frequency of attacks gradually increases. The degree is moderate and it is significantly aggravated after fatigue. Occasional mild dyspnea has also worsened. The electrocardiogram shows abnormalities such as ST segment depression. Comprehensive considerations suggest that myocardial ischemia may be aggravated due to coronary atherosclerosis. The medication regimen needs to be clarified as soon as possible to relieve symptoms and prevent the disease from worsening." The server extracts features from each element based on this enhanced descriptive information and the preset descriptive information feature dictionary. For example, for the element "chest pain", the corresponding code in the feature dictionary may be [1,0,0,0,0] (assuming that the dictionary encodes chest pain, chest tightness, dyspnea, and other symptoms separately); for the element "history of hypertension", the corresponding code may be [0,1,0,0,0]; for the element "ST segment depression", the corresponding code may be [0,0,1,0,0], etc. By performing such feature extraction operations on all elements in the enhanced description information one by one, the server finally determines the descriptive features corresponding to the enhanced description information of the target symptom, and combines these codes to form a descriptive feature vector. For drugs to be matched: Take nitroglycerin as an example, for its enhanced description information, "nitroglycerin is a classic drug for the treatment of angina pectoris. For patients like Mr. Zhang who are suspected of having angina pectoris and whose myocardial ischemia is aggravated by coronary atherosclerosis, it can rapidly dilate coronary arteries, quickly increase myocardial blood supply, and effectively relieve angina pectoris symptoms such as chest pain and chest tightness. However, given that the patient has a history of hypertension, special attention should be paid to the occurrence of adverse reactions of hypotension when using it, and it should be ensured that it is reasonably combined with other antihypertensive drugs being taken to avoid further increase in the risk of hypotension." The server extracts features from each element of nitroglycerin based on the enhanced description information of nitroglycerin and the preset description information feature dictionary. For example, for the element "dilate coronary arteries", the corresponding code in the feature dictionary may be [1,0,0,0,0] (assuming that the dictionary encodes different mechanisms of action, adverse reactions, applicable diseases, etc. of the drug); for the element "adverse reactions of hypotension", the corresponding code may be [0,1,0,0,0]; for the element "relieve angina symptoms", the corresponding code may be [0,0,1,0,0], and so on. By performing such feature extraction operations on all elements in the enhanced description information one by one, the server finally determines the description features corresponding to the enhanced description information of nitroglycerin, and combines these codes to form a description feature vector.In the same way, the server also performs feature extraction operations on other drugs to be matched, such as aspirin and atorvastatin, to determine the descriptive feature vectors corresponding to their respective enhanced description information. After obtaining the descriptive feature vectors corresponding to the enhanced description information of the target symptom and the descriptive feature vectors corresponding to the enhanced description information of each drug to be matched, the server performs an associated adaptation operation through the preset drug decision model to determine the fitness coefficient of the target symptom for each drug to be matched. This drug decision model is also a deep learning model trained with a large amount of cardiovascular disease-related case data and corresponding medication effect data. Taking nitroglycerin as an example, the server loads the descriptive feature vectors corresponding to the enhanced description information of the target symptom (Mr. Zhang's case) and the descriptive feature vectors corresponding to the enhanced description information of nitroglycerin into the preset drug decision model. A series of adaptation indicators are set inside the drug decision model, such as symptom relief degree indicator, adverse reaction probability indicator, drug mechanism of action and symptom matching indicator, etc. For the symptom relief index, the model will give an evaluation value for the symptom relief degree based on the characteristics of nitroglycerin dilating coronary arteries and increasing myocardial blood supply to relieve angina symptoms, combined with the severity of Mr. Zhang's current angina symptoms such as chest pain and chest tightness and the expected relief effect. For the probability of adverse reactions, given that Mr. Zhang has a history of hypertension, the model will consider the probability of adverse reactions of hypotension that may be caused by the use of nitroglycerin, and combine the occurrence of this adverse reaction in previous similar cases to give an evaluation value for the probability of adverse reactions. For the drug mechanism and symptom matching index, the model will analyze the matching degree between the mechanism of action of nitroglycerin dilating coronary arteries and Mr. Zhang's suspected angina attack and coronary atherosclerosis leading to worsening myocardial ischemia, and give a corresponding evaluation value. Then, the drug decision model comprehensively calculates the fitness coefficient of nitroglycerin according to the evaluation values ​​of these different fitness indicators and a certain weight calculation method (these weights are determined during the model training process). Assume that after calculation, the fitness coefficient of nitroglycerin is 0.8. In the same process, the server also performs the above-mentioned association adaptation operation on other drugs to be matched, such as aspirin and atorvastatin, and determines their respective fitness coefficients. For example, the fitness coefficient of aspirin may be 0.7, and the fitness coefficient of atorvastatin may be 0.6, etc. After obtaining the fitness coefficients of each drug to be matched, the server determines the target drug to be matched and associated with the target symptom based on these coefficients. Calculate the adaptation confidence: The server first determines the adaptation confidence corresponding to each fitness coefficient based on the fitness indicator array corresponding to each fitness coefficient and the importance factor of at least two fitness indicators in the fitness indicator array corresponding to each fitness coefficient. Taking nitroglycerin as an example, assuming that its fitness coefficient is 0.8, its adaptation indicator array includes symptom relief degree index, adverse reaction probability index, and drug action mechanism and symptom matching index.During the model training process, the importance factor of the symptom relief index was determined to be 0.4, the importance factor of the adverse reaction probability index was 0.3, and the importance factor of the drug mechanism and symptom matching index was 0.3. Then, the fitness confidence of nitroglycerin = 0.8×0.4+0.8×0.3+0.8×0.3=0.8 (here it is calculated in a simple weighted average manner, and the actual situation may be more complicated). Similarly, for aspirin, assuming that its fitness coefficient is 0.7, its fitness index array includes the symptom relief index, the gastrointestinal bleeding risk index, the drug mechanism and symptom matching index, and the importance factor of the symptom relief index is 0.5, the importance factor of the gastrointestinal bleeding risk index is 0.3, and the importance factor of the drug mechanism and symptom matching index is 0.2. Then the fitness confidence of aspirin = 0.7×0.5+0.7×0.3+0.7×0.3=0.7. For atorvastatin, assuming that its fitness coefficient is 0.6, its fitness index array includes blood lipid regulation effect index, liver function damage risk index, drug action mechanism and symptom matching index, and the importance factor of blood lipid regulation effect index is 0.4, the importance factor of liver function damage risk index is 0.3, and the importance factor of drug action mechanism and symptom matching index is 0.3. Then the fitness confidence of atorvastatin = 0.6×0.6+0.6×0.3+0.6×0.3=0.6. Screening target drugs: The server arranges the fitness confidence of each drug in descending order from high to low. Assume that after the arrangement, nitroglycerin has the highest fitness confidence, followed by aspirin, and then atorvastatin. If the preset ranking is the top two, then nitroglycerin and aspirin in the top two will be used as target drugs for matching and associating with Mr. Zhang's target symptoms. In other words, the server will recommend nitroglycerin and aspirin to the doctor as a reference for medication for Mr. Zhang's current target symptoms such as chest pain, chest tightness, and suspected angina attack. The doctor can further make a comprehensive judgment and make the final medication decision based on the patient's specific situation, such as age, medical history, and the severity of current symptoms.

[0017] In an embodiment of the present invention, based on the description information of the target symptom and the description information of each drug to be matched, a description information enhancement operation is performed by presetting a first expert model to determine the enhanced description information of the target symptom and the enhanced description information of each drug to be matched, which can be implemented through the following example.

[0018] According to the description information of the target symptom and the preset first guide word, a description information enhancement operation is performed by a preset first expert model to determine the enhanced description information of the target symptom; According to the description information of each drug to be matched and the preset second guide word, the description information enhancement operation is performed through the first expert model to determine the enhanced description information of each drug to be matched.

[0019] In an embodiment of the present invention, for example, it is assumed that the server is the core server of the intelligent medication assistance system of a comprehensive hospital, which is responsible for handling medication decision-making related matters for many patients. A patient named Ms. Wang came to see a doctor. She described to the doctor that she always felt dizzy and had a headache recently, and this symptom would be significantly aggravated when standing for a long time or standing up suddenly, accompanied by mild tinnitus. After the doctor asked Ms. Wang in detail about her symptoms, living habits, medical history and other information, he recorded these contents in the electronic medical record system. The server obtained the description information about Ms. Wang’s target symptoms from the electronic medical record system: "Patient Ms. Wang has symptoms of dizziness and headache. The dizziness is aggravated when standing for a long time or standing up suddenly. The headache is mild to moderate, accompanied by mild tinnitus. There is no obvious history of trauma recently. The daily routine is regular but the work pressure is high." The server has a preset first guide word, which is specifically designed for symptoms like Ms. Wang's that may involve blood circulation or the nervous system. It is roughly as follows: "Combining the patient's symptoms, lifestyle, medical history, and environmental factors that may affect the symptoms, we comprehensively and deeply analyze the potential causes and development trends of the symptoms, paying special attention to the relationship between symptoms and possible diseases, and providing more targeted and detailed symptom description information for precise medication." The server then merges the description information of Ms. Wang's target symptoms into the preset first guide word. For example, "Patient Ms. Wang has symptoms of dizziness and headache, which worsens when she stands for a long time or stands up suddenly, with mild to moderate headache, accompanied by mild tinnitus, no recent history of obvious trauma, regular daily routine but high work pressure" is integrated into the guide words to form the optimized first guide words, making it more suitable for Ms. Wang's specific situation, such as "Combined with Ms. Wang's symptoms of dizziness and headache (dizziness worsens when she stands for a long time or stands up suddenly, with mild to moderate headache, accompanied by mild tinnitus, no recent history of obvious trauma, regular daily routine but high work pressure), living habits, past medical history and current environmental factors that may affect the symptoms, the potential causes and development trends of the symptoms are comprehensively and deeply analyzed, with particular attention to the association between symptoms and possible diseases, providing more targeted and detailed symptom description information for precise medication". Then, the server loads this optimized first guide word into the preset first expert model. This first expert model is deeply trained with a large amount of medical case data and professional medical knowledge. It can use deep learning algorithms to perform reasoning and analysis based on the input guide word content. After deep learning reasoning of the first expert model, the server obtained enhanced description information of Ms. Wang's target symptoms: "Ms. Wang, 35 years old, works in an office, sits for a long time in front of a computer screen, and has high work pressure, which causes her to be in a state of tension for a long time. The dizziness and headache symptoms that appeared this time, combined with her living habits and work environment analysis, are likely to be caused by cervical spondylosis compressing blood vessels and nerves, thereby affecting the blood supply and oxygen supply to the brain.Dizziness is not only aggravated when standing for a long time or standing up suddenly, but also aggravated after working at a desk for a long time. The headache is mild to moderate and intermittent, accompanied by mild tinnitus. Over time, if not intervened in time, the symptoms may worsen further, affecting daily work and life. It is urgent to clarify and effectively use the medicine to relieve the symptoms and prevent the disease from worsening. "Based on Ms. Wang's symptoms, the server initially screened a group of drugs to be matched based on its built-in medical knowledge base and algorithm logic, such as betahistine mesylate (used to improve inner ear circulation and relieve symptoms such as dizziness and tinnitus), ibuprofen (can relieve mild to moderate headaches), and Jingfukang granules (for cervical spine-related diseases). For Betahistine Mesylate, the server obtains the following description: "Betahistine Mesylate is a histamine drug, mainly used to treat dizziness, tinnitus and other symptoms caused by inner ear circulation disorders. It relieves the symptoms by improving the inner ear microcirculation. Common adverse reactions include dry mouth and stomach discomfort." The second guide word preset in the server is set according to the characteristics of the drug and its compatibility with the current patient's symptoms. The content is roughly as follows: "Combined with the drug's own pharmacological effects, applicable diseases, adverse reactions and other characteristics, as well as the current patient's specific symptom characteristics, living habits, previous medical history and other factors, deepen the drug description information and highlight the drug's current The server integrates the description information of betahistine mesylate into the preset second guide word to form an optimized second guide word, for example: "Combined with the pharmacological effects, applicable diseases, adverse reactions and other characteristics of betahistine mesylate (a histamine drug, mainly used to treat dizziness, tinnitus and other symptoms caused by inner ear circulation disorders, it relieves the symptoms by improving the inner ear microcirculation, and common adverse reactions include dry mouth, stomach discomfort, etc.), as well as the specific symptoms of the current patient, Ms. Wang (dizziness, headache symptoms, dizziness aggravated by standing for a long time or standing up suddenly, mild to moderate headache, accompanied by mild ear Then, the optimized second guide word is loaded into the first expert model, and the enhanced description information of betahistine mesylate is determined through deep learning reasoning: "Betahistine mesylate is a commonly used drug to improve inner ear circulation. For patients like Ms. Wang, who suffer from dizziness and tinnitus caused by cervical spondylosis compressing blood vessels and nerves, thereby affecting blood supply and oxygen supply to the brain, it can effectively improve the inner ear microcirculation, and the effect of relieving dizziness and tinnitus may be more significant. However, given that Ms. Wang has been sitting for a long time and facing the computer screen, she has a lot of work pressure and may have weak gastrointestinal function. When using it, special attention should be paid to the occurrence of adverse reactions such as dry mouth and stomach discomfort. At the same time, the effect of the medication should be comprehensively evaluated in combination with her headache symptoms."For ibuprofen, its original description information is: "Ibuprofen is a non-steroidal anti-inflammatory drug that can relieve mild to moderate headaches. It has antipyretic, analgesic and anti-inflammatory effects. Common adverse reactions include gastrointestinal discomfort, etc.". According to the above process, the description information of ibuprofen is fused with the preset second guide word to form an optimized second guide word and loaded into the first expert model for reasoning to obtain the enhanced description information of ibuprofen: "As a commonly used drug for relieving mild to moderate headaches, ibuprofen can effectively relieve the headache symptoms caused by cervical spondylosis compressing blood vessels and nerves. However, considering that Ms. Wang sits for a long time and faces a computer screen, she has high work pressure and may have weak gastrointestinal function. When using it, special attention should be paid to the occurrence of adverse reactions such as gastrointestinal discomfort, and the effect of the medication should be comprehensively evaluated in combination with her symptoms of dizziness and tinnitus." For Jingfukang Granules, the original description information is such as: "Jingfukang Granules are mainly used to treat dizziness, headache, upper limb numbness and other discomfort symptoms caused by cervical spine related diseases. Through the synergistic effect of multiple Chinese medicinal ingredients, it has the effects of promoting blood circulation, removing blood stasis, unblocking meridians and relieving pain. Common adverse reactions are relatively few, but some patients may experience nausea, rash and other conditions". According to the same steps, the description information of Jingfukang Granules is merged with the preset second guide word. After the first expert model reasoning, the enhanced description information of Jingfukang Granules is obtained: "Jingfukang Granules is aimed at the patient Ms. Wang's dizziness, headache and other discomfort symptoms caused by cervical spine lesions compressing blood vessels and nerves. Its effects of promoting blood circulation, removing blood stasis, unblocking meridians and relieving pain can specifically relieve the symptoms. Since Ms. Wang has been sitting for a long time and facing the computer screen, she has a lot of work pressure and may have relatively weak physical resistance. Although common adverse reactions are relatively few when using it, it is still necessary to pay attention to the nausea and rash that may occur in individual patients. At the same time, it is necessary to combine her tinnitus symptoms to comprehensively evaluate the effect of the medication." Through the above detailed description information enhancement operation, the server can provide a more comprehensive, in-depth and practical information basis for the subsequent accurate determination of the compatibility of each drug with Ms. Wang's target symptoms.

[0020] In an embodiment of the present invention, the description information enhancement operation is performed based on the description information of the target symptom and the preset first guide word through a preset first expert model to determine the enhanced description information of the target symptom, which can be implemented through the following example.

[0021] Merging the description information of the target symptom into the preset first guide word to obtain an optimized first guide word; Loading the optimized first guide word into a preset first expert model, and determining enhanced description information of the target symptom through deep learning reasoning; The preset first guide words include symptom-related characteristic tendencies of symptoms, past symptom analysis records of symptoms, and a first medication request for the first expert model. The first medication request includes medication target information, and the medication target information is used to characterize the enhanced descriptive information of the target symptom output by the first expert model corresponding to the medication target information.

[0022] In the embodiment of the present invention, for example, a patient named Mr. Li comes to see a doctor. Mr. Li reported that he had a series of discomfort symptoms recently, mainly manifested as abdominal pain, bloating, loss of appetite, and occasional nausea and vomiting. These symptoms have lasted for a period of time and have had a significant impact on his daily life and work. After the doctor asked Mr. Li in detail about the details of his symptoms, the course of the disease, his eating habits, his medical history, and other information, he recorded them in full in the electronic medical record system. The server then obtained descriptive information about Mr. Li's target symptoms from the electronic medical record system: "Mr. Li, a 42-year-old patient, had abdominal pain, which was mainly concentrated in the right upper abdomen and was intermittent and dull. The pain was moderate and sometimes worsened to colic; accompanied by obvious abdominal distension, which worsened after eating; loss of appetite, especially aversion to greasy food; occasional nausea and vomiting, and the vomit was mostly stomach contents. Before the onset of the disease, he had attended a dinner party and consumed a lot of greasy food. He had a history of mild fatty liver disease." The first guide word preset in the server is specially designed for digestive system-related symptoms. Its content is as follows: "For various symptoms of the digestive system, it is necessary to comprehensively consider the symptom-related characteristic tendencies of the symptoms, such as the location, nature, and frequency, etc., historical analysis records of symptoms, including the diagnosis, treatment and recovery effects of similar symptoms in this hospital or other medical institutions, etc., combined with the first medication request for this first expert model. The medication target information in the first medication request is intended to allow the model to output enhanced target symptom description information that matches the medication target information, so as to provide a more targeted basis for subsequent precision medication. Specifically, for the current digestive system symptoms, it is necessary to focus on analyzing their relationship with dietary factors and past medical history, predict the development trend of symptoms, and clarify the direction of medication to relieve pain, improve digestive function, and prevent possible complications. "The server then integrates the description information of Mr. Li's target symptoms into the preset first guide word. For example, "Patient Mr. Li, 42 years old, had abdominal pain symptoms, mainly concentrated in the right upper abdomen, with intermittent dull pain, moderate degree, sometimes aggravated into colic; accompanied by obvious abdominal distension, which worsened after eating; loss of appetite, especially aversion to greasy food; occasional nausea and vomiting, and vomitus were mostly gastric contents. Before the onset of the disease, he had attended a dinner party and ate a lot of greasy food. He had a history of mild fatty liver disease.” is integrated into the above-mentioned first guide word to form the optimized first guide word: "For various symptoms of the digestive system, it is necessary to comprehensively consider the symptom-related characteristic tendencies of the symptoms. For example, the abdominal pain symptoms of patient Mr. Li this time, the pain is mainly concentrated in the right upper abdomen, presenting as intermittent dull pain, of moderate degree, and sometimes aggravated into colic, which reflects the characteristic tendencies of the location, nature, and frequency of pain; past analysis records of symptoms, including the diagnosis, treatment, and recovery effects of similar symptoms in this hospital or other medical institutions (currently for patient Mr. Li, although there is no clear and complete record of similar symptoms in the past, the general diagnosis and treatment of digestive system-related diseases can be referred to); at the same time, combined with the first medication request for this first expert model. The medication target information in the first medication request is intended to allow the model to output enhanced target symptom description information that matches the medication target information, so as to provide a more targeted basis for subsequent precise medication. Specifically, for the digestive system symptoms currently occurring in patient Mr. Li, it is important to analyze their association with dietary factors and past medical history. For example, if he had attended a dinner party before the onset of the disease and consumed a lot of greasy food, and had a history of mild fatty liver disease, the development trend of the symptoms should be predicted, and the direction of medication should be clarified to relieve pain, improve digestive function, and prevent possible complications. "The server loads the optimized first guide word into the preset first expert model. This first expert model is a deep learning model that has been deeply trained with a large amount of digestive system disease-related case data, professional medical knowledge, and clinical experience. It can use complex deep learning algorithms to perform reasoning and analysis based on the input optimized guide word content, so as to dig out deeper and more comprehensive information about the target symptoms. When the server inputs the optimized first guide word into the first expert model, the model begins deep learning reasoning. It first parses each element in the guide word and identifies key content such as detailed characteristic tendencies of the symptoms, past analysis records, and medication target information. Based on the understanding of these contents, the model will combine the large amount of digestive system disease knowledge graphs stored in it and related medical logic. For example, based on the location of Mr. Li's abdominal pain (right upper abdomen), nature (intermittent dull pain, sometimes exacerbated into colic), and association with dietary factors (onset after eating more greasy food) and past medical history (mild fatty liver), the model will refer to the diagnosis and treatment of similar digestive system diseases in the past to infer the potential causes of these symptoms. After a series of complex reasoning operations, the server obtained enhanced description information of Mr. Li's target symptoms through the first expert model: "Patient Mr. Li, age 42, based on the current digestive system symptoms such as abdominal pain, bloating, loss of appetite, nausea and vomiting, combined with his consumption of more greasy food before the onset of the disease and his history of mild fatty liver, after comprehensive analysis, it is very likely to be a problem with the gallbladder, such as cholecystitis or gallstones.The abdominal pain is mainly concentrated in the right upper abdomen, presenting as intermittent dull pain of moderate severity, but the frequency of attacks tends to increase gradually, and the pain is significantly aggravated during colic attacks, which may be related to abnormal gallbladder contraction function and poor bile excretion. The abdominal distension is more serious, and the feeling of abdominal distension is further aggravated after eating, mainly because the digestive system function is affected and gastrointestinal motility is slowed down. Loss of appetite persists, and the aversion to greasy food is stronger. This is not only due to the current digestive system discomfort, but also related to the fact that gallbladder disease may affect bile secretion and thus affect fat digestion. Nausea and vomiting occur occasionally, and the vomitus is mostly gastric contents, but as the disease progresses, if not intervened in time, the frequency of vomiting may increase, and bile reflux may occur. From the perspective of medication goals, in order to relieve Mr. Li's pain symptoms, improve his digestive function, and prevent possible complications such as acute attacks of cholecystitis and gallstone incarceration, it is necessary to clarify an effective medication plan as soon as possible, and pay close attention to symptom changes during medication, and adjust medication in time according to actual conditions. "Through such detailed description information enhancement operations, the server provides more accurate, comprehensive and targeted target symptom description information for the subsequent determination of drugs that match Mr. Li's target symptoms, which helps to improve the accuracy of medication decisions and better serve the patient's treatment needs. In actual medical scenarios, the server will continuously handle similar situations of many patients. Through this deep learning-based medication decision-making method, it provides powerful assistance to doctors and improves the efficiency and quality of the entire medical process.

[0023] In the embodiment of the present invention, the description information enhancement operation is performed by the first expert model according to the description information of each drug to be matched and the preset second guide word to determine the enhanced description information of each drug to be matched, which can be implemented through the following example.

[0024] Merging the description information of each drug to be matched into the preset second guide word to obtain the optimized second guide word; Loading the optimized second guide word into the first expert model, and determining the enhanced description information of each to-be-matched drug through deep learning reasoning; The preset second guide word includes a second medication request for the first expert model, the second medication request includes medication target information, and the medication target information is used to represent the enhanced description information of each to-be-matched drug outputted by the first expert model corresponding to the medication target information.

[0025] In an embodiment of the present invention, for example, there is a patient named Ms. Zhao who has a series of symptoms. As mentioned above, after the server processes and analyzes her symptom information, a group of drugs to be matched is preliminarily determined. Here, three of the drugs are used as examples for illustration, namely omeprazole, domperidone and anti-inflammatory choleretic tablets. The server first obtains the description information of omeprazole: "Omeprazole is a proton pump inhibitor, which is mainly used to inhibit gastric acid secretion. It is commonly used to treat gastric ulcers, duodenal ulcers, reflux esophagitis and other diseases related to excessive gastric acid. It acts specifically on the secretory microtubules composed of the apical membrane of gastric parietal cells and the proton pumps on the intracellular tubular vesicles, thereby inhibiting the final formation link of gastric acid and reducing gastric acid secretion. Common adverse reactions include headache, diarrhea, constipation, abdominal pain, etc., and long-term use may increase the risk of fractures, etc." The second guide word preset in the server is set as follows for this type of drug: "In combination with the second medication request for the first expert model, The medication target information in the medication request is intended to allow the model to output enhanced description information of the drugs to be matched that matches the medication target information. For the current drugs to be matched, it is necessary to comprehensively consider their pharmacological effects, applicable diseases, adverse reactions and other characteristics, as well as their compatibility with the patient's current symptoms, highlighting the drug's advantageous effects under the current condition and potential risks that require special attention, so as to provide a more targeted basis for the subsequent accurate assessment of the compatibility of drugs and symptoms. "The server integrates the description information of omeprazole into the preset second guide word to form an optimized second guide word, for example: "Combined with the second medication request for the first expert model, this medication The target information is designed to allow the model to output enhanced description information of the drug to be matched that matches the target information of the medication. For omeprazole, it is a proton pump inhibitor, mainly used to inhibit gastric acid secretion, and is commonly used to treat diseases related to excessive gastric acid, such as gastric ulcers, duodenal ulcers, and reflux esophagitis. It inhibits the final formation link of gastric acid and reduces gastric acid secretion by specifically acting on the secretory microtubules composed of the apical membrane of gastric parietal cells and the proton pumps on the tubular vesicles in the cells. Common adverse reactions include headache, diarrhea, constipation, abdominal pain, etc., and long-term use may increase the risk of fractures, etc. It is necessary to comprehensively consider its pharmacological effects, applicable diseases, adverse reactions, etc. Characteristics, as well as compatibility with the patient Ms. Zhao's current symptoms (as mentioned above, there are digestive system-related symptoms, which may involve abnormal gastric acid secretion, etc.), highlighting the drug's advantages under the current condition and potential risks that require special attention, so as to provide a more targeted basis for the subsequent accurate assessment of the compatibility of the drug with the symptoms. "For domperidone, its original description information is: "Domperidone is a prokinetic drug that promotes gastrointestinal motility and is mainly used to treat symptoms caused by insufficient gastrointestinal motility, such as indigestion, bloating, nausea, vomiting, and abdominal pain. It blocks dopamine receptors and promotes the release of acetylcholine, thereby enhancing gastrointestinal motility.Common adverse reactions include dry mouth, headache, diarrhea, rash, etc., and when used in combination with certain drugs, the risk of arrhythmia may be increased. "Similarly, the description information of domperidone is integrated into the preset second guide word to obtain the optimized second guide word: "In combination with the second medication request for the first expert model, the medication target information is intended to allow the model to output enhanced description information of the drug to be matched that matches the medication target information. For domperidone, it is a prokinetic drug that promotes gastrointestinal motility and is mainly used to treat symptoms caused by insufficient gastrointestinal motility, such as indigestion, bloating, nausea, vomiting, and abdominal pain. It blocks dopamine receptors and promotes the release of acetylcholine, thereby enhancing gastrointestinal motility. Common adverse reactions include dry mouth, headache, diarrhea, rash, etc., and when used in combination with certain drugs, the risk of arrhythmia may be increased. It is necessary to comprehensively consider its pharmacological effects, applicable diseases, dry mouth, and compatibility with Ms. Zhao's current symptoms (as mentioned above, there are digestive system-related symptoms, which may involve gastrointestinal motility deficiency, etc.), highlight the drug's advantages under the current condition and potential risks that need special attention, so as to provide a more targeted basis for the subsequent accurate assessment of the compatibility of the drug with the symptoms. "The description of Anti-inflammatory and Choleretic Tablets is: "Anti-inflammatory and Choleretic Tablets is a Chinese patent medicine, mainly composed of Chinese medicinal ingredients such as Andrographis paniculata, Cyperus rotundus, and Acanthopanax chinensis. It has the effects of clearing heat, removing dampness, and promoting bile, and is commonly used to treat cholecystitis, cholangitis and other biliary system diseases. Its mechanism of action is to promote bile secretion and excretion and reduce inflammatory response by regulating the physiological function of the biliary system. Common adverse reactions are relatively rare, and some patients may experience nausea, vomiting, rash, etc. "The description information of the anti-inflammatory and choleretic tablets is integrated into the preset second guide word to generate an optimized second guide word: "Combined with the second medication request for the first expert model, the medication target information is intended to allow the model to output enhanced description information of the drugs to be matched that matches that medication target information. As for the anti-inflammatory and choleretic tablets, it is a traditional Chinese medicine, mainly composed of Chinese medicinal ingredients such as Andrographis paniculata, Cyperus rotundus, and Acanthopanax odoratus. It has the effects of clearing heat, removing dampness, and promoting bile secretion. It is commonly used to treat cholecystitis, cholangitis and other diseases of the biliary system. Its mechanism of action is to promote bile secretion and excretion and reduce inflammatory response by regulating the physiological function of the biliary system. Common adverse reactions are relatively rare, and some patients may experience nausea, vomiting, rash, etc. It is necessary to comprehensively consider its pharmacological effects, applicable diseases, adverse reactions and other characteristics, as well as the compatibility with Ms. Zhao's current symptoms (as mentioned above, there are digestive system-related symptoms, which may involve biliary system diseases, etc.), highlight the advantages of the drug under the current condition and the potential risks that need special attention, so as to provide a more targeted basis for the subsequent accurate assessment of the compatibility of the drug and symptoms. "The server loads the optimized second guide word about omeprazole into the first expert model.This first expert model has been trained with a large amount of digestive system disease-related case data and drug knowledge, and can perform deep learning reasoning based on the input information. After the model's reasoning operation, the enhanced description information of omeprazole was determined: "As a proton pump inhibitor, omeprazole is particularly important for the patient Ms. Zhao's current digestive system symptoms (such as discomfort caused by abnormal gastric acid secretion). It can accurately act on the proton pump of gastric wall cells to reduce the formation of gastric acid from the source, thereby effectively alleviating the development of potential diseases such as gastric ulcers and duodenal ulcers that may be caused by excessive gastric acid. In view of the complexity of Ms. Zhao's possible condition, such as the possible concomitant gastrointestinal motility deficiency, although omeprazole can effectively inhibit gastric acid when used, it is also necessary to pay attention to its possible adverse reactions, such as headache. Pain, diarrhea, constipation, abdominal pain, etc., especially long-term use may increase the risk of fractures and requires close monitoring. Moreover, when used in combination with other drugs that may be used to treat Ms. Zhao’s digestive system symptoms, special attention should be paid to drug interactions to avoid aggravation of adverse reactions. "Similarly, the optimized second guide word about domperidone is loaded into the first expert model for deep learning reasoning to obtain enhanced description information of domperidone: "Domperidone is a gastrointestinal prokinetic drug. For the patient Ms. Zhao’s current digestive system symptoms (such as abdominal distension, nausea, vomiting, etc. caused by gastrointestinal motility deficiency), its effect of promoting gastrointestinal motility can quickly improve the condition of gastrointestinal motility deficiency. , making the transportation of food in the gastrointestinal tract smoother, thereby effectively relieving symptoms such as abdominal distension, nausea, and vomiting. However, because Ms. Zhao's condition may be more complicated, when using domperidone, in addition to paying attention to its common adverse reactions, such as dry mouth, headache, diarrhea, rash, etc., special attention should be paid to the possible increased risk of arrhythmia when used in combination with other drugs, especially when used at the same time as some drugs that may affect heart function, the patient's heart condition should be closely monitored to ensure the safety of medication. "The optimized second guide word about Anti-inflammatory and Choleretic Tablets is loaded into the first expert model for deep learning reasoning, and the enhanced description information of Anti-inflammatory and Choleretic Tablets is determined: "Anti-inflammatory and Choleretic Tablets as a traditional Chinese medicine , for the digestive system symptoms that Ms. Zhao may currently have (such as discomfort caused by biliary system diseases), its heat-clearing, dampness-removing and bile-promoting effects can effectively regulate the physiological functions of the biliary system, promote bile secretion and excretion, and reduce inflammatory responses, thereby preventing and treating potential diseases such as cholecystitis and cholangitis. Although its common adverse reactions are relatively rare, some patients may experience nausea, vomiting, rash, etc. During Ms. Zhao's use, she still needs to pay close attention to the occurrence of these adverse reactions. Moreover, since the ingredients of Chinese patent medicines are relatively complex, when used in combination with other drugs, special attention should be paid to drug interactions to avoid aggravation of adverse reactions or reduced efficacy."Through such description information enhancement operations, the server can provide more comprehensive, in-depth and practical drug description information for subsequent accurate evaluation of the compatibility of these drugs to be matched with Ms. Zhao's target symptoms, which helps to make more accurate medication decisions.

[0026] In an embodiment of the present invention, feature extraction is performed based on the enhanced description information of the target symptom and the enhanced description information of each drug to be matched to determine the description features corresponding to the enhanced description information of the target symptom and the description features corresponding to the enhanced description information of each drug to be matched. This can be implemented through the following examples.

[0027] Determining an original description feature vector of the target symptom and an original description feature vector of each of the drugs to be matched according to the enhanced description information of the target symptom and the enhanced description information of each of the drugs to be matched; According to the original description feature vector of the target symptom and the original description feature vector of each drug to be matched, feature extraction is performed by presetting a second expert model to determine the description features corresponding to the enhanced description information of the target symptom and the description features corresponding to the enhanced description information of each drug to be matched.

[0028] In the embodiment of the present invention, for example, a patient named Mr. Sun comes to see a doctor and he has a series of symptoms. After the previous steps, the server has obtained the enhanced description information of Mr. Sun's target symptoms and the enhanced description information of each drug to be matched that is initially screened for these symptoms. Here, several typical drugs are used as examples for explanation, such as amoxicillin, montmorillonite powder and bifidobacterium tetravalent live tablets, which are respectively targeted at different possible symptoms. The enhanced description information of Mr. Sun’s target symptoms obtained by the server is as follows: "The patient, Mr. Sun, aged 50, has recently developed diarrhea symptoms, with diarrhea occurring 5-6 times a day. The stool is watery, accompanied by abdominal cramps, moderate pain and frequent attacks, especially after eating. At the same time, the patient feels physically weak and in poor mental state. Looking back on his medical history, the patient had previously experienced intestinal infections, and may have recently developed intestinal problems again due to unclean diet. Taking all factors into consideration, the diarrhea symptoms this time may be caused by bacterial infection or intestinal flora imbalance, and if not treated in time, it may cause complications such as dehydration. It is urgent to clarify and effectively use the medication plan to relieve symptoms and prevent the disease from worsening." There is a preset descriptive information feature dictionary in the server. This dictionary is constructed based on a large amount of relevant knowledge such as medical diseases, symptoms and drug properties, and various features are represented by dummy variables. For example, for the feature "diarrhea", the corresponding code in the feature dictionary may be [1,0,0,0,0] (assuming that diarrhea, constipation, abdominal pain, other intestinal symptoms, and non-intestinal symptoms are encoded separately in the dictionary); the code corresponding to "watery stool" may be [0,1,0,0,0]; the code corresponding to "abdominal cramps" may be [0,0,1,0,0]; the code corresponding to "physical fatigue" may be [0,0,0,1,0]; the code corresponding to "having a history of intestinal infection" may be [0,0,0,0,1], and so on. Based on the enhanced description information of Mr. Sun's target symptoms and the encoding rules of the feature dictionary, the server encodes each element one by one, and then combines these codes to determine the original description feature vector of the target symptom. For example, combining the feature codes mentioned above may result in an original description feature vector similar to [1,1,1,1,1] (the actual situation will be more complicated depending on the specific encoding situation). Taking amoxicillin as an example, its enhanced description information is as follows: "Amoxicillin is a broad-spectrum antibiotic, mainly used to treat various bacterial infectious diseases, especially intestinal bacterial infections. For Mr. Sun, whose diarrhea symptoms may be caused by bacterial infection, amoxicillin can kill pathogenic bacteria by inhibiting the synthesis of bacterial cell walls, thereby achieving the purpose of treating diarrhea. However, amoxicillin may cause some adverse reactions during use, such as diarrhea, nausea, vomiting, rash, etc., and attention should be paid to interactions with other drugs when using it to avoid affecting the efficacy of the drug or increasing the occurrence of adverse reactions."Similarly, according to the preset descriptive information feature dictionary, each element in the enhanced description information of amoxicillin is encoded. For example, the code corresponding to "broad-spectrum antibiotics" may be [1,0,0,0,0] (assuming that the dictionary encodes different types of drug properties such as antibiotics, probiotics, antidiarrheal drugs, etc.); the code corresponding to "treating bacterial infections" may be [0,1,0,0,0]; the code corresponding to "inhibiting bacterial cell wall synthesis" may be [0,0,1,0,0]; the code corresponding to "adverse reaction diarrhea" may be [0,0,0,1,0]; the code corresponding to "interaction with other drugs" may be [0,0,0,0,1] and so on. Combining these codes, the original descriptive feature vector of amoxicillin is determined, which may be a vector similar to [1,1,1,1,1] (the actual situation will be more complicated according to the specific coding situation). For montmorillonite powder, its enhanced description information is: "Montmorillonite powder is an antidiarrheal drug that mainly plays an antidiarrheal role by absorbing pathogens, toxins and excess water in the intestines. For Mr. Sun's watery stools and frequent diarrhea, montmorillonite powder can quickly absorb harmful substances in the intestines, reduce the frequency of diarrhea, and relieve diarrhea symptoms. It has relatively few adverse reactions, the most common of which is mild constipation, and generally does not interact with other drugs, but when using it, you should also pay attention to observe the patient's bowel movements to ensure that the antidiarrheal effect is not excessive. "According to the above method, the enhanced description information of montmorillonite powder is encoded according to the feature dictionary to determine its original description feature vector. For example, the code corresponding to "antidiarrhea drug" may be [1,0,0,0,0]; the code corresponding to "adsorbing pathogens, toxins, and moisture" may be [0,1,0,0,0]; the code corresponding to "mild constipation adverse reaction" may be [0,0,0,1,0]; the code corresponding to "no interaction with other drugs" may be [0,0,0,0,1], and so on. Combining these codes, the original description feature vector of montmorillonite powder is obtained. For Bifidobacterium tetravalent live bacteria tablets, the enhanced description information is: "Bifidobacterium tetravalent live bacteria tablets are a probiotic preparation, mainly used to regulate intestinal flora and improve intestinal microecological environment. For the diarrhea symptoms caused by intestinal flora imbalance that may exist in patient Mr. Sun, Bifidobacterium tetravalent live bacteria tablets can supplement the beneficial flora in the intestine, inhibit the growth of harmful bacteria, thereby restoring normal intestinal function and alleviating diarrhea symptoms. Pay attention to the storage conditions when using, and avoid high temperature, humidity and other environments, otherwise it may affect the number of live bacteria and reduce the efficacy of the drug. "Similarly, the enhanced description information of the Bifidobacterium Tetravalent Live Bacteria Tablets is encoded according to the feature dictionary to determine its original description feature vector. For example, the code corresponding to "probiotic preparation" may be [1,0,0,0,0]; the code corresponding to "regulate intestinal flora" may be [0,1,0,0,0]; the code corresponding to "supplement beneficial flora" may be [0,0,0,1,0]; the code corresponding to "pay attention to storage conditions" may be [0,0,0,0,1] and so on.These codes are combined to obtain the original description feature vector of the Bifidobacterium Tetravalent Live Bacteria Tablets.

[0029] The server loads the original descriptive feature vector of the target symptom and the original descriptive feature vector of each drug to be matched into the preset second expert model. This second expert model is a deep learning model trained with a large amount of medical data related to symptoms, drugs and their interactions, and is designed to extract more refined and comprehensive descriptive features from the original descriptive feature vector. When the original descriptive feature vector of the target symptom is loaded into the second expert model, the model begins to perform feature extraction operations based on its internal algorithm and the knowledge base built based on a large amount of medical knowledge and previous case data. For example, the model may identify certain feature combinations in the vector that are related to the severity of diarrhea, the frequency of abdominal pain, and the potential risk of dehydration. It may then map these combinations to more abstract and meaningful descriptive features. For example, if the original feature vector indicates a high frequency of diarrhea, moderate abdominal pain, and a history of intestinal infection, the model may extract descriptive features such as "severe acute diarrhea with potential risk of dehydration due to previous infection and current pain symptoms." This extracted descriptive feature is more comprehensive and targeted than a single element in the original descriptive feature vector. It takes into account multiple aspects of the target symptom and allows people to understand the situation in more detail. Take amoxicillin as an example. When the original descriptive feature vector of amoxicillin is loaded into the second expert model, the model also performs feature extraction operations. The model may identify key features related to its antibiotic properties, such as the mechanism of inhibiting bacterial cell wall synthesis and the effectiveness of treating bacterial infections. It also considers potential adverse reactions and drug interactions. Based on these analyses, the model may extract descriptive features such as "a potent antibiotic used to treat bacterial infections with potential adverse reactions and drug interactions." For montmorillonite powder, after loading its original descriptive feature vector into the second expert model, the model may extract descriptive features such as "an effective antidiarrheal drug that can absorb harmful substances in the intestine, has very few adverse reactions and no obvious drug interactions." For Bifidobacterium tetravalent live tablets, the model may extract descriptive features such as "a probiotic preparation used to regulate intestinal flora, restore normal intestinal function, and has specific preservation requirements." Through these feature extraction operations of the second expert model, the server can obtain more detailed and targeted descriptive features for the target symptoms and each drug to be matched. These descriptive features will be used in subsequent steps to accurately determine the compatibility coefficient between the target symptoms and each drug to be matched, thereby providing more accurate and reliable drug recommendation information for doctors to make more precise medication decisions.

[0030] In an embodiment of the present invention, determining the original description feature vector of the target symptom and the original description feature vector of each drug to be matched based on the enhanced description information of the target symptom and the enhanced description information of each drug to be matched can be implemented through the following example.

[0031] Determine, according to the enhanced description information of the target symptom and a preset description information feature dictionary, the description information feature corresponding to each element in the enhanced description information of the target symptom, and determine, according to the description information feature corresponding to each element in the enhanced description information of the target symptom, an original description feature vector of the target symptom, wherein the description information features in the description information feature dictionary are represented by dummy variables; According to the enhanced description information of each drug to be matched and a preset description information feature dictionary, the description information feature corresponding to each element in the enhanced description information of each drug to be matched is determined, and according to the description information feature corresponding to each element in the enhanced description information of each drug to be matched, the original description feature vector of each drug to be matched is determined.

[0032] In an embodiment of the present invention, for example, it is assumed that the server is the core server of an intelligent medication decision support system of a large comprehensive hospital, which is responsible for processing medication-related information of many patients to assist doctors in making more accurate medication decisions. In this scenario, a patient named Ms. Chen came to see a doctor. She had a series of complex symptoms. After a series of previous processing procedures, the server has obtained enhanced description information of Ms. Chen's target symptoms and enhanced description information of each drug to be matched that has been preliminarily screened for these symptoms. Here, several common and representative drugs are used as examples for detailed description, including cefixime, omeprazole, and Huoxiang Zhengqi Water. The enhanced description information of Ms. Chen's target symptoms obtained by the server is as follows: "The patient, Ms. Chen, aged 45, has recently developed more complex gastrointestinal discomfort symptoms. The specific manifestations are upper abdominal pain, which is persistent and dull, and occasionally worsens to stabbing pain, with a moderate to severe pain level; accompanied by frequent nausea and vomiting, and the vomitus are mostly stomach contents, and the vomiting symptoms are significantly aggravated after eating; there is also abdominal distension, and the abdomen feels bloated and uncomfortable, especially after eating a lot or eating greasy food; in addition, the patient reported a bitter taste in the mouth and bad breath, and a poor mental state, feeling weak and tired. Looking back on the medical history, Ms. Chen had a history of chronic gastritis before the onset of this symptom. He had eaten some spicy and greasy food, and had been under a lot of work pressure recently, with irregular work and rest. Taking all factors into consideration, the gastrointestinal discomfort symptoms may be caused by the recurrence of chronic gastritis, coupled with improper diet and mental stress factors. If not treated in time, it may further affect gastrointestinal function and cause more serious digestive problems. It is urgent to clarify the effective medication plan to relieve symptoms and prevent the disease from worsening. "There is a preset descriptive information feature dictionary in the server. This dictionary is carefully constructed based on a large amount of medical diseases, symptoms, drug properties and their relationship and other related knowledge. Various features are represented by dummy variables to facilitate subsequent data analysis and processing.For example, for the feature "upper abdominal pain", the corresponding code in the feature dictionary may be [1,0,0,0,0,0,0,0] (assuming that the dictionary has codes for pain in different parts of the body, such as upper abdominal pain, lower abdominal pain, chest pain, headache, joint pain, muscle pain, pain in other parts of the body, and no pain); the code for "persistent dull pain" may be [0,1,0,0,0,0,0,0]; the code for "occasionally aggravated to stabbing pain" may be [0,0,1,0,0,0,0,0]; the code for "moderate to severe pain" may be [0,0,0,1,0 ,0,0,0]; the code for "frequent nausea" may be [0,0,0,0,1,0,0,0]; the code for "vomiting" may be [0,0,0,0,0,1,0,0]; the code for "abdominal distension" may be [0,0,0,0,0,0,1,0]; the code for "bitter mouth and bad breath" may be [0,0,0,0,0,0,0,1]; the code for "history of chronic gastritis" may be [0,0,0,0,0,0,0,1] (here it is assumed that the medical history-related features are also similarly encoded and the encoding method is unified with other symptom features), etc. Based on the enhanced description information of Ms. Chen's target symptoms and the encoding rules of the feature dictionary, the server encodes each element one by one to determine the descriptive information feature corresponding to each element. For example, for the element "upper abdominal pain", according to the above-mentioned assumed encoding rules, the corresponding description information feature is [1,0,0,0,0,0,0,0,0]; the corresponding description information feature of "persistent dull pain" is [0,1,0,0,0,0,0,0,0]; and so on, all elements in the enhanced description information are encoded in this way. Then, the server combines these encoded description information features in a certain order (usually according to the order in which the elements appear in the symptom description or according to the preset sorting order of the features in the feature dictionary) to determine the original description feature vector of the target symptom. Assume that after encoding and combination, the original description feature vector of Ms. Chen's target symptom may be a vector similar to [1,1,1,1,1,1,1,1] (the actual situation will be more complicated depending on the specific encoding situation and the number of elements). 2. According to the enhanced description information of each drug to be matched and the preset description information feature dictionary, the description information feature corresponding to each element in the enhanced description information of each drug to be matched is determined, and the original description feature vector of each drug to be matched is determined. The enhanced description information of cefixime is as follows: "Cefixime is a broad-spectrum antibiotic, which is mainly used to treat a variety of bacterial infectious diseases, especially for bacterial infections in the respiratory tract, urinary tract and digestive tract. It has a good effect.For patients like Ms. Chen who may have potential bacterial infection caused by gastrointestinal discomfort (considering that she has a history of chronic gastritis and current gastrointestinal dysfunction, which may be prone to bacterial growth), cefixime can inhibit the synthesis of bacterial cell walls, thereby killing pathogenic bacteria, achieving the purpose of treating infection and relieving gastrointestinal discomfort. However, cefixime may cause some adverse reactions during use, such as diarrhea, rash, nausea, vomiting, etc., and attention should be paid to interactions with other drugs when using it to avoid affecting the efficacy or increasing the occurrence of adverse reactions. "Similarly, according to the preset description information feature dictionary, encode each element in the enhanced description information of cefixime. For example, for the feature of "broad-spectrum antibiotics", the corresponding code in the feature dictionary may be [1,0,0,0,0,0,0,0] (assuming that the dictionary encodes different types of drug properties such as antibiotics, antivirals, antifungals, analgesics, antipyretics, stomach medicines, other drug properties, and no properties); the corresponding code for "treating a variety of bacterial infectious diseases" may be [0,1,0,0,0,0,0,0]; the corresponding code for "inhibiting bacterial cell wall synthesis" may be [0,0,1,0,0,0,0,0]; the corresponding code for "adverse reaction diarrhea" may be [0,0,0,1,0,0,0,0]; the corresponding code for "adverse reaction rash" may be [0,0,0,0,1,0,0,0]; The code corresponding to "adverse reactions of nausea and vomiting" may be [0,0,0,0,0,1,0,0]; the code corresponding to "interaction with other drugs" may be [0,0,0,0,0,0,1,0] and so on. Such encoding operation is performed on all elements in the enhanced description information of cefixime, and then these encoded description information features are combined in a certain order to determine the original description feature vector of cefixime. Assume that after encoding and combination, the original description feature vector of cefixime may be a vector similar to [1,1,1,1,1,1,1,1] (the actual situation will be more complicated according to the specific encoding situation and the number of elements). The enhanced description information of omeprazole is as follows: "Omeprazole is a proton pump inhibitor, which is mainly used to inhibit gastric acid secretion and is commonly used to treat diseases related to excessive gastric acid, such as gastric ulcer, duodenal ulcer, reflux esophagitis, etc. Regarding the current gastrointestinal discomfort symptoms of Ms. Chen, considering that she has upper abdominal pain, nausea, vomiting, etc., which may be related to excessive gastric acid secretion, omeprazole can specifically act on the secretory microtubules composed of the apical membrane of gastric parietal cells and the proton pumps on the tubular vesicles in the cells, thereby inhibiting the final formation of gastric acid, reducing gastric acid secretion, and thus relieving symptoms such as upper abdominal pain, nausea, and vomiting. However, omeprazole may cause some adverse reactions during use, such as headache, diarrhea, constipation, abdominal pain, etc., and long-term use may increase the risk of fractures, etc. When using it, you need to pay attention to the interaction with other drugs to avoid affecting the efficacy or increasing the occurrence of adverse reactions.According to the above method, each element in the enhanced description information of omeprazole is encoded according to the preset description information feature dictionary. For example, for the feature of "proton pump inhibitor", the corresponding code in the feature dictionary may be [1,0,0,0,0,0,0,0]; the corresponding code of "inhibit gastric acid secretion" may be [0,1,0,0,0,0,0,0]; the corresponding code of "treating gastric ulcer, duodenal ulcer, reflux esophagitis and other diseases" may be [0,0, 0,0,0,0,0,1]; the code corresponding to "adverse reaction headache" may be [0,0,0,1,0,0,0,0]; the code corresponding to "adverse reaction diarrhea" may be [0,0,0,0,1,0,0,0]; the code corresponding to "adverse reaction constipation" may be [0,0,0,0,0,1,0,0]; the code corresponding to "adverse reaction abdominal pain" may be [0,0,0,0,0,0,1,0]; the code corresponding to "long-term use increases the risk of fracture" may be [0,0,0,0,0,0,0,1]; the code corresponding to "interaction with other drugs" may be [0,0,0,0,0,0,1,0] and so on. Such encoding operation is performed on all elements in the enhanced description information of omeprazole, and then these encoded description information features are combined in a certain order to determine the original description feature vector of omeprazole. Assume that after encoding and combination, the original description feature vector of omeprazole may be a vector similar to [1,1,1 ,1,1,1,1,1] (the actual situation will be more complicated according to the specific coding situation and the number of elements). The enhanced description information of Huoxiang Zhengqi Water is as follows: "Huoxiang Zhengqi Water is a Chinese patent medicine, mainly composed of Chinese medicinal ingredients such as Atractylodes, Tangerine Peel, Magnolia Bark, Angelica Dahurica, Poria, Big Belly Peel, Raw Pinellia, Licorice, Patchouli, Perilla Leaf, etc. It has the functions of dispelling dampness, regulating qi and harmonizing the middle, and strengthening the body and eliminating evil. It is often used to treat symptoms such as cold caused by exogenous wind and cold, internal injury and dampness, or colds and gastrointestinal discomfort caused by summer and autumn heat and dampness. For the current gastrointestinal discomfort symptoms of patient Ms. Chen, considering that it may be caused by improper diet, irregular work and rest, and mental stress, etc., Huoxiang Zhengqi Water can promote gastrointestinal motility, improve digestive function, and relieve symptoms such as abdominal distension, nausea, and vomiting by regulating the physiological function of the stomach and intestines. Although the adverse reactions of Huoxiang Zhengqi Water are relatively few, some patients may experience allergic reactions, such as rashes and itching, and attention should be paid to interactions with other drugs when using it to avoid affecting the efficacy of the drug or increasing the occurrence of adverse reactions. "Similarly, according to the preset descriptive information feature dictionary, each element in the enhanced description information of Huoxiang Zhengqi Water is encoded.For example, for the feature "Chinese patent medicine", the corresponding code in the feature dictionary may be [1,0,0,0,0,0,0,0,0]; the corresponding code for "the effects of dispelling dampness, regulating qi and harmonizing the middle, and strengthening the body's resistance and eliminating evil" may be [0,1,0,0,0,0,0,0]; the corresponding code for "treating exogenous wind-cold, internal injury and dampness, or colds, gastrointestinal discomfort and other symptoms caused by summer heat and dampness in summer and autumn" may be [0,0,0,0,0,0,0,1]; the corresponding code for "regulating gastrointestinal physiological functions" may be [0,0,0,0,0,0,1,0]; the corresponding code for "promoting gastrointestinal motility" may be [0,0,0,0,0,0,0,1]; the corresponding code for "adverse reactions and allergic reactions (rash, itching, etc.)" may be [0,0,0,1,0,0,0,0]; the corresponding code for "interaction with other drugs" may be [0,0,0,0,0,0,1,0], and so on. Such encoding operation is performed on all elements in the enhanced description information of Huoxiang Zhengqi Water, and then the features of these encoded description information are combined in a certain order to determine the original description feature vector of Huoxiang Zhengqi Water. Assume that after encoding and combination, the original description feature vector of Huoxiang Zhengqi Water may be a vector similar to [1,1,1,1,1,1,1,1] (the actual situation will be more complicated according to the specific encoding situation and the number of elements). Through the above detailed steps, the server determines the original description feature vector of the target symptom and the original description feature vector of each drug to be matched according to the enhanced description information of the target symptom and the enhanced description information of each drug to be matched. These vectors will serve as the basis for subsequent more in-depth feature extraction and analysis, so as to accurately determine the fitness coefficient between the target symptom and each drug to be matched, and thus provide doctors with a more accurate reference for medication decision-making.

[0033] In an embodiment of the present invention, feature extraction is performed based on the original descriptive feature vector of the target symptom and the original descriptive feature vector of each drug to be matched by presetting a second expert model to determine the descriptive features corresponding to the enhanced descriptive information of the target symptom and the descriptive features corresponding to the enhanced descriptive information of each drug to be matched, which can be implemented through the following examples.

[0034] Loading the original description feature vector of the target symptom into a preset second expert model, and determining a plurality of feature mapping description features corresponding to the original description feature vector of the target symptom through feature mapping processing; Determine the description feature corresponding to the enhanced description information of the target symptom through feature integration processing according to the multiple feature mapping description features corresponding to the original description feature vector of the target symptom; Loading the original description feature vectors of the drugs to be matched into the second expert model, and determining a plurality of feature mapping description features corresponding to the original description feature vectors of the drugs to be matched through feature mapping processing; According to the multiple feature mapping description features corresponding to the original description feature vectors of the drugs to be matched, the description features corresponding to the enhanced description information of the drugs to be matched are determined through feature integration processing.

[0035] In an embodiment of the present invention, illustratively, it is assumed that the server is the core server of an intelligent medication decision support system of a large hospital, which is responsible for processing medication-related information of many patients to assist doctors in making more accurate medication decisions. In this scenario, a patient named Mr. Lin comes to see a doctor, and he has a series of symptoms. After the previous steps, the server has obtained the original description feature vector of Mr. Lin's target symptoms and the original description feature vector of each drug to be matched that has been preliminarily screened for these symptoms. Here, several common drugs are used as examples to illustrate, such as azithromycin, aluminum carbonate magnesium chewable tablets and stomach-strengthening and digestion-promoting tablets. The original description feature vector of Mr. Lin's target symptoms obtained by the server is based on a detailed analysis of his symptoms before and is encoded according to a preset description information feature dictionary. For example, Mr. Lin's symptoms are mainly manifested as cough, sputum, fever, and sore throat. After encoding these symptoms according to the feature dictionary, a specific original description feature vector is formed. When the server loads the original description feature vector of this target symptom into the preset second expert model, the model begins to perform feature mapping processing. This second expert model is a deep learning model trained with a large amount of medical data (covering symptom manifestations, disease development, drug efficacy and other related data), and has strong feature analysis and extraction capabilities. During the feature mapping process, the model will analyze the input original description feature vector based on its internal complex algorithm and pre-learned knowledge patterns. Taking Mr. Lin's symptoms as an example, assuming that the part of the code representing "cough" in the original description feature vector will be mapped into a series of feature mapping description features related to coughing, such as "characteristics of dry cough", "characteristics of wet cough", "characteristics of cough frequency", "characteristics of cough severity", etc. Similarly, for the feature of "sputum", it may be mapped into "characteristics of sputum color", "characteristics of sputum texture", "characteristics of the difficulty of sputum", etc.; "fever" will be mapped into "characteristics of fever temperature range", "characteristics of fever duration", "characteristics of other symptoms associated with fever", etc.; "swelling and sore throat" will be mapped into "characteristics of the degree of redness and swelling of the throat", "characteristics of the degree of pain in the throat", "characteristics of whether there is a foreign body sensation in the throat", etc. Through such feature mapping processing, the server obtains multiple feature mapping description features corresponding to the original description feature vector of the target symptom from the second expert model. These feature mapping description features are more detailed and specific, and can reflect various aspects of the target symptoms more deeply. After obtaining the multiple feature mapping description features corresponding to the original description feature vector of the target symptom, the server then performs feature integration processing. The purpose of feature integration processing is to comprehensively analyze these scattered and refined feature mapping description features, so as to extract a description feature that can more comprehensively and accurately describe the enhanced description information of the target symptom. Taking Mr. Lin’s symptoms as an example, the above-mentioned refined features were obtained after feature mapping processing.During the feature integration process, the model will consider the relationship between these features and their impact on the overall symptoms. For example, if the "dry cough characteristics" are more obvious, and the "temperature range characteristics of fever" show low fever, the "degree of throat swelling characteristics" are also relatively mild, and the "frequency characteristics of cough" are occasional coughs, then after feature integration, the model may extract such a descriptive feature: "It is a mild upper respiratory tract infection, mainly dry cough, accompanied by low fever and mild throat swelling, and the frequency of coughing is also relatively low." For another example, if the "wet cough characteristics" are very prominent, the "sputum color characteristics" are yellow, the "fever duration characteristics" are relatively long, and the "throat pain characteristics" are also very severe, then after feature integration, the model may extract the following descriptive features: "It is a relatively serious upper respiratory tract infection, manifested as a wet cough accompanied by yellow sputum, with persistent fever, and obvious throat pain." Through such feature integration processing, the server determines the descriptive features corresponding to the enhanced description information of the target symptoms. This descriptive feature can more accurately summarize Mr. Lin's symptom status and provide a more targeted basis for subsequent analysis and the compatibility of the drugs to be matched. The server will also perform similar processing on each drug to be matched. Taking azithromycin as an example, its original descriptive feature vector is encoded according to the preset descriptive information feature dictionary based on its drug characteristics, applicable diseases, adverse reactions and other information. When the server loads the original descriptive feature vector of azithromycin into the second expert model, the model begins to perform feature mapping processing. For azithromycin, during the feature mapping process, it is assumed that the part of the code representing "antibiotics" in its original descriptive feature vector will be mapped into a series of feature mapping description features related to antibiotics within the model, such as "characteristics of the antibacterial spectrum", "characteristics of the antibacterial mechanism", "characteristics of antibiotic resistance", etc. The feature of "treating respiratory infections" will be mapped into "characteristics of effects on different respiratory pathogens", "characteristics of onset time of treating respiratory infections", "characteristics of treatment course requirements for treating respiratory infections", etc.; the part of "adverse reactions" will be mapped into "characteristics of types of common adverse reactions", "characteristics of probability of occurrence of adverse reactions", "characteristics of severity of adverse reactions", etc. Similarly, for aluminum carbonate chewable tablets, after its original description feature vector is loaded into the second expert model, in the feature mapping process, for example, the part of the code representing "antacid" will be mapped into "characteristics of gastric acid neutralization ability", "characteristics of effects on different gastric acid secretion conditions", "characteristics of duration of antacid effect", etc.; the feature of "relieving stomach discomfort" will be mapped into "characteristics of relief effect on stomach pain", "characteristics of relief effect on bloating", "characteristics of relief effect on heartburn symptoms", etc.; the part of "adverse reactions" will be mapped into "characteristics of types of common adverse reactions", "characteristics of probability of occurrence of adverse reactions", "characteristics of severity of adverse reactions", etc.For Jianweixiaoshi tablets, after its original description feature vector is loaded into the second expert model, in the feature mapping process, for example, the part of the code representing "digestive aid" will be mapped to "characteristics of promoting digestive ability", "characteristics of the effect on different indigestion conditions", "characteristics of the duration of digestive effect", etc.; the feature of "relieving abdominal distension and fullness" will be mapped to "characteristics of the effect of relieving abdominal distension", "characteristics of the effect of relieving fullness", "characteristics of the effect of increasing appetite", etc.; the part of "adverse reactions" will be mapped to "characteristics of the types of common adverse reactions", "characteristics of the probability of adverse reactions", "characteristics of the severity of adverse reactions", etc. Through such feature mapping processing, the server obtains multiple feature mapping description features corresponding to the original description feature vectors of each drug to be matched from the second expert model. These feature mapping description features can reflect the characteristics of each drug to be matched in more detail and more specifically.

[0036] After obtaining multiple feature mapping description features corresponding to the original description feature vectors of each drug to be matched, the server then performs feature integration processing. Taking azithromycin as an example, after feature mapping processing, if the "characteristics of antibacterial spectrum" shows that its antibacterial spectrum is relatively broad, the "characteristics of antibacterial mechanism" shows that it exerts antibacterial effects by inhibiting protein synthesis, the "characteristics of common adverse reactions" show that the most common adverse reactions are nausea and diarrhea, and the "characteristics of effects on different respiratory pathogens" show that it has a good effect on common respiratory pathogens such as Streptococcus pneumoniae and Haemophilus influenzae, then after feature integration processing, the server may extract such a description feature: "It is a broad-spectrum antibiotic that exerts antibacterial effects by inhibiting protein synthesis, has a good effect on common respiratory pathogens, and common adverse reactions are nausea and diarrhea." For aluminum carbonate chewable tablets, if the "characteristics of neutralizing gastric acid ability" show that it has a strong ability to neutralize gastric acid, the "characteristics of the effect on different gastric acid secretion conditions" show that it has a good effect on relieving stomach discomfort such as bloating and fullness, and the "characteristics of the types of common adverse reactions" show that the most common adverse reactions are constipation and mild abdominal discomfort, then after feature integration processing, the server may extract the following descriptive feature: "It is an antacid with a strong ability to neutralize gastric acid and a good effect on relieving stomach discomfort. Common adverse reactions are constipation and mild abdominal discomfort." For Jianweixiaoshi tablets, if the "characteristics of promoting digestion ability" show that it has a strong ability to promote digestion, the "characteristics of the effects on different indigestion conditions" show that it has a good effect on relieving abdominal distension, fullness and increasing appetite, and the "characteristics of the types of common adverse reactions" show that the most common adverse reactions are mild nausea and abdominal discomfort, then after feature integration processing, the server may extract such a descriptive feature: "It is a digestive aid with a strong ability to promote digestion, and has a good effect on relieving abdominal distension, fullness and increasing appetite. Common adverse reactions are mild nausea and abdominal discomfort." Through such feature integration processing, the server determines the descriptive features corresponding to the enhanced description information of each drug to be matched. These descriptive features can more accurately summarize the characteristics of each drug to be matched, and provide a more targeted basis for the subsequent analysis of the compatibility of the target symptoms and each drug to be matched.

[0037] In an embodiment of the present invention, the description features corresponding to the enhanced description information of the target symptom and the description features corresponding to the enhanced description information of each drug to be matched are used to determine the fitness coefficient of the target symptom for each drug to be matched through an associated adaptation operation, which can be implemented through the following example.

[0038] The descriptive features corresponding to the enhanced descriptive information of the target symptom and the descriptive features corresponding to the enhanced descriptive information of each drug to be matched are loaded into a preset drug decision model, and the fitness coefficient of the target symptom for each drug to be matched is determined through an associated adaptation operation. The fitness coefficient for each drug to be matched is determined by an adaptation indicator array, and the adaptation indicator array includes at least two adaptation indicators.

[0039] In an embodiment of the present invention, for example, it is assumed that the server is the core server of an intelligent medication decision support system of a large comprehensive hospital, which is responsible for processing medication-related information of many patients to assist doctors in making more accurate medication decisions. In this scenario, a patient named Mr. Zhang comes to see a doctor. He has a series of symptoms. After a series of processing steps, the server has obtained the description features corresponding to the enhanced description information of Mr. Zhang's target symptoms and the description features corresponding to the enhanced description information of each drug to be matched (here, amoxicillin, ibuprofen, and Jianwei Xiaoshi tablets are taken as examples) that are initially screened for these symptoms. The server first obtains the description features corresponding to the enhanced description information of Mr. Zhang's target symptoms. For example, Mr. Zhang's symptoms are mainly abdominal pain, abdominal distension, loss of appetite, and occasional nausea and vomiting. After the previous analysis and processing, the description features corresponding to the enhanced description information may be "digestive system disorders caused by improper diet, gastrointestinal inflammation may be present, manifested as moderate abdominal pain, abdominal distension, obvious loss of appetite, occasional nausea and vomiting, and there is a risk of developing more serious gastrointestinal problems." For amoxicillin, the descriptive feature corresponding to its enhanced description information may be "a broad-spectrum antibiotic that is effective against a variety of bacterial infections, especially suitable for gastrointestinal bacterial infections. It kills bacteria by inhibiting bacterial cell wall synthesis. There is a risk of adverse reactions such as diarrhea and rash. When using it, attention should be paid to interactions with other drugs." For ibuprofen, the descriptive feature corresponding to its enhanced description information may be "a non-steroidal anti-inflammatory drug that can relieve mild to moderate pain and inflammation. It has a certain effect on relieving abdominal pain, but may cause gastrointestinal discomfort such as nausea, vomiting, indigestion and other adverse reactions. It should be used with caution in patients with a history of gastrointestinal diseases." For Jianwei Xiaoshi Tablets, the descriptive feature corresponding to its enhanced description information may be "a Chinese patent medicine that is mainly used to promote digestion and improve appetite. It can regulate gastrointestinal function and relieve symptoms such as abdominal distension and loss of appetite. It has relatively few adverse reactions, but individual differences may exist. The server loads the descriptive features corresponding to the enhanced description information of these target symptoms and the descriptive features corresponding to the enhanced description information of each drug to be matched into the preset drug decision model. This preset drug decision model is a deep learning model that has been deeply trained with a large amount of medical case data, drug use effect data, and symptom-drug adaptation relationship data. It can perform complex association adaptation analysis based on the input feature information. In the preset drug decision model, a series of adaptation indicator arrays are set for amoxicillin. These adaptation indicators include symptom relief effect, probability of adverse reactions, drug mechanism of action and symptom matching, etc. Symptom relief effect indicator: The model will analyze the possibility of symptom relief based on the mechanism of amoxicillin's inhibition of bacterial cell wall synthesis and sterilization, as well as the symptoms (abdominal pain, abdominal distension, etc.) caused by Mr. Zhang's possible gastrointestinal bacterial infection.If we look at past case data and medical knowledge, amoxicillin has a higher probability of effectively relieving abdominal pain, bloating and other symptoms when treating symptoms similar to those caused by gastrointestinal bacterial infections, then a higher score will be given on this indicator, such as 0.8 points (assuming a full score of 1 point). Adverse reaction probability indicator: Since amoxicillin has the risk of adverse reactions such as diarrhea and rash, the model will combine Mr. Zhang's individual situation (such as age, physical condition, etc.) and the probability data of adverse reactions to the use of amoxicillin in the past to assess the possibility of these adverse reactions occurring in Mr. Zhang. Assuming that according to the analysis, Mr. Zhang's probability of adverse reactions such as diarrhea and rash is relatively moderate, then 0.4 points may be given on this indicator. Drug mechanism of action and symptom matching indicator: The mechanism of action of amoxicillin against bacterial infection is highly matched with Mr. Zhang's possible gastrointestinal bacterial infection, so a higher score will be given on this indicator, such as 0.9 points. Then, the drug decision model calculates the fitness coefficient of amoxicillin according to the scores of these different fitness indicators and a certain weight calculation method (these weights are determined during the model training process). Assuming that the weight of the symptom relief effect indicator is 0.4, the weight of the adverse reaction probability indicator is 0.3, and the weight of the drug mechanism and symptom matching indicator is 0.3, then the fitness coefficient of amoxicillin = 0.8×0.4+0.4×0.3+0.9×0.3=0.65. Similarly, for ibuprofen, the preset drug decision model also has a corresponding array of fitness indicators. Symptom relief effect indicator: Ibuprofen can relieve mild to moderate pain and inflammation, and has a certain effect on Mr. Zhang's abdominal pain. However, considering that it may cause adverse reactions such as gastrointestinal discomfort, and Mr. Zhang himself has symptoms of digestive system disorders, the model will comprehensively evaluate its actual effect on abdominal pain relief. Assuming that according to the analysis, ibuprofen has a relatively limited effect on Mr. Zhang's abdominal pain relief, and 0.5 points may be given on this indicator. Probability index of adverse reactions: Since ibuprofen may cause gastrointestinal discomfort such as nausea, vomiting, indigestion and other adverse reactions, combined with Mr. Zhang's situation, the model evaluates that the probability of these adverse reactions is relatively high, for example, 0.6 points are given. Drug mechanism of action and symptom matching index: The anti-inflammatory mechanism of ibuprofen has a certain degree of matching with the possible gastrointestinal inflammation of Mr. Zhang, but it is not a completely accurate match. 0.7 points may be given on this indicator. According to the above weight calculation method (assuming the weight is the same as amoxicillin), the fitness coefficient of ibuprofen = 0.5×0.4+0.6×0.3+0.7×0.3=0.57. For Jianwei Xiaoshi tablets, the array of adaptation indicators also works. Symptom relief effect index: Jianwei Xiaoshi tablets are mainly used to promote digestion and improve appetite. They can regulate gastrointestinal function and relieve symptoms such as abdominal distension and loss of appetite. Regarding Mr. Zhang's symptoms such as abdominal distension and loss of appetite, based on past case data and medical knowledge, there is a high probability that these symptoms can be effectively alleviated, so 0.8 points may be given on this indicator.Adverse reaction probability index: Jianweixiaoshi tablets have relatively few adverse reactions, but individual differences may result in different reactions. Combined with Mr. Zhang's situation, the model assesses that the probability of adverse reactions is relatively low, for example, 0.2 points are given. Drug mechanism of action and symptom matching index: The mechanism of action of Jianweixiaoshi tablets in regulating gastrointestinal function is highly matched with Mr. Zhang's digestive system disorder, and 0.9 points may be given on this boutique indicator. According to the above weight calculation method (assuming the weight is the same as amoxicillin), the fitness coefficient of Jianweixiaoshi tablets = 0.8×0.4+0.2×0.3+0.9×0.3=0.71. Through such an association adaptation operation, the server uses the preset drug decision model to determine the fitness coefficients of the target symptoms for each drug to be matched (amoxicillin, ibuprofen, Jianweixiaoshi tablets), which are 0.65, 0.57, and 0.71 respectively. These fitness coefficients will provide an important basis for the subsequent determination of target drugs for matching and associating with target symptoms, and assist doctors in making more accurate medication decisions.

[0040] In the embodiment of the present invention, the determination of the target drug matched with the target symptom according to each fitness coefficient can be implemented through the following example.

[0041] Determine the adaptation confidence corresponding to each fitness coefficient according to the adaptation index array corresponding to each fitness coefficient in each fitness coefficient and the importance factor of at least two adaptation indicators in the adaptation index array corresponding to each fitness coefficient; Arrange the confidence levels of each adaptive configuration in descending order from high to low, and determine multiple adaptive configuration confidence levels that are before a preset ranking; The drugs to be matched corresponding to the multiple adaptation confidences that are located before the preset ranking are used as target drugs to be matched and associated with the target symptom.

[0042] In the embodiment of the present invention, it is assumed that the server is the core server of the hospital intelligent medication decision support system, and processes the medication decision-related information of Ms. Li. The fitness coefficients of the drugs to be matched for Ms. Li's target symptoms (such as cefaclor, omeprazole, and Huoxiang Zhengqi water) have been obtained, and the target drug is now determined based on this. Cefaclor has a fitness coefficient of 0.7, and the fitness index array contains symptom relief effect, adverse reaction probability, and drug mechanism and symptom matching, and the corresponding importance factors are set to 0.4, 0.3, and 0.3, respectively. Calculate the fitness confidence: symptom relief effect contribution: 0.7×0.4=0.28; adverse reaction probability contribution: 0.7×0.3=0.21; drug mechanism and symptom matching contribution: 0.7×0.3=0.21. The fitness confidence is: 0.28+0.21+0.21=0.7. Omeprazole, fitness coefficient 0.65, importance factors are set to 0.35, 0.3, 0.35. Contribution to symptom relief effect: 0.65×0.35=0.2275; Contribution to the probability of adverse reactions: 0.65×0.3=0.195; Contribution to the matching degree between drug mechanism and symptoms: 0.65×0.35=0.2275. The fitness confidence is: 0.2275+0.195+0.2275=0.65. Huoxiang Zhengqi Water: fitness coefficient 0.6, importance factors are set to 0.4, 0.25, 0.35. Contribution to symptom relief effect: 0.6×0.4=0.24; Contribution to the probability of adverse reactions: 0.6×0.25=0.15; Contribution to the matching degree between drug mechanism and symptoms: 0.6×0.35=0.21. The appropriateness reliability is: 0.24+0.15+0.21=0.6. Arrange the appropriateness reliability of each drug in descending order: cefaclor 0.7, omeprazole 0.65, Huoxiang Zhengqi Water 0.6. Assume that the preset ranking is the top two, and the ones before this ranking are cefaclor 0.7 and omeprazole 0.65. Cefaclor and omeprazole corresponding to the appropriateness reliability in the above ranking are used as target drugs associated with Ms. Li's target symptoms. The doctor can refer to this result and determine the final medication plan based on Ms. Li's specific situation.

[0043] In the embodiment of the present invention, before obtaining the description information of the target symptom and the description information of each to-be-matched drug in the to-be-matched drug group, the following implementation manner is also provided.

[0044] Obtaining the descriptive features corresponding to the enhanced descriptive information of the symptom instance and the descriptive features corresponding to the enhanced descriptive information of each drug instance in the drug instance group; Loading the description features corresponding to the enhanced description information of the symptom instance and the description features corresponding to the enhanced description information of each drug instance into the initial drug decision model, and determining the fitness coefficient of the symptom instance for each drug instance through an associated adaptation operation; Determine the target cost function state according to the fitness coefficient for each drug instance and the preset actual fitness index; According to the target cost function state, the initial drug decision model is tuned until a preset state is reached according to the target cost function state, and the initial drug decision model is tuned to obtain a preset drug decision model.

[0045] In an embodiment of the present invention, the server first selects a group of symptom instances of respiratory tract infection from a large number of existing clinical case data. For example, there is a symptom instance of a patient named Xiao Wang, and the corresponding description feature of the enhanced description information after preliminary processing is: "Continuous fever, body temperature between 38.5℃-39.5℃, accompanied by severe cough, thick sputum when coughing, obvious sore throat, slight wheezing when breathing, it is speculated that it is a moderate respiratory tract infection caused by bacterial infection, and the patient's physical condition is general, and may be sensitive to adverse drug reactions". At the same time, a group of drug instances for respiratory tract infection, such as amoxicillin, azithromycin, cough syrup, etc., are selected from the drug library. Taking amoxicillin as an example, the description feature corresponding to its enhanced description information is: "Broad-spectrum antibiotics, have inhibitory effects on a variety of bacteria, especially suitable for respiratory bacterial infections, kill bacteria by inhibiting bacterial cell wall synthesis, common adverse reactions include diarrhea, rash, etc., and attention should be paid to interactions with other drugs when using it". The server loads the description features of the above Xiao Wang symptom instance and the description features of each drug instance such as amoxicillin into the initial drug decision model. This initial model is built based on certain medical knowledge and preliminary data, but it is not accurate enough. In the model, for amoxicillin, the fitness coefficient is determined by associating adaptation operations. The model will take into account multiple factors. For example, in terms of symptom relief effect, it will analyze the possibility of amoxicillin relieving Xiao Wang's fever, cough, thick sputum and other symptoms; in terms of the probability of adverse reactions, it will evaluate the probability of adverse reactions of amoxicillin in combination with the situation that Xiao Wang may be sensitive to adverse reactions; in terms of the matching degree between the mechanism of action of the drug and the symptoms, it will judge the degree of matching between the mechanism of action of amoxicillin for bacterial infection and the respiratory tract infection symptoms caused by Xiao Wang's inferred bacterial infection. After these analyses and calculations, it is assumed that the fitness coefficient of amoxicillin for Xiao Wang's symptom instance is 0.6. In the same way, the fitness coefficients of other drug instances such as azithromycin and cough syrup for Xiao Wang's symptom instance can be determined, such as 0.55 for azithromycin and 0.65 for cough syrup. There are preset actual adaptation indicators in the server. These indicators are set based on a large number of real medication effects and expert experience, and are used to measure the deviation between the model prediction results and the actual situation. Taking amoxicillin as an example, assume that the preset actual adaptation indicators include symptom relief accuracy, adverse reaction prediction accuracy, etc. If the adaptation coefficient of amoxicillin is 0.6, and according to the actual medication records and expert evaluation, for symptoms like Xiao Wang, the actual symptom relief accuracy index value that amoxicillin should achieve should be 0.7, and the adverse reaction prediction accuracy index value should be 0.8. By comparing the adaptation coefficient with these actual adaptation indicators, the difference between the two is calculated, and then the state of the target cost function is determined.For example, it is found that amoxicillin has a certain deviation in the accuracy of symptom relief and the accuracy of adverse reaction prediction. These deviations together constitute the current target cost function state for amoxicillin, indicating that the model's prediction of amoxicillin deviates from the actual situation and needs to be tuned. Based on the target cost function state of the above-mentioned amoxicillin and other drug instances, the server begins to tune the initial drug decision model. The tuning process adjusts the parameters inside the model, such as the weight of the neural network, to reduce the target cost function value and make it closer to the preset state. Assume that the preset state requires that the deviation between the fitness coefficient predicted by the model and the actual fitness index is within a certain range. During the tuning process, the server continuously repeats the above steps with new symptom instances and drug instance data, and continuously monitors the target cost function state. After multiple iterations and tuning, when the target cost function state reaches the preset state, that is, the deviation between the model's predicted fitness coefficient and the actual fitness index for each drug instance is within an acceptable range, the initial drug decision model at this time is optimized to a preset drug decision model, which can be used for subsequent more accurate medication decision analysis.

[0046] In the embodiment of the present invention, before obtaining the description information of the target symptom and the description information of each to-be-matched drug in the to-be-matched drug group, the following implementation manner is also provided.

[0047] Obtaining enhanced description information of the symptom instance and enhanced description information of each drug instance in the drug instance group; According to the enhanced description information of the symptom instance and the enhanced description information of each drug instance, a training process is performed on the initial second expert model and the initial drug decision model to obtain a preset second expert model and a preset drug decision model.

[0048] In an embodiment of the present invention, the server selects a large number of symptom instances and corresponding drug use cases from the hospital's rich medical record database. For example, for a patient named Xiao Zhao who suffers from cold symptoms, the enhanced description information of his symptom instance is: "Fever to about 38°C, accompanied by cough, runny nose, sore throat, and physical fatigue. It is speculated that it is a viral cold, and the patient has been tired recently and has a slightly low immunity." At the same time, the enhanced description information of the relevant drug instances is obtained from the medication record. Taking the commonly used cold medicine ibuprofen as an example, its enhanced description information is: "Non-steroidal anti-inflammatory drugs can relieve mild to moderate pain and fever, and have a relieving effect on fever and headache caused by colds. There may be adverse reactions such as gastrointestinal discomfort. Medication should pay attention to dosage and interaction with other drugs." The server uses Xiao Zhao's symptom instance enhanced description information and ibuprofen and other drug instances as training data. For the initial second expert model, these data are used to train it to perform better operations such as feature extraction. For example, the model can accurately extract key features from symptom instances, such as fever degree, cough characteristics, etc., and extract features such as drug mechanism of action and adverse reaction characteristics from drug instances. For the initial drug decision model, training is performed based on these data so that it can accurately judge the compatibility of symptoms and drugs. For example, the compatibility of Xiao Zhao's cold symptoms with ibuprofen in relieving fever and pain is analyzed, and the model parameters are continuously adjusted to make the prediction results closer to the actual medication effect. After repeated training of a large number of such symptom examples and drug example data, the initial second expert model and the initial drug decision model are gradually optimized, and finally the preset second expert model and preset drug decision model that can be used for actual medication decisions are obtained.

[0049] In an embodiment of the present invention, after the standardized drug recommendation information is pushed to the preset medication decision interface through the preset interface, the following implementation method is also provided.

[0050] Receiving feedback information from the preset medication decision interface, the feedback information including confirmation of the standardized medication recommendation information, feedback on the efficacy after medication, and whether new symptom information has appeared; If the feedback information contains uncertainty about the standardized drug recommendation information, the basic information of the target symptom is re-acquired according to the feedback information, and the re-acquired basic information is processed again to update the standardized drug recommendation information and re-pushed to the preset medication decision interface; If the feedback information contains new symptom information, the new symptom information is merged with the basic information of the original target symptom as the new target symptom basic information, and the standardized drug recommendation information for the new target symptom is re-determined and pushed to the preset medication decision interface.

[0051] In an embodiment of the present invention, for example, after a patient takes medication according to the standardized drug recommendation information pushed by the server, the doctor enters relevant feedback information in the preset medication decision interface, and the server immediately receives it. For example, the feedback information shows that although the patient's fever symptoms have been alleviated after taking the medication, the cough has not improved significantly, and a mild headache has occurred. At the same time, the doctor has doubts about the usage and dosage of some recommended drugs. This includes the uncertainty of the standardized drug recommendation information and the new symptom information. After the server detects the doctor's doubts about the usage and dosage of some recommended drugs in the feedback information, it will re-acquire the basic information of the patient's target symptoms from the hospital's electronic medical record system and other data sources based on this feedback information. For example, re-examine the patient's detailed course of illness, the specific details of the symptom manifestations, etc. Then the re-acquired basic information is processed again according to the previous processing flow, through a series of operations such as large language model processing and cooperation with the expert model to determine the target drug, update the standardized drug recommendation information, and finally push it back to the preset medication decision interface for the doctor's reference. When the server recognizes the new symptom information of mild headache in the patient in the feedback information, it will merge the new symptom information with the basic information of the original target symptoms (such as fever, cough, etc.) to form the new target symptom basic information. Then, just like the initial processing of the medication decision process, the standardized drug recommendation information for the new target symptom is re-determined, including re-screening matching drugs, integrating drug-related information, etc., and then pushed to the preset medication decision interface so that the doctor can make a more appropriate medication decision based on the new situation.

[0052] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned deep learning-based medication decision-making method. Figure 2 As shown, Figure 2 The block diagram of the computer device 100 provided in the embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112 and a communication unit 113. To achieve data transmission or interaction, the memory 111, the processor 112 and the communication unit 113 are directly or indirectly electrically connected to each other. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0053] For purposes of illustration, the foregoing description has been made with reference to specific embodiments. However, the illustrative discussions above are not intended to be exhaustive or to limit the disclosure to the precise forms disclosed.

Claims

1. A medication decision-making method based on deep learning, characterized in that: include: Obtain basic information on target symptoms; Inputting the basic information into a large language model for processing to obtain description information of the target symptom and description information of each to-be-matched drug in the to-be-matched drug group for the target symptom; According to the description information of the target symptom and the description information of each to-be-matched drug in the to-be-matched drug group, a target drug to be matched and associated with the target symptom is determined by presetting a first expert model; Integrate the relevant information of the target drug and convert it into a standard format to obtain standardized drug recommendation information, wherein the relevant information includes drug name, ingredients, efficacy, applicable symptoms, usage and dosage, adverse reactions, and interactions with other drugs; The standardized drug recommendation information is pushed to the preset medication decision interface through a preset interface.

2. The method according to claim 1, characterized in that The step of determining the target drug to be matched and associated with the target symptom by presetting a first expert model according to the description information of the target symptom and the description information of each drug to be matched in the drug group to be matched includes: Obtaining description information of the target symptom and description information of each to-be-matched drug in the to-be-matched drug group; According to the description information of the target symptom and the description information of each to-be-matched drug, a description information enhancement operation is performed by the preset first expert model to determine enhanced description information of the target symptom and enhanced description information of each to-be-matched drug; Determine, according to the enhanced description information of the target symptom and a preset description information feature dictionary, the description information feature corresponding to each element in the enhanced description information of the target symptom, and determine, according to the description information feature corresponding to each element in the enhanced description information of the target symptom, an original description feature vector of the target symptom, wherein the description information features in the description information feature dictionary are represented by dummy variables; Determine, according to the enhanced description information of each to-be-matched drug and a preset description information feature dictionary, the description information feature corresponding to each element in the enhanced description information of each to-be-matched drug, and determine, according to the description information feature corresponding to each element in the enhanced description information of each to-be-matched drug, the original description feature vector of each to-be-matched drug; Loading the original description feature vector of the target symptom into a preset second expert model, and determining a plurality of feature mapping description features corresponding to the original description feature vector of the target symptom through feature mapping processing; Determine the description feature corresponding to the enhanced description information of the target symptom through feature integration processing according to the multiple feature mapping description features corresponding to the original description feature vector of the target symptom; Loading the original description feature vectors of the drugs to be matched into the second expert model, and determining a plurality of feature mapping description features corresponding to the original description feature vectors of the drugs to be matched through feature mapping processing; According to the plurality of feature mapping description features corresponding to the original description feature vectors of the drugs to be matched, determining the description features corresponding to the enhanced description information of the drugs to be matched through feature integration processing; According to the description features corresponding to the enhanced description information of the target symptom and the description features corresponding to the enhanced description information of each to-be-matched drug, determining the fitness coefficient of the target symptom for each to-be-matched drug through an associated adaptation operation; According to each fitness coefficient, a target drug that is matched and associated with the target symptom is determined, and the group of drugs to be matched includes the target drug.

3. The method according to claim 2, characterized in that The step of performing a description information enhancement operation based on the description information of the target symptom and the description information of each to-be-matched drug by using the preset first expert model to determine the enhanced description information of the target symptom and the enhanced description information of each to-be-matched drug includes: According to the description information of the target symptom and the preset first guide word, a description information enhancement operation is performed by a preset first expert model to determine the enhanced description information of the target symptom; According to the description information of each drug to be matched and the preset second guide word, the description information enhancement operation is performed through the first expert model to determine the enhanced description information of each drug to be matched.

4. The method according to claim 3, characterized in that: The step of performing a description information enhancement operation based on the description information of the target symptom and the preset first guide word by using a preset first expert model to determine the enhanced description information of the target symptom includes: Merging the description information of the target symptom into the preset first guide word to obtain an optimized first guide word; Loading the optimized first guide word into a preset first expert model, and determining enhanced description information of the target symptom through deep learning reasoning; The preset first guide words include symptom-related characteristic tendencies of symptoms, past symptom analysis records of symptoms, and a first medication request for the first expert model. The first medication request includes medication target information, and the medication target information is used to characterize the enhanced descriptive information of the target symptom output by the first expert model corresponding to the medication target information.

5. The method according to claim 2, characterized in that: The step of performing a description information enhancement operation by the first expert model according to the description information of each to-be-matched drug and the preset second guide word to determine the enhanced description information of each to-be-matched drug includes: Merging the description information of each drug to be matched into the preset second guide word to obtain the optimized second guide word; Loading the optimized second guide word into the first expert model, and determining the enhanced description information of each to-be-matched drug through deep learning reasoning; The preset second guide word includes a second medication request for the first expert model, the second medication request includes medication target information, and the medication target information is used to represent the enhanced description information of each to-be-matched drug outputted by the first expert model corresponding to the medication target information.

6. The method according to claim 2, characterized in that The determining, through an associative adaptation operation based on the description features corresponding to the enhanced description information of the target symptom and the description features corresponding to the enhanced description information of each to-be-matched drug, the fitness coefficient of the target symptom for each to-be-matched drug comprises: The descriptive features corresponding to the enhanced descriptive information of the target symptom and the descriptive features corresponding to the enhanced descriptive information of each drug to be matched are loaded into a preset drug decision model, and the fitness coefficient of the target symptom for each drug to be matched is determined through an associated adaptation operation. The fitness coefficient for each drug to be matched is determined by an adaptation indicator array, and the adaptation indicator array includes at least two adaptation indicators.

7. The method according to claim 6, characterized in that Determining the target drug matched and associated with the target symptom according to each fitness coefficient includes: Determine the adaptation confidence corresponding to each fitness coefficient according to the adaptation index array corresponding to each fitness coefficient in each fitness coefficient and the importance factor of at least two adaptation indicators in the adaptation index array corresponding to each fitness coefficient; Arrange the confidence levels of each adaptive configuration in descending order from high to low, and determine multiple adaptive configuration confidence levels that are before a preset ranking; The drugs to be matched corresponding to the multiple adaptation confidences that are located before the preset ranking are used as target drugs to be matched and associated with the target symptom.

8. The method according to claim 2, characterized in that: Before obtaining the description information of the target symptom and the description information of each to-be-matched drug in the to-be-matched drug group, the method further includes: Obtaining the descriptive features corresponding to the enhanced descriptive information of the symptom instance and the descriptive features corresponding to the enhanced descriptive information of each drug instance in the drug instance group; Loading the description features corresponding to the enhanced description information of the symptom instance and the description features corresponding to the enhanced description information of each drug instance into the initial drug decision model, and determining the fitness coefficient of the symptom instance for each drug instance through an associated adaptation operation; Determine the target cost function state according to the fitness coefficient for each drug instance and the preset actual fitness index; According to the target cost function state, the initial drug decision model is tuned until the target cost function state reaches a preset state, and the initial drug decision model is tuned to obtain a preset drug decision model; Also includes: Obtaining enhanced description information of the symptom instance and enhanced description information of each drug instance in the drug instance group; According to the enhanced description information of the symptom instance and the enhanced description information of each drug instance, a training process is performed on the initial second expert model and the initial drug decision model to obtain a preset second expert model and a preset drug decision model.

9. The method according to claim 1, characterized in that: After the standardized drug recommendation information is pushed to the preset medication decision interface through the preset interface, the method further includes: Receiving feedback information from the preset medication decision interface, the feedback information including confirmation of the standardized medication recommendation information, feedback on the efficacy after medication, and whether new symptom information has appeared; If the feedback information contains uncertainty about the standardized drug recommendation information, the basic information of the target symptom is re-acquired according to the feedback information, and the re-acquired basic information is processed again to update the standardized drug recommendation information and re-pushed to the preset medication decision interface; If the feedback information contains new symptom information, the new symptom information is merged with the basic information of the original target symptom as the new target symptom basic information, and the standardized drug recommendation information for the new target symptom is re-determined and pushed to the preset medication decision interface.

10. A server system, characterized in that: The method comprises a server, wherein the server is used to execute the method described in any one of claims 1 to 9.

Images