Hospital inventory management method assisted by parsed electronic medical record based on LMKE

Abstract

The application discloses a kind of based on LMKE's analytic electronic medical record auxiliary hospital inventory management method, comprising the following steps: by image recognition algorithm, the medical record scanned copy is converted into free text and is stored into existing electronic medical record system library;The personnel basic information in electronic medical record, disease, use medicine and medical instrument are extracted;By the word item score calculation method of LMKE, the front as keyword is sorted;Obtain the basic information of character;Generate character basic label;Establish character basic label library;Compare patient basic label library;Compare corresponding chronic disease medical pathological knowledge graph, calculate the consumption of such drugs and instruments by patient number and treatment scheme;Carry out the statistical classification of medicine, calculate the number of existing drug varieties in inventory;Calculate the annual consumption of medicine, and generate a report periodically to submit inventory manager.The application solves the effective information extraction problem caused by a large amount of unstructured information extraction in hospital system electronic medical record.

Description

A Hospital Inventory Management Method Based on LMKE Analysis of Electronic Medical Records Technical Field

[0001] This invention belongs to the field of medical inventory management, and specifically relates to a method for analyzing electronic medical records to assist in hospital inventory management. Background Technology

[0002] Electronic medical records primarily come from two sources: scanned copies of medical records and diagnostic records and explanations entered into the system by doctors. Generally, hospitals need to retain patient medical records for 15 to 30 years. During this period, hospitals introduce electronic medical record systems. Previous medical record information needs to be supplemented with existing patient information through computer vision recognition.

[0003] Electronic medical record systems record patient information electronically. This includes the patient's medical history, examination and test results, medical orders, surgical records, nursing records, etc. It contains both structured information and unstructured free text, such as treatment plans.

[0004] LMKE is a keyword extraction method based on a language model. On experimental datasets, LMKE outperforms traditional keyword extraction methods TF-IDF and textRank in terms of evaluation metrics P@K and nDCG@K.

[0005] This invention aims to assist hospitals in inventory management by parsing free text and extracting diagnostic keywords. Summary of the Invention

[0006] The purpose of this invention is to provide a method for hospital inventory management based on LMKE-based parsing of electronic medical records, in order to solve the problem of extracting effective information from a large amount of unstructured electronic medical record information in hospital systems.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for hospital inventory management based on LMKE-based parsing of electronic medical records includes the following steps:

[0009] Step 1: Convert the scanned medical records into free text using an image recognition algorithm, store the text, and integrate it into the existing electronic medical record system database;

[0010] Step 2: Compile electronic medical records by extracting basic information about individuals, diseases, medications used, and medical devices from the electronic medical records and storing them in a corpus.

[0011] Step 3: Using LMKE's term score calculation method, the top-ranked terms are selected as keywords;

[0012] Step four: Obtain basic information about the individuals by retrieving their personal files;

[0013] Step 5: Generate basic tags for the individual based on their personal information and electronic medical records;

[0014] Step 6: Establish a basic tag library for individuals and store their tags;

[0015] Step 7: Extract keywords and compare them with the patient's basic tag database;

[0016] Step 8: Compare with the corresponding chronic disease pharmacology and pathology knowledge graph, and calculate the usage of such drugs and devices based on the number of patients and treatment plans;

[0017] Step nine: Classify hospital drugs and compare them with similar drugs, perform statistical classification of drugs, and calculate the quantity of each type of drug currently in stock;

[0018] Step 10: Based on the patient's tags, calculate the amount of medicine used in medical records, calculate the annual medicine consumption, and regularly generate reports to submit to the inventory administrator.

[0019] In step two, the jieba word segmentation tool is used to automatically segment the text, tag its parts of speech, remove stop words, and select parts of speech, and then store it in a corpus.

[0020] In step two, the patient's disease, medications, and medical devices are considered valid keywords. Entity recognition is performed by extracting these keywords.

[0021] Step three includes:

[0022] (31) Take all the medical records of patients in the electronic medical records as the corpus, add all the words to the medical record corpus dictionary V, and for any patient's medical record, calculate the terms generated from the medical record corpus. probability word frequency :

[0023] The calculation formula is:

[0024]

[0025] in, and All of these represent any term in the electronic medical record corpus. Indicates terms Document set frequency;

[0026] (32) Adopt a popularity-based strategy; the more frequently a drug appears, the more valuable it is to increase inventory.

[0027] This represents a medical record and consultation information. From the medical records Generate terms The formula for calculating the probability is as follows:

[0028]

[0029] in, Indicates terms In medical records Frequency of occurrence in;

[0030] (33) Calculate word frequencies according to the LMKE formula:

[0031]

[0032] in, This represents a medical record or visit information; Indicates from medical records Generate terms The probability of; The smoothing exponent;

[0033] (34) Given all electronic medical records for a given disease, score the terms for each case. The final term score is the sum of the scores for all medical record terms.

[0034] .

[0035] In step four, the basic information of the person includes the time of consultation, the department visited, age, and medical insurance usage.

[0036] In step five, the tag generation steps are as follows:

[0037] (51) Obtain the patient's disease and frequency based on the patient's electronic medical record extracted from the keyword;

[0038] (52) Disease classification is obtained by querying the medical pathology knowledge graph;

[0039] (53) Filter out chronic diseases and seasonally prevalent diseases, and label patients accordingly;

[0040] (54) By comparing the frequency of visits among patients, the probability of each patient having the disease in the population can be obtained, and the disease label description can be supplemented.

[0041] In step eight, the calculation of the dosage for chronic diseases is as follows:

[0042] (81) Search for chronic diseases in the medical pathology knowledge graph.

[0043] (82) Calculate the annual medication dosage based on the treatment cycle of chronic diseases and drug treatment;

[0044] (83) Based on the chronic diseases generated in step five, perform statistics on patients with the same chronic disease to obtain the number of patients;

[0045] (84) Calculate the annual inventory requirement based on the number of people and the amount of medicine used.

[0046] In step nine, the statistical classification steps for the drugs are as follows:

[0047] (91) In the drug knowledge graph system, query the inventory of drugs;

[0048] (92) Search for categories of the stocked medicines and classify them;

[0049] (93) Query the related similar drugs in the inventory and compare whether they exist in the inventory;

[0050] (94) Extract the attributes of the existing related and similar drugs in the inventory and add new categories based on the drug attributes.

[0051] Beneficial Effects: This invention addresses the issue of accumulated old and new medical records in electronic medical record systems, reducing manual medication dosage statistics and improving the efficiency of annual medication statistics. Simply relying on pharmacy usage for statistical prediction ignores the individual characteristics of patients, hindering model improvement. This invention, with a patient-centered approach and incorporating existing LMKE technology, aims to enhance model prediction performance. It is applicable to hospital queuing and patient triage systems.

[0052] In this invention, combining the business requirements of assisting hospital decision-making, tags can be extracted from patients through electronic medical records to predict the usage of medicines and medical devices. This can significantly improve the utilization rate of drug inventory. Previously, there were no related patents on automatically extracting tags from electronic medical records to calculate drug usage based on content. Only patents have achieved entity recognition in electronic medical records, converting unstructured free text into structured data. This invention can fill the technological gap in optimizing hospital inventory management. Attached Figure Description

[0053] Figure 1 is a flowchart of the present invention. Detailed Implementation

[0054] The invention will now be further explained with reference to the accompanying drawings.

[0055] As shown in Figure 1, the present invention provides a method for hospital inventory management based on LMKE-based parsing of electronic medical records, comprising the following steps:

[0056] Step 1: Convert the scanned medical records into free text using an image recognition algorithm, store it, and integrate it into the existing electronic medical record system database.

[0057] Step two involves compiling electronic medical records, extracting basic information about individuals, diseases, medications used, and medical devices; and then using the jieba word segmentation tool to automatically segment the text, tag parts of speech, remove stop words, and select parts of speech. The resulting data is then stored in a corpus.

[0058] Step 3: Using LMKE's term score calculation method, the top-ranked terms are selected as keywords;

[0059] LMKE scores terms based on each electronic medical record. The final term score is the sum of the scores for all drug and medical device terms.

[0060] The patient's illness, medications, and medical devices are considered valid keywords. Entity recognition is then performed by extracting these keywords.

[0061] Step three specifically includes:

[0062] (31) Take all the medical records in the electronic medical records as the corpus, add all the words to the vocabulary list V, and for any patient's medical record, calculate the terms generated from the medical record corpus. probability word frequency :

[0063] The calculation formula is:

[0064]

[0065] in Words document set frequency, and Each represents any term in the electronic medical record corpus, while V is the medical record corpus dictionary.

[0066] (32) In order to optimize inventory management by statistically analyzing commonly used drugs, a popularity-based strategy was adopted. The more frequently a drug appears, the more valuable it is to increase its inventory.

[0067] This represents a medical record and consultation information. From the medical records Generate terms The formula for calculating the probability is as follows:

[0068]

[0069] Indicates terms In medical records Frequency of occurrence.

[0070] (33) Calculate word frequencies according to the LMKE formula:

[0071]

[0072] This represents a medical record or visit information; Indicates from medical records Generate terms The probability of; It is the smoothing index.

[0073] (34) Given all the electronic medical records of a disease, the model can score the terms for each case, and the final term score is the sum of the terms scores of all medical records.

[0074] .

[0075] Step 4, Extracting Basic Information of the Person: By retrieving the patient's personal file, obtain the person's basic information, including the time of consultation, the department visited, age, and medical insurance usage.

[0076] Step 5: Based on the individual's personal information and electronic medical record information, generate basic tags for the individual, such as whether they suffer from seasonally recurring diseases, how often the patient suffers from the disease, and the severity of the recurrence (high, medium, or low).

[0077] The steps to generate tags are as follows:

[0078] (51) Obtain the patient's disease and frequency based on the patient's electronic medical record extracted from the keyword;

[0079] (52) Disease classification is obtained by querying the medical pathology knowledge graph;

[0080] (53) Filter out chronic diseases and seasonally prevalent diseases, and label patients with corresponding hypertension, seasonal allergies, etc.;

[0081] (54) Compare the frequency of visits among patients to obtain the probability of patients having the disease in the population and supplement the disease label description; such as seasonal influenza_high, seasonal allergy_moderate.

[0082] Step 6: Establish a basic tag library for people and use Elasticsearch to store the tags.

[0083] Step 7: Extract keywords and compare them with the patient's basic tag database.

[0084] Step 8: Compare with the corresponding chronic disease pharmacology and pathology knowledge graph, and calculate the usage of such drugs and devices based on the number of patients and treatment plans;

[0085] The steps for calculating the dosage for chronic diseases are as follows:

[0086] (81) Search for chronic diseases in the medical pathology knowledge graph.

[0087] (82) Calculate the annual medication dosage based on the treatment cycle of chronic diseases and drug treatment;

[0088] (83) Based on the chronic diseases generated in step five, such as hypertension, statistical analysis was conducted on patients with the same chronic diseases to obtain the number of patients;

[0089] (84) Calculate the annual inventory requirement based on the number of people and the amount of medicine used.

[0090] Step nine: Classify hospital drugs and compare them with similar drugs, perform statistical classification of drugs, and calculate the quantity of each type of drug currently in stock;

[0091] The statistical classification steps for pharmaceuticals are as follows:

[0092] (91) In the drug knowledge graph system, query the inventory of drugs;

[0093] (92) Search for categories of the stocked medicines and classify them;

[0094] (93) Query the related similar drugs in the inventory and compare whether they exist in the inventory;

[0095] (94) Extract the attributes of the existing related and similar drugs in the inventory and add new categories based on the drug attributes.

[0096] Step 10: Based on the patient's tags, calculate the amount of medicine used in medical records, calculate the annual medicine consumption, and regularly generate reports to submit to the inventory administrator.

[0097] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for hospital inventory management based on LMKE-based parsing of electronic medical records, characterized in that: The steps include: Step 1, converting the scanned medical records into free text using an image recognition algorithm and storing it in the existing electronic medical record system library; Step 2, summarizing the electronic medical records, extracting basic information about the personnel, diseases, medications used, and medical devices from the electronic medical records, and storing them in the corpus; Step 3, using the LMKE term score calculation method, selecting the top-ranked terms as keywords; Step 3 includes: (31) using all the consultation information in the electronic medical records as corpus, listing all words in the medical record corpus dictionary V, and for any patient's medical record, calculating the terms generated from the medical record corpus. probability word frequency ： The calculation formula is: in, and All of these represent any term in the electronic medical record corpus. Indicates terms The document set frequency; (32) Adopt a popularity-based strategy, the more frequently a drug appears, the more valuable it is to increase inventory. This represents a medical record and consultation information. From the medical records Generate terms The formula for calculating the probability is as follows: in, Indicates terms In medical records Frequency of occurrence in; (33) Calculate word frequency according to the LMKE formula: in, This represents a medical record or visit information; Indicates from medical records Generate terms The probability of; For smoothing index; (34) Given all electronic medical records of a disease, score the terms according to each case, and the final term score is the sum of the terms scores of all medical records. Step 4: Obtain basic information about the person by retrieving the patient's personal file; Step 5: Generate basic tags for the person based on personal information and electronic medical record information; The tag generation steps are as follows: (51) Obtain the patient's disease and frequency based on the patient's electronic medical record extracted by keywords; (52) Obtain the disease classification by querying the medical pathology knowledge graph; (53) Filter out chronic diseases and seasonally prevalent diseases and assign corresponding tags to the patients; (54) Compare the frequency of visits among patients to obtain the probability of the patient's disease in the population and supplement the disease tag description; Step 6: Establish a basic tag library for the person and store the person tags; Step 7: Compare the patient's basic tag library by extracting keywords; Step 8: Compare the corresponding chronic diseases... The knowledge graph of sexually transmitted diseases and medical pathology calculates the usage of such drugs and devices based on the number of patients and treatment plans; the steps for calculating the usage of chronic diseases are as follows: (81) In the knowledge graph of medical pathology, query chronic diseases; (82) Calculate the annual drug usage based on the treatment cycle and drug treatment of chronic diseases; (83) Combine the chronic diseases generated in step five, make statistics on patients with the same chronic diseases, and obtain the number of patients; (84) Calculate the annual inventory demand based on the number of patients and the amount of drugs used; Step nine, classify the hospital drugs and compare similar drugs, perform statistical classification of drugs, and calculate the number of existing drug types in the inventory; Step ten, calculate the drug usage of medical records based on the person's tags, calculate the annual drug consumption, and regularly generate reports to submit to the inventory administrator.

2. The method for hospital inventory management based on LMKE-based parsing of electronic medical records as described in claim 1, characterized in that: In step two, the jieba word segmentation tool is used to automatically segment the text, tag its parts of speech, remove stop words, and select parts of speech, and then store it in a corpus.

3. The method for hospital inventory management based on LMKE-based parsing of electronic medical records as described in claim 1, characterized in that: In step two, the patient's disease, medications, and medical devices are considered valid keywords. Entity recognition is performed by extracting these keywords.

4. The method for hospital inventory management based on LMKE-based parsing of electronic medical records as described in claim 1, characterized in that: In step four, the basic information of the person includes the time of consultation, the department visited, age, and medical insurance usage.

5. The method for hospital inventory management based on LMKE-based parsing of electronic medical records as described in claim 1, characterized in that: In step nine, the statistical classification steps for drugs are as follows: (91) Query the inventory of drugs in the drug knowledge graph system; (92) Query the categories of the inventory drugs and classify the drugs; (93) Query the related similar drugs of the inventory drugs and compare whether they exist in the inventory; (94) Extract the attributes of the existing related similar drugs in the inventory and add new categories based on the drug attributes.

Images