CN111191020B - Prescription recommendation method and system based on machine learning and knowledge graph - Google Patents
Prescription recommendation method and system based on machine learning and knowledge graph Download PDFInfo
- Publication number
- CN111191020B CN111191020B CN201911373211.9A CN201911373211A CN111191020B CN 111191020 B CN111191020 B CN 111191020B CN 201911373211 A CN201911373211 A CN 201911373211A CN 111191020 B CN111191020 B CN 111191020B
- Authority
- CN
- China
- Prior art keywords
- patient
- prescription
- recommendation
- diagnosis
- medicine
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/335—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
- G16H20/10—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H70/00—ICT specially adapted for the handling or processing of medical references
- G16H70/40—ICT specially adapted for the handling or processing of medical references relating to drugs, e.g. their side effects or intended usage
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Public Health (AREA)
- Primary Health Care (AREA)
- Computational Linguistics (AREA)
- Chemical & Material Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Medicinal Chemistry (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- Animal Behavior & Ethology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Pharmacology & Pharmacy (AREA)
- Toxicology (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The invention discloses a prescription recommendation method and system based on machine learning and a knowledge graph. The recommendation method comprises the following steps: obtaining and analyzing diagnosis information to obtain one or more diagnosis details including patient information, doctor information and disease diagnosis; inputting the diagnosis details into a pre-constructed machine learning model for reasoning to obtain an initial recommendation list aiming at the diagnosis details; acquiring a relationship graph network associated with the medicine from a medicine knowledge graph according to the medicine information contained in the initial recommendation list; and screening and eliminating tabu medicines according to the relation diagram network, and then giving a final recommendation list. The invention combines the quick learning capability of machine learning and the mining, association and display capability of the knowledge graph on the data, can realize accurate prescription recommendation, simultaneously improves the prescription generation efficiency, can meet the prescription requirement of outpatient service doctors, improves the working efficiency of the doctors and reduces medical errors.
Description
Technical Field
The invention relates to the technical field of medical informatization, in particular to a personalized prescription recommendation method and a personalized prescription recommendation system.
Background
In the medical service process, doctors diagnose the illness state of patients and then prescribe the corresponding prescriptions, and when prescribing the prescriptions, various factors such as age, sex, physical characteristics and the like need to be comprehensively considered, so that even though the patients with the same illness have different prescriptions and dosages, the patients with different prescriptions need different individuals. With the increase of patients, the clinic doctor's clinical pressure increases, the possibility of doctor's mistakes increases, and especially the doctor with less young experience is more likely to be repeatedly modified in the absence of powerful medical aid decision-making means.
Disclosure of Invention
The invention aims to: in order to solve the problems in the prior art, the invention provides a prescription recommendation method and a system based on machine learning and a knowledge graph, which realize accurate prescription recommendation and improve prescription generation efficiency.
The technical scheme is as follows: according to a first aspect of the present invention, there is provided a prescription recommendation method based on machine learning and knowledge graph, comprising the steps of:
obtaining and analyzing diagnosis information to obtain one or more diagnosis details including patient information, doctor information and disease diagnosis;
inputting the diagnosis details into a pre-constructed machine learning model for reasoning to obtain an initial recommendation list aiming at the diagnosis details;
acquiring a relationship graph network associated with the medicine from a medicine knowledge graph according to the medicine information contained in the initial recommendation list;
and screening and eliminating tabu medicines according to the relation diagram network, and then giving a final recommendation list.
The pre-constructed machine learning model comprises a collaborative filtering-based recommendation model, a knowledge-based recommendation model and a neural network-based recommendation model, and training is carried out through a collaborative filtering algorithm, a knowledge-based recommendation algorithm and a neural network learning algorithm to obtain three basic recommendation models; the method for obtaining the initial recommendation list comprises the following steps: and respectively inputting the diagnosis detail information into three basic recommendation models to obtain the respective recommended prescription medicines, removing repeated items, and sorting according to the predicted adoption probability to obtain an initial recommendation list.
According to a second aspect of the present invention, there is provided a prescription recommendation system based on machine learning and knowledge graph, comprising:
the diagnosis information analysis module is used for acquiring and analyzing diagnosis information to obtain one or more diagnosis details including patient information, doctor information and disease diagnosis;
the machine learning module is used for inputting the diagnosis detail data into the machine learning model for training to obtain a recommendation model for recommending prescriptions according to the patient information and the disease diagnosis;
the medicine relation diagram module is used for acquiring a relation diagram network associated with the medicine from a medicine knowledge graph according to medicine information;
the prescription recommendation module is used for recommending optional prescriptions according to the input patient information, doctor information and disease diagnosis;
and the recommendation result screening module is used for filtering and eliminating the recommended optional prescriptions according to the relation diagram network.
The beneficial effects are that: the invention provides a prescription recommendation method and a system based on machine learning and a knowledge graph, which realize accurate prescription recommendation and improve prescription generation efficiency. The recommendation system adopts a model fusion method, combines three machine learning algorithm training models of collaborative filtering, knowledge-based recommendation and neural network to conduct prescription recommendation, and provides the prescription recommendation for the outpatient. The invention can meet the requirements of the outpatient service doctors in the prescription, improve the working efficiency of the doctors and reduce medical errors.
Drawings
FIG. 1 is a flow chart of a recipe recommendation method provided by the present invention;
FIG. 2 is a schematic diagram of a drug knowledge graph in accordance with an embodiment of the invention;
FIG. 3 is a network of relationship graphs obtained from drug knowledge-graph in accordance with an embodiment of the invention;
FIG. 4 is a block diagram of a prescription recommendation system provided by the present invention.
Detailed Description
The technical scheme of the invention is further described below with reference to the accompanying drawings.
Referring to fig. 1, the invention provides a prescription recommendation method based on machine learning and knowledge graph, comprising the following steps:
and S1, acquiring and analyzing the diagnosis information to obtain one or more diagnosis details.
Along with the rapid development of medical informatization technology, massive prescription records are saved in electronic information systems of many hospitals, data can be exchanged through an open interface of a system such as a Hospital Information System (HIS)/Electronic Medical Record (EMR), diagnostic information in the electronic medical record is acquired, and diagnosis details are obtained through analysis of natural language processing technology. Specifically, acquiring text data of an electronic medical record of an outpatient service, including basic fields such as gender, birth date, pregnancy, lactation period, height, weight and payment mode of a patient and text contents of an allergy history, extracting the name of the allergic reaction medicine of the patient from the field of the allergy history by using a keyword matching method, and inquiring through a pre-obtained medicine alias library to obtain all names of the allergic reaction medicine; simultaneously acquiring outpatient record data, classifying through ICD10 codes of a diagnosis field, excluding diagnosis containing invalid or error codes, and acquiring a large number of 'disease diagnosis < - > prescription medicine' data pairs of the outpatient records, wherein the large number refers to acquiring more than five hundred percent of prescription records for each clinical diagnosis, so as to be used for training of the following model.
And S2, inputting the diagnosis details into a machine learning model for reasoning to obtain an initial recommendation list aiming at the diagnosis details.
The machine learning model is required to be trained firstly, and three original models are obtained by training through three machine learning algorithms of collaborative filtering, knowledge-based recommendation and a neural network respectively by taking disease diagnosis and basic patient information as input parameters and medicines as output parameters after the acquired data are cleaned and tidied; and fusing the three original models by using a model fusion technology to obtain a fused prescription recommendation model. When the method is applied to an outpatient service, patient information and doctor diagnosis are provided as input parameters to the fused prescription recommendation model, the model returns a plurality of recommendation results, the recommendation results are ranked from high to low according to the similarity, and the recommendation results are returned to the outpatient service doctor.
When training the original model, the collaborative filtering algorithm used is as follows:
s21, firstly, carrying out data normalization processing, mapping the field values of gender, age at diagnosis and pregnancy or not in the basic information of the patient to a space with 16 digits (hereinafter referred to as a fingerprint of the patient), and calculating the relative distance of the patient by using an AND relationship, such as gender men, age 10 years old, whether pregnancy can be mapped to 000000 01 0001010 0, gender women, age 11 years old, and whether pregnancy can be mapped to; 000000 00 0001011 0, the two numbers are bit by bit and the calculated 000000 000001010 0 is the calculated similar distance of the fingerprint value, and the smaller the similar distance value is, the higher the similarity is.
S21, sorting the data according to the fingerprint of the patient, the prescription medicine and the disease diagnosis triad, calculating the grading data of the basic condition of the patient and the disease diagnosis, and obtaining a patient similarity matrix by using the grading data as an index for measuring the similarity degree between the diseases of the patient and a distance calculation method.
Further, the step of calculating the scoring data of the patient base condition and disease diagnosis is as follows:
s21-1) reading stored patient fingerprint, prescription drug and disease diagnosis (user, article and score) triplet data;
s21-2) starting a merging process, extracting the number of the prescription medicine as a main key, taking the patient and the disease diagnosis as values, merging the data with the same prescription medicine number in storage according to the formats of the prescription medicine, the disease diagnosis and the fingerprint sequence pair of the patient;
s21-3) carrying out merging operation on the data of the same main key, and outputting a final result, wherein the result takes the prescription drug number as the main key, takes the fingerprint and disease diagnosis sequence pair of the patient as values, and stores the values in a database as prescription drug scoring data.
Further, the patient similarity matrix calculation step includes:
s21-a) reading locally stored patient fingerprint and disease diagnosis scoring data;
s21-b) rearranging data by taking a patient fingerprint pair as a primary key and taking a score pair of the patient fingerprint and disease diagnosis as a value, calculating the score distance between the patient fingerprint and the disease diagnosis, outputting the primary key as the patient fingerprint pair and taking the score distance between every two patient fingerprints for the same disease diagnosis as the value; wherein the scoring distance is encoded by the same disease diagnosis ICD10, and the similar distance of the patient fingerprint is used as the scoring distance;
s21-c) starting a merging process, merging data of the same patient fingerprint pairs to obtain distance values of all the same disease diagnosis scores of the two patient fingerprints, wherein a primary key of the merging result is a patient to be treated, the value is a score distance between the treated patient and the patient to be treated, and the merging result is stored and used as patient similarity matrix data.
The method comprises the following steps of:
a) Reading locally stored prescription medicine scoring data and patient similarity matrix data, sorting the prescription medicine scoring data, taking a patient fingerprint pair as a primary key, sorting the data of the scores of the prescription medicines and the patient for disease diagnosis as values, sorting the data of the patient similarity matrix, taking the patient fingerprint pair as a primary key, taking the patient distance and the sum of the distances as values, merging the two types of data, searching the scores of the prescription medicines and the scores of the prescription medicines which are not treated but are cured by the treated patient, calculating the recommended value of the patient to be treated for the prescription medicines, and storing the recommended value in a database as the recommended value of the prescription medicines; wherein the patient distance refers to the value calculated in step S21 and the sum of the distances refers to the patient distance plus the degree of similarity of the diagnoses (since there may be a plurality of diagnoses, and the diagnoses may be different in the same order, the degree of similarity is highest in which the order and the diagnosis are identical);
b) Ordering all recommended prescription medicines of the patient to be treated according to the recommended values of the prescription medicines, and selecting the prescription medicine with the highest recommended value as a recommended result.
The knowledge recommendation-based algorithm used in the original model training is as follows: according to the latest clinical guideline and expert consensus, a recommendation relation between fixed diagnosis (ICD 10 code) and prescriptions is formed, one code corresponds to one or more drug names and standard doses, the doses can be adjusted by doctors according to actual weights of patients, the recommendation based on knowledge is one of recommendation algorithms and is mainly used for directly using a medication proposal recommended by the guideline when similar patients or the same diagnosis cannot be found in a cold start condition, namely, the prescription record. When in use, the ICD10 coding format diagnosis result is input, the predicted recommendation result is output, and the knowledge-based recommendation algorithm is mainly used for solving the cold start problem, namely the condition of lacking pre-training data.
The neural network recommendation algorithm is used in personalized recommendation taking doctor historical prescription record as a theme as supplement of collaborative filtering. When the original model is trained, the neural network algorithm used is as follows:
S2A, obtaining a user behavior log reported by a doctor client, wherein the user behavior log comprises a doctor ID, ICD10 diagnosis codes, basic patient information (such as sex, birth date, pregnancy, lactation period, height, weight and payment mode of the patient), a prescription medicine list and dosages, and storing the prescription medicine list and dosages in a server basic database;
S2B, performing data acquisition, cleaning, standardization processing and feature combination and extraction on the user behavior log data by constructing a feature collector to obtain uniform and standard feature vectors which meet the mathematical modeling requirements;
further, constructing a feature collector to obtain a unified feature vector comprises the following specific steps:
S2B-1, performing the following treatment on the data: time sequence data is separated into dimensions required by each basic operation model according to different time units; performing Z-score standardization processing on the numerical data; carrying out semantic analysis on the text data; adopting single-heat coding to process enumeration classification data;
S2B-2, extracting characteristics of each item of processed data, and performing dimension reduction processing on the extracted characteristic vectors, wherein the characteristics are the information of each field given in the step S2A;
S2B-3, generating a feature vector which has unified specification and can be used for model training after dimension reduction processing, wherein the feature vector comprises the following components: user feature vectors, prescription drug feature vectors, and disease feature vectors.
S2C, the feature vector is used as input to be transmitted to a multi-layer neural network, and each layer of the network is a global neural network layer; the invention adopts the convolutional neural network for training, because the input data is not pictures and cannot be convolved in a split channel way, the convolution is carried out on all input data (vector matrix), and therefore, each layer is a convolutional neural network, called a global neural network layer, and is used for carrying out convolution operation on the input data, and the convolution result of the former layer is used as the input of the latter layer;
the output matrix becomes smaller after the convolution is carried out on the previous layer, so that the node number of the global neural network layer is reduced from the front to the back, and the node number of the global neural network layer of the previous layer is 2 times that of the global neural network layer of the next layer;
taking the prescription drug feature vector as an output target, taking the user feature vector and the disease feature vector which are subjected to dimension reduction processing as inputs, respectively mapping the user feature vector and the disease feature vector into dense vectors, and outputting the dense vectors to a first global neural network layer;
the method comprises the steps of training a neural network to obtain interlayer weight information of the neural network, and obtaining output information of a currently input user characteristic vector and a disease characteristic vector in the neural network, namely similarity with a prescription medicine characteristic vector, based on the trained interlayer weight information;
fully connecting the hidden layers of the last layer of the neural network, outputting predicted scores of the user and the disease to the prescription, constructing a user-disease-prescription score prediction table based on the predicted scores, and storing the user-disease-prescription score prediction table into a database;
when recommending prescriptions, a user-disease-prescription score prediction table is read from a database, prescriptions available for current diagnosis are ordered in descending order based on the prediction scores in the user-disease-prescription score prediction table, N prescriptions with the highest ordering are selected as a recommendation data table of the current user and recommended, and the N values can be dynamically adjusted.
Furthermore, the invention adopts a k-fold cross validation method to train each basic recommendation model, and in the training stage, each basic recommendation model adopts a grid search method to carry out parameter tuning to obtain optimal parameters, and a prescription medicine recommendation list of each user under each basic recommendation model and the utilization rate of each prescription medicine in the preliminary prescription medicine list by the corresponding user are generated.
The training of each basic recommendation model by adopting the k-fold cross validation method comprises the following steps:
a) Dividing the training sample set of each basic recommendation model into k subsets which are identical in size and mutually exclusive in content;
b) And carrying out k iterations, wherein each iteration adopts a union set of k-I subsets as a training set, the rest subsets are used as test sets, and the obtained k groups of training sets and test sets are used for training the basic recommendation model. Therefore, under the condition that the data volume is not large enough, the utilization rate of the data can be improved, and the condition of over fitting or under fitting is avoided. The value of I is adjusted according to experience and model training, and is usually 1.
Through training each basic recommendation model by adopting a k-fold cross validation method, the generalization capability of a single basic recommendation model is improved.
After the construction of the three basic recommendation models is completed, the fusion processing of the models is carried out. According to the input user and diagnosis information, calculation is carried out, the collaborative filtering model needs to calculate fingerprint and scoring distance, the knowledge-based recommendation model needs patient and diagnosis information, the neural network model needs feature vectors, a preliminary prescription medicine list of corresponding users for disease diagnosis under each basic recommendation model and the adoption probability of corresponding users for various prescription medicines in the preliminary prescription medicine list are obtained, diagnosis conditions corresponding to the medicine recommendation list with the adoption probability returned by the three models are obtained according to the similarity between patient information and doctor diagnosis and the currently input patient information and diagnosis, and the higher the adoption probability is considered if the similarity is close. Under the default condition, personalized recommendation based on a neural network > collaborative filtering > knowledge-based recommendation, taking the adoption probability obtained by each recommendation model as a new feature vector input, and sequencing from high to low according to the predicted adoption probability. After the recommended prescriptions obtained by all the recommended models are ordered, repeated recommendations are removed and displayed to a clinician in a list mode, the clinician can select whether to adopt the recommended prescriptions or not, and the actual adoption or not is used as feedback of the fusion recommended models, so that the recommended sequence of the fusion recommended models is readjusted.
And step S3, acquiring a relation graph network associated with the medicine from the medicine knowledge graph according to the medicine information contained in the initial recommendation list.
Firstly, acquiring a prescription list returned by a recommendation engine, extracting all medicine lists contained in each prescription, then acquiring records of corresponding entities of the medicines in a knowledge graph according to the names of the medicines, and acquiring an entity and a relation set associated with the medicines, namely, a set of attributes such as indication, medicine belongings, medicine interaction, contraindications and the like of the medicines, if multiple medicines of the same prescription can be associated through interaction, a communicated graph subset is formed, interaction relation of medicines in the subset is indicated, and otherwise, the medicines of the prescription are indicated to have no interaction. Fig. 2 is a schematic diagram of a drug knowledge graph, and fig. 3 is an example of a subset of graphs of a certain drug obtained from fig. 2.
And S4, screening medicines according to the medicine map subset network, inquiring the contraindications of all medicines for the map subset of each prescription returned by the recommendation engine, judging whether the medicines in the prescription have use contraindications according to the age, pregnancy, allergy history and other information of the patient, and eliminating the contraindications from a prescription recommendation result list if the contraindications are found.
Preferably, the medicine price can be obtained while recommending the prescription return, and the total price of a plurality of prescriptions is automatically calculated, and the steps comprise:
summarizing all the drug prices of the single recommended prescriptions to obtain the total price of the single recommended prescriptions;
a summarizing step is carried out on all recommended prescriptions to respectively obtain the total price of all prescriptions;
the prescriptions are ranked according to price, and prescriptions exceeding a preset total price threshold provide independent visual identification, so that a prompting effect is achieved for a clinician to control a large prescription.
According to another embodiment of the present invention, there is provided a prescription recommendation system based on machine learning and knowledge graph, as shown in fig. 4, including: the diagnostic information analysis module 100 is configured to obtain diagnostic information from an existing electronic medical record and analyze the diagnostic information to obtain one or more diagnostic details, including basic fields such as gender, birth date, pregnancy, lactation period, height, weight, payment mode, and text content of allergy history, extract names of allergic reaction drugs of the patient from the allergy history fields by using a keyword matching method, and obtain all names of allergic reaction drugs by querying a pre-obtained drug alias library; simultaneously acquiring outpatient record data, classifying through ICD10 codes of a diagnosis field, and excluding diagnosis containing invalid or error codes to obtain a large number of 'disease diagnosis < - > prescription medicine' data pairs of outpatient records;
the machine learning module 200 is configured to input the diagnosis detail data and the disease diagnosis data into a machine learning model for training, obtain a recommendation model for recommending a prescription according to patient information and the diagnosis detail, and obtain three recommendation models based on collaborative filtering, a neural network and knowledge according to a subject of the input data, such as a department, a doctor or a guideline; if the subject is a history prescription record from a doctor, obtaining a neural network model; if the subject is prescription data from all doctors in the department (multiple experts), a model based on collaborative filtering is obtained; if the topic is from guideline data, a knowledge-based model is derived. The training process of the original model is the same as the steps in the method embodiment, and will not be described here again.
The prescription recommendation module 300 is used for recommending optional prescriptions according to the input patient information, doctor information and disease diagnosis.
Specifically, the recommendation module 300 includes a recommendation unit 301 for transmitting input data to three recommendation models based on collaborative filtering, neural networks, and knowledge, and receiving initial prescription recommendation results from the recommendation models;
and a fusion unit 302, which integrates the initial prescription recommendation results provided by the three recommendation models, removes repeated items and sorts the results according to the similarity of patients.
The medicine relationship diagram module 400 is configured to obtain a relationship diagram network associated with the medicine from the medicine knowledge graph according to the medicine information included in the prescription recommended by the prescription recommendation module. The medicine relation diagram module 400 firstly acquires knowledge graph information of medicines and returns relation data composed of relevant medicine information nodes and sides representing association relations.
The recommendation result screening module 500 is configured to screen and exclude contraindicated drugs according to the relationship graph network.
Specifically, the recommendation result screening module 500 includes a matching unit 501, configured to match drug information included in a prescription returned by the prescription recommendation model with node and relationship information pre-stored in a drug knowledge graph, so as to obtain a network relationship including the recommended drug name, a drug indication, a drug belonging, a drug interaction, and a contraindication;
a filtering unit 502, configured to filter out prescriptions of entries containing contraindicated drugs in the recommendation result according to the aforementioned network relationship containing drugs and drug indications, drug belongings, drug interactions, and contraindicated drugs;
and a sending unit 503, configured to send the filtered recommended prescription information to the user.
Preferably, the recommendation result screening module 500 further includes a drug rule configuration unit 504, and controls the overall drug price of the prescription based on the relevant rule, and integrates the national medical insurance basic drug catalog, and prompts the outpatient physician to control the ratio of the basic drugs.
Preferably, the recommendation system further comprises a nurse station operation guide module 600, and the instructions of medicine injection mode, dripping speed, proportioning and the like are directly transmitted to the nurse station, so that risks caused by improper medication are reduced.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical aspects of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the above embodiments, it should be understood by those of ordinary skill in the art that: modifications and equivalents may be made to the specific embodiments of the invention without departing from the spirit and scope of the invention, which is intended to be covered by the claims.
Claims (5)
1. The prescription recommendation method based on machine learning and knowledge graph is characterized by comprising the following steps:
obtaining and analyzing diagnosis information to obtain one or more diagnosis details including patient information, doctor information and disease diagnosis;
inputting the diagnosis details into a pre-built machine learning model for reasoning to obtain an initial recommendation list aiming at the diagnosis details, wherein the pre-built machine learning model comprises a recommendation model based on collaborative filtering, a recommendation model based on knowledge and a recommendation model based on a neural network, and training is carried out through a collaborative filtering algorithm, a recommendation algorithm based on knowledge and a neural network learning algorithm respectively to obtain three basic recommendation models; the method for obtaining the initial recommendation list comprises the following steps: inputting the diagnosis detail information into three basic recommendation models respectively to obtain respective recommended prescription medicines, removing repeated items and sorting according to the predicted adoption probability to obtain an initial recommendation list;
acquiring a relationship graph network associated with the medicine from a medicine knowledge graph according to medicine information contained in the initial recommendation list, wherein the acquiring the relationship graph network associated with the medicine from the medicine knowledge graph comprises the following steps: acquiring an initial recommendation list, extracting all medicines contained in each prescription, acquiring records of corresponding entities of the medicines in a knowledge graph according to the names of the medicines, and acquiring entity and relation sets associated with the medicines, namely, an indication of the medicines, medicine interaction and a set of tabu attribute, wherein if a plurality of medicines of the same prescription can be associated through interaction, a communicated graph subset is formed and the interaction relation of the medicines in the subset is indicated, otherwise, the medicines of the prescription are indicated to have no interaction;
screening and eliminating tabu medicines according to the relation diagram network and then giving a final recommendation list;
the training of the original model by the collaborative filtering algorithm comprises the following steps:
s21, mapping the field values of gender, age at diagnosis and pregnancy in the basic information of the patient into a space with 16 digits as the fingerprint of the patient, and calculating the relative distance of the patient by using the AND relationship;
s22, sorting data according to the fingerprint of the patient, the prescription medicine and the disease diagnosis triad mode, calculating scoring data of the basic condition and the disease diagnosis of the patient, and obtaining a patient similarity matrix by a distance calculation method by taking the scoring data as an index for measuring the similarity degree between the diseases of the patient; wherein, the calculation steps of the scoring data of the basic condition and disease diagnosis of the patient are as follows:
a) Reading stored patient fingerprint, prescription drug and disease diagnosis triplet data;
b) Starting a merging process, extracting the prescription medicine number as a main key, taking the patient and the disease diagnosis as values, merging the data with the same prescription medicine number in storage according to the formats of the prescription medicine, the disease diagnosis and the fingerprint sequence pair of the patient;
c) And merging the data of the same main key, outputting a final result, wherein the result takes the prescription medicine number as the main key, takes the fingerprint and disease diagnosis sequence pair of the patient as values, and stores the values in a database as prescription medicine scoring data.
2. The prescription recommendation method according to claim 1, wherein the patient similarity matrix calculation step is:
a) Reading locally stored patient fingerprint and disease diagnosis scoring data;
b) The method comprises the steps of rearranging data by taking a patient fingerprint pair as a primary key, calculating the grading distance between the patient fingerprint pair and the disease diagnosis, outputting the primary key as the patient fingerprint pair, and taking the grading distance between every two patient fingerprints and the same disease diagnosis as the patient fingerprint pair;
c) And starting a merging process, merging the data of the fingerprint pairs of the same patient to obtain distance values of all the disease diagnosis scores of the fingerprints of the two patients, wherein a primary key of the merging result is a patient to be treated, the value is the score distance between the treated patient and the patient to be treated, and the value is the score distance between the treated patient and the patient to be treated and the score distance is stored as patient similarity matrix data.
3. The prescription recommendation method according to claim 1, wherein the training of the raw model by the knowledge-based recommendation algorithm comprises: based on the latest clinical guidelines and expert consensus, a recommendation of a fixed diagnostic ICD10 code to a prescription is formed.
4. The prescription recommendation method according to claim 1, wherein training of the neural network learning algorithm for the raw model comprises:
s2a, obtaining a user behavior log reported by a doctor client, wherein the user behavior log comprises a doctor ID, ICD10 diagnosis codes, basic patient information, a prescription drug list and doses, and the user behavior log is stored in a server basic database;
s2b, carrying out data acquisition, cleaning, standardization processing and feature combination and extraction on the user behavior log data by constructing a feature collector to obtain feature vectors which are unified and standard and can be used for model training, wherein the feature vectors comprise user feature vectors, prescription medicine feature vectors and disease feature vectors;
s2c, transmitting the feature vector into a multi-layer neural network, taking the user feature vector and the disease feature vector as input and taking the prescription drug feature vector as an output target, wherein each layer of the network is a global neural network layer, carrying out convolution operation on input data, and taking a convolution result of the former layer as input of the latter layer;
the method comprises the steps of training a neural network to obtain interlayer weight information of the neural network, and obtaining output information of a currently input user characteristic vector and a disease characteristic vector in the neural network, namely similarity with a prescription medicine characteristic vector, based on the trained interlayer weight information;
and fully connecting the last layer of the neural network, outputting predicted scores of the user and the disease prescription, constructing a user-disease-prescription score prediction table based on the predicted scores, and storing the user-disease-prescription score prediction table into a database.
5. A machine learning and knowledge graph based prescription recommendation system, comprising:
the diagnosis information analysis module is used for acquiring and analyzing diagnosis information to obtain one or more diagnosis details including patient information, doctor information and disease diagnosis;
the machine learning module is used for inputting diagnosis detail data into the machine learning model for training to obtain a recommendation model for recommending prescriptions according to patient information and disease diagnosis, and comprises a collaborative filtering model, a knowledge-based recommendation model and a neural network learning model, wherein three basic recommendation models are obtained by training through a collaborative filtering algorithm, a knowledge-based recommendation algorithm and a neural network algorithm respectively, and then the three basic recommendation models are fused to obtain the prescription recommendation model;
the prescription recommendation module is used for recommending optional prescriptions according to the input patient information, doctor information and disease diagnosis, and comprises the following steps: inputting the diagnosis detail information into three basic recommendation models respectively to obtain respective recommended prescription medicines, removing repeated items and sorting according to the predicted adoption probability to obtain an initial recommendation list;
the medicine relation diagram module is used for acquiring a relation diagram network associated with the medicine from a medicine knowledge graph according to medicine information in a prescription, and comprises the following components: extracting all medicines contained in each prescription according to an initial recommendation list, acquiring records of corresponding entities of the medicines in a knowledge graph according to the names of the medicines, and acquiring entity and relation sets associated with the medicines, namely, an indication of the medicines, medicine interaction and a set of tabu attribute, wherein if multiple medicines of the same prescription can be associated through interaction, a communicated graph subset is formed and the interaction relation of the medicines in the subset is indicated, otherwise, the medicines of the prescription are indicated to have no interaction;
the recommendation result screening module is used for filtering and removing the recommended optional prescriptions according to the relationship diagram network;
the training of the original model by the collaborative filtering algorithm comprises the following steps:
s21, mapping the field values of gender, age at diagnosis and pregnancy in the basic information of the patient into a space with 16 digits as the fingerprint of the patient, and calculating the relative distance of the patient by using the AND relationship;
s22, sorting data according to the fingerprint of the patient, the prescription medicine and the disease diagnosis triad mode, calculating scoring data of the basic condition and the disease diagnosis of the patient, and obtaining a patient similarity matrix by a distance calculation method by taking the scoring data as an index for measuring the similarity degree between the diseases of the patient; wherein, the calculation steps of the scoring data of the basic condition and disease diagnosis of the patient are as follows:
a) Reading stored patient fingerprint, prescription drug and disease diagnosis triplet data;
b) Starting a merging process, extracting the prescription medicine number as a main key, taking the patient and the disease diagnosis as values, merging the data with the same prescription medicine number in storage according to the formats of the prescription medicine, the disease diagnosis and the fingerprint sequence pair of the patient;
c) And merging the data of the same main key, outputting a final result, wherein the result takes the prescription medicine number as the main key, takes the fingerprint and disease diagnosis sequence pair of the patient as values, and stores the values in a database as prescription medicine scoring data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911373211.9A CN111191020B (en) | 2019-12-27 | 2019-12-27 | Prescription recommendation method and system based on machine learning and knowledge graph |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911373211.9A CN111191020B (en) | 2019-12-27 | 2019-12-27 | Prescription recommendation method and system based on machine learning and knowledge graph |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111191020A CN111191020A (en) | 2020-05-22 |
| CN111191020B true CN111191020B (en) | 2023-09-22 |
Family
ID=70707677
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911373211.9A Active CN111191020B (en) | 2019-12-27 | 2019-12-27 | Prescription recommendation method and system based on machine learning and knowledge graph |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111191020B (en) |
Families Citing this family (46)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111627515B (en) * | 2020-05-29 | 2023-07-18 | 上海商汤智能科技有限公司 | Medicine recommendation method, device, electronic equipment and medium |
| CN111667890A (en) * | 2020-05-29 | 2020-09-15 | 平安科技（深圳）有限公司 | Medication recommendation method, device, equipment and storage medium based on artificial intelligence |
| CN111798954A (en) * | 2020-06-11 | 2020-10-20 | 西北工业大学 | Drug combination recommendation method based on time attention mechanism and graph convolution network |
| CN113808751A (en) * | 2020-06-15 | 2021-12-17 | 中国中医科学院中国医史文献研究所 | Traditional Chinese medicine prescription evaluation and pushing method and device and electronic equipment |
| CN111753543B (en) * | 2020-06-24 | 2024-03-12 | 北京百度网讯科技有限公司 | Medicine recommendation method, device, electronic equipment and storage medium |
| CN111968715B (en) * | 2020-06-30 | 2022-11-01 | 厦门大学 | Drug recommendation modeling method based on medical record data and drug interaction risk |
| CN111833984B (en) * | 2020-07-16 | 2023-06-20 | 平安科技（深圳）有限公司 | Medicine quality control analysis method, device, equipment and medium based on machine learning |
| CN111883228B (en) * | 2020-07-28 | 2023-07-07 | 平安科技（深圳）有限公司 | Knowledge graph-based health information recommendation method, device, equipment and medium |
| CN111951915A (en) * | 2020-08-10 | 2020-11-17 | 智业软件股份有限公司 | Medicine and medicine usage recommendation method, terminal device and storage medium |
| CN112037880B (en) * | 2020-08-31 | 2022-07-15 | 康键信息技术（深圳）有限公司 | Medication recommendation method, device, equipment and storage medium |
| CN112017747B (en) * | 2020-09-04 | 2022-09-27 | 四川大学 | Intelligent recommendation method for enhanced CT contrast agent injection scheme |
| CN112133425A (en) * | 2020-09-11 | 2020-12-25 | 北京欧应信息技术有限公司 | System for assisting doctor to optimize diagnosis and treatment process based on intelligent recommendation |
| CN112071431B (en) * | 2020-09-18 | 2024-02-23 | 众阳健康科技集团有限公司 | Clinical path automatic generation method and system based on deep learning and knowledge graph |
| CN112216364A (en) * | 2020-10-10 | 2021-01-12 | 福建健康之路信息技术有限公司 | Medication prescription pushing method and system based on artificial intelligence |
| CN112100202B (en) * | 2020-11-12 | 2021-04-23 | 上海聚音信息科技有限公司 | Product identification and product information completion method, storage medium and robot |
| CN112435761A (en) * | 2020-12-04 | 2021-03-02 | 中国信息通信研究院 | Information recommendation method and device |
| CN112541351B (en) * | 2020-12-08 | 2023-09-05 | 济南市城乡建设数字化中心 | Method and system for dispatching government service heat line work orders in living construction field |
| CN112700838B (en) * | 2020-12-30 | 2024-03-01 | 平安科技（深圳）有限公司 | Big data-based medication scheme recommendation method and device and related equipment |
| CN112802573B (en) * | 2021-01-26 | 2023-06-20 | 中国科学技术大学 | Medicine package recommendation method, device, computer system and readable storage medium |
| CN112951362A (en) * | 2021-02-23 | 2021-06-11 | 上海商汤智能科技有限公司 | Medicine recommendation method, device, equipment and storage medium |
| CN112802575B (en) * | 2021-04-10 | 2021-09-03 | 浙江大学 | Medication decision support method, device, equipment and medium based on graphic state machine |
| CN113436746B (en) * | 2021-06-30 | 2024-04-12 | 平安科技（深圳）有限公司 | Medication recommendation method, device, equipment and storage medium based on sorting algorithm |
| CN113628716A (en) * | 2021-08-05 | 2021-11-09 | 翼健(上海)信息科技有限公司 | Prescription recommendation system |
| CN113628726B (en) * | 2021-08-10 | 2023-12-26 | 海南榕树家信息科技有限公司 | Traditional Chinese medicine diagnosis and treatment recommendation system and method based on graph neural network and electronic equipment |
| CN113434626B (en) * | 2021-08-27 | 2021-12-07 | 之江实验室 | Multi-center medical diagnosis knowledge map representation learning method and system |
| CN113707261A (en) * | 2021-08-30 | 2021-11-26 | 平安国际智慧城市科技股份有限公司 | Artificial intelligence-based medicine recommendation method and device and related equipment |
| CN113744867B (en) * | 2021-08-30 | 2024-03-08 | 平安科技（深圳）有限公司 | Medicine recommendation evidence-based support method, device, equipment and storage medium |
| CN113707262A (en) * | 2021-08-31 | 2021-11-26 | 平安医疗健康管理股份有限公司 | Medicine use recommendation method and device, computer equipment and storage medium |
| CN114464312B (en) * | 2022-01-04 | 2022-12-02 | 北京欧应信息技术有限公司 | System and storage medium for assisting disease reasoning |
| CN114283925A (en) * | 2022-01-17 | 2022-04-05 | 厦门大学 | Intelligent prescription recommendation method and storage medium |
| CN114386528B (en) * | 2022-01-18 | 2024-05-14 | 平安科技（深圳）有限公司 | Model training method and device, computer equipment and storage medium |
| CN114783571A (en) * | 2022-04-06 | 2022-07-22 | 北京交通大学 | Traditional Chinese medicine dynamic diagnosis and treatment scheme optimization method and system based on deep reinforcement learning |
| CN114582463A (en) * | 2022-05-05 | 2022-06-03 | 成都尚医信息科技有限公司 | Personalized motion guidance system and method based on machine learning |
| CN114722976A (en) * | 2022-06-09 | 2022-07-08 | 青岛美迪康数字工程有限公司 | Medicine recommendation system and construction method |
| CN115620886B (en) * | 2022-12-19 | 2023-04-28 | 北京大学第三医院（北京大学第三临床医学院） | Data auditing method and device |
| CN116486939B (en) * | 2022-12-26 | 2024-01-23 | 北京左医健康技术有限公司 | Data mining method and system for medicine knowledge graph and electronic equipment |
| CN115658877B (en) * | 2022-12-27 | 2023-03-21 | 神州医疗科技股份有限公司 | Medicine recommendation method and device based on reinforcement learning, electronic equipment and medium |
| CN115691741B (en) * | 2023-01-03 | 2023-03-14 | 四川大学华西医院 | Medical information-based information transmission and information combination method |
| CN116364240B (en) * | 2023-02-02 | 2024-01-26 | 复旦大学附属肿瘤医院 | Remote nutrition information processing method and system based on Internet |
| CN116110533B (en) * | 2023-02-27 | 2023-09-01 | 之江实验室 | Event map-based drug type and dosage recommendation system and method |
| CN116469527B (en) * | 2023-04-21 | 2023-09-08 | 脉景（杭州）健康管理有限公司 | Optimized recommendation method, system and equipment for traditional Chinese medicine prescription |
| CN116467471A (en) * | 2023-06-20 | 2023-07-21 | 山东志诚普惠健康科技有限公司 | Intelligent prescription filling method and device based on medical knowledge graph |
| CN116738064A (en) * | 2023-08-14 | 2023-09-12 | 江南大学附属医院 | Method and system for recommending common medicines for specific diseases and symptoms based on big data |
| CN117034202B (en) * | 2023-10-09 | 2024-01-05 | 北京医百科技有限公司 | Big data-based medication prompt method, system, equipment and storage medium |
| CN117252664A (en) * | 2023-11-10 | 2023-12-19 | 浙江口碑网络技术有限公司 | Medicine recommendation reason generation method, device, medium and equipment |
| CN117334352B (en) * | 2023-11-24 | 2024-03-08 | 北京邮电大学 | Hypertension diagnosis and treatment decision reasoning method and device based on multiple role knowledge graph |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170277841A1 (en) * | 2016-03-23 | 2017-09-28 | HealthPals, Inc. | Self-learning clinical intelligence system based on biological information and medical data metrics |
| CN108932975A (en) * | 2018-07-18 | 2018-12-04 | 中山大学 | A kind of recommender system based on mixing collaborative filtering doctors and patients information |
| CN109087691A (en) * | 2018-08-02 | 2018-12-25 | 科大智能机器人技术有限公司 | A kind of OTC drugs recommender system and recommended method based on deep learning |
| CN110223751A (en) * | 2019-05-16 | 2019-09-10 | 平安科技（深圳）有限公司 | Prescription evaluation method, system and computer equipment based on medical knowledge map |
-
2019
- 2019-12-27 CN CN201911373211.9A patent/CN111191020B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170277841A1 (en) * | 2016-03-23 | 2017-09-28 | HealthPals, Inc. | Self-learning clinical intelligence system based on biological information and medical data metrics |
| CN108932975A (en) * | 2018-07-18 | 2018-12-04 | 中山大学 | A kind of recommender system based on mixing collaborative filtering doctors and patients information |
| CN109087691A (en) * | 2018-08-02 | 2018-12-25 | 科大智能机器人技术有限公司 | A kind of OTC drugs recommender system and recommended method based on deep learning |
| CN110223751A (en) * | 2019-05-16 | 2019-09-10 | 平安科技（深圳）有限公司 | Prescription evaluation method, system and computer equipment based on medical knowledge map |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111191020A (en) | 2020-05-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111191020B (en) | Prescription recommendation method and system based on machine learning and knowledge graph | |
| WO2022041729A1 (en) | Medication recommendation method, apparatus and device, and storage medium | |
| Sahoo et al. | Intelligence-based health recommendation system using big data analytics | |
| KR102088980B1 (en) | System and Method for Providing personalized hospital information | |
| US11232365B2 (en) | Digital assistant platform | |
| CN109036545B (en) | Medical information processing method, apparatus, computer device and storage medium | |
| CN110291555B (en) | Systems and methods for facilitating computational analysis of health conditions | |
| El-Gayar et al. | Opportunities for business intelligence and big data analytics in evidence based medicine | |
| US11450434B2 (en) | Implementation of machine-learning based query construction and pattern identification through visualization in user interfaces | |
| US20210398660A1 (en) | Time-based resource allocation for long-term integrated health computer system | |
| US20210057106A1 (en) | System and Method for Digital Therapeutics Implementing a Digital Deep Layer Patient Profile | |
| CN115497631A (en) | Clinical scientific research big data analysis system | |
| CN111968715A (en) | Drug recommendation modeling method based on medical record data and drug interaction risk | |
| US10832822B1 (en) | Methods and systems for locating therapeutic remedies | |
| KR101295785B1 (en) | Apparatus and Method for Constructing Gene-Disease Relation Database | |
| Chen et al. | Using data mining strategies in clinical decision making: a literature review | |
| CN110473636B (en) | Intelligent medical advice recommendation method and system based on deep learning | |
| WO2023240837A1 (en) | Service package generation method, apparatus and device based on patient data, and storage medium | |
| CN114743647A (en) | Medical data processing method, device, equipment and storage medium | |
| Cui et al. | Intelligent recommendation for departments based on medical knowledge graph | |
| CN114722217A (en) | Content pushing method based on link prediction and collaborative filtering | |
| Nayak et al. | An Intelligent Disease Prediction and Drug Recommendation Prototype by Using Multiple Approaches of Machine Learning Algorithms | |
| Yee et al. | Big data: Its implications on healthcare and future steps | |
| Jamshidi et al. | A Hybrid Health Journey Recommender System Using Electronic Medical Records. | |
| Shrestha et al. | A Bayesian method for the automatic extraction of meaningful clinical sequences from large clinical databases |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |