CN117038026A - Recommendation method, electronic equipment and medium for hospital specialists - Google Patents
Recommendation method, electronic equipment and medium for hospital specialists Download PDFInfo
- Publication number
- CN117038026A CN117038026A CN202211480801.3A CN202211480801A CN117038026A CN 117038026 A CN117038026 A CN 117038026A CN 202211480801 A CN202211480801 A CN 202211480801A CN 117038026 A CN117038026 A CN 117038026A
- Authority
- CN
- China
- Prior art keywords
- disease
- patient
- model
- entity
- hospital
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 44
- 201000010099 disease Diseases 0.000 claims abstract description 120
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims abstract description 120
- 208000024891 symptom Diseases 0.000 claims abstract description 57
- 238000003745 diagnosis Methods 0.000 claims abstract description 18
- 238000013145 classification model Methods 0.000 claims abstract description 10
- 239000013598 vector Substances 0.000 claims abstract description 5
- 238000012163 sequencing technique Methods 0.000 claims abstract 3
- 238000011282 treatment Methods 0.000 claims description 24
- 238000000605 extraction Methods 0.000 claims description 22
- 238000012549 training Methods 0.000 claims description 22
- 239000003814 drug Substances 0.000 claims description 16
- 229940079593 drug Drugs 0.000 claims description 15
- 239000011159 matrix material Substances 0.000 claims description 15
- 238000012360 testing method Methods 0.000 claims description 11
- 238000010606 normalization Methods 0.000 claims description 10
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 6
- 240000005369 Alstonia scholaris Species 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 4
- 238000011277 treatment modality Methods 0.000 claims description 4
- 238000001647 drug administration Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000002483 medication Methods 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 claims description 2
- 238000010845 search algorithm Methods 0.000 claims description 2
- 230000011218 segmentation Effects 0.000 claims description 2
- 238000011269 treatment regimen Methods 0.000 claims description 2
- 238000013508 migration Methods 0.000 abstract description 2
- 230000005012 migration Effects 0.000 abstract description 2
- 238000011156 evaluation Methods 0.000 description 6
- 230000003993 interaction Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000001228 spectrum Methods 0.000 description 3
- 230000001419 dependent effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 206010020772 Hypertension Diseases 0.000 description 1
- 208000035965 Postoperative Complications Diseases 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000001684 chronic effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000012774 diagnostic algorithm Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 206010022000 influenza Diseases 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
Classifications
-
- 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
- G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
- G16H40/20—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the management or administration of healthcare resources or facilities, e.g. managing hospital staff or surgery rooms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/237—Lexical tools
- G06F40/247—Thesauruses; Synonyms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
- G06F40/295—Named entity recognition
-
- 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
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Public Health (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Epidemiology (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Primary Health Care (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Pathology (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The invention discloses a hospital expert recommending method, which comprises the steps of firstly obtaining standardized disease information according to patient complaint information, including disease parts, symptoms and disease time, then obtaining possible disease types and corresponding probability of a patient through a classification model, sequencing the possible disease types according to probability, respectively calculating the scores of all the experts in a hospital according to the disease types of N before ranking and combining expert-adept disease information of all the experts in the hospital, sequencing the experts according to the score, and finally recommending M experts before ranking to the patient, wherein N, M is a natural number. The recommendation party comprehensively gives recommendation specialists based on the multidimensional vector of the patient in the diagnosis and the bert model trained by the real data in the hospital, and the recommendation specialists can be used in a migration mode in different scenes without depending on building a corresponding database, so that a recommendation result with high accuracy can be obtained.
Description
Technical Field
The invention relates to the technical field of medical software systems, in particular to a recommendation method, electronic equipment and media of hospital specialists.
Background
With the development of medical technology and the improvement of the living standard of people, the demands of patients on medical accurate services are gradually improved. Patients have not met with a need to visit only common or wide-ranging outpatients, with an increasing need for expert outpatients, special needs, and special disease outpatients. However, the medical reserves of patients are of limited knowledge and knowledge of the areas of expertise of individual medical professionals is poor, which makes it often difficult to select an expert suitable for oneself. On the other hand, medical diagnosis and treatment technologies are rapidly developed and gradually move to be refined, medical knowledge is rapidly updated in iteration, and the expert is good at the field and tends to be accurate and specialized. How to efficiently match patients and experts and reasonably utilize resources by means of technology becomes a problem to be solved urgently.
To solve this problem, xie Rudan and the like propose a method and a system for accurate medical treatment, which input patient information into a machine learning model to infer the kind of a patient disease and a department to which the patient belongs, and match doctors in a doctor evaluation system constructed in advance. When the patient information is input, if the patient is diagnosed, the recommendation is made according to the diagnosis result of the diagnosis, and if the patient is not diagnosed, the disease type of the patient needs to be deduced first. The doctor recommendation strategy is to rank the doctors in the field of each disease, and recommend the doctors with the comprehensive ranks at the top. In this technical implementation, the recommended doctor is strongly dependent on the doctor's assessment hierarchy library. However, the obtaining, evaluating and assigning of the doctor evaluation system library is difficult to accurately, objectively and truly reflect the actual diagnosis and treatment level of the doctor in the actual operation. Specifically, indexes included in the doctor evaluation library, such as academic, average operation duration, postoperative complications, medical praise and the like, are multi-factor influence indexes, and cannot objectively reflect the professional level of the doctor. Taking the average duration of the operation as an example, the index is strongly related to the difficulty of the operation, the individual condition of the patient, the operation mode method and the like, and an evaluation library constructed based on the weight is out of balance.
Geng Shuang et al provide another method and apparatus for recommending doctors by extracting patient features and doctor features, constructing an adaptation function, further determining feature weights corresponding to the respective doctor features, and determining doctor information that best matches the patient information based on the feature weights. Specifically, the optimal weight combination is extracted, and then the feature similarity calculation results in the collaborative filtering algorithm are combined by using a weight method, and the doctor recommendation ordering results which are correspondingly matched are output. This approach is based on patient characteristics, so that the patient's past characteristics will have a greater impact, possibly affecting physician recommendations. For example, if a patient has a history of chronic hypertension and is repeatedly treated at an expert clinic in cardiology, the above features may affect other purposes, and if the patient is treated with influenza, the weight may be affected.
Tan Xingsheng an interactive-based intelligent client online reservation management system is provided, which is used for recommending consultation doctors by establishing a consultation interaction record database for consultation of past patients to the doctors and combining the consultation interaction satisfaction coefficient. The system is used for matching the basic information and the illness state type of the patient treated in the database according to the basic information and the illness state type of the client, judging the patient as an old user if the matching is successful, and recommending a designated doctor by combining the past consultation content and the patient satisfaction. Wherein, the patient satisfaction degree considers the interaction reply rate of doctors, the civilization expression degree and the like. If the data matching with the database is unsuccessful, judging as a new user, entering a new user matching flow, and matching the judging disease parts, the characteristics and the like. The method has the following limitation that firstly, the dependence on the disease database is high, and the recommendation of the old user is strongly dependent on the quantity and quality of database construction. Secondly, matching of past data is not carried out by adding data algorithm learning, accuracy is difficult to guarantee, and personalized treatment requirements of patients in treatment are not considered enough. Meanwhile, the inquiry and the visit should firstly consider the professional matching degree, and the weight of satisfaction evaluation is increased, so that professional proportion judgment is easily affected. The evaluation parameters of satisfaction degree are greatly influenced by the main view, and the accuracy degree is insufficient, such as the medical care interaction frequency and the recovery frequency, are strongly related to the degree of idleness and the disease difficulty degree of doctors, and cannot accurately reflect the professional level of the doctors.
Disclosure of Invention
Aiming at part or all of the problems in the prior art, the invention provides a recommendation method of hospital specialists, which comprises the following steps:
obtaining disease information of standardized description according to the complaint information of the patient, wherein the disease information comprises disease parts, symptoms and disease time;
based on the illness state information, the possible illness types and the corresponding probabilities of the patient are obtained through a classification model, and the possible illness types and the corresponding probabilities are ranked according to the probability;
according to the disease types of N ranked in advance, combining the disease information of each expert in the hospital, respectively calculating the scores of each expert in the hospital, and sorting according to the scores, wherein N is a natural number; and
the top M experts are recommended to the patient, where M is a natural number.
Further, obtaining the standardized description of the illness state information according to the complaint information of the patient comprises:
inputting the complaint information of the patient into the extraction model for decoding to obtain corresponding symptom words, and the disease parts and the disease time corresponding to the symptoms;
combining the symptom scholartree with the disease part, and inputting the combination into a normalization model to obtain a corresponding standard symptom scholartree; and
and inputting the disease parts into a normalization model to obtain corresponding standard part words.
Further, the extraction model adopts a beam search (beam search) algorithm to realize decoding.
Further, the extraction model is obtained according to the following steps:
the method comprises the steps of performing word segmentation on a dictation main complaint text of a patient in historical data, and marking symptoms, diseases, treatment modes, medication, body parts and existence conditions in the dictation main complaint text to obtain marked data; and
dividing the labeling data into a training subset and a testing subset according to a preset proportion, and training based on an UNL I M model to obtain an extraction model.
Further, the annotation data does not exceed 1000 pieces; and/or
The preset ratio is 8:2.
Further, the normalization model is obtained according to the following steps:
collecting a synonym table of a preset number of disease symptoms;
each standard word and each synonym in the synonym word list are inferred by using a PCLBert model to obtain corresponding word vectors; and
and constructing a corresponding unsupervised normalized transfer matrix by using a BERT-whiten i ng algorithm to obtain a normalized model.
Further, the classification model is obtained according to the following steps:
obtaining a training subset and a testing subset, comprising:
acquiring a specified number of outpatient case data, wherein each data comprises data I d, patient gender and age, patient complaints, current medical history, past history, diagnosis of disease, and I CD code for diagnosing disease;
according to the distribution of the diseases, sampling corresponding case texts of each diagnosis disease in the clinic medical record data of the preset quantity, and marking entities and relations including symptoms, diseases, treatment modes, medication, body parts and existence conditions, wherein each disease is marked with at least one case text, and the difference value between the integral disease probability distribution and the original distribution is not larger than a preset value;
and
dividing the annotation data into a training subset and a testing subset according to a preset proportion;
based on the Bert+ G l oba l Po i nter model, training an entity and relationship joint extraction model, wherein the entity is marked as an entity triplet through G l oba l Po i nter, the relationship is marked as two relationship triples, and the presence flag is one presence triplet, wherein each triplet comprises a type, a start position and an end position, and the training target comprises: the score of the positive example triplet is larger than 0, and the score of the negative example triplet is smaller than 0;
extracting model reasoning residual model data through the entity and relation combination to obtain symptoms, diseases, treatment modes, medication, body parts and existence conditions, wherein the method comprises the following steps:
obtaining a corresponding triplet score matrix according to the case text;
analyzing the entity part of the triplet score matrix, and selecting entity triples with scores greater than 0 as entity extraction results;
selecting a symptom, a starting position and an ending position of a disease entity, searching according to the existence condition part of the triplet scoring matrix, and if the score of the existence condition triplet is greater than 0, the entity exists, otherwise, the entity does not exist;
selecting a starting position and an ending position of a symptom and a part entity, searching according to a relation part of a triplet score matrix, and if the scores of two relation triples are larger than 0, the relation exists, otherwise, the relation does not exist; and
mapping all obtained position information back to the position in the original text;
combining the symptoms, diseases, treatment regimens, medications, body parts, and presence into text in the following format: patient age + patient gender + symptoms + disease + treatment modality, medication to be used as input for diagnosing the Bert model; and
based on the input, a classification model is trained with the diagnostic disease with the I CD code in the case as output.
Further, the acquiring of the text includes:
sampling a preset number of symptom words and disease words according to poisson distribution with the mean value of 2;
combining the body part corresponding to the symptom word and the existence condition into a text format of existence condition + body part + symptom word, and combining the existence condition corresponding to the disease word into a text format of existence condition + disease word;
the rest treatment modes and the drug administration entities are combined into corresponding texts in sequence;
dividing the ages of patients into neonates, infants, patients and elderly patients, and converting the ages into texts;
patient gender was converted to text: male and female; and
combining the text into a finally input text of the model according to a preset format: patient age, patient gender, symptoms, disease, mode of treatment.
Based on the hospital expert recommendation method as described above, a second aspect of the invention provides an electronic device for recommending hospital experts, comprising a memory and a processor, wherein the memory is configured to store a computer program which, when run by the processor, performs the hospital expert recommendation method as described above.
The third aspect of the present invention also provides a computer readable storage medium for recommending hospital specialists, storing a computer program which, when run on a processor, performs a hospital specialist recommending method as described above.
According to the recommendation method of the hospital specialist, the recommendation specialist is comprehensively given by the algorithm based on the multidimensional vector of the patient's present visit, and the recommendation specialist can be used in a migration mode in different scenes without depending on building a corresponding database. Specifically, the recommendation method provides personalized and accurate expert recommendation service according to the complaints, the main symptoms and the like of each patient visit, current expert recommendation cannot be influenced by historical visit information, in addition, the recommendation method takes the department matching degree, expert proficiency professional fields and the like as learning training materials based on a bert model trained by real data in a hospital, the data is objective and reliable, and further, a recommendation result with high accuracy can be obtained. In short, the technical effect of the invention is that, on the first hand, the department and expert with the highest degree of matching with the complaints or symptoms are provided for the user, namely, the recommendation precision of the hospital expert is improved, on the other hand, the method does not depend on the patient database, so compared with the recommendation method which needs to carry out matching searching in a larger or larger patient database, the operation time can be greatly shortened, and the occupation of computer hardware is reduced.
It should be noted that the present invention does not relate to a method of diagnosis and treatment of a disease, but merely provides information related to medical treatment, belonging to an intelligent system, i.e. the present invention is neither intended to identify a disease of a patient nor to provide a certain parameter or index for diagnosing a disease nor a disease prescreening method. In contrast, the information provided by the solution of the present invention cannot be used for diagnosis and treatment of diseases, but the corresponding diagnosis and treatment should be provided to the user by the hospital/doctor.
Drawings
To further clarify the above and other advantages and features of embodiments of the present invention, a more particular description of embodiments of the invention will be rendered by reference to the appended drawings. It is appreciated that these drawings depict only typical embodiments of the invention and are therefore not to be considered limiting of its scope. In the drawings, for clarity, the same or corresponding parts will be designated by the same or similar reference numerals.
Fig. 1 shows a flow diagram of a method of recommending hospital specialists according to an embodiment of the present invention.
Detailed Description
In the following description, the present invention is described with reference to various embodiments. One skilled in the relevant art will recognize, however, that the embodiments may be practiced without one or more of the specific details, or with other alternative and/or additional methods or components. In other instances, well-known structures or operations are not shown or described in detail to avoid obscuring aspects of the invention. Similarly, for purposes of explanation, specific numbers and configurations are set forth in order to provide a thorough understanding of embodiments of the present invention. However, the invention is not limited to these specific details.
Reference throughout this specification to "one embodiment" or "the embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment.
It should be noted that the embodiments of the present invention describe the steps of the method in a specific order, however, this is merely for the purpose of illustrating the specific embodiments, and not for limiting the order of the steps. In contrast, in different embodiments of the present invention, the sequence of each step may be adjusted according to the adjustment of the actual requirement.
The BERT (B i d i rect i ona l Encoder Representat i on from Transformers) model is a pre-trained language characterization model. The model is not pre-trained by adopting a traditional unidirectional language model or a method for shallow splicing of two unidirectional language models, but adopts a new MLM (masked l anguage mode l) model to generate a deep bidirectional language representation. Based on the excellent performance of the BERT model in the aspects of information learning and processing, the invention provides a method for fully integrating rich semantic meaning and information into an expert model by adopting the BERT model and fully combining department diagnosis guiding and recommending information so as to effectively improve the accuracy of hospital expert recommendation.
The embodiments of the present invention will be further described with reference to the drawings.
Fig. 1 shows a flow diagram of a method of recommending hospital specialists according to an embodiment of the present invention. As shown in the figure, a recommendation method of a hospital expert includes:
first, in step 101, standardized condition information is acquired. And obtaining the disease information of standardized description according to the complaint information of the patient, wherein the disease information comprises disease parts, symptoms and disease time. In one embodiment of the present invention, the information of the disease location, symptom, disease time, etc. of the patient is obtained based on the complaint information of the patient mainly through the word splitting and normalization algorithm, specifically including:
inputting the complaint information of the patient into the extraction model for decoding to obtain the corresponding symptom words, and the corresponding disease parts and disease time of the symptoms. In one embodiment of the invention, the extraction model is decoded using a beam search algorithm, the dictionary tree is limited, and corresponding symptom words, and the disease sites, disease times, etc. corresponding to the symptoms are generated. In a further embodiment of the invention, the extraction model is obtained or trained
The training comprises the following steps:
acquiring and reading a large number of patient dictation main complaint texts, namely, splitting words of dictation main complaint texts of a specified number of patients in historical data, marking entities and relations of symptoms, diseases, treatment modes, medicines, body parts, existence conditions and the like in the dictation main complaint texts, and obtaining marking data, wherein in one embodiment of the invention, the total amount of the marking data is limited to about 1 kilo; and
injecting the labeling data into a training subset and a testing subset according to a preset proportion, dividing the labeling data into a training subset and a testing subset, and training based on an UNL I M model to obtain an extraction model; and
and combining the disease-causing part and the disease-causing part, inputting the combined disease-causing part and the disease-causing part into a normalization model to obtain a corresponding standard disease-causing part, and independently inputting the disease-causing part into the normalization model to obtain a corresponding standard part word. In one embodiment of the invention, the normalization
The model is obtained according to the following steps:
collecting a synonym table of a preset number of disease symptoms;
each standard word and each synonym in the synonym word list are inferred by using a PCLBert model to obtain corresponding word vectors; and
constructing a corresponding unsupervised normalized transfer matrix by using a BERT-wh iten i ng algorithm to obtain a normalized model;
next, at step 102, the likely disease type and probability are obtained. Based on the illness state information, the possible illness types and the corresponding probabilities of the patient are obtained through a classification model, and the possible illness types and the corresponding probabilities are ranked according to the probability. In one embodiment of the invention, the classification model is trained according to the following:
obtaining a training subset and a testing subset, comprising:
acquiring a large number of outpatient cases, in one embodiment of the present invention, the number of outpatient cases is about 100 ten thousand, wherein each data includes data I d, patient basic information, such as gender, age, etc., patient complaints, current medical history, past history, diagnosis of disease, I CD code for diagnosing disease, etc.;
according to the distribution of the diseases, sampling the corresponding case text of each diagnosis disease in the outpatient medical record data, marking the entities and relations of symptoms, diseases, treatment modes, medicines, body parts, existence conditions and the like, wherein each disease is marked with at least one case text, the difference between the integral disease probability distribution and the original distribution is not more than a preset value, namely, the integral disease probability distribution and the original distribution are not greatly ensured,
in one embodiment of the invention, the total amount of annotation data is limited to about 1 ten thousand;
and
dividing the annotation data into a training subset and a testing subset according to a preset proportion, such as 8:2;
based on the Bert+ G l oba l Po i nter model, training to obtain an entity and relation joint extraction model based on the training subset and the testing subset, wherein the entity is marked as an entity triplet through G l oba lPo i nter: (entity type, start position, end position); the relationship is marked as two relationship triples: (relationship type, entity 1 start position, entity 2 start position) (relationship type, entity 1 end position, entity 2 end position); presence flag bit one presence triplet: (where present, start position, end position), the entity and relationship joint extraction model training targets include: the score of the positive example triplet is larger than 0, and the score of the negative example triplet is smaller than 0;
the residual model data is inferred through the entity and relation joint extraction model, and the entities and relations of symptoms, diseases, treatment modes, medication, body parts, existence and the like in the model data are obtained, wherein the method comprises the following steps:
inputting a case text, and obtaining a corresponding triplet score matrix through the entity and relation joint extraction model;
analyzing the entity part of the triplet score matrix, and selecting entity triples (entity types, starting positions and ending positions) with scores greater than 0 as entity extraction results;
selecting a symptom, a starting position and an ending position of a disease entity, searching according to the existence condition part of the triplet scoring matrix, and if the score of the existence condition triplet (existence condition, starting position and ending position) is more than 0, the entity exists, otherwise, the entity does not exist;
selecting the starting position and the ending position of the symptom, the part entity, searching according to the relation part of the triplet score matrix, if the scores of the two relation triples (symptom-part relation, entity 1 starting position and entity 2 starting position) and (symptom-part relation, entity 1 ending position and entity 2 ending position) are both more than 0, the relation exists, and otherwise, the relation does not exist; and
mapping all obtained position information back to the position in the original text;
and combining the entities and the relation sets of symptoms, diseases, treatment modes, medication, body parts, existence and the like obtained in the steps into the input of the diagnosis Bert model according to a certain rule. In one embodiment of the invention, the diagnostic Bert model is input in the format of: patient age + patient gender + symptoms + disease + treatment modality, medication to be used as input to a diagnostic Bert model, the acquisition of the formatted text includes:
sampling a preset number of symptom words and disease words according to poisson distribution with the mean value of 2;
combining the body part corresponding to the symptom word and the existence condition into a text format of existence condition + body part + symptom word, and combining the existence condition corresponding to the disease word into a text format of existence condition + disease word;
the rest treatment modes and the drug administration entities are combined into corresponding texts in sequence;
patient ages are divided into neonates, infants, patients, elderly patients and converted to text, in one embodiment of the invention, 5. Patient ages are divided into 48 days, 14 years, 60 years and converted to text: neonates, infants, patients, elderly patients;
patient gender was converted to text: male and female; and
combining the text into a finally input text of the model according to a preset format: age of patient +
Sex + symptom + disease + mode of treatment of patient, use medicine. And
taking the text in the format as the input, taking the diagnosis diseases with the I CD codes in the cases as the output, and training to obtain a classification model;
next, at step 103, an expert score is calculated. According to the disease types of N in the ranking, the scores of all the experts in the hospital are calculated respectively in combination with the disease information of the experts in the hospital, and are ranked according to the score, wherein N is a natural number, and in one embodiment of the invention, the value of N is preferably 5. In one embodiment of the invention, hospital incumbent specialist name, job number, department and disease-adequacy information may be collected, and each specialist's disease-adequacy spectrum is established in accordance with the I CD-10 standard for matching integration. In one embodiment of the invention, expert score computation includes:
confirming from each expert's disease-adept spectrum whether the top-ranked N disease category is included, adding the included diagnostic algorithm probabilities to each expert to calculate a score, specifically, extracting each expert's disease-adept set in the expert-adept spectrum, and then traversing and summing the corresponding disease probabilities in the probability distribution of the disease in which the individual expert is adept at the output in step 102 as the expert score, i.e., the probability that it is selected; and
finally, at step 104, the expert is recommended. The top M experts are recommended to the patient, where M is a natural number. In one embodiment of the present invention, the value of N is preferably 3.
Based on the hospital expert recommendation method as described above, the invention further provides an electronic device for recommending hospital experts, comprising a memory and a processor, wherein the memory is configured to store a computer program which, when run by the processor, performs the hospital expert recommendation method as described above.
Furthermore, the present invention provides a computer readable storage medium for recommending hospital specialists, which stores a computer program which, when run on a processor, performs the hospital specialist recommending method as described above.
While various embodiments of the present invention have been described above, it should be understood that they have been presented by way of example only, and not limitation. It will be apparent to those skilled in the relevant art that various combinations, modifications, and variations can be made therein without departing from the spirit and scope of the invention. Thus, the breadth and scope of the present invention as disclosed herein should not be limited by any of the above-described exemplary embodiments, but should be defined only in accordance with the following claims and their equivalents.
Claims (10)
1. A method of recommending hospital specialists, comprising the steps of:
obtaining disease information of standardized description according to the complaint information of the patient, wherein the disease information comprises disease parts, symptoms and disease time;
based on the illness state information, the possible illness types and the corresponding probabilities of the patient are obtained through a classification model, and the possible illness types are ordered according to the probability;
according to the disease types of N ranked in advance, combining the disease information of each expert in the hospital, respectively calculating the scores of each expert in the hospital, and sequencing the experts according to the scores, wherein N is a natural number; and
the top M experts are recommended to the patient, where M is a natural number.
2. The recommendation method of claim 1, wherein obtaining the standardized descriptive patient condition information based on patient complaint information comprises:
inputting the complaint information of the patient into the extraction model for decoding to obtain corresponding symptom words, and the disease parts and the disease time corresponding to the symptoms;
combining the symptom scholartree with the disease part, and inputting the combination into a normalization model to obtain a corresponding standard symptom scholartree; and
and inputting the disease parts into a normalization model to obtain corresponding standard part words.
3. The recommendation method of claim 2, wherein the extraction model employs a bundle search algorithm to effect decoding.
4. The recommendation method according to claim 2, wherein said extraction model is obtained according to the steps of:
the method comprises the steps of performing word segmentation on a dictation main complaint text of a patient in historical data, and marking symptoms, diseases, treatment modes, medication, body parts and existence conditions in the dictation main complaint text to obtain marked data; and
dividing the labeling data into a training subset and a testing subset according to a preset proportion, and training based on an UNLIM model to obtain an extraction model.
5. The recommendation method of claim 4, wherein the annotation data does not exceed 1000 pieces; and/or
The preset ratio is 8:2.
6. The recommendation method according to claim 2, wherein said normalization model is obtained according to the steps of:
collecting a synonym table of a preset number of disease symptoms;
each standard word and each synonym in the synonym word list are inferred by using a PCLBert model to obtain corresponding word vectors; and
and constructing a corresponding unsupervised normalized transfer matrix by using a BERT-whistening algorithm to obtain a normalized model.
7. The recommendation method according to claim 1, wherein said classification model is obtained according to the steps of:
obtaining a training subset and a testing subset, comprising:
acquiring a specified number of outpatient case data, wherein each data comprises data i d, patient gender and age, patient complaints, current medical history, past history, diagnosis of disease, ICD encoding of diagnosis of disease;
according to the distribution of the diseases, sampling corresponding case texts of each diagnosis disease in the clinic medical record data of the preset quantity, and marking entities and relations, wherein the entities and relations comprise symptoms, diseases, treatment modes, medication, body parts and existence conditions, each disease is marked with at least one case text, and the difference value between the integral disease probability distribution and the original distribution is not larger than a preset value; and
dividing the annotation data into a training subset and a testing subset according to a preset proportion;
training an entity and relation joint extraction model based on a Bert+Global Pointer model, so that the score of a positive case triplet is greater than 0, and the score of a negative case triplet is less than 0, wherein the entity is marked as an entity triplet through the Global Pointer, the relation is marked as two relation triples, and the existence condition mark is located as an existence condition triplet, wherein each triplet comprises a type, a starting position and an ending position;
deriving, by the entity and relationship joint extraction model, symptoms, diseases, treatment modalities, medications, body parts, and presence based on the remaining-based outpatient medical record data, including:
obtaining a corresponding triplet score matrix according to the corresponding case text;
analyzing the entity part of the triplet score matrix, and selecting entity triples with scores greater than 0 as entity extraction results;
selecting a symptom, a starting position and an ending position of a disease entity, searching according to the existence condition part of the triplet scoring matrix, and if the score of the existence condition triplet is greater than 0, the entity exists, otherwise, the entity does not exist;
selecting a starting position and an ending position of a symptom and a part entity, searching according to a relation part of a triplet score matrix, and if the scores of two relation triples are larger than 0, the relation exists, otherwise, the relation does not exist; and
mapping all obtained position information back to the position in the original text;
combining the symptoms, diseases, treatment regimens, medications, body parts, and presence into text in the following format: patient age + patient gender + symptoms + disease + treatment modality, medication to be used as input for diagnosing the Bert model; and
based on the input, a classification model is trained with the diagnostic disease with the I CD code in the case as output.
8. The recommendation method of claim 7, wherein the obtaining of text comprises:
sampling a preset number of symptom words and disease words according to poisson distribution with the mean value of 2;
combining the body part corresponding to the symptom word and the existence condition into a text format of existence condition + body part + symptom word, and combining the existence condition corresponding to the disease word into a text format of existence condition + disease word;
the rest treatment modes and the drug administration entities are combined into corresponding texts in sequence;
dividing the ages of patients into neonates, infants, patients and elderly patients, and converting the ages into texts;
patient gender was converted to text: male and female; and
combining the text into a finally input text of the model according to a preset format: patient age, patient gender, symptoms, disease, mode of treatment.
9. An electronic device for recommending hospital specialists, characterized by comprising a memory and a processor, wherein the memory is configured to store a computer program which, when run by the processor, performs the recommending method of a hospital specialist according to any of claims 1 to 8.
10. A computer readable storage medium for recommending hospital specialists, characterized in that a computer program is stored, which computer program, when run on a processor, performs the hospital specialist recommending method according to any of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211480801.3A CN117038026A (en) | 2022-11-24 | 2022-11-24 | Recommendation method, electronic equipment and medium for hospital specialists |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211480801.3A CN117038026A (en) | 2022-11-24 | 2022-11-24 | Recommendation method, electronic equipment and medium for hospital specialists |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117038026A true CN117038026A (en) | 2023-11-10 |
Family
ID=88625001
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211480801.3A Pending CN117038026A (en) | 2022-11-24 | 2022-11-24 | Recommendation method, electronic equipment and medium for hospital specialists |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117038026A (en) |
-
2022
- 2022-11-24 CN CN202211480801.3A patent/CN117038026A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11928570B2 (en) | Artificial intelligence methods and systems for generation and implementation of alimentary instruction sets | |
| US11289198B2 (en) | Systems and methods for generating alimentary instruction sets based on vibrant constitutional guidance | |
| US10553319B1 (en) | Artificial intelligence systems and methods for vibrant constitutional guidance | |
| WO2020123723A1 (en) | System and method for providing health information | |
| US11275985B2 (en) | Artificial intelligence advisory systems and methods for providing health guidance | |
| CN111897967A (en) | Medical inquiry recommendation method based on knowledge graph and social media | |
| US10593431B1 (en) | Methods and systems for causative chaining of prognostic label classifications | |
| US11468363B2 (en) | Methods and systems for classification to prognostic labels using expert inputs | |
| EP3298518A1 (en) | Computer implemented method for determining clinical trial suitability or relevance | |
| US11568351B2 (en) | Systems and methods for arranging transport of alimentary components | |
| WO2023178971A1 (en) | Internet registration method, apparatus and device for seeking medical advice, and storage medium | |
| US11157822B2 (en) | Methods and systems for classification using expert data | |
| US10936962B1 (en) | Methods and systems for confirming an advisory interaction with an artificial intelligence platform | |
| US20200320414A1 (en) | Artificial intelligence advisory systems and methods for vibrant constitutional guidance | |
| US11392854B2 (en) | Systems and methods for implementing generated alimentary instruction sets based on vibrant constitutional guidance | |
| US11915827B2 (en) | Methods and systems for classification to prognostic labels | |
| US20200321115A1 (en) | Systems and methods for generating alimentary instruction sets based on vibrant constitutional guidance | |
| WO2022227203A1 (en) | Triage method, apparatus and device based on dialogue representation, and storage medium | |
| Liao et al. | Medical data inquiry using a question answering model | |
| CN114639489B (en) | Mutual learning-based inquiry quick reply recommendation method and device and electronic equipment | |
| US20210133627A1 (en) | Methods and systems for confirming an advisory interaction with an artificial intelligence platform | |
| CN114743647A (en) | Medical data processing method, device, equipment and storage medium | |
| US11222727B2 (en) | Systems and methods for generating alimentary instruction sets based on vibrant constitutional guidance | |
| CN117038026A (en) | Recommendation method, electronic equipment and medium for hospital specialists | |
| CN110289065A (en) | A kind of auxiliary generates the control method and device of medical electronic report |
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 |