Medical imaging diagnosis and inspection guidance system, method and equipment based on artificial intelligence

Through the medical imaging guidance system based on artificial intelligence, combined with AI question-and-answer system and knowledge graph, the gap in the existing technology in imaging applications has been solved, and intelligent auxiliary recommendations for medical imaging professional examination projects and medical treatment plans have been realized. , improves medical service efficiency and patient experience.

CN112786194BActive Publication Date: 2024-11-01BEIJING YIMAI SUNSHINE MEDICAL INFORMATION TECH CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202110121696.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-01-28
Publication Date
2024-11-01
Estimated Expiration
2041-01-28

AI Technical Summary

Technical Problem

The existing intelligent diagnosis guidance system is incomplete in terms of the subdivision of the service objects, especially in imaging applications, which leads to confusion in patients' choice of departments and low medical services.

Method used

Adopt medical imaging guidance and inspection system based on artificial intelligence, combined with AI question and answer system, knowledge graph, risk prediction and recommendation system, and use machine learning through consultation module, case generation module, data resource module, matching module and guidance module and deep learning technology, assisting patient self-examination and doctor diagnosis, and providing professional and standardized examination programs and medical treatment plans.

Benefits of technology

It has realized intelligent auxiliary recommendations for medical imaging professional examination projects and medical treatment plans, improved medical service efficiency and patient experience, helped patients to conduct self-examination and doctor diagnosis, and filled the gaps in imaging applications.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112786194B_ABST
    Figure CN112786194B_ABST
Patent Text Reader

Abstract

The medical imaging guidance and inspection system based on artificial intelligence disclosed in the present invention includes: a data resource module for creating and integrating medical imaging data resources such as chronic disease knowledge graphs, imaging graphs, and imaging professional diagnostic materials; a matching module for building a matching model, inputting case information into the matching model for matching, and obtaining a multi-channel fusion rough screening and recall result set; a guidance module for extracting patient business-related features from the rough screening and recall result set as data corpus, sending it to the refined sorting model for training and screening, and obtaining scientific guidance, knowledge, risk prediction and other result sets; a guidance and inspection module for extracting doctor business-related features from the rough screening and recall result set as data corpus, sending it to the refined sorting model for training and screening, and obtaining professional diagnosis and treatment plans, diagnosis reminders and other guidance and inspection service result sets. It can realize AI-assisted recommendation and question-and-answer consultation for medical imaging examination items, treatment plans and services, and assist patients in self-examination and doctor diagnosis.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of smart medical technology, and in particular to a medical imaging diagnosis and inspection guidance system, method, equipment and medium based on artificial intelligence. Background Art

[0002] During the medical consultation process, many patients often encounter confusion in choosing a department; medical knowledge is too professional, the clinical manifestations of diseases are complex, and there are a large number of diseases with similar symptoms, which causes patients to be confused or even make mistakes when choosing a department; currently, manual guidance is not only time-consuming and labor-intensive, but also greatly affects the efficiency of medical services and the patient's medical experience. Existing intelligent guidance, inspection guidance, and case structured products have comprehensive coverage in terms of technical functions and service applications, but the field segmentation of service objects is still incomplete, and imaging applications are vacant. Summary of the invention

[0003] In response to the defects in the prior art, the embodiments of the present invention provide a medical imaging diagnosis and inspection guidance system, method, device and medium based on artificial intelligence, which specifically involve the application of several major artificial intelligence technology fields such as AI question-answering system, knowledge graph, risk prediction and recommendation system. Artificial intelligence technology is used to provide auxiliary recommendations for medical imaging specialties, standardized examination items, treatment plans and services, to assist patients in self-examination and doctors in diagnosis.

[0004] In a first aspect, an embodiment of the present invention provides a medical imaging diagnosis and inspection guidance system based on artificial intelligence, comprising: a diagnosis inquiry module, a case generation module, a data resource module, a matching module, a diagnosis guidance module and an inspection guidance module.

[0005] The consultation module is used to collect basic patient information, chief complaint information, and AI consultation communication information;

[0006] The case generation module is used to generate structured cases from patient information and push the cases to the patient end and the doctor end respectively;

[0007] The data resource module is used to create and integrate chronic disease knowledge graphs, imaging examination graphs, and imaging professional diagnostic materials and medical imaging data resources;

[0008] The matching module is used to construct a matching model based on the integrated matching algorithm, matching rules, and knowledge graph retrieval rules, extract features from medical image data resources, create an image data corpus, train the matching model using a machine learning method, input case information into the matching model for matching, and obtain a multi-channel fusion rough screening recall result set;

[0009] The guidance module is used to extract features associated with the patient's business from the multi-channel fusion rough screening recall result set, and send it to the first fine sorting model as a supplementary corpus for training, sorting and screening to obtain guidance information, disease knowledge and risk prediction result sets, and send recommendation results to the patient end;

[0010] The inspection guidance module is used to extract features related to the doctor's business from the multi-channel fusion rough screening recall result set, and send it to the second fine sorting model as a supplementary data corpus for training, sorting and screening to obtain the diagnosis and treatment plan and diagnosis reminder service result set, and send the recommendation results to the doctor end.

[0011] In a second aspect, an embodiment of the present invention provides a medical imaging diagnosis and inspection guidance method based on artificial intelligence, comprising the following steps:

[0012] Obtain the patient's basic information, chief complaint information, and AI consultation communication information;

[0013] Generate structured cases based on patient information and push them to the patient and doctor ends respectively;

[0014] Create and integrate chronic disease knowledge graphs, imaging examination graphs, and medical imaging data resource libraries for imaging professional diagnostic materials;

[0015] A matching model is constructed based on the integrated matching algorithm, matching rules, and knowledge graph retrieval rules. Features are extracted from the medical image data resource library to create an image data corpus. The matching model is trained using machine learning methods. Case information is input into the matching model for matching to obtain a multi-channel fusion rough screening recall result set.

[0016] Extract features related to the patient's business from the multi-channel fusion rough screening recall result set and send it to the first fine sorting model as supplementary corpus for training, sorting and screening to obtain the guidance information, disease knowledge and risk prediction result set, and send the recommendation results to the patient end;

[0017] The features related to the doctor's business are extracted from the multi-channel fusion rough screening recall result set and sent to the second fine sorting model as a supplementary data corpus for training, sorting and screening to obtain the diagnosis and treatment plan and diagnosis reminder service result set, and send the recommendation results to the doctor.

[0018] In a third aspect, an embodiment of the present invention provides an intelligent device, comprising a processor, an input device, an output device and a memory, wherein the processor, input device, output device and memory are interconnected, the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method described in the above embodiment.

[0019] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor executes the method described in the above embodiment.

[0020] Beneficial effects of the present invention:

[0021] The embodiments of the present invention provide a medical imaging diagnosis and inspection guidance system, method, device and medium based on artificial intelligence. Specifically, the system integrates computer artificial intelligence natural language processing technology solutions including machine learning, deep learning, entity recognition, similarity matching, intent recognition, knowledge graphs, recommendation systems, question-and-answer systems, risk prediction, etc., and processes, trains and machines industry data including professional, large-scale and systematic real corpus data of professional imaging reports and professional academic materials on imaging examinations. It can realize artificial intelligence technology-assisted recommendation and question-and-answer consultation for professional and standardized examination items, treatment plans and services for medical imaging, and assist patients in self-examination and doctors in diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the description of the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0023] Figure 1 A structural block diagram of a medical imaging diagnosis and inspection guidance system based on artificial intelligence provided by a first embodiment of the present invention is shown;

[0024] Figure 2 A flowchart of a medical imaging diagnosis and inspection guidance method based on artificial intelligence provided by a second embodiment of the present invention is shown;

[0025] Figure 3 A structural block diagram of an intelligent device provided by the third embodiment of the present invention is shown. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0028] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0029] It should be further understood that the term "and / or" used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0030] As used in this specification and the appended claims, the term "if" may be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if [described condition or event] is detected" may be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0031] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should have the common meanings understood by those skilled in the art to which the present invention belongs.

[0032] like Figure 1 As shown, a structural block diagram of a medical imaging diagnosis and inspection guidance system based on artificial intelligence provided by the first embodiment of the present invention is shown, the system includes: a diagnosis inquiry module, a case generation module, a data resource module, a matching module, a diagnosis guidance module and an inspection guidance module, wherein:

[0033] The consultation module is used to collect basic patient information, chief complaint information, and AI consultation communication information;

[0034] The case generation module is used to generate structured cases from patient information and push the cases to the patient end and the doctor end respectively;

[0035] The data resource module is used to create and integrate medical imaging data resources such as chronic disease knowledge graphs, imaging examination graphs, and imaging professional diagnostic materials;

[0036] The matching module is used to construct a matching model based on the integrated matching algorithm, matching rules, and knowledge graph retrieval rules, extract features from the medical image data resource library, create an image data corpus, use machine learning methods to train the matching model, input case information into the matching model for matching, and obtain a multi-channel fusion rough screening recall (recommendation) result set;

[0037] The guidance module is used to extract features associated with the patient's business from the multi-channel fusion rough screening recall result set, and send it to the first fine sorting model as a supplementary corpus for training, sorting, and screening to obtain a result set of popular science guidance information, disease knowledge, risk prediction, etc., and send the recommendation results to the patient end;

[0038] The inspection guidance module is used to extract features related to the doctor's business from the multi-channel fusion rough screening recall result set, and send it to the second fine sorting model as a supplementary data corpus for training, sorting, and screening to obtain service result sets such as professional treatment plans and diagnosis reminders, and send recommendation results to the doctor side.

[0039] In this embodiment, the consultation module includes a patient self-report unit, a doctor recommendation unit, a big data question and answer retrieval unit, an AI customer service question and answer system unit, a doctor-patient communication unit, a recommended expert question and answer unit, a patient information recording unit and an image data corpus unit; the patient self-report unit is used to collect and obtain user personal information, including name, age, gender, region, symptom description, symptom attribute description, including onset site, degree, frequency, etc., medical history, allergy history and other one-complaint and five-history information; the doctor recommendation unit is used to recommend related doctors based on the patient's self-report information to facilitate doctor-patient communication; the big data question and answer retrieval unit is used to retrieve the mainstream medical question and answer information on the Internet. The AI ​​customer service question and answer system unit is used for patients to communicate with AI customer service online, ask questions and get answers; the patient information recording unit is used to record patient self-report information, patient chat information with AI customer service, doctor-patient communication information, and expert question and answer information, and store them in the database; the image data corpus unit is used to retrieve patient self-report information, patient chat information with AI customer service, doctor-patient communication information, and expert question and answer information from the database, analyze and enter them into the image data corpus, which is returned to the AI ​​customer service question and answer system unit for analysis and organization of dialogue. The data is recycled and used synchronously for subsequent matching module training and learning.

[0040] The AI ​​customer service question-and-answer system unit includes an entity recognition unit, an intent recognition unit, and a graph retrieval unit. The entity recognition unit is used to perform strict search matching and similarity approximate matching entity recognition on the questions input by the patient, that is, synonym replacement, to generate standard expression questions; the intent recognition unit is used to send the patient's standard expression questions after matching or synonym replacement processing into the machine learning model training classification to perform question classification, such as: sentences containing keywords such as "how to treat" and "what medicine to take" tend to be questions about consulting treatment methods, and sentences containing keywords such as "how long" and "recovery" tend to be questions about consulting treatment cycles; the graph retrieval unit is used to guide patient questions according to question classification to answer templates, such as: treatment methods The answer template for method-related questions is "The treatment methods for disease {} are {}", and the answer template for questions about treatment cycles is "The cure period for disease {} is {}". Keywords are searched for data resources such as chronic disease knowledge graphs and imaging knowledge graphs to fill in the template. For example, for the keyword "otitis media", the treatment method is retrieved as "using ear drops phenol glycerin and oral anti-inflammatory drugs", and the cure period is retrieved as "10-14 days". The complete answer is returned to the patient, "The treatment methods for otitis media are: using ear drops phenol glycerin and oral anti-inflammatory drugs", and "The cure period for otitis media is 10-14 days".

[0041] The case generation module includes an information acquisition unit and a case generation unit. The information acquisition module is used to read data such as patient's chief complaint information, and the case generation unit is used to generate structured case result data from patient information.

[0042] The data resource module includes a chronic disease knowledge graph unit, an imaging examination graph unit, an imaging report unit, a big data expert question and answer unit and a self-expanding knowledge graph unit. The chronic disease knowledge graph unit is used to generate graph data of entity and relationship pairs from relational chronic disease knowledge data such as disease names, introductions, causes, symptoms, treatments, prevention, and daily life, and import them into the chronic disease knowledge graph database; the imaging examination graph unit is used to generate graph data of entity and relationship pairs from relational imaging professional data such as symptoms, examinations, indications, treatment plans, equipment, scanning methods, and examination body positions, and import them into the imaging examination knowledge graph database; the imaging report unit is used to collect and obtain the original imaging report data knowledge base; the big data expert question and answer unit is used to collect Internet medical consultation expert question and answer data; the self-expanding knowledge graph unit is used to extract entity and relationship pairs from patient cases, reports, and Internet expert answers, and automatically create an extended knowledge graph in real time.

[0043] The matching module includes a data processing unit, a graph retrieval unit, a neural network model unit, a similarity feature extraction unit, a recommendation model unit, and a recall result set unit.

[0044] Among them, the data processing unit is used to obtain data from chronic disease knowledge graphs, imaging examination knowledge graphs, imaging report knowledge bases, big data expert Q&A, and self-expanding knowledge graphs, analyze and process data, extract features, represent them with mathematical vectorized symbols, and create an imaging data corpus for machine learning training intent and matching.

[0045] The graph retrieval unit is used to retrieve the chronic disease management knowledge graph, imaging examination knowledge graph, and self-expanding knowledge graph based on patient case data, find the associated matching result entity relationship pairs, and obtain the graph retrieval recall result set.

[0046] The neural network model unit is used to build a neural network model, feed the imaging data corpus into the model, perform machine learning and training, perform intent analysis and prediction on patient case information, generate question-answer pairs and risk prediction candidate sets, obtain a neural network result set, and assist in graph retrieval and matching.

[0047] The similarity feature extraction unit is used to extract similarity features from the data in the image data corpus to make matching sentence pairs required for model training. Similarity feature extraction includes: sentence length similarity extraction, longest common substring extraction, longest common subsequence extraction, edit distance feature extraction, n-gram feature word extraction, jieba (stammer) segmentation feature word extraction, question word feature extraction, rule weight keyword feature extraction, and word vector feature extraction. Sentence length similarity extraction is used to extract the sentence length features of both matching sentences, and the difference in sentence length is used to measure the similarity of intentions; the longest common substring extraction is used to extract the longest identical string length features contained in both matching sentences, and the longest repeated word in the sentence is used to measure the similarity; the longest common subsequence extraction is used to extract the cumulative value features of the longest identical string length contained in both matching sentences, and the length of repeated words in the sentence is used to measure the similarity; the edit distance feature extraction is used to extract the transformation complexity of both matching parties, and the transformation complexity of both parties is used to measure the similarity features of sentence length, sentence order and repeated word dimensions; n-gram feature extraction is used to extract the repeated word features of two-character combinations and three-character combinations of both matching parties, and the similarity is measured by the number of adjacent characters in the sentence; jieba segmentation feature extraction is used to extract the number of regular words that are the same after jieba segmentation of both matching parties, and the similarity is measured by the repetitiveness index of regular words in the sentence; question word feature extraction is used Extract and compare the same question word features of both parties, use question words to locate the question type, and grasp the similarity of the sentence body, such as: "What is?" represents that the answer tends to be an explanatory result, such as the introduction of the disease, and "How long?" represents that the answer is a time, such as recovery time; rule weight keyword feature extraction is used to add weight to the weight keywords contained in both parties, such as: both parties contain "knee", which means that the patient's current round of questions revolves around the core of "knee", and the feature words containing "knee", "knee joint", and other knee joint-related features in the corpus can be weighted to improve the score reference for similarity calculation; word vector feature extraction is used to perform mathematical vectorization symbol processing on the sentences of both parties, and directly send them to the BiLSTM (bidirectional recurrent long short-term memory) deep learning model and the Attention mechanism model (Bert) to learn similarity rules.

[0048] The recommendation model unit is used to send the features extracted by the similarity feature extraction unit into the matching model to learn similarity features: including collaborative filtering algorithm, matrix decomposition, GBDT+LR (decision tree model + logistic regression) model, Wide&Deep deep learning model, rule weight recall; the collaborative filtering algorithm is used to calculate the result set based on patient (UserCF) similarity and the result set based on disease (ItemCF) similarity; matrix decomposition is used to decompose the co-occurrence matrix in UserCF or ItemCF to calculate the association similarity of the implicit semantics of patient and disease information, such as : Patient matrix, Zhang San: [chest tightness 0.8 nausea 0.6 arrhythmia 0.9], symptom matrix, gallbladder heart disease: [chest tightness 0.81 nausea 0.61 arrhythmia 0.91], then Zhang San's score for gallbladder heart disease is the frequency of Zhang San's chest tightness * the probability of gallbladder heart disease having chest tightness + the frequency of Zhang San's nausea * the probability of gallbladder heart disease having nausea + the frequency of Zhang San's arrhythmia * the probability of gallbladder heart disease having arrhythmia = 0.8*0.81+0.6*0.61+0.9*0.91, the co-occurrence part of the two implicit matrices of the patient matrix and the symptom matrix is ​​calculated and condensed The GBDT+LR (decision tree model + logistic regression) model is used to automatically perform feature screening and combination of context information using the decision tree machine learning model, generate new discrete feature vectors, and generate prediction results through the LR (logistic regression) model, which makes up for the deficiency that collaborative filtering only uses the associated interactive information between patients and symptoms and ignores the characteristics of patients themselves and the characteristics of symptom information. The Wide&Deep deep learning model is used to use the Wide part of the deep learning model to quickly associate rules to roughly screen similarities, which enhances the "direct memory ability" of the model, and the Deep part to deeply abstract the training model, which enhances the "abstract generalization ability" of the model, so as to achieve the purpose of combining efficiency and accuracy. The rule weight recall is used to weight the recall results of the weighted feature words in the similarity extraction features to obtain the rule weight result set. For example, for "knee" related issues, the "knee" related term information in the image data corpus is weighted to calculate the similarity ranking.

[0049] The recall result set unit is used to splice multiple recall result sets such as graph retrieval result set, neural network result set, matching model result set, rule weight result set, and recommendation model result set, and normalize them into dimensionless and scalar indicator probability values ​​between 0 and 1 to obtain a multi-channel fusion coarse screening recall result set.

[0050] In this embodiment, the medical guidance module includes a first precise sorting unit and a first recommendation application unit. The first precise sorting unit includes a first unit for extracting supplementary business features and a first precise sorting recommendation ranking model unit. The first unit for extracting supplementary business features is used to extract the multi-channel recall result set to strengthen the features associated with the patient-side business, including departments, examination items, disease knowledge, bio-information guidance, and risk prediction. The first precise sorting recommendation ranking model unit is used to generate a new data corpus from the supplementary extracted business features and feed it into the model to perform refined machine learning, training, similarity calculation, sorting, and screening to obtain the patient-side recommendation result set. The first recommendation application unit is used to combine the patient-side recommendation results, including departments, examination items, disease knowledge, bio-information guidance, and risk prediction with patient cases, and recommend them to the patient-side application.

[0051] In this embodiment, the doctor-side inspection guidance module includes a second fine sorting unit of the recall result set and a second recommendation application unit, and the second fine sorting unit includes a second extraction and supplementary business feature unit, a second fine sorting recommendation ranking model unit and a key word reminder unit; the second extraction and supplementary business feature unit is used to extract multiple recall result sets to strengthen the features associated with the doctor-side business, including professional examination items, project introductions, indications, equipment, scanning methods, examination body positioning, and treatment plans; the second fine sorting recommendation ranking model unit is used to generate a new data corpus from the supplementary extracted business features and send it into the model to perform refined machine learning, training, calculation of similarity, sorting, and screening to obtain the doctor-side recommended result set; the key word reminder unit is used to highlight and quickly remind key words such as symptoms and risks in patient cases, and identify the symptoms and risk key words in patient cases; the second recommendation unit is used to combine the doctor-side recommendation results, including examination items, project introductions, indications, equipment, scanning methods, examination body positioning, treatment plans, and key word quick reminders with patient cases, and recommend them to the doctor-side application.

[0052] The embodiment of the present invention provides a medical imaging diagnosis and inspection guidance system based on artificial intelligence. Specifically, the system integrates a complete set of computer artificial intelligence natural language processing (NLP) technical solutions including machine learning, deep learning, similarity matching, intent recognition, knowledge graphs, recommendation systems, question-and-answer systems, risk prediction, etc., and processes, trains, and machine-learns industry data including professional, large-scale, and systematic real corpus data of professional imaging reports and professional academic materials on imaging examinations. It can realize artificial intelligence technology-assisted recommendations and question-and-answer consultations for professional and standardized examination items, treatment plans, and services for medical imaging, and assist patients in self-examination and doctors in diagnosis.

[0053] In the above first embodiment, a medical imaging diagnosis and inspection guidance system based on artificial intelligence is provided. Correspondingly, another embodiment of the present application provides a medical imaging diagnosis and inspection guidance method based on artificial intelligence. Please refer to Figure 2, shows a flow chart of a medical imaging diagnosis and inspection method based on artificial intelligence provided by another embodiment of the present application. Since the method embodiment is basically similar to the device embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the device embodiment. The method embodiment described below is only illustrative.

[0054] like Figure 2 FIG. 1 is a flowchart of a medical imaging diagnosis and inspection guidance method based on artificial intelligence provided by another embodiment of the present invention, and the method includes the following steps:

[0055] S1: Obtain the patient’s basic information, chief complaint information, and AI consultation communication information.

[0056] Specifically, the patient's main complaint information such as symptoms, medical history, personal information, etc. entered in the chat window and other chat information with the AI ​​customer service are analyzed and stored in the database.

[0057] Specifically include:

[0058] Collect user personal information, including name, age, gender, region, symptom description, symptom attribute description, including onset location, degree, frequency, etc., medical history, allergy history and other information;

[0059] Recommend related doctors based on the patient's self-reported information for doctor-patient communication;

[0060] Search the mainstream medical Q&A information on the Internet to recommend experts to answer questions;

[0061] Patients communicate with AI customer service online, ask questions, and get answers. The specific methods used are: entity recognition, intent recognition, and graph retrieval. The entity recognition step performs strict search matching and similarity approximate matching entity recognition on the questions input by the patients, that is, synonym replacement, to generate standard statement questions; the intent recognition step sends the patient's standard statement questions after matching or synonym replacement processing to the machine learning model training classification for question classification, such as: sentences containing keywords such as "how to treat" and "what medicine to take" tend to be questions about consulting treatment methods, and sentences containing keywords such as "how long" and "recovery" tend to be questions about consulting treatment cycles; the graph retrieval step guides patient questions according to question classification to answer templates, such as: treatment methods The answer template for questions about treatment methods is "The treatment methods for disease {} are {}", and the answer template for questions about treatment cycles is "The cure cycle for disease {} is {}". The keywords are searched for data resources such as chronic disease knowledge graphs and imaging knowledge graphs to fill in the templates. For example, for the keyword "otitis media", the treatment method "using ear drops phenol glycerin and oral anti-inflammatory drugs" is retrieved, and the cure cycle is retrieved as "10-14 days". The complete answer is returned to the patient, "The treatment methods for otitis media are: using ear drops phenol glycerin and oral anti-inflammatory drugs", and "The cure cycle for otitis media is 10-14 days".

[0062] Record patient self-report information, patient chat information with AI customer service, doctor-patient communication information, and expert Q&A information, and store them in the database;

[0063] The records of the patient's own complaints, the patient's chat information with the AI ​​customer service, the doctor-patient communication information, and the expert Q&A information are retrieved from the database, analyzed and entered into the image data corpus, and used to return to the AI ​​customer service question and answer system unit to analyze and organize the dialogue. The data is recycled and used synchronously for subsequent matching module training and learning.

[0064] S2: Generate structured cases based on patient information and push the cases to the patient and doctor ends respectively.

[0065] Specifically, data such as patient complaints are obtained;

[0066] Generate structured case outcome data from patient information;

[0067] Obtain and generate case result data and display the information on the patient side;

[0068] Obtain generated case result data and display case lists and information on the doctor's side.

[0069] S3: Create and integrate medical imaging data resource libraries such as chronic disease knowledge graphs, imaging examination graphs, and imaging professional diagnostic materials.

[0070] Specifically, the relational disease name, introduction, cause, symptoms, treatment, prevention, daily life and other chronic disease knowledge data are used to generate graph data of entity and relationship pairs and stored in the chronic disease knowledge graph database;

[0071] Generate graph data of entity and relationship pairs from relational disease, examination, indication, treatment plan, equipment, scanning method, examination body position and other imaging professional data, and store them in the imaging examination knowledge graph database;

[0072] Obtain the original imaging report data knowledge base;

[0073] Obtain data on questions and answers from Internet medical consultation experts;

[0074] Extract entity and relationship pairs from patient cases, reports, and Internet expert Q&A data, and automatically create an extended knowledge graph in real time.

[0075] S4: Build a matching model based on the integrated matching algorithm, matching rules, and knowledge graph retrieval rules, extract features from the medical image data resource library, create an image data corpus, use machine learning methods to train the matching model, input case information into the matching model for matching, and obtain a multi-channel fusion coarse screening recall (recommendation) result set.

[0076] Specifically, data from the chronic disease knowledge graph, imaging examination knowledge graph, imaging report knowledge base, big data expert Q&A, and self-expanding knowledge graph are obtained, analyzed and processed, features are extracted, and represented with mathematical vectorized symbols to create an imaging data corpus for machine learning training intent and matching. According to the patient case data, the chronic disease management knowledge graph, imaging examination knowledge graph, and self-expanding knowledge graph are retrieved to find the entity relationship pairs of the associated matching results and obtain the graph retrieval recall result set. A neural network model is built, and the imaging data corpus is fed into the model for machine learning and training. The patient case information is analyzed and predicted for intent, which is used to generate question-answer pairs and risk prediction candidate sets, and a neural network result set is obtained to assist in graph retrieval and matching.The data in the image data corpus are used to make the matching sentence pair data required for model training, and similarity features are extracted: the sentence length features of both sides of the matching sentences are extracted, and the sentence length difference is used to measure the similarity of intention; the longest identical string length features of both sides of the matching sentences are extracted, and the longest repeated word in the sentence is used to measure the similarity; the cumulative value features of the longest identical string lengths contained in both sides of the matching sentences are extracted, and the length of repeated words in the sentence is used to measure the similarity; the transformation complexity of both sides is extracted, and the transformation complexity of both sides is used to measure the similarity features of sentence length, sentence order and repeated word dimensions; the repeated word features of two-character combinations and three-character combinations of both sides are extracted, and the length of adjacent words in the sentence is used to measure the similarity of similar words in the sentence. The similarity is measured by the number of identical characters; the number of identical regular words after Jieba segmentation of the two matching parties is extracted, and the repetition index of regular words in the sentence is used to measure the similarity; the interrogative word features that are the same in the two matching parties are extracted and compared, and the question words are used to locate the question type and grasp the similarity of the main body of the sentence; the weights of the weighted keywords contained in the two matching parties are extracted to improve the score reference for similarity calculation; the sentences of the two matching parties are mathematically vectorized and symbolized, and are directly sent to the BiLSTM (bidirectional recurrent long short-term memory) deep learning model and the Attention mechanism model (Bert) to learn the similarity law; the features extracted by the similarity feature extraction unit are sent to the matching model to learn the similarity features: the similarity result set based on the patient (UserCF) and the result set based on the similarity of the symptom (ItemCF) are calculated; the co-occurrence matrix in UserCF or ItemCF is decomposed to calculate the association similarity of the implicit semantics of the patient and symptom information; the decision tree machine learning model is used to automatically perform feature screening and combination on the context information to generate a new discrete feature vector, and the prediction result is generated through the LR (logistic regression) model to make up for the fact that collaborative filtering only uses the association interaction information between the patient and the symptom, and ignores the patient's own characteristics The shortcomings of the symptom and symptom information features are overcome; the Wide part of the deep learning model is used to quickly associate rules to roughly screen similarities and enhance the "direct memory ability" of the model; the Deep part is used to deeply abstract the training model and enhance the "abstract generalization ability" of the model to achieve the purpose of combining efficiency and accuracy; the weighted feature words in the similarity extraction features are weighted to weight the recall results; the graph retrieval result set, neural network result set, matching model result set, rule weight result set, recommendation model result set and other multiple recall result sets are spliced ​​and normalized to dimensionless and scalar indicator probability values ​​between 0 and 1 to obtain a multi-channel fusion rough screening recall result set.

[0077] S5: Extract the features associated with the patient's business from the multi-channel fusion rough screening recall result set, and send it to the first fine sorting model as a supplementary data corpus for training, sorting, and screening to obtain a result set of popular science guidance information, disease knowledge, risk prediction, etc., and send the recommendation results to the patient end.

[0078] Specifically, the features associated with the patient-side business are extracted from the multi-channel recall result set as the first extracted supplementary business features, including department, examination items, disease knowledge, bio-information guidance, and risk prediction;

[0079] The first supplementary extraction of business features generates a new data corpus and feeds it into the first refined sorting model to perform refined machine learning, training, similarity calculation, sorting, and screening to obtain a patient-side popular science recommendation result set;

[0080] Recommend results to patients, including departments, examination items, disease knowledge, biomedical guidance, risk prediction and patient cases, and recommend them to patients for application.

[0081] S6: Extract the features related to the doctor's business from the multi-channel fusion rough screening recall result set, and send it to the second fine sorting model as a supplementary data corpus for training, sorting, and screening to obtain service result sets such as professional treatment plans and diagnosis reminders, and send recommendation results to the doctor.

[0082] Specifically, the multi-channel recall result set is extracted to strengthen the features associated with the doctor's business as the second extraction of supplementary business features, including professional examination items, project introduction, indications, equipment, scanning methods, examination body positioning, and diagnosis and treatment plans;

[0083] The second supplementary extraction of business features generates a new data corpus and feeds it into the second refined ranking model for refined machine learning, training, similarity calculation, sorting, and screening to obtain a professional doctor-side recommendation result set;

[0084] Highlight key words such as symptoms and risks in patient records for quick reminders;

[0085] The results are recommended to doctors, including examination items, project introductions, indications, equipment, scanning methods, body positioning, treatment plans, quick reminders of key words, and patient cases, and recommended to doctors.

[0086] The above is a description of an embodiment of a medical imaging diagnosis and inspection guidance method based on artificial intelligence provided by the second embodiment of the present invention.

[0087] The artificial intelligence-based medical image diagnosis and inspection guidance method provided by the present invention is based on the same inventive concept as the above-mentioned artificial intelligence-based medical image diagnosis and inspection guidance system and has the same beneficial effects, which will not be repeated here.

[0088] like Figure 3As shown, an embodiment of the present invention also provides a structural block diagram of an intelligent device, the device includes a processor, an input device, an output device and a memory, the processor, input device, output device and memory are interconnected, the memory is used to store a computer program, the computer program includes program instructions, the processor is configured to call the program instructions to execute the method described in the above second embodiment.

[0089] It should be understood that in the embodiments of the present invention, the processor referred to may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0090] Input devices may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint direction information), a microphone, etc., and output devices may include a display (LCD, etc.), a speaker, etc.

[0091] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0092] In a specific implementation, the processor, input device, and output device described in the embodiments of the present invention may execute the implementation described in the method embodiment provided in the embodiments of the present invention, or may execute the implementation described in the system embodiment described in the embodiments of the present invention, which will not be described in detail here.

[0093] The present invention further provides an embodiment of a computer-readable storage medium, wherein the computer storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor executes the method described in the second embodiment.

[0094] The computer-readable storage medium may be an internal storage unit of the terminal described in the foregoing embodiments, such as a hard disk or memory of the terminal. The computer-readable storage medium may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the terminal. Further, the computer-readable storage medium may also include both an internal storage unit of the terminal and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by the terminal. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.

[0095] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0096] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the terminals and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here.

[0097] In the several embodiments provided in the present application, it should be understood that the disclosed terminal and method can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or it can be an electrical, mechanical or other form of connection.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A medical imaging diagnosis and inspection guidance system based on artificial intelligence, characterized in that: include: The consultation module, case generation module, data resource module, matching module, guidance module and inspection guidance module are The consultation module is used to collect basic patient information, chief complaint information, and AI consultation communication information; The case generation module is used to generate structured cases from patient information and push the cases to the patient end and the doctor end respectively; The data resource module is used to create and integrate chronic disease knowledge graphs, imaging examination graphs, and medical imaging data resource libraries for imaging professional diagnostic materials; The matching module is used to construct a matching model based on the integrated matching algorithm, matching rules, and knowledge graph retrieval rules, extract features from medical image data resources, create an image data corpus, train the matching model using a machine learning method, input case information into the matching model for matching, and obtain a multi-channel fusion rough screening recall result set; The guidance module is used to extract features associated with the patient's business from the multi-channel fusion rough screening recall result set, and send it to the first fine sorting model as a supplementary corpus for training, sorting, and screening to obtain guidance information, disease knowledge, and risk prediction result sets, and send recommendation results to the patient end; The diagnosis guidance module includes a first sorting unit and a first recommended application unit. The first fine sorting unit is used to extract the service features associated with the patient from the multi-channel recall result set, generate new data corpus from the extracted service features associated with the patient and send it to the first fine sorting model for training, sorting and screening, obtain the guidance information, disease knowledge and risk prediction result set, and obtain the patient-side recommendation result set; The first application recommendation unit is used to recommend applications to the patient terminal according to the patient terminal recommendation result set; The first precision sorting unit includes a first extraction and supplementary service feature unit and a first precision sorting recommendation ranking model unit; the first extraction and supplementary service feature unit is used to extract and strengthen the features associated with the patient-side service from the multi-channel recall result set, including departments, examination items, disease knowledge, health knowledge guidance, and risk prediction; The first refined recommendation ranking model unit is used to generate new data corpus by supplementing and extracting business features and feed it into the model, perform refined machine learning, training, calculate similarity, sort, and filter to obtain a patient-side recommendation result set. The first recommendation application unit is used to combine the patient-side recommendation results, including departments, examination items, disease knowledge, health knowledge guidance, and risk prediction with patient cases, and recommend them to the patient-side application; The inspection guidance module includes a second sorting unit and a second recommended application unit. The second fine sorting unit is used to extract the business features associated with doctors from the multi-channel recall result set, generate new data corpus from the extracted business features associated with doctors and send it to the second fine sorting model for training, sorting and screening, obtain the diagnosis and treatment plan and diagnosis reminder service result set, obtain the doctor-side recommendation result set, and identify the symptoms and risk key words in the patient's case; The second application recommendation unit is used to recommend applications to the doctor side according to the doctor side recommendation result set; The inspection guidance module is used to extract the features associated with the doctor's business from the multi-channel fusion rough screening recall result set, and send it to the second fine sorting model as a supplementary data corpus for training, sorting, and screening to obtain the diagnosis and treatment plan and diagnosis reminder service result set, and send the recommendation results to the doctor end; The matching module includes a data processing unit and a graph retrieval unit. The data processing unit obtains data from a chronic disease knowledge graph, an imaging examination knowledge graph, an imaging report knowledge base, a big data expert Q&A, and a self-expanding knowledge graph, analyzes and processes the data, extracts features, and represents them with mathematical vectorized symbols to create an image data corpus for machine learning training intent and matching; the graph retrieval unit is used to retrieve a chronic disease management knowledge graph, an imaging examination knowledge graph, and a self-expanding knowledge graph based on patient case data, find associated matching result entity relationship data, and obtain a graph retrieval recall result set; The matching module also includes a neural network model unit, a similarity feature extraction unit, a recommendation model unit and a recall result set unit. The neural network model unit is used to construct a neural network model, send medical image data to the neural network model for training, perform intention analysis and prediction on patient case information, and generate a neural network result set; The similarity feature extraction unit is used to make the data in the image data corpus into matching sentence pair data required for neural network model training, and extract similarity features; The recommendation model unit is used to send the similarity features into the matching model to learn the similarity features, obtain a result set similar to the patient's main complaint information, and recommend a result set; The recall result set unit is used to splice the obtained multi-channel recall result sets, normalize them to dimensionless and scalar indicator probability values ​​between 0 and 1, and obtain a multi-channel fusion coarse screening recall result set.

2. The system according to claim 1, characterized in that The consultation module includes an AI customer service question and answer system unit, and the AI ​​customer service question and answer system unit includes an entity recognition unit, an intention recognition unit and a graph retrieval unit; The entity recognition unit is used to perform retrieval matching and similarity approximate matching entity recognition on the questions input by the patient to generate standard expression questions; The intention recognition unit is used to send the standard expression question into the machine learning model training classification to perform question classification; The graph retrieval unit is used to guide the patient's questions to answer templates according to question classification.

3. A medical imaging diagnosis and inspection method based on artificial intelligence, characterized in that: The following steps are involved: Obtain the patient's basic information, chief complaint information, and AI consultation communication information; Generate structured cases based on patient information and push them to the patient and doctor ends respectively; Create and integrate chronic disease knowledge graphs, imaging examination graphs, and medical imaging data resource libraries for imaging professional diagnostic materials; A matching model is constructed based on the integrated matching algorithm, matching rules, and knowledge graph retrieval rules. Features are extracted from the medical image data resource library to create an image data corpus. The matching model is trained using machine learning methods. Case information is input into the matching model for matching to obtain a multi-channel fusion rough screening recall result set. Extract features related to the patient's business from the multi-channel fusion rough screening recall result set and send it to the first fine sorting model as supplementary corpus for training, sorting and screening to obtain the guidance information, disease knowledge and risk prediction result set, and send the recommendation results to the patient end; Extract features related to patient-side services from the multi-channel recall result set as the first extracted supplementary business features, including departments, examination items, disease knowledge, health knowledge guidance, and risk prediction; The first supplementary extraction of business features generates a new data corpus and feeds it into the first refined sorting model to perform refined machine learning, training, similarity calculation, sorting, and screening to obtain a patient-side popular science recommendation result set; Recommend results to patients, including departments, examination items, disease knowledge, health guidance, risk prediction and patient cases, and recommend them to patients for application; Extract the features related to the doctor's business from the multi-channel fusion rough screening recall result set and send it to the second fine sorting model as supplementary data corpus for training, sorting and screening to obtain the diagnosis and treatment plan and diagnosis reminder service result set, and send the recommendation results to the doctor end; Integrating chronic disease knowledge graphs, imaging examination graphs, and imaging professional diagnostic data into medical imaging data resources specifically includes: Generate graph data of entity and relationship pairs from relational disease names, introductions, causes, symptoms, treatments, preventions, and daily chronic disease knowledge data, and store them in the chronic disease knowledge graph database; Generate graph data of entity and relationship pairs from relational disease, examination, indication, treatment plan, equipment, scanning method and examination body position imaging professional data, and store it in the imaging examination knowledge graph database; Collect and obtain the original image report data knowledge base; Collect data on the answers given by Internet medical consultation experts; Extract entity and relationship pairs from patient cases, reports, and Internet expert Q&A data, and automatically create an extended knowledge graph in real time; The method of extracting features from medical image data resources, creating an image data corpus, using a machine learning method to train a matching model, inputting case information into the matching model for matching, and obtaining a multi-channel fusion rough screening recall result set specifically includes: Obtain data from chronic disease knowledge graphs, imaging examination knowledge graphs, imaging report knowledge bases, big data expert Q&A, and self-expanding knowledge graphs, analyze and process them, extract features, represent them with mathematical vectorized symbols, and create an imaging data corpus for machine learning training intent and matching; Retrieve the chronic disease management knowledge graph, imaging examination knowledge graph, and self-expanding knowledge graph based on the patient case data, find the associated matching result entity relationship pairs, and obtain the graph retrieval recall result set; Build a neural network model, feed the image data corpus into the neural network model, machine learn and train, perform intention analysis and prediction on patient case information, generate question-answer pairs, risk prediction candidate sets, assist in graph retrieval and matching, and obtain the neural network result set; The data in the image data corpus is made into matching sentence pair data required for neural network model training, and similarity features are extracted; The similarity features are sent to the matching model to learn the similarity features, and a result set similar to the patient's chief complaint information is obtained, and a recommended result set is obtained; The obtained multi-channel recall result sets are concatenated and normalized to dimensionless and scalarized indicator probability values ​​between 0 and 1 to obtain a multi-channel fusion coarse screening recall result set.

4. An intelligent device, comprising a processor, an input device, an output device and a memory, wherein the processor, the input device, the output device and the memory are interconnected, the memory is used to store a computer program, the computer program comprises program instructions, and is characterized in that: The processor is configured to call the program instructions to execute the method according to claim 3.

5. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor, the processor is caused to perform the method according to claim 3.

Citation Information

Cited By

  • Auxiliary diagnosis method and system based on clinical big data

    CN119444744A