Patient-expert intelligent matching system and triage method based on diagnosis and treatment path

Abstract

The invention discloses a patient-expert intelligent matching system and a triage method based on a diagnosis and treatment path, and accurate mapping from pathological features to expert capabilities is realized by constructing a patient personalized diagnosis and treatment path and expert capability map. According to the method, a dynamic triage mechanism is adopted, deep learning and reinforcement learning technologies are combined, and the matching process of patients and experts is optimized according to the diagnosis and treatment path requirements of the patients and the real-time states of medical resources. The problems that traditional triage is low in efficiency and inaccurate in expert matching are solved, a new technical scheme is provided for optimal configuration of medical resources, and the method can be widely applied to intelligent triage scenes of medical institutions at all levels.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical information, and in particular to a patient-expert intelligent matching system and a triage method based on a diagnosis and treatment path. BACKGROUND

[0002] In the modern medical service system, pre-examination triage is the first key link for patients entering the hospital, and its efficiency and accuracy directly affect the patient's medical experience and the overall operation efficiency of the hospital. However, the traditional pre-examination triage mode has many problems to be solved. First, the complex process leads to the patient's unfamiliarity with the department distribution and the process, and often needs to go back and forth several times, which not only increases the patient's time and energy cost, but also reduces the service efficiency of the hospital. Second, the triage standard is not unified, and the judgment of different medical staff may differ, resulting in long patient waiting time and unreasonable resource allocation. In addition, the shortage of manpower is a common problem in busy hospitals, which further exacerbates the phenomenon of long patient waiting time, seriously affecting the quality of medical services.

[0003] With the popularization of artificial intelligence technology in the medical field, some intelligent guide diagnosis systems have emerged. For example, the intelligent guide diagnosis app based on medical big model launched by Longhua District, Shenzhen, guides the patient to state the illness through human-computer interaction dialogue, and realizes the recommendation of the hospital, department and doctor. However, most of the existing technologies are based on symptom matching for triage, and lack dynamic analysis of the complete diagnosis and treatment path of the patient, making it difficult to accurately match the patient's illness and the professional ability of the expert. For example, the disease triage and path optimization method disclosed in Chinese patent, although it considers the optimization of pre-examination items, it is still insufficient in expert matching.

[0004] On the other hand, clinical pathway management, as a standardization method of diagnosis and treatment, has been applied in hospitals, aiming to standardize the diagnosis and treatment process and reduce unnecessary diagnosis and treatment behaviors. However, how to organically combine clinical pathway with intelligent matching system to realize the accurate docking of patient diagnosis and treatment needs and expert ability is still a technical problem to be solved. In addition, existing expert matching systems, such as the real-time expert matching patent of LivePerson Inc., mainly target general customer consultation scenarios and do not fully consider the professionalism and complexity of the medical field, especially the dynamic matching of diagnosis and treatment path and expert ability.

[0005] In summary, the existing technology lacks dynamic analysis of the patient's diagnosis and treatment path in the triage process, making it difficult to predict subsequent diagnosis and treatment needs; expert matching is mostly based on static information such as department or title, without deep integration with the professional ability map of the expert; and the clinical pathway management and intelligent matching system are not effectively integrated, resulting in low resource allocation efficiency. Therefore, there is an urgent need for a system and method that can dynamically triage based on diagnosis and treatment path and realize accurate matching of patients and experts. SUMMARY

[0006] The application aims to provide a patient-expert intelligent matching system and a diagnosis method based on a diagnosis path.

[0007] To achieve the above-mentioned purpose, the application is implemented according to the following technical scheme:

[0008] The application comprises the following steps:

[0009] Obtaining a patient complaint, extracting a disease keyword in the patient complaint through a natural language processing technology, combining physical sign data and a history of the disease, constructing a standardized pre-examination data set, and pre-processing the pre-examination data set;

[0010] Inputting the pre-examination data set into a multi-modal diagnosis path generation model, and predicting a potential disease development path according to a pathological feature of the patient; the multi-modal diagnosis path generation model fuses a knowledge graph and a deep learning technology;

[0011] Extracting an expert ability requirement of each key node based on the disease development path, constructing an ability requirement vector, and giving different weights according to a priority of a path node; the ability requirement includes a disease diagnosis ability, a treatment technology requirement, and a complication handling experience;

[0012] Performing semantic matching in the ability requirement vector and an expert ability graph, and screening an expert candidate set;

[0013] Introducing a real-time state parameter of a medical resource, optimizing and sorting the expert candidate set through a reinforcement learning model, and generating a final expert recommendation list;

[0014] Generating a diagnosis guide according to an expert recommendation result, pushing patient information to a corresponding expert, and monitoring a patient condition change in a diagnosis and treatment process in real time, and if a path variation occurs, generating a diagnosis and treatment path again and updating an expert matching result.

[0015] Further, a method for constructing the expert ability graph comprises:

[0016] Integrating basic information from an expert information base to form an initial ability evaluation, calculating a disease coverage breadth index, evaluating a technical operation proficiency, and quantifying a complication handling experience; generating an expert feature code according to the initial ability evaluation, the disease coverage breadth index, the technical operation proficiency, and the complication handling experience; wherein the initial ability evaluation includes a comprehensive degree of a specialist disease diagnosis and treatment, a proportion of common diseases and rare diseases, a proficiency of various diagnosis and treatment technologies, an operation proficiency of basic technologies and high-difficulty technologies, a complication handling ability, and a handling experience of cases with different severity levels;

[0017] Constructing a relationship network according to a cooperative diagnosis and treatment relationship between experts, recording a case referral path and an effect, and labeling a teaching guidance relationship.

[0018] The multi-layer graph neural network fuses the expert's own characteristics and related expert characteristics, fuses multi-dimensional information by using a nonlinear activation function, and finally generates a low-dimensional vector representation to form a dynamic atlas reflecting disease coverage, technical proficiency, and complication handling capacity.

[0019] Further, the method for constructing the ability demand vector comprises:

[0020] The ability dimensions of each node of the diagnosis and treatment path are extracted, and a weighted ability demand vector is constructed.

[0021] Further, the method for optimizing and sorting the expert candidate set by the reinforcement learning model comprises:

[0022] The expert ability matching degree and the medical resource utilization rate are taken as optimization objectives, and a reinforcement learning model is used to generate an optimal expert recommendation.

[0023] Further, the method for constructing a patient-expert intelligent matching system according to the final recommendation list comprises:

[0024] A data acquisition and preprocessing module is configured to acquire clinical data of a patient and expert information, perform standardized processing on the clinical data, construct a structured patient data set, and construct an expert information database according to the expert information; the clinical data includes chief complaints, current medical history, past medical history, physical sign data, and test results; the expert information includes professional background, diagnosis and treatment experience, areas of expertise, and research achievements.

[0025] A diagnosis and treatment path construction module is configured to construct a personalized diagnosis and treatment path of the patient based on the structured patient data set in combination with clinical guidelines, historical cases, and a machine learning model; the personalized diagnosis and treatment path includes a diagnosis process, possible treatment schemes, expected examination items, and time nodes.

[0026] An expert ability modeling module is configured to construct an expert ability atlas based on the expert information database; the ability atlas includes disease field coverage, diagnosis and treatment technical proficiency, and complication handling experience of the expert.

[0027] An intelligent matching engine is configured to design a dynamic matching algorithm based on the diagnosis and treatment path of the patient and the ability atlas of the expert to obtain patient-expert matching; the dynamic matching algorithm adopts a double-layer matching mechanism: firstly, an expert candidate set is obtained by screening according to key node requirements of the diagnosis and treatment path; and then, a reinforcement learning model is used to optimize the matching result to generate an optimal expert recommendation list in combination with a real-time medical resource state.

[0028] Real-time feedback and update module: collect the diagnosis and treatment result data after the patient-expert matching, evaluate the matching effect, and feed back to the diagnosis and treatment path construction module and the expert ability modeling module, and continuously optimize and update based on the final recommendation list using online learning algorithm.

[0029] Further, the method for generating the personalized diagnosis and treatment path comprises:

[0030] A multi-modal deep learning model is used to generate a dynamic diagnosis and treatment path containing diagnosis process, treatment scheme and time node in combination with clinical guidelines and historical case data.

[0031] The diagnosis and treatment path nodes of the dynamic diagnosis and treatment path are preliminarily screened to retain key nodes with a probability of occurrence greater than a probability threshold; the key nodes are evaluated in multiple dimensions in combination with diagnosis and treatment value and resource consumption, the highest comprehensive benefit node combination is selected through an optimization algorithm, and a preliminary optimization path is formed;

[0032] The preliminary optimization path is matched with expert resources in real time according to the matching degree of expert expertise and case demand, the current workload of the expert, and the available time of the examination equipment, the order of the key nodes of the preliminary optimization path is adjusted through a dynamic adjustment mechanism, and the preliminary optimization path after adjustment of the key nodes is output as the personalized diagnosis and treatment path.

[0033] The dynamic adjustment mechanism: when the total time consumption exceeds a preset threshold, the key nodes with high diagnosis and treatment value are preferentially retained, and when multiple examinations need the same equipment, the examination item with the highest cost performance is selected.

[0034] Further, the personalized diagnosis and treatment path issuance variation re-generates the diagnosis and treatment path and updates the method for matching the expert result, comprising:

[0035] A multi-modal deep learning model is used to generate a dynamic path containing key nodes and variation processing in combination with a knowledge graph and clinical guidelines.

[0036] The present application has the following advantages:

[0037] The present application is a patient-expert intelligent matching system and a triage method and system based on diagnosis and treatment path, which has the following technical effects compared with the prior art:

[0038] The application realizes accurate intelligent matching of patients and medical experts by constructing personalized diagnosis and treatment paths and expert ability maps, combining deep learning and reinforcement learning technology. The system significantly improves triage efficiency and matching accuracy, and can dynamically adjust diagnosis and treatment schemes and expert recommendations according to patient conditions. Through intelligent optimization of medical resource allocation, not only the patient waiting time is shortened, the risk of complications is reduced, but also the reasonable allocation of expert workload is realized, which provides an efficient and reliable intelligent triage solution for medical institutions, effectively improves the overall medical service quality and patient medical experience. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 A flowchart of the steps of the patient-expert intelligent matching system and the triage method based on the diagnosis and treatment path of the application. DETAILED DESCRIPTION

[0040] The application will be further described below through specific embodiments. The illustrative embodiments of the application and the description are used to explain the application, but not as a limitation of the application.

[0041] The patient-expert intelligent matching system and the triage method based on the diagnosis and treatment path of the application include the following steps:

[0042] As shown in Figure 1 In this embodiment, the following steps are included:

[0043] Obtain the patient's complaint, extract the disease keywords in the patient's complaint through natural language processing technology, combine the physical sign data and the medical history, construct a standardized pre-examination data set, and pre-process the pre-examination data set;

[0044] Input the pre-examination data set into a multi-modal diagnosis and treatment path generation model, and predict the potential disease development path according to the patient's pathological characteristics; the multi-modal diagnosis and treatment path generation model integrates knowledge graph and deep learning technology;

[0045] Extract the expert ability requirements of each key node based on the disease development path, construct an ability requirement vector, and give different weights according to the priority of the path node; the ability requirement includes disease diagnosis ability, treatment technology requirement and complication handling experience;

[0046] According to the ability requirement vector, semantic matching is performed in the expert ability map to obtain an expert candidate set;

[0047] Introduce the real-time state parameters of medical resources, optimize and sort the expert candidate set through a reinforcement learning model, and generate a final expert recommendation list;

[0048] In the actual assessment, the patient's complaint of "right lower abdominal pain with vomiting for 1 day" is understood semantically by the BERT model, extracting the keywords "right lower abdominal pain", "vomiting", and "fever", and integrating them with data such as body temperature and blood routine into structured input;

[0049] Path generation: The diagnosis and treatment path generation model predicts a 70% probability of acute appendicitis, a 20% probability of acute gastroenteritis, and a 10% probability of other abdominal cavity diseases based on the input data. Accordingly, a three-level diagnosis and treatment path is generated: the first-level path (mandatory) includes emergency ultrasound and blood biochemical examination; the second-level path (selected according to ultrasound results) enters preoperative preparation if the ultrasound suggests appendiceal swelling, otherwise consider gastroscopy; the third-level path (deal with variations) such as perforation signs immediately surgery;

[0050] Capability requirement analysis: The first-level path requires "acute abdominal pain differential diagnosis ability" and "ultrasound image interpretation ability"; the second-level path requires "appendicitis diagnosis and treatment experience" and "laparoscopic surgery skills"; the third-level path requires "experience in handling acute abdominal symptoms complications"; each requirement dimension is assigned a weight according to the priority of the path node (e.g., first-level path weight 0.5, second-level 0.3, third-level 0.2);

[0051] Candidate set screening: In the expert capability map, 10 experts who meet the requirements of the first-level path are selected through semantic matching, and 5 experts who meet the requirements of the second-level path are further selected, and finally 3 experts who meet the requirements of the third-level path are selected;

[0052] Dynamic optimization: Considering the current load and expected idle time of the experts, the DQN model is used to calculate the optimal matching strategy; Expert B has handled 150 cases of acute appendicitis with a laparoscopic surgery success rate of 98%, and currently has no patients to be diagnosed, with the highest matching score (0.92), and is recommended as the first choice;

[0053] Triage execution and path update: The patient is triaged to Expert B, and the ultrasound examination suggests appendiceal swelling, confirming acute appendicitis, entering the second-level path; Expert B handles the preoperative preparation scheme recommended by the system; if perforation of the appendix is found during the operation (path variation), the system automatically triggers the third-level path, calling Expert C with experience in handling acute abdominal symptoms complications for consultation, achieving dynamic matching and updating;

[0054] Generate triage guidelines based on expert recommendations, and push patient information to corresponding experts, while monitoring patient condition changes in real-time during diagnosis and treatment, if path variation occurs, regenerate diagnosis and treatment path and update expert matching results;

[0055] In the actual evaluation, the time from admission to diagnosis of the patient with abdominal pain was shortened to 45 minutes, which was 50% higher than the traditional process (average 90 minutes); the expert matching accuracy reached 92%, the incidence of postoperative complications was reduced by 18% compared with historical data; at the same time, the system realized the balanced allocation of emergency department physician resources, and the daily average number of cases treated by expert B was controlled within a reasonable range (8-10 cases), avoiding excessive workload.

[0056] In this embodiment, the method for constructing the expert ability map comprises:

[0057] According to the integration of basic information in the expert information base, an initial ability evaluation is formed, a disease coverage index is calculated, a technical operation proficiency is evaluated, and a complication handling experience is quantified; according to the initial ability evaluation, the disease coverage index, the technical operation proficiency, and the complication handling experience, an expert feature code is generated; wherein the initial ability evaluation includes the overall degree of specialist disease diagnosis and treatment, the proportion of common diseases and rare diseases, the proficiency of various diagnosis and treatment technologies, the operation proficiency of basic technologies and high-difficulty technologies, the complication disposal ability, and the handling experience of cases with different severity grades;

[0058] According to the establishment of cooperative diagnosis and treatment relationship among experts, the recording of case referral path and effect, and the marking of teaching guidance relationship, a relationship network is constructed.

[0059] A multi-layer graph neural network is used to fuse the expert's own features and associated expert features, a nonlinear activation function is used to fuse multidimensional information, and finally a low-dimensional vector representation is generated to form a dynamic map reflecting disease coverage, technical proficiency, and complication disposal ability.

[0060] In this embodiment, the method for constructing the ability requirement vector comprises:

[0061] The ability dimensions of each node in the diagnosis and treatment path are extracted to construct a weighted ability requirement vector.

[0062] In this embodiment, the method for optimizing and sorting the expert candidate set by using a reinforcement learning model comprises:

[0063] The expert ability matching degree and the medical resource utilization rate are used as optimization objectives, and a reinforcement learning model is used to generate an optimal expert recommendation.

[0064] In this embodiment, the method for constructing a patient-expert intelligent matching system according to the final recommendation list comprises:

[0065] The data acquisition and preprocessing module is used for acquiring clinical data and expert information of a patient, performing standardized processing on the clinical data, constructing a structured patient data set, and constructing an expert information library according to the expert information. The clinical data includes chief complaints, current medical history, past medical history, physical sign data and test results. The expert information includes professional background, diagnosis and treatment experience, areas of expertise and research achievements.

[0066] The diagnosis and treatment path construction module is used for constructing a personalized diagnosis and treatment path of the patient based on the structured data set of the patient in combination with clinical guidelines, historical cases and a machine learning model. The personalized diagnosis and treatment path includes a diagnosis process, possible treatment schemes, expected examination items and time nodes.

[0067] The expert ability modeling module is used for constructing an expert ability graph based on the expert information library. The ability graph includes disease field coverage, diagnosis and treatment technology proficiency and complication handling experience of the expert.

[0068] The intelligent matching engine is used for designing a dynamic matching algorithm based on the diagnosis and treatment path of the patient and the ability graph of the expert, and obtaining patient-expert matching. The dynamic matching algorithm adopts a double-layer matching mechanism. First, an expert candidate set is obtained by screening according to key node requirements of the diagnosis and treatment path. Then, an optimal expert recommendation list is generated by optimizing the matching result through a reinforcement learning model in combination with real-time medical resource status.

[0069] The real-time feedback and updating module is used for collecting diagnosis and treatment result data after patient-expert matching, evaluating the matching effect, feeding back to the diagnosis and treatment path construction module and the expert ability modeling module, and continuously optimizing and updating based on the final recommendation list by using an online learning algorithm.

[0070] In the embodiment, the method for generating the personalized diagnosis and treatment path includes:

[0071] A multi-modal deep learning model is used to generate a dynamic diagnosis and treatment path including a diagnosis process, a treatment scheme and time nodes in combination with clinical guidelines and historical case data.

[0072] The diagnosis and treatment path nodes of the dynamic diagnosis and treatment path are preliminarily screened to retain key nodes with a probability greater than a probability threshold. The key nodes are evaluated in multiple dimensions in combination with diagnosis and treatment values and resource consumption. The node combination with the highest comprehensive benefit is selected by an optimization algorithm to form a preliminary optimized path.

[0073] The preliminary optimized path is matched with expert resources in real time according to a matching degree of expert expertise and case requirements, a current work load of the expert and an available time of an examination device. The key node order of the preliminary optimized path is adjusted by a dynamic adjustment mechanism. The preliminary optimized path after adjustment of the key nodes is output as the personalized diagnosis and treatment path.

[0074] The dynamic adjustment mechanism: when the total time consumption exceeds the preset threshold, the key node with high diagnosis and treatment value is preferentially reserved, and when multiple examinations need the same equipment, the examination item with the highest cost performance is selected.

[0075] In the embodiment, the personalized diagnosis and treatment path issuance variation re-generates the diagnosis and treatment path and updates the expert matching result method, comprising:

[0076] Using a multi-modal deep learning model, combining a knowledge graph and a clinical guideline, predicting the development trend of the patient's condition, generating a dynamic path containing key nodes and variation processing.

[0077] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A patient-specialist intelligent matching system and a diagnosis path-based triage method, characterized in that, The method comprises the following steps: Obtaining patient complaints, extracting disease keywords from the patient complaints through natural language processing technology, combining sign data and medical history to construct a standardized pre-examination data set, and preprocessing the pre-examination data set; Inputting the pre-examination data set into a multi-modal diagnosis and treatment path generation model to predict potential disease development paths according to the patient's pathological characteristics; the multi-modal diagnosis and treatment path generation model integrates knowledge graph and deep learning technology; Based on the disease development path, the expert ability requirement of each key node is extracted, the ability requirement vector is constructed, and different weights are given according to the priority of the path node; the ability requirement includes disease diagnosis ability, treatment technology requirement and complication handling experience; According to the ability requirement vector, semantic matching is carried out in the expert ability graph to obtain an expert candidate set; Introducing real-time state parameters of medical resources, optimizing and sorting the expert candidate set through a reinforcement learning model to generate a final expert recommendation list; According to the expert recommendation result, a triage guide is generated, and the patient information is pushed to the corresponding expert, and the patient's condition is monitored in real time during the diagnosis and treatment process. If the path changes, a new diagnosis and treatment path is generated and the expert matching result is updated. 2.The patient-specialist intelligent matching system and triage method based on diagnosis and treatment path of claim 1, wherein, The method for constructing the expert ability graph comprises: According to the expert information base, the basic information is integrated to form an initial ability evaluation, the disease coverage breadth index is calculated, the technical operation proficiency is evaluated, and the complication handling experience is quantified; the expert feature code is generated according to the initial ability evaluation, the disease coverage breadth index, the technical operation proficiency and the complication handling experience; wherein the initial ability evaluation includes the comprehensive degree of specialist disease diagnosis and treatment, the proportion of common diseases and rare diseases, the proficiency of various diagnosis and treatment technologies, the operation proficiency of basic technologies and high difficulty technologies, the complication disposal ability, and the handling experience of different severity cases; According to the cooperation diagnosis and treatment relationship between experts, the case referral path and effect are recorded, and the teaching guidance relationship is labeled to construct a relationship network; A multi-layer graph neural network is used to integrate the expert's own features and related expert features, a nonlinear activation function is used to integrate multidimensional information, and finally a low-dimensional vector representation is generated to form a dynamic graph reflecting disease coverage, technical proficiency and complication disposal ability. 3.The patient-specialist intelligent matching system and triage method based on diagnosis and treatment path of claim 1, wherein, The method for constructing the ability requirement vector comprises: Extracting the ability dimensions of each node of the diagnosis and treatment path to construct a weighted ability requirement vector. 4.The patient-specialist intelligent matching system and triage method based on diagnosis and treatment path of claim 1, wherein, The method for optimizing and sorting the expert candidate set through a reinforcement learning model comprises: Taking the expert ability matching degree and the medical resource utilization rate as the optimization target, and using a reinforcement learning model to generate an optimal expert recommendation. 5.The patient-specialist intelligent matching system and triage method based on diagnosis and treatment path of claim 1, wherein, The method for constructing a patient expert intelligent matching system according to the final recommendation list comprises: A data acquisition and preprocessing module is used to obtain clinical data of patients and expert information, and to standardize the clinical data, construct a structured patient data set, and construct an expert information base according to the expert information; the clinical data includes complaints, current medical history, past medical history, sign data and test results; the expert information includes professional background, diagnosis and treatment experience, field of expertise and research achievements; The diagnosis and treatment path construction module: based on the structured data set of the patient combined with clinical guidelines, historical cases and machine learning models, the personalized diagnosis and treatment path of the patient is constructed; the personalized diagnosis and treatment path includes diagnosis process, possible treatment scheme, expected examination item and time node; The expert ability modeling module: based on the expert information base, the expert ability graph is constructed; the ability graph includes the disease field coverage, diagnosis and treatment technology proficiency and complication handling experience of the expert; The intelligent matching engine: based on the diagnosis and treatment path of the patient and the ability graph of the expert, a dynamic matching algorithm is designed to obtain the patient-expert matching; the dynamic matching algorithm adopts a double-layer matching mechanism: firstly, the expert candidate set is obtained according to the key node demand of the diagnosis and treatment path; then, through the reinforcement learning model combined with the real-time medical resource state, the matching result is optimized to generate the optimal expert recommendation list; The real-time feedback and updating module: the diagnosis and treatment result data after the patient-expert matching is collected, the matching effect is evaluated, and the feedback is fed back to the diagnosis and treatment path construction module and the expert ability modeling module, and the online learning algorithm is used to continuously optimize and update based on the final recommendation list. 6.The patient-specialist intelligent matching system and triage method based on diagnosis and treatment path of claim 5, wherein, The method for generating the personalized diagnosis and treatment path comprises: A multi-modal deep learning model is used to combine clinical guidelines and historical case data to generate a dynamic diagnosis and treatment path including diagnosis process, treatment scheme and time node; The diagnosis and treatment path nodes of the dynamic diagnosis and treatment path are preliminarily screened to retain key nodes with a probability greater than a probability threshold; the key nodes are evaluated in multiple dimensions in combination with diagnosis and treatment value and resource consumption, the highest comprehensive benefit node combination is selected through an optimization algorithm, and a preliminary optimized path is formed; According to the matching degree of the expert's specialty and the case demand, the current workload of the expert and the available time of the examination equipment, the preliminary optimized path is matched with the expert resources in real time, the key node order of the preliminary optimized path is adjusted through a dynamic adjustment mechanism, and the preliminary optimized path after the key nodes are adjusted is output as the personalized diagnosis and treatment path; The dynamic adjustment mechanism: when the total time consumption exceeds a preset threshold, the key nodes with high diagnosis and treatment value are preferentially retained, and when multiple examinations need the same equipment, the examination item with the highest cost performance is selected. 7.The patient-specialist intelligent matching system and triage method based on diagnosis and treatment path of claim 1, wherein, The method for re-generating the diagnosis and treatment path and updating the expert matching result when the personalized diagnosis and treatment path issues a variation comprises: Using a multi-modal deep learning model, combined with a knowledge graph and clinical guidelines, the patient's disease development trend is predicted, and a dynamic path including key nodes and variation processing is generated.

Images