A physician training and assessment system based on digital patients

Abstract

The present application provides a physician training and assessment system based on digital patients, which relates to the field of pre-job training technology. The system includes: a digital patient generation module, an assessment strategy configuration module, an assessment module, an assessment feedback module, a data analysis module and a training module; obtaining case data, and constructing a virtual patient according to the case data; generating an assessment and training strategy configuration according to historical assessment data and a training plan; configuring the virtual patient according to the assessment and training strategy configuration to obtain a target virtual patient, and assessing the physician through the target virtual patient to obtain physician assessment data; analyzing and evaluating the physician assessment data to obtain an assessment and evaluation report; performing data analysis on the assessment and evaluation report and historical assessment data to obtain a physician capability analysis report; optimizing the training plan according to the physician capability analysis report to obtain an optimized training plan. The above scheme can improve the effect of physician training.

Description

Technical Field

[0001] The present application relates to the technical field of pre-job training, and in particular to a physician training and assessment system based on digital patients. Background Art

[0002] With the continuous development of medical science and the growing demand for medical care, physician training and assessment are facing new challenges. Traditional physician training and assessment methods mainly rely on theoretical examinations, on-site operation assessments, etc., which have certain limitations. First, traditional assessments often use standardized cases or manikin, which are difficult to fully reflect the complexity and diversity of clinical practice, and there is a certain gap between the assessment scenario and actual work. Usually, a "one-size-fits-all" assessment standard and training program are adopted, which does not fully consider factors such as the physician's previous ability foundation, learning characteristics, and development potential, making it difficult to teach students in accordance with their aptitude and with strong pertinence. Summary of the invention

[0003] The present application provides a physician training and assessment system based on digital patients, which can improve the effect of physician training.

[0004] In a first aspect, the present application provides a physician training and assessment system based on digital patients, the system comprising: a digital patient generation module, an assessment strategy configuration module, an assessment module, an evaluation feedback module, a data analysis module and a training module;

[0005] A digital patient generation module is used to obtain case data and construct a virtual patient based on the case data;

[0006] The case acquisition module is used to acquire case data;

[0007] The data preprocessing module is used to preprocess the case data to obtain standard case data;

[0008] The feature extraction module is used to extract features from the standard case data to obtain feature case data; the feature extraction module includes: a structured feature extraction subunit, a text feature extraction subunit, an image feature extraction subunit and a feature integration unit;

[0009] The knowledge fusion module is used to fuse the characteristic case data with preset medical knowledge to construct a knowledge representation of a virtual patient;

[0010] The patient modeling module is used to construct the virtual patient according to the knowledge representation of the virtual patient;

[0011] An assessment strategy configuration generation module is used to obtain the historical assessment data of the physician to be assessed and the training plan of the physician to be assessed, and to generate an assessment training strategy configuration based on the historical assessment data and the training plan;

[0012] The assessment module is used to configure the virtual patient according to the assessment training strategy configuration to obtain the target virtual patient, and to assess the physician through the target virtual patient to obtain the physician assessment data;

[0013] Evaluation feedback module, used to analyze and evaluate physician evaluation data and obtain an evaluation report;

[0014] The data analysis module is used to analyze the assessment report and historical assessment data to obtain a physician capability analysis report;

[0015] The training module is used to optimize the training plan according to the physician capability analysis report to obtain an optimized training plan.

[0016] By adopting the above technical solutions, the digital patient generation module constructs highly simulated virtual patients through in-depth mining of real case data, providing high-quality cases that are close to clinical practice, consistent in standards, and sufficient in number for assessment. This not only expands the coverage of the assessment, but also allows doctors to practice multiple cases in a safe and realistic environment, making up for the problems of limited real case resources and difficult scene settings. Secondly, the assessment strategy configuration module introduces a generation mechanism for personalized assessment plans. It comprehensively analyzes the doctor's past performance and actual needs, and intelligently matches personalized assessment strategies, which not only highlights the doctor's shortcomings and weaknesses, but also takes into account the actual work needs, so as to teach students in accordance with their aptitude and accurately portray them. This highly targeted and flexible assessment design is conducive to mobilizing the enthusiasm of doctors and promoting their intensive training in weak areas and key capabilities. Thirdly, based on the assessment strategy configuration, the assessment module conducts scene design and intelligent regulation of virtual patients, highly simulating the entire process of real diagnosis and treatment. During the assessment, doctors interact with virtual patients in multiple rounds, fully demonstrating their clinical thinking and practical skills, and objectively reflecting their real level. The evaluation feedback module generates a quantitative report immediately after the assessment, helping doctors to accurately locate the crux of the problem, improve diagnosis and treatment ideas, accelerate the learning of lessons from the assessment, and strengthen the learning effect. In addition, the data analysis module combines the assessment report with historical assessment data, and uses data mining technology to dynamically portray the development trajectory of doctors' abilities in multiple dimensions. The physician ability analysis report condenses deep insights into the growth rules and potential predictions of doctors, providing a scientific basis for teaching students in accordance with their aptitude and formulating training strategies in a targeted manner.

[0017] Finally, the training module optimizes and adjusts the original training plan based on the physician capability analysis report to match the actual needs and characteristics of the physician. The optimized training content is richer and more flexible, and the allocation of training resources is more efficient and reasonable, which helps to fully unleash the learning potential of physicians and accelerate the leap in their capabilities.

[0018] In a second aspect of the present application, a method for physician training and assessment is provided, the method comprising:

[0019] Obtain case data and construct a virtual patient based on the case data;

[0020] Obtain the historical assessment data and training plan of the physician to be assessed, and generate assessment and training strategy configuration based on the historical assessment data and training plan;

[0021] The virtual patient is configured according to the assessment and training strategy configuration to obtain a target virtual patient, and the physician is assessed through the target virtual patient to obtain physician assessment data;

[0022] Analyze and evaluate physician assessment data to obtain assessment reports;

[0023] Conduct data analysis on the assessment report and historical assessment data to obtain a physician capability analysis report;

[0024] The training plan is optimized according to the physician capability analysis report to obtain an optimized training plan.

[0025] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0026] 1. The digital patient generation module of this application has achieved a leap from "small sample" to "large scale" through "infinite recursion" of real cases. Physicians can repeatedly hone their diagnosis and treatment skills in highly simulated assessment scenarios, breaking through the shackles of limited real case resources and single training scenarios, and achieving rapid iterative improvement of their level of ability.

[0027] 2. The feature extraction module of this application adopts a "divide and conquer" strategy. For different types of feature information in standard case data, special feature extraction subunits are set up to process them in a "categorized" manner. The structured feature extraction subunit is responsible for feature extraction of structured data, the text feature extraction subunit is responsible for semantic mining of text information, and the image feature extraction subunit is responsible for deep feature learning of medical images. This extraction method that teaches students in accordance with their aptitude fully considers the multi-source heterogeneous characteristics of medical data, and can maximize the intrinsic value of various types of data, so that every piece of data can be "fully utilized".

[0028] 3. The assessment strategy configuration generation module of this application is based on the physician's historical performance data, and uses the physician data analysis module to comprehensively "portray" the physician's ability characteristics, strengths and weaknesses. This data-driven analysis method can objectively and accurately understand the growth trajectory and ability level of each physician, and explore the ability development rules hidden behind the massive assessment data. The physician's ability portrait is like a "magic mirror", which can not only accurately locate the physician's current ability "coordinates", but also deeply deconstruct the key "code" that affects the physician's ability improvement, so that the assessment strategy can be formulated with a purpose and basis. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 An architectural diagram of a physician training and assessment system based on digital patients provided in an embodiment of the present application;

[0030] Figure 2 An architecture diagram of a digital patient generation module provided in an embodiment of the present application;

[0031] Figure 3 An architectural diagram of an assessment strategy configuration generation module provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] In order to enable technicians in this field to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0033] In the description of the embodiments of the present application, words such as "for example" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "for example" or "for example" is intended to present related concepts in a specific way.

[0034] In the description of the embodiments of the present application, the meaning of the term "multiple" refers to two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "include", "comprise", "have" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0035] In order to facilitate understanding of the method and system provided by the embodiments of the present application, before introducing the embodiments of the present application, the background of the embodiments of the present application is first introduced.

[0036] First, traditional physician assessment methods are mostly limited to theoretical examinations and on-site operation assessments, which makes it difficult to comprehensively and truly test physicians' comprehensive clinical abilities. Written examinations often focus on mechanical memorization of medical theoretical knowledge, which is divorced from actual diagnosis and treatment work situations. Although on-site operation assessments focus on practical skills, they are also difficult to highly simulate the complexity and variability of clinical practice due to limitations such as venues, equipment, and assessment objects. Existing assessments cannot deeply assess physicians' core abilities such as diagnostic thinking and treatment decisions, and there is still a gap with the increasingly complex requirements of clinical practice. Secondly, the pertinence and flexibility of the physician training system need to be further improved. Due to the current conditions of limited resources and insufficient faculty, most physician training currently adopts a unified training program and curriculum setting, which does not fully consider the physician's knowledge base, work background, career development plan, etc., and ignores the diversity of the physician group. The "one size fits all" training model is difficult to meet the personalized needs of physicians and restricts the full release of physician potential. How to teach students in accordance with their aptitude and provide more accurate and efficient training programs for physicians of different levels and types is a realistic problem that needs to be solved in physician training. Thirdly, the traditional physician training path is relatively single and lacks a continuous and systematic training and assessment mechanism. At different stages of their careers, physicians face multiple tasks such as improving their professional skills, expanding their disciplines, and changing their roles, which places higher demands on their continuous learning and self-transcendence. However, after completing standardized residency training, the subsequent continuing education and capacity building of physicians mainly rely on personal career planning and lack systematic training support. Training content and assessment evaluation are often limited to the professional knowledge level, and insufficient attention is paid to non-technical abilities of physicians such as humanistic qualities, teamwork, and social responsibility, making it difficult to adapt to the new requirements of industry development.

[0037] After the background introduction of the above content, those skilled in the art can understand the problems existing in the prior art. The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0038] Based on the above background technology, please refer to Figure 1 , Figure 1 A schematic diagram of the architecture of a physician training and assessment system based on a digital patient provided in an embodiment of the present application. The system can be implemented by a computer program or can be run as an independent tool application. Specifically, in an embodiment of the present application, the method can be applied on a server, but can also be applied on electronic devices such as a server. The physician training and assessment system based on a digital patient includes: a digital patient generation module 1, an assessment strategy configuration module 2, an assessment module 3, an evaluation feedback module 4, a data analysis module 5 and a training module 6;

[0039] A digital patient generation module 1 is used to obtain case data and construct a virtual patient based on the case data;

[0040] Specifically, a large amount of real case data is obtained by connecting with the hospital's electronic medical record system, medical image storage system, etc. These data include patient symptom descriptions, physical sign observations, laboratory test results, imaging data, treatment plans and prognosis results, etc., covering the entire cycle of disease diagnosis and treatment. The purpose of obtaining these case data is to establish a comprehensive and real disease knowledge base as the material basis for the subsequent construction of virtual patients. Next, the digital patient generation module cleans, annotates and classifies the massive case data obtained. Through natural language processing technology, key information elements in the case, such as chief complaint, current medical history, laboratory indicators, etc., are extracted and standardized according to unified medical terminology standards. At the same time, the disease is classified according to clinical manifestations, pathological characteristics, etc., to form a structured and standardized case knowledge base. The role of this step is to prepare data for the intelligent generation of virtual patients and improve the efficiency and accuracy of subsequent processing.

[0041] On this basis, the digital patient generation module applies machine learning algorithms to intelligently extract the typical characteristics and evolution patterns of the disease from the case knowledge base. By comparing and analyzing a large number of similar cases, the common manifestations of the disease in terms of symptoms, signs, and auxiliary examination results are identified, and the basic context and key nodes of the disease progression are summarized. This process can be regarded as an intelligent summary and refinement of the doctor's clinical experience, forming a reusable disease model and injecting realistic clinical characteristics into the virtual patient.

[0042] Finally, the digital patient generation module constructs a series of virtual patients based on the disease model and in combination with the requirements of assessment difficulty and coverage. Each virtual patient has complete case information, including basic information, chief complaint, medical history, physical examination findings, auxiliary examination results, etc., and also sets reasonable clues to the development of the disease and a variety of possible outcomes. The generated virtual patients not only conform to the clinical manifestations of real diseases, but also take into account the assessment goals and evaluation indicators, and can objectively and comprehensively examine the physician's diagnosis and treatment ideas and practical ability.

[0043] Based on the above embodiment, as an optional embodiment, please refer to Figure 2 ,The digital patient generation module includes : case acquisition module, data preprocessing module, feature extraction module, knowledge fusion module and patient modeling module;

[0044] A case acquisition module 11 is used to acquire case data;

[0045] Specifically, the case acquisition module mainly obtains case data through the following three ways: first, using the hospital's own electronic medical record system to directly extract structured case texts; second, sharing data with other medical institutions in the medical alliance to obtain more diverse case sources in the region; third, grabbing high-quality cases from professional medical literature databases. Through this "internal and external" multi-source data acquisition method, the case acquisition module can combine the massive amount of real diagnosis and treatment data accumulated by the hospital with typical teaching cases in the literature, which has both quantitative scale and professional depth, providing a solid corpus for the design of digital patients.

[0046] After the case acquisition module automatically collects data from multi-source heterogeneous environments, it will simultaneously consider the volume and quality of the acquired case data. On the one hand, the module will automatically count the number of cases and the comprehensiveness of disease coverage to evaluate whether the current data scale can support the subsequent generation of digital patients; on the other hand, the module will also sample and analyze the structural integrity and content compliance of the cases to preliminarily determine whether the data quality meets the standards. Only on the basis of meeting the requirements of both data volume and quality, will the case acquisition module officially start data capture and deliver the full amount of case data to the downstream data preprocessing module. This evaluation mechanism that emphasizes both quantity and quality ensures the efficiency and accuracy of case acquisition work, and lays a solid data foundation for the subsequent digital patient design.

[0047] After the case acquisition module continuously delivers a large amount of case "raw materials" from multiple sources to the digital patient generation module, the data preprocessing module begins to standardize these original case data. Many of the original case materials have not been systematically sorted and organized, and inevitably have some "defects" such as missing data, confusing logic, and different formats. This requires the data preprocessing module to use various data cleaning algorithms to "finely process" the case data, and finally extract a batch of standardized case data with regular structure, clear logic, and complete content. Only through this preprocessing can the essence be extracted from the vast case raw materials to provide high-quality data "fuel" for the generation of digital patients.

[0048] The data preprocessing module 12 is used to preprocess the case data to obtain standard case data;

[0049] Specifically, the case acquisition module uses crawler technology to automatically collect large-scale medical record samples from the hospital's electronic medical record system and medical literature database. The system sets parameters such as the target disease range, sample quantity threshold, and quality standards for medical record collection, and strategically controls the medical record collection process. For example, in response to the need to build a virtual patient for a certain disease, the system can be set to only collect medical records of that disease, and each case must cover different stages, different severities, and different comorbid symptoms. At the same time, it requires that the medical records be completed and the examination reports are complete to ensure that the collected case data is comprehensive, rich, and vertically coherent.

[0050] The case acquisition module uses web crawlers to achieve batch and automated medical record collection, and can collect large-scale medical record samples in a short period of time. The collected medical record data will form a structured medical record sample library, which has a huge amount of data, extensive content, and reliable sources, providing a solid data foundation for the subsequent digital patient construction.

[0051] After obtaining the original medical record data, the digital patient generation module will start the data preprocessing module to perform a series of standardized processing on the medical record data. The reason for preprocessing the original medical record data is that the medical record data accumulated by the hospital over the years often has problems such as non-standardization, non-uniformity, redundant noise, etc. Direct use will affect the quality of the digital patient model. In order to facilitate subsequent feature extraction and knowledge mining, the original medical record data must be converted into a standardized form with unified format and standardized content through preprocessing. The cleaned medical record data has higher quality and clearer feature expression, which can improve the accuracy of subsequent patient modeling.

[0052] The feature extraction module 13 is used to extract features from the standard case data to obtain feature case data; the feature extraction module includes: a structured feature extraction subunit, a text feature extraction subunit, an image feature extraction subunit and a feature integration unit;

[0053] Specifically, the feature extraction module comprehensively utilizes artificial intelligence technologies such as medical natural language processing and disease knowledge graphs to fully explore the hidden features behind standard case data. The first thing to bear the brunt is semantic feature extraction. Based on the medical natural language processing model, the feature extraction module deeply analyzes the semantic logic behind the medical record text. Through symptom terminology standardization, negative and positive recognition, semantic dependency analysis and other processing, the system accurately extracts the semantic essence of the main symptoms, medical history, clinical signs, auxiliary examination results, etc. in the medical record, and organizes them into a structured semantic feature knowledge base. This lays a semantic foundation for the subsequent construction of a complete clinical manifestation of the disease. Next is symptom feature extraction. The feature extraction module uses a rich medical disease manifestation knowledge graph to automatically identify various disease symptoms appearing in the medical record, and performs structured extraction based on attributes such as symptom frequency, duration, and severity. Through the systematic combing of symptom characteristics, not only can a clinical "portrait" of the disease be constructed, but also the causal relationship between the symptoms can be revealed, forming a complete "chain" of disease manifestations.

[0054] Based on the above embodiment, as an optional embodiment, the feature extraction module includes: a structured feature extraction subunit, a text feature extraction subunit, an image feature extraction subunit and a feature integration unit;

[0055] A structured feature extraction subunit is used to extract structured features from standard case data to obtain structured feature data;

[0056] Specifically, the structured feature extraction subunit first preprocesses the standard case data. It converts the unstructured case text into a standardized structured representation through a series of data preprocessing operations such as data cleaning and format conversion, laying the data foundation for subsequent feature extraction. This preprocessing step is like a "facelift" of the messy case data, giving it a brand new look and clearing the way for accurate extraction of structured features.

[0057] Based on standardized structured cases, the structured feature extraction subunit begins to exert its "clairvoyance" ability and extracts features from structured data at multiple granularities and angles. It uses pattern matching technologies such as regular expressions and dictionary matching to accurately locate structured features such as basic patient information, symptoms and signs, laboratory tests, imaging tests, and treatment plans, and extracts key attribute values, such as the location, degree, and nature of symptoms, and the type, indicators, and results of examinations. These refined structured feature data not only depict the static phenotype of the disease, but also outline the dynamic context of diagnosis and treatment, laying a solid foundation for shaping the "data skeleton".

[0058] In the process of extracting structured features, the structured feature extraction subunit will also cluster and integrate similar features. It uses medical knowledge such as domain dictionaries and ontology knowledge bases to map and align synonymous and near-synonymous symptoms and signs, examination indicators and other structured features, and merge them into a unified standardized representation. This semantic feature integration eliminates the differences in the description of case data, making the extracted structured features more standardized and unified, and also provides a semantic basis for subsequent knowledge fusion.

[0059] A text feature extraction subunit is used to extract text features from standard case data to obtain text feature data;

[0060] Specifically, the text feature extraction subunit first performs text preprocessing on the standard case data. It converts the unstructured medical record text into a structured representation that can be understood by computers through a series of natural language processing operations such as word segmentation, part-of-speech tagging, and named entity recognition. This preprocessing step is like "purifying" the messy text data, filtering out text noise and laying a linguistic foundation for accurately extracting text features.

[0061] Based on the structured text, the text feature extraction subunit begins to play its "sharp eye for pearls" ability, using a variety of text mining technologies to perform all-round, multi-level feature extraction of medical record text. It uses bag-of-words models and word embedding models such as TF-IDF and Word2Vec to extract linguistic features such as keywords, phrases, and sentences from medical record texts, and characterize the subject content and semantic information of medical record texts. At the same time, it also uses sequence annotation models such as conditional random fields and recursive neural networks to identify key medical entities such as symptoms, signs, diseases, drugs, and surgeries in medical record texts and their attribute modifiers, and mine semantic relationships between entities. After the extraction of these text features, the unstructured scattered text is given a structured "knowledge skeleton", disease information is highly concentrated, and clinical knowledge is deeply condensed.

[0062] In the process of extracting text features, the text feature extraction subunit also uses medical domain knowledge to semantically associate text features. It uses knowledge bases such as disease ontologies and medical dictionaries to map extracted medical entities to standardized terminology representations, and organizes key entities such as symptoms, signs, examinations, and diagnoses into a semantically coherent disease knowledge graph. This feature semantic association that integrates domain knowledge eliminates the ambiguity of medical record text representation, enhances the semantic basis of text features, and also provides a "learning map" for subsequent knowledge reasoning.

[0063] An image feature extraction subunit is used to extract image features from standard case data to obtain image feature data;

[0064] Specifically, the image feature extraction subunit first preprocesses the medical images in the standard case data. It converts medical images under different imaging conditions into a unified standardized representation through a series of image processing operations such as image calibration, noise removal, and grayscale normalization. This preprocessing step is like "beautifying" the uneven image data, eliminating imaging differences and laying the foundation for accurate extraction of image features.

[0065] Based on standardized medical images, the image feature extraction subunit begins to play its "eagle eye" ability, and uses a variety of medical image analysis technologies to extract multi-scale and multi-dimensional features of medical images. It uses traditional image processing algorithms, such as image segmentation, edge detection, texture analysis, etc., to extract the morphological features of medical images, such as the size, shape, boundary, density, etc. of lesions. At the same time, it also introduces deep learning models, such as convolutional neural networks and generative adversarial networks, to learn the deep semantic features of medical images and explore the abstract patterns and intrinsic representations contained in the images. Through the precise depiction of these image features, the disease information in medical images is highly condensed, and the clinical diagnosis clues are deeply revealed.

[0066] In the process of extracting image features, the image feature extraction subunit also uses the knowledge of radiology experts to semantically describe the image features. It uses domain knowledge bases such as radiology report templates and imaging terminology sets to associate the extracted imaging features with the corresponding clinical explanations, and uses professional terms to describe the cardiac shadow findings of the lesions. This feature semantic description that integrates expert knowledge gives the image features a clinical interpretation, enhances the professionalism of image understanding, and provides a semantic bridge for subsequent multimodal fusion with clinical manifestations.

[0067] The feature integration unit is used to integrate the structured feature data, text feature data and image feature data to obtain feature case data.

[0068] Specifically, the feature integration unit first performs feature alignment on the three types of feature data extracted. It uses data alignment technologies such as common primary key mapping and timestamp synchronization to associate and align the three types of heterogeneous data, namely structured features, text features, and image features, with "patient-visit" as the main line, to ensure the semantic consistency of different types of feature data. This feature alignment process is like leveling the pitch of uneven instruments, laying the foundation for harmonious performance.

[0069] Based on the aligned heterogeneous features, the feature integration unit begins to exert its ability of "comprehensive integration" and deeply integrates the three types of feature data with data fusion technology as the baton. It uses multi-view feature representation technology, such as multi-view learning and multimodal representation learning, to map structured features, text features, and image features to a shared feature space and learn the unified representation of disease features. In this representation space, multi-dimensional heterogeneous information such as clinical manifestations, symptoms and signs, and imaging manifestations of the disease are organically integrated, which not only retains the unique perspectives of various features, but also explores the internal connections between different features, and depicts the internal mechanism of the disease from multiple angles.

[0070] On the basis of multi-perspective feature learning, the feature integration unit further uses cutting-edge models such as attention mechanisms and graph neural networks to accurately calibrate and collaboratively model disease features from different perspectives. The attention mechanism dynamically adjusts the importance weights of different features to highlight the key features of disease manifestations; the graph neural network is used to model the topological relationship between features to reveal the inherent laws of disease occurrence and development.

[0071] The knowledge fusion module 14 is used to fuse the characteristic case data with the preset medical knowledge to construct the knowledge representation of the virtual patient;

[0072] Specifically, the knowledge fusion module uses the medical knowledge graph as the top-level framework to perform multi-angle and multi-level semantic association and fusion of characteristic case data.

[0073] The first is the semantic association of clinical features. The knowledge fusion module uses medical ontology to semantically map the refined clinical feature entities such as symptoms, signs, examinations, diagnosis, and treatment, and constructs a clinical manifestation knowledge graph with the disease as the core. By revealing the semantic relationship between various clinical features, such as symptoms leading to signs, examinations confirming diagnosis, and diagnosis guiding treatment, the clinical manifestation map not only depicts the overall picture of the disease, but also forms a complete logical chain of disease occurrence and development. The semantic association of clinical features organizes scattered data fragments into an organic and unified disease knowledge system, which lays a solid foundation for shaping digital patients with rigorous logic and rich expressions.

[0074] On the basis of the association of clinical features, the knowledge fusion module further introduces the knowledge of medical diagnosis and treatment standards to logically reconstruct the diagnosis and treatment ideas of digital patients. By integrating structured medical knowledge such as disease diagnosis and treatment guidelines and clinical pathways, the system simulates the doctor's diagnosis and treatment reasoning process and establishes an evidence-based diagnosis and treatment decision logic diagram. From medical history collection to physical examination, from auxiliary examination to differential diagnosis, from diagnostic reasoning to treatment selection, rigorous and meticulous medical logic is reflected behind each diagnosis and treatment decision node. The integration of clinical features and diagnosis and treatment standards not only highlights the professionalism and standardization of the diagnosis and treatment process of digital patients, but also makes their diagnosis and treatment reasoning logic closer to clinical reality, making them veritable "smart" patients.

[0075] The patient modeling module 15 is used to construct a virtual patient based on the knowledge representation.

[0076] Specifically, the patient modeling module takes the virtual patient knowledge representation as the core, and comprehensively uses a variety of artificial intelligence modeling technologies to shape the virtual patient's image at multiple levels and from multiple perspectives. The first thing to bear the brunt is the visual modeling of disease phenotypes. The patient modeling module uses computer graphics technology to render the clinical manifestation entities in the virtual patient knowledge representation in two-dimensional and three-dimensional visualization. Through the graphical presentation of key phenotypic nodes such as symptoms, signs, examinations, and diagnoses, the full picture of the virtual patient's disease is intuitively and three-dimensionally portrayed. Converting textual clinical characteristics into graphical disease phenotypes makes abstract medical knowledge vivid and concrete, and the clinical manifestations of the disease are clearer at a glance. This visual disease phenotype modeling not only enhances the perceptual cognition of the virtual patient's image, but also lays an intuitive foundation for the subsequent development of intelligent applications such as phenotyping and auxiliary diagnosis.

[0077] Assessment strategy configuration generation module 2, used to obtain historical assessment data of the physician to be assessed and the training plan of the physician to be assessed, and generate assessment training strategy configuration according to the historical assessment data and the training plan;

[0078] Specifically, by connecting data with the hospital information system, assessment management system, etc., past assessment records related to the physician to be assessed are extracted. These historical assessment data include detailed information such as the time, content, results, and evaluation of the physician's participation in various assessments, reflecting the physician's ability level and development changes at different stages and dimensions. The purpose of obtaining these data is to fully understand the current status of the physician's basic ability and clinical practice characteristics, so as to generate more accurate and personalized assessment strategies in the future. At the same time, the assessment strategy configuration generation module also needs to obtain the training plan of the physician to be assessed. The training plan is usually formulated uniformly by the physician's department or hospital based on factors such as medical service needs and physician career development goals, and clearly stipulates the theoretical knowledge, clinical skills, and humanistic qualities that the physician needs to focus on learning and mastering in a certain period of time in the future. The assessment strategy configuration generation module obtains the specific training plan for each physician to be assessed by accessing the hospital's training management system. The purpose of obtaining training plan information is to ensure that the generated assessment strategy can be consistent with the actual training needs of the physician, so as to promote the continuous improvement of the physician's ability.

[0079] Based on the above embodiment, as an optional embodiment, please refer to Figure 3 ,Assessment strategy configuration generation module, including: physician data analysis module and strategy generation module;

[0080] The physician data analysis module 21 is used to analyze the historical assessment data to obtain the physician's ability portrait;

[0081] Specifically, the physician data analysis module first cleans and preprocesses the historical assessment data. It converts assessment data from different sources and formats into structured and standardized analysis samples through a series of data purification operations such as missing value filling, outlier correction, and data format unification. This preprocessing step is like "removing the false and retaining the true" from the chaotic assessment data, filtering out data noise and laying a data foundation for accurately portraying the physician's ability portrait.

[0082] Based on the structured assessment data, the physician data analysis module begins to play its "eagle eye" ability, using a variety of data mining and statistical analysis techniques to conduct a comprehensive perspective on the physician's assessment performance. It uses unsupervised learning algorithms such as cluster analysis and association rule mining to explore the hidden patterns in the physician's assessment results in multiple dimensions such as professional skills, clinical operations, and doctor-patient communication, and identify physician groups with different professional expertise and different ability levels. At the same time, it also uses dynamic modeling methods such as time series analysis and growth curve models to depict the growth trajectory of physicians' abilities and predict their potential for professional development. Through the multi-angle perspective of these ability dimensions, the unique professional characteristics of each physician gradually emerge, and a personalized ability profile is vividly displayed on paper.

[0083] Based on the above embodiment, as an optional embodiment, the physician data analysis module includes: an assessment data processing submodule, a knowledge graph construction submodule and a learning behavior analysis submodule;

[0084] The assessment data processing submodule is used to pre-process the historical assessment data to obtain standard historical assessment data;

[0085] Specifically, the assessment data processing submodule first cleans the original historical assessment data. It corrects incomplete and inaccurate assessment records through a series of data purification operations such as missing value filling and outlier correction to ensure the integrity and accuracy of each assessment data. It is like "gold panning" in the messy assessment data, removing the sand and impurities, leaving only pure "data gold sand". This data cleaning process not only improves the quality of the assessment data, but also lays a high-quality data foundation for subsequent analysis.

[0086] The knowledge graph construction submodule is used to construct a physician's personal knowledge graph based on standard historical assessment data;

[0087] Specifically, the knowledge graph construction submodule first extracts physician competency elements based on standard historical assessment data. It uses natural language processing technology to semantically analyze the competency elements contained in the assessment index system, and extracts competency characteristics in multiple dimensions such as professional skills, clinical operations, academic achievements, and doctor-patient communication. This process is like classifying building blocks into different competency elements in a standardized "building block system" based on characteristics such as shape, color, and function, preparing for the construction of a personalized knowledge graph.

[0088] On the basis of the extraction of competence elements, the knowledge graph construction submodule further constructs the physician's personal knowledge graph. By defining the semantic relationship between competence elements, the logical links of different competence elements in the physician's growth path are portrayed, forming a three-dimensional knowledge network with physicians as the core and radiating multiple competence dimensions. This construction process is like a skilled painter who uses a brush to connect scattered competence elements into lines, and interweaves lines into surfaces, outlining a series of unique portraits of physician capabilities. In this knowledge graph, each physician's professional expertise is like a towering mountain, clinical experience is like a rushing river, and development potential is like a crisscrossing road, outlining a series of clearly visible career development trajectories.

[0089] The learning behavior analysis submodule is used to build a physician capability profile based on the physician’s personal knowledge graph.

[0090] Specifically, the learning behavior analysis submodule first extracts the learning behavior characteristics of physicians based on their personal knowledge graph. It uses data mining technology to deeply analyze the evolution path and logical association of the ability elements in the knowledge graph, and identifies the behavior patterns of physicians in professional learning, skill training, academic research, etc.

[0091] The strategy generation module 22 is used to generate the assessment and training strategy configuration according to the physician capability profile and training plan.

[0092] Specifically, the strategy generation module first integrates the physician's ability portrait with the training plan. Through data fusion technologies such as common primary key mapping and feature splicing, it seamlessly connects the ability portrait describing the physician's professional expertise with the training plan that reflects the hospital's development goals, and builds a feature space with two-way connection between "individuals and organizations". In this integrated feature space, the physician's current ability status and future development direction echo each other, and personal characteristics and organizational goals resonate at the same frequency. This "individual-organization" two-way fusion feature perspective respects the inherent laws of physician growth while taking into account the overall strategy of hospital construction, laying a feature foundation for customizing personalized and global assessment and training strategies.

[0093] On the basis of the fusion features, the strategy generation module begins to exert its "strategic" ability, making full use of various intelligent optimization technologies such as operational optimization, heuristic search, and reinforcement learning to explore the optimal solution for the assessment and training strategy. It takes the improvement of physician capabilities as the optimization goal and the hospital's talent training system as the constraint condition. It intelligently traverses and iteratively optimizes in the strategy search space to find the best strategy combination that can both teach students in accordance with their aptitude and adapt to the overall situation. In this process, the assessment focus, training methods, incentive mechanisms and other aspects of each physician are carefully designed, which not only fits their own ability shortcomings, but also integrates into the hospital's development direction to form a personalized capacity improvement plan. At the same time, the assessment and training strategies of physicians throughout the hospital have achieved goal coordination and resource coordination at the macro level, maximizing the overall effectiveness of training resources.

[0094] Assessment module 3, configured to configure the virtual patient according to the assessment training strategy configuration, obtain the target virtual patient, and assess the physician through the target virtual patient to obtain the physician assessment data;

[0095] Specifically, the assessment module will read the personalized assessment strategy configuration file of each physician to be assessed, extract key parameters such as the assessment focus and difficulty requirements set therein, and use these parameters as constraints to screen and match the existing digital cases in the virtual patient database. For cases that fully meet the assessment requirements, the system will directly retrieve them as the target virtual patients for this assessment; for cases that deviate from the assessment requirements, the system will intelligently reprocess them on this basis, and modify them to target virtual patients that meet the assessment requirements by making appropriate adjustments to the medical history details, clinical indicators, and physical signs descriptions. For example, if the assessment strategy requires that the physician's ability to diagnose rare diseases be focused on, the target virtual patient needs to select cases of rare diseases and set interference factors in the description of the condition to increase the difficulty of diagnosis; if the assessment strategy focuses on evaluating the physician's level of handling patients with multiple complications, the target virtual patient needs to select complex cases with multiple diseases coexisting and presenting cross-mixed clinical manifestations; if the assessment strategy emphasizes the examination of the physician's humanistic care and ethical decision-making ability, the medical record of the target virtual patient must include the patient's special psychological state, family background and other humanistic information, and set some ethical dilemmas.

[0096] When the configuration of the target virtual patient is completed, the assessment module will officially start the assessment process. The physician to be assessed needs to carry out diagnosis and treatment activities facing the target virtual patient in a virtual simulation environment. From consultation and physical examination to auxiliary examination, diagnosis and decision-making, and then to treatment plan formulation and efficacy evaluation and tracking, the physician must demonstrate the diagnosis and treatment ideas and practical skills throughout the whole process, just like facing real patients. During the interactive process, the assessment module will record every step of the physician's operation, including the comprehensiveness of medical history collection, the standardization of physical examination, the rationality of diagnostic ideas, the thoroughness of differential diagnosis, the suitability of treatment plans and other information. The system will eventually generate detailed physician assessment data, covering quantitative scoring of key operations and qualitative analysis of operation procedures.

[0097] Based on the above embodiment, as an optional embodiment, the assessment module includes: a configuration module and a target assessment module;

[0098] A configuration module is used to determine configuration parameters according to the assessment and training strategy configuration, and configure the virtual patient according to the configuration parameters to obtain a target virtual patient;

[0099] Specifically, the configuration module first determines a series of configuration parameters based on the assessment and training strategy configuration. It integrates factors such as the physician's ability level, training focus, and assessment goals, and generates various parameter settings for virtual patients through intelligent algorithm analysis, such as the complexity of the disease, the completeness of medical records, and the urgency of the treatment scenario. This process is like drawing a detailed "parameter drawing" for the shaping of the virtual patient under the guidance of the assessment blueprint, ensuring that each configuration parameter accurately corresponds to the physician's actual level and assessment requirements.

[0100] Based on the configuration parameters, the configuration module further configures the virtual patient. Based on the configuration parameters, it selects qualified case models from the virtual patient knowledge base, and adjusts and polishes the case details according to the parameter settings, giving the virtual patient vivid clinical characteristics.

[0101] The target assessment module is used to assess the physician through the target virtual patient and obtain the physician assessment data.

[0102] Specifically, the target assessment module first starts the assessment scenario of the target virtual patient. It presents the virtual case carefully designed by the configuration module to the physician, and sets the assessment tasks and time limits. The physician needs to analyze the condition, formulate diagnosis and treatment strategies, implement simulated treatments, and complete a series of clinical decisions and operations based on the case information within the specified time. This process is like an exciting "clinical battle", with various "traps" and "problems" set by the target virtual patient emerging one after another, constantly testing the physician's clinical thinking and adaptability.

[0103] Evaluation feedback module 4 is used to analyze and evaluate the physician evaluation data and obtain an evaluation report;

[0104] Specifically, after the evaluation feedback module receives the physician assessment data from the assessment module, it first needs to conduct a comprehensive analysis and evaluation of the data. The purpose of the analysis and evaluation is to objectively present the physician's comprehensive performance in this assessment, accurately diagnose the physician's strengths and areas of improvement, and form personalized improvement opinions and follow-up learning suggestions based on this. Evaluation feedback can help physicians gain a deeper understanding of themselves, find the right direction to work hard, and truly realize the transformation of assessment results into learning improvement.

[0105] When implementing the analysis and evaluation, the evaluation feedback module adopts a multi-dimensional and multi-angle evaluation index system and evaluation model. First, the system organizes and classifies the physician assessment data in a structured manner according to the dimensions of clinical information collection, medical knowledge application, diagnostic reasoning thinking, clinical operation specifications, and humanistic care communication, and extracts the key data points of the physician's performance in each dimension. This process can objectively portray the physician's ability level and behavioral characteristics in terms of condition analysis, diagnosis and treatment decision-making, and operation execution when facing the target virtual patient. Secondly, the evaluation feedback module compares the physician's performance data in each dimension with the preset evaluation index system to calculate the physician's score on each indicator. The evaluation index system covers multiple aspects such as the standardization, comprehensiveness, accuracy, rationality, and humanity of physician behavior, and achieves a detailed portrayal of the physician's ability level through quantitative scoring. For example, the evaluation of the physician's medical history collection should not only see whether it touches all key issues, but also examine whether the logical order of its inquiries is reasonable and whether the language expression is appropriate; the evaluation of the physician's diagnostic thinking should not only see the accuracy of its conclusions, but also analyze the rigor of its logical reasoning and the comprehensiveness of differential diagnosis. This three-dimensional, multi-level indicator evaluation can present all aspects of a physician's abilities in a panoramic manner.

[0106] Data analysis module 5, used to analyze the assessment report and historical assessment data to obtain a physician capability analysis report;

[0107] Specifically, the evaluation feedback module uses artificial intelligence technologies such as natural language processing, knowledge graphs, and deep learning to conduct intelligent mining and analysis of physician assessment data. First, the system extracts multi-dimensional features from the physician's full-process operation data in the virtual case assessment, including a series of quantitative indicators such as consultation time, completeness of medical history collection, physical examination coverage, diagnostic keyword hit rate, and suitability of treatment plans, which are used to characterize the standardization and accuracy of physician diagnosis and treatment operations. At the same time, the system will also apply natural language processing technology to perform semantic analysis on text data during the interaction between physicians and virtual patients. By identifying the professionalism of the physician's consultation language, the popularization of the explanation, and the appropriateness of emotional comfort, the physician's soft power in doctor-patient communication and humanistic care is evaluated.

[0108] Based on the extraction of the above assessment performance characteristics, the evaluation feedback module will associate and compare them with the medical knowledge graph. The medical knowledge graph is constructed in advance and covers standardized knowledge such as the etiology, clinical manifestations, diagnosis and treatment plans of various specialized diseases. Through comparative analysis, the system can identify the errors and omissions in the physician's diagnosis and treatment ideas and decision-making plans, and accurately locate their knowledge blind spots and ability shortcomings. The evaluation feedback module will also match the physician's assessment performance with the pre-set scoring rules, and automatically generate assessment scores for various ability dimensions. At the same time, the system will also compare the physician's assessment scores with other physicians horizontally and generate ranking information, so that the physician can intuitively understand his relative level among physicians at the same level.

[0109] Based on the above embodiment, as an optional embodiment, the data analysis module includes: a descriptive statistical analysis module, a longitudinal trend analysis module and a multi-dimensional comprehensive evaluation module;

[0110] The descriptive statistical analysis module is used to conduct descriptive analysis on the assessment report and historical assessment data to obtain descriptive analysis data;

[0111] Specifically, the descriptive statistical analysis module first conducts a comprehensive review of the assessment report and historical assessment data. It systematically integrates the physician's scores in different assessment dimensions, performance at different assessment time points, and evaluations in different assessment scenarios to form a structured data set. This process is like a diligent "data organizer", stringing together the scattered data "pearls" into an orderly "necklace", which facilitates subsequent statistical analysis.

[0112] On the basis of data collation, the descriptive statistical analysis module uses various statistical methods to describe the overall distribution and typical characteristics of physicians' abilities. It calculates the average score, median, standard deviation and other indicators of physicians in various assessment dimensions to outline the overall ability level of physicians; it collects data such as the error rate and reaction time of physicians in different assessment scenarios to reveal the characteristics of physicians when facing different clinical tasks; it sorts out the score rankings and distribution frequency of evaluation levels of physicians in previous assessments to show the relative position of physicians among the same population. This process is like a "capability sketch artist", which sketches a quick sketch of physicians' abilities with concise statistical indicators and intuitive distribution charts, directly hitting the core essence of physicians' abilities.

[0113] The longitudinal trend analysis module is used to integrate the assessment report and historical assessment data according to the time series to obtain the time series data, and to perform data analysis on the time series data to obtain the longitudinal trend data;

[0114] Specifically, the longitudinal trend analysis module first integrates the assessment report and historical assessment data in time series. It uses the time point of each physician assessment as a reference, extracts the assessment score, evaluation level and other data at that time point, and arranges them in chronological order to form a set of orderly time series data. This process is like a rigorous "time chronicler", "connecting" the "fragments" of assessment data scattered in different time and space in chronological order, and weaving a "chronicle" reflecting the growth process of the physician.

[0115] Based on time series data, the longitudinal trend analysis module uses data analysis technology to explore the longitudinal laws of physicians' ability development. It can calculate the growth rate of physicians' scores in various assessment dimensions and determine the speed of physicians' progress at different levels of ability; it can fit the trend line of physicians' previous assessment scores and predict the trend of physicians' ability changes in the future; it can also perform cluster analysis on physicians' assessment performance and identify the key "turning points" and "bottleneck periods" in the physicians' development process. This process is like an insightful "growth mentor" who identifies hidden growth laws in the complex assessment data and grasps the characteristics that best reflect the essence of physician growth in the tortuous development path.

[0116] The multi-dimensional comprehensive evaluation module is used to perform weighted scoring of physician capabilities based on longitudinal trend data and descriptive analysis data to obtain a physician capability analysis report.

[0117] Specifically, the multi-dimensional comprehensive evaluation module first conducts in-depth research and judgment on the longitudinal trend data and descriptive analysis data. It examines the performance of various abilities of physicians from different angles and levels, and conducts a three-dimensional evaluation of physicians' clinical skills, doctor-patient communication, teamwork, professional accomplishments, and other aspects by comprehensively analyzing the physicians' scores in various assessment dimensions, the improvement trends of previous assessments, and the comparison results with similar populations. This process is like a rigorous and meticulous "examiner", comprehensively collecting the "evidence" of the physician's performance, objectively measuring the "weight" of the physician's ability, and ensuring the comprehensiveness and accuracy of the evaluation results.

[0118] Based on comprehensive analysis, the multi-dimensional comprehensive evaluation module uses a weighted scoring method to quantitatively evaluate the various competence indicators of physicians. It sets differentiated weights for each assessment dimension based on the talent development strategies of different hospitals, highlighting the competence factors that hospitals value most; it refers to industry standards and expert opinions, and scientifically divides and sets scores for each competence level to ensure the rationality and comparability of the scoring system.

[0119] The training module 6 is used to optimize the training plan according to the physician capability analysis report to obtain an optimized training plan.

[0120] Specifically, the training module uses artificial intelligence technologies such as rule mining, reinforcement learning, and recommendation algorithms to conduct intelligent and in-depth mining of individual and group physician capability analysis reports, and implement personalized optimization of training plans accordingly. First, the system will compare and analyze the current ability level of individual physicians with the requirements of the job competency model, and automatically identify the key competency items that the physician urgently needs to improve. For these competency items, the system intelligently recommends the most matching learning content and training programs from the hospital's training resource library. For example, for the lack of diagnostic thinking ability, the system will recommend relevant professional skills training courses; for the lack of humanistic communication skills, the system will recommend corresponding special lectures on doctor-patient communication. This personalized and precise optimization of training content allows physicians to spend their time and energy on the cutting edge, hit the key points, and achieve twice the result with half the effort.

[0121] Based on the above embodiment, as an optional embodiment, the training module includes: a training demand diagnosis module, a training resource matching module and a training plan optimization module;

[0122] The training demand diagnosis module is used to diagnose and analyze the physician's ability needs based on the physician's ability analysis report to obtain training needs;

[0123] Specifically, the training needs diagnosis module uses the physician's ability analysis report as a "medical record" and uses methods such as needs analysis and ability diagnosis to conduct a comprehensive "consultation" on the physician's ability needs. It carefully compares the physician's scores in various ability dimensions, identifies the "shortcomings" dimensions with low scores, and judges the importance of these dimensions to the physician's development in combination with industry standards, and then determines the key capabilities that need to be improved most urgently; it systematically examines the physician's progress trend in previous assessments, finds out the "bottleneck" links where the ability improvement speed is relatively slow, analyzes the deep-seated reasons for the bottleneck, and digs out the ability shortcomings hidden under the surface; it also widely consults the physician's subjective willingness to improve his or her ability, deeply understands the physician's professional development demands, integrates the physician's subjective initiative into the training needs analysis, and maximizes the physician's endogenous motivation. This process is like an experienced "capability doctor", who accurately grasps the "lesions" of the physician's ability improvement through the complex "symptoms" and prescribes the most accurate and targeted "training prescription".

[0124] The training resource matching module is used to match the training requirements with the preset training resources to obtain a training resource list;

[0125] Specifically, the training resource matching module uses a rich set of preset training resources as a "database" and physicians' training needs as "search terms" to quickly screen massive resources using intelligent matching algorithms. First, it comprehensively sorts out and systematically classifies various types of training resources inside and outside the hospital, and establishes a training resource "supermarket" covering courses, teachers, teaching materials, venues, equipment and other elements; secondly, it uses natural language processing technology to extract keywords for each training need to form a search term library, and then conducts a global search in the resource "supermarket" to find the training resources with the highest relevance to the search terms; thirdly, it comprehensively considers factors such as the pertinence, systematicness, and accessibility of training resources, and uses intelligent algorithms to perform multiple rounds of matching and optimization combinations, and finally customizes the most appropriate training resource "package" for each training need. This process is like an efficient and professional "matchmaker", who not only understands the personalized needs of each "customer", but is also familiar with the massive resources in the "market", and through precise matching, it makes the needs and resources a "match made in heaven".

[0126] The training plan optimization module is used to optimize the training plan according to the training needs and the training resource list to obtain an optimized training plan.

[0127] Specifically, the training plan optimization module uses system theory and optimization methods to dynamically optimize the training plan from a global perspective. First, based on the hospital's development strategy and talent training goals, it systematically sorts out the training needs and training resource lists, finds out the connections and logic, and determines the overall framework and implementation path of the training plan at the macro level; secondly, it comprehensively considers factors such as the physician's knowledge base, learning characteristics, and work arrangements, and uses mathematical modeling and optimization algorithms to set parameters and optimize the combination of various elements of the training plan at the micro level, forming a set of courses, teachers, time planning, etc. with the highest matching degree and the strongest feasibility; thirdly, it introduces a dynamic feedback mechanism, and adjusts and continuously improves the training plan in real time according to the effect evaluation and physician feedback during the implementation of the training, to ensure the pertinence and effectiveness of the training activities. This process is like a "designer" with unique ingenuity, who can not only plan the blueprint from a global perspective, but also implement the local details to carve out meticulously, and through system optimization, let the training plan be perfected in a dynamic way.

[0128] Based on the above embodiment, as an optional embodiment, the training plan optimization module includes: an optimization target setting module, an optimization model building module and an optimization plan generating module;

[0129] Optimization target setting module, used to set optimization targets according to training needs;

[0130] Specifically, the optimization goal setting module uses the methods of goal management and demand analysis to set optimization goals based on sufficient research and judgment. First, it comprehensively collects the "demand list" formed by doctors in the training demand diagnosis module, and through questionnaires, interviews and other forms, deeply understands the detailed demands of doctors in terms of professional ability, professional quality, personal development, etc.; secondly, it is based on the hospital's talent development strategy and education and training plan, and systematically sorts out the physician training needs from the macro, meso and micro levels, finds out the logical connection and primary and secondary relationship between the needs, and forms a scientific and reasonable demand hierarchy; thirdly, it comprehensively weighs the personal development of doctors and the overall goals of the hospital, and through scientific methods such as the Delphi method and hierarchical analysis method, determines a quantitative optimization goal that takes into account both physician satisfaction and organizational performance, and further decomposes it into several specific and assessable sub-goals.

[0131] The optimization module builds a module for building an optimization model according to preset training resources and preset optimization logic of resource configuration;

[0132] Specifically, the optimization model construction module uses theoretical methods such as operations research and systems engineering to build an optimization model based on a systematic combing. First, it comprehensively inventories the hospital's various training resources such as teachers, teaching materials, equipment, and venues, and transforms these resources into digital information that can be processed by computers through system classification and data standardization. Secondly, it deeply studies the inherent laws and constraints of resource allocation, extracts the quantitative and logical relationships between resources and between resources and demand, and expresses them rigorously in mathematical language to form a complete set of optimization logic equations. Thirdly, it selects appropriate mathematical models, such as linear programming, integer programming, dynamic programming, etc., and incorporates resource constraints, demand goals, optimization logic and other elements into the model framework to build an optimization model with a complete structure and rigorous logic. This process is like a skilled "designer" who builds seemingly chaotic resource "building blocks" into optimized "algorithm models" according to specific logical rules.

[0133] The optimization plan generation module is used to input the optimization objectives and training resource list into the optimization model and output the optimization training plan.

[0134] Specifically, the optimization scheme generation module follows the basic logic of "input-output" and makes full use of the optimization model to dynamically optimize resource allocation. First, it uses the training optimization target input model established by the optimization target setting module as the "baton" for resource allocation optimization, and imports the training resource list formed by the training resource matching module into the model as the "raw material" for resource combination optimization; secondly, under the "guidance" of constraints and objective functions, the optimization model starts to run at high speed, searching for the "optimal solution" in a large number of resource combination schemes through mathematical methods such as multi-objective programming and integer programming, and further refines it into an executable training plan "construction drawing"; thirdly, it actively absorbs the opinions and suggestions of front-line teaching experts and training managers, evaluates and demonstrates the optimization scheme from the dimensions of professionalism, pertinence, feasibility, etc., and dynamically adjusts the model parameters accordingly to generate more mature and finalized training "solutions".

[0135] Based on the above embodiment, as an optional embodiment, the training resource matching module includes: a training demand analysis module and the resource matching recommendation module;

[0136] The training demand analysis module is used to analyze the training demand and obtain key demand tags;

[0137] Specifically, the training demand analysis module plays a "problem-solving" role and deeply deconstructs the training needs submitted by doctors. When doctors submit training demand applications in the system, they often explain the demand background, goals, key demands, etc. in text form. For unstructured demand texts, this module uses natural language processing technology to convert them into structured and standardized key demand tags through steps such as word segmentation, part-of-speech tagging, named entity recognition, and semantic dependency analysis. For example, a doctor submitted "I hope to improve the imaging diagnosis capabilities of lung tumors, especially CT diagnosis". After analysis and refinement, key demand tags such as "lung tumors", "imaging diagnosis", and "CT" were obtained. The core of this step is to retain the core essence of the demand to the greatest extent, eliminate complex information, condense keywords, and lay the foundation for subsequent matching.

[0138] The resource matching recommendation module is used to match the key requirement tags with the preset training resources to obtain the training resource list.

[0139] Specifically, the resource matching recommendation module receives the key demand tags output by the training demand analysis module. These tags highly refine the core points of physician needs, such as "coronary heart disease", "interventional treatment", "operation specifications", etc. At the same time, the system background maintains a preset training resource library, covering various training resources such as monographs, tutorials, cases, and practical training projects. When each resource is entered into the library, a series of key attribute tags will be extracted, such as disease, specialty, equipment, difficulty, etc., to describe the characteristic attributes of the resource.

[0140] Next, the resource matching recommendation module uses artificial intelligence technologies such as semantic analysis and machine learning to calculate the semantic association between key demand tags and resource attribute tags. Specifically, this module builds a set of semantic matching algorithm models that can deeply understand the semantic connotations behind the tags and reveal the intrinsic connections between the tags. For example, the key demand tags of doctors are "pediatric pneumonia" and "antibiotic selection". The algorithm model can find that "pediatrics" and "respiratory diseases" are highly semantically related to "pediatric pneumonia", and "drug treatment" is closely semantically related to "antibiotic selection". On this basis, the system automatically matches a batch of training resources whose resource attribute tags best match the key demand tags of doctors, generates a list of training resources and pushes them to doctors.

[0141] The above are only exemplary embodiments of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification and the truth of practice, those skilled in the art will easily think of other embodiments of the present disclosure.

[0142] This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art not described in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A physician training and assessment system based on digital patients, characterized in that: The system includes: a digital patient generation module, an assessment strategy configuration module, an assessment module, an evaluation feedback module, a data analysis module and a training module; The digital patient generation module is used to obtain case data and construct a virtual patient according to the case data; the digital patient generation module includes: a case acquisition module, a data preprocessing module, a feature extraction module, a knowledge fusion module and a patient modeling module; The case acquisition module is used to acquire case data; The data preprocessing module is used to preprocess the case data to obtain standard case data; The feature extraction module is used to extract features from the standard case data to obtain feature case data; the feature extraction module includes: a structured feature extraction subunit, a text feature extraction subunit, an image feature extraction subunit and a feature integration unit; The knowledge fusion module is used to fuse the characteristic case data with preset medical knowledge to construct a knowledge representation of a virtual patient; The patient modeling module is used to construct the virtual patient according to the knowledge representation of the virtual patient; The assessment strategy configuration generation module is used to obtain the historical assessment data of the physician to be assessed and the training plan of the physician to be assessed, and generate the assessment and training strategy configuration according to the historical assessment data and the training plan; the assessment strategy configuration generation module includes: a physician data analysis module and a strategy generation module; the physician data analysis module is used to analyze the historical assessment data to obtain the physician's ability portrait; the strategy generation module is used to generate the assessment and training strategy configuration according to the physician's ability portrait and the training plan; the physician data analysis module includes: an assessment data processing submodule, a knowledge graph construction submodule and a learning behavior analysis submodule; the assessment data processing submodule is used to preprocess the historical assessment data to obtain standard historical assessment data; the knowledge graph construction submodule is used to construct the physician's personal knowledge graph based on the standard historical assessment data; the learning behavior analysis submodule is used to construct the physician's ability portrait based on the physician's personal knowledge graph; The assessment module is used to configure the virtual patient according to the assessment training strategy configuration to obtain a target virtual patient, and to assess the physician through the target virtual patient to obtain physician assessment data; the assessment module includes: a configuration module and a target assessment module; the configuration module is used to determine configuration parameters according to the assessment training strategy configuration, and to configure the virtual patient according to the configuration parameters to obtain the target virtual patient; the target assessment module is used to assess the physician through the target virtual patient to obtain physician assessment data; The evaluation feedback module is used to analyze and evaluate the physician evaluation data to obtain an evaluation report; The data analysis module is used to perform data analysis on the assessment report and the historical assessment data to obtain a physician capability analysis report; The training module is used to optimize the training plan according to the physician capability analysis report to obtain an optimized training plan.

2. The system according to claim 1, characterized in that The feature extraction module includes: a structured feature extraction subunit, a text feature extraction subunit, an image feature extraction subunit and a feature integration unit; The structured feature extraction subunit is used to extract the structured features in the standard case data to obtain structured feature data; The text feature extraction subunit is used to extract text features from the standard case data to obtain text feature data; The image feature extraction subunit is used to extract the image features in the standard case data to obtain image feature data; The feature integration unit is used to integrate the structured feature data, the text feature data and the image feature data to obtain the feature case data.

3. The system according to claim 1, characterized in that The data analysis module includes: a descriptive statistical analysis module, a longitudinal trend analysis module and a multi-dimensional comprehensive evaluation module; The descriptive statistical analysis module is used to perform descriptive analysis on the assessment report and the historical assessment data to obtain descriptive analysis data; The longitudinal trend analysis module is used to integrate the assessment report and the historical assessment data according to the time series to obtain time series data, and perform data analysis on the time series data to obtain longitudinal trend data; The multi-dimensional comprehensive evaluation module is used to perform weighted scoring on the physician's ability based on the longitudinal trend data and the descriptive analysis data to obtain a physician's ability analysis report.

4. The system according to claim 1, characterized in that The training module includes: a training demand diagnosis module, a training resource matching module and a training plan optimization module; The training demand diagnosis module is used to diagnose and analyze the physician's ability needs according to the physician's ability analysis report to obtain training needs; The training resource matching module is used to match the training requirements with preset training resources to obtain a training resource list; The training plan optimization module is used to optimize the training plan according to the training requirements and the training resource list to obtain the optimized training plan.

5. The system according to claim 4, characterized in that The training plan optimization module includes: an optimization target setting module, an optimization model building module and an optimization plan generation module; The optimization target setting module is used to set the optimization target according to the training requirements; The optimization module construction module is used to construct an optimization model according to the preset training resources and the preset optimization logic of resource configuration; The optimization scheme generating module is used to input the optimization target and the training resource list into the optimization model and output the optimization training plan.

6. The system according to claim 4, characterized in that The training resource matching module: a training demand analysis module and a resource matching recommendation module; The training demand analysis module is used to analyze the training demand and obtain key demand tags; The resource matching recommendation module is used to match the key requirement tags with the preset training resources to obtain the training resource list.

Images