Image department doctor training method and system based on generative artificial intelligence

Abstract

The invention provides an imaging department doctor training method and system based on generative artificial intelligence, and relates to the technical field of artificial intelligence, and the method comprises the steps: calling a predetermined case parameterization strategy to carry out the parameterization processing of a real-time case, and obtaining a real-time parameter feature; extracting a first case in the image database, and obtaining a first parameter feature of the first case; generating a virtual case in combination with the real-time parameter features and the first parameter features, and introducing medical knowledge graph analysis to obtain a case reasonable coefficient of the virtual case; if the case reasonable coefficient is in a preset coefficient threshold value, constructing a virtual clinical environment of the virtual case; in a virtual clinical environment, image training is carried out on a target doctor in combination with a virtual case, and the problems that an existing image department doctor training system depends on fixed textbooks and standardized training cases and lacks a dynamic adjustment mechanism based on a real-time case, so that dynamic deduction and immersive interactive learning of doctors are difficult to realize, and the training efficiency is poor are solved. And the technical problem that the image diagnosis skill is difficult to improve efficiently is solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for training radiologists based on generative artificial intelligence. Background Art

[0002] In modern medical education, the training of radiologists faces numerous challenges, particularly given the rapid advancements in medical imaging technology. With the widespread adoption of advanced imaging technologies such as computed tomography (CT), magnetic resonance imaging (MRI), and digital radiography (DR), the knowledge base required of radiologists is becoming increasingly large and complex. In addition to a solid theoretical foundation, physicians must master the operation of various imaging devices, possess precise image interpretation skills, and possess the ability to make accurate clinical decisions.

[0003] However, existing training methods for radiologists have limitations. Traditional teaching models typically rely on classroom instruction and textbook learning, lacking interactivity and practical application. This approach reduces physicians' interest and initiative in learning, impacting learning outcomes. Furthermore, with the rapid advancement of medical science, traditional textbooks and training content often lag behind and fail to reflect current medical trends. The scarcity of real-life case resources and the need to protect patient privacy also make it difficult for each physician to obtain sufficient practical opportunities during training, limiting their skill development. Summary of the Invention

[0004] This application provides a method and system for training radiologists based on generative artificial intelligence, which is used to solve the technical problem that the existing radiologist training system relies on fixed textbooks and standardized training cases, lacks a dynamic adjustment mechanism based on real-time cases, making it difficult to achieve dynamic deduction and immersive interactive learning for doctors, resulting in difficulty in efficiently improving imaging diagnostic skills.

[0005] In view of the above problems, this application provides a method and system for training radiologists based on generative artificial intelligence.

[0006] In the first aspect, the present application provides a method for training radiologists based on generative artificial intelligence, the method comprising: calling a predetermined case parameterization strategy to perform parameterization processing on a real-time case to obtain real-time parameter features; extracting a first case from an imaging database and obtaining a first parameter feature of the first case; generating a virtual case by combining the real-time parameter feature with the first parameter feature, and introducing medical knowledge graph analysis to obtain a case rationality coefficient of the virtual case; if the case rationality coefficient is within a predetermined coefficient threshold, constructing a virtual clinical environment for the virtual case; and in the virtual clinical environment, performing imaging training on the target doctor in combination with the virtual case.

[0007] In the second aspect, the present application provides an radiologist training system based on generative artificial intelligence, and the system includes: a parameter processing unit: calling a predetermined case parameterization strategy to perform parameter processing on the real-time case to obtain real-time parameter features; a parameter feature acquisition unit: extracting the first case in the image database and obtaining the first parameter feature of the first case; a case rationality coefficient acquisition unit: combining the real-time parameter feature and the first parameter feature to generate a virtual case, and introducing medical knowledge graph analysis to obtain the case rationality coefficient of the virtual case; a virtual clinical environment construction unit: if the case rationality coefficient is within a predetermined coefficient threshold, constructing a virtual clinical environment for the virtual case; an image training unit: in the virtual clinical environment, performing image training on the target doctor in combination with the virtual case.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: The radiologist training method and system based on generative artificial intelligence provided in this application relate to the field of artificial intelligence technology. They solve the technical problems of limited training resources for radiologists, inability to meet personalized learning needs, and lack of interactivity and immersion. They achieve personalized and immersive training experience for radiologists, and improve learning efficiency and clinical decision-making ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A flowchart of the radiologist training method based on generative artificial intelligence is provided for this application.

[0010] Figure 2 A structural diagram of the radiologist training system based on generative artificial intelligence is provided for this application.

[0011] Explanation of the accompanying symbols: parameter processing unit 11, parameter feature acquisition unit 12, case rationality coefficient acquisition unit 13, virtual clinical environment construction unit 14, image training unit 15. DETAILED DESCRIPTION

[0012] This application provides a method and system for training radiologists based on generative artificial intelligence to solve the technical problem that the existing radiologist training system relies on fixed textbooks and standardized training cases, lacks a dynamic adjustment mechanism based on real-time cases, making it difficult to achieve dynamic deduction and immersive interactive learning for doctors, resulting in difficulty in efficiently improving imaging diagnostic skills.

[0013] Example 1, as Figure 1 As shown, the embodiment of the present application provides a method for training radiologists based on generative artificial intelligence, the method comprising: Step S100: Retrieve the parameterization strategy of the predetermined case and perform parameterization processing on the real-time case to obtain real-time parameter features.

[0014] Specifically, after the target doctor enters the radiologist training system through the login function provided by the management module, the real-time case will be parameterized. This management module is the core of the radiologist training system, responsible for the target doctor management, authority control, data maintenance and system update of the entire system. It provides each target doctor with registration and login functions to ensure that each doctor can safely access his or her training materials and progress. Back-end administrators can manage each target doctor through the management module, including adding, deleting and modifying permissions to ensure the security and confidentiality of the training content; during the parameterization process, the patient's morphological features, such as morphology, edge, density, etc., will be extracted from the real-time case, and then the predetermined feature vector quantization mechanism in the predetermined case parameterization strategy will be used to convert the extracted morphological features into a format that is easier to analyze and process, forming a real-time morphological vector; subsequently, the patient's real-time clinical feature information is converted into a real-time clinical vector using the same method, and then the real-time morphological vector and real-time clinical vector are added to the real-time parameter feature. This real-time parameter feature can describe the specific situation of the real-time case and be used for subsequent virtual case generation or clinical environment construction.

[0015] Furthermore, step S100 of the present application also includes: Step S110: Acquire the first real-time image of the real-time case in the first stage; Step S120: Perform morphological feature extraction on the first real-time image to obtain first real-time morphological feature information; Step S130: Read the predetermined feature vectorization mechanism in the predetermined case parameterization strategy, and perform vectorization processing on the first real-time morphological feature information according to the predetermined feature vectorization mechanism to obtain a first real-time morphological vector; Step S140: Construct the real-time parameter feature based on the first real-time morphological vector.

[0016] Specifically, after obtaining the real-time case, the real-time image of the first stage in the real-time case will be extracted as the first real-time image. This first stage refers to the preliminary examination of the patient. Subsequently, the first real-time image is input into the pre-constructed morphological feature convolution channel. This morphological feature convolution channel is constructed based on the convolutional neural network. During the construction process, the sample image data and the sample morphological feature data are input into the initialized convolutional neural network for forward propagation. The data is transmitted layer by layer through the input layer, convolution layer, pooling layer, fully connected layer, output layer, etc., and the prediction results including morphological features are calculated. Then, the cross entropy loss function is used. The loss value between the prediction result and the sample morphological feature data is calculated, and the gradient of the loss to the weight of each layer is calculated layer by layer through back propagation. The Adam optimizer is then used to optimize the network parameters and adjust the weights to minimize the value of the loss function. The above process is repeated until the maximum number of iterations is reached. After the training is completed, the network performance is tested using data not used for training to evaluate the accuracy of the network in the feature classification task. If the accuracy meets the preset accuracy, the current convolutional neural network is output as the final morphological feature convolution channel. Otherwise, the learning rate, number of training batches and other hyperparameters are adjusted to further improve the calculation of the convolutional neural network. According to the calculation of the morphological feature convolution channel, morphological features including morphology, edge, density, structure, distribution and other features can be obtained, and these extracted morphological features are used as the first real-time morphological feature information. This first real-time morphological feature information can help to quantify and standardize the description of the image content. Afterwards, a predetermined feature vectorization mechanism is extracted from the predetermined case parameterization strategy. This predetermined feature vectorization mechanism can convert each feature in the first real-time morphological feature information into numerical data, which has a unique hot encoding representation corresponding to each feature. For example, the circle in the morphological feature is [1,0,0,0], the ellipse is [1,0,0,0], and the ellipse is [1,0,0,0]. The circle is [0,1,0,0], the lobed shape is [0,0,1,0], the clarity of the edge features is [1,0,0,0], the blur is [0,1,0,0], and the burr shape is [0,0,1,0]. Through this predetermined feature vectorization mechanism, a one-hot encoding can be mapped for each feature in the first real-time morphological feature information, and then these one-hot encodings are spliced according to the splicing order in the predetermined feature vectorization mechanism to obtain the first real-time morphological vector. Then, this first real-time morphological vector is added to a data set as a real-time parameter feature, laying the foundation for subsequent virtual case generation or clinical environment construction.

[0017] Furthermore, step S120 further includes: Step S121: The morphological features include morphological features, edge features, density features, structural features and distribution features.

[0018] Specifically, morphological features include information on shape, edge, density, structure and distribution. Morphological features describe the geometric shape of objects in the image, including circular, elliptical, lobed and irregular shapes. Circles usually represent areas with more symmetrical structures, ellipses are slightly elongated or flat, lobed shapes indicate that the shape presents multiple branches or protrusions, and irregular shapes describe asymmetric and irregular shapes, which are usually associated with heterogeneous lesions or atypical structures. Edge features focus on the edge quality of objects in the image. Clear edges usually represent well-defined structures, which may be benign areas. Blurred edges may suggest the spread of lesions. Burr-like edges are more common in malignant tumors. Lobulated edges indicate that the structure is more complex or diffusive. These edge features help to judge the nature of the lesion and its possible degree of malignancy. Density and signal intensity features reflect the brightness or density of the area in the image. High-density / high-signal areas usually appear in calcified or hardened tissues, low-density / low-signal areas are common in fluid accumulation or cysts, equal-density / equal-signal areas indicate that the signal or density of the area is similar to that of the surrounding tissue, and mixed density Mixed signal indicates that the tissue composition of the area is uneven and is usually associated with complex lesions. Structural features further refine the material composition of the area in the image, including solid, cystic, cystic-solid, calcification, and fat. Solid areas have no obvious cavity or fluid components and are commonly seen in tumors or nodules. Cystic areas have obvious fluid components and are commonly seen in lesions such as cysts or abscesses. Cystic-solid areas are a combination of solid and fluid components and are usually seen in mixed masses or cysts. Calcification areas appear as high-density areas and usually suggest calcification deposits in certain lesions. Fatty areas have low density or low signal and are commonly seen in lesions such as lipomas. Distribution features describe the distribution of lesions or abnormal areas in the image. Single lesions appear in only one location and usually indicate a relatively limited lesion and may be a benign lesion. Multiple lesions indicate abnormalities in multiple areas and may be diffuse lesions or multiple diseases. Diffuse lesions indicate that the lesion is distributed over a wide area and are commonly seen in malignant tumors or severe inflammatory reactions. Localized lesions exist only in a certain area and usually have clear boundaries and may be early lesions or localized lesions.

[0019] Furthermore, step S140 of the present application includes: Step S141: Acquire the real-time clinical feature information of the real-time case; Step S142: Vectorize the real-time clinical feature information according to the predetermined feature vectorization mechanism to obtain a real-time clinical vector; Step S143: Add the real-time clinical vector to the real-time parameter feature.

[0020] Specifically, real-time clinical feature information is extracted from real-time cases, including basic patient information (gender, age, weight, etc.), medical history (past medical history, family medical history, etc.), physical signs (blood pressure, pulse, body temperature, respiratory rate, etc.), laboratory test results (imaging examinations, etc.), etc.; then, a predetermined feature vectorization mechanism is used to vectorize each feature in the real-time clinical feature information, that is, it is converted into a one-hot encoding representation, for example, males are mapped to [1,0], those with a history of hypertension are mapped to [1], and those without a history of hypertension are mapped to [0]; then, the one-hot encodings of these features are concatenated in a predetermined order to obtain a real-time clinical vector, and then the real-time clinical vector is added to the real-time parameter feature, so that a multi-dimensional feature vector can be constructed, which integrates imaging features and clinical information. In this way, imaging features and clinical features can be considered simultaneously in imaging training to help improve the diagnostic capabilities of complex cases.

[0021] Step S200: extracting a first case from an image database and obtaining a first parameter feature of the first case.

[0022] Specifically, a training case is randomly extracted from the image database as the first case. This image database belongs to the knowledge base module. This knowledge base module is the core of the imaging physician training system and contains rich medical imaging data and educational resources. It can provide a comprehensive medical imaging question bank, such as a practicing physician examination question bank, an imaging residency training completion assessment question bank, an imaging intermediate professional title question bank, an imaging database, etc., to meet the learning needs of doctors at different levels. In addition, since the first case is any case among the previous real cases, the real diagnosis and other records of the first case are used as the basis for the training case, which can ensure the accuracy and effectiveness of the training materials and thus improve the quality of training. After obtaining the first case, the same operation is performed on the first case in the same way as the real-time parameter features extracted from the real-time case to obtain the first parameter features, wherein the first parameter features include the clinical diagnosis and pathological diagnosis features of the first case in the past real clinical process. This first parameter feature can provide more comprehensive and personalized training data for the target doctor, helping to improve the diagnostic ability of complex cases.

[0023] Step S300: Generate a virtual case by combining the real-time parameter feature and the first parameter feature, and introduce medical knowledge graph analysis to obtain the case rationality coefficient of the virtual case.

[0024] Specifically, after obtaining the real-time parameter features and the first parameter features, the artificial intelligence module is called to randomly combine the real-time parameter features obtained from the real-time case with the first parameter features of the first case extracted from the imaging database to obtain a virtual case. This virtual case is an imaginary case generated based on existing data. It has the key features of a real case but does not rely on the data of real patients. This artificial intelligence module is also the core of the imaging physician training system, which stores the medical knowledge graph, which can verify the rationality of the randomly generated virtual cases and help generate various clinical cases that are in line with reality for the target doctors to conduct simulated diagnosis and treatment training to improve the practicality of the training. This medical knowledge graph is a A structured knowledge base contains a large amount of knowledge in the medical field, such as the relationship between diseases, symptoms, treatment methods, medical images, etc. It is constructed by extracting key entities (such as diseases, drugs, symptoms, etc.) and their relationships (such as diseases and symptoms, drugs and treatments, etc.) from medical record data obtained from various sources such as academic literature, medical textbooks, electronic medical records, and medical databases (such as PubMed, UpToDate, disease classification systems, etc.) through natural language processing technology. These extracted entities and relationships are organized into a graph structure, in which entities serve as nodes and relationships serve as edges connecting these nodes. Then a medical ontology is created, which defines the categories, attributes, and relationships between entities in the knowledge graph. Through ontology modeling, the consistency, extensibility and semantic clarity of the medical knowledge graph can be ensured. The ontology can include medical knowledge at different levels, for example, disease → symptoms → examination → treatment method; through this medical knowledge graph, it is possible to understand and analyze the characteristics of the case, determine whether these characteristics are reasonable, and further verify the case based on existing medical knowledge; during the verification process, the characteristics of the virtual case will be compared with the known medical information in the knowledge graph to verify whether the various characteristics of the virtual case are consistent with medical common sense and real-life case manifestations, and to evaluate whether these characteristics have potential inconsistencies or errors; then, based on the analysis results of the medical knowledge graph, a case rationality coefficient will be calculated for the virtual case. This coefficient reflects the rationality of the virtual case and indicates the degree of consistency between the characteristics of the virtual case and typical cases in reality. If the characteristics of the virtual case are consistent with the known standard case characteristics in the medical knowledge graph, the rationality coefficient is high, indicating that the virtual case is more reliable. If there is a large deviation in the characteristics of the virtual case, the rationality coefficient is low, indicating that the virtual case may not conform to medical laws, thereby providing more accurate training data.

[0025] Furthermore, step S300 includes: Step S310: Obtain the virtual parameter features of the virtual case; Step S320: Match the virtual parameter features in the medical knowledge graph to obtain a virtual matching result; Step S330: Determine whether the first matching result in the virtual matching result conforms to the predetermined medical logic; Step S340: When the first matching result conforms to the predetermined medical logic, add the first matching result to the reasonable list; Step S350: Take the ratio of the number of matching results in the reasonable list to the number of matching results in the virtual matching result, and record it as the case rationality coefficient.

[0026] Specifically, after obtaining the virtual case, the recombined parameter features are extracted as virtual parameter features, including the imaging features (such as morphology, edge, density, structure, etc.) and clinical features (such as medical history, physical signs, examination results, etc.) of the virtual case; then, the virtual parameter features are input into the medical knowledge graph for matching. Through the relationships and rules in the graph, the parameters of the virtual case are compared with the known case features in the graph to generate a virtual matching result. This virtual matching result contains multiple matching results, and each matching result corresponds to a virtual parameter feature. For example, the matching result of morphological features (such as round) may be benign tumors, lipomas, and adipose tissue lesions. The matching result of edge features (such as clear) may be benign tumors, lipomas, and liver cysts. The matching result of density features (such as low density) may be lipomas, cysts, and adipose tissue lesions. This virtual matching result also has a pathological identifier of the same type, that is, the virtual matching result appears twice. The pathology with the largest number, such as lipoma; then, the matching result corresponding to the first virtual parameter feature is extracted from the virtual matching results as the first matching result, and a medical logic judgment is performed on the first matching result, that is, whether the same type of pathology identifier exists in the first matching result. If so, it means that the first matching result conforms to the predetermined medical logic, and the first matching result will be added to the reasonable list. This medical logic judgment is to determine whether the case feature combination is reasonable, that is, to determine whether all virtual parameter features can point to a pathology at the same time, so as to avoid certain features that can only appear at the same time in a specific disease context and thus do not match other features; then, the number of matching results in the reasonable list is calculated by ratio with the number of matching results in the virtual matching results to obtain the case rationality coefficient. This coefficient reflects the rationality of the virtual case in medical logic. The higher the coefficient, the more consistent the virtual case is with medical common sense and the stronger the rationality, thereby further optimizing the generation and application of virtual cases.

[0027] Step S400: If the case rationality coefficient is within a predetermined coefficient threshold, a virtual clinical environment for the virtual case is constructed.

[0028] Specifically, the calculated case rationality coefficient is compared with a predetermined coefficient threshold. This threshold is a pre-set standard, which usually represents that a virtual case meets the minimum requirements of medical common sense. If the case rationality coefficient is greater than or equal to the threshold, it means that the characteristics of the virtual case are considered reasonable and meet the conditions for building a virtual clinical environment. At this time, the scene module will be called to build a virtual clinical environment. This scene module includes scenes such as the image reading room (training the ability to interpret various imaging materials), the diagnostic simulation room (training diagnostic skills), the imaging clinic room (training the ability to communicate with patients), the multidisciplinary discussion room (training the ability to collaborate in teams) and the virtual operating room (training the ability to navigate and operate images during surgery). It enables the target doctor to practice and improve his skills in a risk-free environment. In the process of using the scene module to build a virtual clinical environment for the virtual case, 3D modeling technology will be used to build an initial virtual scene, and then the virtual patient developed based on the virtual case will be imported into the initial virtual scene, thereby constructing the virtual clinical environment of the virtual case. In this virtual clinical environment, the target doctor can interact with the virtual patient for further diagnosis and treatment practice, helping the target doctor to better understand and handle the case.

[0029] Furthermore, step S400 includes: Step S410: constructing an initial virtual scene using 3D modeling technology; Step S420: developing a virtual patient for the virtual case using natural language processing and speech recognition technology; Step S430: importing the virtual patient into the initial virtual scene to generate the virtual clinical environment.

[0030] Specifically, based on the training needs of the target doctors, for example, if image interpretation skills need to be trained, an image reading room needs to be built; if diagnostic skills need to be trained, a diagnostic simulation room needs to be built, etc. Each training target corresponds to a different scenario type. According to different needs, modeling data of the corresponding scenario type is obtained from the scenario module, such as room structure data, workbench data, equipment data, etc., and 3D modeling software connected to the scenario module, such as Unity, Blender, Maya, etc., is used to build the basic three-dimensional structure of the scene based on the modeling data, such as the three-dimensional structure of the virtual room, the medical equipment involved, etc., thereby building the initial virtual scene; Subsequently, through natural language processing (NLP) and speech recognition technology, virtual patients in virtual cases are developed, allowing virtual patients to interact and communicate with doctors based on case data. In this process, data such as medical history, symptoms and signs in the virtual case will be analyzed through natural language processing technology. These data include basic information of the patient (such as age, gender) and medical history (such as hypertension, diabetes, etc.). NLP technology (such as BERT, GPT-3, Rasa, etc.) can convert this text information into a format that can be understood by the machine, allowing the virtual patient to simulate the condition feedback of the actual patient. In order to enhance the interactive experience, speech recognition technology is also combined. When the target doctor has a voice conversation with the virtual patient, speech recognition technology (such as Google Speech-to-Text or Microsoft Azure The virtual patient's responses are converted to speech using speech synthesis technology (such as Google Text-to-Speech or Amazon Polly), ensuring that the target doctor can communicate fluently with the virtual patient through voice, simulating real-world clinical communication. Once the virtual scene and virtual patient are prepared, the virtual patient is introduced into the initial virtual scene. At this point, the virtual patient is not only a static 3D model but also responds dynamically to the doctor's questions and actions, simulating a real-world clinical environment. In this virtual clinical environment, the target doctor can interact with the virtual patient and conduct diagnostic procedures such as physical examinations, medical history taking, and imaging examinations. The virtual patient can respond to the virtual case similarly to a real patient, enhancing the target doctor's clinical response to different cases. This virtual clinical environment provides a low-risk, highly interactive, and realistic training platform for the target doctor, thereby improving their clinical diagnosis and decision-making capabilities.

[0031] Step S500: In the virtual clinical environment, imaging training is performed on the target doctor in combination with the virtual case.

[0032] Specifically, in a virtual clinical environment, the target doctor can be exposed to various medical imaging data just like in a real hospital. These imaging data are determined based on virtual cases. In this virtual environment, the target doctor can interpret the images independently, including zooming in, rotating, switching between different levels of images, etc., which are very similar to the imaging tools used in real diagnosis. In addition, the virtual patient's condition, medical history, physical signs and other information can also be provided simultaneously to help doctors combine other clinical information of the patient when analyzing the images to make a more comprehensive diagnosis. In this way, the target doctor can not only practice how to interpret imaging data without actual risks, but also practice repeatedly when encountering complex cases. The images of each virtual case are highly personalized and diverse, simulating different types of diseases and pathological changes, giving the target doctor the opportunity to be exposed to various typical and atypical imaging cases in a virtual environment. After the training is completed, the training effect will be evaluated to determine the target doctor's diagnosis and treatment coefficient in this training, and the corresponding experience value will be determined based on this diagnosis and treatment coefficient (positive or negative, depending on whether the diagnosis and treatment coefficient meets the standard). This experience value The level assessment module will be used to determine the level of the target doctor. This level assessment module is a key component of the radiologist training system. The module divides users into five stages: novice, advanced novice, competent, proficient and expert. Each stage has clear evaluation criteria and training objectives. In addition, the stage level determined by the level assessment module will be used to activate the corresponding function in the game task module. The game task module is an important part of the radiologist training system to improve learning interest and participation. It can provide multiple game task modes such as single-player practice (focusing on improving personal skills), multiplayer battles (focusing on competition between doctors), team tasks (focusing on cooperation between doctors) and case analysis (focusing on solving practical problems), thereby enhancing the training effect. This game task module will also be connected to the voice module. This voice module can provide rich audio feedback and interactive experience, including game background music, operation music, voice prompts, voice rewards, user voice recognition and user voice dialogue, etc. Through these audio elements, doctors can get real-time feedback and motivation during the training process, enhancing the immersion and interactivity of learning. In summary, this virtual imaging training process can help doctors improve their accuracy and efficiency in interpreting imaging data, while enhancing their confidence in making decisions in clinical work. Through repeated practice in a virtual environment, doctors' image interpretation ability has been enhanced and can be better applied in real clinical practice.

[0033] Furthermore, step S500 further includes: Step S510: Obtain target diagnosis and treatment data of the target doctor for diagnosing and treating the virtual case in the virtual clinical environment; Step S520: Compare the target diagnosis and treatment data with the predetermined diagnosis and treatment data to obtain a target diagnosis and treatment coefficient, wherein the target diagnosis and treatment coefficient is used to characterize the diagnosis and treatment level of the target doctor and the training effect on the target doctor.

[0034] Specifically, in the virtual clinical environment, the target doctor will conduct diagnosis and treatment based on the specific circumstances of the virtual case, which includes the doctor conducting a comprehensive clinical assessment of the virtual patient, asking about the medical history, reviewing imaging materials, performing physical examinations, proposing diagnostic hypotheses, selecting relevant examination items, etc. Through interaction in the virtual clinical environment, the data generated by the doctor during the diagnosis and treatment process will be automatically recorded, including diagnostic results, selected examination items, treatment plans, etc. These data constitute the target diagnosis and treatment data, which are used to reflect the doctor's diagnosis and treatment process in the virtual environment; then, the target doctor's diagnosis and treatment data will be compared with the predetermined diagnosis and treatment data. The predetermined diagnosis and treatment data is the reference data set based on the optimal historical diagnosis of the virtual case. These data are verified by medical experts or standard treatment guidelines. By inputting the target diagnosis and treatment data and the predetermined diagnosis and treatment data together into the training effect evaluation model built into the grade evaluation module, this training effect evaluation model can be based on the multi-layer perceptron regression model (MLP Regressor), and the training method is the same as the above, which is carried out through steps such as forward propagation, loss calculation, back propagation, and parameter optimization. The difference lies in the model structure and training data. The model structure used here is input layer, hidden layer, and output layer, and the training data used are sample diagnosis and treatment data, sample scheduled diagnosis and treatment data, and sample diagnosis and treatment coefficient. Through the mapping calculation of the training effect evaluation model, the target diagnosis and treatment coefficient of the target doctor in this training can be obtained. This target diagnosis and treatment coefficient is used to measure the diagnosis and treatment level and training effect of the target doctor, and reflects the accuracy of the doctor's judgment in a specific case. , rationality and the ability to execute medical processes; then, calculate the difference between the target diagnosis and treatment coefficient and the diagnosis and treatment pass coefficient, and then calculate the ratio of the difference to the diagnosis and treatment pass coefficient, and multiply this ratio by the fixed experience value to obtain the experience value that can be obtained in this training, which is used for the grade evaluation module to determine the grade of the target doctor; in addition, when the calculated difference is a negative number, it means that there is an abnormality in this training. At this time, the data in the target diagnosis and treatment data that is significantly different from the predetermined diagnosis and treatment data will be marked and fed back to the target doctor to help the doctor find deficiencies and improve the diagnostic level, thereby helping the target doctor to continuously improve in clinical practice.

[0035] Furthermore, step S520 further includes: Step S521: Activate the intelligent auxiliary diagnosis platform; Step S522: Perform image recognition on any image of the virtual case through the abnormality recognition and annotation module in the intelligent auxiliary diagnosis platform, and annotate to obtain a virtual lesion node set; Step S523: Perform reasoning analysis on the first virtual lesion node in the virtual lesion node set through the lesion reasoning suggestion module in the intelligent auxiliary diagnosis platform to obtain a first diagnostic suggestion; Step S524: Organize the predetermined diagnosis and treatment data based on the first diagnostic suggestion.

[0036] Specifically, during the image training process, the intelligent auxiliary diagnosis platform is activated, which includes an abnormality recognition and annotation module and a lesion reasoning and suggestion module to help determine the target doctor's scheduled diagnosis and treatment data corresponding to the virtual case in this training; after the platform is started, the abnormality recognition and annotation module is called, which is constructed by the convolutional neural network in deep learning, and can automatically process the image data of the virtual case. The construction method of this abnormality recognition and annotation module is the same as the above, and will not be repeated here; this abnormality recognition and annotation module will scan the key areas in the image, automatically identify potential abnormal features, such as masses, lesions or other abnormal signals, and accurately annotate them on the image to generate a virtual lesion node set containing multiple abnormal annotations. Each lesion node contains specific image feature information, such as location, morphology, boundary characteristics, etc.; then, the abnormality recognition and annotation module is called. Using the lesion reasoning and suggestion module, each lesion node in the virtual lesion node set is subjected to reasoning and analysis one by one. Taking the first virtual lesion node as an example, the key imaging feature parameters of the node (such as morphology, density, edge characteristics, etc.) are extracted and matched with the image database. The reasoning and analysis module combines the historical data in the image database for comprehensive analysis to generate the first diagnostic suggestion for the lesion node, including potential disease diagnosis, diagnostic suggestions, and recommendations for further examinations. Afterwards, the first diagnostic suggestion provided by the reasoning and analysis module and the diagnostic suggestions of other virtual lesion nodes are added to a data set to form predetermined diagnosis and treatment data that meets medical standards. These data not only include diagnostic conclusions, but also recommended examination items, treatment plans, etc., forming a systematic diagnosis and treatment reference to provide support for the training effect evaluation and feedback of target doctors.

[0037] Furthermore, step S523 further includes: Step S5231: Collect the first node feature parameters of the first virtual lesion node; Step S5232: Traverse the first node feature parameters in the lesion node database to obtain a first traversal result; wherein, the lesion node database refers to a database based on any historical lesion node and any corresponding historical diagnosis in the image database; Step S5233: Compare multiple historical diagnoses in the first traversal result and determine the first optimal historical diagnosis; Step S5234: Use the first optimal historical diagnosis as the first diagnostic recommendation.

[0038] Specifically, a first virtual lesion node is identified from the virtual case image. The node is a specific lesion area in the image, such as a tumor, inflammation or other lesions, and the node features of the first virtual lesion node are extracted to obtain first node feature parameters, including but not limited to morphological features, edge features, density, etc.; then, the first node feature parameters are matched with historical lesion features in the lesion node database. The lesion node database is a database containing a large number of historical cases and corresponding lesion nodes, wherein each historical case includes the feature parameters of the lesion node and the corresponding historical diagnosis information. This database is constructed based on the historical lesion nodes in the image database and their corresponding historical diagnosis results. By traversing all lesion nodes in the database, according to the feature phase, the first node feature parameters are matched. Similarity (such as Hamming distance, Jaccard similarity, etc.) finds historical lesion nodes with similar features to the first node, and returns multiple historical case information matching the first node features (that is, the matching degree is greater than or equal to the preset matching degree), forming the first traversal result. This first traversal result includes the lesion node features of each historical case and the corresponding diagnosis results. Taking the Hamming distance as an example, the circle is [1,0,0,0] and the ellipse is [0,1,0,0], then the Hamming distance is 2, because the two vectors are different in the first and second positions; then, the one with the largest matching degree is extracted from multiple historical diagnoses in the first traversal result as the first optimal historical diagnosis, and this first optimal historical diagnosis is used as the final first diagnosis recommendation to help doctors improve diagnostic accuracy and efficiency.

[0039] To sum up, the radiologist training method based on generative artificial intelligence provided by the embodiment of the present application includes at least the following technical effects, which realizes a personalized and immersive training experience for radiologists and improves learning efficiency and clinical decision-making ability.

[0040] Example 2, based on the same inventive concept as the method for training radiologists based on generative artificial intelligence in the above-mentioned embodiment, Figure 2As shown, the present application provides an radiologist training system based on generative artificial intelligence, and the system includes: a parameter processing unit 11: calling a predetermined case parameterization strategy to perform parameter processing on a real-time case to obtain a real-time parameter feature; a parameter feature acquisition unit 12: extracting a first case from an image database, and obtaining a first parameter feature of the first case; a case rationality coefficient acquisition unit 13: combining the real-time parameter feature with the first parameter feature to generate a virtual case, and introducing a medical knowledge graph analysis to obtain a case rationality coefficient of the virtual case; a virtual clinical environment construction unit 14: if the case rationality coefficient is within a predetermined coefficient threshold, constructing a virtual clinical environment for the virtual case; an image training unit 15: in the virtual clinical environment, performing image training on the target doctor in combination with the virtual case.

[0041] Furthermore, the system also includes: A diagnosis and treatment data acquisition unit is configured to acquire target diagnosis and treatment data of the target doctor diagnosing and treating the virtual case in the virtual clinical environment; a diagnosis and treatment data comparison unit is configured to compare the target diagnosis and treatment data with predetermined diagnosis and treatment data to obtain a target diagnosis and treatment coefficient, wherein the target diagnosis and treatment coefficient is configured to characterize the diagnosis and treatment level of the target doctor and the training effect on the target doctor.

[0042] Furthermore, the system also includes: Platform activation unit: activates the intelligent auxiliary diagnosis platform; image recognition unit: performs image recognition on any image of the virtual case through the abnormality recognition and annotation module in the intelligent auxiliary diagnosis platform, and annotates it to obtain a virtual lesion node set; reasoning and analysis unit: performs reasoning and analysis on the first virtual lesion node in the virtual lesion node set through the lesion reasoning suggestion module in the intelligent auxiliary diagnosis platform to obtain a first diagnostic suggestion; scheduled diagnosis and treatment data formation unit: forms the scheduled diagnosis and treatment data based on the first diagnostic suggestion.

[0043] Furthermore, the system also includes: A feature parameter collection unit is configured to collect the first node feature parameters of the first virtual lesion node; a database traversal unit is configured to traverse the first node feature parameters in the lesion node database to obtain a first traversal result; wherein the lesion node database refers to a database formed based on any historical lesion node and any corresponding historical diagnosis in the image database; a historical diagnosis comparison unit is configured to compare multiple historical diagnoses in the first traversal result and determine a first optimal historical diagnosis; a diagnosis suggestion acquisition unit is configured to use the first optimal historical diagnosis as the first diagnosis suggestion.

[0044] Furthermore, the system also includes: A real-time image acquisition unit is configured to acquire a first real-time image of the real-time case in the first stage; a morphological feature extraction unit is configured to perform morphological feature extraction on the first real-time image to obtain first real-time morphological feature information; a vectorization processing unit is configured to read a predetermined feature vectorization mechanism in the predetermined case parameterization strategy, and to perform vectorization processing on the first real-time morphological feature information according to the predetermined feature vectorization mechanism to obtain a first real-time morphological vector; a real-time parameter feature construction unit is configured to construct the real-time parameter feature based on the first real-time morphological vector.

[0045] Furthermore, the system also includes: The morphological features include morphological features, edge features, density features, structural features and distribution features.

[0046] Furthermore, the system also includes: A clinical feature information acquisition unit is configured to acquire the real-time clinical feature information of the real-time case; a second vectorization processing unit is configured to vectorize the real-time clinical feature information according to the predetermined feature vectorization mechanism to obtain a real-time clinical vector; and a clinical vector adding unit is configured to add the real-time clinical vector to the real-time parameter feature.

[0047] Furthermore, the system also includes: A virtual parameter feature acquisition unit is used to obtain the virtual parameter features of the virtual case; a virtual parameter feature matching unit is used to match the virtual parameter features in the medical knowledge graph to obtain a virtual matching result; a matching result judgment unit is used to judge whether the first matching result in the virtual matching result conforms to the predetermined medical logic; a matching result adding unit is used to add the first matching result to the reasonable list when the first matching result conforms to the predetermined medical logic; a quantity ratio calculation unit is used to take the ratio of the number of matching results between the reasonable list and the virtual matching result, and record it as the case reasonable coefficient.

[0048] Furthermore, the system also includes: An initial virtual scene construction unit is used to construct an initial virtual scene using 3D modeling technology; a virtual patient development unit is used to develop a virtual patient for the virtual case using natural language processing and speech recognition technology; and a virtual clinical environment generation unit is used to import the virtual patient into the initial virtual scene to generate the virtual clinical environment.

[0049] Through the above detailed description of the radiologist training method based on generative artificial intelligence in this specification, those skilled in the art can clearly understand the radiologist training system based on generative artificial intelligence in this embodiment. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0050] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for training radiologists based on generative artificial intelligence, characterized by: include: Retrieve the parameterization strategy of the scheduled case to perform parameterization processing on the real-time case to obtain the real-time parameter characteristics; Extracting a first case from an image database and obtaining a first parameter feature of the first case; Combining the real-time parameter feature with the first parameter feature to generate a virtual case, and introducing medical knowledge graph analysis to obtain a case rationality coefficient of the virtual case; If the case rationality coefficient is within a predetermined coefficient threshold, constructing a virtual clinical environment for the virtual case; In the virtual clinical environment, imaging training is performed on target doctors in combination with the virtual cases.

2. The method for training radiologists based on generative artificial intelligence according to claim 1, characterized in that: In the virtual clinical environment, imaging training is performed on the target doctor in combination with the virtual case, and then the following steps are further included: Acquire target diagnosis and treatment data of the target doctor diagnosing and treating the virtual case in the virtual clinical environment; The target diagnosis and treatment data are compared with the predetermined diagnosis and treatment data to obtain a target diagnosis and treatment coefficient, wherein the target diagnosis and treatment coefficient is used to characterize the diagnosis and treatment level of the target doctor and the training effect on the target doctor.

3. The method for training radiologists based on generative artificial intelligence according to claim 2, characterized in that: Comparing the target diagnosis and treatment data with the predetermined diagnosis and treatment data to obtain a target diagnosis and treatment coefficient includes: Activate the intelligent auxiliary diagnosis platform; Performing image recognition on any image of the virtual case through the abnormality recognition and annotation module in the intelligent auxiliary diagnosis platform, and annotating and obtaining a virtual lesion node set; Performing reasoning and analysis on the first virtual lesion node in the virtual lesion node set by the lesion reasoning and suggestion module in the intelligent auxiliary diagnosis platform to obtain a first diagnosis suggestion; The predetermined diagnosis and treatment data is constructed based on the first diagnosis suggestion.

4. The method for training radiologists based on generative artificial intelligence according to claim 3, characterized in that: Performing reasoning and analysis on the first virtual lesion node in the virtual lesion node set by the lesion reasoning and suggestion module in the intelligent auxiliary diagnosis platform to obtain a first diagnostic suggestion includes: collecting first node characteristic parameters of the first virtual lesion node; Traversing the first node characteristic parameter in the diseased node database to obtain a first traversal result; The lesion node database refers to a database based on any historical lesion node and any corresponding historical diagnosis in the image database; Comparing multiple historical diagnoses in the first traversal result and determining a first optimal historical diagnosis; The first optimal historical diagnosis is used as the first diagnosis suggestion.

5. The method for training radiologists based on generative artificial intelligence according to claim 1, characterized in that: Retrieve the parameterization strategy of the scheduled case and perform parameterization processing on the real-time case to obtain real-time parameter features, including: Acquire a first real-time image of the real-time case in the first stage; performing morphological feature extraction on the first real-time image to obtain first real-time morphological feature information; Reading a predetermined feature vectorization mechanism in the predetermined case parameterization strategy, and performing vectorization processing on the first real-time morphological feature information according to the predetermined feature vectorization mechanism to obtain a first real-time morphological vector; The real-time parameter feature is constructed based on the first real-time morphological vector.

6. The method for training radiologists based on generative artificial intelligence according to claim 5, characterized in that: The morphological features include morphological features, edge features, density features, structural features and distribution features.

7. The method for training radiologists based on generative artificial intelligence according to claim 5, characterized in that: The method further comprises: forming the real-time parameter feature based on the first real-time morphological vector; Obtaining real-time clinical characteristic information of the real-time case; performing vectorization processing on the real-time clinical feature information according to the predetermined feature vectorization mechanism to obtain a real-time clinical vector; The real-time clinical vector is added to the real-time parameter feature.

8. The method for training radiologists based on generative artificial intelligence according to claim 1, characterized in that: Combining the real-time parameter feature with the first parameter feature to generate a virtual case, and introducing medical knowledge graph analysis to obtain a case rationality coefficient of the virtual case, including: Obtaining virtual parameter characteristics of the virtual case; Matching the virtual parameter features in the medical knowledge graph to obtain a virtual matching result; Determining whether a first matching result in the virtual matching results conforms to a predetermined medical logic; When the first matching result conforms to the predetermined medical logic, adding the first matching result to a reasonable list; The ratio of the number of matching results in the reasonable list to the number of matching results in the virtual matching result is taken as the case rationality coefficient.

9. The method for training radiologists based on generative artificial intelligence according to claim 1, characterized in that: Constructing a virtual clinical environment for the virtual case, including: Use 3D modeling technology to build the initial virtual scene; developing a virtual patient for the virtual case using natural language processing and speech recognition technology; The virtual patient is introduced into the initial virtual scene to generate the virtual clinical environment.

10. A radiologist training system based on generative artificial intelligence, characterized by: The system is used to execute the method for training radiologists based on generative artificial intelligence according to any one of claims 1 to 9, comprising: Parameterized processing unit: calls the parameterized strategy of the predetermined case to perform parameterized processing on the real-time case to obtain the real-time parameter characteristics; A parameter feature acquisition unit is configured to extract a first case from an image database and acquire a first parameter feature of the first case; A case rationality coefficient acquisition unit: generates a virtual case by combining the real-time parameter feature and the first parameter feature, and obtains the case rationality coefficient of the virtual case by introducing medical knowledge graph analysis; A virtual clinical environment construction unit: if the case rationality coefficient is within a predetermined coefficient threshold, constructing a virtual clinical environment for the virtual case; Image training unit: in the virtual clinical environment, image training is performed on target doctors in combination with the virtual cases.

Images