Simulation Teaching and Management System Based on Medical Imaging Archives of Real Medical Records in Radiology Department

Abstract

The present application provides a simulation teaching and management system based on medical imaging archives of real cases in the radiology department, including: an archive classification module for collecting medical imaging archives of real cases in the radiology department and classifying them; a sample extraction module for extracting multiple training samples from each type of medical imaging archives of real cases in the radiology department; a model training module for training each training sample according to the decision tree algorithm; a model combination module for combining multiple classifiers to obtain a simulation teaching and management model when the training results all meet the standards; a solution formulation module for inputting the learning labels of students into the simulation teaching and management model to formulate a simulation teaching and management solution; and a learning data evaluation module for recording the learning data of students completing the simulation teaching and management solution and evaluating the learning situation of students. The present invention realizes customizing personalized teaching solutions for each student and automatically evaluating the learning effects.

Description

Technical Field

[0001] This application relates to the technical field of intelligent teaching and management of medical image archives. Specifically, this application relates to a simulation teaching and management system based on medical image archives of real medical records in the radiology department. Background Art

[0002] The PACS (Picture Archiving and Communication System), that is, the image storage and transmission system, is a comprehensive application system for collecting, storing, managing, diagnosing, and processing digital medical image information generated by digital medical devices in hospitals.

[0003] The cases in the PACS system are all based on the real world, with clinical medical records having complete clinical and imaging data. After being reviewed by the scientific research and education department and the radiology teaching and research section of the hospital and anonymized by the system, they are included in the teaching PACS system according to disease systems, teaching purposes, and difficulty coefficients. The purpose of the review by the teaching and research section is to make all cases meet the requirements of the radiology residency training syllabus and be typical and / or rare cases with certain teaching purposes, which have a better teaching effect than general clinical cases. Trainees have their own PACS accounts and, in addition to completing their daily work, complete the teaching plans formulated by the PACS system. However, the vast majority of existing PACS systems are clinical work systems and cannot formulate teaching plans for each trainee according to classification methods such as the teaching syllabus, disease systems, or difficulty levels, and cannot meet the needs of trainees of different natures and levels.

[0004] For example, in the technical solution with the patent application number 202311182717.8, an internship function is incorporated into the original formal PACS system, an internship account is established and corresponding permissions are assigned. After the inspection data enters the system, it can be operated by both formal physicians and interns at the same time. After logging into the formal PACS system through the internship account, the intern writes a practice report. After the completed report is set with an internship label, it is stored separately. The formal doctor enters the formal PACS system, writes a formal report and sets a formal label and stores it in the original path. The reviewing doctor conducts the review in the formal PACS system by distinguishing through labels; however, this PACS system cannot automatically formulate suitable teaching plans for trainees according to different teaching staff (interns, clinical residency trainees, radiology residency personnel, ultrasound residency personnel), the difficulty of teaching medical records, and the disease system, with clear requirements, purposes, and difficulties, and has a low level of intelligence.

[0005] In the technical solution with the patent application number 202211268259.5, it obtains medical imaging data through the management terminal and generates teaching videos for scientific research personnel to conduct scientific research and for learning personnel to study. It conducts learning of teaching videos through the learning terminal and conducts scientific research for scientific research personnel through the scientific research terminal. Thus, it can provide references for teaching and scientific research through the medical imaging data in the hospital PACS database, which is convenient to use and improves the learning efficiency of medical staff. However, it cannot achieve the purpose of in-service training in the industry, still cannot customize personalized teaching plans for each trainee, and cannot meet the needs of different trainees. Summary of the Invention

[0006] The main purpose of this application is to provide a simulation teaching and management system based on the medical imaging archives of real cases in the radiology department, so as to customize personalized teaching plans for each trainee by using the simulation teaching and management model, and realize the personalized customization of teaching content and methods.

[0007] To achieve the above invention purpose, this application provides a simulation teaching and management system based on the medical imaging archives of real cases in the radiology department, including:

[0008] An archive classification module, used to collect medical imaging archives of real cases in the radiology department, classify the medical imaging archives of real cases in the radiology department, and obtain various types of medical imaging archives of real cases in the radiology department;

[0009] A sample extraction module, used to extract multiple training samples from each type of the medical imaging archives of real cases in the radiology department, and each training sample includes a medical imaging of a real case in the radiology department and the corresponding reference diagnosis result;

[0010] A model training module, used to train multiple training samples of each type according to the decision tree algorithm respectively to obtain multiple classifiers and the corresponding training results;

[0011] A model combination module, used to combine multiple classifiers to obtain a simulation teaching and management model when it is determined that the training results of each classifier meet the standards;

[0012] A plan formulation module, used to obtain the trainee's file, determine the learning label of the trainee based on the trainee's file, input the learning label of the trainee into the simulation teaching and management model, use the simulation teaching and management model to formulate a simulation teaching and management plan for the trainee, and send the simulation teaching and management plan to the trainee;

[0013] A learning data evaluation module, used to record the learning data of the trainee completing the simulation teaching and management plan, and evaluate the learning situation of the trainee according to the learning data.

[0014] Further, the system further includes a diagnostic report evaluation module for:

[0015] Obtain a preset number of diagnostic reports written by the trainee;

[0016] Use the simulation teaching and management model to evaluate each of the diagnostic reports to obtain an evaluation value for each of the diagnostic reports;

[0017] Compare the evaluation value of each diagnostic report with a preset evaluation value corresponding to the diagnostic report type, and calculate the proportion of diagnostic reports whose evaluation value is greater than the preset evaluation value;

[0018] When the proportion is greater than a preset proportion, determine that the simulation teaching quality of the trainee meets the standard and generate an evaluation report;

[0019] Randomly send the evaluation report to the tutor of the trainee for revision. After the tutor completes the revision, send the evaluation report to the teaching and research section for review and archiving.

[0020] Further, the system further includes a trainee assessment module for:

[0021] When the number of diagnostic reports written by the trainee reaches a preset number, determine the assessment target, trainee grade, and practice record of the trainee;

[0022] Randomly generate assessment questions according to the assessment target, trainee grade, and practice record, and set the completion time of the assessment questions;

[0023] Assess the trainee according to the assessment questions and the completion time to generate an assessment result;

[0024] Adjust the simulation teaching and management plan of the trainee according to the assessment result.

[0025] Further, the system further includes a model update module for:

[0026] Regularly collect medical image files of real cases in the radiology department to obtain candidate medical image files of real cases in the radiology department;

[0027] After classifying the candidate medical image files of real cases in the radiology department, input them into the simulation teaching and management model to iteratively update the simulation teaching and management model.

[0028] Preferably, the learning data evaluation module is further used for:

[0029] Perform data preprocessing on the learning data, and screen out data related to learning behaviors from the learning data after data preprocessing to obtain target learning data;

[0030] Convert the target learning data into a transaction list, where each transaction in the transaction list contains a series of items;

[0031] Perform frequency statistics on the items in each of the transactions and sort them in descending order of frequency;

[0032] Insert each of the transactions into a pre-constructed frequent pattern tree in the order of the frequency of the items. If an item in the transaction already exists in the frequent pattern tree, increase the count of the transaction; otherwise, create a new node;

[0033] Recursively mine frequent item sets from the frequent pattern tree until no more frequent item sets can be mined from the frequent pattern tree;

[0034] Generate association rules from the frequent item sets and check the relationship between each subset of the frequent item sets and the entire item set, calculate the confidence and lift. The confidence refers to the conditional probability that the consequent appears in the transactions where the antecedent appears, and the lift refers to the ratio of the probability that the consequent appears given the antecedent to the probability that the consequent appears by itself;

[0035] Evaluate the effectiveness of the association rules based on the confidence and lift. After passing the evaluation, analyze the learning data according to the association rules and generate a learning evaluation report for the trainee.

[0036] Preferably, the data preprocessing methods include removing noise, filling in missing values, and data normalization processing.

[0037] Preferably, the model combination module is further used for:

[0038] Based on the training results of each classifier, calculate the loss value of each classifier respectively based on a preset cross-entropy loss function;

[0039] Judge whether the loss value of each classifier is lower than a preset loss value;

[0040] If so, determine that the training results of each classifier are all qualified;

[0041] If not, screen out the classifiers with unqualified training results to obtain candidate classifiers;

[0042] Use the stochastic gradient descent method to adjust the parameters of the candidate classifiers;

[0043] Randomly select multiple training samples of one type to retrain the candidate classifiers with adjusted parameters until the loss value of the candidate classifiers is lower than the preset loss value to obtain classifiers with all qualified training results.

[0044] Preferably, the model combination module is further configured to:

[0045] Randomly select a training sample from multiple said training samples to obtain a target training sample;

[0046] Calculate the gradient of each original parameter of the candidate classifier under the condition of the target training sample according to the cross-entropy loss function;

[0047] Calculate the difference between the gradient of each original parameter with respect to the loss value of the candidate classifier and the gradient of each corresponding original parameter on the target training sample to obtain a gradient difference;

[0048] Calculate the difference between the logarithm of the gradient difference and the logarithm of a reference gradient difference to obtain a target gradient difference, where the reference gradient difference is 0.18;

[0049] Continuously adjust the original parameters of the candidate classifier in the negative gradient direction according to the target gradient difference.

[0050] Preferably, the preset cross-entropy loss function includes:

[0051]

[0052] where N is the number of training samples, i is the i-th training sample, e is the base of the natural logarithm, k is the diagnostic result predicted by the i-th training sample, and y i,k is the vector corresponding to the diagnostic result k of the i-th training sample, and p i,k is the probability that the diagnostic result k predicted by the i-th training sample is the reference diagnostic result.

[0053] Furthermore, the system further includes a probability calculation module, which is configured to:

[0054] Vectorize the diagnostic result k predicted by the i-th training sample to obtain a first vector;

[0055] Query the reference diagnostic result corresponding to the i-th training sample, and vectorize the reference diagnostic result corresponding to the i-th training sample to obtain a second vector;

[0056] Calculate the Euclidean distance between the first vector and the second vector to obtain the probability that the diagnostic result k predicted by the i-th training sample is the reference diagnostic result.

[0057] A simulation teaching and management system based on real medical record medical image files in the radiology department provided by this application. The file classification module classifies the real medical record medical image files in the radiology department through automation technology, improving the efficiency and accuracy of classification; the sample extraction module automatically extracts training samples from the classified medical image files, reducing the workload of manually selecting samples and ensuring the representativeness and diversity of the samples; the model training module uses the decision tree algorithm to train the extracted training samples to generate multiple classifiers to process a large amount of data and can learn complex patterns from the data; the model combination module combines them into a comprehensive simulation teaching and management model after confirming that the training results of each classifier meet the standards. This integration method can improve the generalization ability and prediction accuracy of the model; the solution formulation module customizes personalized teaching plans for each student using the simulation teaching and management model according to the learning files and labels of the students, realizing the personalization of teaching content and methods to meet the needs of different students. The learning data evaluation module records the learning data of the students and evaluates the learning situation of the students based on these data, providing feedback for teaching to help teachers and students understand the learning progress and effects. The entire system improves the efficiency of medical image teaching and management through automated and intelligent methods, reduces the need for manual operations, and improves the quality of decision-making through data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 FIG. is a schematic structural diagram of a simulation teaching and management system based on real medical record medical image files in the radiology department according to an embodiment of this application;

[0059] Figure 2 FIG. is a schematic structural diagram of a simulation teaching and management system based on real medical record medical image files in the radiology department according to another embodiment of this application;

[0060] Figure 3 FIG. is a schematic structural diagram of a simulation teaching and management system based on real medical record medical image files in the radiology department according to another embodiment of this application;

[0061] Figure 4 FIG. is a schematic structural diagram of a simulation teaching and management system based on real medical record medical image files in the radiology department according to another embodiment of this application.

[0062] The realization, functional characteristics, and advantages of the purpose of this application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] In order to make the purpose, technical solutions, and advantages of this application clearer, the following further details this application in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0064] Reference Figure 1 As shown, in one of the embodiments, the present application provides a simulation teaching and management system based on radiology real medical record medical image files. The system includes:

[0065] An archive classification module 11, configured to collect radiology real medical record medical image files, classify the radiology real medical record medical image files, and obtain various types of radiology real medical record medical image files;

[0066] A sample extraction module 12, configured to extract a plurality of training samples from each type of the radiology real medical record medical image files respectively. Each training sample includes a radiology real medical record medical image and a corresponding reference diagnosis result;

[0067] A model training module 13, configured to train each type of the plurality of training samples respectively according to a decision tree algorithm to obtain a plurality of classifiers and corresponding training results;

[0068] A model combination module 14, configured to combine the plurality of classifiers to obtain a simulation teaching and management model when it is determined that the training results of each classifier meet the standards;

[0069] A solution formulation module 15, configured to obtain a student file of a student, determine a learning label of the student based on the student file, input the learning label of the student into the simulation teaching and management model, formulate a simulation teaching and management solution for the student by using the simulation teaching and management model, and send the simulation teaching and management solution to the student;

[0070] A learning data evaluation module 16, configured to record learning data of the student for completing the simulation teaching and management solution, and evaluate the learning situation of the student according to the learning data.

[0071] In this embodiment, the archive classification module 11 of the system collects real medical record medical image files from the radiology department, such as CT, MRI, X-ray, etc. At the same time, the image files are automatically classified according to features such as disease type, body part, etc., and can also be automatically classified according to examination plans, disease systems to which they belong, assessment difficulty, assessor classification, etc. And all the system cases are anonymized, error points are modified, it is verified whether the cases are available or typical, the difficulty coefficient is scored (easy - medium - difficult), and the real-time update and maintenance of the medical records in the database are carried out, etc. Finally, various types of radiology real medical record medical image files are obtained.

[0072] From each category, for standardized training students of different grades and natures, the system's sample extraction module 12 can extract multiple training samples respectively according to the classification mode of the disease system and difficulty. Each training sample includes a medical image of a real case in the radiology department and the corresponding reference diagnosis result, which includes evaluations such as the evaluation of the report writing content, the evaluation of the diagnostic accuracy, and the evaluation of whether the differential diagnosis is complete, etc., for training a machine learning model to identify and diagnose specific diseases.

[0073] The model training module 13 uses the decision tree algorithm to iteratively train multiple training samples of each type. After the training is completed, multiple classifiers are generated. The decision tree algorithm is a machine learning algorithm that mimics the human decision-making process. It predicts the classification or result of unknown data by learning data features and decision rules. The decision tree shows the decision-making process in the form of a tree diagram, where each internal node represents a test on a feature, each branch represents the result of the test, and each leaf node represents a decision result or classification. In addition, the decision tree algorithm constructs a model by learning the relationship between image features and diagnostic results.

[0074] When the training results of all classifiers reach the predetermined accuracy standard, the model combination module 14 combines all classifiers into a comprehensive simulation teaching and management model. For example, different weights are assigned to each classifier, and the weights are set based on the performance or prediction accuracy of the classifier, and then all classifiers are spliced according to the weight level. Among them, the simulation teaching and management model is used to formulate a simulation teaching and management plan for students based on real cases.

[0075] The system's plan formulation module 15 obtains the student's file information, including learning content, learning duration, student level, etc., and determines the student's learning labels, such as learning style, knowledge level, etc. According to the learning labels, the system uses the simulation teaching and management model to customize a personalized simulation teaching and management plan for each student, and sends the simulation teaching and management plan to the student, the full-time tutor and the teaching management staff. At the same time, the system conducts a preliminary evaluation of the simulation teaching and management plan through AI and generates an AI preliminary evaluation report. The simulation teaching and management plan includes simulation cases, simulation videos, exercises and learning plans, etc.

[0076] The system's learning data evaluation module 16 records the interactions and performances of students in simulation teaching activities, including learning data such as diagnostic accuracy rate and learning time, evaluates the learning situation of students using the learning data, generates an evaluation result, and uses the evaluation result to formulate a suitable learning plan for the student. The learning data is used to evaluate the learning progress and effect of students in order to further optimize the teaching plan.

[0077] For example, if a radiology resident wants to improve their ability to diagnose lung nodules, the system first collects the hospital's lung CT image archives from the past few years and classifies them into two categories: "benign nodules" and "malignant nodules". From the "malignant nodules" category, the system extracts 200 CT images and their diagnostic results as training samples. Using the decision tree algorithm, the system trains a classifier capable of identifying malignant lung nodules. This classifier is combined with other disease classifiers in the system to form a comprehensive lung disease diagnosis model. The system analyzes the resident's learning profile and finds that they need to improve in identifying lung nodules. Therefore, the system develops a teaching plan focused on lung nodule identification for them. After completing a series of simulated diagnostic training, the system records the physician's diagnostic accuracy rate and learning time. The evaluation results show that the physician's diagnostic ability has been significantly improved. Through this technical solution, radiologists can improve their diagnostic skills through a learning environment that simulates real cases, while teaching managers can evaluate and optimize teaching effectiveness based on learning data.

[0078] A simulation teaching and management system based on real medical image archives of radiology departments provided by this application. The archive classification module classifies real medical image archives of radiology departments through automation technology, improving the efficiency and accuracy of classification; the sample extraction module automatically extracts training samples from the classified medical image archives, reducing the workload of manually selecting samples and ensuring the representativeness and diversity of the samples; the model training module uses the decision tree algorithm to train the extracted training samples to generate multiple classifiers to process large amounts of data and can learn complex patterns from the data; the model combination module combines them into a comprehensive simulation teaching and management model after confirming that the training results of each classifier meet the standards. This integration method can improve the generalization ability and prediction accuracy of the model; the plan formulation module customizes personalized teaching plans for each student using the simulation teaching and management model based on the student's learning profile and labels, realizing the personalization of teaching content and methods to meet the needs of different students. The learning data evaluation module records the learning data of the students and evaluates the students' learning situation based on this data, providing feedback for teaching to help teachers and students understand the learning progress and effectiveness. The entire system improves the efficiency of medical image teaching and management through automated and intelligent methods, reduces the need for manual operations, and improves the quality of decision-making through data analysis.

[0079] In one embodiment, referring to Figure 2 as shown, the system further includes a diagnostic report evaluation module 17, specifically used for:

[0080] Obtain a preset number of diagnostic reports written by the student;

[0081] Evaluate each of the diagnostic reports using the simulation teaching and management model to obtain the evaluation value of each diagnostic report;

[0082] Compare the evaluation value of each diagnostic report with the preset evaluation value of the corresponding diagnostic report type, and calculate the proportion of diagnostic reports whose evaluation value is greater than the preset evaluation value;

[0083] When the proportion is greater than the preset proportion, determine that the simulation teaching quality of the trainee meets the standard and generate an evaluation report;

[0084] Randomly send the evaluation report to the tutor of the trainee for revision. After the tutor completes the revision, send the evaluation report to the teaching and research section for review and archiving.

[0085] This embodiment realizes a trainee evaluation process based on a simulation teaching and management system, aiming to evaluate the diagnostic report writing ability and clinical thinking of trainees through automated and standardized methods. Specifically, trainees write a certain number of diagnostic reports in a simulation teaching environment. These diagnostic reports are based on simulated cases or virtual patients. The system monitors the number of diagnostic reports written by the trainee in real time. When the number of diagnostic reports written by the trainee reaches the standard, the simulation teaching and management model is used to evaluate the content of each diagnostic report to generate an evaluation value; when the number of diagnostic reports written by the trainee does not reach the standard, the trainee is automatically reminded of the gap between the number of diagnostic reports and the target.

[0086] Determine the type of the diagnostic report, query the preset evaluation value corresponding to the diagnostic report based on the type of the diagnostic report, compare the evaluation value of each diagnostic report with the preset evaluation criteria of the corresponding diagnostic report type to determine whether the performance of the trainee meets the expected level, and calculate the proportion of diagnostic reports whose evaluation value is greater than the preset evaluation value.

[0087] If the passing proportion exceeds the preset proportion, it is considered that the simulation teaching quality of the trainee meets the standard, and a formal evaluation report is generated. This evaluation report is randomly sent to the full-time tutor of the trainee for revision to ensure the accuracy and fairness of the evaluation. The evaluation report revised by the full-time tutor is submitted to the teaching and research section for final review and archiving, serving as an official record of the trainee's learning achievements, and the full-time tutor updates the trainee's learning plan according to the actual situation.

[0088] For example, assume that a medical student is participating in a simulation teaching program with the goal of improving their ability to write diagnostic reports. If the trainee writes 10 diagnostic reports on different cases in a simulated environment, the simulation teaching and management model evaluates each report separately, gives a score from 0 to 100, sets the preset evaluation value to 70 points, and sets the passing rate to 70%. If 8 of the trainee's reports have a score higher than 70 points, since the passing rate exceeds the preset 70%, the system generates an evaluation report for the trainee, deems that the quality of the trainee's simulation teaching meets the standard, and sends the evaluation report to the trainee's tutor. The tutor revises the trainee's evaluation report, puts forward some improvement suggestions, and the revised evaluation report is sent to the teaching and research section for final review and archived as the official learning achievement record of the trainee.

[0089] This embodiment ensures the accuracy, consistency, and standardization of the evaluation by analyzing the proportion of passing diagnostic reports and using model evaluation, and also facilitates the trainee to timely understand their performance and areas for improvement. At the same time, through the review by the tutor and the teaching and research section, the quality and reliability of the evaluation report are ensured. The archiving of the evaluation report also helps to record and track the trainee's learning progress and achievements.

[0090] In one embodiment, referring to Figure 3 as shown, the system further includes a trainee assessment module 18, which can be specifically used for:

[0091] When the number of diagnostic reports written by the trainee reaches a preset number, determine the trainee's assessment goal, trainee grade, and practice record;

[0092] Randomly generate assessment questions according to the assessment goal, trainee grade, and practice record, and set the completion time of the assessment questions;

[0093] Assess the trainee according to the assessment questions and the completion time, and generate an assessment result;

[0094] Adjust the trainee's simulation teaching and management plan according to the assessment result.

[0095] In this embodiment, after the number of diagnostic reports written by the trainee reaches the preset number, the system will determine the trainee's assessment goal, grade, and practice record. The assessment goal includes promotion assessment, monthly assessment, or year-end assessment, etc. The system randomly generates assessment questions based on the trainee's assessment goal, grade, and practice record, aiming to evaluate the trainee's knowledge and skills in a specific field. At the same time, a completion time is set for each assessment question to ensure the fairness and challenge of the assessment.

[0096] The trainee completes all assessment questions within the specified time, and the system records the trainee's answers and the time taken. The system evaluates the trainee's answers and generates an assessment result, including a score, rating, or other forms of feedback, which is simultaneously provided to the trainee himself / herself, the full-time tutor, and the teaching management staff.

[0097] Based on the assessment result, the system adjusts the trainee's simulated teaching and management plan to strengthen the trainee's weaknesses and improve their overall performance.

[0098] For example, assume that an intern in a medical school is preparing for an internship assessment in internal medicine. The trainee has completed a certain number of diagnostic reports. The system determines the assessment content based on the trainee's goals (such as improving the ability to diagnose heart diseases), grade (such as the third year), and practice records. The system generates a series of case analysis questions about heart diseases, including electrocardiogram (ECG) interpretation, medical history analysis, and treatment plan design. The completion time for each question is set at 30 minutes, with a total assessment time of 90 minutes. The trainee answers all the questions in a simulated exam environment and submits the answers within the specified time. The system evaluates the trainee's answers, gives a score and feedback, pointing out the trainee's deficiencies in ECG interpretation. Finally, based on the assessment result, the system recommends more simulated training for ECG interpretation to the trainee and arranges an online consultation meeting with a heart disease expert.

[0099] Through the above technical solution, the system can generate assessment questions according to the specific situation of the trainee, ensuring the personalization and pertinence of the learning content. The trainee can also timely understand their performance in the assessment, which helps to quickly identify and improve deficiencies. In addition, the simulated teaching and management plan model can adjust the teaching content and methods in a timely manner according to the trainee's performance, thereby improving the teaching effect and the trainee's learning outcome. The dynamic adjustment of the teaching plan also helps the trainee to learn more effectively, improving the learning efficiency and quality. Finally, through automated assessment and evaluation, the objectivity and accuracy of the assessment are also improved.

[0100] In one embodiment, referring to Figure 4 as shown, the system further includes a model update module 19, which can be specifically used for:

[0101] Regularly collect real medical image files of radiology cases to obtain candidate real medical image files of radiology cases;

[0102] After classifying the candidate real medical image files of radiology cases, input them into the simulated teaching and management model to iteratively update the simulated teaching and management model.

[0103] This embodiment implements a radiology medical image file processing and teaching update system based on a simulation teaching and management model. This system regularly collects and classifies real medical record medical image files of the radiology department, and then uses these files to iteratively update the simulation teaching and management model to improve the teaching quality and effect. Specifically, the model update module 19 of the system regularly collects real medical record medical image files of the radiology department, and the files can include image data obtained by various imaging techniques (such as X-ray, CT, MRI, etc.). Then, the collected files are classified, which can be classified based on criteria such as disease type, image characteristics, body parts, etc. The classified medical image files are input into the simulation teaching and management model to iteratively update the simulation teaching and management model using the newly collected files, so as to ensure that the model can reflect the latest medical knowledge and technological progress.

[0104] By continuously updating the simulation teaching model in this embodiment, teaching cases closer to real clinical situations can be provided, improving the learning effect of trainees; iterative update helps to improve the diagnostic accuracy of the model, enabling it to better simulate the doctor's diagnostic decision-making process.

[0105] In one embodiment, the learning data evaluation module 16 can also be specifically used for:

[0106] Perform data preprocessing on the learning data, and screen out the data related to learning behaviors from the preprocessed learning data to obtain target learning data;

[0107] Convert the target learning data into a transaction list, where each transaction in the transaction list contains a series of items;

[0108] Perform frequency statistics on the items in each transaction and sort them in descending order of frequency;

[0109] Insert each transaction into the pre-constructed frequent pattern tree in the order of the frequency of the items. If the item in the transaction already exists in the frequent pattern tree, increase the count of the transaction; otherwise, create a new node;

[0110] Recursively mine frequent item sets from the frequent pattern tree until no more frequent item sets can be mined from the frequent pattern tree;

[0111] Generate association rules from the frequent item sets, and check the relationship between the subsets of each frequent item set and the entire item set, calculate the confidence and lift. The confidence refers to the conditional probability that the consequent appears in the transactions where the antecedent appears, and the lift refers to the ratio of the probability that the consequent appears given the antecedent to the probability that the consequent appears itself;

[0112] Evaluate the effectiveness of the association rule based on the confidence and lift. After the evaluation passes, analyze the learning data according to the association rule and generate a learning evaluation report for the student.

[0113] This embodiment can first preprocess the learning data, including removing noise, filling in missing values, and performing data normalization to ensure data quality. Screen out the data related to learning behavior from the preprocessed learning data as the target learning data for analysis. For example, match the learning data with the features describing learning behavior to determine which learning data is related to learning behavior and use it as the target learning data.

[0114] Convert the target learning data into a transaction list. Each transaction represents a learning event and contains a series of items (such as learning time, learning content, learning achievements, etc.). Count the frequencies of the items in the transaction and sort them in descending order of frequency to form a frequency order to identify the most frequently occurring items. Insert each transaction into the frequent pattern tree according to the frequency order of the corresponding items. During the insertion process, if the item in the transaction already exists in the frequent pattern tree, increase the count of the transaction; otherwise, create a new node and insert the transaction into the new node. Among them, the frequent pattern tree is a data structure for mining frequent item sets. The frequent item set refers to an item set that appears more than a certain preset threshold (i.e., the minimum support) in the dataset.

[0115] Recursively mine frequent item sets from the frequent pattern tree until no more frequent item sets can be mined from the frequent pattern tree. Generate association rules from the frequent item sets and calculate the confidence and lift of each rule. Evaluate the effectiveness of the association rule based on the confidence and lift. The confidence represents the conditional probability that the consequent appears in the transactions where the antecedent appears; the lift represents the ratio of the probability that the consequent appears given the antecedent to the probability that the consequent appears by itself.

[0116] Generate a learning evaluation report for the student according to the evaluation results. The report includes an analysis of the student's learning behavior and suggestions.

[0117] This embodiment can discover the internal connections between learning behaviors, such as the relationship between certain learning habits and learning achievements, through association rule mining. Based on the analysis results of the association rules, personalized learning suggestions and intervention measures can be provided for the student.

[0118] For example, suppose there is an online learning platform that collects the behavioral data of students on the learning platform, including learning duration, learning frequency, types of videos watched, quiz scores completed, etc. The system first filters out the data related to learning behaviors, such as learning duration, types of videos watched, and quiz scores, and converts the learning behaviors of each student into a transaction list, where each transaction contains a series of items. A frequent pattern tree is constructed, and the transactions are inserted into the frequent pattern tree to form frequent item sets. Frequent item sets are mined from the frequent pattern tree. For example, it is found that "watching teaching videos" and "completing quizzes" often occur together. Association rules are generated, such as "if a student watches a teaching video, then they are more likely to complete the quiz". At the same time, the confidence and lift of the association rules are calculated to evaluate the effectiveness of the rules. According to the association rules, a learning assessment report for the students is generated, which indicates the relationship between the students' learning habits and learning outcomes and provides improvement suggestions to gain a deeper understanding of the students' learning behaviors, provide personalized learning suggestions for the students, and also provide a basis for the model to optimize teaching strategies.

[0119] Preferably, the data preprocessing method includes removing noise, filling in missing values, and data normalization processing.

[0120] Data preprocessing is a crucial step in data analysis, which involves cleaning, transforming, and organizing the original data to improve data quality and ensure that the data is suitable for subsequent analysis.

[0121] In this embodiment, noise refers to errors or interference information in the data, such as random pixel points in an image, errors in sensor readings, etc. Removing noise can be achieved through filters (such as Gaussian filtering, median filtering) or more advanced techniques (such as wavelet transform), which are not specifically limited here.

[0122] Data missing values can be filled in by various methods, such as using the mean, median, mode, or a prediction model based on other variables to estimate the missing values.

[0123] Data normalization processing refers to scaling the data proportionally so that it falls into a specific small interval, such as [0, 1], to avoid biases caused by features with different dimensions during the analysis process. This normalization method includes min-max normalization and Z-score standardization.

[0124] In this embodiment, through data preprocessing, the accuracy of data analysis can be improved. At the same time, normalization and noise removal can reduce the time and resources required for data analysis and improve the efficiency of the system for data analysis.

[0125] In one embodiment, the model combination module 14 can also be specifically used for:

[0126] Based on the training results of each classifier, calculate the loss value of each classifier respectively based on a preset cross-entropy loss function;

[0127] Determine whether the loss value of each classifier is lower than a preset loss value;

[0128] If so, determine that the training results of each classifier meet the standards;

[0129] If not, filter out the classifiers whose training results do not meet the standards to obtain candidate classifiers;

[0130] Use the stochastic gradient descent method to adjust the parameters of the candidate classifier;

[0131] Randomly select multiple training samples of one type to retrain the candidate classifier with adjusted parameters until the loss value of the candidate classifier is lower than the preset loss value, and obtain classifiers whose training results all meet the standards.

[0132] In this embodiment, the cross-entropy loss function can be used to calculate the loss value of each classifier. This cross-entropy loss function is a loss function used in classification tasks and is used to measure the difference between the probability distribution predicted by the model and the probability distribution of the true labels.

[0133] Compare the loss value of each classifier with the preset loss value to determine whether the model has reached the expected performance standard. This preset loss value can be custom-set, such as 0.3.

[0134] If the performance of the classifier does not meet the standards, use the stochastic gradient descent method (SGD) to adjust the parameters of the candidate classifier that does not meet the standards. SGD is an optimization algorithm that updates the parameters of the model by randomly selecting samples in the parameter space, thereby reducing the loss value of the model.

[0135] Then, randomly select multiple training samples of one type and use these samples to retrain the classifier with adjusted parameters until the loss value of the candidate classifier is also lower than the preset loss value.

[0136] For example, after training the model, first calculate the cross-entropy loss value of the model on the validation set and find that the loss value of the model is higher than the set threshold, which means that the performance of the model still needs to be improved. Then use the SGD method to adjust the parameters of the model, randomly selecting one or a small part of the image samples each time to update the weights of the model. After several rounds of parameter adjustment, use specific training samples to retrain the model until the loss value of the model is lower than the preset loss value.

[0137] In this embodiment, through continuous parameter adjustment and retraining, the prediction accuracy and generalization ability of the model can be improved. At the same time, compared with calculating the gradient of the entire training set, the SGD method only uses one or a small part of the samples to update the parameters each time, which can greatly reduce the consumption of computing resources. In addition, the SGD method can also update the parameters faster, thereby accelerating the training speed of the model. By continuously updating the parameters during the training process, the SGD method also helps the model better adapt to the training data and avoid overfitting.

[0138] In one embodiment, the model combination module 14 may further be specifically configured to:

[0139] Randomly select a training sample from multiple said training samples to obtain a target training sample;

[0140] Calculate the gradient of each original parameter of the candidate classifier under the condition of the target training sample according to the cross-entropy loss function;

[0141] Calculate the difference between the gradient of each said original parameter with respect to the loss value of the candidate classifier and the gradient of each said original parameter on the target training sample to obtain a gradient difference;

[0142] Calculate the difference between the logarithm of the gradient difference and the logarithm of a reference gradient difference to obtain a target gradient difference, where the reference gradient difference is 0.18;

[0143] Continuously adjust the original parameters of the candidate classifier in the negative gradient direction according to the target gradient difference.

[0144] The model combination module of this embodiment randomly selects a training sample from multiple training samples as the target training sample for the subsequent parameter update process. According to the cross-entropy loss function, calculate the gradient of each original parameter of the target training sample under the current model parameters. This gradient represents the direction in which the derivative of a multivariate function is the largest at a given point, indicating the direction in which the function value increases fastest. In machine learning, especially when training a model, the gradient is used to guide how to adjust the model parameters (such as weights and biases) to minimize the loss function.

[0145] Calculate the gradient of the loss function with respect to each parameter, which represents how the change of each parameter affects the value of the loss function under the current parameters. At the same time, calculate the difference between the gradient of each parameter and the gradient of the corresponding parameter on the target training sample to obtain a gradient difference to evaluate the difference between the current parameter update direction and the ideal update direction. Calculate the difference between the logarithm of the gradient difference and the logarithm of a preset reference gradient difference (such as 0.18) to obtain a target gradient difference to convert the gradient difference into a more refined and more convenient form for system processing.

[0146] Adjust the original parameters of the model in the direction of the negative gradient according to the target gradient difference. This negative gradient direction is the direction in which the loss function decreases most rapidly. Therefore, updating the parameters in this direction can effectively reduce the loss value of the model.

[0147] In this embodiment, by continuously optimizing the model parameters, the performance of the model on the training data can be improved, thereby improving the prediction accuracy of the model. A suitable parameter update strategy can accelerate the speed at which the model converges to the minimum loss value and reduce the training time. At the same time, by continuously adjusting the parameters during the training process, overfitting of the model to the training data can be reduced, and the generalization ability of the model can be improved.

[0148] Preferably, the preset cross-entropy loss function includes:

[0149]

[0150] where N is the number of training samples, i is the i-th training sample, e is the base of the natural logarithm, k is the diagnostic result predicted for the i-th training sample, y i,k is the vector corresponding to the diagnostic result k of the i-th training sample, and p i,k is the probability that the diagnostic result k predicted for the i-th training sample is the reference diagnostic result.

[0151] In one embodiment, the system further includes a probability calculation module, which can be specifically used for:

[0152] Vectorize the diagnostic result k predicted for the i-th training sample to obtain a first vector;

[0153] Query the reference diagnostic result corresponding to the i-th training sample, and vectorize the reference diagnostic result corresponding to the i-th training sample to obtain a second vector;

[0154] Calculate the Euclidean distance between the first vector and the second vector to obtain the probability that the diagnostic result k predicted for the i-th training sample is the reference diagnostic result.

[0155] In this embodiment, the predicted diagnostic result k of the i-th training sample is converted into a numerical vector, and this process is called vectorization. Vectorization is used to convert text into a format that can be processed by a machine. For example, the word2vec word vectorization algorithm is used to vectorize the diagnostic result k in text form to obtain a first vector.

[0156] Query the reference diagnostic result corresponding to the i-th training sample, and also vectorize the reference diagnostic result using the word2vec word vectorization algorithm to obtain a second vector. The reference diagnostic result can be provided by experts or clinical experience data and used as a standard for evaluating the prediction accuracy of the model.

[0157] Calculate the Euclidean distance between two vectors, which is used to measure the straight-line distance between two points in a multi-dimensional space. Based on the calculated Euclidean distance, evaluate the similarity between the predicted diagnosis result and the reference diagnosis result. The smaller the Euclidean distance, the closer the predicted diagnosis result is to the reference diagnosis result, and the higher the accuracy of the model diagnosis.

[0158] For example, assume that the model analyzes a lung CT image, and the predicted result is "lung adenocarcinoma", which is converted into vector A. The expert reference diagnosis result of this image is also "lung adenocarcinoma", which is converted into vector B. Calculate the Euclidean distance between vector A and vector B to obtain a value representing their similarity. If the Euclidean distance is very small and close to zero, then it can be considered that the predicted result of the model is very close to the reference diagnosis, and the performance of the model is good. Thus, through this process, the accuracy of the diagnostic report generated by the machine learning model can be automatically evaluated, and the model parameters can be adjusted according to the evaluation results to improve the accuracy of future diagnoses.

[0159] This embodiment can achieve the quantification of the accuracy of the diagnostic report by comparing the predicted result and the reference result, which helps to improve the diagnostic quality; at the same time, through the continuous evaluation and feedback of the diagnostic results, the model can also be continuously optimized to improve the accuracy of future diagnoses.

[0160] In summary, a simulation teaching and management system based on the real medical record medical image files of the radiology department provided by the present application. The file classification module classifies the real medical record medical image files of the radiology department through automation technology, improving the efficiency and accuracy of classification; the sample extraction module automatically extracts training samples from the classified medical image files, reducing the workload of manually selecting samples and ensuring the representativeness and diversity of the samples; the model training module uses the decision tree algorithm to train the extracted training samples to generate multiple classifiers to process a large amount of data and can learn complex patterns from the data; the model combination module combines them into a comprehensive simulation teaching and management model after confirming that the training results of each classifier meet the standards. This integration method can improve the generalization ability and prediction accuracy of the model; the solution formulation module customizes personalized teaching solutions for each student using the simulation teaching and management model according to the learning files and labels of the students, realizing the personalization of teaching content and methods to meet the needs of different students. The learning data evaluation module records the learning data of the students and evaluates the learning situation of the students based on these data to provide feedback for teaching, helping teachers and students understand the learning progress and effects. The entire system improves the efficiency of medical image teaching and management through automated and intelligent methods, reduces the need for manual operations, and improves the quality of decision-making through data analysis.

[0161] It should be noted that in this text, the term "including", "comprising", or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or elements that are inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, apparatus, article, or method comprising such element.

[0162] The above are only the preferred embodiments of the present application, and thus do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.

Claims

1. A simulation teaching and management system based on medical imaging archives of real medical records in the radiology department, characterized in that, Including: A file classification module, which is used to collect the medical image files of real medical records in the radiology department, classify the medical image files of real medical records in the radiology department, and obtain various types of medical image files of real medical records in the radiology department. The classification basis includes: disease type, body part, disease system to which it belongs, or assessment difficulty; A sample extraction module, which is used to extract multiple training samples from each type of the medical image files of real medical records in the radiology department. Each training sample includes a medical image of a real medical record in the radiology department and a corresponding reference diagnosis result; A model training module, which is used to train multiple training samples of each type according to the decision tree algorithm respectively, and obtain multiple classifiers and corresponding training results; A model combination module, which is used to combine multiple classifiers to obtain a simulation teaching and management model when it is determined that the training results of each classifier meet the standards; A plan formulation module, which is used to obtain the student file of the student, determine the learning label of the student based on the student file, input the learning label of the student into the simulation teaching and management model, use the simulation teaching and management model to formulate a simulation teaching and management plan for the student, and send the simulation teaching and management plan to the student; A learning data evaluation module, which is used to record the learning data of the student completing the simulation teaching and management plan, and evaluate the learning situation of the student according to the learning data; Among them, the learning data evaluation module is also used for: Performing data preprocessing on the learning data, and screening out the data related to learning behavior from the learning data after data preprocessing to obtain target learning data; Converting the target learning data into a transaction list, and each transaction in the transaction list contains a series of items; Performing frequency statistics on the items in each transaction, and arranging them in descending order of frequency; Inserting each transaction into the pre-constructed frequent pattern tree in the order of the frequency of the items. If the item in the transaction already exists in the frequent pattern tree, the count of the transaction is increased. Otherwise, a new node is created; Recursively mining frequent item sets from the frequent pattern tree until no more frequent item sets can be mined from the frequent pattern tree; Generating association rules from the frequent item sets, and checking the relationship between the subsets of each frequent item set and the entire item set, calculating the confidence and lift. The confidence refers to the conditional probability that the posterior item also appears in the transactions where the anterior item appears. The lift refers to the ratio of the probability that the posterior item appears given the anterior item to the probability that the posterior item appears itself; Evaluating the effectiveness of the association rules according to the confidence and lift. After the evaluation passes, analyzing the learning data according to the association rules, and generating a learning evaluation report of the student.

2. The system according to claim 1, wherein It also includes a diagnostic report evaluation module, which is used for: Obtaining a preset number of diagnostic reports written by the student; using the simulation teaching and management model to evaluate each diagnostic report, and obtaining the evaluation value of each diagnostic report; Compare the evaluation value of each of the said diagnostic reports with the preset evaluation value corresponding to the type of diagnostic report, and calculate the proportion of diagnostic reports whose evaluation value is greater than the preset evaluation value; When the said proportion is greater than the preset proportion, determine that the simulated teaching quality of the said trainee meets the standard, and generate an evaluation report; Randomly send the said evaluation report to the tutor of the said trainee for revision, and after the tutor completes the revision, send the said evaluation report to the teaching and research section for review and archiving.

3. The system according to claim 2, wherein It further includes a trainee assessment module, which is used for: When the number of diagnostic reports written by the said trainee reaches the preset number, determine the assessment goal, trainee grade and practice record of the said trainee; Randomly generate assessment questions according to the said assessment goal, trainee grade and practice record, and set the completion time of the said assessment questions; Assess the said trainee according to the said assessment questions and completion time, and generate an assessment result; Adjust the simulated teaching and management plan of the said trainee according to the said assessment result.

4. The system according to claim 1, characterized in that, It further includes a model update module, which is used for: Regularly collect the medical image files of real cases in the radiology department to obtain candidate medical image files of real cases in the radiology department; After classifying the said candidate medical image files of real cases in the radiology department, input them into the simulated teaching and management model, and perform iterative update on the simulated teaching and management model.

5. The system according to claim 1, wherein The said data preprocessing methods include removing noise, filling missing values, and data normalization processing.

6. The system according to claim 1, characterized in that The said model combination module is further used for: According to the training results of each of the said classifiers, calculate the loss value of each of the said classifiers respectively based on the preset cross-entropy loss function; Judge whether the loss value of each of the said classifiers is lower than the preset loss value; If so, determine that the training results of each of the said classifiers meet the standard; If not, screen out the classifiers whose training results do not meet the standard to obtain candidate classifiers; Use the stochastic gradient descent method to adjust the parameters of the said candidate classifiers; Randomly select multiple said training samples of one type to retrain the said candidate classifiers with adjusted parameters until the loss value of the said candidate classifiers is lower than the preset loss value, and obtain classifiers whose training results all meet the standard.

7. The system according to claim 6, characterized in that, The said model combination module is further used for: Randomly select a training sample from multiple said training samples to obtain a target training sample; Calculate the gradient of each of the said original parameters corresponding to the said candidate classifier under the condition of the said target training sample according to the said cross-entropy loss function; Calculate the difference between the gradient of each of the said original parameters with respect to the loss value corresponding to the said candidate classifier and the gradient of each of the said original parameters on the said target training sample to obtain a gradient difference; Calculate the difference between the logarithm of the said gradient difference and the logarithm of the reference gradient difference to obtain a target gradient difference, and the reference gradient difference is 0.18; Continuously adjust the original parameters of the said candidate classifier in the negative gradient direction according to the said target gradient difference.

8. The system according to claim 6, characterized in that The said preset cross-entropy loss function includes: where N is the number of training samples, i is the i-th training sample, e is the base of the natural logarithm, k is the predicted diagnostic result of the i-th training sample, and y i,k is the vector of the i-th training sample corresponding to the diagnostic result k, and p i,k is the probability that the predicted diagnostic result k of the i-th training sample is the reference diagnostic result.

9. The system according to claim 8, characterized in that, It further includes a probability calculation module, which is used for: Vectorize the diagnostic result k predicted by the i-th training sample to obtain a first vector; Query the reference diagnosis result corresponding to the i-th training sample, and perform vectorization processing on the reference diagnosis result corresponding to the i-th training sample to obtain a second vector; Calculate the Euclidean distance between the first vector and the second vector to obtain the probability that the predicted diagnosis result k of the i-th training sample is the reference diagnosis result.

Images