System and method for judging knee joint motion range loss degree after trauma

Abstract

The invention provides a system and a method for judging knee joint activity loss degree after trauma, a data acquisition module acquires knee joint judicial expertise archive data as original sample data, and a data processing module processes the original sample data into sample loss characteristic data; the model training module is used for training the initially constructed machine learning algorithm model by using the sample loss feature data to obtain a loss degree discrimination model; the data acquisition module also acquires knee joint clinical data of a wounded person, and the data processing module also processes the knee joint clinical data into to-be-discriminated feature data; and the model discrimination module processes the to-be-discriminated feature data by using a loss degree discrimination model, and outputs a model discrimination result of the knee joint activity loss degree. The accuracy and efficiency of knee joint motion range diagnosis are improved, and the influence of human subjective judgment on the diagnosis result is reduced.

Description

Technical Field

[0001] The present invention relates to the technical fields of medical information processing and forensic identification, and particularly relates to a discrimination system and method for the degree of loss of knee joint mobility after trauma. Background Art

[0002] The knee joint is one of the largest weight-bearing joints in the human body. Knee joint injuries account for a high proportion in limb injury incidents. Knee joint function injuries are closely related to traffic accidents, industrial injury accidents, criminal cases, etc. Strong violence will directly cause fractures of the bones that make up the knee joint. In the field of forensic clinical medicine, there are a relatively large number of cases of identification of knee joint movement function disorders caused by trauma. The goal of knee joint function repair is to restore the stability and movement function of the joint. The degree of repair is related to factors such as age, occupation, primary injury, medical intervention, and rehabilitation exercise. The repair of tissue structure and the degree of joint function recovery are directly related to subsequent judicial issues such as disability assessment grades and compensation. According to the relevant provisions of the "Classification of Disability Degrees of Human Injury", a 25% loss of function is the boundary for whether the disability degree is reached. Therefore, correctly evaluating the degree of knee joint movement limitation is of great significance for disability degree assessment.

[0003] Currently, the degree of loss of knee joint mobility mainly depends on the comprehensive judgment of the appraiser on medical record data and instrument measurement examinations. Although the existing knee joint mobility measurement methods have been widely used in forensic clinical practice and can meet the needs of daily identification to a certain extent, they are greatly affected by subjective factors, and thus still face some significant challenges and deficiencies in the field of forensic identification. First, in the case of manual or instrument measurement, it cannot be guaranteed that the appraisee fully cooperates. The appraisee may deliberately exaggerate the degree of loss of joint mobility for reasons such as economic compensation. Second, analyzing medical records and physical examinations depends on the appraiser's appraisal experience, and there are certain subjective errors. Summary of the Invention

[0004] Based on the above, the present invention provides a discrimination system and method for the degree of loss of knee joint mobility after trauma, aiming to solve the technical problems such as the excessive subjectivity and large disputes in the existing knee joint mobility identification.

[0005] A discrimination system for the degree of loss of knee joint mobility after trauma, comprising:

[0006] A data acquisition module, configured to collect data of a number of closed forensic identification files regarding the knee joint as original sample data, and each sample in the original sample data includes the actual measurement result of the degree of loss of knee joint mobility;

[0007] A data processing module, connected to the data acquisition module, configured to process the original sample data to obtain sample loss feature data;

[0008] A model training module, connected to the data processing module, is used to construct training data based on sample feature data, using the actual measured value of the degree of knee joint mobility loss as a label to train the initially constructed machine learning algorithm model to obtain a loss degree discrimination model;

[0009] The data acquisition module is also used to collect the knee joint clinical data of the injured person, and the data processing module is also used to process the knee joint clinical data of the injured person to obtain the feature data to be discriminated;

[0010] A model discrimination module, also connected to the data processing module, is used to process the feature data to be discriminated using the deployed and trained loss degree discrimination model, discriminate the degree of knee joint mobility loss of the injured person, and output the model discrimination result of the degree of knee joint mobility loss.

[0011] Furthermore, the data processing module includes:

[0012] A feature extraction unit, used to extract original feature information from the original sample data;

[0013] A feature encoding unit, connected to the feature extraction unit, is used to encode the extracted original feature information to obtain feature encoding information, and the sample feature data is composed of the feature encoding information.

[0014] Furthermore, the system also includes a database;

[0015] The forensic appraisal archive data includes an appraisal opinion and imaging materials, and one piece of forensic appraisal archive data serves as one sample in the original sample data;

[0016] The imaging materials include archived X-ray films and DICOM images;

[0017] The data processing module is also connected to the database;

[0018] The feature extraction unit stores the extracted original feature information in the database in the form of a first two-dimensional table. The columns in the first two-dimensional table are the feature fields of the original feature information, and the rows in the first two-dimensional table are each sample;

[0019] The data processing module also includes a data storage unit, and the data storage unit is used for:

[0020] Storing the archived X-ray films in the database in the form of a second two-dimensional table, and the second two-dimensional table records the relative storage paths of the archived X-ray films of each sample;

[0021] Storing the DICOM images in the database in the form of a third two-dimensional table, and the third two-dimensional table records the relative storage paths of the DICOM images of each sample.

[0022] Further, the original feature information includes the actual measurement of the degree of knee joint mobility loss, personal basic information, fracture site, fracture type, fracture healing condition, whether the implant is in place, whether surgical treatment is performed, whether there is displacement, whether the joint surface is involved, and the time interval between the first and last radiographs.

[0023] Further, the model training module includes:

[0024] A model construction unit for constructing multiple initial loss degree discrimination models based on multiple machine learning algorithms;

[0025] A model training unit, connected to the model construction unit, for constructing a training set and a test set according to the sample feature data, and using the training set to train each initially constructed machine learning algorithm model to obtain a loss degree discrimination model;

[0026] A first screening unit, connected to the model training unit, for performing a first screening on the trained loss degree discrimination model using the test set;

[0027] The model training unit is also used to verify the loss degree discrimination model screened out for the first time based on the five-fold cross-validation method;

[0028] A second screening unit, respectively connected to the first screening unit and the model training unit, for: performing a second screening on the loss degree discrimination model obtained by using the five-fold cross-validation method, and jointly constructing the loss degree discrimination model screened out for the second time into a final loss degree discrimination model;

[0029] The model discrimination module is used to: discriminate the degree of knee joint mobility loss of the injured person based on the finally constructed loss degree discrimination model in combination with the soft voting method, and output the model discrimination result of the degree of knee joint mobility loss.

[0030] Further, the feature encoding unit encodes the actual measurement of the degree of knee joint mobility loss according to whether it reaches the disability level;

[0031] During training, the model training module uses the encoding result of the actual measurement of the degree of knee joint mobility loss reaching the disability level as the label of the positive sample, and the encoding result of not reaching the disability level as the label of the negative sample, so as to train the initially constructed loss degree discrimination model.

[0032] Further, the personal basic information includes age;

[0033] The feature encoding unit encodes the age according to the age range.

[0034] A method for discriminating the degree of knee joint mobility loss after trauma, using the aforementioned discrimination system for the degree of knee joint mobility loss after trauma, includes:

[0035] Step A1, collect the forensic appraisal archive data of several closed cases regarding the knee joint as the original sample data. Each sample in the original sample data includes the actual measurement results of the degree of loss of knee joint mobility.

[0036] Step A2, process the original sample data to obtain sample feature data.

[0037] Step A3, construct training data based on the sample feature data, using the actual measurement value of the degree of loss of knee joint mobility as the label to train the initially constructed machine learning algorithm model to obtain a loss degree algorithm model.

[0038] Step A4, collect the clinical data of the knee joint of the injured person, and process the clinical data of the knee joint of the injured person to obtain the feature data to be discriminated.

[0039] Step A5, use the deployed and trained loss degree discrimination model to process the feature data to be discriminated, discriminate the degree of loss of knee joint mobility of the injured person, and output the model discrimination result of the degree of loss of knee joint mobility.

[0040] Further, step A2 includes:

[0041] Step A21, extract the original feature information from the original sample data.

[0042] Step A22, encode the extracted original feature information to obtain feature encoding information, and the sample feature data is composed of the feature encoding information.

[0043] Further, the forensic appraisal archive data includes an appraisal opinion and imaging materials. One piece of forensic appraisal archive data serves as one sample in the original sample data.

[0044] The imaging materials include archived images and DICOM images.

[0045] In step A21, the extracted original feature information is also stored in the database in the form of a first two-dimensional table. The columns in the first two-dimensional table are the feature fields of the original feature information, and the rows in the first two-dimensional table are each sample.

[0046] In step A21, the archived images are also stored in the database in the form of a second two-dimensional table, and the second two-dimensional table records the relative storage paths of the archived images of each sample.

[0047] In step A21, the DICOM images are also stored in the database in the form of a third two-dimensional table, and the third two-dimensional table records the relative storage paths of the DICOM images of each sample.

[0048] The beneficial technical effects of the present invention are as follows: The present invention proposes an intelligent and automated system based on artificial intelligence to improve the accuracy and efficiency of diagnosing the range of motion of the knee joint, and can more accurately determine whether the degree of loss of the range of motion of the knee joint of the appraised person reaches the disability level, reducing the influence of subjective human judgment on the diagnosis result. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figures 1-3 It is a module schematic diagram of a discriminant system for the degree of loss of knee joint range of motion after trauma according to the present invention;

[0050] Figures 4-5 It is a step flowchart of a discriminant method for the degree of loss of knee joint range of motion after trauma according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0052] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0053] Next, the present invention will be further described in conjunction with the accompanying drawings and specific embodiments, but it is not a limitation of the present invention.

[0054] See Figure 1 , a discriminant system for the degree of loss of knee joint range of motion after trauma according to the present invention, includes:

[0055] A data acquisition module (1) for collecting data of a number of closed judicial appraisal files on the knee joint as original sample data, and each sample in the original sample data includes the actual measurement result of the degree of loss of knee joint range of motion;

[0056] A data processing module (2), connected to the data acquisition module (1), for processing the original sample data to obtain sample feature data;

[0057] A model training module (3), connected to the data processing module (2), for constructing training data according to the sample feature data, using the actual measurement value of the degree of loss of knee joint range of motion as a label, and training the initially constructed machine learning algorithm model to obtain a loss degree discriminant model;

[0058] The data acquisition module (1) is also used to collect the clinical data of the knee joint of the injured person, and the data processing module (2) is also used to process the clinical data of the knee joint of the injured person to obtain the feature data to be discriminated.

[0059] The model discrimination module (4), which is also connected to the data processing module (2), is used to process the feature data to be discriminated using the trained loss degree discrimination model deployed, discriminate the loss degree of the knee joint mobility of the injured person, and output the model discrimination result of the loss degree of the knee joint mobility.

[0060] The present invention proposes an intelligent and automated system based on artificial intelligence to improve the accuracy and efficiency of the diagnosis of knee joint mobility, and can more accurately judge whether the loss degree of the knee joint mobility of the appraised person reaches the disability level, reducing the influence of subjective human judgment on the diagnosis result.

[0061] Each sample in the original sample data is a forensic appraisal file data of a knee joint. The forensic appraisal file data includes unstructured text carried by the appraisal opinion and imaging materials in different formats, containing a large amount of appraisal-related information, such as: personal basic information (case number, gender, age), injury-related information (injury situation and time, whether surgery is performed), imaging materials (mainly the first and last radiographs such as X-rays, anteroposterior and lateral views), the appraisal time limit, whether the implant is in place, the fracture healing situation (normal healing, mild malunion, severe malunion), etc. that can be obtained from the imaging materials.

[0062] Specifically, each forensic appraisal file data has a case number.

[0063] See Figure 2 , further, the data processing module (2) includes:

[0064] The feature extraction unit (21) is used to extract the original feature information from the original sample data;

[0065] The feature encoding unit (22), which is connected to the feature extraction unit (21), is used to encode the extracted original feature information to obtain the feature encoding information, and the sample feature data is composed of the feature encoding information.

[0066] Further, the system also includes a database (5);

[0067] The forensic appraisal file data includes the appraisal opinion and imaging materials, and one forensic appraisal file data is used as one sample in the original sample data;

[0068] The imaging materials include archived imaging films and DICOM imaging films;

[0069] The data processing module (2) is also connected to the database (5);

[0070] The feature extraction unit (21) stores the extracted original feature information in the database in the form of a first two-dimensional table. The columns in the first two-dimensional table are the feature fields of the original feature information, and the rows in the first two-dimensional table are each sample;

[0071] The data processing module (2) further includes a data storage unit (23), and the data storage unit (23) is used for:

[0072] Storing the archived image films in the database in the form of a second two-dimensional table, and the second two-dimensional table records the relative storage paths of the archived image films of each sample;

[0073] Storing the DICOM image films in the database in the form of a third two-dimensional table, and the third two-dimensional table records the relative storage paths of the DICOM image films of each sample.

[0074] Specifically, the imaging data includes archived image films and DICOM image films.

[0075] The present invention uses a MySQL database to establish a knee joint injury database. The database takes appraisal cases, archived image films, and DICOM image films as three entities, and the extracted original feature information is the attribute of each entity. Each appraisal case can have multiple archived image films or DICOM image films, and each archived image film can correspond to multiple DICOM image films.

[0076] MySQL: An open-source relational database management system that uses Structured Query Language (SQL) for data management. MySQL is widely popular for its speed, reliability, flexibility, and ease of use, and is suitable for applications of various scales, from embedded systems to large-scale Internet applications.

[0077] DICOM (Digital Imaging and Communications in Medicine): DICOM is an international standard (ISO 12052) for medical images and related information. The standard defines the format and protocol for storing and transmitting medical images and their related information (such as patient information, device parameters, etc.).

[0078] The database contains three two-dimensional tables for storing data, and the first two-dimensional table is a text set two-dimensional table.

[0079] The second two-dimensional table is an archived image set two-dimensional table.

[0080] The third two-dimensional table is the DICOM image set two-dimensional table. The archived image films are the pictures after scanning, such as the pictures after scanning by the appraisal institute. The DICOM image films are the original pictures that can be obtained, such as the original pictures that can be obtained by the appraisal institute. The forensic appraisal archive data includes archived image films, but does not necessarily include DICOM image films.

[0081] The database contains three two-dimensional tables for storing data. The table names can be the text set, the archived image set, and the DICOM image set respectively. The archived image films and DICOM image films are unstructured data. The storage paths of the image films are recorded using relative paths in the two-dimensional table. The rest of the data is obtained manually by extracting from the appraisal opinions and reading the films. The obtained original feature information is entered into the first two-dimensional table of the database through the feature extraction unit. The original copies of the attachments, such as archived image films and DICOM image films, are stored in the form of relative paths, which can save space and improve efficiency.

[0082] Furthermore, the original feature information includes the actual measured value of the degree of loss of knee joint mobility, personal basic information, fracture site, fracture type, fracture healing condition, whether the implant is in place, whether surgical treatment has been performed, whether there is displacement, whether the joint surface is involved, and the time interval between the first and last radiographs.

[0083] Specifically, the original feature information also includes the appraisal time limit.

[0084] The appraisal time limit refers to the time from when the knee joint is injured to when the appraisal is conducted, and is encoded in months.

[0085] The "year", "category", and "case number" of the appraisal opinion can be extracted to form a unique identifier (primary key). The features such as "degree of loss of knee joint mobility", "age", "appraisal time limit", and "gender" should be extracted from the appraisal opinion and its attached materials. Combining with medical imaging data, the features such as "fracture site", "fracture type", "whether there is displacement", "whether the joint surface is involved", "whether surgical treatment has been performed", "fracture healing condition", "time interval between the first and last radiographs", and "whether the implant is in place" are extracted.

[0086] The actual measured value of the degree of loss of knee joint mobility is obtained from the appraisal opinion and is mainly used as a label when training the model. When collecting the forensic appraisal archive data that has been closed from relevant institutions, it is necessary to include the complete forensic appraisal opinion and the forensic clinical physical examination record, and the measurement result of the degree of loss of knee joint mobility must be included. The forensic appraisal archive data with nerve and muscle dysfunction left by fractures adjacent to the knee joint, non-union of fractures, and limb loss is excluded from the original sample data.

[0087] After feature extraction, the extracted features are also encoded. To ensure that each feature can be used for model training, first, the database is exported as a csv file, with columns for each feature and rows for each sample.

[0088] Further, the feature encoding unit encodes the actual measurement value of the degree of knee joint mobility loss according to whether it reaches the disability level;

[0089] During training, the model training module uses the encoding result of the actual measurement value of the degree of knee joint mobility loss reaching the disability level as the label of the positive sample, and the encoding result of not reaching the disability level as the label of the negative sample, so as to train the initially constructed discrimination model of the degree of loss.

[0090] Further, the personal basic information includes age and gender.

[0091] The feature encoding unit encodes age according to age ranges.

[0092] Encode each feature of each sample. Divide age into multiple age intervals and encode according to the divided age intervals. For example, use "1", "2", and "3" to encode three age groups respectively.

[0093] Specifically, the "degree of knee joint mobility loss" is the actually measured value of joint mobility loss, which is calculated according to "Forensic Clinical Examination Specification" 7.11.7.2.3b)2).

[0094] For example, "age" is divided into a three-classification index with 60 years old and 75 years old as the boundaries, that is, 0 - 60 years old is an age interval, encoded as 1, 60 - 75 years old is an age interval, encoded as 2, and over 75 years old is an age interval, encoded as 3. Set a code for each interval. If an individual's age is in a certain age interval, the encoding of the age feature corresponds to the code of the age interval to which it belongs.

[0095] For "gender" and "whether surgical treatment is performed", binary encoding of 0 and 1 is carried out. For example, male gender is encoded as 1, and female gender is encoded as 0. For example, if surgical treatment is performed, it is encoded as 1, and if no surgical treatment is performed, it is encoded as 0. "Whether the joint surface is involved" and "whether the implant is in place" are also encoded as 0 and 1;

[0096] For the feature of fracture healing situation, hierarchical encoding is carried out. According to the fracture healing situation in the last radiograph, encode from small to large according to the severity, and the encoding values range from 1 to 3. For example, normal healing is encoded as 1, mild malunion is encoded as 2, and severe malunion is encoded as 3.

[0097] The encoding of some features can be carried out according to numerical mapping encoding, and the encoding values start from 1. For example, "fracture site" is a multi-classification index, and its values are all non-empty subsets of the set S = {patella, tibia, fibula, femur}. "Fibula" is encoded as 1, "patella" is encoded as 2, etc. Specifically, in "fracture site", "fibula", "patella", "femur", "tibia", "fibula and femur", "patella and femur", "tibia and fibula", "patella and tibia", "tibia and femur", "patella, tibia and fibula", "tibia, fibula and femur", "patella, tibia and femur" are encoded from 1 to 12.

[0098] In the fracture type, "avulsion fracture", "linear fracture", and "comminuted fracture" are encoded as 1, 2, and 3 respectively.

[0099] The "time interval between the first and last radiographs" is the time from the first radiograph after injury to the last radiograph before the appraisal.

[0100] The main purposes of these encodings are twofold. One is to classify some features hierarchically so that machine learning algorithms can more easily understand and analyze data; the other is to convert the original non-numerical data into numerical form for the training and prediction of machine learning models.

[0101] See Figure 3 , further, the model training module (3) includes:

[0102] A model construction unit (31) for constructing multiple initial loss degree discrimination models based on multiple machine learning algorithms;

[0103] A model training unit (32), connected to the model construction unit (31), for constructing a training set and a test set according to the sample feature data, and using the training set to train each initially constructed machine learning algorithm model to obtain a loss degree discrimination model;

[0104] A first screening unit (33), connected to the model training unit (32), for performing a first screening on the trained loss degree discrimination model using the test set;

[0105] The model training unit (32) is also used to verify the loss degree discrimination model screened out for the first time based on the five-fold cross-validation method;

[0106] A second screening unit (34), respectively connected to the first screening unit (33) and the model training unit (32), for: performing a second screening on the loss degree discrimination model obtained by the five-fold cross-validation method, and jointly constructing the loss degree discrimination model screened out for the second time into a final loss degree discrimination model;

[0107] The model discrimination module is used to: based on the finally constructed loss degree discrimination model, combine the soft voting method to discriminate the loss degree of the knee joint range of motion of the injured person, and output the model discrimination result of the loss degree of the knee joint range of motion.

[0108] The sample feature data formed by the encoded data after data processing is divided into a training set, a validation set, and a test set, and five-fold cross-validation is used to ensure that each piece of data can fully participate in model training and testing. The cross-validation method can enhance the robustness of the model, avoid overfitting of the model, and at the same time ensure the generalization ability of the model on real data. The training set is used for model training, the validation set is used to measure and optimize the performance of the model during the training process, and the test set is used to evaluate the actual performance of the model after training is completed. To address the problem of sample imbalance, the present invention also uses the SMOTE technique to generate virtual samples, increasing the number of minority class samples and introducing more diversity to the classifier, which helps to improve the classifier's ability to recognize the minority class.

[0109] Specifically, first divide the training set and the test set according to 8:2 as the initial screening data set of the model for the first screening of the model. For the models used for ensemble learning after the first screening, the data set is divided using five-fold cross-validation, and the data set division for each fold ensures that all data can be utilized in the training, validation, and testing phases to avoid overfitting of the model, thereby obtaining the second model screening result.

[0110] SMOTE (Synthetic Minority Over-sampling Technique): SMOTE is a method for solving class imbalance in classification problems. It increases the amount of data in the minority class by synthesizing minority class samples, that is, randomly selecting samples within the minority class sample space and generating new samples, thereby improving the model's ability to recognize the minority class.

[0111] In this study, various machine learning algorithms such as SVC, RF, LR, GB, KNN, and XGBoost were used to establish a binary classification discrimination model. Each encoded structured feature, that is, the sample feature information, was used as the input. Among them, the two continuous variables of age and appraisal time limit exist in the form of interval division encoding. Samples with a knee joint range of motion loss less than or equal to 25% were used as the negative class, and those greater than 25% were used as the positive class as sample labels. During model training, first, 80% of the total samples were taken as the training set to train all the initially constructed models, and the remaining samples were used as the test set to test the model performance. Three models with the best performance were selected. In this invention, the best three models, namely SVC, RF, and XGBoost, were selected. The best three machine learning algorithm models can be selected according to the AUC value. Then, 5-fold cross-validation was used to test the classification ability of the selected models again, and the voting method was used to combine the models with good performance in both validations to construct the final discrimination model. Specifically, the 5-fold cross-validation method was used to use 20% of the data for testing each time.

[0112] The research of this invention shows that in the repeated verification of 5-fold cross-validation, the best training models among the three models of SVC, RF, and XGBoost were combined to construct the loss degree discrimination model of this invention.

[0113] SVC (support vector classification): SVC is the application of support vector machine (SVM, Support Vector Machine) in classification tasks. SVM is a supervised learning model that finds a hyperplane to separate data points of different classes in the best way. SVC is especially suitable for high-dimensional spaces and can effectively handle a relatively small number of samples.

[0114] RF (random forest): Random forest is an ensemble learning method that classifies or regresses by constructing multiple decision trees and aggregating their results (usually voting or averaging). Random forest can reduce the overfitting risk of a single decision tree and provide higher accuracy and stability.

[0115] LR (logistic regression): Logistic regression is a widely used statistical analysis method for describing the relationship between data and binary outcomes. Although the name contains the word "regression", it is actually a classification algorithm and is often used to predict the probability of an event occurring.

[0116] GB (gradient boosting): Gradient boosting is a machine learning technique that gradually improves the existing model by iteratively adding new models (usually decision trees). Each newly added model attempts to correct the errors of the previous model, and finally forms a powerful prediction model.

[0117] KNN (k-nearest neighbors): The K-nearest neighbor algorithm is a simple and intuitive classification and regression method. For a new input instance, the KNN algorithm finds the K nearest neighbors in the training set that are most similar to it, and then determines the class or value of the new instance based on the majority class or average value of these neighbors.

[0118] XGBoost (extreme gradient boosting): XGBoost is an implementation of the gradient boosting framework, aiming to optimize speed and performance. It provides an efficient implementation of the gradient boosting algorithm, has regularization functions to help prevent overfitting, and supports parallel processing, making it suitable for large-scale datasets.

[0119] Each validation evaluates the model performance through multiple key metrics, including F1 score, Youden's Index, and Area Under the Curve (AUC) of the Receiver Operating Characteristic curve, etc. These metrics comprehensively reflect the actual performance of the model in complex waveform classification, ensuring the reliability and accuracy of the final judgment results.

[0120] F1 score: The F1 score is the harmonic mean of Precision and Recall, used to measure the accuracy of a classifier. The F1 score ranges from 0 to 1, and the larger the value, the better the classification effect. It is a commonly used evaluation metric when Precision and Recall are equally important.

[0121] Youden's Index: Youden's Index (also known as the informed score or maximum diagnostic benefit) is a linear combination of sensitivity and specificity, used to evaluate the accuracy of a diagnostic test. Its calculation formula is: sensitivity + specificity - 1. The maximum value of Youden's Index is 1, indicating no false positives or false negatives; the minimum value is -1, indicating that all predictions are incorrect.

[0122] AUC (Area Under the Curve): The Area Under the Curve (AUC) of the Receiver Operating Characteristic curve is a metric for evaluating the performance of a binary classification model. The ROC curve plots the variation of the true positive rate against the false positive rate. The AUC value ranges from 0.5 to 1, and the closer the value is to 1, the better the performance of the classifier. An AUC of 0.5 means that the model's prediction effect is no better than random guessing, while an AUC of 1 indicates perfect classification.

[0123] The loss degree discrimination model of the present invention is a binary classification output of reaching the disability level and not reaching the disability level.

[0124] In this study, various injury characteristics that are relatively easy to obtain during the forensic appraisal process were extracted, and a discriminant model was constructed by combining machine learning algorithms. The model has high efficiency in judging whether the degree of loss of knee joint range of motion reaches the disability level, and has good repeatability and strong objectivity.

[0125] Specifically, when formally conducting the appraisal, the knee joint clinical data collected does not include the appraisal opinion letter, and mainly includes personal basic information (such as gender, age, etc.), imaging data and other information. For the data processing method, the original feature information is also extracted manually, such as extracting from personal basic information and extracting by reading imaging data. The original feature information extracted from the knee joint clinical data is also encoded to form the feature data to be discriminated, and the feature data to be discriminated is input into the deployed degree-of-loss discriminant model for discrimination, and finally the discriminant result is output by the model.

[0126] See Figure 4 , the present invention also provides a method for discriminating the degree of loss of knee joint range of motion after trauma, using the aforementioned discriminant system for the degree of loss of knee joint range of motion after trauma, including:

[0127] Step A1, collecting the forensic appraisal archive data of several closed cases regarding the knee joint as the original sample data, and each sample in the original sample data includes the actual measurement result of the degree of loss of knee joint range of motion;

[0128] Step A2, processing the original sample data to obtain sample feature data;

[0129] Step A3, constructing training data according to the sample feature data, using the actual measurement value of the degree of loss of knee joint range of motion as the label, and training the initially constructed machine learning algorithm model to obtain the degree-of-loss algorithm model;

[0130] Step A4, collecting the knee joint clinical data of the injured person, and processing the knee joint clinical data of the injured person to obtain the feature data to be discriminated;

[0131] Step A5, using the deployed and trained degree-of-loss discriminant model to process the feature data to be discriminated, discriminating the degree of loss of knee joint range of motion of the injured person, and outputting the model discriminant result of the degree of loss of knee joint range of motion.

[0132] See Figure 5 , further, step A2 includes:

[0133] Step A21, extracting the original feature information from the original sample data;

[0134] Step A22, encoding the extracted original feature information to obtain the feature encoding information, and the sample feature data is composed of the feature encoding information.

[0135] Furthermore, the forensic appraisal archive data includes appraisal opinions and imaging materials, and one piece of forensic appraisal archive data serves as one sample in the original sample data;

[0136] The imaging materials include archived images and DICOM images;

[0137] In step A21, the extracted original feature information is also stored in the database in the form of a first two-dimensional table. The columns in the first two-dimensional table are the feature fields of the original feature information, and the rows in the first two-dimensional table are each sample;

[0138] In step A21, the archived images are also stored in the database in the form of a second two-dimensional table. The second two-dimensional table records the relative storage paths of the archived images of each sample;

[0139] In step A21, the DICOM images are also stored in the database in the form of a third two-dimensional table. The third two-dimensional table records the relative storage paths of the DICOM images of each sample.

[0140] Through the integration of advanced technologies, the present invention significantly improves the intelligent and automated level of the diagnosis of post-traumatic motor dysfunction of the knee joint, and comprehensively optimizes the accuracy, efficiency and objectivity of the diagnosis.

[0141] Intelligence and automation: By introducing advanced algorithms, machine learning and artificial intelligence technologies, the present invention realizes the precise detection of post-traumatic motor dysfunction of the knee joint. The system automatically analyzes each injury feature, reduces manual intervention, and significantly improves the accuracy and consistency of the diagnosis results. The results of the five-fold cross-validation test show that the AUC reaches 0.91.

[0142] Reducing subjective interpretation errors: The automated system reduces the subjective judgment deviation of the operator in injury analysis. The deep learning model trained with a large amount of data can effectively identify each injury feature and reduce the influence of human factors on the diagnosis results.

[0143] Improving the diagnosis efficiency: The adoption of automated technology significantly improves the speed of the diagnosis process of motor dysfunction, quickly processes a large amount of data, reduces the time for manual analysis and measurement, and at the same time reduces the operation cost. The overall diagnosis efficiency has been significantly improved.

[0144] Enhancing scientificity and objectivity: The intelligent diagnosis system provides analysis results with higher precision, realizes the precise determination of post-traumatic motor dysfunction of the knee joint, improves the scientificity and fairness of forensic appraisal, and provides a more reliable basis for court trials or loss compensation.

[0145] The above are only the preferred embodiments of the present invention, and do not limit the implementation manners and protection scope of the present invention. For those skilled in the art, it should be realized that all the solutions obtained by equivalent substitution and obvious changes made by using the description and illustrations of the present invention should be included within the protection scope of the present invention.

Claims

1. A system for determining the degree of loss of knee joint range of motion after trauma, characterized in that: include: A data collection module, used to collect a number of closed forensic identification file data on knee joints as original sample data, wherein the forensic identification file data includes actual measurement results of the degree of loss of knee joint range of motion; A data processing module, connected to the data acquisition module, for processing the original sample data to obtain sample characteristic data; A model training module, connected to the data processing module, for constructing training data according to the sample feature data, using the actual measured value of the degree of loss of knee joint activity as a label, so as to train the initially constructed machine learning algorithm model to obtain a loss degree discrimination model; The data acquisition module is also used to collect clinical data of the knee joint of the injured person, and the data processing module is also used to process the clinical data of the knee joint of the injured person to obtain feature data to be identified; The model discrimination module is also connected to the data processing module, and is used to use the deployed and trained loss degree discrimination model to process the feature data to be discriminated, discriminate the degree of loss of knee joint range of motion of the injured person, and output the model discrimination result of the degree of loss of knee joint range of motion.

2. A system for determining the degree of loss of knee joint range of motion after trauma as claimed in claim 1, characterized in that: The data processing module comprises: A feature extraction unit, used to extract original feature information from the original sample data; The feature encoding unit is connected to the feature extraction unit and is used to encode the extracted original feature information to obtain feature encoding information, and the feature encoding information constitutes the sample feature data.

3. A system for determining the degree of loss of knee joint range of motion after trauma as claimed in claim 2, characterized in that: The system also includes a database; The judicial appraisal file data includes the appraisal opinion and imaging data, and one copy of the judicial appraisal file data is used as one sample of the original sample data; The imaging data include archived images and DICOM images; The data processing module is also connected to the database; The feature extraction unit stores the extracted original feature information in the database in the form of a first two-dimensional table, wherein the columns in the first two-dimensional table are feature fields of the original feature information, and the rows in the first two-dimensional table are the samples; The data processing module further includes a data storage unit, which is used to: storing the archived image pieces in the database in the form of a second two-dimensional table, wherein the second two-dimensional table records the relative storage paths of the archived image pieces of each sample; The DICOM image slices are stored in the database in the form of a third two-dimensional table, and the third two-dimensional table records the relative storage path of the DICOM image slices of each sample.

4. A system for determining the degree of loss of knee joint range of motion after trauma as claimed in claim 2, characterized in that: The original characteristic information includes the actual measurement value of the degree of loss of knee joint range of motion, basic personal information, fracture site, fracture type, fracture healing status, whether the implant is in place, whether surgical treatment is performed, whether it is displaced, whether the joint surface is involved, and the time interval between the first and last radiographs.

5. A system for determining the degree of loss of knee joint range of motion after trauma as claimed in claim 1, characterized in that: The model training module includes: A model building unit, used to build multiple initial loss degree discrimination models based on multiple machine learning algorithms; A model training unit, connected to the model building unit, is used to build a training set and a test set according to the sample feature data, and use the training set to train each of the initially built machine learning algorithm models to obtain a loss degree discrimination model; A first screening unit, connected to the model training unit, for performing a first screening on the trained loss degree discrimination model using the test set; The model training unit is also used to verify the loss degree discrimination model screened out for the first time based on a five-fold cross validation method; The second screening unit is connected to the first screening unit and the model training unit, respectively, and is used to: perform a second screening on the loss degree discrimination model obtained by using the five-fold cross validation method, and jointly construct the loss degree discrimination model obtained by the second screening into a final loss degree discrimination model; The model discrimination module is used to: discriminate the degree of loss of knee joint activity of the injured person based on the finally constructed loss degree discrimination model in combination with the soft voting method, and output the model discrimination result of the degree of loss of knee joint activity.

6. A system for determining the degree of loss of knee joint range of motion after trauma as claimed in claim 4, characterized in that: The feature coding unit codes the actual measured value of the degree of knee joint range of motion loss according to whether it reaches the disability level; During training, the model training module uses the coding result of the actual measurement value of the degree of knee joint range of motion loss that reaches the disability level as the label of the positive sample, and uses the coding result that does not reach the disability level as the label of the negative sample, thereby training the initially constructed loss degree discrimination model.

7. A system for determining the degree of loss of knee joint range of motion after trauma as claimed in claim 4, characterized in that: The basic personal information includes age; The feature encoding unit encodes the age according to age segments.

8. A method for determining the degree of loss of knee joint range of motion after trauma, characterized in that: A system for determining the degree of loss of range of motion of a knee joint after trauma as claimed in any one of claims 1 to 7, comprising: Step A1, collecting a number of closed forensic identification file data on knee joints as original sample data, wherein each sample in the original sample data includes an actual measurement result of the degree of loss of knee joint range of motion; Step A2, processing the original sample data to obtain sample feature data; Step A3, constructing training data according to the sample feature data, using the actual measured value of the knee joint range of motion loss degree as a label to train the initially constructed machine learning algorithm model to obtain a loss degree algorithm model; Step A4, collecting clinical data of the knee joint of the injured person, and processing the clinical data of the knee joint of the injured person to obtain feature data to be identified; Step A5, using the deployed trained loss degree discrimination model to process the feature data to be discriminated, discriminate the degree of loss of knee joint range of motion of the injured person, and output the model discrimination result of the degree of loss of knee joint range of motion.

9. A method for determining the degree of loss of knee joint range of motion after trauma as claimed in claim 8, characterized in that: The step A2 comprises: Step A21, extracting original feature information from the original sample data; Step A22: Encode the extracted original feature information to obtain feature coding information, and use the feature coding information to form the sample feature data.

10. A method for determining the degree of loss of knee joint range of motion after trauma as claimed in claim 9, characterized in that: The judicial appraisal file data includes the appraisal opinion and imaging data, and one copy of the judicial appraisal file data is used as one sample of the original sample data; The imaging data include archived images and DICOM images; In the step A21, the extracted original feature information is also stored in a database in the form of a first two-dimensional table, wherein the columns in the first two-dimensional table are feature fields of the original feature information, and the rows in the first two-dimensional table are the samples; In the step A21, the archived image pieces are also stored in the database in the form of a second two-dimensional table, and the second two-dimensional table records the relative storage paths of the archived image pieces of each sample; In the step A21, the DICOM image slices are also stored in the database in the form of a third two-dimensional table, and the third two-dimensional table records the relative storage path of the DICOM image slices of each sample.

Images