Method, system and device for constructing temporal-mandibular joint disc anterior displacement screening model

Abstract

A method for constructing a temporal-mandibular joint disc anterior displacement screening model comprises the following steps: collecting clinical examination data and head shadow measurement data of a qualified patient, taking a temporal-mandibular joint nuclear magnetism upper temporal-mandibular joint disc anterior displacement diagnosis result as a real label, screening features for screening, and finally constructing the model. In the model construction process, a basic model based on the constructed model is determined firstly, then model parameters are defined respectively, and a required screening model is formed. After network deployment of the constructed model, primary hospitals and clinics can also use related indexes of a skull lateral position film in orthodontic routine examination and clinical examination results to screen the risk of temporal-mandibular joint disc anterior displacement before orthodontic treatment, so that the popularization rate of joint screening before orthodontic treatment can be greatly improved; therefore, the orthodontic treatment effect is ensured.

Description

Technical Field

[0001] This invention relates to a technical solution for screening the occurrence of diseases, and more particularly to a method for constructing a model for screening anterior displacement of the temporomandibular joint disc, as well as the construction system and apparatus thereof. Background Technology

[0002] Temporomandibular joint anterior disc displacement (TMJ ADD) is one of the most common clinically diagnosed types of temporomandibular joint disorders (TMD). According to current medical consensus, ADD can be defined as a pathological condition in which the temporomandibular joint disc shifts anteriorly relative to the condyle. Based on the disc repositioning characteristics, it can be divided into two clinical subtypes: irreversible anterior displacement (DDWoR) and reversible anterior displacement (DDWR). The prevalence of this condition in the general population is as high as 5-12%, accounting for approximately 41% of clinically diagnosed TMD cases. About one-third of these patients have no obvious clinical symptoms, and this insidious development greatly increases the difficulty of early diagnosis.

[0003] Numerous studies have shown that TMJ ADD can affect the normal development of the condylar bone and cartilage, and is associated with degenerative joint disease and reduced mandibular ramus height. This effect is even greater in adolescents who are still growing and developing, leading not only to impaired mandibular function but also to dentofacial deformities, airway narrowing, and even sleep apnea syndrome.

[0004] Orthodontic patients often have temporomandibular joint (TMD) problems. Ignoring the temporomandibular joint issue and directly proceeding with orthodontic or orthognathic treatment may lead to undesirable results, such as recurrence of malocclusion and temporomandibular joint pain.

[0005] Although temporomandibular joint MRI (TMJ MRI) is considered the gold standard for assessing the structure of the temporomandibular joint, accurately determining the position of the articular disc, the condition of the cortical bone, and joint effusion, its high cost makes it difficult to use as a routine screening method, resulting in asymptomatic patients not receiving early diagnosis. Secondly, while the diagnostic criteria for temporomandibular joint disorder (DC / TMD) have good validity for diagnosing myalgia, their reliability in diagnosing disc displacement is insufficient, and the consistency between clinical examination and MRI results is poor, particularly with a high misdiagnosis rate for irreversible anterior disc displacement.

[0006] On the one hand, MRI is considered the gold standard for assessing the structure of the temporomandibular joint, but its high testing cost makes it difficult to use as a routine screening method, resulting in asymptomatic patients and orthodontic patients who need to be screened for joint problems having difficulty getting an early diagnosis.

[0007] On the other hand, the widely used diagnostic criteria for temporomandibular joint disorder (DC / TMD) have good diagnostic validity for myalgia, but their diagnostic reliability for disc displacement is insufficient, and the rate of missed diagnosis for reversible and irreversible anterior disc displacement is relatively high.

[0008] Patients with TMJ ADD have unique craniofacial morphological features. Therefore, there is an urgent need to establish an effective risk screening model based on clinical examination and lateral cephalometric radiographs to enable early identification and intervention during orthodontic treatment. Summary of the Invention

[0009] One objective of this invention is to provide a method for constructing a screening model for anterior temporomandibular joint disc displacement (TMJ ADD) in order to achieve early screening of TMJ ADD.

[0010] Another objective of this invention is to provide a system for screening anterior displacement of the temporomandibular joint disc, thereby improving the convenience and automation of the screening process.

[0011] Another objective of this invention is to provide a device for screening anterior temporomandibular joint disc displacement, which, as a medical device, facilitates the screening and early diagnosis of anterior temporomandibular joint disc displacement in clinical practice.

[0012] Lateral cephalometric radiographs are of significant value in routine monitoring of facial and occlusal changes in orthodontic patients. Therefore, using cephalometric data combined with clinical examination indicators to detect articular disc displacement is both scientifically sound and practically feasible. Given that most orthodontic patients undergo lateral cephalometric radiographs, utilizing existing cephalometric parameters for TMJ ADD detection can significantly improve the efficiency of clinicians in screening and monitoring patients' joint conditions.

[0013] A method for constructing a screening model for anterior temporomandibular joint disc displacement (TMJ ADD) is proposed. Using TMJ MRI results as the gold standard, a machine learning screening model based on clinical examination indicators and cephalometric indicators is developed to achieve early screening of TMJ ADD. The steps include:

[0014] The methods for constructing the TMJ ADD screening model include, in order:

[0015] First, acquire data by collecting clinical examination data and cephalometric data from qualified patients;

[0016] Further filter features by selecting features from patient data for screening;

[0017] Finally, the model is constructed. First, the basic model on which the constructed model is based is determined, then the parameters of each model are defined, and the required screening model is formed.

[0018] The method of the present invention includes deploying a screening model on a terminal device or network platform and displaying the model on a human-machine interface.

[0019] The method of this invention uses the diagnostic results of temporomandibular joint disc anterior displacement on temporomandibular joint MRI as the true label.

[0020] The method of the present invention involves obtaining TMJ MRI and lateral cephalometric radiographs of at least 200 patients, and randomly dividing the patients into a training set and an internal validation set at a ratio of 7:3.

[0021] The clinical examination features of the method of the present invention include: mouth opening path, joint clicking, chin deviation, mouth opening degree, protrusion movement, left lateral movement of the mandible, right lateral movement of the mandible, posterior crossbite, posterior reverse crossbite, anterior open bite, upper tooth crowding degree, and lower tooth crowding degree.

[0022] The method of the present invention includes the following characteristics for cephalic shadow measurement: ANB angle, ALFH / PLFH ratio, ANS-Me / Na-Me ratio, Co-Go, FMA angle, FH-NA angle, FH-NPo angle, FMIA angle, IMPA angle, LL-EP distance, L1-MP distance, L1-NB angle, MP-SN angle, NA-Apo angle, OP-FH angle, Overbite, Overjet, Po-NB distance, S-Go / N-Me ratio, S Vert-Co distance, SN distance, SNB angle, SN / GoMe angle, UL-EP distance, ULL, U1-L1 angle, U1-NA angle, U1-PP distance, U6-PP distance, Wits value, Y-axis angle, G'-Sn-Pog' angle, N'-Vert-Pog' distance, Sn to G Vert distance, Pog' to G Vert distance, ANS-Me, Na-Me, S-Go, and Go-Me. This information was obtained from lateral cephalometric radiographs (X-rays).

[0023] The method of this invention uses LASSO regression and Boruta algorithm to screen key features from patient data and then takes the intersection of the features.

[0024] In the LASSO regression, λ = 0.029 was set.

[0025] The method of the present invention, through feature screening, obtains nine key features, namely: mouth opening, left mandibular lateral movement, right mandibular lateral movement, upper dentition crowding, Wits value, FH-Npo angle, Co-Go, ALFH / PLFH ratio, and G'-Sn-Pog' angle.

[0026] The method of this invention uses an extreme gradient boosting model as its base model, and the relevant parameters set include: booster, objective, eval_metric, eta, max_depth, min_child_weight, subsample, colsample_bytree, colsample_bylevel, alpha, nfold, and validate_parameters.

[0027] The method of this invention uses an extreme gradient boosting model as its base model, and the parameter values ​​for the relevant parameters are as follows:

[0028] The Booster parameter value is: gbtree;

[0029] The objective parameter value is: binary:logistic;

[0030] The value of the eval_metric parameter is: logloss;

[0031] The eta parameter value is: 0.05;

[0032] The value of the max_depth parameter is 4;

[0033] The value of the min_child_weight parameter is: 1;

[0034] The subsample parameter value is: 0.8;

[0035] The value of the colsample_bytree parameter is 0.8;

[0036] The value of the colsample_bylevel parameter is 0.4;

[0037] The value of the alpha parameter is 0;

[0038] The value of the nfold parameter is 5;

[0039] The validate_parameters parameter value is TRUE.

[0040] The method of the present invention, by inputting the corresponding values ​​of the above nine feature variables, outputs the risk probability of TMJ ADD of the sample by the constructed screening model.

[0041] To achieve visualization, the Shapley Additive Interpretation (SHAP) framework is used to interpret the model, outputting the feature results of the constructed algorithm model.

[0042] The screening model constructed using the method of this invention is deployed on a network platform to form a screening system for use by clinicians. After inputting the corresponding values ​​of nine feature variables into the human-computer interface, feedback information on the risk of TMJ ADD in the sample can be obtained from the screening system on the human-computer interface.

[0043] The screening model constructed by the method of this invention is loaded into a screening device. By directly reading the corresponding values ​​of the nine characteristic variables, feedback information on the TMJ ADD risk of the sample can be obtained on the human-machine interface of the device.

[0044] The device typically includes interfaces for acquiring data, such as Bluetooth, infrared, Wi-Fi, or hardware ports, allowing the device to directly acquire data generated by other devices (such as MRI scans). This facilitates screening within a closed environment and prevents patient data from being leaked on network platforms.

[0045] To implement the method of the present invention, an apparatus or system is also required to facilitate the rapid and efficient generation of the screening model of the present invention. Such an apparatus or system includes:

[0046] Data acquisition module: It is used to collect clinical examination data and cephalometric data from qualified patients;

[0047] Feature filtering module: It uses one or more algorithms to filter key features for screening;

[0048] Model building module: It determines the base model from various known models, then determines the values ​​of each parameter of the base model accordingly, and generates a screening model based on the selected features;

[0049] The results output module outputs feedback on the input data based on the screening model in the human-machine interface, making the TMJ ADD risk probability calculation visual.

[0050] Temporomandibular joint (TMJ) structural stability is crucial for orthodontic treatment outcomes. The presence of temporomandibular joint (TMJ) aversion disorder (ADD) can lead to recurrence of malocclusion after orthodontic treatment, easily causing medical disputes. TMJ MRI is expensive and not widely available. The model constructed using the method of this invention and its online deployment allow primary care hospitals and clinics to screen for the risk of anterior temporomandibular joint disc displacement (TMJ disc ADD) before orthodontic treatment using relevant indicators from routine orthodontic examinations such as lateral cephalometric radiographs and clinical examination results. This significantly improves the accessibility of pre-orthodontic joint screening, thereby ensuring the effectiveness of orthodontic treatment.

[0051] Verification has shown that the screening model of this invention can reduce the cost of early screening and improve diagnostic accessibility. Addressing the current situation where MRI examinations are expensive and difficult to use as a routine screening method, this invention combines clinical examination and cephalometric data to construct a low-cost, high-efficiency screening system, enabling asymptomatic patients and patients requiring joint assessment before orthodontic treatment to have early screening opportunities.

[0052] On the other hand, the screening model of this invention improves the accuracy of disc displacement screening. Addressing the insufficient reliability of existing DC / TMD standards for diagnosing disc displacement, this invention optimizes the diagnostic model by applying TMJ MRI results as the gold standard, reducing the misdiagnosis rate and improving the ability to differentiate disc displacement. Attached Figure Description

[0053] Figure 1 Flowchart for constructing a screening model for anterior displacement of the temporomandibular joint disc;

[0054] Figure 2 This is a summary map of the features output by the screening model built based on extreme gradient boosting, interpreted using the SHAP framework; where A is the honeycomb map and B is the feature importance map.

[0055] Figure 3 A single-sample explanation diagram for the model;

[0056] Figure 4 For the subject operating characteristic curves of the internal and external validation datasets;

[0057] Figure 5 Operating characteristic curves for internal and external asymptomatic subset subjects;

[0058] Figure 6 This represents the clinical decision curve. Detailed Implementation

[0059] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings. The embodiments of the present invention are only used to illustrate the technical solution of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solution of the invention without departing from the spirit and scope of the technical solution of the present invention, and all such modifications and substitutions should be covered within the scope of the claims of the present invention.

[0060] Patients with TMJ ADD have unique craniofacial morphological features. Therefore, there is an urgent need to establish an effective risk screening model based on clinical examination and lateral cephalometric radiographs to enable early identification and intervention during orthodontic treatment.

[0061] This embodiment uses the diagnostic results of anterior temporomandibular joint disc displacement on temporomandibular joint MRI as the true label for model construction, and includes the following steps:

[0062] First, acquire data by collecting clinical examination data and cephalometric data from qualified patients;

[0063] Further feature selection was performed, using LASSO regression and Boruta algorithm to select key features from the patient data, and the intersection of the features was taken.

[0064] Next, the model is built. First, the basic model on which the model is based is determined, then the parameters of each model are defined, and the required screening model is formed.

[0065] Finally, the screening model is deployed on terminal devices or network platforms and displayed on the human-machine interface, outputting the TMJ ADD risk probability and a visual explanation (e.g., using the SHAP framework to provide model explanation, including feature contribution analysis and individualized risk explanation).

[0066] The selected features are used to build a screening model based on the extreme gradient boosting (XGBoost) model.

[0067] Patients were randomly assigned to the training set and the internal validation set in a 7:3 ratio. All patients underwent clinical examination, high-quality TMJ MRI, and lateral cephalometric radiographs. The joint condition of all patients was diagnosed based on the TMJ MRI.

[0068] Key factors were screened using LASSO (Least Absolute Contraction Selection) regression and the Boruta algorithm in the training set, and six machine learning methods were used to construct screening models. Internal and external validation of the models was performed using the area under the receiver operating characteristic (AUC) curve, and the optimal model was selected. Finally, the optimal model was deployed as a user-friendly web interface for clinical application.

[0069] The screening model of this invention is constructed, validated, visualized, and deployed on the network using R (version 4.2.2; RFoundation for Statistical Computing, Vienna, Austria).

[0070] The method for constructing the temporomandibular joint disc anterior displacement screening model in this embodiment is as follows: Figure 1 As shown, it specifically includes:

[0071] 1. Data Acquisition

[0072] Raw data of 1,720 patients who underwent clinical examinations between 2009 and 2023 were collected. After screening the raw data, clinical examination data and cephalometric data of 532 patients were retained. The true labels of all patients were obtained by temporomandibular joint surgeons based on TMJ MRI diagnosis.

[0073] The data selection criteria are as follows:

[0074] 1) Age between 18 and 50 years old;

[0075] 2) MRI examination is performed within one week before or one week after lateral cephalometric angiography.

[0076] The data exclusion criteria meet one of the following:

[0077] 1) Blurred images on MRI or lateral cephalometric radiographs (unable for doctors to interpret);

[0078] 2) Maxillofacial deformities;

[0079] 3) History of craniofacial surgery or orthodontic treatment;

[0080] 4) History of temporomandibular joint treatment;

[0081] 5) There are obvious symptoms of joint pain, and treatment is urgently needed.

[0082] (1) Clinical examination characteristics and data obtained, including:

[0083] age;

[0084] gender;

[0085] Opening path: Observe whether there is deviation when the mandible opens and closes (normal / deviation);

[0086] Joint clicking: Check for clicking during temporomandibular joint movement (yes / no);

[0087] Chin deviation: Measure the distance (mm) that the chin deviates from the midline of the face;

[0088] Mouth opening: The distance between the upper and lower incisors when the mouth is fully open (mm);

[0089] Protrusion movement: Record the distance between the incisal edges of the upper and lower central incisors at rest, and then measure it again at maximum protrusion. Calculate the protrusion range (mm) by subtracting the resting measurement from the maximum protrusion measurement.

[0090] Lateral movement of the mandible on the left: displacement (mm) of the mandibular central incisor during the maximum left lateral movement;

[0091] Lateral movement of the mandible on the right: displacement (mm) of the mandibular central incisor during the maximum right lateral movement;

[0092] Posterior crossbite: The lingual slope of the lingual cusp of the maxillary posterior teeth is located on the buccal side of the buccal slope of the buccal cusp of the mandibular posterior teeth (yes / no);

[0093] Posterior crossbite: One or more maxillary posterior teeth bite on the lingual side of the lower posterior teeth (yes / no);

[0094] Anterior open bite: Check for vertical gaps between the upper and lower anterior teeth at the point of cusp intercuspation (yes / no);

[0095] Maxillary crowding: The degree of crowding of the maxillary teeth, measured in mm by the total difference between available space and the space required for proper tooth alignment; and

[0096] Mandibular crowding: The degree of crowding of the mandibular teeth is measured by the total difference (mm) between the available space and the space required for proper tooth alignment.

[0097] (2) Cephalometric features, including:

[0098] ANB angle: The angle between maxillary point A (upper alveolar seat) - N (nasal root point) - mandibular point B (lower alveolar seat), reflecting the anterior-posterior relationship between the maxilla and mandible;

[0099] ALFH / PLFH ratio: The ratio of the anterior inferior height to the posterior inferior height, used to assess the vertical proportion of the face; draw perpendicular lines from points A and Ptm (pterygomaxillary cleft) to the palatal plane (ANS-PNS), with the feet of the perpendiculars being A' and Ptm'; draw perpendicular lines from points B and J (inner mandibular angle) to the mandibular plane (Go-Gn), with the feet of the perpendiculars being B' and J'; the anterior inferior height is the distance between A' and B'; the posterior inferior height is the distance between Ptm' and J'.

[0100] ANS-Me / Na-Me ratio: The lower part of the face is higher than the overall height, reflecting the vertical development of the face;

[0101] Co-Go: Distance from the condyle to the angle of the mandible, representing the length of the mandibular ramus;

[0102] FMA angle: The angle between the orbitoauricular plane and the mandibular plane, used to determine vertical growth pattern;

[0103] FH-NA angle: The angle between the orbitoauricular plane and the NA line, used to assess the position of the maxilla;

[0104] FH-NPo angle: The angle between the orbitoauricular plane and the NPog line, used to assess the position of the chin;

[0105] FMIA angle: The angle between the orbitoauricular plane and the long axis of the mandibular incisor, reflecting the inclination of the mandibular incisor;

[0106] IMPA angle: The angle between the mandibular plane and the long axis of the mandibular incisor, used to assess mandibular incisor tilt;

[0107] LL-EP distance: The distance from the lower lip to the aesthetic plane, used to measure the protrusion of the lower lip;

[0108] L1-MP distance: Vertical distance from the incisal edge of the lower incisor to the mandibular plane;

[0109] L1-NB angle: The angle between the long axis of the mandibular incisor and the NB line;

[0110] MP-SN angle: The angle between the mandibular plane and the SN plane, used to assess vertical growth pattern;

[0111] NA-Apo angle: The angle between the NA line and the aesthetic plane;

[0112] OP-FH angle: the angle between the occlusal plane and the orbitoauricular plane;

[0113] Overbite: Vertical overlap of anterior teeth;

[0114] Overjet: the horizontal overlap of the anterior teeth;

[0115] Po-NB distance: The perpendicular distance from the anterior chin point to the NB line;

[0116] S-Go / N-Me ratio: The ratio of the lower rear height to the overall front height;

[0117] S Vert-Co distance: the horizontal distance from the vertical line of the saddle to the condyle;

[0118] SN distance: the distance from the sella turcica to the root of the nose, used to measure the length of the anterior skull base;

[0119] SNB angle: Angle between the saddle point, the root of the nose, and the mandibular point B, used to assess the position of the mandible;

[0120] SN / GoMe angle: The angle between the SN plane and the mandibular plane;

[0121] UL-EP distance: The distance from the upper lip to the aesthetic plane, used to measure the protrusion of the upper lip;

[0122] ULL: Upper lip length (from the base of the nose to the lower edge of the lip);

[0123] U1-L1 angle: the angle between the long axes of the upper and lower incisors, reflecting the relationship between the anterior teeth;

[0124] U1-NA angle: The angle between the long axis of the upper incisor and the NA line;

[0125] U1-PP distance: The vertical distance from the incisal edge of the maxillary incisor to the palatal plane;

[0126] U6-PP distance: Vertical distance from the maxillary molar to the palatal plane;

[0127] Wits value: AO-BO distance, used to assess the anterior-posterior relationship of the jaw;

[0128] Y-axis angle: The angle between the line connecting the saddle point and the chin vertex and the orbitoauricular plane;

[0129] G'-Sn-Pog' angle: the angle between the frontal point, the subnasal point, and the anterior chin point, used to assess soft tissue profile;

[0130] N'-Vert-Pog' distance: distance from the soft tissue anterior chin point to the vertical reference line;

[0131] Sn to G Vert distance: the vertical distance from the subnasal point to the forehead point;

[0132] Pog'to G Vert distance: the vertical distance from the premental point of the soft tissue to the forehead point;

[0133] ANS-Me: Distance from the anterior nasal spine to the submental point (anterior inferior level);

[0134] Na-Me: Distance from the root of the nose to the submental point (frontal facial height);

[0135] S-Go: Distance from the saddle point to the mandibular angle point (posterior lower elevation); and

[0136] Go-Me: Distance from the angle of the mandible to the submental point (length of the mandibular body).

[0137] 2. Screen and identify key features related to TMJ ADD.

[0138] The 532 patients were randomly divided into a training set (373 cases) and an internal validation set (159 cases) at a ratio of 7:3.

[0139] LASSO regression and Boruta algorithm were used to select features from the training set data for the above two types of indicators (a total of 53 candidate features):

[0140] When λ = 0.029, 14 features were identified by LASSO regression (namely, chin deviation, clicking, upper dentition crowding, Wits value, G'-Sn-Pog' angle, ANB angle, ALFH / PLFH ratio, Overjet, FMIA angle, FH-NPo angle, mouth opening, Co-Go, left mandibular lateral movement, and right mandibular lateral movement).

[0141] The Boruta algorithm identified 22 features (i.e., upper dentition crowding, left mandibular lateral movement, right mandibular lateral movement, Wits value, G'-Sn-Pog' angle, ANB angle, Overjet, NA-Apo angle, N'-Vert-Pog' distance, FH-NPo angle, FMIA angle, mouth opening, Co-Go, ALFH / PLFH ratio, SNB angle, Pog' to G Vert distance, MP-SN angle, S-Go / N-Me ratio, L1-NB angle, FMA angle, Y-axis angle, L1-NB angle);

[0142] The intersection of the two algorithms is taken, and convergence is achieved to 12 features;

[0143] Pearson correlation analysis was performed on these 12 features. Among the features with a correlation coefficient greater than 0.7, the features with higher importance as assessed by the Boruta algorithm were retained. Finally, 9 features were determined as the input features required for screening, namely: mouth opening, left mandibular lateral movement, right mandibular lateral movement, upper dentition crowding, Wits value, FH-NPo angle, Co-Go, ALFH / PLFH ratio, and G'-Sn-Pog' angle.

[0144] 3. Construct a screening algorithm model

[0145] After validation with six algorithm models, the model was validated internally and externally using the area under the receiver operating characteristic curve (AUC). Ultimately, the extreme gradient boosting model (XGBoost) was determined. The main parameters and their settings are shown in Table 1 below, and a model file for screening in this embodiment was generated.

[0146] Table 1

[0147]

[0148] 4. Algorithm Model

[0149] The Shapley Additive Interpretation (SHAP) framework is used to interpret the model, and the output is the feature result of the constructed algorithm model. A summary plot of the nine selected features is shown below. Figure 2 As shown.

[0150] To further interpret the model, this embodiment also employed an individualized interpretation method to conduct a model interpretability analysis on a randomly selected asymptomatic individual (with bilateral irreducible anterior disc displacement verified by MRI) (see [link to relevant documentation]). Figure 3 Patient numbered Sample 5, although without joint symptoms, had a 97.1% risk of developing the disease, suggesting a possible hidden joint structural abnormality. The contribution of various indicators to the risk is as follows: Figure 3 .

[0151] Example 3 Model Deployment

[0152] Install R software on the terminal, and the obtained screening model file can be used for screening based on R software.

[0153] Furthermore, the Shiny platform is an open-source framework within the R language ecosystem for building interactive web applications. It allows users to quickly create interactive data visualization applications using only R code, without needing in-depth knowledge of front-end technologies such as HTML, CSS, and JavaScript. Therefore, the model files obtained in the above embodiments can also be deployed on the Shiny platform, enabling clinicians to directly perform clinical screening without installing R software or possessing programming expertise. For example, screening can be achieved by inputting relevant feature values ​​through a human-computer interface.

[0154] Example 4 Model Validation

[0155] The internal data refers to the aforementioned internal validation dataset composed of the number of patients in the three tiers.

[0156] The external validation dataset consisted of 81 patients collected and provided by the applicant's partner hospital, used to validate the universality of the screening model.

[0157] The receiver operating characteristic (ROC) curves show that the area under the curve (AUC) for the model is 0.890 (0.832–0.947) on the internal validation dataset and 0.862 (0.780–0.943) on the external validation dataset. Figure 4 As shown.

[0158] The receiver operating characteristic (ROC) curve showed that the area under the curve (AUC) for the model was 0.953 (0.910–0.997) in the internal asymptomatic subset and 0.879 (0.761–0.943) in the external asymptomatic subset. Figure 5 As shown.

[0159] The obtained screening model was tested and found to have an accuracy of 89.1%, precision of 89.1%, recall of 95.3%, and F1 score of 92.1% for early screening.

[0160] like Figure 6 As shown, the decision curve analysis indicates that the obtained screening model has significant clinical practical value, exhibiting a higher net benefit rate across a wide threshold range (17%–96%).

[0161] The economic benefits of the deployed screening model are detailed in Table 2 below.

[0162] Table 2

[0163]

Claims

1. A method for constructing a screening model for anterior displacement of the temporomandibular joint disc, characterized in that, include: First, acquire data by collecting clinical examination data and cephalometric data from qualified patients; Further filter features by selecting features from patient data for screening; Finally, the model is constructed. First, the basic model on which the constructed model is based is determined, then the parameters of each model are defined, and the required screening model is formed.

2. The method according to claim 1, characterized in that, The results of anterior temporomandibular joint disc displacement diagnosis from temporomandibular joint MRI were used as the true labels for model construction.

3. The method according to claim 1, characterized in that, Clinical examination features include: mouth opening path, joint clicking, chin deviation, mouth opening degree, protrusion movement, left lateral mandibular movement, right lateral mandibular movement, posterior crossbite, posterior reverse crossbite, anterior open bite, upper and lower tooth crowding.

4. The method according to claim 1, characterized in that, Characteristics of cephalic shadow measurements include: ANB angle, ALFH / PLFH ratio, ANS-Me / Na-Me ratio, Co-Go, FMA angle, FH-NA angle, FH-NPo angle, FMIA angle, IMPA angle, LL-EP distance, L1-MP distance, L1-NB angle, MP-SN angle, NA-Apo angle, OP-FH angle, Overbite, Overjet, Po-NB distance, S-Go / N-Me ratio, S Vert-Co distance, SN distance, SNB angle, SN / GoMe angle, UL-EP distance, ULL, U1-L1 angle, U1-NA angle, U1-PP distance, U6-PP distance, Wits value, Y-axis angle, G'-Sn-Pog' angle, N'-Vert-Pog' distance, Sn to G Vert distance, Pog' to G Vert distance, ANS-Me, Na-Me, S-Go, and Go-Me.

5. The method according to claim 1, characterized in that, LASSO regression and Boruta algorithm were used to screen key features from patient data, and the intersection of the features was taken.

6. The method according to claim 5, characterized in that... In the LASSO regression, λ = 0.029 was set.

7. The method according to claim 1, characterized in that... The screening model outputs the TMJ ADD risk probability and a visual explanation based on the nine key input features. The key features are mouth opening, left mandibular lateral movement, right mandibular lateral movement, upper dentition crowding, Wits value, FH-NPo angle, Co-Go, ALFH / PLFH ratio, and G'-Sn-Pog' distance.

8. The method according to claim 1, characterized in that, The base model is the XGBoost model, and the relevant parameters include: booster, objective, eval_metric, eta, max_depth, min_child_weight, subsample, colsample_bytree, colsample_bylevel, alpha, nfold, and validate_parameters.

9. The method according to claim 8, characterized in that, The parameter values ​​for the relevant settings are as follows: The Booster parameter value is: gbtree; The objective parameter value is: binary:logistic; The value of the eval_metric parameter is: logloss; The eta parameter value is: 0.05; The value of the max_depth parameter is 4; The value of the min_child_weight parameter is: 1; The subsample parameter value is 0.8; The value of the colsample_bytree parameter is 0.8; The value of the colsample_bylevel parameter is 0.4; The value of the alpha parameter is 0; The value of the nfold parameter is 5; The validate_parameters parameter value is TRUE.

10. A TMJ ADD screening model, characterized in that, Obtained by the method according to any one of claims 1 to 9.

11. A system for screening TMJ ADDs, deployed on a network platform, characterized in that, Includes the screening model described in any one of claims 1 to 9.

12. A device for screening TMJ ADD, characterized in that, Includes the screening model described in any one of claims 1 to 9.

13. A medical device, characterized in that, Includes the screening model described in any one of claims 1 to 9.

Images