Temporomandibular joint disorder prediction method and system based on artificial intelligence

Abstract

The invention relates to the technical field of medical imaging, and discloses an artificial intelligence-based temporomandibular joint disorder prediction method, which comprises the following steps: acquiring an MRI image of a patient; performing data resampling according to the average spatial resolution of the patient image; cutting the patient data three-dimensional image layer by layer to obtain cut patient three-dimensional image data; introducing the cut three-dimensional image data of the patient into an artificial intelligence algorithm; and grouping verification is carried out on the cut three-dimensional image data of the patient, so that a temporomandibular joint disorder classification prediction model is established for prediction. The application proposes that MRI iconomics examination has significant advantages in diagnosis of temporomandibular joint disorder, data are collected to construct an MRI iconomics diagnosis model, clinical factors of a patient suffering from temporomandibular joint disorder are combined with an iconomics score calculation formula (Radio-score, Rad-score), an iconomics-clinical combined prediction model of temporomandibular joint disorder cases is established, and the prediction accuracy of the temporomandibular joint disorder cases is improved. And a powerful tool is provided for accurate diagnosis of temporomandibular joint disorder.

Description

Technical Field

[0001] The present invention relates to the technical field of large medical models, and in particular to a prediction method and system for temporomandibular joint disorder based on artificial intelligence. Background Art

[0002] Temporomandibular disorders (TMD) are a group of diseases that involve the temporomandibular joint (TMJ) and / or the masticatory muscle system and have the same or similar clinical symptoms. The main symptoms include pain in the joint area and / or the masticatory muscles, mandibular movement disorders, joint snapping and murmurs, and may be accompanied by ear, shoulder, neck and head pain symptoms. As the most common chronic painful disease in the oral and maxillofacial region, it can seriously affect the patient's work and life and reduce the oral health-related quality of life. However, a considerable number of dentists are currently unable to accurately interpret MRI and other images of temporomandibular joint lesions, and misdiagnosis is prone to occur during the diagnosis process. Since the above misdiagnosis can easily lead to poor prognosis and adverse clinical outcomes for patients, early and accurate diagnosis of TMD is particularly important for the prognosis and outcome of the patient's disease.

[0003] Studies have shown that imaging examinations can assist in the diagnosis of disc displacement and joint degenerative diseases. Among them, MRI has been widely used in the clinical examination of TMD. However, there is currently no universal inspection specification in the industry in China. The quality of some MRI image data cannot meet the requirements of clinical diagnosis. Ineffective inspections and repeated inspections often occur, and may even lead to incorrect diagnosis and treatment. Radiomics, which is a multidisciplinary cross-integration of multiple imaging technologies, has gradually entered the perspective of researchers. Taking TMD as the research object, quantitative analysis of lesion characteristics can obtain information that traditional imaging cannot provide. It relies on computers for image segmentation and feature extraction, and lesion characteristics are quantitative indicators, making the evaluation more objective. At the same time, it also realizes large-scale, efficient and automatic processing of images, making big data analysis and artificial intelligence diagnosis possible.

[0004] Artificial intelligence (AI) is the most widely used in assisting clinicians in imaging analysis and diagnosis. Deep learning convolution neural network (CNN) algorithms are one of the branches of AI machine learning. CNN uses a multi-layer processing network to extract and transform features of the images to be analyzed, uses the output information of the previous layer as the input information of the next layer, analyzes layer by layer, and learns with or without supervision, and finally outputs accurate recognition results. Recent studies have pointed out that artificial intelligence deep learning CNN can directly extract deep feature information of images and automatically learn and train models. There are also studies reporting that with the help of imaging omics based on artificial intelligence deep learning, MRI imaging data can be mined in multiple dimensions and high throughput to build a prediction model, which has good application prospects in the diagnosis and clinical decision-making of TMD. However, the features extracted by deep learning-based classification are high-dimensional abstract features generated autonomously by the model in an end-to-end manner. Its decision-making process has a "black box" attribute and lacks clinical interpretability. In contrast, the TMD classification based on radiomics artificially designs visual feature parameters such as texture, morphology, and grayscale (such as gray-level co-occurrence matrix and wavelet energy map). The image markers it screens have clear biomedical significance and can provide clinicians with a traceable basis for decision-making.

[0005] Therefore, it is of great significance to study the artificial intelligence-assisted diagnosis of temporomandibular joint disorders based on MRI imaging in order to provide a solid theoretical basis for the clinical examination of temporomandibular joint disorders. Summary of the invention

[0006] The embodiment of the present invention discloses a prediction method and system for temporomandibular joint disorder based on artificial intelligence, which can provide an evaluation standard for temporomandibular joint disorder through MRI images and provide judgment accuracy and reliability.

[0007] A first aspect of an embodiment of the present invention discloses a method for predicting temporomandibular joint disorder based on artificial intelligence, comprising: Obtaining MRI images of the patient; Data resampling was performed based on the average spatial resolution of the patient's images; Cropping the patient data three-dimensional image layer by layer to obtain cropped patient three-dimensional image data; The cropped three-dimensional image data of the patient is introduced into an artificial intelligence algorithm; and group verification is performed on the cropped three-dimensional image data of the patient, and then a temporomandibular joint disorder classification prediction model is established for prediction.

[0008] Furthermore, the step of resampling data according to the average spatial resolution of the patient's image includes: First, calculate the average spatial resolution of the patient image; The patient image is then resampled according to the average spatial resolution.

[0009] Furthermore, the step of cropping the patient data three-dimensional image layer by layer to obtain cropped patient three-dimensional image data includes: A three-dimensional stereoscopic frame containing the entire lesion is constructed for the entire temporomandibular joint; Performing layer-by-layer image cropping on the patient image data; Collect the cropped image data for each layer.

[0010] Furthermore, the steps of introducing the cropped three-dimensional image data of the patient into an artificial intelligence algorithm and grouping the cropped three-dimensional image data of the patient to establish a classification model for verification, and forming a standard method for evaluating temporomandibular joint disorders include: Call medical drawing software to outline the ROI area and add three-dimensional labels; Perform quantitative feature extraction of ROI area; A feature screening method from rough selection to fine selection is used to perform ROI feature screening, thereby obtaining finely selected ROI feature data; Importing the selected ROI feature data into logistic regression, support vector machine, random forest, decision tree, k-nearest neighbor, and then completing the construction of a temporomandibular joint disorder classification prediction model; And by analyzing and comparing the normal MRI images without disease and the diseased MRI image data, the evaluation index of the diseased MRI image is given.

[0011] Furthermore, the quantitative feature extraction is performed on the ROI region, wherein the extracted features include: intensity features, shape features, texture features and transformation features.

[0012] Furthermore, the step of performing ROI feature screening using the feature screening method from rough selection to fine selection includes: First, the variance analysis F test feature selection method is used to perform rough feature screening; Secondly, further fine-tuning is performed by introducing the least absolute shrinkage and selection operators.

[0013] Furthermore, the step of importing the selected ROI feature data into logistic regression, support vector machine, and random forest to complete the construction of a temporomandibular joint disorder classification prediction model includes: The study cohort was randomly divided into training and validation sets with a ratio of 8:2; Optimizing model parameters using a five-fold cross validation method in the training set; The mean of the predicted probabilities of all models generated during the five-fold cross-validation method in the training set is used to evaluate the model performance; Among them, grid search is used to determine the optimal values ​​of model hyperparameters.

[0014] Furthermore, the step of providing an evaluation index of the diseased MRI image by analyzing and comparing the normal MRI image without the disease and the diseased MRI image data includes: Draw a receiver operating characteristic curve graph of the model on the test set and the validation set, wherein the curve graph is drawn with the false positive rate (FPR) as the horizontal axis and the true positive rate (TPR) as the vertical axis; Calculate AUC and use AUC as the main evaluation indicator; The indicators are calculated to comprehensively evaluate the model performance, including: F1 score, which is the harmonic mean of precision and recall; accuracy, which is the ratio of samples correctly predicted by the model to the total number of samples; precision, which is the ratio of samples predicted by the model to be positive that are actually positive; sensitivity, which is the ratio of samples correctly predicted by the model to be positive, and specificity, which is the ratio of samples correctly predicted by the model to be negative.

[0015] A second aspect of an embodiment of the present invention discloses a prediction system, the system comprising: An image acquisition unit, used for acquiring an MRI image of a patient to be tested; The data collection unit re-collects data according to the average spatial resolution of the patient's images; An image cropping unit, which crops the patient data three-dimensional image layer by layer to obtain cropped patient three-dimensional image data; The model building unit introduces the cropped three-dimensional image data of the patient into an artificial intelligence algorithm; and groups the cropped three-dimensional image data of the patient to establish a classification model for verification, so as to form a method for evaluating the temporomandibular joint disorder.

[0016] A third aspect of an embodiment of the present invention discloses a computer including a prediction system, wherein the computer including the prediction system includes: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute any one of the methods of evaluating temporomandibular joint disorders based on artificial intelligence.

[0017] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: In an embodiment of the present invention, a prediction method for temporomandibular joint disorder based on artificial intelligence is provided, including: obtaining an MRI image of a patient; re-collecting data according to the average spatial resolution of the patient image; cropping the three-dimensional image of the patient data layer by layer to obtain cropped three-dimensional image data of the patient; introducing the cropped three-dimensional image data of the patient into an artificial intelligence algorithm; and grouping and verifying the cropped three-dimensional image data of the patient, thereby establishing a classification prediction model for temporomandibular joint disorder for prediction. The present application proposes that MRI imaging examination has significant advantages in diagnosing temporomandibular joint disorder, collects data to construct an MRI radiomics diagnosis model, combines the clinical factors of patients with temporomandibular joint disorder with the radiomics score calculation formula (Radiomics score, Rad-score), and establishes a radiomics-clinical joint prediction model for temporomandibular joint disorder cases, providing a powerful tool for accurate diagnosis of temporomandibular joint disorder. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is a flow chart of a method for predicting temporomandibular joint disorder based on artificial intelligence disclosed in an embodiment of the present invention; Figure 2 It is a flow chart of another method for predicting temporomandibular joint disorder based on artificial intelligence disclosed in an embodiment of the present invention; Figure 3 It is a flow chart of another method for predicting temporomandibular joint disorder based on artificial intelligence disclosed in an embodiment of the present invention; Figure 4 It is a flow chart of another method for predicting temporomandibular joint disorder based on artificial intelligence disclosed in an embodiment of the present invention; Figure 5 It is a schematic diagram of reversible anterior disc displacement and irreversible anterior disc displacement in an abnormal MRI image and a normal MRI image disclosed in an embodiment of the present invention; Figure 6 It is a flow chart of another method for predicting temporomandibular joint disorder based on artificial intelligence disclosed in an embodiment of the present invention; Figure 7 It is a flow chart of another method for predicting temporomandibular joint disorder based on artificial intelligence disclosed in an embodiment of the present invention; Figure 8It is a flow chart of another method for predicting temporomandibular joint disorder based on artificial intelligence disclosed in an embodiment of the present invention; Fig. 9 A schematic diagram of the structure of a prediction system disclosed in an embodiment of the present invention; Fig.10a , Fig.10b The trajectory of the volume of interest (VOI) in the radiographic (3D Slice) platform is shown for MRI images; Fig.11a Schematic diagram of the diagnostic performance of the five models for (A) classifying normal and abnormal cases based on the training set using the classification prediction model; Fig.11b (B) Diagram of diagnostic performance of five models for classifying normal and abnormal cases based on the test set using the classification prediction model; Fig.12a Schematic diagram of the diagnostic performance of five models for classifying anterior disc displacement with reduction (ADDwR) and without reduction (ADDwoR) using the classification prediction model for (A) the training set; Figure 12b Schematic diagram of the diagnostic performance of five models for classifying anterior disc displacement with reduction (ADDwR) and without reduction (ADDwoR) using the classification prediction model for the (B) test set. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] It should be noted that the terms "first", "second", "third", "fourth", etc. in the specification and claims of the present invention are used to distinguish different objects rather than to describe a specific order. The terms "including" and "having" in the embodiments of the present invention and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0022] The embodiment of the present invention discloses a prediction method and a prediction system for temporomandibular joint disorder based on artificial intelligence, which can improve the accuracy and reliability of temporomandibular joint disorder prediction.

[0023] The following is a detailed description with reference to the accompanying drawings.

[0024] Embodiment 1 See also Figure 1 , Figure 1 FIG. 1 is a flow chart of a method for predicting temporomandibular joint disorder based on artificial intelligence disclosed in an embodiment of the present invention. Figure 1 As shown, the prediction method of temporomandibular joint disorder based on artificial intelligence may include the following steps.

[0025] A first aspect of an embodiment of the present invention discloses a method for predicting temporomandibular joint disorder based on artificial intelligence, comprising: 101. Obtain an MRI image of the patient; 102. Resample the data according to the average spatial resolution of the patient's images; 103. Crop the patient data three-dimensional image layer by layer to obtain cropped patient three-dimensional image data; 104. Introduce the cropped three-dimensional image data of the patient into an artificial intelligence algorithm; and perform group verification on the cropped three-dimensional image data of the patient, and then establish a temporomandibular joint disorder classification prediction model for prediction.

[0026] It should be noted that, with respect to the MRI images of patients, the diseased images are MRI images of subjects suffering from temporomandibular joint disorder; and the normal images are MRI images of normal subjects without disease.

[0027] The purpose of data resampling is to improve the generalization of the model based on the average spatial resolution of the patient's images, so the image data needs to be resampled. Because the image formation is based on different devices or sources, the resolution of some image data may be inconsistent. Therefore, the average spatial resolution of the included patient images is calculated first, and in order to avoid resampling too much data, all image data are uniformly resampled to the average spatial resolution.

[0028] As an optional implementation, before segmenting the lesion, the MR image is corrected by N4 bias field; the image is resampled to a voxel size of 1×1×1 mm 3 , to standardize the voxel spacing; the voxel intensity values ​​are discretized by using a fixed bin width of 5 SI to reduce image noise and standardize the intensity, thereby achieving stable intensity resolution in all images; the images are normalized to normalize the signal intensity to 1~100 SI to reduce the difference in signal intensity of images collected by different machines; similarly, MRI images can also perform Z-score normalization on image grayscale values ​​to reduce the impact of inconsistent imaging parameters on the variation of imaging genomics features.

[0029] At the same time, it should be noted that the same resampling operation was used for the lesion judgment criteria of the patients.

[0030] It should also be noted that in order to reduce the interference of the background information of the MRI image on the network model, all patient images were cropped. The gold standard was outlined on the cross-section of the patient's MRI image by calling ITK software. This outline range includes the entire temporomandibular joint (condyle) to form a three-dimensional frame that includes the entire lesion. Then, each layer of the patient data image was cropped according to the three-dimensional frame to obtain the cropped image data.

[0031] As an optional implementation, the patient's general information is collected based on a self-compiled information collection form, a hospital HIS system or a medical record system, and the imaging data used to train and verify the imaging genomics model is extracted from the medical record system; the patient's MRI image is obtained by calling the patient's case information.

[0032] It should be noted that the scanning equipment involved in the present invention is: using a GE 3.0 T Discovery superconducting magnetic resonance imaging device, fixing the subject's head with a sponge pad, and administering 10% chloral hydrate orally to all ASD subjects in the case group, and then using a scanner to collect images after they fall asleep.

[0033] Among them, the MR image scanning procedure and parameters: a GE 3.0 T Discovery 750 superconducting MR scanner and a SENSE 8-channel head coil were used.

[0034] Scanning method: bilateral temporomandibular joint oblique coronal scan (T2WI), and bilateral temporomandibular joint oblique sagittal open and closed mouth scan (proton density-weighted spin echo image). Oblique coronal position: the positioning line is parallel to the transverse axis of the condyle; oblique sagittal closed mouth position: the positioning line is perpendicular to the transverse axis of the condyle. Open mouth position: a plastic mouth opener is placed between the patient's upper and lower anterior teeth to achieve the maximum comfortable open mouth position.

[0035] Among them, T2WI (TR / TE=2740.0 ms / 68.0ms, NEX=4), FOV: 16cm×16cm; bandwidth: 35.71; slice thickness / slice spacing: 2.0mm / 0.5mm; matrix: 288×256; scanning time: 2 min39s). PDWI (TR / TE=1951.0 ms / 35.0ms, NEX=4), FOV: 16cm×16cm; bandwidth: 35.71; slice thickness / slice spacing: 2.0mm / 0.5mm; matrix: 288×224; scanning time: 2 min24s.

[0036] Embodiment 2 See also Figure 2 , Figure 2FIG. 1 is a flow chart of another method for predicting temporomandibular joint disorder based on artificial intelligence disclosed in an embodiment of the present invention. Figure 2 , based on the average spatial resolution of the patient's images, the steps for data reacquisition include: 201. First, calculate the average spatial resolution of the patient image; 202. Resample the patient image according to the average spatial resolution.

[0037] Embodiment 3 See also Figure 3 , Figure 3 FIG. 1 is a flow chart of another method for predicting temporomandibular joint disorder based on artificial intelligence disclosed in an embodiment of the present invention. Figure 3 The step of cropping the patient data three-dimensional image layer by layer to obtain cropped patient three-dimensional image data includes: 301. Construct a three-dimensional frame containing the entire lesion for the entire temporomandibular joint; 302. Performing layer-by-layer image cropping on the patient image data; 303. Collect the cropped image data of each layer.

[0038] Embodiment 4 See also Figure 4 , Figure 5 , Fig.10a , Fig.10b , Figure 4 It is a flow chart of another method for predicting temporomandibular joint disorder based on artificial intelligence disclosed in an embodiment of the present invention. Figure 5 A schematic diagram of reversible anterior disc displacement and irreversible anterior disc displacement in an abnormal MRI image and a normal MRI image disclosed in an embodiment of the present invention; Fig.10a , Fig.10b The trajectory of the volume of interest (VOI) in the radiographic (3DSlice) platform is shown for MRI images; Figure 4 , introducing the cropped three-dimensional image data of the patient into an artificial intelligence algorithm; and grouping the cropped three-dimensional image data of the patient to establish a classification model for verification, the steps for forming a standard method for evaluating temporomandibular joint disorders include: 401. Calling medical drawing software, outlining the ROI area, and adding a three-dimensional label; 402. Perform quantitative feature extraction of the ROI region; 403. Using a feature screening method from rough selection to fine selection, ROI feature screening is performed to obtain finely selected ROI feature data; 404. Import the selected ROI feature data into logistic regression, support vector machine, random forest, decision tree, k-nearest neighbor, and then complete the construction of the temporomandibular joint disorder classification prediction model; 405. By analyzing and comparing the normal MRI image data without disease and the MRI image data of disease, an evaluation index of the diseased MRI image is given.

[0039] Here, it should be noted that the entire temporomandibular joint (condyle) is outlined on the patient's MRI image using INK-SNAP software to form a three-dimensional frame (label = 1) that includes the entire lesion.

[0040] As an optional implementation, Figure 5 As shown, ROI annotation: Both positive and abnormal MRI images contain open and closed positions, and abnormal images can be divided into reversible anterior disc displacement (ADDwR) and irreversible anterior disc displacement (ADDwoR). The diseased image is annotated with case ROI to obtain a diseased image; the diseased label annotates the abnormal condylar position. The normal image is annotated with a position label to obtain a normal annotated image.

[0041] In magnetic resonance imaging (MRI) of normal and abnormal anterior disc positions with and without reduction, the posterior band of the disc is located on top of the condyles. If the posterior band of the disc is located at the 12 o'clock position between the condyles, the disc position is considered normal. A disc located below 11 o'clock is considered anteriorly displaced. In the open oblique sagittal view, if the relationship of the disc to the condyles is restored to normal, the anterior disc displacement is reversible. If the condylar position of the disc cannot be restored to the open oblique sagittal view, it is an irreversible anterior disc displacement.

[0042] It should be noted that (1) The normal shape of the articular disc is bow-tie shaped. Round, biconvex, biplanar, folded and posterior thickening are all deformations of the articular disc.

[0043] (2) Angle of articular disc displacement: On closed-mouth oblique sagittal T1WI, the posterior disc band is located above the condylar apex, which is a normal disc-condylar relationship. If the posterior disc band exceeds the condylar process, it is anterior disc displacement. On open-mouth oblique sagittal T1WI, if the disc-condylar position relationship returns to normal, it is reducible anterior disc displacement; if it cannot return to normal and the articular disc is still in front of the condylar process, it is irreducible anterior disc displacement. Draw a line between the lowest point of the temporomandibular joint tubercle and the condylar tubercle, and mark the midpoint of the condylar part of this line. Measure the angle between the vertical line of the midpoint of the condyle and the posterior edge of the posterior disc band. 31° to 50° is mild displacement, 51° to 80° is moderate displacement, and greater than 80° is severe displacement.

[0044] (3) Condylar morphology: On the closed-mouth oblique sagittal PDWI images, the condylar head is divided into oval, flat, and bird-beak shapes according to its morphology.

[0045] (4) Medullary cavity edema: high signal appears in the condylar medullary cavity on T2WI oblique sagittal view.

[0046] (5) Joint effusion: High signal appears in the joint cavity on two consecutive levels of T2WI oblique sagittal.

[0047] Embodiment 5 like Figure 4 As shown, for step 402, quantitative feature extraction is performed on the ROI region, wherein the extracted features include: intensity features, shape features, texture features, and transformation features.

[0048] It should be noted that after the ROI is delineated, quantitative feature extraction is performed based on the ROI area. Radiomics features include: intensity features, shape features, texture features, and transformation features. According to the ROI delineated by senior doctors, the present invention uses the Python-based radiomics toolkit pyradiomics to automatically extract radiomics features, and finally obtains 1130 radiomics features of the ROI area, including 14 shape features, 216 first-order features, and 900 high-order features, and these features are max-min normalized.

[0049] Embodiment 6 See also Figure 6 , Figure 6 FIG. 1 is a flow chart of another method for predicting temporomandibular joint disorder based on artificial intelligence disclosed in an embodiment of the present invention. Figure 6 The step of performing ROI feature screening using the feature screening method from rough selection to fine selection includes: 601. Use the variance analysis F test feature selection method to perform rough feature screening; 602. Further fine screening is performed by introducing the minimum absolute shrinkage and selection operators.

[0050] It should be noted that due to the large dimension of the extracted features, which contain a large number of redundant features, overfitting and "one-size-fits-all disaster" may occur. In order to ensure that the features most relevant to the prediction of temporomandibular joint disorders are retained, the present invention adopts a feature screening method from rough selection to fine selection.

[0051] The first step is to use the variance analysis F test feature selection method to perform rough feature screening. The importance of the feature is evaluated by calculating the F statistic between each radiomics feature and the prediction of temporomandibular joint disorder and non-category. The F statistic is calculated based on the ratio of the between-group variance to the within-group variance. A higher F value indicates that the feature has a higher statistical significance for distinguishing different categories. After preliminary screening, the 50 most statistically significant features were finally obtained. The second step is to introduce the Least Absolute Shrinkage and Selection Operator (LASSO) to perform further fine screening on the basis of rough screening. 𝜆 is a regularization parameter that controls the strength of the L1 penalty. By adjusting the value of 𝜆, the complexity of the model can be controlled. The larger the value of 𝜆, the greater the penalty, and the more the model tends to be simple (that is, more coefficients are compressed to 0), thereby realizing the automatic selection of features. The present invention uses a 10-fold cross-validation method during the training process to select the optimal parameters to obtain the optimal number of features and avoid overfitting. Finally, based on the coarsely screened features, further fine screening is performed, and the present invention retains the radiomics features most relevant to temporomandibular joint disorder; if the most relevant features are not selected in this fine screening, the step of fine screening features is skipped and the next step of model construction is entered.

[0052] Embodiment 7 See also Figure 7 , Figure 7 FIG. 1 is a flow chart of another method for predicting temporomandibular joint disorder based on artificial intelligence disclosed in an embodiment of the present invention. Figure 7 The steps of importing the selected ROI feature data into logistic regression, support vector machine, random forest, decision tree, k-nearest neighbor, and then completing the construction of the temporomandibular joint disorder classification prediction model include: 701. The study cohort was randomly divided into a training set and a validation set with a ratio of 8:2; 702. Optimize model parameters using a five-fold cross validation method in the training set; 703. Using the mean of the predicted probabilities of all models generated during the five-fold cross validation method in the training set to evaluate model performance; 704. Use grid search to determine the optimal values ​​of model hyperparameters.

[0053] It should be noted that the common machine learning classifiers Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), and k-Nearest Neighbors (kNN) were used to develop a model for predicting the classification of temporomandibular joint disorders. The study cohort was randomly divided into a training set and a validation set with a ratio of 8:2.

[0054] Among them, the five-fold cross-validation method is used to optimize the model parameters in the training cohort. In this process, four folds are used as training data sets and one fold is used as a validation data set. This process is repeated five times, and the average prediction probabilities from the five validation data sets are used to evaluate the model performance. For independent testing, the mean of the predicted probabilities of all models generated during cross-validation is used to evaluate the model performance. The hyperparameters of the model are determined by grid search to determine the optimal value.

[0055] As an optional embodiment, the classification model construction adopts three classifiers: logistic regression, support vector machine, random forest, decision tree, and k-nearest neighbor to develop a model for the classification prediction of temporomandibular joint disorder and normal, as well as the classification prediction of ADDwR and ADDwoR.

[0056] Embodiment 8 See also Figure 8 , Figure 8 FIG. 1 is a flow chart of another method for predicting temporomandibular joint disorder based on artificial intelligence disclosed in an embodiment of the present invention. Figure 8 The step of analyzing and comparing the normal MRI image without disease and the MRI image data with disease to provide the evaluation index of the diseased MRI image comprises: 801. Draw a receiver operating characteristic curve graph of the model on the test set and the validation set, wherein the curve graph is drawn with the false positive rate (FPR) as the horizontal axis and the true positive rate (TPR) as the vertical axis; 802. Calculate AUC and use AUC as the main evaluation indicator; 803. Calculate indicators to comprehensively evaluate model performance, including: F1 score, which is the harmonic mean of precision and recall; accuracy, which is the ratio of samples correctly predicted by the model to the total number of samples; precision, which is the ratio of samples predicted by the model to be positive that are actually positive; sensitivity, which is the ratio of samples correctly predicted by the model to be positive; and specificity, which is the ratio of samples correctly predicted by the model to be negative.

[0057] It should be noted that the prediction method in the present invention involves a binary classification problem. Therefore, the present invention draws a receiver operating characteristic (ROC) curve of the model on the validation set and the test set. The curve is drawn with the false positive rate (FPR) as the horizontal axis and the true positive rate (TPR) as the vertical axis to show the performance of the model at different thresholds. AUC is used as the main evaluation indicator, and the model performance is comprehensively evaluated by calculating the indicator.

[0058] As an optional embodiment, Figure 5 As shown, Figure 5 Comparison of reversible anterior disc displacement (ADDwR) and irreversible anterior disc displacement (ADDwoR) in abnormal MRI images with normal MRI images.

[0059] Embodiment 9 Table 1 shows the diagnostic performance of the normal and abnormal TMJ classification models in terms of AUC (area under the curve), recall, F1 score, sensitivity, specificity, and precision in the training and validation sets. Our results show that the RF classifier outperforms other methods in predicting normal and abnormal. In the training set, the normal and abnormal classification models achieved AUC values ​​of 0.854 (DT), 0.847 (KNN), 0.890 (SVM), 0.899 (LR), and 0.889 (RF), respectively; in the validation test, the normal and abnormal classification models achieved AUC values ​​of 0.801 (DT), 0.852 (KNN), 0.863 (SVM), 0.841 (LR), and 0.874 (RF), respectively.

[0060] Table 1

[0061]

[0062] Table 1 is an evaluation table of the diagnostic performance of the training and validation cohorts using the five indicators of normal and abnormal classification in the normal-abnormal classification model.

[0063] Among them, SVM: support vector machine; LR: logistic regression; RF: random forest; DT: decision tree; KNN: K-nearest neighbor; CI: confidence interval; AUC, area under the receiver operating characteristic curve P value is the significance level for comparing AUC with random cases (AUC=0.5).

[0064] Table 2 shows the diagnostic performance of the classification models for reversible and irreversible anterior displacement in the training and validation sets in terms of AUC (area under the curve), recall, F1 score, sensitivity, specificity, and precision. Our results show that the KNN classifier outperforms other methods in predicting reversible and irreversible anterior displacement. In the training set, the normal and abnormal classification models achieved AUC values ​​of 0.732 (DT), 0.733 (KNN), 0.747 (LR), 0.733 (RF), and 0.731 (SVM), respectively; in the validation test, the normal and abnormal classification models achieved AUC values ​​of 0.699 (DT), 0.712 (KNN), 0.705 (LR), 0.693 (RF), and 0.701 (SVM), respectively.

[0065] Table 2

[0066]

[0067] Table 2 shows the evaluation of the diagnostic performance of the training and validation cohorts using the five indicators of ADDwR and ADDwoR classification in the ADDwR and ADDwoR classification models.

[0068] Among them, SVM: support vector machine; LR: logistic regression; RF: random forest; DT: decision tree; KNN: K-nearest neighbor; CI: confidence interval; AUC, area under the receiver operating characteristic curve P value is the significance level for comparing AUC with random cases (AUC=0.5).

[0069] As shown in Table 1 , the diagnostic performance of the normal and abnormal classification of the temporomandibular joint is shown in the following indicators: AUC (area under the curve), recall rate, F1 score, sensitivity, specificity, and precision in the training set and validation set.

[0070] As shown in Table 2 , the diagnostic performance of indicators such as AUC (area under the curve), recall, F1 score, sensitivity, specificity, and precision in the training and validation sets are shown for the classification of ADDwR and ADDwoR for the TMJ.

[0071] At the same time, reference Fig.11a , Fig.11b , Fig.12a , Figure 12b ;like Fig.11a , Fig.11b As shown in Table 1, the diagnostic performance of five models for classifying normal and abnormal cases using classification prediction models in the training set (A) and validation set (B) cohorts. Fig.12a , Figure 12bAs described, corresponding to Fig. 2 Diagnostic performance of five models for classifying anterior disc displacement with reduction (ADDwR) and without reduction (ADDwoR) using classification prediction models in the training set (A) and validation set (B) cohorts.

[0072] Embodiment 10 See also Fig. 9 , Fig. 9 Schematic diagram of a prediction system disclosed in the second aspect of the present invention. Fig. 9 , the prediction system 900 may include a graphics acquisition unit 901, a data collection unit 902, an image cropping unit 903 and a model building unit 904, wherein: The image acquisition unit 901 is used to acquire the MRI image of the patient to be tested; The data collection unit 902 performs data recollection according to the average spatial resolution of the patient's image; An image cropping unit 903 crops the patient data three-dimensional image layer by layer to obtain cropped patient three-dimensional image data; The model building unit 904 introduces the cropped three-dimensional image data of the patient into an artificial intelligence algorithm; and groups the cropped three-dimensional image data of the patient to establish a classification model for verification, so as to form a method for evaluating the temporomandibular joint disorder.

[0073] A third aspect of an embodiment of the present invention discloses a computer including a prediction system, wherein the computer including the prediction system includes: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute any one of the methods of evaluating temporomandibular joint disorders based on artificial intelligence.

[0074] A fourth aspect of an embodiment of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute an artificial intelligence-based method for determining temporomandibular joint disorder disclosed in the first aspect of an embodiment of the present invention.

[0075] A fifth aspect of an embodiment of the present invention discloses a computer program product, which, when executed on a computer, enables the computer to execute part or all of the steps of any one of the artificial intelligence-based methods for determining temporomandibular joint disorders of the first aspect.

[0076] A sixth aspect of an embodiment of the present invention discloses an application publishing platform, which is used to publish a computer program product. When the computer program product runs on a computer, the computer executes part or all of the steps of any one of the artificial intelligence-based temporomandibular joint disorder judgment methods of the first aspect.

[0077] The embodiment of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute Figure 1~Figure 2 Any method for determining temporomandibular joint disorder based on artificial intelligence.

[0078] The embodiments of the present invention further disclose a computer program product, wherein when the computer program product is run on a computer, the computer is enabled to execute part or all of the steps of the methods in the above method embodiments.

[0079] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0080] The above is a detailed introduction to a method and prediction system for judging temporomandibular joint disorders based on artificial intelligence disclosed in an embodiment of the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A prediction method for temporomandibular joint disorder based on artificial intelligence, characterized in that: include: Obtaining MRI images of the patient; Data resampling was performed based on the average spatial resolution of the patient's images; Cropping the patient data three-dimensional image layer by layer to obtain cropped patient three-dimensional image data; introducing the cropped three-dimensional image data of the patient into an artificial intelligence algorithm; The cropped three-dimensional image data of the patient are grouped and verified, and then a temporomandibular joint disorder classification prediction model is established for prediction.

2. The method according to claim 1, characterized in that The step of re-collecting data based on the average spatial resolution of the patient's image comprises: First, the average spatial resolution of the patient image is calculated; Then, the patient images are resampled to a uniform spatial resolution according to the average spatial resolution.

3. The method according to claim 1, characterized in that The step of cropping the patient data three-dimensional image layer by layer to obtain cropped patient three-dimensional image data comprises: A three-dimensional stereoscopic frame containing the entire lesion is constructed for the entire temporomandibular joint; Performing layer-by-layer image cropping on the patient image data; Collect the cropped image data for each layer.

4. The method according to claim 1, characterized in that The steps of introducing the cropped three-dimensional image data of the patient into an artificial intelligence algorithm; and grouping the cropped three-dimensional image data of the patient to establish a classification model for verification, and forming a standard method for evaluating temporomandibular joint disorders include: Call medical drawing software to outline the ROI area and add three-dimensional labels; Perform quantitative feature extraction of ROI area; A feature screening method from rough selection to fine selection is used to perform ROI feature screening, thereby obtaining finely selected ROI feature data; Importing the selected ROI feature data into logistic regression, support vector machine, random forest, decision tree, k-nearest neighbor, and then completing the construction of a temporomandibular joint disorder classification prediction model; And by analyzing and comparing the normal MRI images without disease and the diseased MRI image data, the evaluation index of the diseased MRI image is given.

5. The method according to claim 4, characterized in that The quantitative feature extraction is performed on the ROI area, wherein the extracted features include: intensity features, shape features, texture features and transformation features.

6. The method according to claim 4, characterized in that The steps of performing ROI feature screening using the feature screening method from rough selection to fine selection include: First, the variance analysis F test feature selection method is used to perform rough feature screening; Secondly, further fine-tuning is performed by introducing the least absolute shrinkage and selection operators.

7. The method according to claim 4, characterized in that The step of importing the selected ROI feature data into logistic regression, support vector machine, random forest, decision tree, k-nearest neighbor, and then completing the construction of the temporomandibular joint disorder classification prediction model includes: The study cohort was randomly divided into training and validation sets with a ratio of 8:2; Optimizing model parameters using a five-fold cross validation method in the training set; The mean of the predicted probabilities of all models generated during the five-fold cross-validation method in the training set is used to evaluate the model performance; Among them, grid search is used to determine the optimal values ​​of model hyperparameters.

8. The method according to claim 4, characterized in that The step of providing an evaluation index of the diseased MRI image by analyzing and comparing the normal MRI image without the disease and the diseased MRI image data comprises: Draw a receiver operating characteristic curve graph of the model on the test set and the validation set, wherein the curve graph is drawn with the false positive rate (FPR) as the horizontal axis and the true positive rate (TPR) as the vertical axis; Calculate AUC and use AUC as the main evaluation indicator; The indicators are calculated to comprehensively evaluate the model performance, including: F1 score, which is the harmonic mean of precision and recall; accuracy, which is the ratio of samples correctly predicted by the model to the total number of samples; precision, which is the ratio of samples predicted by the model to be positive that are actually positive; sensitivity, which is the ratio of samples correctly predicted by the model to be positive, and specificity, which is the ratio of samples correctly predicted by the model to be negative.

9. A prediction system, characterized in that: The system comprises: An image acquisition unit, used for acquiring an MRI image of a patient to be tested; The data collection unit performs data resampling according to the average spatial resolution of the patient's images; An image cropping unit, which crops the patient data three-dimensional image layer by layer to obtain cropped patient three-dimensional image data; The model building unit introduces the cropped three-dimensional image data of the patient into an artificial intelligence algorithm; and groups the cropped three-dimensional image data of the patient to establish a classification model for verification, so as to form a method for evaluating the temporomandibular joint disorder.

10. A computer including a prediction system, characterized in that: The computer containing the prediction system comprises: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the artificial intelligence-based prediction method for temporomandibular joint disorder as described in any one of claims 1-8.

Images