Temporomandibular joint disc double-circle measurement model recognition system

Abstract

The invention, which relates to the technical field of the medical identification system, discloses a temporomandibular joint disc double-circle measurement model identification system comprising an image preprocessing and labeling module, a key point detection network construction module based on a deep learning network, a double-circle model construction module and a shift parameter calculation module. According to the method, a double-circle geometric model is provided, the center of a circle is positioned through a three-tangency-point normal intersection method, key points are automatically detected in combination with an optimized ResNet-34 deep learning network, the displacement direction is quantified based on an angular bisector included angle and an intersection point distance, the severity degree is evaluated through geometric parameters, single linear measurement is broken through, multi-dimensional morphological analysis is achieved, and compared with a traditional method, the method has the advantages that the method is simple in operation and convenient to implement. The measurement accuracy is improved, the measurement time is shortened, and a high-precision and automatic diagnosis standard is provided for temporal-mandibular joint disc displacement.

Description

Technical Field

[0001] The invention relates to the technical field of medical identification systems, in particular to a temporomandibular joint disc double-circle measurement model identification system. Background Art

[0002] The temporomandibular joint is a joint that connects the mandible and the skull. It plays a vital role in many human functions, including chewing, speaking, breathing, and swallowing. Temporomandibular joint dysfunction is a common clinical disease, manifested by joint pain, movement disorders, noise, and loss of joint function. Medial and lateral displacement of the temporomandibular joint is one of the common symptoms of temporomandibular joint dysfunction, which is usually caused by malposition of the articular disc or damage to the joint capsule and ligaments. This displacement not only affects the patient's chewing function, but may also cause long-term pain and discomfort, affecting the patient's quality of life. Therefore, accurate measurement and diagnosis of the medial and lateral displacement of the temporomandibular joint is crucial.

[0003] However, the current evaluation of the medial and lateral displacement of the temporomandibular joint often relies on a variety of imaging techniques, such as MRI and then measurement. The traditional measurement method measures the e value, that is, the medial displacement, which is the horizontal distance between the medial edge of the articular disc and the outermost inner edge of the condyle, the f value, that is, the lateral displacement, which is the horizontal distance between the outermost outer edge of the articular disc and the outermost outer edge of the condyle, and the ef value, that is, the medial and lateral difference. The three key distance values ​​of e value, f value and ef value are used to analyze the medial and lateral displacement characteristics of the articular disc. However, this manual measurement method is highly dependent on the manual operation of the imaging software. The doctor marks the edges of the articular disc and the condyle according to the image. Measuring the e value and f value is not only time-consuming, but also depends on the operator's skills and experience in terms of accuracy. Especially when the image edge is not clear, there may be large differences in the measurement results between different doctors, which leads to increased subjectivity in the diagnosis and increases the error of repeated measurements. In addition, the traditional measurement method only evaluates the displacement characteristics of the articular disc based on linear distance, that is, e-value, f-value and ef-value, which is difficult to fully capture the subtle changes in the position offset of the articular disc. This method ignores the complex morphology of the articular disc in the condylar structure and cannot analyze the displacement of the articular disc from a more comprehensive geometric feature, which limits the measurement accuracy and the effectiveness of clinical diagnosis. Therefore, it is necessary to explore a new method that can simplify the measurement process, improve measurement accuracy, and more intuitively reflect the direction and severity of displacement. Summary of the invention

[0004] The purpose of the present invention is to provide a temporomandibular joint disc double circle measurement model recognition system, which solves the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a temporomandibular joint disc double circle measurement model recognition system, comprising: The image preprocessing and annotation module is used to obtain the MRI image of the patient's temporomandibular joint, and then perform standardization and edge enhancement on the MRI image, and perform regional cropping and scaling operations on the MRI image; A key point detection network module is built based on a deep learning network to input preprocessed MRI images, then optimize the network structure, annotate key points, and finally train the model; The key points include: the center of the large circle on the condylar head, the center of the small circle on the condylar neck, the left boundary point of the articular disc, the right boundary point of the articular disc, the left boundary point of the condyle, and the right boundary point of the condyle; A double circle model building module is used to build a large circle model based on the condylar head and a small circle model based on the condylar neck, and calculate the center and radius of the large circle model and the small circle model; The displacement parameter calculation module generates the condylar angle bisector and the articular disc angle bisector based on the key points, and locates the intersection based on the relationship between the condylar angle bisector and the articular disc angle bisector and the condyle and the articular disc, outputs the intersection A and the intersection B, and then measures the angle α of the angle bisector and the distance between the intersection A and the intersection B to intuitively display the displacement of the articular disc.

[0006] Optionally, the specific process of the image preprocessing and annotation module is as follows: Standardization processing: The brightness and contrast of MRI images are unified through the CLAHE algorithm, and Gaussian filtering is used to eliminate the noise of MRI images; Edge enhancement processing: The boundaries of the condylar head, condylar neck and articular disc are extracted and highlighted by using the Sobel operator; Regional cropping and scaling operations: The temporomandibular joint region in the MRI image is cropped and scaled to a uniform resolution to ensure the size consistency of the input deep learning network.

[0007] Optionally, the optimization of the network structure in the key point detection network module based on the deep learning network includes: 1×1 convolution replacement and semi-residual module; 1×1 convolution replacement: In the deep learning network, 1x1 convolution kernels are introduced in the deeper layers, i.e., near the output layer, and the 3x3 convolution kernels in the shallow layers are retained to reduce the computational complexity while fully extracting local detail features and ensuring effective modeling of key points; Semi-residual module: To enhance the feature expression capability of deep learning networks, the network structure is designed in combination with the semi-residual module. The semi-residual module introduces a learnable weighting coefficient into the traditional residual structure, and dynamically adjusts the fusion ratio of residual features and input features through the following formula: in: Refers to the weighted fusion result of multi-branch features; Refers to the features after being processed by the convolutional layer; Refers to the input features; Both refer to learning parameters, which are used to adaptively adjust the feature fusion method; The optimized deep learning network gives full play to the advantages of shallow and deep convolutional features, and enhances the feature fusion capability through the semi-residual module.

[0008] Optionally, the model training in the key point detection network module based on the deep learning network is specifically: Pair the annotated MRI images with the key point coordinates to form a training data set for model training. During the model training process, the images are rotated and scaled to increase data diversity and adapt the model to different MRI image angles and qualities. The mean square error loss function is used to measure the error between the key point coordinates predicted by the deep learning network and the actual coordinates. The calculation formula is as follows: in: Loss refers to the error between the predicted key point coordinates and the actual coordinates, that is, the loss value; N Refers to the total number of samples, the total number of coordinate points in all training data; ( x i ，y i ) refers to the true coordinate value in the i-th sample; Refers to the i-th sample x and y The predicted value of the coordinates; Evaluate the model performance on the test set, calculate the error between the predicted key points and the actual key points according to the formula, and adjust the model parameters or perform hyperparameter tuning based on the test results to improve the model's performance in key point prediction.

[0009] Optionally, the purpose of the double circle model construction module is to determine the center and radius of the large and small circles by the three-tangent point normal intersection method, and the steps are as follows: first, a large circle is drawn inside the condylar head, so that it is tangent to the upper surface and the inner and outer sides of the condylar head, and then a small circle is drawn inside the condylar neck, so that it is tangent to the inner and outer edges of the condylar neck and the lower boundary of the large circle, so as to form a standard geometric shape of the joint structure; The process of constructing the great circle model based on the condylar head is as follows: Draw normal lines at the three tangent points of the condylar structure, analyze the intersection relationship of the three tangent point normal lines, and calculate the coordinates of the center of the circle and the radius; If the three tangent point normals intersect each other; The intersection points of the three tangent normal lines are Construct the centroid of the triangle as the center of the circle Position; then the radius of the great circle model based on the condylar head is calculated as the center of the great circle when the three tangent point normals intersect each other. The distance to the normal line of the three tangent points is taken as the radius of the circle; If the three tangent point normals do not intersect, they are parallel; first, the tangent point normal L1: And the tangent point normal 2 L2: Then obtain the intersection of the third tangent point normal L3 with the tangent point normal 1 L1 and the tangent point normal 2 L2, and then obtain two valid intersection points. Then use the midpoint of these two intersection points as the center of the replacement large circle Then combine the remaining non-parallel tangent point normal line 3 L3 and make its tangent point and We can calculate the center of gravity by weighting, further correct the position of the center of the circle, improve the stability of the center of the circle positioning, and obtain the center of the large circle when the three tangent point normals do not intersect. ; Similarly, according to the three tangent points, the center of the great circle is , calculate the distance from the center of the circle to the three tangent point normals and use the median method to determine the radius of the circle. For the equation of the tangent point normal Li Center The distance to the tangent point normal Li is Take the median of the normal distances of the three tangent points as the radius.

[0010] Optionally, the specific process of constructing the small circle model based on the condylar neck in the double circle model construction module is as follows: Similarly, based on the large circle model of the condylar head, the small circle model based on the condylar neck is constructed by drawing normals at the three tangent points of the condylar neck structure, analyzing the relationship between the three normals, and calculating the coordinates and radius of the center of the circle when the tangent point normals intersect and when the tangent point normals are parallel.

[0011] Optionally, the generation of the condylar angle bisector and the articular disc angle bisector based on the key points in the displacement parameter calculation module is specifically: Connect the center of the small circle with the inner and outer convex points of the temporomandibular joint disc and the inner and outer convex points of the condyle and draw an angle bisector, the inner and outer convex points of the temporomandibular joint disc correspond to the left boundary point of the joint disc and the right boundary point of the joint disc, and the inner and outer convex points of the condyle correspond to the left boundary point of the condyle and the right boundary point of the condyle.

[0012] Optionally, the angle α of the angle bisector measured in the displacement parameter calculation module is used to reflect the severity of the displacement of the temporomandibular joint disc, and the distance between the intersection point A and the intersection point B is analyzed to reflect the displacement direction of the temporomandibular joint disc.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention transforms traditional measurement into quantitative analysis of geometric morphology by constructing a double circle geometric model of the condylar head large circle and the condylar neck small circle. The double circle model can stably and efficiently determine the center and radius of the circle through the three-tangent point normal intersection method, and uses the angle bisector and the intersection distance to intuitively reflect the displacement direction and severity of the articular disc. The present invention not only reduces the dependence on multiple variables in traditional measurement, but also reduces the subjective error of manual marking through a standardized model, thereby improving the repeatability and consistency of measurement.

[0014] 2. In order to solve the difficulty in locating the cutting points caused by blurred MRI image boundaries, the present invention introduces an optimized ResNet-34 deep learning network, combined with the optimization of the semi-residual module and the 1x1 convolution kernel, to achieve automatic detection of key points such as the condylar boundary, articular disc boundary and the center of the double circle. The model can efficiently extract local detail features, overcome the dependence of traditional manual labeling on operator experience, and solve the problem of inaccurate key point positioning caused by blurred images.

[0015] 3. The present invention generates an angle bisector and calculates the angle α and the intersection distance AB by connecting the center of the small circle with the inner and outer convex points of the articular disc and the inner and outer convex points of the condyle. The angle α and AB are used to quantify the severity of the displacement, and the complex anatomical structure relationship is converted into geometric parameters, making the diagnostic results more intuitive and convenient for clinicians to quickly judge the condition. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a corresponding structural diagram of the temporomandibular joint disc image of the present invention; Figure 2 A schematic diagram of determining the center and radius of a large circle when the tangent point normal lines intersect in the present invention; Figure 3 A schematic diagram for determining the center and radius of a large circle when the normal lines of the tangent points do not intersect; Figure 4 This is a schematic diagram of determining the center and radius of a small circle when the tangent point normal lines intersect in the present invention; Figure 5 A schematic diagram for determining the center and radius of a small circle when the normal lines of the tangent points do not intersect; Figure 6 A flowchart of the steps of the double-circle measurement method for temporomandibular joint disc displacement of the present invention; Figure 7 The deep learning network of the present invention cooperates with the semi-residual module to optimize the network graph; Figure 8 This is a schematic diagram of the overall structure of the temporomandibular joint disc double circle measurement model recognition system. DETAILED DESCRIPTION

[0017] 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.

[0018] For example, see Figures 1 to 8 This embodiment provides a technical solution: a temporomandibular joint disc double circle measurement model recognition system, including: The image preprocessing and annotation module is used to obtain the MRI image of the patient's temporomandibular joint, and then perform standardization and edge enhancement on the MRI image, and perform regional cropping and scaling operations on the MRI image; A key point detection network module is built based on a deep learning network to input preprocessed MRI images, then optimize the network structure, annotate key points, and finally train the model; The key points include: the center of the large circle on the condylar head, the center of the small circle on the condylar neck, the left boundary point of the articular disc, the right boundary point of the articular disc, the left boundary point of the condyle, and the right boundary point of the condyle; A double circle model building module is used to build a large circle model based on the condylar head and a small circle model based on the condylar neck, and calculate the center and radius of the large circle model and the small circle model; The displacement parameter calculation module generates the condylar angle bisector and the articular disc angle bisector based on the key points, and locates the intersection based on the relationship between the condylar angle bisector and the articular disc angle bisector and the condyle and the articular disc, outputs the intersection A and the intersection B, and then measures the angle α of the angle bisector and the distance between the intersection A and the intersection B to intuitively display the displacement of the articular disc.

[0019] In this embodiment: the present invention is different from the prior art. The traditional technology may only rely on linear distance measurement, while ignoring the geometric relationship between the articular disc and the condyle, and cannot capture subtle morphological changes. The present invention converts the three-dimensional structure of the condylar head and neck into a standardized geometric shape through a double circle model, and combined with angle bisector analysis, it can more comprehensively reflect the displacement characteristics of the articular disc, such as rotation and tilt, breaking through the limitations of traditional linear measurement. In the prior art, doctors may need to manually mark image boundary points, and the measurement efficiency is low and the error is large. The present invention realizes automatic detection of key points through an improved ResNet-34 deep learning network, and combines the semi-residual module and dynamic feature fusion technology to significantly improve the model's recognition ability for fuzzy boundaries, so that the measurement process is automated and standardized, and the error caused by manual intervention is reduced. In the prior art, the displacement may be evaluated by the ef difference, while the present invention reflects the displacement angle through the angle α and the intersection distance AB reflects the two parameters of the displacement distance. Combined with positive and negative signs, multi-dimensional quantification of the displacement direction, angle and distance is realized, providing richer decision-making basis for clinical diagnosis.

[0020] See also Figures 1 to 8 ,The specific process of the image preprocessing and annotation module is as follows: Standardization processing: The brightness and contrast of MRI images are unified through the CLAHE algorithm, and Gaussian filtering is used to eliminate the noise of MRI images; Edge enhancement processing: The boundaries of the condylar head, condylar neck and articular disc are extracted and highlighted by using the Sobel operator; Regional cropping and scaling operations: The temporomandibular joint region in the MRI image is cropped and scaled to a uniform resolution to ensure the size consistency of the input deep learning network.

[0021] In this embodiment: the image preprocessing and annotation module can pre-process the MRI image efficiently, so as to meet the use of the subsequent deep learning network and improve the accuracy and efficiency of the overall recognition system.

[0022] See also Figures 1 to 8 ,The optimization of the network structure in the key point detection network module based on deep learning network includes: 1×1 convolution substitution and semi-residual module; 1×1 convolution replacement: In the deep learning network, 1x1 convolution kernels are introduced in the deeper layers, i.e., near the output layer, and the 3x3 convolution kernels in the shallow layers are retained to reduce the computational complexity while fully extracting local detail features and ensuring effective modeling of key points; Semi-residual module: To enhance the feature expression capability of deep learning networks, the network structure is designed in combination with the semi-residual module. The semi-residual module introduces a learnable weighting coefficient into the traditional residual structure, and dynamically adjusts the fusion ratio of residual features and input features through the following formula: in: Refers to the weighted fusion result of multi-branch features; Refers to the features after being processed by the convolutional layer; Refers to the input features; Both refer to learning parameters, which are used to adaptively adjust the feature fusion method; the optimized deep learning network ResNet-34 gives full play to the advantages of shallow and deep convolutional features, and enhances the feature fusion capability through the semi-residual module, thereby significantly improving the training efficiency and the generalization performance of the model, providing strong support for efficient and accurate key point detection.

[0023] In this embodiment: the core of the present invention to construct a double circle model by using the three-tangent point normal intersection method is to find the three-tangent points to locate the large circle, and then the three-tangent points to locate the small circle. Considering that the boundary clarity of actual MRI images is limited, the boundary tangent points are not easy to find and thus the center positions of the large and small circles cannot be accurately determined. The present invention uses ResNET-34 to optimize the deep learning network to assist in measuring the key point positions; Based on the relationship between the center of the large circle, the center of the small circle, the boundary point of the zygomatic process and the boundary point of the articular disc, the present invention adopts the deep learning network ResNET-34 to optimize the deep learning network algorithm to build a model to identify the key points on the articular disc, thereby improving the positions of the centers of the large and small circles.

[0024] A key point detection network is built based on a deep learning network, namely ResNet-34. A semi-residual module is introduced into the residual module of ResNet. The fusion ratio of residual features and input features is dynamically adjusted using learnable weighting coefficients. The model outputs the coordinates of multiple key points. A regression layer is added to the last layer of ResNet to directly output the coordinate values ​​(x, y) of the key points. The goal is to predict the center coordinates of large and small circles and the coordinates of other boundary points. Aiming at the difficulty of locating the cutting point caused by blurred MRI image boundaries, the present invention introduces an optimized ResNet-34 deep learning network, combines the semi-residual module and 1x1 convolution kernel optimization, and realizes automatic detection of key points such as the condylar boundary, articular disc boundary and the center of the double circle. By dynamically adjusting the fusion ratio of residual features and input features, the model can adaptively extract local detail features, overcoming the dependence of traditional manual labeling on operator experience and solving the problem of inaccurate key point positioning caused by blurred images.

[0025] See also Figures 1 to 8 , the model training in the key point detection network module based on the deep learning network is as follows: Pair the annotated MRI images with the key point coordinates to form a training data set for model training. During the model training process, the images are rotated and scaled to increase data diversity and adapt the model to different MRI image angles and qualities. The mean square error loss function is used to measure the error between the key point coordinates predicted by the deep learning network and the actual coordinates. The calculation formula is as follows: in: Loss refers to the error between the predicted key point coordinates and the actual coordinates, that is, the loss value; N Refers to the total number of samples, the total number of coordinate points in all training data; ( x i ，y i ) refers to the true coordinate value in the i-th sample; Refers to the i-th sample x and y The predicted value of the coordinates; evaluate the model performance on the test set, calculate the error between the predicted key points and the actual key points according to the formula, and adjust the model parameters or perform hyperparameter tuning based on the test results to improve the model's performance in key point prediction.

[0026] In the preprocessed MRI images, the positions of the key points, including the center of the large circle, the center of the small circle, the left boundary point of the articular disc, the right boundary point of the articular disc, the left boundary point of the condyle, and the right boundary point of the condyle, are marked as training labels for the ResNet model.

[0027] The purpose of the double circle model construction module is to determine the center and radius of the large and small circles by the three-tangent point normal intersection method. The steps are as follows: first, draw a large circle inside the condylar head so that it is tangent to the upper surface and the inner and outer sides of the condylar head, and then draw a small circle inside the condylar neck so that it is tangent to the inner and outer edges of the condylar neck and the lower boundary of the large circle to form the standard geometric shape of the joint structure.

[0028] The process of constructing the great circle model based on the condylar head is as follows: Draw normal lines at the three tangent points of the condylar structure, analyze the intersection relationship of the three normal lines, and calculate the coordinates of the center and radius of the circle; Three tangent points: Upper tangent point, the highest point of the condylar head is tangent to the upper surface; Medial tangent point: the most convex point on the medial side of the condylar head is tangent to the medial edge; Lateral tangent point: the most convex point on the outer side of the condylar head is tangent to the lateral edge; Draw three tangent point normals based on the three tangent points; If the three tangent point normals intersect each other, such as Figure 3 As shown; The intersection points of the three tangent normal lines are Construct the centroid of the triangle as the center of the circle Position, the calculation formula based on the center of the great circle model of the condylar head is as follows: Then the radius of the great circle model based on the condylar head is calculated as the center of the great circle when the three tangent point normals intersect each other. The distance to the normal line of the three tangent points is taken as the circle radius. The specific calculation process and formula are as follows: The calculation equation for the tangent point normal Li is: Center of big circle The distance to the tangent point normal Li is ,So The calculation formula is as follows: in: Refers to the center of the circle The distance to the normal line Li of the tangent point; Both refer to the equation coefficients of the tangent point normal line Li. In the calculation formula of the tangent point normal line Li, Both refer to the directional coefficient of the tangent point normal, which is used to determine the slope of the tangent point normal, and Refers to the intercept of the tangent point normal line, which is used to determine the position of the tangent point normal line; Furthermore, the median of the normal distances of the three tangent points is taken as the radius, and the calculation formula is as follows: in: R refers to the radius of the great circle model when the three tangent point normals intersect each other; Median refers to substitution ( d1, d2, d3 ) in the middle; If the three tangent point normals do not intersect, they are parallel; First, the tangent point normal L1: And the tangent point normal 2 L2: Then obtain the intersection of the third tangent point normal L3 with the tangent point normal 1 L1 and the tangent point normal 2 L2, and then obtain two valid intersection points. The calculation formula is as follows: Then use the midpoint of these two intersection points as the center of the replacement large circle The calculation formula is as follows: Then combine the remaining non-parallel tangent point normal line 3 L3 and make its tangent point and We can calculate the center of gravity by weighting, further correct the position of the center of the circle, improve the stability of the center of the circle positioning, and obtain the center of the large circle when the three tangent point normals do not intersect. , the calculation formula is as follows: Similarly, according to the three tangent points, the center of the great circle is , calculate the distance from the center of the circle to the three tangent point normals and use the median method to determine the radius of the circle. For the equation of the tangent point normal Li Center The distance to the tangent point normal Li is : Take the median of the normal distances of the three tangent points as the radius, and the calculation formula is as follows; .

[0029] In this embodiment: a double-circle measurement model of the temporomandibular joint disc is constructed in the MRI image: first, a circle is drawn inside the condylar head, tangent to the upper surface and the inner and outer sides of the condylar head, and then a small circle is drawn on the condylar neck, tangent to the inner and outer edges of the condylar neck and the lower boundary of the large circle, to form a standard geometric shape of the joint structure; In the present invention, the large circle and the small circle intersect with the three tangent points of the condylar head and the neck respectively, which has certain difficulties in the actual implementation process. Therefore, the present invention introduces the three-tangent point normal intersection method to determine the center and radius of the large and small circles; In the present invention, three tangent points where the large circle is tangent to the condylar head are found and normal lines are marked respectively, the relationship between the three normal lines is analyzed, the center position of the circle constituting the large circle model is calculated according to the formula, and the distance from the center of the circle to the normal lines of the three tangent points is calculated, and the median value is taken as the radius to form a large circle inside the condylar head.

[0030] Based on the large circle model, the center and radius of the small circle are determined in the same way. The two tangent points where the small circle is tangent to the condylar neck and the tangent points where the small circle is tangent to the large circle are found, and normal lines are drawn respectively. The relationship between the three normal lines is analyzed, and the position of the center of the circle that constitutes the small circle model is calculated according to the formula. The distance from the center of the circle to the three tangent point normals is calculated, and the median value is taken as the radius to form a small circle on the condylar neck.

[0031] See also Figures 1 to 8 The specific process of constructing the small circle model based on the condylar neck in the double circle model construction module is as follows: Similarly, based on the large circle model of the condylar head, the small circle model based on the condylar neck is constructed by drawing normals at the three tangent points of the condylar neck structure, analyzing the relationship between the three normals, and calculating the coordinates and radius of the center of the circle when the tangent point normals intersect and when the tangent point normals are parallel.

[0032] In this embodiment: a double-circle model is used to construct a module to analyze and determine the medial and lateral displacement of the temporomandibular joint disc. According to the positional relationship of key points, the displacement direction is indicated by positive and negative signs. The medial displacement of the joint disc is recorded as -, and the lateral displacement is recorded as +. The medial and lateral displacement of the joint disc are marked, and the severity of the displacement is reflected by measuring geometric parameters such as angle and distance. This geometric method of building modules through the double-circle model not only reduces the dependence on multiple variables in traditional measurement, but also reduces the subjective error of manual operation through standardized models, significantly improving the repeatability and consistency of measurements.

[0033] Embodiment 2: Based on the above embodiment: See also Figures 1 to 8 , the condylar angle bisector and articular disc angle bisector generated based on key points in the displacement parameter calculation module are as follows: Connect the center of the small circle with the medial and lateral convex points of the temporomandibular joint disc and the medial and lateral convex points of the condyle and draw an angle bisector. The medial and lateral convex points of the temporomandibular joint disc correspond to the left boundary point of the joint disc and the right boundary point of the joint disc, and the medial and lateral convex points of the condyle correspond to the left boundary point of the condyle and the right boundary point of the condyle. The angle α of the angle bisector measured in the displacement parameter calculation module is used to reflect the severity of the temporomandibular joint disc displacement, and the distance between the intersection A and the intersection B is analyzed to reflect the displacement direction of the temporomandibular joint disc.

[0034] In this embodiment: the specific diagram is Figure 7 In the fifth step, the specific angle bisector angle α can be calculated using the cosine theorem, and the distance between the intersection A and the intersection B can be calculated using the Euclidean distance calculation formula, so that the angle bisector angle α and the distance between the intersection A and the intersection B can be quickly output. The present invention can convert complex anatomical structure relationships into geometric parameters through the shift parameter calculation module, making the diagnosis result more intuitive and facilitating clinicians to quickly judge the condition.

[0035] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The double-circle measurement model recognition system for the temporomandibular joint disc is characterized by: include: The image preprocessing and annotation module is used to obtain the MRI image of the patient's temporomandibular joint, and then perform standardization and edge enhancement on the MRI image, and perform regional cropping and scaling operations on the MRI image; A key point detection network module is built based on a deep learning network to input preprocessed MRI images, then optimize the network structure, annotate key points, and finally train the model; The key points include: the center of the large circle on the condylar head, the center of the small circle on the condylar neck, the left boundary point of the articular disc, the right boundary point of the articular disc, the left boundary point of the condyle, and the right boundary point of the condyle; A double circle model building module is used to build a large circle model based on the condylar head and a small circle model based on the condylar neck, and calculate the center and radius of the large circle model and the small circle model; The displacement parameter calculation module generates the condylar angle bisector and the articular disc angle bisector based on the key points, and locates the intersection based on the relationship between the condylar angle bisector and the articular disc angle bisector and the condyle and the articular disc, outputs the intersection A and the intersection B, and then measures the angle α of the angle bisector and the distance between the intersection A and the intersection B to intuitively display the displacement of the articular disc.

2. The temporomandibular joint disc double circle measurement model recognition system according to claim 1, characterized in that: The specific process of the image preprocessing and annotation module in the steps is as follows: Standardization processing: The brightness and contrast of MRI images are unified through the CLAHE algorithm, and Gaussian filtering is used to eliminate the noise of MRI images; Edge enhancement processing: The boundaries of the condylar head, condylar neck and articular disc are extracted and highlighted by using the Sobel operator; Regional cropping and scaling operations: The temporomandibular joint region in the MRI image is cropped and scaled to a uniform resolution to ensure the size consistency of the input deep learning network.

3. The temporomandibular joint disc double circle measurement model recognition system according to claim 2, characterized in that: The optimization of the network structure in the key point detection network module based on the deep learning network includes: 1×1 convolution replacement and semi-residual module; 1×1 convolution replacement: In the deep learning network, 1x1 convolution kernels are introduced in the deeper layers, i.e., near the output layer, and the 3x3 convolution kernels in the shallow layers are retained to reduce the computational complexity while fully extracting local detail features and ensuring effective modeling of key points; Semi-residual module: To enhance the feature expression capability of deep learning networks, the network structure is designed in combination with the semi-residual module. The semi-residual module introduces a learnable weighting coefficient into the traditional residual structure, and dynamically adjusts the fusion ratio of residual features and input features through the following formula: in: Refers to the weighted fusion result of multi-branch features; Refers to the features after being processed by the convolutional layer; Refers to the input features; Both refer to learning parameters, which are used to adaptively adjust the feature fusion method; The optimized deep learning network gives full play to the advantages of shallow and deep convolutional features, and enhances the feature fusion capability through the semi-residual module.

4. The temporomandibular joint disc double circle measurement model recognition system according to claim 3, characterized in that: The model training in the key point detection network module based on the deep learning network is specifically as follows: Pair the annotated MRI images with the key point coordinates to form a training data set for model training. During the model training process, the images are rotated and scaled to increase data diversity and adapt the model to different MRI image angles and qualities. The mean square error loss function is used to measure the error between the key point coordinates predicted by the deep learning network and the actual coordinates. The calculation formula is as follows: in: Loss refers to the error between the predicted key point coordinates and the actual coordinates, that is, the loss value; N Refers to the total number of samples, the total number of coordinate points in all training data; ( x i ，y i ) refers to the true coordinate value in the i-th sample; Refers to the i-th sample x and y The predicted value of the coordinates; Evaluate the model performance on the test set, calculate the error between the predicted key points and the actual key points according to the formula, and adjust the model parameters or perform hyperparameter tuning based on the test results to improve the model's performance in key point prediction.

5. The temporomandibular joint disc double circle measurement model recognition system according to claim 4, characterized in that: The purpose of the double circle model building module is to determine the center and radius of the large and small circles by the three-tangent point normal intersection method. The steps are as follows: first, a large circle is drawn inside the condylar head, so that it is tangent to the upper surface and the inner and outer sides of the condylar head, and then a small circle is drawn inside the condylar neck, so that it is tangent to the inner and outer edges of the condylar neck and the lower boundary of the large circle, so as to form a standard geometric shape of the joint structure; The process of constructing the great circle model based on the condylar head is as follows: Draw normal lines at the three tangent points of the condylar structure, analyze the intersection relationship of the three tangent point normal lines, and calculate the coordinates of the center of the circle and the radius; If the three tangent point normals intersect each other; The intersection points of the three tangent normal lines are Construct the centroid of the triangle as the center of the circle Position; then the radius of the great circle model based on the condylar head is calculated as the center of the great circle when the three tangent point normals intersect each other. The distance to the normal line of the three tangent points is taken as the radius of the circle; If the three tangent point normals do not intersect, they are parallel; first, the tangent point normal L1: And the tangent point normal 2 L2: Then obtain the intersection of the third tangent point normal L3 with the tangent point normal 1 L1 and the tangent point normal 2 L2, and then obtain two valid intersection points. Then use the midpoint of these two intersection points as the center of the replacement large circle Then combine the remaining non-parallel tangent point normal line 3 L3 and make its tangent point and We can calculate the center of gravity by weighting, further correct the position of the center of the circle, improve the stability of the center of the circle positioning, and obtain the center of the large circle when the three tangent point normals do not intersect. ; Similarly, according to the three tangent points, the center of the great circle is , calculate the distance from the center of the circle to the three tangent point normals and use the median method to determine the radius of the circle. For the equation of the tangent point normal Li Center The distance to the tangent point normal Li is Take the median of the normal distances of the three tangent points as the radius.

6. The temporomandibular joint disc double circle measurement model recognition system according to claim 5, characterized in that: The specific process of constructing the small circle model based on the condylar neck in the double circle model construction module is as follows: Similarly, based on the large circle model of the condylar head, the small circle model based on the condylar neck is constructed by drawing normals at the three tangent points of the condylar neck structure, analyzing the relationship between the three normals, and calculating the coordinates and radius of the center of the circle when the tangent point normals intersect and when the tangent point normals are parallel.

7. The temporomandibular joint disc double circle measurement model recognition system according to claim 6, characterized in that: The generation of the condylar angle bisector and the articular disc angle bisector based on the key points in the displacement parameter calculation module is specifically as follows: Connect the center of the small circle with the inner and outer convex points of the temporomandibular joint disc and the inner and outer convex points of the condyle and draw an angle bisector, the inner and outer convex points of the temporomandibular joint disc correspond to the left boundary point of the joint disc and the right boundary point of the joint disc, and the inner and outer convex points of the condyle correspond to the left boundary point of the condyle and the right boundary point of the condyle.

8. The temporomandibular joint disc double circle measurement model recognition system according to claim 7, characterized in that: The angle α of the angle bisector measured in the displacement parameter calculation module is used to reflect the severity of the displacement of the temporomandibular joint disc, and the distance between the intersection point A and the intersection point B is analyzed to reflect the displacement direction of the temporomandibular joint disc.

Images