An alveolar bone fracture detection system, device, medium and program product

Abstract

The present invention discloses an alveolar bone fracture detection system, device, medium and program product, relating to the technical field of data recognition, including: extracting multi-scale features from the acquired oral image data, and segmenting to obtain a fracture region and a root region; extracting a centerline from the fracture region, accumulating the three-dimensional Euclidean distance along the centerline voxel by voxel to obtain the fracture line length, calculating the global average curvature after representing the centerline as a function of arc length, detecting the number of branch points in the fracture region, calculating the minimum distance between the fracture region and the root region, and thus identifying the fracture type according to the above characteristic parameters; taking the centroid of the fracture region as a seed point, and determining the gray value of the seed point, and merging the new voxels that belong to the neighborhood expansion region of the original fracture region and the difference between the gray value and the gray value of the seed point meets the set threshold into the original fracture region to obtain an optimized fracture region, and calculating the regional volume of the optimized fracture region. The objectivity, accuracy and efficiency of diagnosis are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data recognition, and in particular, to an alveolar bone fracture detection system, device, medium, and program product. Background Art

[0002] In recent years, intelligent algorithms, such as deep learning and image segmentation models, have made significant progress in medical image diagnosis, but there are still the following technical bottlenecks in the accurate diagnosis of alveolar bone fractures.

[0003] (1) Insufficient adaptability to complex fracture morphologies.

[0004] (1-1) Existing models, such as convolutional neural networks and Mask R-CNN (Mask Region-based Convolutional Neural Network, a deep learning model for object detection and instance segmentation), are mostly designed based on regular-shaped objects, and have limited segmentation accuracy for slender, multi-branched, and low-contrast fracture lines of alveolar bone fractures. For example, the detection of branches in comminuted fractures often results in breaks due to feature loss (false negative rate > 15%).

[0005] (1-2) In 3D segmentation, traditional models (such as 3D convolutional neural networks), which rely on local context, are difficult to model the continuity of long fracture lines. Especially when the inter-slice resolution of CBCT (Cone beam CT) is insufficient, "stepped artifacts" are likely to occur.

[0006] (2) Challenges of small samples and domain generalization.

[0007] (2-1) The imaging data of alveolar bone fractures is scarce and the annotation cost is high. Existing methods rely on large-scale annotated data for training, and are prone to overfitting in small-sample scenarios (such as rare periapical fractures), and the generalization performance drops significantly (the cross-center test accuracy drops by 20% - 30%).

[0008] (2-2) Although transfer learning can alleviate the data shortage, the pre-trained model has a large difference from the medical image domain, and it is difficult to align the feature spaces, resulting in poor adaptability of the model to the gray distribution and noise patterns.

[0009] (3) Defects in the fusion of morphological-semantic features.

[0010] (3-1) Most studies only rely on end-to-end segmentation, lacking explicit modeling of fracture morphology (length, curvature) and anatomical structure (tooth root position), resulting in the disconnection between classification results and clinical standards.

[0011] (3-2) The fusion strategy of multi-modal features (such as gray scale, texture, and spatial coordinates) is single, and traditional concatenation or weighting methods are difficult to capture high-order interactions, restricting the discriminant ability of the classifier.

[0012] (4)Error accumulation in three-dimensional volume measurement.

[0013] (4-1)Existing volume calculations are mostly based on layer-by-layer two-dimensional mask accumulation, ignoring the interlayer continuity. The volume underestimation of oblique or spiral fractures can reach 10%-20%.

[0014] (4-2)Partial volume effect compensation methods (such as linear interpolation) fail at complex fracture boundaries. Region growing algorithms rely on fixed thresholds and it is difficult to adapt to the bone density differences of different patients. Summary of the Invention

[0015] To solve the above problems, the present invention proposes an alveolar bone fracture detection system, device, medium and program product, which can accurately locate the fracture line, automatically distinguish the fracture type, calculate the volume of the fracture area, and significantly improve the objectivity, accuracy and efficiency of diagnosis.

[0016] To achieve the above object, the present invention adopts the following technical solutions:

[0017] In a first aspect, the present invention provides an alveolar bone fracture detection system, including:

[0018] A feature extraction module configured to extract multi-scale features from the acquired oral image data, thereby segmenting the fracture area and the root area;

[0019] A type recognition module configured to extract the centerline of the fracture area, accumulate the three-dimensional Euclidean distance along the centerline voxel by voxel to obtain the fracture line length, calculate the global average curvature after representing the centerline as a function of arc length, detect the number of branch points in the fracture area, calculate the minimum distance between the fracture area and the root area, and thereby identify the fracture type according to the fracture line length, global average curvature, number of branch points, and the minimum distance between the fracture area and the root area;

[0020] A volume calculation module configured to use the centroid of the fracture area as a seed point, determine the gray value of the seed point, and merge the new voxels that belong to the neighborhood expansion area of the original fracture area and the difference between the gray value and the gray value of the seed point meets the set threshold into the original fracture area to obtain an optimized fracture area, and calculate the regional volume of the optimized fracture area.

[0021] As an alternative implementation, in the feature extraction module, during the process of extracting multi-scale features, an adaptive weight is calculated for each feature layer:

[0022] ;

[0023] Wherein, is the step of the oral image data of the Layer feature map; and are the mean and standard deviation of the layer feature map respectively; is the smoothing factor; is the th layer feature's adaptive weight; is the spatial position of each voxel in

[0024] Multi-scale feature fusion is performed based on the adaptive weight:

[0025] ;

[0026] wherein, is the fused feature, is the total number of feature layers.

[0027] As an alternative implementation, in the type recognition module, the fracture line length is:

[0028] ;

[0029] wherein, is the coordinate of the th centerline voxel; is the coordinate of the th centerline voxel; is the total number of centerline voxels.

[0030] As an alternative implementation, in the type recognition module, the process of calculating the global mean curvature includes: representing the centerline as a function of the arc length , and the global mean curvature is:

[0031] ;

[0032] wherein, is the first derivative of the function; is the second derivative of the function; is the total number of centerline voxels.

[0033] As an alternative implementation, in the type recognition module, the process of detecting the number of branch points in the fracture area includes: traversing the skeleton points on the centerline, determining whether there are more than a set threshold number of connected directions in the neighborhood of the current skeleton point, and if so, regarding the current skeleton point as a branch point, thereby determining the number of branch points in the fracture area.

[0034] As an alternative implementation, in the type recognition module, the process of recognizing the fracture type includes:

[0035] If the length of the fracture line is greater than or equal to the set length threshold, the global average curvature is less than or equal to the set first curvature threshold, and the number of branch points is less than or equal to the set first number threshold, it is a linear fracture;

[0036] If the number of branch points is greater than or equal to the set second number threshold, or the global average curvature is greater than the set second curvature threshold, it is a comminuted fracture;

[0037] If the minimum distance between the fracture area and the root area is less than the distance threshold, it is a periapical fracture.

[0038] As an alternative implementation, in the volume calculation module, the centroid coordinates are ; where are the coordinates of all fracture voxels in the binary mask of the fracture area, is the total number of all fracture voxels in the fracture area;

[0039] New voxel The condition for being merged is: satisfying ; where is the gray value of the seed point, Tolerance, neighborhood represents the neighborhood expansion area of the original fracture area;

[0040] Regional volume [[ID=AF=35]] is:

[0041] ;

[0042] Where is the cross-sectional area of the th layer of voxels in the original fracture area; is the slice spacing of the oral image data; is the volume of the new voxel, and M is the number of new voxels.

[0043] In a second aspect, the present invention provides an electronic device, including a memory, a processor, and computer instructions stored on the memory and running on the processor. When the computer instructions are run by the processor, an alveolar bone fracture detection method is completed. The alveolar bone fracture detection method includes:

[0044] Extract multi-scale features from the acquired oral image data, and thereby segment the fracture area and the root area;

[0045] Extract the centerline of the fracture region, accumulate the three-dimensional Euclidean distance voxel by voxel along the centerline to obtain the fracture line length, calculate the global average curvature after representing the centerline as a function of arc length, detect the number of branch points in the fracture region, calculate the minimum distance between the fracture region and the root region, and thus identify the fracture type based on the fracture line length, global average curvature, number of branch points, and the minimum distance between the fracture region and the root region;

[0046] Use the centroid of the fracture region as the seed point, determine the gray value of the seed point, and merge the new voxels that belong to the neighborhood expansion region of the original fracture region and whose gray value difference from the seed point gray value meets the set threshold into the original fracture region to obtain the optimized fracture region, and calculate the regional volume of the optimized fracture region.

[0047] In a third aspect, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the alveolar bone fracture detection method. The alveolar bone fracture detection method includes:

[0048] Extract multi-scale features from the obtained oral image data, and thus segment the fracture region and the root region;

[0049] Extract the centerline of the fracture region, accumulate the three-dimensional Euclidean distance voxel by voxel along the centerline to obtain the fracture line length, calculate the global average curvature after representing the centerline as a function of arc length, detect the number of branch points in the fracture region, calculate the minimum distance between the fracture region and the root region, and thus identify the fracture type based on the fracture line length, global average curvature, number of branch points, and the minimum distance between the fracture region and the root region;

[0050] Use the centroid of the fracture region as the seed point, determine the gray value of the seed point, and merge the new voxels that belong to the neighborhood expansion region of the original fracture region and whose gray value difference from the seed point gray value meets the set threshold into the original fracture region to obtain the optimized fracture region, and calculate the regional volume of the optimized fracture region.

[0051] In a fourth aspect, the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the alveolar bone fracture detection method. The alveolar bone fracture detection method includes:

[0052] Extract multi-scale features from the obtained oral image data, and thus segment the fracture region and the root region;

[0053] Extract the centerline of the fracture area, accumulate the three-dimensional Euclidean distance voxel by voxel along the centerline to obtain the fracture line length, represent the centerline as a function of arc length and then calculate the global average curvature, detect the number of branch points within the fracture area, calculate the minimum distance between the fracture area and the root area, and thereby identify the fracture type based on the fracture line length, global average curvature, number of branch points, and the minimum distance between the fracture area and the root area;

[0054] Use the centroid of the fracture area as the seed point, determine the gray value of the seed point, and merge the new voxels that belong to the neighborhood expansion area of the original fracture area and whose gray value difference from the seed point gray value meets the set threshold into the original fracture area to obtain the optimized fracture area, and calculate the area volume of the optimized fracture area.

[0055] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0056] The present invention extracts multi-scale features from oral imaging data and segments the fracture area and the root area; during the process of extracting multi-scale features, an adaptive feature extraction mechanism is introduced. When processing each layer of features, the weights are automatically adjusted according to the saliency of different regions, which can optimize the model according to the different region features in the oral imaging data, making the detection of key regions (such as fracture lines) more sensitive and avoiding the over-concern of traditional methods for unimportant regions. At the same time, by introducing a multi-scale feature fusion strategy, the edge information of the low layer and the semantic information of the high layer are effectively combined, which not only improves the accuracy and robustness of image segmentation, but also enhances the model's processing ability for complex images (such as noise interference or edge-blurred regions) and improves the diagnostic accuracy of fracture lines.

[0057] The present invention accurately locates the fracture line by improving the multi-scale feature fusion mechanism of the segmentation model, introducing conditional random field optimization, and the dynamic region growth algorithm. According to the fracture line morphology, position, and relationship with the root, it automatically distinguishes types such as linear fracture, comminuted fracture, and periapical fracture. Through three-dimensional reconstruction and the region growth algorithm, it accurately calculates the volume of the fracture area; integrates the diagnostic results and outputs a quantitative report including fracture type, location, and volume, significantly improving the objectivity, accuracy, and efficiency of diagnosis.

[0058] Based on the three-dimensional voxel model of the fracture area binary mask, the present invention directly calculates the actual space volume, avoiding the errors caused by projection superposition in traditional two-dimensional measurements. It supports the volume integration of complex fracture morphologies (such as multi-regions of comminuted fractures), and at the same time, the volume error compensation part can dynamically compensate for the missing volume of scattered fracture fragments. The voxel refinement is performed using the region growth algorithm, and by precisely controlling the growth conditions, the continuity between the newly added voxels and the fracture area is ensured, effectively compensating for the volume measurement errors caused by image noise or incomplete segmentation and improving the accuracy of volume calculation.

[0059] Advantages of additional aspects of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to the provided drawings.

[0061] Figure 1 Schematic diagram of the alveolar bone fracture detection system provided in Embodiment 1 of the present invention;

[0062] Figure 2 Flowchart of the alveolar bone fracture detection method implemented by the alveolar bone fracture detection system provided in Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] The present invention will be further described below in conjunction with the drawings and embodiments.

[0064] It should be noted that the following detailed description is exemplary and is intended to provide further illustration of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0065] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "comprising" and "including" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0066] In the case of no conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0067] Embodiment 1

[0068] This embodiment proposes an alveolar bone fracture detection system, as Figure 1 - Figure 2 shown, including:

[0069] A feature extraction module configured to extract multi-scale features from the acquired oral image data, thereby segmenting the fracture area and the root area;

[0070] A type recognition module, configured to extract a center line from a fracture region, accumulate the three-dimensional Euclidean distance along the center line voxel by voxel to obtain the fracture line length, calculate the global average curvature after representing the center line as a function of arc length, detect the number of branch points in the fracture region, calculate the minimum distance between the fracture region and the root region, and thereby identify the fracture type according to the fracture line length, the global average curvature, the number of branch points, and the minimum distance between the fracture region and the root region;

[0071] A volume calculation module, configured to use the centroid of the fracture region as a seed point, determine the gray value of the seed point, merge the new voxels that belong to the neighborhood expansion region of the original fracture region and the difference between the gray value and the gray value of the seed point satisfies a set threshold into the original fracture region to obtain an optimized fracture region, and calculate the regional volume of the optimized fracture region.

[0072] In this embodiment, the system further includes a data acquisition and preprocessing module, which acquires the oral image data of the patient through a CBCT (Cone beam CT) device and performs preprocessing using an adaptive denoising algorithm to remove noise while ensuring the retention of details, ensuring the high quality of the oral image data and the accuracy of subsequent analysis.

[0073] The denoising optimization formula is:

[0074] ;

[0075] Wherein, is the preprocessed oral image data; is the pixel value of the original oral image data at the coordinate ; is the pixel value of the preprocessed oral image data at the coordinate ; is a region defined on the oral image data, indicating summation within this region; is a balance parameter for adjusting the weight between denoising and detail retention; is the gradient of the preprocessed oral image data at the coordinate , indicating the degree of image change.

[0076] This embodiment introduces an adaptive denoising technique during the denoising process, which can remove noise while retaining details, significantly improving the image quality. Compared with traditional denoising methods, the method of this embodiment can better handle complex background noise, ensuring that the image is more accurate in terms of details and contours, providing a reliable image basis for subsequent feature extraction.

[0077] In this embodiment, in the feature extraction module, the Segment Anything Model (SAM) is used as the feature extraction network to perform fine segmentation and feature capture on the fracture area.

[0078] The SAM model consists of an encoder and a decoder. The encoder is responsible for extracting multi-scale features of the input oral image data, and the decoder segments the fracture area based on the extracted multi-scale features.

[0079] Its basic process is expressed as:

[0080] ;

[0081] Among them, is the oral image data of the th step, and are the weight parameters of the encoder and decoder of the th step respectively, is the segmentation result of the th step.

[0082] The goal of the encoder is to extract multi-scale features of the input oral image data and capture hierarchical information from local details (such as the edge of the fracture line) to global semantics (such as the jawbone anatomical structure). ResNet-50 is used as the encoder backbone network, and the hierarchical structure is shown in Table 1. Among them, Conv is the convolutional layer, MaxPool is the maximum pooling layer, and ResBlock is the residual block.

[0083] Table 1 Hierarchical structure;

[0084] .

[0085] The structure of the residual block (ResBlock) is: ; Among them, is the convolution operation, which includes two 3×3 convolutional layers, each followed by a batch normalization layer and a ReLU activation function; is the input feature map, is the output feature map; represents the set of weight parameters in the convolution operation, specifically including the kernel weights and bias terms of each 3×3 convolutional layer, used to perform convolution operations on the input feature map to generate .

[0086] The encoder outputs feature maps of 5 scales (corresponding to stages 1-5), which are respectively marked as: ; Among them, It is defined as a low-level feature that retains detailed information such as fracture line edge and texture; It is defined as a high-level feature that encodes the overall morphology of the jaw and the semantic context of the fracture area.

[0087] The goal of the decoder is to gradually restore the spatial resolution, fuse the multi-scale features of the encoder, and generate a segmentation mask for the fracture area. Specifically, it includes:

[0088] This embodiment introduces adaptive weight calculation in the feature extraction process of the SAM model to automatically adjust the importance of features in different regions. For each feature layer, the adaptive weight is:

[0089] ;

[0090] in, It is The first step of oral imaging data Layer feature map; and They are The mean and standard deviation of the layer feature map; is the smoothing factor; It is Step Adaptive weights of layer features, Reflection feature layer The significance of , a high weight indicates that the layer feature contributes more to the fracture line segmentation; for The spatial location of each voxel (or pixel) in ; Represents the feature map All spatial locations in The voxels are traversed and summed.

[0091] Based on adaptive weights, multi-scale feature fusion is performed to combine features of different scales for information enhancement. Multi-scale feature fusion is expressed as:

[0092] ;

[0093] in, is the fusion feature, is the total number of feature layers.

[0094] Fusion Features Directly input to the decoder as the initial input feature of the decoder, and then use the Skip Connection method to connect the encoder stage Feature map Feature maps corresponding to the decoder layer Stitching, gradually restore the spatial resolution and generate segmentation masks. For example, stage 4 decoding: ; wherein, is the channel dimension concatenation; is the transposed convolution, stride = 2, kernel size = 3×3.

[0095] After the decoder generates the segmentation result, the semantic information and edge details of the fused features are combined, and then further optimized through the conditional random field. The conditional random field smooths the segmentation edges by modeling the spatial relationship between pixels, ensuring the continuity and accuracy of the fracture lines.

[0096] In this embodiment, the advantages of this module are as follows:

[0097] Adaptive weight calculation mechanism: An adaptive feature extraction mechanism is introduced. When processing each layer of features, the weights are automatically adjusted according to the saliency of different regions. This mechanism can optimize the model according to the different regional features in the oral cavity image data, making the detection of key regions (such as fracture lines) more sensitive and avoiding the over - attention of traditional methods to unimportant regions.

[0098] Multi - scale feature fusion: By introducing a multi - scale feature fusion strategy, the low - level edge information and high - level semantic information are effectively combined, not only improving the accuracy and robustness of image segmentation, but also enhancing the model's ability to process complex images (such as noise - disturbed or edge - blurred regions), and improving the diagnostic accuracy of fracture lines.

[0099] In this embodiment, after the fracture region is segmented, the conditional random field (CRF) is used to optimize the segmentation result to improve the accuracy of the segmentation edge and the reliability of the overall diagnosis. Its basic principle is to regard the segmentation result as nodes and optimize the segmentation by establishing the relationship between nodes.

[0100] The CRF energy function is:

[0101] ;

[0102] Wherein, is the output label of the model; is the input image; is the cost of a single node (i.e., the segmentation confidence of each pixel); is the smooth cost between adjacent nodes, which can ensure the classification consistency of adjacent pixels.

[0103] Single - point cost calculation: The single - point cost of a node is calculated based on the pixel value and classification probability of the segmented image, specifically: ; wherein, is the pixel is the probability of being classified into a certain category.

[0104] Double - point cost calculation: The double - point cost is used to enhance the smoothness of the segmentation boundary and is usually achieved through a Gaussian kernel function: ; where is the feature vector of pixel and . is a parameter that controls the degree of smoothness.

[0105] Parameter optimization: The Iterative Conditional Mode (ICM) algorithm is used to optimize the energy function. The class label of each pixel is iteratively updated until convergence. In each iteration, the goal of optimization is to reduce the value of the energy function: ; where is the label at the th step and the th iteration, is the label at the th step and the th iteration, is the learning rate, is the gradient of the energy function at the th iteration. Through continuous iteration, the label prediction is optimized to ensure that the segmentation result is accurate and consistent with the feature extraction in the previous stage.

[0106] In this embodiment, a conditional random field is used to post - process the preliminary segmentation result to further optimize the segmentation edge. The CRF model can smooth adjacent pixels by modeling the spatial relationship between pixels, enhance the continuity of the edge, and make the segmentation result more refined and realistic.

[0107] In this embodiment, the type recognition module realizes the automatic classification of the alveolar bone fracture type based on the morphological features, spatial distribution of the fracture line, and its relationship with the surrounding anatomical structures (such as tooth roots).

[0108] Specifically, it includes:

[0109] (1) Extraction of fracture line morphological features.

[0110] (1 - 1) Fracture line skeletonization and centerline extraction.

[0111] According to the binary mask of the fracture area optimized by CRF (1 represents the fracture area, 0 represents the normal area), the centerline coordinates of the fracture area are extracted :

[0112] .

[0113] Skeletonization is an image processing technique used to simplify the fracture area into a centerline while preserving its basic shape and topological structure. The skeletonization and the identification of skeleton points are based on the binary mask of the fracture area optimized by CRF. It is implemented;

[0114] Specifically, it includes: calculating the distance from each point in the fracture area to the nearest boundary, and based on the obtained distance results, extracting the skeleton through iterative erosion operations. The erosion operation gradually shrinks the area until only the centerline remains. Each skeleton point is a coordinate point on the centerline of the fracture line, representing a key position of the fracture line.

[0115] (1-2) Calculation of characteristic parameters.

[0116] (a) Accumulating the three-dimensional Euclidean distance along the centerline voxel by voxel to calculate the fracture line length. Its physical meaning is to reflect the spatial extension degree of the fracture line. If the fracture line length is greater than the set threshold, it is regarded as a long fracture line;

[0117] ;

[0118] Among them, voxel is the abbreviation of Volume Pixel. is the coordinate of the th centerline voxel; is the coordinate of the th centerline voxel; is the total number of centerline voxels.

[0119] (b) Calculating the global average curvature to reflect the local bending degree of the centerline. If the average curvature is greater than the set threshold (for example, C>0.5), it is regarded as a high curvature value, indicating that there is a sharp turn in the fracture line, which is common in comminuted fractures.

[0120] Specifically, it includes: representing the centerline as a function of arc length to parameterize the centerline: ;

[0121] The curvature formula is: ; Among them, is the first derivative (tangent direction); is the second derivative (curvature direction);

[0122] The global average curvature is: .

[0123] (c) Traverse the skeleton points and determine whether there are more than a set threshold number of connected directions in the neighborhood of the current skeleton point. If so, regard the current skeleton point as a branch point to detect the number of branch points in the fracture area. , It indicates that there are multiple bifurcations in the fracture line, which is a typical feature of comminuted fractures.

[0124] Specifically, for each skeleton point, check other points within a set range around this skeleton point (i.e., the neighborhood). If there are 3 or more directions in the neighborhood that extend from this skeleton point (i.e., there are also fracture lines in these directions), then this skeleton point is considered a branch point.

[0125] (d) By traversing all fracture area voxels and root area voxels , calculate the Euclidean distance between the two, and finally take the minimum value as the minimum distance between the fracture area and the root area ; that is, by calculating the minimum distance between the fracture area and the root area of , if is less than the set distance threshold (such as ), it is prompted that the fracture involves the root and needs to be additionally marked as "periapical fracture";

[0126] ;

[0127] Among them, is the three-dimensional coordinate of the th voxel in the fracture area, representing the actual spatial position of this voxel in the image data; is the three-dimensional coordinate of the th voxel in the root area, representing the spatial position of a specific voxel in the root area.

[0128] (2) Classification logic and threshold setting.

[0129] (a) If the fracture line length is greater than or equal to the set length threshold, and the global average curvature is less than or equal to the set first curvature threshold, and the number of branch points is less than or equal to the set first number threshold (such as ), then it is a linear fracture; the clinical basis is that it is long, has a low curvature and no bifurcation, which conforms to the morphological characteristics of a linear fracture.

[0130] (b) If the number of branch points is greater than or equal to the set second number threshold, or the global average curvature is greater than the set second curvature threshold (such as ), then it is a comminuted fracture; the clinical basis is that multiple branches or high curvature indicate a complex fracture line, which conforms to the definition of a comminuted fracture.

[0131] (c) If the minimum distance between the fracture region and the root region is less than the distance threshold (for example ), it is a periapical fracture; the clinical basis is that the fracture involves the root, which conforms to the definition of periapical fracture.

[0132] In this embodiment, when calculating the volume of the fracture region , the region growing algorithm is first used to further refine the voxels to ensure the accuracy of volume measurement.

[0133] Specifically:

[0134] Seed selection: Use the centroid coordinates of the original fracture region as the seed point to ensure that the algorithm focuses on the core region; the centroid coordinates are: ; where are the coordinates of all fracture voxels in the binary mask of the fracture region, is the total number of all fracture voxels in the binary mask of the fracture region.

[0135] The conditions for the newly added voxel to be merged need to be satisfied simultaneously: ; where is the gray value of the seed point (consistent with the typical value of the fracture region, such as 200 - 400 HU); , representing the tolerance. In the CBCT image, the gray value of the fracture region is usually within a certain range (such as 200 - 400 HU), and a tolerance range of 50 HU can cover the gray value fluctuations caused by image noise or partial volume effect. is set based on clinical experience and experimental verification, which can balance sensitivity and specificity, neither missing real fracture voxels nor misclassifying non - fracture voxels; the neighborhood represents the neighborhood expansion region of the binary mask of the original fracture region to ensure the continuity of the newly added voxel with the fracture region.

[0136] Thus, by combining the volume of the newly added voxel grown by the region growing algorithm, the volume of the fracture region is calculated as:

[0137] ;

[0138] ;

[0139] where is the cross - sectional area of the th layer voxel in the original fracture region, obtained by converting pixel counting and the physical size of a single pixel; is the CBCT slice spacing (from data acquisition); Add the voxel volume for regional growth, and M is the number of newly added voxels. is the volume error compensation part; is the width of the newly added voxel, is the height of the newly added voxel, is the layer spacing of the newly added voxels. For isotropic voxels, , that is, the sizes of each voxel in the three directions are the same.

[0140] In this embodiment, based on the three-dimensional voxel model of the fracture region binary mask, the actual space volume is directly calculated, avoiding the error caused by projection superposition in traditional two-dimensional measurements. It supports the volume integration of complex fracture morphologies (such as multi-regional comminuted fractures), and at the same time, the volume error compensation part can dynamically compensate for the missing volume of scattered fracture fragments.

[0141] In this embodiment, the region growing algorithm is used for voxel refinement. By precisely controlling the growth conditions, the continuity between the newly added voxels and the fracture region is ensured, effectively compensating for the volume measurement error caused by image noise or incomplete segmentation, and improving the accuracy of volume calculation.

[0142] In this embodiment, by extracting the centroid coordinates and the volume of the lesion area from the segmentation results, accurate fracture positions and quantified volume data can be provided for doctors. It can not only help doctors quickly understand the severity of the fracture, but also provide quantitative support for treatment decisions. And a complete diagnostic report is generated according to the automatic diagnosis results, including detailed information such as fracture type, position, and volume. By automatically generating reports, the workload of doctors is reduced, and the accuracy and consistency of the reports are ensured.

[0143] The alveolar bone fracture detection system proposed in this embodiment, through the deep integration of deep learning, three-dimensional reconstruction, and adaptive optimization technologies, etc., realizes the efficient and accurate diagnosis and quantitative analysis of alveolar bone fractures. Taking CBCT images as input, relying on the improved SAM model combined with the conditional random field optimization algorithm, the fine segmentation of the fracture line is completed; through the joint analysis of multi-scale morphological features (length, curvature, number of branches) and spatial relationships (distance from the tooth root), the automatic classification of linear fractures, comminuted fractures, and periapical fractures is realized; and based on the three-dimensional mask and the dynamic region growing algorithm, the high-precision measurement of the fracture volume is completed.

[0144] Compared with the traditional diagnosis method that relies on doctors' experience, the system in this embodiment has the following core advantages:

[0145] Full-process automation: From image preprocessing to report generation, it reduces manual intervention and improves the diagnostic efficiency.

[0146] Multi-dimensional quantitative output: It provides fracture type, position coordinates, volume ratio, and morphological parameters, providing quantitative support for surgical planning and prognosis evaluation.

[0147] Clinical interpretability: All diagnostic results are based on traceable morphological and imaging features (such as curvature calculation, gray-scale constraint), which conform to the cognitive logic of doctors.

[0148] Enhanced robustness: Mechanisms such as adaptive denoising and region-growing compensation significantly suppress the influence of image noise and partial volume effect.

[0149] It should be noted that the acquisition of all data is based on compliance with laws and regulations and user consent, and the data is legally applied.

[0150] In more embodiments, there is also provided:

[0151] An electronic device, including a memory, a processor, and computer instructions stored on the memory and running on the processor. When the computer instructions are run by the processor, an alveolar bone fracture detection method is completed. The alveolar bone fracture detection method includes:

[0152] Extract multi-scale features from the acquired oral image data, and thus segment the fracture area and the root area;

[0153] Extract the centerline of the fracture area, accumulate the three-dimensional Euclidean distance along the centerline voxel by voxel to obtain the fracture line length, represent the centerline as a function of arc length, then calculate the global average curvature, detect the number of branch points in the fracture area, calculate the minimum distance between the fracture area and the root area, and thus identify the fracture type according to the fracture line length, global average curvature, number of branch points, and the minimum distance between the fracture area and the root area;

[0154] Use the centroid of the fracture area as the seed point, and determine the gray-scale value of the seed point. Merge the new voxels that belong to the neighborhood expansion area of the original fracture area and the difference between the gray-scale value and the seed point gray-scale value meets the set threshold into the original fracture area to obtain the optimized fracture area, and calculate the regional volume of the optimized fracture area.

[0155] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0156] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A part of the memory may also include a non-volatile random memory. For example, the memory may also store information about the device type.

[0157] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete an alveolar bone fracture detection method. The alveolar bone fracture detection method includes:

[0158] Extract multi-scale features from the acquired oral imaging data, and thereby segment a fracture region and a tooth root region;

[0159] Extract a centerline from the fracture region, accumulate the three-dimensional Euclidean distance voxel by voxel along the centerline to obtain the fracture line length, calculate the global average curvature after representing the centerline as a function of arc length, detect the number of branch points in the fracture region, calculate the minimum distance between the fracture region and the tooth root region, and thereby identify the fracture type according to the fracture line length, the global average curvature, the number of branch points, and the minimum distance between the fracture region and the tooth root region;

[0160] Use the centroid of the fracture region as a seed point, and determine the gray value of the seed point. Merge the new voxels that belong to the neighborhood expansion region of the original fracture region and whose gray value difference from the seed point gray value satisfies the set threshold into the original fracture region to obtain an optimized fracture region, and calculate the regional volume of the optimized fracture region.

[0161] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory, and the processor reads the information in the memory and combines its hardware to complete the steps of the above method. To avoid repetition, it will not be described in detail here.

[0162] A computer program product includes a computer program, which, when executed by a processor, implements an alveolar bone fracture detection method. The alveolar bone fracture detection method includes:

[0163] Extract multi-scale features from the acquired oral imaging data, and thereby segment a fracture region and a tooth root region;

[0164] Extract a centerline from the fracture region, accumulate the three-dimensional Euclidean distance voxel by voxel along the centerline to obtain the fracture line length, calculate the global average curvature after representing the centerline as a function of arc length, detect the number of branch points in the fracture region, calculate the minimum distance between the fracture region and the tooth root region, and thereby identify the fracture type according to the fracture line length, the global average curvature, the number of branch points, and the minimum distance between the fracture region and the tooth root region;

[0165] Taking the centroid of the fracture region as the seed point, determining the gray value of the seed point, and merging the new voxels that belong to the neighborhood expansion region of the original fracture region and whose difference in gray value from the seed point gray value satisfies the set threshold into the original fracture region to obtain the optimized fracture region, and calculating the regional volume of the optimized fracture region.

[0166] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which are executed in a device on a target real or virtual processor to perform the processes / methods as described above. Generally, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided as needed. The machine-executable instructions for program modules can be executed within a local or distributed device. In a distributed device, program modules can be located in local and remote storage media.

[0167] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing devices, such that when the program code is executed by the computer or other programmable data processing devices, the functions / operations specified in the flowchart and / or block diagram are implemented. The program code can be executed entirely on the computer, partially on the computer, as an independent software package, partially on the computer and partially on a remote computer, or entirely on a remote computer or server.

[0168] In the context of the present invention, the computer program code or related data can be carried by any suitable carrier so that a device, apparatus, or processor can perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, etc. Examples of signals can include electrical, optical, radio, acoustic, or other forms of propagated signals, such as carrier waves, infrared signals, etc.

[0169] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in conjunction with this embodiment can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of this application.

[0170] Although the specific implementation manners of the present invention have been described above in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications or deformations that can be made without creative efforts on the basis of the technical solutions of the present invention are still within the protection scope of the present invention.

Claims

1. An alveolar bone fracture detection system, characterized in that, Including: A feature extraction module, configured to extract multi-scale features from the acquired oral imaging data, thereby segmenting a fracture region and a root region; A type recognition module, configured to extract a centerline from the fracture region, accumulate the three-dimensional Euclidean distance voxel by voxel along the centerline to obtain the fracture line length, calculate the global average curvature after representing the centerline as a function of arc length, detect the number of branch points in the fracture region, calculate the minimum distance between the fracture region and the root region, and thereby identify the fracture type according to the fracture line length, the global average curvature, the number of branch points, and the minimum distance between the fracture region and the root region; A volume calculation module, configured to use the centroid of the fracture region as a seed point, determine the gray value of the seed point, merge the new voxels that belong to the neighborhood expansion region of the original fracture region and whose gray value difference from the seed point gray value satisfies a set threshold into the original fracture region to obtain an optimized fracture region, and calculate the regional volume of the optimized fracture region.

2. The alveolar bone fracture detection system according to claim 1, wherein In the feature extraction module, during the process of extracting multi-scale features, an adaptive weight is calculated for each feature layer: ； Among them, is the th layer feature map of the th step oral cavity image data; and are the mean and standard deviation of the th layer feature map respectively; is the smoothing factor; is the th step th layer feature's adaptive weight; is the spatial position of each voxel in Multi-scale feature fusion is performed based on the adaptive weight: ； Among them, is the fused feature, is the total number of feature layers.

3. The alveolar bone fracture detection system according to claim 1, characterized in that In the type recognition module, the length of the fracture line is as follows: ； Among them, is the coordinate of the th central line voxel; is the coordinate of the th central line voxel; is the total number of central line voxels.

4. The alveolar bone fracture detection system according to claim 1, characterized in that, In the type recognition module, the process of calculating the global average curvature includes: representing the center line as a function of arc length function , the global average curvature is as follows: ； Among them, is the first-order derivative of the function; is the second-order derivative of the function; is the total number of centerline voxels.

5. The alveolar bone fracture detection system according to claim 1, wherein, In the type recognition module, the process of detecting the number of branch points in the fracture region includes: traversing the skeleton points on the centerline, determining whether there are more than a set threshold number of connected directions in the neighborhood of the current skeleton point, and if so, regarding the current skeleton point as a branch point, thereby determining the number of branch points in the fracture region.

6. The alveolar bone fracture detection system according to claim 1, characterized in that, In the type recognition module, the process of identifying the fracture type includes: If the fracture line length is greater than or equal to a set length threshold, and the global average curvature is less than or equal to a set first curvature threshold, and the number of branch points is less than or equal to a set first number threshold, it is a linear fracture; If the number of branch points is greater than or equal to a set second number threshold, or the global average curvature is greater than a set second curvature threshold, it is a comminuted fracture; If the minimum distance between the fracture region and the root region is less than a distance threshold, it is a periapical fracture.

7. The alveolar bone fracture detection system according to claim 1, characterized in that, In the volume calculation module, the centroid coordinates are ; where are the coordinates of all fracture voxels in the binary mask of the fracture region, is the total number of all fracture voxels in the fracture region; New voxel The condition for merging is: meet ; where is the gray value of the seed point, tolerance, neighborhood represents the neighborhood expansion area of the original fracture area; Region volume is as follows: ； Among them, is the cross-sectional area of the th voxel in the original fracture area; is the layer spacing of the oral image data; is the volume of the newly added voxels, and M is the number of newly added voxels.

8. An electronic device, characterized in that, Including a memory, a processor, and computer instructions stored on the memory and running on the processor. When the computer instructions are run by the processor, an alveolar bone fracture detection method is completed. The alveolar bone fracture detection method includes: Extracting multi-scale features from the acquired oral imaging data, thereby segmenting a fracture region and a root region; Extracting a centerline from the fracture region, accumulating the three-dimensional Euclidean distance voxel by voxel along the centerline to obtain the fracture line length, calculating the global average curvature after representing the centerline as a function of arc length, detecting the number of branch points in the fracture region, calculating the minimum distance between the fracture region and the root region, and thereby identifying the fracture type according to the fracture line length, the global average curvature, the number of branch points, and the minimum distance between the fracture region and the root region; Using the centroid of the fracture region as a seed point, determining the gray value of the seed point, merging the new voxels that belong to the neighborhood expansion region of the original fracture region and whose gray value difference from the seed point gray value satisfies a set threshold into the original fracture region to obtain an optimized fracture region, and calculating the regional volume of the optimized fracture region.

9. A computer-readable storage medium, characterized in that, For storing computer instructions, when the computer instructions are executed by a processor, an alveolar bone fracture detection method is completed. The alveolar bone fracture detection method includes: Extracting multi-scale features from the acquired oral image data, and thereby segmenting a fracture region and a tooth root region; Extracting a centerline from the fracture region, accumulating the three-dimensional Euclidean distance voxel by voxel along the centerline to obtain the fracture line length, calculating the global average curvature after representing the centerline as a function of arc length, detecting the number of branch points in the fracture region, calculating the minimum distance between the fracture region and the tooth root region, and thereby identifying the fracture type according to the fracture line length, the global average curvature, the number of branch points, and the minimum distance between the fracture region and the tooth root region; Taking the centroid of the fracture region as a seed point, and determining the gray value of the seed point, merging new voxels that belong to the neighborhood expansion region of the original fracture region and the difference between whose gray value and the gray value of the seed point satisfies a set threshold into the original fracture region to obtain an optimized fracture region, and calculating the regional volume of the optimized fracture region.

10. A computer program product, characterized in that, Including a computer program, when the computer program is executed by a processor, an alveolar bone fracture detection method is implemented. The alveolar bone fracture detection method includes: Extracting multi-scale features from the acquired oral image data, and thereby segmenting a fracture region and a tooth root region; Extracting a centerline from the fracture region, accumulating the three-dimensional Euclidean distance voxel by voxel along the centerline to obtain the fracture line length, calculating the global average curvature after representing the centerline as a function of arc length, detecting the number of branch points in the fracture region, calculating the minimum distance between the fracture region and the tooth root region, and thereby identifying the fracture type according to the fracture line length, the global average curvature, the number of branch points, and the minimum distance between the fracture region and the tooth root region; Taking the centroid of the fracture region as a seed point, and determining the gray value of the seed point, merging new voxels that belong to the neighborhood expansion region of the original fracture region and the difference between whose gray value and the gray value of the seed point satisfies a set threshold into the original fracture region to obtain an optimized fracture region, and calculating the regional volume of the optimized fracture region.

Images