Image segmentation method and device for the maxillary alveolar sinus artery

Abstract

This invention provides an image segmentation method and apparatus for the maxillary sinus artery, relating to the field of image processing technology. The method includes: extracting a binary mask of the maxillary sinus mucosa, a binary mask of the sinus airway, and a binary mask of the maxilla based on the original CBCT volume data; generating an anatomical constraint probability map based on the binary masks of the maxillary sinus mucosa, the sinus airway, and the maxilla, utilizing the symmetrical morphological expansion of the bone-sinus geometric relationship; stitching together the original CBCT volume data, the binary masks of the maxillary sinus mucosa, the sinus airway, the maxilla, and the anatomical constraint probability map to obtain a five-channel input tensor; inputting the five-channel input tensor into a pre-trained vessel segmentation prediction model to obtain a segmentation prediction result; and post-processing the segmentation prediction result to obtain a binary segmentation mask for the maxillary sinus artery, thereby achieving rapid and accurate automatic segmentation of the maxillary sinus vessels.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and specifically to an image segmentation method and apparatus for the maxillary alveolar sinus artery. Background Technology

[0002] The alveolar maxillary artery (AAA) is an anastomotic branch of the posterior superior alveolar artery and the infraorbital artery. It passes through the maxillary sinus and branches intraosseously or extraosseously. Identification of the alveolar maxillary artery is crucial for dental implantology and related procedures involving the maxillary sinus.

[0003] AAA injury is one of the most common complications during maxillary sinus floor elevation. Although vascular injury is usually not life-threatening, it can lead to significant bleeding that is difficult to control. This not only severely affects the operative field of vision and prolongs the operation time, but may also force the operation to be stopped due to difficulty in hemostasis.

[0004] Because maxillary sinus vessels (AAAs) have sub-millimeter diameters and complex, individual-specific morphologies, traditional preoperative imaging assessments can only identify them in 47%-67% of images. Cone-beam computed tomography (CBCT) can provide three-dimensional assessments of residual bone height, sinus anatomy, and vascular structure. However, in non-contrast CBCT, the Hounsfield Unit (HU) value of AAAs is approximately 40-60, almost completely overlapping with the HU value (approximately 40-80) of the surrounding sinus mucosa, making purely grayscale-based segmentation methods unreliable. To address this issue, some studies have employed artificial intelligence (AI) techniques for automated AAA segmentation, achieving some progress. However, current techniques primarily focus on segmenting relatively large structures such as sinuses and bone, and are not suitable for solving the problem of maxillary sinus vessel segmentation.

[0005] Existing technologies have the problem of difficulty in automatically segmenting the maxillary sinus vessels. Summary of the Invention

[0006] This invention provides an image segmentation method and apparatus for the maxillary sinus artery, which solves the problem of difficulty in automatically segmenting maxillary sinus vessels in the prior art, and realizes rapid and accurate automatic segmentation of maxillary sinus vessels.

[0007] This invention provides an image segmentation method for the alveolar maxillary sinus artery, comprising: The binary mask of the maxillary sinus mucosa, the binary mask of the sinus airway, and the binary mask of the maxilla were extracted from the original CBCT body data. Based on the maxillary sinus mucosa binary mask, the sinus airway binary mask, and the maxillary bone binary mask, an anatomical constraint probability map is generated using the symmetrical morphological expansion bone-sinus geometric relationship. The original CBCT volume data, the maxillary sinus mucosa binary mask, the sinus airway binary mask, the maxillary bone binary mask, and the anatomical constraint probability map are spliced ​​together to obtain a five-channel input tensor; The five-channel input tensor is input into a pre-trained blood vessel segmentation prediction model to obtain the segmentation prediction result. The blood vessel segmentation prediction model is based on a three-dimensional full-resolution nnU-Net backbone network and is trained using a pre-constructed sample dataset. The segmentation prediction results are post-processed to obtain a binary segmentation mask for the maxillary artery.

[0008] According to the present invention, an image segmentation method for the maxillary sinus artery is provided, which extracts a binary mask of the maxillary sinus mucosa, a binary mask of the sinus airway, and a binary mask of the maxilla based on the original CBCT volume data, including: The original CBCT volume data is input into a pre-trained first mask extraction model to obtain the maxillary sinus mucosa binary mask; The original CBCT volume data is input into a pre-trained second mask extraction model to obtain the binary mask of the sinus airway. The original CBCT volume data is input into a pre-trained third mask extraction model to obtain the maxillary binary mask; The first mask extraction model, the second mask extraction model, and the third mask extraction model are based on the nnU-Net framework and are trained using the maxillary sinus mucosa binary mask training dataset, the sinus airway binary mask training dataset, and the maxillary bone binary mask training dataset, respectively.

[0009] According to the present invention, an image segmentation method for the maxillary sinus artery is provided, which generates an anatomically constrained probability map based on the maxillary sinus mucosa binary mask, the sinus airway binary mask, and the maxillary bone binary mask, utilizing the symmetrical morphological expansion bone-sinus geometric relationship, including: The maxillary sinus mucosa binary mask and the sinus cavity airway binary mask are logically ORed to obtain a composite sinus cavity region. Using structuring elements of a preset size, morphological erosion extraction is performed on the binary mask of the maxilla to obtain the surface of the maxilla. Using structural elements of a preset size, morphological erosion extraction is performed on the composite sinus cavity region to obtain the sinus cavity surface; Calculate the Euclidean distance from each voxel in the original CBCT volume data to the surface of the maxilla to obtain the first distance field; The Euclidean distance from each voxel in the original CBCT volume data to the surface of the sinus cavity is calculated to obtain the second distance field; Using a preset Gaussian kernel function, the first distance field is mapped to a nearest neighbor score to obtain the first score; Using the preset Gaussian kernel function, the second distance field is mapped to a nearest neighbor score to obtain the second score; The joint interface probability is obtained based on the first score and the second score; By using the complement of the binary mask of the sinus cavity airway, the probability value inside the airway in the joint interface probability is set to zero to obtain the probability map. The region of interest is obtained by performing three iterations of dilation on the binary mask of the maxilla; A masking operation is performed based on the region of interest and the probability map to obtain a restricted probability map; The constraint probability map is normalized to obtain the anatomical constraint probability map.

[0010] According to the present invention, an image segmentation method for the alveolar maxillary sinus artery is provided, wherein the segmentation prediction result is post-processed to obtain a binary segmentation mask for the alveolar maxillary sinus artery, including: The segmentation prediction results are subjected to morphological smoothing to obtain the first processing result; Remove connected components in the first processing result whose volume is less than a preset volume threshold or whose elongation is less than a preset elongation threshold to obtain the first set of remaining connected components; Based on the inertial tensor eigenvalues, the morphological composite score of the connected components in the first set of remaining connected components is calculated. Remove the connected components in the first set of remaining connected components whose morphological composite scores are lower than a preset score threshold to obtain the second set of remaining connected components; Based on the morphological composite score, the connected regions in the second set of remaining connected regions are sorted, and the first preset number of connected regions are extracted as the binary segmentation mask for the maxillary artery.

[0011] According to the image segmentation method of the maxillary alveolar sinus artery provided by the present invention, based on the inertial tensor feature value, the morphological compound score of the connected regions in the first set of remaining connected regions is calculated, including: Using the first preset formula, calculate the morphological composite score of the connected components in the first set of remaining connected components; The first preset formula includes: s= · + · + · ; =min(V / 5000,1); =min(E / 20,1); =max(0, 1-|C-1.2| / 2); E= / ; C= / ; in, , , These are the normalized volume, normalized elongation, and normalized cylindricity scores, respectively, where E is the elongation and C is the cylindricity score. ≥ ≥ , These are the ordered eigenvalues ​​of the inertial tensor of the connected component; These are the weighting coefficients.

[0012] According to the present invention, an image segmentation method for the alveolar maxillary sinus artery is provided, the method further includes: calculating the three-dimensional position coordinates of the alveolar maxillary sinus artery, the vascular course path, and the spatial relationship between the vascular course path and the operation window area based on the binary segmentation mask of the alveolar maxillary sinus artery.

[0013] The present invention also provides an image segmentation device for the alveolar maxillary sinus artery, comprising: The mask extraction unit is used to extract the binary mask of the maxillary sinus mucosa, the binary mask of the sinus airway, and the binary mask of the maxilla based on the original CBCT volume data. The probability graph unit is used to generate an anatomically constrained probability graph based on the maxillary sinus mucosa binary mask, the sinus airway binary mask, and the maxillary bone binary mask, using the symmetrical morphological expansion bone-sin geometric relationship. Tensor construction unit is used to stitch together the original CBCT volume data, the maxillary sinus mucosa binary mask, the sinus cavity airway binary mask, the maxillary bone binary mask and the anatomical constraint probability map to obtain a five-channel input tensor; The segmentation prediction unit is used to input the five-channel input tensor into a pre-trained blood vessel segmentation prediction model to obtain the segmentation prediction result. The blood vessel segmentation prediction model is based on a three-dimensional full-resolution nnU-Net backbone network and is trained using a pre-constructed sample dataset. The post-processing unit is used to post-process the segmentation prediction results to obtain a binary segmentation mask for the maxillary artery.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the image segmentation method for the alveolar maxillary sinus artery as described above.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the image segmentation method for the alveolar maxillary sinus artery as described above.

[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the image segmentation method for the alveolar maxillary sinus artery as described above.

[0017] This invention provides an image segmentation method and apparatus for the alveolar maxillary sinus artery. It extracts binary masks of the maxillary sinus mucosa, sinus airway, and maxillary bone from raw CBCT volume data, and performs image segmentation of the alveolar maxillary sinus artery using non-enhanced CBCT data. Based on the acquired binary masks of the maxillary sinus mucosa, sinus airway, and maxillary bone, it utilizes the symmetrical morphological expansion of the bone-sinus geometry to narrow the network's search space from the entire CBCT volume data to a narrow region of the subperiosteal groove, generating an anatomical constraint probability map, effectively suppressing false positives in the sinus cavity and soft tissue transition zone. A multi-channel prior fusion strategy is employed to stitch together the raw CBCT volume data, the three anatomical structure mask data, and the anatomical constraint probability map to obtain a five-channel input tensor, which serves as the model's input data. The three anatomical structure masks are then combined with the constraint probability map. Figure 1 As an additional input channel, it provides explicit spatial prior guidance for the segmentation model without modifying the network architecture. The five-channel input tensor is input into the pre-trained blood vessel segmentation prediction model to obtain the segmentation prediction results. The blood vessel segmentation prediction based on spatial prior guidance can better identify the differences between the maxillary sinus anterior alveolar artery and the surrounding tissues. The segmentation prediction results are post-processed to remove false positive fragments and retain anatomically reasonable blood vessel candidates to obtain a binary segmentation mask for the maxillary sinus artery, thus achieving fast and accurate automatic segmentation of the maxillary sinus vessels. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1This is a flowchart illustrating the image segmentation method for the alveolar maxillary sinus artery provided by the present invention; Figure 2 This is a schematic diagram illustrating the construction process of the anatomical constraint probability map in the alveolar maxillary sinus artery image segmentation method provided by the present invention; Figure 3 This is a visualization comparing the image segmentation method of the alveolar maxillary sinus artery provided by this invention with the segmentation results of existing technologies; Figure 4 This is a schematic diagram of the image segmentation device for the alveolar maxillary sinus artery provided by the present invention; Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] The following is combined with Figures 1-3 The present invention describes an image segmentation method for the maxillary artery of the alveolar bone. Figure 1 This is a schematic flowchart illustrating the image segmentation method for the maxillary alveolar sinus artery provided by the present invention. Figure 1 As shown, the method includes: Step 110: Extract the binary mask of the maxillary sinus mucosa, the binary mask of the sinus airway, and the binary mask of the maxilla based on the original CBCT body data.

[0022] Raw CBCT volume data refers to the raw CBCT (cone-beam computed tomography) volume data V of the user of the maxillary sinus artery to be segmented, which includes at least the maxillary sinus region.

[0023] Based on the original CBCT body data, the entire maxillary skeletal structure, including the alveolar bone and sinus floor bone wall, is segmented to obtain a maxillary bone binary mask. The air cavity region inside the maxillary sinus is segmented to obtain a sinus airway binary mask. The soft tissue mucosa layer lining the inner side of the sinus bone wall is segmented to obtain a maxillary sinus mucosa binary mask.

[0024] It is important to emphasize that the three structural masks play complementary roles: the sinus airway binary mask provides stable geometric anchors, and its large-volume gas-filled cavities are easy to reliably segment, defining the inner boundary of the interface area; the maxillary sinus mucosa binary mask tracks the soft tissue lining adjacent to the blood vessels, acting as the inner fence of the periosteal band; and the maxillary bone binary mask captures the bony shell that physically accommodates the AAA (alveolar maxillary sinus artery), and its surface serves as a reference for periosteal distance measurement.

[0025] Segmentation can be achieved using a segmentation model, which can be an existing model or a model trained based on a publicly available model or network structure. This invention does not impose any restrictions on this.

[0026] Step 120: Based on the maxillary sinus mucosa binary mask, the sinus airway binary mask, and the maxillary bone binary mask, generate an anatomical constraint probability map using the symmetrical morphological expansion bone-sinus geometric relationship.

[0027] Based on the binary mask of the maxillary sinus mucosa, the binary mask of the sinus airway, and the binary mask of the maxilla, the search space of the network is reduced from the entire CBCT volume data to a narrow band region of the subperiosteal groove by utilizing the symmetrical morphological expansion of the bone-sinus geometric relationship, generating an anatomical constraint probability map that does not require training.

[0028] Understandably, anatomically constrained probability maps are used to encode the spatial likelihood of vascularity, showing a high probability value in the subperiosteal groove region at the bone-sinus interface.

[0029] In the specific implementation process, according to the example embodiment, based on the maxillary sinus mucosa binary mask, the sinus airway binary mask and the maxillary bone binary mask, an anatomical constraint probability map is generated through surface extraction, double Euclidean distance transformation, Gaussian nearest neighbor mapping, anatomical exclusion and periosteal reinforcement normalization operations.

[0030] Step 130: The original CBCT volume data, the maxillary sinus mucosa binary mask, the sinus airway binary mask, the maxillary bone binary mask, and the anatomical constraint probability map are spliced ​​together to obtain a five-channel input tensor.

[0031] A multi-channel input tensor was constructed. Specifically, the original CBCT volume data, the maxillary sinus mucosa binary mask, the sinus airway binary mask, the maxillary bone binary mask, and the anatomical constraint probability map were stitched together to form a five-channel input tensor.

[0032] This invention does not limit the dimensions of splicing. According to the example embodiment, the above data is spliced ​​along the channel dimension.

[0033] Step 140: Input the five-channel input tensor into the pre-trained blood vessel segmentation prediction model to obtain the segmentation prediction result. The blood vessel segmentation prediction model is trained using a pre-constructed sample dataset based on a three-dimensional full-resolution nnU-Net backbone network.

[0034] The five-channel input tensor is fed into the blood vessel segmentation prediction model for segmentation inference, and the model outputs the segmentation prediction results.

[0035] The blood vessel segmentation prediction model is built on a 3D full-resolution nnU-Net backbone network and trained using a pre-built sample dataset.

[0036] The sample dataset includes five-channel input tensor samples labeled with segmentation results.

[0037] It should be noted that, compared to standard single-channel CT input, the three-dimensional full-resolution nnU-Net backbone network of this embodiment replaces the number of input channels from 1 to 5. The network uses Dice loss. With cross-entropy loss Composite loss function Conduct training: .

[0038] The composite loss function of this invention is particularly effective for severely imbalanced vessel segmentation tasks. Other training settings can be configured as needed to ensure that the performance gain is entirely attributable to the multi-channel prior input representation.

[0039] According to the example implementation, the Adam optimizer (learning rate 1×10^(-4), cosine annealing scheduling) was used for training for 300 epochs, with a patch size of 128×128×128. The voxel spacing was resampled to 0.4 mm³ isotropic, and training was performed on five NVIDIA 3090 GPUs (total 120 GB of VRAM). A sliding window strategy was used during inference, with an overlap rate of 0.5.

[0040] Step 150: Post-process the segmentation prediction results to obtain a binary segmentation mask for the maxillary artery.

[0041] The segmentation prediction results are post-processed to remove false positive fragments, retain anatomically reasonable vascular candidates, and output the final binary segmentation mask for the alveolar maxillary sinus artery.

[0042] According to an example embodiment, post-processing includes morphological smoothing and multi-feature connected component filtering.

[0043] This invention provides an image segmentation method for the maxillary sinus artery, aiming to solve the technical challenge of automatically segmenting the anterior maxillary sinus artery in non-enhanced CBCT images due to its small diameter and high overlap of gray values ​​with surrounding tissues. The method extracts binary masks of the maxillary sinus mucosa, sinus airway, and maxilla from the original CBCT volume data, and uses non-enhanced CBCT data to achieve image segmentation of the maxillary sinus artery. Based on the obtained binary masks of the maxillary sinus mucosa, sinus airway, and maxilla, the search space of the network is reduced from the entire CBCT volume data to a narrow band region of the subperiosteal groove by utilizing the symmetrical morphological expansion of the bone-sinus geometry, generating an anatomical constraint probability map, effectively suppressing false positives in the sinus cavity and soft tissue transition zone. A multi-channel prior fusion strategy is employed to stitch together the original CBCT volume data, the three anatomical structure mask data, and the anatomical constraint probability map to obtain a five-channel input tensor, which serves as the input data for the model. The three anatomical structure masks and the constraint probability map are then integrated. Figure 1 As an additional input channel, it provides explicit spatial prior guidance for the segmentation model without modifying the network architecture. The five-channel input tensor is input into the pre-trained blood vessel segmentation prediction model to obtain the segmentation prediction results. The blood vessel segmentation prediction based on spatial prior guidance can better identify the differences between the maxillary sinus anterior alveolar artery and the surrounding tissues. The segmentation prediction results are post-processed to remove false positive fragments and retain anatomically reasonable blood vessel candidates to obtain a binary segmentation mask for the maxillary sinus artery, thus achieving fast and accurate automatic segmentation of the maxillary sinus vessels.

[0044] The following provides a further explanation of step 110. In some embodiments, the binary mask of the maxillary sinus mucosa, the binary mask of the sinus airway, and the binary mask of the maxilla are extracted based on the original CBCT volume data, including steps 111-113.

[0045] Step 111: Input the original CBCT volume data into the pre-trained first mask extraction model to obtain the maxillary sinus mucosa binary mask.

[0046] Step 112: Input the original CBCT volume data into the pre-trained second mask extraction model to obtain the binary mask of the sinus airway.

[0047] Step 113: Input the original CBCT volume data into the pre-trained third mask extraction model to obtain the maxillary binary mask.

[0048] The first mask extraction model, the second mask extraction model, and the third mask extraction model are based on the nnU-Net framework and are trained using the maxillary sinus mucosa binary mask training dataset, the sinus airway binary mask training dataset, and the maxillary bone binary mask training dataset, respectively.

[0049] Specifically, three independently pre-trained nnU-Net models—the first mask extraction model, the second mask extraction model, and the third mask extraction model—are used to infer the original CBCT volume data, resulting in three binary anatomical structure masks: a maxillary sinus mucosa binary mask. Binary masking of sinus airways and maxillary binary mask .

[0050] The first, second, and third mask extraction models are all based on the nnU-Net framework and are trained using the corresponding maxillary sinus mucosa binary mask training dataset, sinus airway binary mask training dataset, and maxillary bone binary mask training dataset. Training parameters can be selected according to actual needs. The maxillary sinus mucosa binary mask training dataset includes CBCT volume data samples labeled with maxillary sinus mucosa binary mask tags; the sinus airway binary mask training dataset includes CBCT volume data samples labeled with sinus airway binary mask tags; and the maxillary bone binary mask training dataset includes CBCT volume data samples labeled with maxillary bone binary mask tags.

[0051] According to the example embodiment, the DSC (Dice Similarity Coefficient) of the sinus airway pre-training model is 0.97, the DSC of the mucosa pre-training model is 0.72, and the DSC of the maxillary pre-training model is 0.90.

[0052] The following provides a further explanation of step 120. In some embodiments, based on the maxillary sinus mucosa binary mask, the sinus airway binary mask, and the maxillary bone binary mask, an anatomical constraint probability map is generated using the symmetrical morphological expansion bone-sinus geometric relationship, including steps 121-1212.

[0053] Step 121: Perform a logical OR operation between the maxillary sinus mucosa binary mask and the sinus cavity airway binary mask to obtain a composite sinus cavity region.

[0054] To extract the complex sinus cavity region, specifically, a logical OR operation is performed between the maxillary sinus mucosa binary mask and the sinus cavity airway binary mask to obtain the complex sinus cavity region S. S = ∨ .

[0055] Step 122: Using a pre-defined structuring element, perform morphological erosion extraction on the maxillary binary mask to obtain the maxillary surface.

[0056] Step 123: Using structural elements of a preset size, perform morphological erosion extraction on the composite sinus cavity region to obtain the sinus cavity surface.

[0057] Morphological erosion was performed on the maxillary binary mask and the complex sinus region using a pre-defined structuring element B, as follows: Figure 2 As shown, its surface is extracted. Specifically, = - ( B), S performs similar calculations.

[0058] According to the example embodiment, the preset size of the structuring element is a 3×3×3 structuring element.

[0059] Step 124: Calculate the Euclidean distance from each voxel in the original CBCT volume data to the surface of the maxilla to obtain the first distance field.

[0060] Step 125: Calculate the Euclidean distance from each voxel in the original CBCT volume data to the surface of the sinus cavity to obtain the second distance field.

[0061] Perform a two-distance transformation, such as Figure 2 As shown, the x-axis of each voxel in the original CBCT volume data to the maxillary bone surface is calculated. to the surface of the sinus cavity Euclidean distance of S (x) and (x). All voxels to the surface of the maxilla. Euclidean distance (x) is denoted as the first distance field, representing all voxels to the surface of the sinus cavity. Euclidean distance of S (x) is denoted as the second distance field.

[0062] Step 126: Using a preset Gaussian kernel function, map the first distance field to a nearest neighbor score to obtain the first score.

[0063] Step 127: Using the preset Gaussian kernel function, map the second distance field to a nearest neighbor score to obtain the second score.

[0064] The two distance fields are mapped to nearest neighbor scores using a Gaussian kernel function. The nearest neighbor score of the first distance field is denoted as the first score, and the nearest neighbor score of the second distance field is denoted as the second score.

[0065] According to the example embodiment, the spatial decay rate parameter σ of the preset Gaussian kernel function is set to 2.5 voxels.

[0066] Step 128: Obtain the joint interface probability based on the first score and the second score.

[0067] like Figure 2 As shown, multiplying the first score and the second score yields the joint interface probability: (x) = exp(- / (2σ²)) · exp(- / (2σ²)), Where σ is the space decay rate parameter, used to control the space decay rate.

[0068] Understandably, the joint interface probability only reaches a high value when the voxel is close to both surfaces simultaneously, thus achieving precise focusing on the bone-sinus interface.

[0069] Step 129: Using the complement of the binary mask of the sinus cavity airway, set the probability value inside the airway in the joint interface probability to zero to obtain the probability map.

[0070] Complementation using binary masking of sinus airways (1- The probability value of the Gaussian tail inside the airway is set to zero to exclude the cavity region. The result of setting it to zero is recorded as a probability map.

[0071] Step 1210: Perform three iterations of dilation on the maxillary binary mask to obtain the region of interest.

[0072] Step 1211: Perform masking operations based on the region of interest and the probability map to obtain a restricted probability map.

[0073] like Figure 2 As shown, the maxillary binary mask is expanded by three iterations ( ⊕ B³), the new mask obtained by dilation is used as the region of interest (ROI), and the region of interest is multiplied pixel by pixel with the original probability map (i.e., mask operation), which restricts the probability map to the periosteal neighborhood, resulting in a restricted probability map.

[0074] Step 1212: Normalize the constraint probability map to obtain the anatomical constraint probability map.

[0075] Normalization processing includes min-max normalization, which normalizes the constraint probability map to a preset interval, thus obtaining the anatomical constraint probability map. ,like Figure 2 As shown.

[0076] Understandably, this involves dissecting the probability diagram of constraints. The peak value is close to 1 in the subperiosteal groove region, and rapidly decays to 0 in other regions, providing spatially compact prior guidance for the network.

[0077] According to the example embodiment, the preset interval is the [0,1] interval.

[0078] Step 150 is further described below. In some embodiments, the segmentation prediction result is post-processed to obtain a binary segmentation mask for the maxillary artery, including steps 151-155.

[0079] Step 151: Perform morphological smoothing on the segmentation prediction results to obtain the first processing result.

[0080] The specific method of morphological smoothing can be selected as needed, and the result obtained is recorded as the first processing result.

[0081] According to the example embodiment, a binary closing operation and a binary opening operation are sequentially performed on the segmentation prediction result using a 6-connected structuring element. The closing operation fills the small holes inside the blood vessel, and the opening operation removes surface protrusions without changing the overall shape, thus obtaining the first processing result.

[0082] Step 152: Remove connected components in the first processing result whose volume is less than a preset volume threshold or whose elongation is less than a preset elongation threshold to obtain the first set of remaining connected components.

[0083] To perform multi-feature connected component filtering, firstly, the volume V or elongation E is filtered using a preset volume threshold V_min and a preset elongation rate E_min to remove connected components that belong to noise spots or spherical artifacts. The remaining connected components are denoted as the first set of remaining connected components.

[0084] According to the example embodiment, the preset volume threshold is set to 300 voxels, and the preset elongation threshold is set to 1.0.

[0085] Step 153: Based on the inertial tensor eigenvalues, calculate the morphological composite score of the connected components in the first set of remaining connected components.

[0086] For the first set of remaining connected components, morphological composite scores are calculated based on the inertial tensor.

[0087] According to the example embodiment, the composite score is a weighted sum of normalized volume, normalized elongation, and normalized cylindricity.

[0088] Step 154: Remove connected components in the first set of remaining connected components whose morphological composite scores are lower than a preset score threshold to obtain a second set of remaining connected components.

[0089] The connected components in the first set of remaining connected components whose morphological composite scores are lower than a preset score threshold are removed to obtain the second set of remaining connected components.

[0090] According to the example embodiment, the preset scoring threshold is 0.35.

[0091] Step 155: Based on the morphological composite score, sort the connected regions in the second set of remaining connected regions, and extract the first preset number of connected regions as the binary segmentation mask for the maxillary artery.

[0092] The connected components in the second set of remaining connected components are sorted by morphological composite score, and the top K (preset number) connected components with the highest scores are retained as binary segmentation masks for the maxillary artery.

[0093] According to the example embodiment, K=2, corresponding to AAA on the left and right sides respectively.

[0094] The following provides a further explanation of step 153. In some embodiments, based on the inertial tensor eigenvalues, the morphological composite score of the connected components in the first set of remaining connected components is calculated, including: Using the first preset formula, calculate the morphological composite score of the connected components in the first set of remaining connected components; The first preset formula includes: s= · + · + · ; =min(V / 5000, 1); =min(E / 20, 1); =max(0, 1-|C-1.2| / 2); E= / ; C= / ; in, , , These are the normalized volume, normalized elongation, and normalized cylindricity scores, respectively, where E is the elongation and C is the cylindricity score. ≥ ≥ , These are the ordered eigenvalues ​​of the inertial tensor of the connected component; These are the weighting coefficients.

[0095] Specifically, elongation and cylindricity Ordered eigenvalues ​​of the inertial tensor Derivation.

[0096] According to the example embodiment, the calculation process of the inertia tensor includes: calculating the arithmetic mean of the coordinates of all voxels in the current connected domain to obtain the centroid coordinates; based on the centroid, calculating the second moment of the coordinate axes of all voxels in the current connected domain to obtain the inertia tensor matrix; and performing eigenvalue decomposition on the inertia tensor matrix to obtain three eigenvalues. .

[0097] For tubular structures like AAA Therefore, the elongation E is high and the cylindricity C≈1.

[0098] The normalized score is calculated as follows: =min(V / 5000,1), =min(E / 20,1), =max(0,1-|C-1.2| / 2).

[0099] In some embodiments, the method further includes: calculating the three-dimensional position coordinates of the alveolar maxillary sinus artery, the vascular path, and the spatial relationship between the vascular path and the operation window area based on the binary segmentation mask of the alveolar maxillary sinus artery.

[0100] Specifically, the three-dimensional location coordinates of the maxillary sinus artery, its vascular path, and its spatial relationship with the fenestration area are calculated. The results of the calculation can provide visual preoperative planning support for maxillary sinus floor elevation surgery.

[0101] To further illustrate the image segmentation method for the maxillary sinus artery provided by the present invention, a specific embodiment is given.

[0102] This invention employed five-fold cross-validation on CBCT data from 38 real patients. Vascular annotation was independently performed by three experienced oral and maxillofacial surgeons using 3D Slicer software, with consensus labels determined for each case through voxel-level majority voting and joint review.

[0103] like Figure 3 As shown, the comparison results between the method provided by the present invention and the existing methods are as follows: The DSC of the present invention (AAA-Seg) is 71.6±7.5%, clDice is 89.3±5.5%, HD95 is 10.2±17.9mm, and ASD is 1.3±2.1mm. It is significantly better than single-channel nnU-Net (DSC=64.8%), TransUNet-3D (DSC=0.3%) and SwinUNETR (DSC=12.7%) in all indicators.

[0104] Table 1 Comparison of Ablation Test Performance

[0105] As shown in Table 1, ablation experiments validated the contributions of each channel: the baseline DSC of a single channel was 64.8%, the DSC of the four-channel model with three structural masks increased to 65.2%, and the DSC of the five-channel model with the addition of an anatomical constraint probability map reached 71.6%. The addition of the anatomical constraint probability map brought the most significant performance improvement (+6.8pp DSC, +7.9pp clDice, and a 19.7mm reduction in HD95), validating the effectiveness of focusing probabilistic mass on the subperiosteal groove rather than the entire interface zone.

[0106] The method of this invention achieved a Dice similarity coefficient (DSC) of 71.6%, a centerline Dice (clDice) of 89.3%, and a 95th percentile Hausdorff distance (HD95) of 10.2 mm in five-fold cross-validation of CBCT data from 38 real patients, which is significantly better than existing methods such as single-channel nnU-Net, TransUNet-3D, and SwinUNETR.

[0107] The image segmentation device for the alveolar maxillary sinus artery provided by the present invention is described below. The image segmentation device for the alveolar maxillary sinus artery described below can be referred to in correspondence with the image segmentation method for the alveolar maxillary sinus artery described above. Figure 4 This is a schematic diagram of the image segmentation device for the alveolar maxillary sinus artery provided by the present invention, as shown below. Figure 4 As shown, the device includes: The mask extraction unit 410 is used to extract the maxillary sinus mucosa binary mask, the sinus airway binary mask, and the maxillary bone binary mask based on the original CBCT body data. The probability graph unit 420 is used to generate an anatomically constrained probability graph based on the maxillary sinus mucosa binary mask, the sinus airway binary mask, and the maxillary bone binary mask, using the symmetrical morphological expansion bone-sin geometric relationship. Tensor construction unit 430 is used to stitch together the original CBCT volume data, the maxillary sinus mucosa binary mask, the sinus cavity airway binary mask, the maxillary bone binary mask and the anatomical constraint probability map to obtain a five-channel input tensor; The segmentation prediction unit 440 is used to input the five-channel input tensor into a pre-trained blood vessel segmentation prediction model to obtain a segmentation prediction result. The blood vessel segmentation prediction model is based on a three-dimensional full-resolution nnU-Net backbone network and is trained using a pre-constructed sample dataset. The post-processing unit 450 is used to post-process the segmentation prediction results to obtain a binary segmentation mask for the maxillary artery.

[0108] The method performs similar functions to the system provided earlier. Other functions can be found in the previous descriptions and will not be repeated here.

[0109] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 can call logic instructions in the memory 530 to execute an image segmentation method for the maxillary sinus artery. This method includes: extracting a binary mask of the maxillary sinus mucosa, a binary mask of the sinus airway, and a binary mask of the maxilla based on the original CBCT volume data; generating an anatomical constraint probability map based on the binary masks of the maxillary sinus mucosa, the sinus airway, and the maxilla, utilizing symmetrical morphological expansion of the bone-sinus geometric relationship; concatenating the original CBCT volume data, the binary masks of the maxillary sinus mucosa, the sinus airway, the maxilla, and the anatomical constraint probability map to obtain a five-channel input tensor; inputting the five-channel input tensor into a pre-trained vessel segmentation prediction model to obtain a segmentation prediction result, wherein the vessel segmentation prediction model is based on a three-dimensional full-resolution nnU-Net backbone network and trained using a pre-constructed sample dataset; and post-processing the segmentation prediction result to obtain a binary segmentation mask for the maxillary sinus artery.

[0110] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0111] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the alveolar maxillary sinus artery image segmentation method provided by the above methods. The method includes: extracting a maxillary sinus mucosa binary mask, a sinus airway binary mask, and a maxillary bone binary mask based on the original CBCT volume data; and using the maxillary sinus mucosa binary mask, the sinus airway binary mask, and the maxillary bone binary mask to expand the bone-sinus geometry using symmetrical morphology. The relationships are analyzed to generate an anatomical constraint probability map. The original CBCT volume data, the maxillary sinus mucosa binary mask, the sinus airway binary mask, the maxillary bone binary mask, and the anatomical constraint probability map are concatenated to obtain a five-channel input tensor. The five-channel input tensor is then input into a pre-trained blood vessel segmentation prediction model to obtain segmentation prediction results. The blood vessel segmentation prediction model is based on a three-dimensional full-resolution nnU-Net backbone network and trained using a pre-constructed sample dataset. The segmentation prediction results are then post-processed to obtain a binary segmentation mask for the alveolar maxillary sinus artery.

[0112] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements an image segmentation method for the alveolar maxillary sinus artery provided by the methods described above. This method includes: extracting a binary mask of the maxillary sinus mucosa, a binary mask of the sinus airway, and a binary mask of the maxilla based on original CBCT volume data; generating an anatomical constraint probability map based on the binary mask of the maxillary sinus mucosa, the binary mask of the sinus airway, and the binary mask of the maxilla, utilizing the symmetrical morphological expansion of the bone-sinus geometric relationship; and... The original CBCT volume data, the maxillary sinus mucosa binary mask, the sinus airway binary mask, the maxillary bone binary mask, and the anatomical constraint probability map are concatenated to obtain a five-channel input tensor. The five-channel input tensor is then input into a pre-trained vessel segmentation prediction model to obtain segmentation prediction results. The vessel segmentation prediction model is based on a three-dimensional full-resolution nnU-Net backbone network and trained using a pre-constructed sample dataset. The segmentation prediction results are then post-processed to obtain a binary segmentation mask for the alveolar maxillary sinus artery.

[0113] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0114] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for image segmentation of the maxillary alveolar sinus artery, characterized in that, include: The binary mask of the maxillary sinus mucosa, the binary mask of the sinus airway, and the binary mask of the maxilla were extracted from the original CBCT body data. Based on the maxillary sinus mucosa binary mask, the sinus airway binary mask, and the maxillary bone binary mask, an anatomical constraint probability map is generated using the symmetrical morphological expansion bone-sinus geometric relationship. The original CBCT volume data, the maxillary sinus mucosa binary mask, the sinus airway binary mask, the maxillary bone binary mask, and the anatomical constraint probability map are spliced ​​together to obtain a five-channel input tensor; The five-channel input tensor is input into a pre-trained blood vessel segmentation prediction model to obtain the segmentation prediction result. The blood vessel segmentation prediction model is based on a three-dimensional full-resolution nnU-Net backbone network and is trained using a pre-constructed sample dataset. The segmentation prediction results are post-processed to obtain a binary segmentation mask for the alveolar maxillary sinus artery; Specifically, based on the maxillary sinus mucosa binary mask, the sinus airway binary mask, and the maxillary bone binary mask, an anatomical constraint probability map is generated using the symmetrical morphological expansion bone-sinus geometric relationship, including: The maxillary sinus mucosa binary mask and the sinus cavity airway binary mask are logically ORed to obtain a composite sinus cavity region. Using structuring elements of a preset size, morphological erosion extraction is performed on the binary mask of the maxilla to obtain the surface of the maxilla. Using structural elements of a preset size, morphological erosion extraction is performed on the composite sinus cavity region to obtain the sinus cavity surface; Calculate the Euclidean distance from each voxel in the original CBCT volume data to the surface of the maxilla to obtain the first distance field; The Euclidean distance from each voxel in the original CBCT volume data to the surface of the sinus cavity is calculated to obtain the second distance field; Using a preset Gaussian kernel function, the first distance field is mapped to a nearest neighbor score to obtain the first score; Using the preset Gaussian kernel function, the second distance field is mapped to a nearest neighbor score to obtain the second score; The joint interface probability is obtained based on the first score and the second score; By using the complement of the binary mask of the sinus cavity airway, the probability value inside the airway in the joint interface probability is set to zero to obtain the probability map. The region of interest is obtained by performing three iterations of dilation on the binary mask of the maxilla; A masking operation is performed based on the region of interest and the probability map to obtain a restricted probability map; The constraint probability map is normalized to obtain the anatomical constraint probability map.

2. The method according to claim 1, characterized in that, Based on the original CBCT volume data, binary masks of the maxillary sinus mucosa, sinus airway, and maxillary bone were extracted, including: The original CBCT volume data is input into a pre-trained first mask extraction model to obtain the maxillary sinus mucosa binary mask; The original CBCT volume data is input into a pre-trained second mask extraction model to obtain the binary mask of the sinus airway. The original CBCT volume data is input into a pre-trained third mask extraction model to obtain the maxillary binary mask; The first mask extraction model, the second mask extraction model, and the third mask extraction model are based on the nnU-Net framework and are trained using the maxillary sinus mucosa binary mask training dataset, the sinus airway binary mask training dataset, and the maxillary bone binary mask training dataset, respectively.

3. The method according to claim 1, characterized in that, The segmentation prediction results are post-processed to obtain a binary segmentation mask for the alveolar maxillary sinus artery, including: The segmentation prediction results are subjected to morphological smoothing to obtain the first processing result; Remove connected components in the first processing result whose volume is less than a preset volume threshold or whose elongation is less than a preset elongation threshold to obtain the first set of remaining connected components; Based on the inertial tensor eigenvalues, the morphological composite score of the connected components in the first set of remaining connected components is calculated. Remove the connected components in the first set of remaining connected components whose morphological composite scores are lower than a preset score threshold to obtain the second set of remaining connected components; Based on the morphological composite score, the connected regions in the second set of remaining connected regions are sorted, and the first preset number of connected regions are extracted as the binary segmentation mask for the maxillary artery.

4. The method according to claim 3, characterized in that, Based on the inertial tensor eigenvalues, the morphological composite score of the connected components in the first set of remaining connected components is calculated, including: Using the first preset formula, calculate the morphological composite score of the connected components in the first set of remaining connected components; The first preset formula includes: s = · + · + · ； = min(V / 5000, 1); = min(E / 20, 1); = max(0, 1 - |C-1.2| / 2); E = / ； C = / ； in, , , These are normalized volume, normalized elongation, and normalized cylindricity scores, respectively, where V represents volume, E represents elongation, and C represents cylindricity. ≥ ≥ , These are the ordered eigenvalues ​​of the inertial tensor of the connected component; These are the weighting coefficients.

5. The method according to claim 1, characterized in that, The method further includes: calculating the three-dimensional position coordinates of the alveolar maxillary sinus artery, the vascular course, and the spatial relationship between the vascular course and the operation window area based on the binary segmentation mask of the alveolar maxillary sinus artery.

6. An image segmentation device for the maxillary alveolar sinus artery, characterized in that, include: The mask extraction unit is used to extract the binary mask of the maxillary sinus mucosa, the binary mask of the sinus airway, and the binary mask of the maxilla based on the original CBCT volume data. The probability graph unit is used to generate an anatomically constrained probability graph based on the maxillary sinus mucosa binary mask, the sinus airway binary mask, and the maxillary bone binary mask, utilizing the symmetrical morphological expansion bone-sinus geometric relationship; specifically including: The maxillary sinus mucosa binary mask and the sinus cavity airway binary mask are logically ORed to obtain a composite sinus cavity region. Using structuring elements of a preset size, morphological erosion extraction is performed on the binary mask of the maxilla to obtain the surface of the maxilla. Using structural elements of a preset size, morphological erosion extraction is performed on the composite sinus cavity region to obtain the sinus cavity surface; Calculate the Euclidean distance from each voxel in the original CBCT volume data to the surface of the maxilla to obtain the first distance field; The Euclidean distance from each voxel in the original CBCT volume data to the surface of the sinus cavity is calculated to obtain the second distance field; Using a preset Gaussian kernel function, the first distance field is mapped to a nearest neighbor score to obtain the first score; Using the preset Gaussian kernel function, the second distance field is mapped to a nearest neighbor score to obtain the second score; The joint interface probability is obtained based on the first score and the second score; By using the complement of the binary mask of the sinus cavity airway, the probability value inside the airway in the joint interface probability is set to zero to obtain the probability map. The region of interest is obtained by performing three iterations of dilation on the binary mask of the maxilla; A masking operation is performed based on the region of interest and the probability map to obtain a restricted probability map; The constraint probability map is normalized to obtain the anatomical constraint probability map; Tensor construction unit is used to stitch together the original CBCT volume data, the maxillary sinus mucosa binary mask, the sinus cavity airway binary mask, the maxillary bone binary mask and the anatomical constraint probability map to obtain a five-channel input tensor; The segmentation prediction unit is used to input the five-channel input tensor into a pre-trained blood vessel segmentation prediction model to obtain the segmentation prediction result. The blood vessel segmentation prediction model is based on a three-dimensional full-resolution nnU-Net backbone network and is trained using a pre-constructed sample dataset. The post-processing unit is used to post-process the segmentation prediction results to obtain a binary segmentation mask for the maxillary artery.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the image segmentation method for the maxillary artery as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the image segmentation method for the maxillary artery as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the image segmentation method for the maxillary artery as described in any one of claims 1 to 5.

Images