Automatic neural tube labeling method and system based on dental arch curve cross section

Abstract

The invention belongs to the technical field of medical image processing, and particularly relates to an automatic neural tube labeling method and system based on a dental arch curve cross section. The labeling method comprises the following steps: acquiring a position of a lower alveolar nerve chin hole projection in a corresponding fitting dental arch curve parameter coordinate system; starting from the position of the inferior alveolar nerve chin hole, acquiring cross section images at each sampling point on the fitting dental arch curve; performing region marking on the cross section image based on the HU value to generate a binary image; after connected domains in the binarized image are extracted, the connected domains containing the target contour are screened out to serve as candidate connected domains; performing spatial collocation on the candidate connected domains; and performing three-dimensional reconstruction of the neural tube according to a spatial arrangement result to complete marking of the neural tube. According to the method, the neural tube can be efficiently and accurately labeled, and the labeling result has relatively high stability.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing technology, and in particular relates to an automatic neural tube annotation method and system based on the cross-section of the dental arch curve. Background Technology

[0002] The inferior alveolar nerve within the jawbone passes through a neural canal; the neural canal is a tubular passage, its cross-section mostly circular or nearly closed, and its interior is usually soft tissue or fat; these neural canals are directly related to the proper functioning of organs such as the lips and tongue. Therefore, it is necessary to mark the neural canals in the jawbone before wisdom tooth extraction or orthognathic surgery.

[0003] Different individuals have different jawbone morphology, bone density, and nerve branch morphology. In the current technology, several key points (usually 4 to 8 key points) of the neural tube are selected point by point on the 3D image of CBCT (cone beam computed tomography). The neural tube trajectory is generated based on the key points and then manually corrected to complete the annotation of the neural tube. The annotation cost is high.

[0004] Furthermore, this method typically takes 3–10 minutes to complete the annotation of a single neural tube; when encountering anatomical variations or image artifacts, it can even take 15–20 minutes. Moreover, because it is manual annotation, it relies on the experience of the annotators, so there is a high possibility that the same neural tube trajectory annotated by different annotators may differ by 1–2 mm; this not only affects the accuracy of the annotation results but also the stability of the accuracy.

[0005] Therefore, how to efficiently and accurately annotate neural tubes and ensure the stability of the annotation results has become a pressing problem in this field. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide an automatic neural tube annotation method based on the cross-section of the dental arch curve, which can efficiently and accurately annotate the neural tube and ensure that the annotation results have high stability.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] An automatic neural tube annotation method based on the cross-section of the dental arch curve includes the following steps:

[0009] Step 1: Obtain the position of the inferior alveolar nerve mental foramen projection in the corresponding fitted dental arch curve parameter coordinate system;

[0010] Step 2: Starting from the position of the mental foramen of the inferior alveolar nerve, obtain cross-sectional images at each sampling point on the fitted dental arch curve;

[0011] Step 3: After marking the regions of the cross-sectional image based on the HU value, a binarized image is generated;

[0012] Step 4: After extracting the connected components in the binarized image, select the connected components containing the target contour as candidate connected components.

[0013] Step 5: Spatial alignment of candidate connected domains. The spatial alignment result includes the central line of the inferior alveolar nerve canal formed by the centroids of all successfully matched candidate connected domains. Then, three-dimensional reconstruction of the nerve canal is performed based on the spatial alignment result to complete the annotation of the nerve canal.

[0014] Preferably, after step 4 and before step 5, step 4' is included: Step 4' matches all adjacent candidate connected components in each sampling point: if the spatial distance and geometric similarity between adjacent candidate connected components both meet the set threshold, then the two currently adjacent candidate connected components are successfully matched; otherwise, they are both unmatched.

[0015] Preferably, step 2 also includes the following:

[0016] Step 21: Using the mental foramen of the inferior alveolar nerve as the starting sampling point, determine several sampling points on the fitted dental arch curve;

[0017] Step 22: Starting from the current sampling point, along the orthogonal direction of the current sampling point on the fitted dental arch curve, sequentially acquire cross-sectional images of the current sampling point in the 3D CT image with the number of layers continuously increasing, with a preset sampling step size; until the cross-sectional images corresponding to all sampling points on the current fitted dental arch curve are acquired.

[0018] Preferably, step 3 also includes the following:

[0019] Step 31: Mark the bone region and cavity region in the cross-sectional image based on the HU value;

[0020] Step 32: Use the bone region as the foreground pixel and the cavity region as the background pixel to generate a binarized image corresponding to each cross-sectional image.

[0021] Preferably, step 4 also includes the following:

[0022] S41, Denoise the binarized image;

[0023] S42, based on foreground and background pixels, extracts connected components from the denoised binarized image and calculates the geometric properties of each connected component; the geometric properties include area A, perimeter P, and roundness index C, where...

[0024] S43, select connected components that satisfy the geometric properties of the target contour as candidate connected components, and record the centroid coordinates and boundary point set corresponding to the candidate connected components.

[0025] Preferably, step 4' further includes the following: the Euclidean distance between the centroids of adjacent candidate connected components and the similarity of the boundary point sets are respectively the spatial distance and geometric similarity between adjacent candidate connected components; if the spatial distance between adjacent candidate connected components is less than the first distance threshold τ d And the geometric similarity is greater than the second similarity threshold τ s If the two adjacent candidate connected components match successfully, then the two adjacent candidate connected components match successfully; otherwise, the two adjacent candidate connected components match unsuccessfully.

[0026] Preferably, step 5 also includes the following:

[0027] Step 51: Connect the centroids of all successfully matched candidate connected components sequentially according to the spatial position of the corresponding sampling point on the fitted dental arch curve and the spatial position of the corresponding cross-sectional image layer, to form the center line of the current inferior alveolar nerve canal in three-dimensional space; at the same time, mark all candidate connected components in three-dimensional space.

[0028] Step 52: Based on the centerline of the inferior alveolar nerve canal in three-dimensional space and all candidate connected regions in three-dimensional space, perform three-dimensional reconstruction of the nerve canal and complete the annotation of the nerve canal.

[0029] This invention also provides an automatic neural tube annotation system based on the cross-section of a dental arch curve, comprising: a projection module, a cross-sectional image acquisition module, a binarized image acquisition module, a filtering module, and an annotation module; the projection module is used to acquire the position of the projection of the mental foramen of the inferior alveolar nerve in the coordinate system of the corresponding fitted dental arch curve parameters, and then send it to the cross-sectional image acquisition module; the cross-sectional image acquisition module is used to acquire cross-sectional images at each sampling point on the fitted dental arch curve starting from the position of the mental foramen of the inferior alveolar nerve, and then send them to the binarized image acquisition module; the binarized image acquisition module is used to generate a binarized image based on the cross-sectional image and then send it to the filtering module; the filtering module is used to filter out candidate connected components from the binarized image and send them to the annotation module; the annotation module is used to perform three-dimensional reconstruction of the neural tube after spatially aligning the candidate connected components; each module is programmed or configured to execute the steps of the automatic neural tube annotation method based on the cross-section of a dental arch curve as described above.

[0030] The present invention also provides a computer-readable storage medium storing a computer program programmed or configured to perform an automatic neural tube annotation method based on the cross-section of a dental arch curve as described above.

[0031] The present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described automatic neural tube annotation method based on the cross-section of a dental arch curve.

[0032] The beneficial effects of this invention are as follows:

[0033] (1) The annotation method of the present invention can be directly applied to CT data. Based on the cross-section of the dental arch curve, the center line of the inferior alveolar nerve canal in three-dimensional space is obtained. Combined with the candidate connected regions of the cross-sectional images of each layer, the nerve canal in three-dimensional space can be reconstructed, and the nerve canal can be labeled.

[0034] (2) Compared with manual annotation, the annotation method of the present invention is highly efficient and saves a lot of labor costs while ensuring extremely high and stable accuracy.

[0035] (3) Compared with deep learning models that require large-scale labeled data for training, the labeling method of the present invention can be directly applied to CT data without labeled data, saving the time of training deep learning models and the manpower cost of using large-scale labeled data. It also avoids the problem that deep learning models are greatly affected by the differences in CT equipment, scanning parameters and jaw structure when labeling. The labeling results of the present invention are more robust. Attached Figure Description

[0036] Figure 1 This is a flowchart of an automatic neural tube annotation method based on the cross-section of the dental arch curve according to the present invention. Detailed Implementation

[0037] To make the technical solution of the present invention clearer and more explicit, the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Solutions derived by those skilled in the art through equivalent substitution and conventional reasoning of the technical features of the present invention without creative effort all fall within the protection scope of the present invention.

[0038] like Figure 1 The diagram shows a flowchart of an automatic neural tube annotation method based on the cross-section of the dental arch curve according to the present invention, which includes the following steps:

[0039] S1, obtain the position of the mental foramen projection of the inferior alveolar nerve in the corresponding fitted dental arch curve parameter coordinate system.

[0040] S2, starting from the mental foramen of the inferior alveolar nerve, acquire cross-sectional images at each sampling point on the fitted dental arch curve.

[0041] S3, after region labeling of the cross-sectional image based on the HU value, generates a binarized image.

[0042] S4. After extracting the connected components from the binarized image, select the connected components that contain the target contour as candidate connected components.

[0043] S5, match all adjacent candidate connected components in each sampling point: if the spatial distance and geometric similarity between adjacent candidate connected components both meet the set threshold, then the two adjacent candidate connected components are matched successfully; otherwise, they are matched unsuccessfully.

[0044] Except for the first and last candidate connected components of a certain sampling point, all candidate connected components of that sampling point must be matched with the two adjacent candidate connected components before and after it.

[0045] S6. Spatial alignment is performed on the candidate connected domains. The spatial alignment result includes the central line of the inferior alveolar nerve canal formed by the centroids of all successfully matched candidate connected domains. Then, the three-dimensional reconstruction of the nerve canal is performed based on the spatial alignment result to complete the annotation of the nerve canal.

[0046] S1 also includes the following: After acquiring a three-dimensional CT image, the technician projects the mental foramen of the inferior alveolar nerve onto the corresponding pre-fitted dental arch curve based on the position of the mental foramen of the inferior alveolar nerve in the three-dimensional coordinate system, thereby obtaining the position of the mental foramen of the inferior alveolar nerve in the dental arch curve parameter coordinate system.

[0047] The corresponding pre-fitted dental arch curve refers to the dental arch curve and the mental foramen of the inferior alveolar nerve belonging to the same person.

[0048] Technicians fit dental arch curves based on 3D CT images, which is an existing technique (such as principal component analysis), and will not be elaborated here.

[0049] S2 also includes:

[0050] S21, using the mental foramen of the inferior alveolar nerve as the starting sampling point, determine several sampling points on the fitted dental arch curve;

[0051] S22, starting from the current sampling point, along the orthogonal direction of the current sampling point on the fitted dental arch curve, sequentially acquire cross-sectional images of the current sampling point in the 3D CT image with the number of layers continuously increasing, with a preset sampling step size; until the cross-sectional images corresponding to all sampling points on the current fitted dental arch curve are acquired.

[0052] The sampling points on the fitted dental arch curve are set by technicians; starting from the initial sampling point, a sampling point is determined at regular intervals along the fitted dental arch curve until the end of the dental arch curve.

[0053] In S22 of this invention, the preset sampling step size is set by a technician based on the slice thickness of the three-dimensional CT image. In this embodiment, the preset sampling step size ranges from 0.1 mm to 0.3 mm.

[0054] Each sampling point has several corresponding cross-sectional images.

[0055] S3 also includes:

[0056] S31, based on HU values, marks the bone region and cavity region in the cross-sectional image;

[0057] S32 uses the bone region as the foreground pixel and the cavity region as the background pixel to generate a binarized image corresponding to each cross-sectional image.

[0058] S31 also includes the following:

[0059] The region with HU values ​​within the first threshold range is marked as the bone region, and the region with HU values ​​within the second threshold range is marked as the cavity region.

[0060] In this embodiment, the first threshold range is 250 HU to 400 HU, and the second threshold range is 0 to 150 HU. The lower limit of the first threshold range corresponding to the bone region is significantly higher than the upper limit of the second threshold range corresponding to the cavity region, which also includes soft tissue.

[0061] S4 also includes the following sub-steps:

[0062] S41, Denoising the binarized image: Perform morphological closing operation on the binarized image to remove isolated noise points;

[0063] S42, based on foreground and background pixels, extracts connected components from the denoised binarized image and calculates the geometric properties of each connected component; the geometric properties include area A, perimeter P, and roundness index C, where...

[0064] S43, select connected components that satisfy the geometric properties of the target contour as candidate connected components, and record the centroid coordinates and boundary point set corresponding to the candidate connected components.

[0065] In S42, extracting connected components in a binarized image based on foreground and background pixels is a prior art technique; for example, a queue-based BFS seed filling algorithm can be used to extract connected components in a binarized image based on foreground and background pixels.

[0066] In this embodiment, the roundness of the target contour ranges from [0.7, 1.0]; the area ranges from [5, 200], unit: mm. 2The perimeter ranges from [10, 80], in mm. When C ∈ [0.7, 1.0], A ∈ [5, 200], and P ∈ [10, 80], the geometric properties of the connected component satisfy the geometric properties of the target contour, i.e., the connected component is a candidate connected component.

[0067] S5 also includes the following:

[0068] The Euclidean distance between the centroids of adjacent candidate connected components and the similarity of their boundary point sets are respectively the spatial distance and geometric similarity of the adjacent candidate connected components; if the spatial distance between adjacent candidate connected components is less than the first distance threshold τ... d And the geometric similarity is greater than the second similarity threshold τ s If the two adjacent candidate connected components match successfully, then the two adjacent candidate connected components match successfully; otherwise, the two adjacent candidate connected components match unsuccessfully.

[0069] The similarity of the boundary point set can be calculated using methods such as region overlap rate or contour Hausdorff distance, which are existing technologies and will not be elaborated here.

[0070] In this embodiment, the first distance threshold τ d =2mm, τ s =0.7.

[0071] For example, for a sampling point corresponding to a cross-sectional image with 7 layers, candidate connected components appear in layers 3, 4, 5, and 6 (the layer number of the binarized image is the same as the layer number of the corresponding cross-sectional image). Therefore, the first candidate connected component for this sampling point appears in layer 3, the second in layer 4, ..., and the fourth in layer 6. During the matching process, regardless of whether the match between the first and second candidate connected components at this sampling point is successful or not, the matching between the second and third candidate connected components must continue. It is possible that the first and second candidate connected components match successfully, but the second and third candidate connected components fail to match; therefore, each successful match preserves at least two candidate connected components for subsequent neural tube annotation.

[0072] S6 also includes the following:

[0073] S61, connect the centroids of all successfully matched candidate connected components sequentially according to the spatial position of the corresponding sampling point on the fitted dental arch curve and the spatial position of the corresponding cross-sectional image layer, and use them as the center line of the current inferior alveolar nerve canal in three-dimensional space; at the same time, mark all candidate connected components in three-dimensional space.

[0074] S62, based on the centerline of the inferior alveolar nerve canal in three-dimensional space and all candidate connected regions in three-dimensional space, perform three-dimensional reconstruction of the nerve canal and complete the annotation of the nerve canal.

[0075] In S61, the centerline of the inferior alveolar nerve canal in three-dimensional space is a continuous three-dimensional spatial path formed by connecting the centroids of successfully matched candidate connected regions. Each centroid has its own corresponding sampling point and cross-sectional image layer. Since it is a three-dimensional spatial path, the centroids corresponding to different sampling points are connected sequentially according to the spatial position of the sampling points on the fitted dental arch curve. However, several centroids corresponding to the same sampling point are connected sequentially according to the spatial position of the corresponding cross-sectional image layer.

[0076] Candidate connected components that fail to match are not involved in the operation of constructing the centerline of the inferior alveolar nerve canal in three-dimensional space using their centroids.

[0077] In S61, "labeling all candidate connected components in three-dimensional space" means labeling them according to the spatial location of the corresponding cross-sectional image layer.

[0078] Optionally, output the standardized annotation results of the neural tube: the three-dimensional neural tube center path and the three-dimensional spatial coordinates of the candidate connected components.

[0079] The following is a comparison of the effects of using the annotation method of this invention with the prior art:

[0080] ①Based on the results of manual review by multiple experts: When processing the same 50 CBCT data, manual annotation takes an average of 8.5 minutes per case, while this invention only takes 15 seconds per case, improving overall efficiency by approximately 34 times. Furthermore, the annotation accuracy of this invention is slightly higher than that of manual annotation. In other words, this invention can significantly reduce annotation time.

[0081] ② In existing technologies, a large number of labeled samples can be used to train a deep learning model to achieve automatic identification and labeling of neural tubes. However, the accuracy of this existing technology is directly affected by the CT equipment, scanning parameters, and differences in jawbone structure. Furthermore, the accuracy of this method directly depends on the number of labeled samples; only a large number of labeled samples can yield good labeling results. This invention, however, does not require large-scale sample labeling, thus directly saving significant manpower and labeling time while ensuring the stability of the labeling accuracy. Tested on 50 CBCT datasets, the labeling accuracy (trajectory error ≤ 1 mm) of this invention is 93%, which is essentially equivalent to the existing pre-trained deep learning model (95%). However, this invention does not require pre-training of the deep learning model. Taking 1000 3D CT datasets as a training set as an example, deep learning methods require hundreds of GPU hours (approximately 2 weeks) of training overhead, while this invention can be directly applied to CBCT data. For the same 50 CBCT datasets, the overall labeling efficiency of this invention is approximately 20 times higher than that of the deep learning model.

[0082] ③ On CT data from different manufacturers with slice thicknesses of 0.2 to 0.5 mm, the accuracy of this invention remained stable in the range of 90% to 95%; while the annotation accuracy of deep learning models was greatly affected by the machine that acquired the CT data and the scale of the training data.

[0083] The annotation method of this invention can be directly applied to CT data. Based on the cross-section of the dental arch curve, the centerline of the inferior alveolar nerve canal in three-dimensional space is obtained. By combining the candidate connected regions of the cross-sectional images of each layer, the nerve canal in three-dimensional space can be reconstructed, and the nerve canal can be marked, which facilitates the determination of various spatial data of the nerve canal.

[0084] Compared with manual annotation, the annotation method of the present invention is highly efficient while ensuring extremely high and stable accuracy, saving a lot of labor costs.

[0085] Compared to deep learning models that require large-scale labeled data for training, the annotation method of this invention can be directly applied to CT data without labeled data, saving the time of training deep learning models and the manpower cost of using large-scale labeled data. Furthermore, it avoids the problem that the annotation of deep learning models is greatly affected by the differences in CT equipment, scanning parameters and jaw structure. The annotation results of this invention are more robust.

[0086] The present invention also provides an automatic neural tube annotation system based on the cross-section of the dental arch curve, comprising: a projection module, a cross-sectional image acquisition module, a binarized image acquisition module, a filtering module, and an annotation module;

[0087] The projection module is used to obtain the position of the projection of the mental foramen of the inferior alveolar nerve in the corresponding fitted dental arch curve parameter coordinate system, and then send it to the cross-sectional image acquisition module;

[0088] The cross-sectional image acquisition module is used to acquire cross-sectional images at each sampling point on the fitted dental arch curve, starting from the position of the mental foramen of the inferior alveolar nerve, and then send them to the binarized image acquisition module;

[0089] The binarized image acquisition module is used to generate a binarized image based on the cross-sectional image and then send it to the filtering module;

[0090] The filtering module is used to filter out candidate connected components from the binarized image and send them to the annotation module;

[0091] The annotation module is used to perform three-dimensional reconstruction of neural tubes after spatial orientation of candidate connected components.

[0092] Each module is programmed or configured to perform the steps of the above-described automatic neural tube annotation method based on the cross-section of the dental arch curve.

[0093] The present invention also provides a computer-readable storage medium storing a computer program programmed or configured to perform the above-described automatic neural tube annotation method based on the cross-section of the dental arch curve.

[0094] The present invention also provides a computer program product, including a computer program / instructions, which are executed by a processor to implement the steps of the above-described automatic neural tube annotation method based on the cross-section of the dental arch curve.

[0095] The technologies, shapes, and structures not described in detail in this invention are all well-known technologies. It should also be noted that the above are merely preferred embodiments of this invention and are not intended to limit the scope of the invention. The components or steps in the embodiments of this invention can be decomposed and / or recombined, and these decompositions and / or recombinations should be considered equivalent solutions to this application and should all fall within the protection scope of this invention.

Claims

1. An automatic neural tube annotation method based on the cross-section of the dental arch curve, characterized in that, Includes the following steps: Step 1: Obtain the position of the inferior alveolar nerve mental foramen projection in the corresponding fitted dental arch curve parameter coordinate system; Step 2: Starting from the position of the mental foramen of the inferior alveolar nerve, obtain cross-sectional images at each sampling point on the fitted dental arch curve; Step 3: After marking the regions of the cross-sectional image based on the HU value, a binarized image is generated; Step 4: After extracting the connected components in the binarized image, select the connected components containing the target contour as candidate connected components. Step 5: Spatial alignment of candidate connected domains. The spatial alignment result includes the central line of the inferior alveolar nerve canal formed by the centroids of all successfully matched candidate connected domains. Then, three-dimensional reconstruction of the nerve canal is performed based on the spatial alignment result to complete the annotation of the nerve canal.

2. The automatic neural tube annotation method based on the cross-section of the dental arch curve according to claim 1, characterized in that, After step 4 and before step 5, there is also step 4': Step 4' matches all adjacent candidate connected components in each sampling point: if the spatial distance and geometric similarity between adjacent candidate connected components both meet the set threshold, then the two adjacent candidate connected components are matched successfully; otherwise, they are matched unsuccessfully.

3. The automatic neural tube annotation method based on the cross-section of the dental arch curve according to claim 1, characterized in that, Step 2 also includes the following: Step 21: Using the mental foramen of the inferior alveolar nerve as the starting sampling point, determine several sampling points on the fitted dental arch curve; Step 22: Starting from the current sampling point, along the orthogonal direction of the current sampling point on the fitted dental arch curve, sequentially acquire cross-sectional images of the current sampling point in the 3D CT image with the number of layers continuously increasing, with a preset sampling step size; until the cross-sectional images corresponding to all sampling points on the current fitted dental arch curve are acquired.

4. The automatic neural tube annotation method based on the cross-section of the dental arch curve according to claim 3, characterized in that, Step 3 also includes the following: Step 31: Mark the bone region and cavity region in the cross-sectional image based on the HU value; Step 32: Use the bone region as the foreground pixel and the cavity region as the background pixel to generate a binarized image corresponding to each cross-sectional image.

5. The automatic neural tube annotation method based on the cross-section of the dental arch curve according to claim 4, characterized in that, In step 4, also Includes the following: S41, Denoise the binarized image; S42, based on foreground and background pixels, extracts connected components from the denoised binarized image and calculates the geometric properties of each connected component; the geometric properties include area A, perimeter P, and roundness index C, where... S43, select connected components that satisfy the geometric properties of the target contour as candidate connected components, and record the centroid coordinates and boundary point set corresponding to the candidate connected components.

6. The automatic neural tube annotation method based on the cross-section of the dental arch curve according to claim 2, characterized in that, In step 4′, also This includes the following: the Euclidean distance between the centroids of adjacent candidate connected components and the similarity of their boundary point sets, representing the spatial distance and geometric similarity of the adjacent candidate connected components, respectively; if the spatial distance between adjacent candidate connected components is less than the first distance threshold τ... d And the geometric similarity is greater than the second similarity threshold τ s If the two adjacent candidate connected components match successfully, then the two adjacent candidate connected components match successfully; otherwise, the two adjacent candidate connected components match unsuccessfully.

7. The automatic neural tube annotation method based on the cross-section of the dental arch curve according to claim 2, characterized in that, Step 5 also includes the following: Step 51: Connect the centroids of all successfully matched candidate connected components sequentially according to the spatial position of the corresponding sampling point on the fitted dental arch curve and the spatial position of the corresponding cross-sectional image layer, to form the center line of the current inferior alveolar nerve canal in three-dimensional space; at the same time, mark all candidate connected components in three-dimensional space. Step 52: Based on the centerline of the inferior alveolar nerve canal in three-dimensional space and all candidate connected regions in three-dimensional space, perform three-dimensional reconstruction of the nerve canal and complete the annotation of the nerve canal.

8. An automatic neural tube annotation system based on the cross-section of the dental arch curve, characterized in that, include: The module includes a projection module, a cross-sectional image acquisition module, a binarized image acquisition module, a filtering module, and an annotation module. The projection module acquires the position of the inferior alveolar nerve mental foramen projection in the corresponding fitted dental arch curve parameter coordinate system and then sends it to the cross-sectional image acquisition module. The cross-sectional image acquisition module acquires cross-sectional images at each sampling point on the fitted dental arch curve, starting from the position of the inferior alveolar nerve mental foramen, and then sends them to the binarization image acquisition module. The binarization image acquisition module generates a binarized image based on the cross-sectional images and then sends it to the filtering module. The filtering module filters out candidate connected components from the binarized image and sends them to the annotation module. The annotation module is used to perform three-dimensional reconstruction of the neural tube after spatially arranging the candidate connected components; each module is programmed or configured to perform the steps of the automatic neural tube annotation method based on the cross section of the dental arch curve as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program that is programmed or configured to perform an automatic neural tube annotation method based on the cross-section of the dental arch curve as described in any one of claims 1-7.

10. A computer program product comprising a computer program / instructions, characterized in that: When executed by a processor, the computer program / instructions implement the steps of an automatic neural tube annotation method based on the cross-section of a dental arch curve as described in any one of claims 1-7.

Images