Method and system for identifying inferior alveolar nerves in CT (Computed Tomography) image

Abstract

The invention relates to a method and system for identifying inferior alveolar nerves in a CT image, and the method comprises the steps: obtaining a CT image sample of an inferior alveolar nerve tube, marking the feature points of the inferior alveolar nerves on the center line of the inferior alveolar nerves in the CT image sample, so as to construct a statistical shape model of the inferior alveolar nerves; a to-be-recognized CT image of the patient is obtained, the starting point and the end point of the inferior alveolar neural tube are detected from the to-be-recognized CT image, the statistical shape model is aligned with the detected starting point and the end point of the inferior alveolar neural tube, all cross sections at the starting point and the end point of the inferior alveolar neural tube in the to-be-recognized CT image are obtained, and multiple to-be-recognized slices are formed; and inputting each slice to be identified into a pre-trained target detection model to obtain a segmentation result of the inferior alveolar nerve, and taking the statistical shape model as priori knowledge in the training process of the target detection model. Compared with the prior art, the method has the advantages that the inferior alveolar nerve recognition precision and efficiency can be improved, and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of inferior alveolar nerve recognition, and in particular to a method and system for recognizing the inferior alveolar nerve in CT images. Background Art

[0002] The inferior alveolar nerve is a part of the cranial nerves and is also known as the mandibular branch of the trigeminal nerve. In the field of oral cavity, preoperative high-resolution three-dimensional images of patients can be obtained through preoperative CBCT scanning, and it is of great clinical significance to detect the inferior alveolar nerve in CBCT images. By segmenting and reconstructing the three-dimensional morphology of the inferior alveolar nerve, doctors can evaluate the position and course of the inferior alveolar nerve of patients before oral surgery, so as to plan a more reasonable surgical operation path, avoid damaging the inferior alveolar nerve during the operation, and reduce potential complications. However, accurately identifying the inferior alveolar nerve from CBCT and reconstructing its three-dimensional morphology is a difficult task. The reasons are as follows. First, there are many soft tissue structures around the inferior alveolar nerve, such as blood vessels, muscles and mucous membranes, which makes it difficult to clearly distinguish the contour of the inferior alveolar nerve in CBCT. In addition, there are individual differences in anatomical structures among different patients, and the morphology and course of the inferior alveolar nerve are not the same. Finally, metal artifacts or scatter artifacts may appear in CBCT images, which pose challenges to the accurate identification of the inferior alveolar nerve. If the inaccurate identification leads to doctor's operation mistakes, what bad consequences will occur, such as facial nerve injury, facial paralysis or even death. Three-dimensional CBCT usually has hundreds of two-dimensional sections, and layer-by-layer processing of two-dimensional CBCT is a time-consuming and laborious task, and has high requirements for the experience and professional training of dentists. To solve the above problems, some computer-aided automatic recognition methods for the inferior alveolar nerve have been proposed.

[0003] Some traditional computer vision-based methods have been proposed for semi-automatic or full-automatic segmentation of the inferior alveolar nerve. For example: methods based on texture analysis technology, methods based on threshold and morphological operations, methods based on the combination of template functions and multi-hypothesis tracking, methods based on statistical shape models, etc. However, these methods require manual setting of some conditions or adjustment of hyperparameters, resulting in poor generalization performance and robustness performance in the detection of the inferior alveolar nerve. In recent years, deep learning-based automatic detection methods for the inferior alveolar nerve have become the mainstream. For example: Jaskari et al. proposed a fully convolutional neural network for automatic detection of the inferior alveolar nerve. The results showed that the deep learning-based method can segment the mandibular canal nerve at the pixel level, effectively improving the segmentation accuracy. Kwak et al. used and compared the results of 2D SegNet, 2D U-Net, and 3D U-Net networks in the automatic detection of the inferior alveolar nerve. The results showed that the method based on the 3D U-Net network has higher accuracy than other comparison methods. Liu et al. proposed an automatic detection method for the third mandibular molar and the mandibular canal based on the U-Net network. The results showed that the deep learning-based automatic detection ability of the inferior alveolar nerve is comparable to that of doctors. In addition, multiple methods have tested the automatic segmentation of the inferior alveolar nerve based on the U-Net network, all using the inferior alveolar nerve manually labeled by doctors as the test gold standard, and all indicating the application potential of the deep learning-based method in detection accuracy and efficiency.

[0004] However, the above deep learning-based methods only consider the semantic information of CBCT image data, without considering the morphological characteristics of the inferior alveolar nerve itself, resulting in poor segmentation in some noise regions and incomplete three-dimensional reconstruction morphology. Summary of the Invention

[0005] The purpose of the present invention is to overcome the above-mentioned defects of the existing technology that only consider the semantic information of CBCT image data without considering the morphological characteristics of the inferior alveolar nerve itself, resulting in poor segmentation in some noise regions and incomplete three-dimensional reconstruction morphology, and to provide a method and system for identifying the inferior alveolar nerve in CT images.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A method for identifying the inferior alveolar nerve in CT images, comprising the following steps:

[0008] Obtain CT image samples of the inferior alveolar nerve canal, mark the feature points of the inferior alveolar nerve on the center line of the inferior alveolar nerve in the CT image samples, so as to construct a statistical shape model of the inferior alveolar nerve;

[0009] Obtain the CT image to be recognized of the patient, detect the starting point and ending point of the inferior alveolar nerve canal from the CT image to be recognized, align the statistical shape model with the detected starting point and ending point of the inferior alveolar nerve canal, and obtain all cross-sections at the starting point and ending point of the inferior alveolar nerve canal in the CT image to be recognized to form multiple slices to be recognized;

[0010] Input each slice to be recognized into a pre-trained object detection model to obtain the segmentation result of the inferior alveolar nerve. During the training process of the object detection model, the statistical shape model is used as prior knowledge.

[0011] Further, the process of constructing the statistical shape model of the inferior alveolar nerve is specifically as follows:

[0012] Mark the feature points of the inferior alveolar nerve on the center line of the inferior alveolar nerve in the CT image sample, perform interpolation processing on the marked feature points to obtain all sample feature points;

[0013] After registering all the sample feature points, model to obtain the statistical shape model of the inferior alveolar nerve.

[0014] Further, the detection process of the starting point and ending point of the inferior alveolar nerve canal includes:

[0015] Use mandibular surgical planning software to identify the condyle vertex, mandibular angle point, submental point, point E, point A 2 point, point O, the zero-order oriented bounding box OBB of the entire mandible MD , and the inferior border plane S of the mandible down ; thereby extracting the starting point and ending point of the inferior alveolar nerve canal.

[0016] Further, the extraction process of the starting point of the inferior alveolar nerve canal includes:

[0017] Obtain the midpoint between point A 2 and point O, and define the point with the maximum curvature within a 10-mm spatial range of this midpoint as the starting point of the inferior alveolar nerve canal.

[0018] Further, the detection process of the ending point of the inferior alveolar nerve canal includes:

[0019] Obtain the center point C MD of the zero-order oriented bounding box OBB MD , and define a reference point P ref = C MD +(Me - C MD )*0.6; obtain the reference plane S down perpendicular to the inferior border plane S of the mandible ref and passing through the reference point P ref ;

[0020] Through the reference plane S ref Cut the mandibular model constructed in the mandibular surgery planning software to obtain the center point C of the cutting contour CT ;

[0021] Taking the center point C of the cutting contour CT The point with the maximum curvature in the mandibular model within a spatial range of 10 mm is the end point of the inferior alveolar nerve canal

[0022] Furthermore, the object detection model includes an input end, a Backbone part, a Neck part, and a Prediction part connected in sequence. The Backbone part adopts a Focus structure and a CSP structure, and the Neck part adopts an FPN+PAN structure

[0023] Furthermore, the expression of the loss function of the object detection model is:

[0024] L = λ 1 *L box ′ + λ 2 *L cLs + λ 3 *L obj

[0025] In the formula, L is the loss function of the object detection model, L bxox ′ is the final localization loss, L cls is the classification loss, L obj is the confidence loss, λ 1 is the localization loss coefficient, λ 2 is the classification loss coefficient, λ 3 is the confidence loss coefficient

[0026] Furthermore, the calculation expression of the final localization loss is:

[0027] L box ′ = L box *(1 - g box )

[0028] In the formula, L box is the localization loss, g box is the Gaussian probability corresponding to the prior box formed after the statistical shape model is aligned with the CT image to be recognized

[0029] Furthermore, the method further includes performing three-dimensional reconstruction and visualization on the segmentation result of the inferior alveolar nerve obtained

[0030] The present invention also provides a recognition device for the inferior alveolar nerve in CT images, including a memory and a processor. The memory stores a computer program, and the processor calls the computer program to execute the steps of the method described above.

[0031] Compared with the prior art, the present invention has the following advantages:

[0032] (1) To overcome the problem that existing deep learning-based methods only consider the semantic information of CBCT image data, the present invention proposes a new deep learning detection method for the inferior alveolar nerve that combines object detection and statistical shape prior models. First, a series of representative inferior alveolar nerve models manually labeled and reconstructed by doctors are used to establish a statistical shape model; then, the starting and ending points of the inferior alveolar nerve canal are determined based on the patient's CT image to achieve alignment with the statistical shape model, obtaining a series of slices with prior boxes of the statistical shape model; finally, the inferior alveolar nerve is segmented based on the slices with prior boxes through an object detection algorithm to construct an inferior alveolar nerve model.

[0033] On the one hand, introducing the statistical shape model as prior knowledge in this process can effectively overcome the disadvantages of poor detection accuracy and discontinuity of the inferior alveolar nerve canal based on two-dimensional CT sections; on the other hand, only object detection needs to be performed on two-dimensional CT sections, which can effectively improve the detection efficiency during deployment.

[0034] (2) In the process of determining the starting and ending points of the inferior alveolar nerve canal in this application, key anatomical landmark points of the mandible identified by mandibular surgical planning software are proposed, and then the starting and ending positions of the inferior alveolar nerve canal are deduced and calculated to achieve efficient and accurate automatic detection.

[0035] (3) In the process of constructing the statistical shape model of the present invention, doctors only need to mark sparse feature points from CT images to construct the statistical shape model, which not only conforms to the operation habits of doctors in clinical practice but also shortens the marking time. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a schematic flowchart of a method for recognizing the inferior alveolar nerve in CT images provided in an embodiment of the present invention;

[0037] Figure 2 is a schematic framework diagram of a method for recognizing the inferior alveolar nerve in CT images provided in an embodiment of the present invention;

[0038] Figure 3 is a schematic diagram of the recognition result of key anatomical landmark points of the mandible provided in an embodiment of the present invention, where a is the automatic recognition result of key anatomical landmark points of the mandible, and b are the zero-order, first-order, and second-order bounding boxes of the mandible;

[0039] Figure 4 Schematic diagram of a deep learning algorithm framework for automatic detection of the inferior alveolar nerve guided by statistical shape prior provided in the embodiment of the present invention;

[0040] Figure 5 Visualization comparison of detection results of the inferior alveolar nerve canal by different methods in the embodiment of the present invention; each column represents the result of one method or ground truth; the first row is the error distribution diagram of the detection result and the gold standard; the second to fourth rows are the contours of the detected inferior alveolar nerve canal in three representative cross-sections, where the red is the predicted contour and the green is the ground truth contour. Detailed implementation manners

[0041] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and illustrated herein can be arranged and designed in various different configurations.

[0042] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0043] It should be noted that: similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0044] Embodiment 1

[0045] As Figure 1 and Figure 2 shown, this embodiment provides a method for identifying the inferior alveolar nerve in CT images, including the following steps:

[0046] S1: Obtain CT image samples of the inferior alveolar nerve canal, mark the feature points of the inferior alveolar nerve on the center line of the inferior alveolar nerve in the CT image samples, so as to construct a statistical shape model of the inferior alveolar nerve;

[0047] S2: Obtain the CT image to be recognized of the patient, detect the starting point and the ending point of the inferior alveolar nerve canal from the CT image to be recognized, align the statistical shape model with the detected starting point and ending point of the inferior alveolar nerve canal, and obtain all cross-sections at the starting point and ending point of the inferior alveolar nerve canal in the CT image to be recognized, so as to form a plurality of slices to be recognized;

[0048] S3: Input each slice to be recognized into a pre-trained object detection model to obtain the segmentation result of the inferior alveolar nerve. During the training process of the object detection model, a statistical shape model is used as prior knowledge.

[0049] Finally, the automatic detection algorithm for the inferior alveolar nerve region can be integrated into the oral implant surgical planning software to execute the operation process.

[0050] Equivalently, this application mainly includes the following processes

[0051] 1. Establish a statistical shape model of the inferior alveolar nerve using sparse key points of the inferior alveolar nerve manually marked by a doctor

[0052] The process of constructing the statistical shape model of the inferior alveolar nerve is specifically as follows:

[0053] Mark the feature points of the inferior alveolar nerve on the center line of the inferior alveolar nerve in the CT image sample, and perform interpolation processing on the marked feature points to obtain all sample feature points;

[0054] After registering all the sample feature points, a statistical shape model of the inferior alveolar nerve is built.

[0055] The specific principle is as follows:

[0056] It can be observed that although there are individual differences in the inferior alveolar nerve of each person, the overall shape presents a regular arc. This shape distribution rule can be used as prior information for the automatic detection algorithm of the inferior alveolar nerve. As a geometric model, the statistical shape model can well represent the shape characteristics of a group of semantically similar objects. In this embodiment, a statistical shape model is first established using a group of three-dimensional models of the inferior alveolar nerve.

[0057] In order to reconstruct the three-dimensional model of the inferior alveolar nerve from CT for constructing the statistical shape model, it is usually necessary to outline the inferior alveolar nerve region in each slice of CT, resulting in a long time consumption. Therefore, in this embodiment, several feature points on the sparse center line of the inferior alveolar nerve are used to represent the shape of the inferior alveolar nerve. The doctor marks sparse feature points from the CT image, which not only conforms to the doctor's operation habits in clinical practice but also shortens the marking time.

[0058] Denote all the marking data used to establish the statistical shape model as where N SSM represents the total number of samples. For the i-th sample, the sparse feature points marked by the doctor are denoted as where represents the total number of sparse feature points. To obtain more feature points, the B-spline interpolation method

[19] is used to interpolate the sparse feature points, and finally, all the feature points used to represent the shape of the i-th sample are denoted as where N S represents the total number of feature points.

[0059] Before establishing the statistical shape model, it is necessary to establish a one-to-one correspondence between the feature points of all samples. For this purpose, in this embodiment, the non-rigid ICP algorithm is used to register all the sample feature points, thereby establishing point-to-point matching. Modeling the shape variation using the normal distribution can be expressed as:

[0060]

[0061] where μ represents the average shape and ∑ represents the covariance matrix. The two can be expressed as:

[0062]

[0063] 2. Propose an automatic detection algorithm for the entrance and exit points of the inferior alveolar nerve canal to achieve the alignment of the average shape model with the actual patient

[0064] For the actual patient, during the inference process of the inferior alveolar nerve canal, since the method adopted in this embodiment is a two-dimensional method, in order to improve the inference efficiency, the cross-section containing the inferior alveolar nerve canal is detected from the complete CT of the patient. To detect all the cross-sections with the inferior alveolar nerve canal, only the starting point and the ending point of the inferior alveolar nerve canal need to be detected. In this embodiment, an automatic detection algorithm for the entrance and exit points of the left and right inferior alveolar nerve canals is proposed, and its basic idea is to use the pre-operative segmented and three-dimensional reconstructed mandibular bone model to achieve automatic detection of the entrance and exit points based on the anatomical morphological characteristics.

[0065] Using the mandibular bone surgical planning software developed in the early stage within the laboratory, the automatic recognition of key anatomical landmark points of the mandible can be achieved, including the condylion (Co), gonion (Go), meton (Me), point E, point A 2 point and point O, as well as the zero-order oriented bounding box OBB of the entire mandible MD , and the lower border surface S of the mandible down , as Figure 3 shown.

[0066] For the starting point of the inferior alveolar nerve canal, the automatic detection method process proposed in this embodiment is: automatically identify point A 2 point and point O, and obtain the midpoint of the two points. Define the point with the maximum curvature in the three-dimensional model of the mandible within the spatial range of 10 mm from the midpoint as the starting point of the inferior alveolar nerve canal.

[0067] For the endpoint of the inferior alveolar nerve canal, first obtain the zero-order oriented bounding box (OBB) of the mandible model MD and its center point, denoted as C MD , and define a reference point P ref = C MD +(Me - C MD ) * 0.6. Obtain a reference plane S down that is perpendicular to S ref and passes through P ref . Use S ref to cut the mandible model to obtain the center point C CT of the cutting profile. Define the endpoint of the inferior alveolar nerve canal as the point with the maximum curvature in the 3D mandible model within a 10 mm spatial range of C CT .

[0068] After obtaining the left and right entry / exit points of the patient's inferior alveolar nerve canal and those of the average shape model respectively, take out all the cross-sections of the patient's CT at the starting point and the ending point of the inferior alveolar nerve canal for the inference of the inferior alveolar nerve canal.

[0069] 3. On this basis, propose an automatic detection algorithm for the inferior alveolar nerve region in CT slices guided by statistical shape prior

[0070] The specific principle is as follows:

[0071] After establishing the statistical shape model, in order to apply the statistical prior information to the detection of the inferior alveolar nerve canal, it is necessary to align the average shape model with the actual patient. In the inferior alveolar nerve canal dataset of this article, both the statistical shape model and the inferior alveolar nerve canal manually marked by doctors are standardized and then manually aligned.

[0072] Use the object detection method to detect the inferior alveolar nerve region in the CT 2D slices. As a mature object detection method, the YOLO series algorithms have very wide applications in medical image object detection. However, if only the YOLO object detection method is used to detect the objects in a single-layer CT slice, the wrong detection results of the single-layer slice will lead to poor overall reconstruction accuracy of the inferior alveolar nerve. For this reason, in this embodiment, the YOLO object detection is combined with the statistical shape prior of the inferior alveolar nerve to propose a new automatic detection framework for the inferior alveolar nerve, and its overall network structure is as Figure 4 shown.

[0073] The adopted object detection framework is YOLO-V5, which is a lightweight and efficient object detection framework. Its network structure is divided into four parts: the input end, Backbone, Neck, and Prediction. Among them, the Backbone part adopts the Focus structure and the CSP structure, and the Neck part adopts the FPN+PAN structure, which improves the receptive field and representation ability of the model.

[0074] The loss function of the YOLO-V5 object detection deep learning network is divided into three parts, namely the localization loss L box , calculated for each object; the classification loss L cls , calculated for each grid; and the confidence loss L obj , calculated for each object. After establishing the statistical shape model and aligning the average shape model with the patient's CT space, for each slice, a Gaussian probability g box can be obtained. Combining the Gaussian probability with the localization loss, a new localization loss function is obtained, defined as:

[0075] L box ′ = L box * (1 - g box )

[0076] It should be noted that since YOLO will perform random data augmentation, there may be more than one bounding box in the current slice. Each bounding box may correspond to multiple two-dimensional Gaussian distributions. In this case, the maximum Gaussian probability value is taken.

[0077] In summary, the loss function of the automatically detected inferior alveolar nerve canal network finally obtained is defined as:

[0078] L = λ 1 * L box ′ + λ 2 * L cls + λ 3 * L obj

[0079] In the formula, L is the loss function of the object detection model, L box ′ is the final localization loss, L c;s is the classification loss, L obj is the confidence loss, λ 1 is the localization loss coefficient, λ 2 is the classification loss coefficient, λ 3 is the confidence loss coefficient.

[0080] Furthermore, the automatic detection algorithm for the inferior alveolar nerve region proposed in this embodiment can also be integrated and applied to oral implant surgical planning software to achieve automatic detection, three-dimensional reconstruction, and visualization of the inferior alveolar nerve. The software constructed in this embodiment is developed in the Python operating environment, and the main running libraries used include VTK (https: / / www.vtk.org), SimpleITK (https: / / simpleitk.org / ), Pytorch (https: / / pytorch.org / ), etc. This software consists of four modules: Segmentation module, Landmarks module, Surgical planning module, and Visualization module.

[0081] In this embodiment, three-dimensional reconstruction and visualization are performed on the detected inferior alveolar nerve canal. The reconstruction results of several representative cases are as Figure 5 shown. It can be seen from the visualization results that the method proposed in this embodiment has better overall accuracy and can better reflect the three-dimensional shape of the inferior alveolar nerve. This is due to the introduction of a statistical shape model in our deep learning network.

[0082] A comparison was made with the current state-of-the-art object detection methods. The results show that this method has achieved the best experimental results in the three evaluation indicators of Precision, Recall, and F1-score, indicating that the method of combining object detection with a statistical shape model proposed in this study has significant advantages.

[0083] The advantages of this technical solution are as follows: (1) Through object detection on two-dimensional CT sections, the detection efficiency can be effectively improved during deployment. (2) By introducing the prior of a statistical shape model, the disadvantages of poor detection accuracy and discontinuity of the inferior alveolar nerve canal based on two-dimensional CT sections can be effectively overcome.

[0084] This embodiment also provides an identification device for the inferior alveolar nerve in CT images, including a memory and a processor. The memory stores a computer program, and the processor calls the computer program to execute the steps of the method described above.

[0085] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative labor. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field of the present invention through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should fall within the protection scope determined by the claims.

Claims

1. A method for identifying the inferior alveolar nerve in a CT image, characterized in that: The following steps are involved: Acquire a CT image sample of the inferior alveolar nerve canal, and mark characteristic points of the inferior alveolar nerve on the center line of the inferior alveolar nerve in the CT image sample to construct a statistical shape model of the inferior alveolar nerve; Acquire a CT image of the patient to be identified, detect the starting point and the end point of the inferior alveolar nerve canal from the CT image to be identified, align the statistical shape model with the detected starting point and the end point of the inferior alveolar nerve canal, and acquire all cross sections at the starting point and the end point of the inferior alveolar nerve canal in the CT image to be identified, to form a plurality of slices to be identified; Each slice to be identified is input into a pre-trained target detection model to obtain the segmentation result of the inferior alveolar nerve. During the training process of the target detection model, the statistical shape model is used as prior knowledge.

2. The method for identifying the inferior alveolar nerve in a CT image according to claim 1, characterized in that: The process of constructing the statistical shape model of the inferior alveolar nerve is specifically as follows: Marking the characteristic points of the inferior alveolar nerve on the center line of the inferior alveolar nerve in the CT image sample, and performing interpolation processing on the marked characteristic points to obtain all the sample characteristic points; After all sample feature points are registered, a statistical shape model of the inferior alveolar nerve is obtained by modeling.

3. The method for identifying the inferior alveolar nerve in a CT image according to claim 1, characterized in that: The process of detecting the starting point and the end point of the inferior alveolar nerve tube includes: The mandibular surgery planning software is used to identify the condylar apex, mandibular angle, submental point, point E, point A2, point O, and the zero-order bounding box OBB of the entire mandible from the CT image to be identified. MD , and the lower edge of the mandible S down ; thereby extracting the starting point and end point of the inferior alveolar nerve canal.

4. The method for identifying the inferior alveolar nerve in a CT image according to claim 3, characterized in that: The extraction process of the starting point of the inferior alveolar nerve tube includes: The midpoint between point A2 and point O was obtained, and the point with the largest curvature within the spatial range of 10 mm of the midpoint was defined as the starting point of the inferior alveolar nerve canal.

5. The method for identifying the inferior alveolar nerve in a CT image according to claim 3, characterized in that: The process of detecting the end point of the inferior alveolar nerve tube includes: Get the zero-order bounding box OBB MD The center point C MD , and define a reference point P ref =C MD +(Me-C MD )*0.6; obtain the surface perpendicular to the lower edge of the mandible S down and passes through the reference point P ref The reference plane S ref ; Through the reference plane S ref Cut the mandibular model constructed in the mandibular surgery planning software to obtain the cutting contour center point C CT ; Cutting contour center point C CT The point with the largest curvature in the mandibular model within the spatial range of 10 mm is the end point of the inferior alveolar nerve canal.

6. The method for identifying the inferior alveolar nerve in a CT image according to claim 1, characterized in that: The target detection model includes an input end, a Backbone part, a Neck part and a Prediction part which are connected in sequence. The Backbone part adopts a Focus structure and a CSP structure, and the Neck part adopts an FPN+PAN structure.

7. The method for identifying the inferior alveolar nerve in a CT image according to claim 1, characterized in that: The loss function of the target detection model is expressed as: L=λ1*L box ′+λ2*L cls +λ3*L obj Where L is the loss function of the target detection model, L box ′ is the final positioning loss, L c;s is the classification loss, L obj is the confidence loss, λ1 is the positioning loss coefficient, λ2 is the classification loss coefficient, and λ3 is the confidence loss coefficient.

8. The method for identifying the inferior alveolar nerve in a CT image according to claim 7, characterized in that: The calculation expression of the final positioning loss is: L box ′=L box *(1-g box ) Where, L box is the positioning loss, g b0x It is the Gaussian probability corresponding to the prior box formed after the statistical shape model is aligned with the CT image to be identified.

9. The method for identifying the inferior alveolar nerve in a CT image according to claim 8, characterized in that: The method also includes three-dimensionally reconstructing and visualizing the acquired segmentation results of the inferior alveolar nerve.

10. A device for identifying the inferior alveolar nerve in a CT image, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor calls the computer program to execute the steps of any one of the methods according to claims 1 to 9.

Images