Medical navigation method, device, equipment and storage medium based on CBCT
Through the medical navigation method based on CBCT, the image registration model is used to automatically align real-time and three-dimensional simulated images in dental surgery, solving the problems of complex navigation and slow registration speed in the prior art, and achieving fast and accurate teeth Root navigation image generation improves the surgical experience.
Patent Information
- Application Number
- CN202411188889.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-28
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-08-28
AI Technical Summary
The existing oral surgical navigation methods require an external 3D stereo infrared navigation camera and calibration plate. The process is complicated and requires manual alignment. It needs to be recalibrated after the camera angle is converted, resulting in slow registration speed during the operation and poor experience.
Using CBCT-based medical navigation method, the pre-trained image registration model, including residual network and registration network, automatically and quickly align real-time images and three-dimensional simulation images through feature extraction and parameter correction, and realize automatic and automatic tooth root model Generate navigation images to reduce manual intervention.
It realizes automatic and rapid alignment in dental surgery, improves registration accuracy and speed, reduces interference during surgery, and improves patient experience.
Smart Images

Figure CN118697473B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of medical technology, and in particular to a CBCT-based medical navigation method, device, equipment and storage medium. Background Art
[0002] At present, oral surgery navigation requires the use of an external 3D stereo infrared navigation camera and a calibration plate clamped on the teeth. The real-time image of the 3D stereo infrared navigation camera is used to synthesize a three-dimensional simulation image through CBCT technology, and then the calibration plate is used for image registration through the 3D stereo infrared navigation camera to project the tooth roots in the three-dimensional simulation image onto the real-time image. Based on the above method, a calibration plate needs to be clamped on the patient's teeth before each operation, and medical staff need to manually align the real-time image and the three-dimensional simulation image according to the calibration plate. The process is complicated and the experience of the user to be operated is not good. After the 3D stereo infrared navigation camera changes the camera angle, it must be recalibrated. Summary of the invention
[0003] The present application provides a CBCT-based medical navigation method, apparatus, device and storage medium for automatically and quickly aligning real-time images and three-dimensional simulation images in dental surgery.
[0004] In a first aspect, an embodiment of the present application provides a CBCT-based medical navigation method, which is applied to a terminal device, wherein the terminal device is configured with a pre-trained image registration model, wherein the image registration model includes a residual network and a registration network, and the method includes:
[0005] Acquire a real-time image of teeth captured by a camera, and acquire a simulation model of teeth generated by a CBCT instrument scanning the teeth, wherein the simulation model includes: a crown model and a root model;
[0006] Determine a model cross-sectional image of the crown model according to a plurality of preset spatial parameters;
[0007] Based on the residual network, feature extraction is performed on the real-time image to obtain a first feature parameter, and feature extraction is performed on each of the model cross-sectional images to obtain multiple second feature parameters, and a first residual parameter of the first feature parameter and each of the second feature parameters is calculated;
[0008] Based on the registration network, determining initial spatial parameters from the preset spatial parameters according to the first residual parameters, and obtaining a model debugging image according to preset debugging parameters, the initial spatial parameters and the simulation model;
[0009] Based on the residual network, second residual parameters of the real-time image and the model debugging image are calculated; based on the registration network, the debugging parameters are corrected according to the second residual parameters to obtain debugging correction parameters, until the second residual parameters converge, and rotation parameters are obtained according to the debugging correction parameters and the initial space parameters;
[0010] According to the rotation parameters, an observation image of the root model on a preset projection plane is determined, and the observation image is superimposed on the real-time image to generate a medical navigation image.
[0011] In a second aspect, an embodiment of the present application provides a CBCT-based medical navigation device, which is applied to a terminal device, wherein the terminal device is configured with a pre-trained image registration model, wherein the image registration model includes a residual network and a registration network, and the device includes:
[0012] A data reading module is used to obtain a real-time image of teeth captured by a camera, and to obtain a simulation model of teeth scanned by a CBCT instrument, wherein the simulation model includes a crown model and a root model;
[0013] An image determination module, used to determine a model cross-sectional image of the crown model according to a plurality of preset spatial parameters;
[0014] A parameter calculation module, configured to perform feature extraction on the real-time image based on the residual network to obtain a first feature parameter, perform feature extraction on each of the model cross-sectional images to obtain a plurality of second feature parameters, and calculate a first residual parameter between the first feature parameter and each of the second feature parameters;
[0015] A parameter debugging module, configured to determine initial spatial parameters from the preset spatial parameters according to the first residual parameters based on the registration network, and obtain a model debugging image according to preset debugging parameters, the initial spatial parameters and the simulation model;
[0016] A result calculation module is used to calculate the second residual parameters of the real-time image and the model debugging image based on the residual network, and to correct the debugging parameters according to the second residual parameters based on the registration network to obtain debugging correction parameters until the second residual parameters converge, and to obtain rotation parameters according to the debugging correction parameters and the initial space parameters;
[0017] The image display module is used to determine the observation image of the root model on a preset projection plane according to the rotation parameters, and to superimpose the observation image on the real-time image to generate a medical navigation image.
[0018] In a third aspect, an embodiment of the present application provides a medical device, the medical device comprising a memory and a processor;
[0019] The memory is used to store computer programs;
[0020] The processor is used to execute the computer program and implement the CBCT-based medical navigation method as described in any one of the embodiments of the present application when executing the computer program.
[0021] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements a CBCT-based medical navigation method as described in any one of the embodiments of the present application.
[0022] The embodiment of the present application provides a medical navigation method based on CBCT, which is applied to a terminal device. The terminal device is configured with a pre-trained image registration model, and the image registration model includes a residual network and a registration network. The method includes: obtaining a real-time image generated by photographing teeth with a camera, and obtaining a simulation model generated by scanning teeth with a CBCT instrument, the simulation model including: a crown model and a root model; determining a model cross-sectional image of a crown model according to multiple preset spatial parameters; based on the residual network, extracting features from the real-time image to obtain a first feature parameter, and extracting features from each model cross-sectional image to obtain multiple second feature parameters, and calculating the first feature parameter. The method comprises the following steps: calculating the first residual parameter of each second feature parameter of the real-time image and the model debugging image based on the residual network, and correcting the debugging parameter based on the second residual parameter to obtain the debugging correction parameter until the second residual parameter converges, and obtaining the rotation parameter based on the debugging correction parameter and the initial spatial parameter; determining the observation image of the root model on the preset projection plane based on the rotation parameter, and superimposing the observation image on the real-time image to generate a medical navigation image. Through the above method, the model cross-sectional images of the simulation model are first obtained at multiple angles based on multiple spatial parameters, and then the real-time image and the model cross-sectional image are compared through the residual network to obtain the model cross-sectional image closest to the real-time image, so as to determine the initial spatial parameters of the registration network. No manual initial alignment is required. The registration network successively aligns the simulation model and the real-time image according to the debugging parameters on the basis of the initial spatial parameters. During the debugging process, the registration network corrects the debugging parameters according to the second residual parameters of the residual network after debugging to ensure the registration accuracy and speed. Finally, the closest rotation parameters are obtained, and the navigation image of the tooth root of the real-time image is obtained according to the rotation parameters and the root model, which realizes following the changes of the real-time image and automatically generating the correct navigation image, and does not require manual intervention in alignment. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.
[0024] Figure 1 A schematic flow chart of a CBCT-based medical navigation method provided in an embodiment of the present application;
[0025] Figure 2 A schematic block diagram of a CBCT-based medical navigation device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0027] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.
[0028] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.
[0029] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.
[0030] CBCT is the abbreviation of Cone Beam Computed Tomography. Compared with traditional CT scans, CBCT can provide high-resolution three-dimensional models with lower radiation doses, and is particularly suitable for diagnosis and treatment planning in the fields of oral and maxillofacial surgery, dentistry, and otolaryngology. At present, the process of medical navigation using the tooth root images corresponding to the CBCT three-dimensional model is relatively complicated, requiring the cooperation of auxiliary devices and alignment and adjustment by medical staff, and automatic projection has not been achieved.
[0031] See also Figure 1 , Figure 1 The embodiment of the present application also provides a schematic flow chart of a medical navigation method based on CBCT. Figure 1 The medical navigation method based on CBCT shown in the figure requires a pre-trained image registration model, which includes a residual network and a registration network, and the real-time image and the three-dimensional simulation model are registered through the image registration model. The specific steps of the medical navigation method based on CBCT include: S101-S106.
[0032] S101, obtaining a real-time image of teeth captured by a camera, and obtaining a simulation model of teeth generated by a CBCT instrument scanning the teeth, wherein the simulation model includes a crown model and a root model.
[0033] Exemplarily, after the patient is scanned by a CBCT instrument, a simulation model of the patient's teeth is generated, and the simulation model includes: a gum model, a tooth crown model and a tooth root model, wherein the crown model and the root model correspond one to one, and the tooth crown model and the tooth root model use the surface of the gum model as a dividing line.
[0034] S102, determining a model cross-sectional image of a crown model according to a plurality of preset spatial parameters.
[0035] Exemplarily, in order to enable the terminal device to achieve initial registration, it is necessary to equip the terminal device with model cross-sectional images from multiple angles, for example, the front, the back, the oblique front and the oblique back, and the selection of these directions needs to cover all angles of the simulation model. Based on these angles and the preset projection surface, multiple preset spatial parameters are generated. After the terminal device obtains the simulation model of the CBCT instrument, it obtains a model cross-sectional image of the simulation model according to the preset spatial parameters, specifically, obtains a model cross-sectional image of the crown model.
[0036] Through the above steps, automated initial registration is achieved, so that during the operation, after the position of the photographic equipment is adjusted significantly, the problem of slow initial registration speed of the registration network when facing the latest real-time image is solved, and manual intervention in registration is not required, thereby improving the registration speed.
[0037] S103, based on the residual network, feature extraction is performed on the real-time image to obtain a first feature parameter, and feature extraction is performed on each model cross-sectional image to obtain multiple second feature parameters, and a first residual parameter between the first feature parameter and each second feature parameter is calculated.
[0038] Exemplarily, a pre-trained or customized residual network (ResNet) is used to forward propagate the real-time photo of the teeth. Based on the residual network, the edge corners of the teeth in the real-time image are identified, for example, the edge corners of the tooth contour and the edge corners of the tooth gap. The output of the last layer or a specific layer can be regarded as the "feature vector" of the image, that is, the first feature parameter. The first feature parameter condenses the important information of the real-time image and can be used for subsequent comparison and alignment. Similarly, the edge corners of the model cross-sectional images of the simulation model at multiple angles are also identified through the residual network to obtain the second feature parameter. The residual network compares the first feature parameter and the second feature parameter by a preset residual comparison method. For example, the preset residual comparison method includes: Euclidean distance, cosine similarity and KL divergence, so as to obtain the first residual parameter of the real-time image and each model cross-sectional image.
[0039] Feature extraction and comparison based on residual networks can improve the accuracy and efficiency of registration in subsequent processes by quantifying feature differences.
[0040] S104, based on the registration network, determine the initial space parameters from the preset space parameters according to the first residual parameters, and obtain the model debugging image according to the preset debugging parameters, the initial space parameters and the simulation model.
[0041] Exemplarily, the registration network selects a target cross-sectional image from multiple model cross-sectional images based on the size of the first residual parameter. The smaller the first residual parameter, the closer the model cross-sectional image is to the real-time target image, and the more suitable it is as the image for initial registration. The registration network obtains the initial spatial parameters corresponding to the target cross-sectional image, and adjusts the spatial relationship between the simulation model and the preset projection surface according to the initial spatial parameters. In order to obtain a projection image that best matches the real-time image, the registration network needs to adjust the spatial relationship between the simulation model and the preset projection surface according to the preset debugging parameters, that is, correct the initial spatial parameters according to the debugging parameters, and obtain the model debugging image after correction. During the correction process, the residual of the model debugging image becomes smaller and smaller with the real-time network, and finally becomes basically the same.
[0042] In some embodiments, the initial spatial parameters are determined from the preset spatial parameters based on the first residual parameters, including: calculating the residual standard deviation of each first residual parameter; setting the first residual parameter corresponding to the largest residual standard deviation as the target residual parameter; setting the model section image corresponding to the target residual parameter as the target section image; and setting the preset spatial parameters corresponding to the target section image as the initial spatial parameters.
[0043] Through the above mechanism, the automation of the initial registration is achieved. If the real-time image changes significantly during the operation, for example, the shooting angle of the camera changes significantly, the registration network can automatically re-register without the need for medical staff to perform manual registration, which helps to reduce interference with the surgical process.
[0044] S105. Based on the residual network, calculate the second residual parameters of the real-time image and the model debugging image, and based on the registration network, correct the debugging parameters according to the second residual parameters to obtain the debugging correction parameters until the second residual parameters converge, and obtain the rotation parameters according to the debugging correction parameters and the initial space parameters.
[0045] Exemplarily, the residual network and the registration network cooperate with each other, and the registration network determines the correction direction of the simulation model according to the change of the residual parameters, and determines the correction amplitude and correction direction of the debugging parameters according to the residual parameters. In this way, the number of registrations can be reduced and the registration speed of the registration network can be improved.
[0046] In some embodiments, based on the registration network, the debugging parameters are corrected according to the second residual parameters to obtain the debugging correction parameters, until the second residual parameters converge, and the rotation parameters are obtained according to the debugging correction parameters and the initial spatial parameters, including: based on the registration network, the first residual parameters and the second residual parameters are compared and analyzed to obtain the residual change parameters. The residual change trend is determined according to the residual change parameter, the feedback parameter is determined according to the residual change trend, and the debugging parameters are corrected according to the feedback parameter to obtain the debugging correction parameters. According to the debugging correction parameters, the initial spatial parameters and the simulation model, a corrected model debugging image is obtained. Based on the registration network, the second residual parameters of the real-time image and the corrected model debugging image are calculated until the second residual parameters converge, and the rotation parameters are determined according to the latest debugging correction parameters and the initial spatial parameters.
[0047] It should be noted that after each registration, the registration network analyzes the registration accuracy according to the second residual parameters before and after the registration. In the first registration, the first residual parameter of the initial registration can be used as the first second residual parameter.
[0048] By comparing the first residual parameter and the second residual parameter, the trend of residual change after each registration can be obtained. If the residual transformation trend shows an increase in residual in a certain direction, it means that the direction needs to be rotated and called back during registration. If the residual transformation trend shows a decrease in residual in a certain direction, it means that it is the correct registration direction. The second residual parameter may also contain residuals of edge corner points that do not exist in the first residual parameter. For example, after rotating the preset angle, the top tooth gap of the teeth appears in the model debugging image, which may be reflected in the second residual parameter. Similarly, the second residual parameter may also lose the residuals of edge corner points in the first residual parameter. These can all be used to know the registration network for re-registration to improve the registration accuracy and speed of the registration network.
[0049] In some embodiments, based on the registration network, a first residual parameter and a second residual parameter are compared and analyzed to obtain a residual change parameter, including: constructing a first residual matrix according to the first residual parameter, and constructing a second residual matrix according to the second residual parameter; performing parameter value detection on the first residual matrix and the second residual matrix respectively, determining the first parameter items that are not zero in the first residual matrix, and determining the second parameter items that are not zero in the second residual matrix; subtracting the second residual matrix from the first residual matrix to obtain a residual change matrix, the residual change matrix including: common terms, incremental terms and decrement terms, the common terms are terms included in both the first parameter terms and the second parameter terms; the incremental terms are terms included in the second parameter terms but not included in the first parameter terms; the decrement terms are terms included in the first parameter terms but not included in the second parameter terms; and the residual change matrix is used as the residual change parameter.
[0050] Exemplarily, the first residual parameter is for the real-time image and the target cross-sectional image, and the second residual parameter is for the real-time image and the model debugging image, both of which are obtained by comparing the residual network on the basis of the edge corners of the real-time image. Therefore, after matrixing, the format size of the two matrices is the same and can be used directly for comparison. The registration network identifies which residuals are added to the second residual parameter before and after the registration, that is, determines the incremental terms, and which residuals are lost in the second residual parameter, that is, the decrement terms, and finally identifies which changes have occurred in the common residuals between the two, that is, the common terms. The common terms are determined before matrix subtraction, and the incremental terms and decrement terms are determined during matrix subtraction.
[0051] Through the above mechanism, the registration accuracy and speed of the registration network are improved based on matrix residual auxiliary registration.
[0052] In some embodiments, a residual change trend is determined according to a residual change parameter, and a feedback parameter is determined according to the residual change trend, including: performing numerical comparison on common items to determine the direction of numerical change in the common items; performing density distribution evaluation on incremental items to obtain a first density change direction of the incremental items; performing density distribution evaluation on decremental items to obtain a second density change direction of the decremental items; determining a residual change trend according to the numerical change direction, the first density change direction, and the second density change direction; and determining a feedback coefficient according to a preset iterative algorithm and the residual change trend.
[0053] Exemplarily, in the registration process, both the incremental term and the decrement term need to be suppressed. The greater the density of the two, the greater the degree of deviation from the real-time image. The first density change direction is: the direction of increase of the unit density of the incremental term. The numerical change of the common term will also have a tendency to increase or decrease, which reflects the change in the similarity of the edge corner points before and after the registration. Therefore, the numerical change direction includes the direction of numerical increase and the direction of numerical decrease. The second density change direction is: the direction of increase of the unit density of the decrement term. In the above process, the size of the density change value and the numerical change value can be used to determine the size of the feedback coefficient, and the numerical change direction, the first density change direction and the second density change direction can determine the adjustment angle of the feedback coefficient. In this way, the single registration accuracy of the registration network is improved, the total number of registrations of the registration network is reduced, and the registration efficiency is improved.
[0054] In some embodiments, after the debugging parameters are corrected according to the second residual parameters based on the registration network to obtain the debugging correction parameters, the method also includes: based on the registration network, obtaining a model debugging image according to the debugging correction parameters, initial spatial parameters and a simulation model until the second residual parameters converge.
[0055] Exemplarily, the registration network needs to go through multiple registrations before successful registration can be achieved. The sign of successful registration is: the second residual parameter converges, for example, the standard value of the second residual parameter is less than the first preset threshold. Before the second residual parameter converges, the following steps are executed in a loop: the residual network calculates the second residual parameter according to the model debugging image, the registration network obtains the debugging correction parameter according to the second residual parameter, and the model debugging image is obtained according to the debugging correction parameter, the initial space parameter and the simulation model.
[0056] It should be noted that if the second residual parameter is suddenly detected to be greater than the second preset threshold value in a stable state, the camera may have a large angle change. At this time, the registration network needs to replace the initial registration image used for registration, that is, jump to S103, so that during the operation, when the camera changes the camera angle, no manual intervention registration is required.
[0057] S106. Determine the observation image of the root model on the preset projection plane according to the rotation parameters, and superimpose the observation image on the real-time image to generate a medical navigation image.
[0058] Exemplarily, after the observed image is processed, a pseudo-color image is obtained, which is superimposed according to the corresponding relationship, and the pseudo-color image is superimposed on the real-time image to generate a medical navigation image, so that medical staff can observe the direction of the tooth root and assist in the operation.
[0059] In some embodiments, determining an observed image of the root model on a preset projection plane according to the rotation parameters includes: adjusting a spatial relationship between the root model and the preset projection plane according to the rotation parameters; and acquiring an observed image of the root model from the preset projection plane.
[0060] Exemplarily, since the real-time image is a two-dimensional image, the required observation image should also be two-dimensional. The preset projection plane can be a yz plane in the xyz three-axis coordinate system, and the rotation parameter is the spatial coordinate of the simulation model in the xyz three-axis coordinate system.
[0061] The embodiment of the present application provides a medical navigation method based on CBCT, which is applied to a terminal device. The terminal device is configured with a pre-trained image registration model, and the image registration model includes a residual network and a registration network. The method includes: obtaining a real-time image generated by photographing teeth with a camera, and obtaining a simulation model generated by scanning teeth with a CBCT instrument, the simulation model including: a crown model and a root model; determining a model cross-sectional image of a crown model according to multiple preset spatial parameters; based on the residual network, extracting features from the real-time image to obtain a first feature parameter, and extracting features from each model cross-sectional image to obtain multiple second feature parameters, and calculating the first feature parameter. The method comprises the following steps: calculating the first residual parameter of each second feature parameter of the real-time image and the model debugging image based on the residual network, and correcting the debugging parameter based on the second residual parameter to obtain the debugging correction parameter until the second residual parameter converges, and obtaining the rotation parameter based on the debugging correction parameter and the initial spatial parameter; determining the observation image of the root model on the preset projection plane based on the rotation parameter, and superimposing the observation image on the real-time image to generate a medical navigation image. Through the above method, the model cross-sectional images of the simulation model are first obtained at multiple angles based on multiple spatial parameters, and then the real-time image and the model cross-sectional image are compared through the residual network to obtain the model cross-sectional image closest to the real-time image, so as to determine the initial spatial parameters of the registration network. No manual initial alignment is required. The registration network successively aligns the simulation model and the real-time image according to the debugging parameters on the basis of the initial spatial parameters. During the debugging process, the registration network corrects the debugging parameters according to the second residual parameters of the residual network after debugging to ensure the registration accuracy and speed. Finally, the closest rotation parameters are obtained, and the navigation image of the tooth root of the real-time image is obtained according to the rotation parameters and the root model, which realizes following the changes of the real-time image and automatically generating the correct navigation image, and does not require manual intervention in alignment.
[0062] See also Figure 2 , Figure 2 The embodiment of the present application also provides a schematic block diagram of a CBCT-based medical navigation device, wherein the CBCT-based medical navigation device 200 is configured with a pre-trained image registration model, the image registration model includes a residual network and a registration network, and is used to execute the aforementioned CBCT-based medical navigation method. The CBCT-based medical navigation device 200 can be configured in a server.
[0063] Among them, the server can be an independent server or a server cluster, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN) and big data and artificial intelligence platforms.
[0064] like Figure 2 As shown, the CBCT-based medical navigation device 200 includes: a data reading module 201 , an image determination module 202 , a parameter calculation module 203 , a parameter debugging module 204 , a result calculation module 205 and an image display module 206 .
[0065] The data reading module 201 is used to obtain a real-time image of teeth captured by a camera, and to obtain a simulation model of teeth scanned by a CBCT instrument. The simulation model includes a crown model and a root model.
[0066] The image determination module 202 is used to determine a model cross-sectional image of a crown model according to a plurality of preset spatial parameters.
[0067] The parameter calculation module 203 is used to perform feature extraction on the real-time image based on the residual network to obtain the first feature parameter, and to perform feature extraction on each model cross-sectional image to obtain multiple second feature parameters, and calculate the first residual parameter between the first feature parameter and each second feature parameter.
[0068] The parameter debugging module 204 is used to determine the initial spatial parameters from the preset spatial parameters according to the first residual parameters based on the registration network, and obtain the model debugging image according to the preset debugging parameters, the initial spatial parameters and the simulation model.
[0069] In some embodiments, when the parameter debugging module 204 is used to determine the initial spatial parameters from the preset spatial parameters according to the first residual parameters, it is specifically used to implement: calculating the residual standard deviation of each first residual parameter; setting the first residual parameter corresponding to the largest residual standard deviation as the target residual parameter; setting the model section image corresponding to the target residual parameter as the target section image; setting the preset spatial parameters corresponding to the target section image as the initial spatial parameters.
[0070] The result calculation module 205 is used to calculate the second residual parameters of the real-time image and the model debugging image based on the residual network, and to correct the debugging parameters according to the second residual parameters based on the registration network to obtain the debugging correction parameters until the second residual parameters converge, and to obtain the rotation parameters according to the debugging correction parameters and the initial space parameters.
[0071] In some embodiments, the result calculation module 205 is used to implement, based on the registration network, correcting the debugging parameters according to the second residual parameters to obtain the debugging correction parameters, until the second residual parameters converge, and obtaining the rotation parameters according to the debugging correction parameters and the initial spatial parameters. Specifically, it is used to implement: based on the registration network, comparing and analyzing the first residual parameters and the second residual parameters to obtain the residual change parameters; determining the residual change trend according to the residual change parameters, determining the feedback parameters according to the residual change trend, correcting the debugging parameters according to the feedback parameters to obtain the debugging correction parameters; obtaining the corrected model debugging image according to the debugging correction parameters, the initial spatial parameters and the simulation model; based on the registration network, calculating the second residual parameters of the real-time image and the corrected model debugging image, until the second residual parameters converge, and determining the rotation parameters according to the latest debugging correction parameters and the initial spatial parameters.
[0072] In some embodiments, when the result calculation module 205 is used to implement a comparative analysis of the first residual parameter and the second residual parameter based on the registration network to obtain the residual change parameter, it is specifically used to implement: constructing a first residual matrix according to the first residual parameter, and constructing a second residual matrix according to the second residual parameter; performing parameter value detection on the first residual matrix and the second residual matrix respectively, determining the first parameter item that is not zero in the first residual matrix, and determining the second parameter item that is not zero in the second residual matrix; subtracting the second residual matrix from the first residual matrix to obtain a residual change matrix, the residual change matrix including: common terms, incremental terms and decrement terms, the common terms are terms included in both the first parameter item and the second parameter item; the incremental terms are terms included in the second parameter item but not included in the first parameter item; the decrement terms are terms included in the first parameter item but not included in the second parameter item; and using the residual change matrix as the residual change parameter.
[0073] In some embodiments, the result calculation module 205, when used to determine the residual change trend according to the residual change parameter and determine the feedback parameter according to the residual change trend, is specifically used to implement: performing numerical comparison on common items to determine the direction of numerical change in the common items; performing density distribution evaluation on incremental items to obtain a first density change direction of the incremental items; performing density distribution evaluation on decremental items to obtain a second density change direction of the decremental items; determining the residual change trend according to the numerical change direction, the first density change direction and the second density change direction; and determining the feedback coefficient according to a preset iterative algorithm and the residual change trend.
[0074] In some embodiments, the result calculation module 205 is used to implement, based on the registration network, to correct the debugging parameters according to the second residual parameters to obtain the debugging correction parameters, and is also specifically used to implement: based on the registration network, according to the debugging correction parameters, the initial spatial parameters and the simulation model, to obtain a model debugging image until the second residual parameters converge.
[0075] The image display module 206 is used to determine the observation image of the root model on the preset projection plane according to the rotation parameters, and to superimpose the observation image on the real-time image to generate a medical navigation image.
[0076] In some embodiments, when the image display module 206 is used to determine the observation image of the root model on a preset projection plane according to the rotation parameters, it is specifically used to implement: adjusting the spatial relationship between the root model and the preset projection plane according to the rotation parameters; and obtaining the observation image of the root model from the preset projection plane.
[0077] An embodiment of the present application provides a medical device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement a CBCT-based medical navigation method as any one of the embodiments of the present application when executing the computer program.
[0078] An embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements any CBCT-based medical navigation method as described in the embodiments of the present application.
[0079] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.
Claims
1. A medical navigation method based on CBCT, characterized in that: Applied to a terminal device, the terminal device is configured with a pre-trained image registration model, the image registration model includes a residual network and a registration network, and the method includes: Acquire a real-time image of teeth captured by a camera, and acquire a simulation model of teeth generated by a CBCT instrument scanning the teeth, wherein the simulation model includes: a crown model and a root model; Determine a model cross-sectional image of the crown model according to a plurality of preset spatial parameters; Based on the residual network, feature extraction is performed on the real-time image to obtain a first feature parameter, and feature extraction is performed on each of the model cross-sectional images to obtain multiple second feature parameters, and a first residual parameter between the first feature parameter and each of the second feature parameters is calculated; Based on the registration network, determining initial spatial parameters from the preset spatial parameters according to the first residual parameters, and obtaining a model debugging image according to preset debugging parameters, the initial spatial parameters and the simulation model; Based on the residual network, second residual parameters of the real-time image and the model debugging image are calculated; based on the registration network, the debugging parameters are corrected according to the second residual parameters to obtain debugging correction parameters, until the second residual parameters converge, and rotation parameters are obtained according to the debugging correction parameters and the initial space parameters; According to the rotation parameters, an observation image of the root model on a preset projection plane is determined, and the observation image is superimposed on the real-time image to generate a medical navigation image.
2. The CBCT-based medical navigation method according to claim 1, characterized in that: The method of correcting the debugging parameters according to the second residual parameters based on the registration network to obtain debugging correction parameters until the second residual parameters converge, and obtaining rotation parameters according to the debugging correction parameters and the initial space parameters, includes: Based on the registration network, comparative analysis is performed on the first residual parameter and the second residual parameter to obtain a residual change parameter; Determine a residual variation trend according to the residual variation parameter, determine a feedback parameter according to the residual variation trend, and correct the debugging parameter according to the feedback parameter to obtain a debugging correction parameter; Obtaining a corrected model debugging image according to the debugging correction parameters, the initial space parameters and the simulation model; Based on the registration network, second residual parameters of the real-time image and the corrected model debugging image are calculated until the second residual parameters converge, and the rotation parameters are determined according to the latest debugging correction parameters and the initial space parameters.
3. The CBCT-based medical navigation method according to claim 2, characterized in that: Based on the registration network, the first residual parameter and the second residual parameter are compared and analyzed to obtain a residual change parameter, including: constructing a first residual matrix according to the first residual parameters, and constructing a second residual matrix according to the second residual parameters; Performing parameter value detection on the first residual matrix and the second residual matrix respectively, determining a first parameter item that is not zero in the first residual matrix, and determining a second parameter item that is not zero in the second residual matrix; Subtracting the second residual matrix from the first residual matrix to obtain a residual change matrix, wherein the residual change matrix includes: common items, incremental items, and decrement items, wherein the common items are items included in both the first parameter item and the second parameter item; the incremental items are items included in the second parameter item but not included in the first parameter item; and the decrement items are items included in the first parameter item but not included in the second parameter item; The residual change matrix is used as the residual change parameter.
4. The CBCT-based medical navigation method according to claim 3, characterized in that: The determining of the residual variation trend according to the residual variation parameter, and determining the feedback parameter according to the residual variation trend, comprises: Comparing the values of the common items to determine the direction of value change in the common items; Performing density distribution evaluation on the incremental item to obtain a first density change direction of the incremental item; Performing density distribution evaluation on the decrement item to obtain a second density change direction of the decrement item; Determine the residual variation trend according to the value variation direction, the first density variation direction, and the second density variation direction; The feedback coefficient is determined according to a preset iterative algorithm and the residual variation trend.
5. The CBCT-based medical navigation method according to claim 1, characterized in that: After the debugging parameter is corrected based on the registration network according to the second residual parameter to obtain the debugging correction parameter, the method further includes: Based on the registration network, a model debugging image is obtained according to the debugging correction parameters, the initial space parameters and the simulation model until the second residual parameters converge.
6. The CBCT-based medical navigation method according to claim 1, characterized in that: Determining the observation image of the root model on a preset projection plane according to the rotation parameter includes: Adjusting the spatial relationship between the root model and the preset projection plane according to the rotation parameter; The observation image of the root model is acquired from a preset projection plane.
7. The CBCT-based medical navigation method according to claim 1, characterized in that: The determining the initial space parameter from the preset space parameter according to the first residual parameter includes: Calculate the residual standard deviation of each of the first residual parameters; Setting the first residual parameter corresponding to the largest residual standard deviation as a target residual parameter; Setting the model cross-sectional image corresponding to the target residual parameter as a target cross-sectional image; The preset spatial parameters corresponding to the target cross-sectional image are set as the initial spatial parameters.
8. A medical navigation device based on CBCT, characterized in that: Applied to a terminal device, the terminal device is configured with a pre-trained image registration model, the image registration model includes a residual network and a registration network, and the CBCT-based medical navigation device includes: A data reading module is used to obtain a real-time image of teeth captured by a camera, and to obtain a simulation model of teeth scanned by a CBCT instrument, wherein the simulation model includes a crown model and a root model; An image determination module, used to determine a model cross-sectional image of the crown model according to a plurality of preset spatial parameters; A parameter calculation module, configured to perform feature extraction on the real-time image based on the residual network to obtain a first feature parameter, perform feature extraction on each of the model cross-sectional images to obtain a plurality of second feature parameters, and calculate a first residual parameter between the first feature parameter and each of the second feature parameters; A parameter debugging module, configured to determine initial spatial parameters from the preset spatial parameters according to the first residual parameters based on the registration network, and obtain a model debugging image according to preset debugging parameters, the initial spatial parameters and the simulation model; A result calculation module is used to calculate the second residual parameters of the real-time image and the model debugging image based on the residual network, and to correct the debugging parameters according to the second residual parameters based on the registration network to obtain debugging correction parameters until the second residual parameters converge, and to obtain rotation parameters according to the debugging correction parameters and the initial space parameters; The image display module is used to determine the observation image of the root model on a preset projection plane according to the rotation parameters, and to superimpose the observation image on the real-time image to generate a medical navigation image.
9. A medical device, characterized in that: The medical device comprises a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the CBCT-based medical navigation method according to any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements the CBCT-based medical navigation method according to any one of claims 1 to 7.
