Children neurosurgery auxiliary guiding method and system based on machine vision

Abstract

The invention discloses a child neurosurgery auxiliary guiding method and system based on machine vision, and relates to the technical field of visual guidance. Neurosurgery image data at different angles are collected in real time and summarized, and meanwhile the image data collected in real time are processed in an image processing mode; after the processing is completed, constructing a three-dimensional model in real time based on the processed image data, and registering the three-dimensional model constructed in real time; after registration is completed, the operation path is planned in a three-dimensional tracking mode, prediction auxiliary guiding is carried out in a digital twinborn mode based on the planned operation path, and the accuracy of operation auxiliary guiding is improved.

Description

Technical Field

[0001] This invention relates to the field of visual guidance technology, and more specifically to a method and system for assisting in pediatric neurosurgery based on machine vision. Background Technology

[0002] Traditional surgery relies heavily on the surgeon's anatomical knowledge and clinical experience. Furthermore, the need for the surgeon to switch between the patient's and the monitor screen during the procedure leads to longer surgery times and reduced surgical precision.

[0003] The prior art, such as the invention patent application with publication number CN118710708A, discloses a clinical anesthesia ultrasound image-assisted positioning and guidance method and system. The method includes: acquiring the needle midline and enhancement boundary distance in a lumbar spine ultrasound image; obtaining the main edge pixels from the lumbar spine ultrasound image; obtaining the lamina edge probability based on the grayscale values ​​of the main edge pixels and the pixels; obtaining the suspected lamina region based on the correlation value and the lamina edge probability; obtaining the relative position angle and lamina-related angle based on the suspected lamina region; obtaining the grayscale enhancement coefficient based on the relative position angle, lamina-related angle, and extension trend; and obtaining an ultrasound-enhanced image based on the grayscale enhancement coefficient, the needle midline, and the enhancement boundary distance, and performing assisted positioning and guidance.

[0004] As can be seen from the above solutions, current medical assistance systems often focus on the analysis of two-dimensional images and provide assistance and guidance based on the analysis results of two-dimensional images. They lack the overall analysis and judgment of the patient and have certain limitations. Summary of the Invention

[0005] The purpose of this invention is to provide a machine vision-based method and system for assisting and guiding pediatric neurosurgery, which solves the problems existing in the background art.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a machine vision-based method for assisted guidance in pediatric neurosurgery, specifically including the following steps: S1. Install a laser camera to collect neurosurgical images from different angles in real time and compile them to build an image training dataset. The neurosurgical image data includes: nerve images and images of the surgical procedure; S2. Based on real-time acquired neurosurgical image data, the real-time acquired image data is processed using image processing methods to obtain processed image data, including the following steps: S21. Image enhancement is performed on the real-time acquired image data to obtain enhanced image data; S22. The enhanced image data is fused using an image fusion method to obtain fused image data; S23. The fused image data is identified using a data recognition method to obtain the processed image data; S3. Construct a 3D model in real time based on the processed image data, and register the constructed 3D model to obtain the registered 3D model. S4. Based on the registered 3D model, surgical path planning is performed using 3D tracking, including the following steps: S41. Divide the registered 3D model into segments using a model cutting method to obtain the segmented 3D model. S42. Based on the partitioned 3D model, perform localization analysis using collision detection methods and output the localization results; S43. Surgical path planning based on positioning results; S5. Based on the planned surgical path, predictive assistance and guidance are provided through digital twins.

[0007] Preferably, the installation of the laser camera to acquire neurosurgical images from different angles in real time and to compile and construct an image training dataset includes the following steps: Multiple reference points are set within the neurosurgical surgical area, and laser cameras are evenly installed around the neurosurgical surgical area at set distances and angles, so that the reference points within the neurosurgical surgical area are evenly distributed throughout the camera's field of view. A three-dimensional coordinate system is constructed based on the installed laser cameras and the center point of the neurosurgical area, and the three-dimensional coordinates of each laser camera are recorded in real time. The image acquisition interval of the laser camera is set, and image data of the neurosurgical area is acquired in real time according to the set acquisition interval. At the same time, the image data of the neurosurgical area acquired in real time by each laser camera are summarized to construct an image training dataset.

[0008] Preferably, the step of enhancing the real-time acquired image data to obtain enhanced image data includes the following steps: Calculate pixel probability distribution based on grayscale value distribution in real-time acquired image data; The formula for pixel probability distribution is as follows: ; in, Represents the pixel probability of image data. Indicates grayscale value The number of pixels, This represents the total number of pixels. Construct the cumulative distribution function of the image based on the pixel probability distribution of the image data; The formula for constructing the cumulative distribution function of an image is shown below: ; Where L represents the total number of grayscale values ​​in the image. Indicates the first The gray values ​​are mapped by the cumulative distribution function; Image enhancement is performed on image data based on the constructed image cumulative distribution function; The gray values ​​mapped by the cumulative distribution function are multiplied by the total number of gray values ​​using a transformation function. Based on the mapping relationship between the cumulative distribution function and the transformation function of the image, and referring to the pixels in the original image, the histogram equalized image is output. The image after histogram equalization is defined as the enhanced image data.

[0009] Preferably, the step of identifying the fused image data using a data recognition method to obtain processed image data includes the following steps: Collect various types of historical neural image data, and input the collected historical neural image data and the fused image data into a convolutional neural network for feature extraction; The feature extraction formula for convolutional networks is shown below: ; in, This represents the extracted image features. This represents the input image data. This represents the weights for feature extraction in a convolutional neural network. Indicates the bias value; The extracted features from historical neural image data and the features from the fused image data are compared. The feature comparison formula is as follows: ; in, Representing features of historical neural image data and image data features Similarity values ​​between them; A similarity threshold is set, and the fused image data is classified based on the set threshold. Image data that meet the similarity threshold are classified into one category, resulting in classified image data. The classified image data is defined as the processed image data.

[0010] Preferably, the real-time construction of a 3D model based on the processed image data, and the registration of the real-time constructed 3D model to obtain the registered 3D model, includes the following steps: A three-dimensional model is constructed in real time based on a set coordinate system, processed image data, and reference points within the neurosurgical surgical area. Based on the real-time constructed 3D model, a point cloud dataset is obtained by summarizing reference points within the neurosurgical surgical area; The aggregated point cloud dataset is registered by calculating the mean and covariance of the point cloud data. Reconstructing a 3D model based on registered point cloud data; set up The reconstructed 3D model, The registered point cloud data; in, These represent the horizontal and vertical coordinates in the registered point cloud data, respectively. This indicates the camera's axis distance parameter; The relationship between the reconstructed 3D model and the registered point cloud data is shown below: ; in, express Axis camera wheelbase, express Axis camera wheelbase, These represent the offset between the pixel plane and the camera imaging plane, respectively. This represents the rotation vector of the camera in the 3D model. This represents the translation vector of the camera in the 3D model; The reconstructed 3D model is set as the registered 3D model.

[0011] Preferably, the step of dividing the registered 3D model into subdivided 3D models by model cutting includes the following steps: Select a vertex on the 3D model outline as the initial point; Based on the selected initial point, the coordinates of the reference points in each plane of the model are traversed sequentially, and a linked list is set up for numbering. If two reference points with the same coordinates are detected in two planes during the traversal, the two planes are set as adjacent planes. Based on the obtained numbered 3D model, slices are made by setting the height and determining the contour of the cutting plane; Set the slice thickness, and according to the set thickness, slice the 3D model evenly from the bottom upwards with a transverse cutting plane perpendicular to the Z-axis. Each slice yields a hyperplane, and the location of the affected area is determined based on the obtained hyperplane. Construct a classification decision function and determine the location of the affected area based on the constructed classification decision function; ; in, For weights, For classification threshold, Represents the classification decision function; The classification decision function is constructed to traverse each pixel in the hyperplane, and the hyperplane is divided into affected areas and non-affected areas based on the traversal. After the division is completed, the affected area is located based on the position of the hyperplane reference point.

[0012] Preferably, the localization analysis based on the partitioned 3D model using a collision detection method, and the output of the localization results, includes the following steps: Images of the surgical process are collected and processed through steps 2 and 3 to obtain a three-dimensional model of the surgical process images. At the same time, the hyperplane of the three-dimensional model of the surgical process images is obtained through model cutting. Collision detection is performed on hyperplanes based on the 3D model of surgical process images and the 3D model of neural images. When the two sets of hyperplanes intersect, it indicates that the two sets of hyperplanes are colliding. The location of the surgical procedure is determined based on the collision results.

[0013] Preferably, the surgical path planning based on the positioning results includes the following steps: The three-dimensional model of the neural image is rasterized, and the surgical path is planned using the A* algorithm based on the localization results and the coordinates of the affected area. The actual path distance and the estimated path distance to the affected area are determined based on the actual path distance of each unit coordinate in the grid space and the length of the path in the grid space. If there exists a path whose estimated path distance is less than or equal to the actual path distance from the current location to the affected area, then this path is designated as the optimal path.

[0014] Preferably, the predictive and assisted guidance based on the planned surgical path using a digital twin includes the following steps: The surgical path, planned based on the current location and the location of the affected area, will be separated into multiple smaller target nodes. Each time a small target node is reached, the current target node is set as the current node, and the next small target node is set as an auxiliary guiding node, continuously iterating to assist doctors in surgical planning.

[0015] The present invention also discloses a machine vision-based pediatric neurosurgical surgical assistance guidance system, which is used to implement a machine vision-based pediatric neurosurgical surgical assistance guidance method. The system includes: a data acquisition module, a data processing module, a model building module, a path planning module, and a prediction-assisted guidance module. The data acquisition module is used to install a laser camera to acquire and summarize neurosurgical image data from different angles in real time. The data processing module is used to process the acquired image data through data processing methods; The model building module is used to perform 3D modeling and registration based on the processed image data; The path planning module is used to segment the constructed 3D model and calculate the planned path; The prediction-assisted guidance module is used to provide assisted guidance based on the path planning results.

[0016] The beneficial effects of this invention are as follows: (1) This invention collects and summarizes neurosurgical image data from different angles in real time, and processes the real-time collected image data through image processing. After processing, a three-dimensional model is constructed in real time based on the processed image data, and the real-time constructed three-dimensional model is registered. After registration, the surgical path is planned through three-dimensional tracking, and the surgical path is predicted and assisted by digital twin based on the planned surgical path, which improves the accuracy of surgical assisted guidance.

[0017] (2) This invention enhances the image data of real-time acquired neurosurgical images, and after the enhancement is completed, the enhanced image data is fused by data fusion. After the fusion is completed, historical image data is acquired, and features are extracted from the acquired historical image data and the fused image data. Recognition is performed by feature comparison, which improves the reliability of image processing.

[0018] (3) The present invention constructs a three-dimensional model in real time and obtains a point cloud dataset by summarizing reference points in the neurosurgical surgical area; at the same time, it reconstructs the three-dimensional model by registering the summarized point cloud dataset, thereby improving the accuracy of the three-dimensional model construction.

[0019] (4) The present invention completes the planning of surgical path by dividing, locating and planning the constructed three-dimensional model, thereby improving the real-time performance and efficiency of surgical path planning. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1This is a schematic diagram of the pediatric neurosurgical surgical assistance and guidance method of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] In a specific embodiment of the present invention, Reference Figure 1 As shown, this invention provides a machine vision-based method for assisting and guiding pediatric neurosurgery, comprising the following steps: S1. Install a laser camera to collect neurosurgical images from different angles in real time and compile them to build an image training dataset. The neurosurgical image data includes: nerve images and images of the surgical procedure; S2. Based on real-time acquired neurosurgical image data, the real-time acquired image data is processed using image processing methods to obtain processed image data, including the following steps: S21. Image enhancement is performed on the real-time acquired image data to obtain enhanced image data; S22. The enhanced image data is fused using an image fusion method to obtain fused image data; S23. The fused image data is identified using a data recognition method to obtain the processed image data; S3. Construct a 3D model in real time based on the processed image data, and register the constructed 3D model to obtain the registered 3D model. S4. Based on the registered 3D model, surgical path planning is performed using 3D tracking, including the following steps: S41. Divide the registered 3D model into segments using a model cutting method to obtain the segmented 3D model. S42. Based on the partitioned 3D model, perform localization analysis using collision detection methods and output the localization results; S43. Surgical path planning based on positioning results; S5. Based on the planned surgical path, predictive assistance and guidance are provided through digital twins; Furthermore, referring to Figure 1 As shown, installing a laser camera to acquire real-time neurosurgical images from different angles and compiling them into an image training dataset includes the following steps: Multiple reference points are set within the neurosurgical surgical area, and laser cameras are evenly installed around the neurosurgical surgical area at set distances and angles, so that the reference points within the neurosurgical surgical area are evenly distributed throughout the camera's field of view. Furthermore, a three-dimensional coordinate system is constructed based on the installed laser cameras and the center point of the neurosurgical area, and the three-dimensional coordinates of each laser camera are recorded in real time. Furthermore, the image acquisition interval of the laser camera is set, and image data of the neurosurgical area is acquired in real time according to the set acquisition interval. At the same time, the image data of the neurosurgical area acquired in real time by each laser camera is summarized to construct an image training dataset. Furthermore, referring to Figure 1 As shown, image enhancement is performed on real-time acquired image data to obtain enhanced image data, including the following steps: Calculate pixel probability distribution based on grayscale value distribution in real-time acquired image data; The formula for pixel probability distribution is as follows: ; in, Represents the pixel probability of image data. Indicates grayscale value The number of pixels, This represents the total number of pixels. Construct the cumulative distribution function of the image based on the pixel probability distribution of the image data; The formula for constructing the cumulative distribution function of an image is shown below: ; Where L represents the total number of grayscale values ​​in the image. Indicates the first The gray values ​​are mapped by the cumulative distribution function; Image enhancement is performed on image data based on the constructed image cumulative distribution function; The gray values ​​mapped by the cumulative distribution function are multiplied by the total number of gray values ​​using a transformation function. Furthermore, based on the mapping relationship between the cumulative distribution function and the transformation function of the image, and referring to the pixels in the original image, the image after histogram equalization is output; The image after histogram equalization is defined as the enhanced image data; Furthermore, referring to Figure 1 As shown, the enhanced image data is fused using an image fusion method to obtain the fused image data, which includes the following steps: Two adjacent sets of enhanced image data are selected, and image fusion is performed based on the positions of each reference point in the enhanced image data. The image fusion formula is shown below: ; in, This represents the fused image data. This represents the first set of enhanced image data. The fusion weight is the ratio of the gray values ​​at corresponding positions in two adjacent enhanced image data sets. Furthermore, referring to Figure 1 As shown, the process of identifying the fused image data using a data recognition method to obtain processed image data includes the following steps: Collect various types of historical neural image data, and input the collected historical neural image data and the fused image data into a convolutional neural network for feature extraction; The feature extraction formula for convolutional networks is shown below: ; in, This represents the extracted image features. This represents the input image data. This represents the weights for feature extraction in a convolutional neural network. Indicates the bias value; Furthermore, the extracted historical neural image data features and the fused image data features are compared. The feature comparison formula is as follows: ; in, Representing features of historical neural image data and image data features Similarity values ​​between them; Furthermore, a similarity threshold is set, and the fused image data is classified based on the set threshold. Image data that meet the similarity threshold are classified into one category to obtain the classified image data. The classified image data is defined as the processed image data; Furthermore, referring to Figure 1 As shown, a 3D model is constructed in real time based on the processed image data, and the constructed 3D model is registered to obtain the registered 3D model. The process includes the following steps: A three-dimensional model is constructed in real time based on a set coordinate system, processed image data, and reference points within the neurosurgical surgical area. Furthermore, based on the real-time constructed 3D model, a point cloud dataset is obtained by summarizing reference points within the neurosurgical surgical area; Furthermore, the aggregated point cloud dataset is registered by calculating the mean and covariance of the point cloud data; Furthermore, a 3D model is reconstructed based on the registered point cloud data; set up The reconstructed 3D model, The registered point cloud data; in, These represent the horizontal and vertical coordinates in the registered point cloud data, respectively. This indicates the camera's axis distance parameter; The relationship between the reconstructed 3D model and the registered point cloud data is shown below: ; in, express Axis camera wheelbase, express Axis camera wheelbase, These represent the offset between the pixel plane and the camera imaging plane, respectively. This represents the rotation vector of the camera in the 3D model. This represents the translation vector of the camera in the 3D model; Furthermore, the reconstructed 3D model is set as the registered 3D model; Furthermore, referring to Figure 1 As shown, the registered 3D model is divided into subdivisions using a model cutting method to obtain the subdivided 3D model, which includes the following steps: Select a vertex on the 3D model outline as the initial point; Based on the selected initial point, the coordinates of the reference points in each plane of the model are traversed sequentially, and a linked list is set up for numbering. Furthermore, when two reference points with the same coordinates are detected in two planes during the traversal, the two planes are set as adjacent planes; Furthermore, based on the obtained numbered 3D model, slicing is performed by setting the height, and the contour of the cutting plane is determined; Set the slice thickness, and according to the set thickness, slice the 3D model evenly from the bottom upwards with a transverse cutting plane perpendicular to the Z-axis. Each slice yields a hyperplane, and the location of the affected area is determined based on the obtained hyperplane. Construct a classification decision function and determine the location of the affected area based on the constructed classification decision function; ; in, For weights, For classification threshold, Represents the classification decision function; Furthermore, the hyperplane is traversed through each pixel by the constructed classification decision function, and the hyperplane is divided into affected and non-affected areas based on the traversal. After the division is completed, the affected area is located based on the position of the hyperplane reference point; Furthermore, referring to Figure 1 As shown, the localization analysis based on the partitioned 3D model using collision detection methods, and the output of the localization results, includes the following steps: Images of the surgical process are collected and processed through steps 2 and 3 to obtain a three-dimensional model of the surgical process images. At the same time, the hyperplane of the three-dimensional model of the surgical process images is obtained through model cutting. Furthermore, collision detection is performed based on the hyperplane of the three-dimensional model of the surgical process image and the hyperplane of the three-dimensional model of the neural image. When the two sets of hyperplanes intersect, it indicates that the two sets of hyperplanes are colliding. Furthermore, the location of the surgical procedure is determined based on the collision results; Furthermore, referring to Figure 1 As shown, surgical path planning based on the localization results includes the following steps: The three-dimensional model of the neural image is rasterized, and the surgical path is planned using the A* algorithm based on the localization results and the coordinates of the affected area. Furthermore, based on the actual path distance of each unit coordinate in the grid space and the length of the path in the grid space, the actual path distance and the estimated path distance to the current location of the affected area are determined; If there exists a path whose estimated path distance is less than or equal to the actual path distance from the current location to the affected area, then this path is designated as the optimal path. Furthermore, referring to Figure 1 As shown, predictive guidance based on the planned surgical path using digital twins includes the following steps: The surgical path, planned based on the current location and the location of the affected area, will be separated into multiple smaller target nodes. Furthermore, each time a small target node is reached, the current target node is taken as the current node, and the next small target node is taken as the auxiliary guidance node, continuously iterating to assist doctors in surgical planning; In one specific embodiment, the machine vision-based pediatric neurosurgical surgical assistance guidance system is used to implement a machine vision-based pediatric neurosurgical surgical assistance guidance method. The system includes: a data acquisition module, a data processing module, a model building module, a path planning module, and a prediction-assisted guidance module. The data acquisition module is used to install a laser camera to acquire and summarize neurosurgical image data from different angles in real time. The data processing module is used to process the acquired image data through data processing methods; The model building module is used to perform 3D modeling and registration based on the processed image data; The path planning module is used to segment the constructed 3D model and calculate the planned path; The prediction-assisted guidance module is used to provide assisted guidance based on the path planning results.

[0024] It should be noted that The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A machine vision-based method for assisted guidance in pediatric neurosurgery, characterized in that, Includes the following steps: S1. Install a laser camera to collect neurosurgical images from different angles in real time and compile them to build an image training dataset. The neurosurgical image data includes: nerve images and images of the surgical procedure; S2. Based on real-time acquired neurosurgical image data, the real-time acquired image data is processed using image processing methods to obtain processed image data, including the following steps: S21. Image enhancement is performed on the real-time acquired image data to obtain enhanced image data; S22. The enhanced image data is fused using an image fusion method to obtain fused image data; S23. The fused image data is identified using a data recognition method to obtain the processed image data; S3. Construct a 3D model in real time based on the processed image data, and register the constructed 3D model to obtain the registered 3D model. S4. Based on the registered 3D model, surgical path planning is performed using 3D tracking, including the following steps: S41. Divide the registered 3D model into segments using a model cutting method to obtain the segmented 3D model. S42. Based on the partitioned 3D model, perform localization analysis using collision detection methods and output the localization results; S43. Surgical path planning based on positioning results; S5. Based on the planned surgical path, predictive assistance and guidance are provided through digital twins.

2. The method for assisting and guiding pediatric neurosurgery based on machine vision according to claim 1, characterized in that, The process of installing a laser camera to acquire real-time neurosurgical images from different angles and then compiling them into an image training dataset includes the following steps: Multiple reference points are set within the neurosurgical surgical area, and laser cameras are evenly installed around the neurosurgical surgical area at set distances and angles, so that the reference points within the neurosurgical surgical area are evenly distributed throughout the camera's field of view. A three-dimensional coordinate system is constructed based on the installed laser cameras and the center point of the neurosurgical area, and the three-dimensional coordinates of each laser camera are recorded in real time. The image acquisition interval of the laser camera is set, and image data of the neurosurgical area is acquired in real time according to the set acquisition interval. At the same time, the image data of the neurosurgical area acquired in real time by each laser camera are summarized to construct an image training dataset.

3. The method for assisting and guiding pediatric neurosurgery based on machine vision according to claim 1, characterized in that, The process of enhancing real-time acquired image data using image enhancement methods to obtain enhanced image data includes the following steps: Calculate pixel probability distribution based on grayscale value distribution in real-time acquired image data; The formula for pixel probability distribution is as follows: ； in, Represents the pixel probability of image data. Indicates grayscale value The number of pixels, This represents the total number of pixels. Construct the cumulative distribution function of the image based on the pixel probability distribution of the image data; The formula for constructing the cumulative distribution function of an image is shown below: ； Where L represents the total number of grayscale values ​​in the image. Indicates the first The gray values ​​are mapped by the cumulative distribution function; Image enhancement is performed on image data based on the constructed image cumulative distribution function; The gray values ​​mapped by the cumulative distribution function are multiplied by the total number of gray values ​​using a transformation function. Based on the mapping relationship between the cumulative distribution function and the transformation function of the image, and referring to the pixels in the original image, the histogram equalized image is output. The image after histogram equalization is defined as the enhanced image data.

4. The method for assisting and guiding pediatric neurosurgery based on machine vision according to claim 1, characterized in that, The process of identifying the fused image data using a data recognition method to obtain processed image data includes the following steps: Collect various types of historical neural image data, and input the collected historical neural image data and the fused image data into a convolutional neural network for feature extraction; The feature extraction formula for convolutional networks is shown below: ； in, This represents the extracted image features. This represents the input image data. This represents the weights for feature extraction in a convolutional neural network. Indicates the bias value; The extracted features from historical neural image data and the features from the fused image data are compared. The feature comparison formula is as follows: ； in, Representing features of historical neural image data and image data features Similarity values ​​between them; A similarity threshold is set, and the fused image data is classified based on the set threshold. Image data that meet the similarity threshold are classified into one category, resulting in classified image data. The classified image data is defined as the processed image data.

5. The method for assisting and guiding pediatric neurosurgery based on machine vision according to claim 1, characterized in that, The process of constructing a 3D model in real time based on the processed image data and registering the constructed 3D model to obtain the registered 3D model includes the following steps: A three-dimensional model is constructed in real time based on a set coordinate system, processed image data, and reference points within the neurosurgical surgical area. Based on the real-time constructed 3D model, a point cloud dataset is obtained by summarizing reference points within the neurosurgical surgical area; The aggregated point cloud dataset is registered by calculating the mean and covariance of the point cloud data. Reconstructing a 3D model based on registered point cloud data; set up The reconstructed 3D model, The registered point cloud data; in, These represent the horizontal and vertical coordinates in the registered point cloud data, respectively. This indicates the camera's axis distance parameter; The relationship between the reconstructed 3D model and the registered point cloud data is shown below: ； in, express Axis camera wheelbase, express Axis camera wheelbase, These represent the offset between the pixel plane and the camera imaging plane, respectively. This represents the rotation vector of the camera in the 3D model. This represents the translation vector of the camera in the 3D model; The reconstructed 3D model is set as the registered 3D model.

6. The method for assisting and guiding pediatric neurosurgery based on machine vision according to claim 1, characterized in that, The process of dividing the registered 3D model into subdivided 3D models by model cutting includes the following steps: Select a vertex on the 3D model outline as the initial point; Based on the selected initial point, the coordinates of the reference points in each plane of the model are traversed sequentially, and a linked list is set up for numbering. If two reference points with the same coordinates are detected in two planes during the traversal, the two planes are set as adjacent planes. Based on the obtained numbered 3D model, slices are made by setting the height and determining the contour of the cutting plane; Set the slice thickness, and according to the set thickness, slice the 3D model evenly from the bottom upwards with a transverse cutting plane perpendicular to the Z-axis. Each slice yields a hyperplane, and the location of the affected area is determined based on the obtained hyperplane. Construct a classification decision function and determine the location of the affected area based on the constructed classification decision function; ； in, For weights, For classification threshold, Represents the classification decision function; The classification decision function is constructed to traverse each pixel in the hyperplane, and the hyperplane is divided into affected areas and non-affected areas based on the traversal. After the division is completed, the affected area is located based on the position of the hyperplane reference point.

7. The machine vision-based assisted guidance method for pediatric neurosurgery according to claim 1, characterized in that, The localization analysis based on the partitioned 3D model using collision detection methods, and the output of the localization results, includes the following steps: Images of the surgical process are collected and processed through steps 2 and 3 to obtain a three-dimensional model of the surgical process images. At the same time, the hyperplane of the three-dimensional model of the surgical process images is obtained through model cutting. Collision detection is performed on hyperplanes based on the 3D model of surgical process images and the 3D model of neural images. When the two sets of hyperplanes intersect, it indicates that the two sets of hyperplanes are colliding. The location of the surgical procedure is determined based on the collision results.

8. The machine vision-based assisted guidance method for pediatric neurosurgery according to claim 1, characterized in that, The surgical path planning based on the positioning results includes the following steps: The three-dimensional model of the neural image is rasterized, and the surgical path is planned using the A* algorithm based on the localization results and the coordinates of the affected area. The actual path distance and the estimated path distance to the affected area are determined based on the actual path distance of each unit coordinate in the grid space and the length of the path in the grid space. If there exists a path whose estimated path distance is less than or equal to the actual path distance from the current location to the affected area, then this path is designated as the optimal path.

9. The machine vision-based assisted guidance method for pediatric neurosurgery according to claim 1, characterized in that, The method of predicting and assisting guidance based on the planned surgical path using digital twins includes the following steps: The surgical path, planned based on the current location and the location of the affected area, will be separated into multiple smaller target nodes. Each time a small target node is reached, the current target node is set as the current node, and the next small target node is set as an auxiliary guiding node, continuously iterating to assist doctors in surgical planning.

10. A system for implementing the machine vision-based assisted guidance method for pediatric neurosurgery as described in claims 1-9, characterized in that, It includes a data acquisition module, a data processing module, a model building module, a path planning module, and a prediction-aided guidance module; The data acquisition module is used to install a laser camera to acquire and summarize neurosurgical image data from different angles in real time. The data processing module is used to process the acquired image data through data processing methods; The model building module is used to perform 3D modeling and registration based on the processed image data; The path planning module is used to segment the constructed 3D model and calculate the planned path; The prediction-assisted guidance module is used to provide assisted guidance based on the path planning results.

Images