Neurosurgical navigation acceleration processing method and system

Abstract

The invention relates to the technical field of surgical navigation, and discloses a neurosurgical navigation acceleration processing method and system. According to the method, three-dimensional reconstruction is carried out by adopting an SART algorithm, a historical and real-time three-dimensional basic model is obtained, and a deep learning module constructed by a U-Net image segmentation network, a YOLO target detection network and a graph neural network is utilized to obtain an organizational structure and a surgical instrument position in real time and carry out path planning. Meanwhile, the data preprocessing module is used for conducting denoising and contrast enhancement processing on the real-time operation image data, and the image definition is improved. According to the method and the system, the precision, the real-time performance and the safety of the neurosurgery operation can be remarkably improved, and the method and the system have important clinical application value.

Description

Technical Field

[0001] The present invention relates to the technical field of surgical navigation, and in particular to a method and system for accelerating processing of neurosurgery navigation. Background Art

[0002] Neurosurgery is a highly complex and delicate medical operation that requires doctors to have superb surgical skills and rich anatomical knowledge. Traditional neurosurgery navigation methods mainly rely on preoperative imaging data and the doctor's experience to plan the surgical path. However, this method has problems such as insufficient accuracy and poor real-time performance. With the rapid development of medical imaging technology and computer technology, three-dimensional reconstruction technology and deep learning algorithms have been widely used in neurosurgery navigation. However, existing neurosurgery navigation methods still have problems such as low computational efficiency and inaccurate path planning when processing large amounts of imaging data, which affects the efficiency and safety of the operation. Summary of the invention

[0003] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to provide a neurosurgery navigation acceleration processing method and system to solve the problems of low computing efficiency and inaccurate path planning in the prior art.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for accelerating a neurosurgery navigation process, comprising: Step S100: Obtain historical neurosurgery imaging data, including historical preoperative CT images, historical preoperative MRI images, and historical intraoperative images; perform three-dimensional reconstruction of the historical preoperative CT images and historical preoperative MRI images using the SART algorithm to obtain a historical three-dimensional basic model; annotate the historical imaging data to mark the tissue structure, the position of the surgical instruments, the surgical target position, and the surgical path information; Step S200: adopting a deep learning-based method to train a neurosurgery navigation model using a historical three-dimensional basic model and historical image data, wherein the deep learning method includes using a U-Net image segmentation network to segment tissue structures in the image, using a YOLO target detection network to track the position of surgical instruments, and using a graph neural network for path planning; Step S300: During the neurosurgery operation, the SART algorithm is used to perform three-dimensional reconstruction based on the preoperative CT image and MRI image to obtain a real-time three-dimensional basic model; neurosurgery image data is acquired in real time, and data preprocessing is performed, including denoising and contrast enhancement; Step S400: The neurosurgery navigation model reads the three-dimensional basic model and pre-processed surgical image data in real time, uses the U-Net image segmentation network and the YOLO target detection network to obtain the tissue structure and surgical instrument position in real time, uses the graph neural network for path planning, generates and updates the surgical path in real time, and displays the surgical path in the navigation system.

[0005] Preferably, in a possible implementation of the first aspect, the three-dimensional reconstruction step of the SART algorithm includes: aligning the historical preoperative CT image and the historical preoperative MRI image to form a unified coordinate system, and iteratively updating the reconstructed image using the SART algorithm until a preset convergence condition is met.

[0006] Preferably, in a possible implementation manner of the first aspect, the U-Net image segmentation network structure includes: The first input layer is used to receive real-time image data; The first feature extraction layer uses an encoder for downsampling; Feature fusion layer, which performs feature fusion through skip connections; Upsampling and segmentation layer, using encoder for upsampling; In the first output layer, a 1x1 convolutional layer is used to convert the feature map into a segmentation probability map, and then converted into a binary segmentation result through a sigmoid activation function.

[0007] Preferably, in a possible implementation of the first aspect, the YOLO target detection network structure includes: The second input layer receives real-time image data; The second feature extraction layer uses the Darknet-53 network structure for feature extraction; Feature pyramid construction layer, constructing feature pyramid; The detection head processing layer applies convolutional layers to each multi-scale feature map to generate bounding box predictions and class predictions; In the post-processing layer, non-maximum suppression algorithm is applied to remove overlapping bounding boxes; The second output layer tracks the position of surgical instruments in real time based on continuous detection results.

[0008] Preferably, in a possible implementation of the first aspect, the graph neural network structure includes: The data preprocessing layer builds a graph structure based on the organizational structure segmented by the U-Net network and the surgical instrument location information detected by the YOLO network; The third feature extraction layer uses the CNN extraction method to extract features; Graph convolution layer, which performs convolution operations on the graph structure to update the node feature representation; Path planning layer, which uses reinforcement learning algorithm or search algorithm for path planning; The third output layer outputs the generated surgical path to the output and feedback step of the neurosurgery navigation system.

[0009] Preferably, in a possible implementation of the first aspect, the data preprocessing step includes using a Gaussian filtering method to remove noise in real-time surgical image data, and using an adaptive contrast enhancement method to improve the visibility of tissue structures and surgical instruments in the image.

[0010] Preferably, in a possible implementation manner of the first aspect, step S400 specifically includes: The neurosurgery navigation model reads the 3D basic model and surgical image data in real time; Use U-Net image segmentation network to segment image data to obtain tissue structure information; Use the YOLO target detection network to detect the location of surgical instruments; The image segmentation results and target detection results are input into the graph neural network for path planning; Generate surgical paths in real time based on path planning results, and update path planning results immediately when the position of surgical instruments or tissue structure changes; The surgical path is displayed on the 3D base model in the navigation system.

[0011] In a second aspect, the present invention provides a neurosurgery navigation acceleration processing system, comprising: A data acquisition module, used to acquire historical imaging data and real-time imaging data of neurosurgery operations; The 3D reconstruction module uses the SART algorithm to perform 3D reconstruction based on historical preoperative CT images and MRI images to obtain a historical 3D basic model and a real-time 3D basic model; Deep learning modules, including U-Net image segmentation network, YOLO target detection network and graph neural network, are used to train neurosurgery navigation models and obtain tissue structure and surgical instrument position in real time for path planning; Data preprocessing module, which performs denoising and contrast enhancement on real-time surgical image data; The navigation display module is used to display the surgical path in real time and adjust the surgical path according to the modification instructions.

[0012] The beneficial effects of the present invention are as follows: the neurosurgery navigation acceleration processing method of the present invention, by adopting the SART algorithm for three-dimensional reconstruction, can efficiently obtain the historical three-dimensional basic model and the real-time three-dimensional basic model, providing an accurate basis for subsequent path planning. At the same time, the deep learning module constructed by using the U-Net image segmentation network, the YOLO target detection network and the graph neural network can obtain the tissue structure and the position of the surgical instruments in real time, and perform path planning, which greatly improves the accuracy and real-time performance of the operation. In addition, the present invention also performs denoising and contrast enhancement processing on the real-time surgical image data through the data preprocessing module, further improving the clarity and visibility of the image data, and providing more intuitive and accurate navigation information for the doctor's surgical operation. In general, the neurosurgery navigation acceleration processing method of the present invention can significantly improve the efficiency and safety of neurosurgery, and has important clinical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0014] Figure 1 A flowchart of a method for accelerating processing of neurosurgery navigation is provided for the present application.

[0015] Figure 2 A structural schematic diagram of a neurosurgery navigation acceleration processing system is provided for this application.

[0016] Explanation of the accompanying drawings: 1-data acquisition module, 2-three-dimensional reconstruction module, 3-deep learning module, 4-data preprocessing module, 5-navigation display module. DETAILED DESCRIPTION

[0017] 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 only 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.

[0018] Example 1: Figure 1 As shown, the present invention provides a method for accelerating processing of neurosurgery navigation, comprising: Step S100: Obtain historical imaging data of neurosurgery, including historical preoperative CT images, historical preoperative MRI images, and historical intraoperative images; use the SART algorithm to perform three-dimensional reconstruction of historical preoperative CT images and historical preoperative MRI images to obtain a historical three-dimensional basic model; annotate the historical imaging data to mark the tissue structure, the position of the surgical instruments, the surgical target position, and the surgical path information.

[0019] Specifically, the three-dimensional reconstruction step of the SART algorithm covers the following operations: first, the historical preoperative CT images and the historical preoperative MRI images are aligned to form a unified coordinate system; then, the SART algorithm is used to iteratively update the reconstructed image until the preset convergence conditions are met.

[0020] In this embodiment, the historical imaging data of neurosurgery is first retrieved from the database, which includes preoperative CT images, preoperative MRI images and intraoperative images, and all images are stored in DICOM format to ensure data compatibility and integrity. The SimpleITK library (an open source image processing library that supports reading and conversion of multiple image formats) is used to read these DICOM images and convert them into an array format that can be used for subsequent processing, laying a solid foundation for subsequent steps.

[0021] Since CT images and MRI images come from different imaging devices and exist in different coordinate systems, image registration is required to fuse the two images into a unified coordinate system. This embodiment uses the registration function of SimpleITK to achieve this goal. First, initialize the registration method and configure parameters such as similarity metric, optimizer, and interpolator. Among them, the similarity metric is used to evaluate the similarity between two images, the optimizer is responsible for finding the transformation matrix that maximizes the similarity, and the interpolator ensures the continuity of the image during the transformation process.

[0022] Subsequently, the registration process is executed. The optimizer continuously adjusts the transformation matrix during the registration process until the similarity between the CT image and the MRI image is maximized. After multiple iterations, the transformation matrix is ​​finally obtained. This transformation matrix is ​​used to resample the MRI image so that it is spatially aligned with the CT image, thereby completing the fusion of the CT image and the MRI image in a unified coordinate system, providing a basis for subsequent 3D reconstruction.

[0023] After obtaining the registered image data, the SART algorithm is used for 3D reconstruction. As an iterative reconstruction algorithm, the SART algorithm approximates the real 3D structure by continuously updating the reconstructed image. First, a reconstructed image with all zeros is initialized; then, multiple iterative updates are performed based on the projection matrix and sine diagram. The projection matrix and sine diagram are generated based on the specific imaging system and data, and describe the distribution and projection relationship of the image data in space.

[0024] In each iteration, the error between the projection of the current reconstructed image and the sinogram is calculated, and the error reflects the difference between the current reconstructed image and the real three-dimensional structure. Then, the reconstructed image is updated using this error. After multiple iterations (10 times in this embodiment), the real three-dimensional structure is gradually approached to obtain the final three-dimensional reconstructed image. This image has high precision and high resolution, and can clearly show the three-dimensional structure of the neurosurgical area.

[0025] After obtaining the 3D reconstructed image, the image data needs to be annotated. This embodiment uses the ITK-SNAP tool for manual annotation. ITK-SNAP is an open source image annotation tool that supports the import and annotation of multiple image formats. Use ITK-SNAP to mark the surgical area, the location of surgical instruments, and the surgical path on the 3D reconstructed image, and save the annotation results as an XML format file.

[0026] These annotation information can provide important reference for subsequent surgical planning. Doctors can understand the three-dimensional structure of the surgical area, the location of surgical instruments, and the surgical path based on the annotation information, so as to formulate a more accurate and safe surgical plan.

[0027] Step S200: Using a deep learning-based method to train a neurosurgery navigation model using historical three-dimensional basic models and historical image data, the deep learning method includes using a U-Net image segmentation network to segment tissue structures in the image, using a YOLO target detection network to track the position of surgical instruments, and using a graph neural network for path planning.

[0028] Specifically, when using the U-Net image segmentation network for tissue structure segmentation, the process is as follows: First, the first layer of the U-Net network is used as the input layer to receive real-time image data. These real-time image data come from the surgical process, are converted into a format suitable for neural network processing after preprocessing, and are input into the network. Next, the first feature extraction layer uses an encoder for downsampling, which consists of multiple convolutional layers and pooling layers to extract features from the image. The convolutional layer captures local features through convolution operations, while the pooling layer reduces the dimension of the data through downsampling while retaining key features. Subsequently, the feature fusion layer fuses the feature maps of different levels in the encoder through jump connections to retain more detailed information and improve the segmentation performance of the network. Next, the upsampling and segmentation layer uses the encoder for upsampling, gradually restores the resolution of the image, and generates the final segmentation result. In this process, a 1x1 convolutional layer is used to convert the feature map into a segmentation probability map, and the probability map is converted into a binary segmentation result through a sigmoid activation function. Finally, the output layer is responsible for outputting the final segmentation result, that is, each pixel is classified as foreground or background.

[0029] In this embodiment, a class named UNet is defined, which inherits from nn.Module. In the initialization method __init__ of the UNet class, two parameters are received: in_channels (indicating the number of channels of the input image) and out_channels (indicating the number of channels of the output segmentation map). Next, the first feature extraction layer self.encoder1 is defined, which is a nn.Sequential module containing two convolutional layers (nn.Conv2d), two ReLU activation functions (nn.ReLU) and a maximum pooling layer (nn.MaxPool2d). In addition, the upsampling and segmentation layers (decoders) self.upconv2 and self.decoder2 are also defined. In the forward method, the forward propagation process of the network is defined, including steps such as feature extraction, feature fusion and upsampling, and finally the segmentation result is obtained.

[0030] When using the YOLO target detection network for surgical instrument position tracking, the process is as follows: First, the input layer of the YOLO target detection network is responsible for receiving real-time image data, which comes from the camera or video stream in the operating room and is preprocessed. Next, the feature extraction layer uses the Darknet-53 network structure to extract features from the input image. Based on the feature extraction, the feature pyramid construction layer improves the model's detection ability for targets of different sizes by fusing feature maps at different scales. The detection head processing layer applies a convolutional layer to each multi-scale feature map in the feature pyramid to generate bounding box predictions and category predictions. The post-processing layer applies the non-maximum suppression algorithm (NMS) to remove overlapping bounding boxes to improve the accuracy and efficiency of detection. Finally, the output layer tracks the position of surgical instruments in real time based on continuous detection results and passes this information to the neurosurgical navigation system.

[0031] In this example, OpenCV is used to capture video frames from a video stream or camera and preprocess the frames. After that, a placeholder class for the YOLO model for object detection is defined, including the Darknet-53 backbone network, the Feature Pyramid Network (FPN), and the detection head. The output of the YOLO model is post-processed to obtain the bounding box, confidence, and category of the object.

[0032] When using graph neural networks for path planning, the process is as follows: First, the data preprocessing layer constructs a graph structure based on the tissue structure obtained by the U-Net network segmentation and the surgical instrument location information detected by the YOLO network. Specifically, it creates an empty graph structure and adds each tissue structure as a node to the graph, setting its type ('tissue') and location attributes. At the same time, each surgical instrument is also added to the graph as a node, setting its type ('instrument') and location attributes. Finally, according to the spatial position relationship or logical relationship between the tissue structures, edges are added to the nodes in the graph. Next, the third feature extraction layer uses a convolutional neural network (CNN) to extract features from the input data. Then, the graph convolution layer performs a convolution operation on the previously constructed graph structure to update the feature representation of the node. Next, the path planning layer uses a reinforcement learning algorithm or a search algorithm to perform path planning and output a path from the start node to the end node. Finally, the third output layer outputs the generated surgical path to the output and feedback step of the neurosurgical navigation system.

[0033] The above process ensures that the neurosurgery navigation model can accurately segment tissue structures in images, track the position of surgical instruments, and plan the optimal surgical path, thereby providing precise and safe navigation support for neurosurgery.

[0034] Step S300: During the neurosurgery operation, the SART algorithm is used to perform three-dimensional reconstruction based on the preoperative CT images and MRI images to obtain a real-time three-dimensional basic model; neurosurgery image data is acquired in real time, and data preprocessing is performed, including denoising and contrast enhancement.

[0035] Specifically, the data preprocessing process includes: using Gaussian filtering technology to eliminate noise in real-time surgical image data; and using adaptive contrast enhancement methods to improve the visibility of tissue structures and surgical instruments in the image.

[0036] In this embodiment, a function named preprocess_image is defined to preprocess the real-time surgical image data. The specific operation involves two steps: image denoising and contrast enhancement.

[0037] The function preprocess_image receives a parameter named image, which is an array of type numpy.ndarray, representing the input surgical image data and supports grayscale images or color images.

[0038] The function first uses the cv2.GaussianBlur function to perform Gaussian filtering on the image to remove image noise. In this processing step, the size of the Gaussian kernel is set to (5, 5), and the standard deviation is automatically calculated based on the kernel size. The filtered image is stored in the denoised_image variable.

[0039] Next, the function selects an appropriate contrast enhancement strategy based on the number of channels of the input image (i.e., whether the image is a grayscale image). If the image is a grayscale image (i.e., the length of image.shape is 2), a contrast-limited adaptive histogram equalization (CLAHE) object is created through the cv2.createCLAHE function, and this CLAHE is applied to the denoised image using the apply method to further enhance the image contrast. The enhanced image is saved in the contrast_enhanced_image variable.

[0040] Finally, the function returns the preprocessed image contrast_enhanced_image.

[0041] Step S400: The neurosurgery navigation model reads the three-dimensional basic model and pre-processed surgical image data in real time, uses the U-Net image segmentation network and the YOLO target detection network to obtain the tissue structure and surgical instrument position in real time, uses the graph neural network for path planning, generates and updates the surgical path in real time, and displays the surgical path in the navigation system.

[0042] Specifically, the neurosurgery navigation model first synchronizes the three-dimensional basic model and surgical image data in real time.

[0043] Next, the U-Net image segmentation network is used to accurately segment the surgical image data to obtain detailed tissue structure information. The real-time acquired surgical image data is input into the pre-trained U-Net image segmentation network, which implements pixel-by-pixel classification of the input image, accurately distinguishes different tissue structures, and outputs the segmented image, in which various tissue structures are clearly identified by different colors or labels.

[0044] At the same time, the YOLO target detection network is used to accurately locate the position of surgical instruments. Real-time surgical image data is also input into the YOLO target detection network, which accurately identifies surgical instruments in the image and accurately outputs their position and category information, including the bounding box coordinates, confidence score and category label of the surgical instrument.

[0045] Subsequently, the tissue structure obtained by U-Net segmentation and the surgical instrument location information detected by YOLO are passed as input to the graph neural network for path planning. The graph neural network builds a complex graph structure based on this information, in which the nodes represent the tissue structure and surgical instruments, and the edges reflect the spatial or logical connection between them. The graph neural network uses the internal graph convolution layer and reinforcement learning or efficient search algorithm to intelligently plan the optimal path from the surgical starting point to the target point.

[0046] Based on the planning results of the graph neural network, the surgical path is dynamically generated and updated in real time. Once there is any change in the position of the surgical instrument or the tissue structure, the path planning results will respond and update immediately, thus ensuring real-time adjustment and optimization of the surgical path.

[0047] Finally, the surgical path is clearly displayed in the navigation system. The generated surgical path information is accurately transmitted to the neurosurgery navigation system. Based on this information, the navigation system overlays the surgical path on the three-dimensional basic model to provide doctors with intuitive and clear surgical guidance. Doctors can perform precise surgical operations based on the displayed surgical path. At the same time, the navigation system will continuously update the display content based on real-time image data and path planning results to ensure the accuracy, safety and reliability of the surgery.

[0048] Example 2: Figure 2 As shown, the present invention provides a neurosurgery navigation acceleration processing system, including 1-data acquisition module, 2-3D reconstruction module, 3-deep learning module, 4-data preprocessing module, 5-navigation display module, wherein: A data acquisition module, used to acquire historical imaging data and real-time imaging data of neurosurgery operations; The 3D reconstruction module uses the SART algorithm to perform 3D reconstruction based on historical preoperative CT images and MRI images to obtain a historical 3D basic model and a real-time 3D basic model; Deep learning modules, including U-Net image segmentation network, YOLO target detection network and graph neural network, are used to train neurosurgery navigation models and obtain tissue structure and surgical instrument position in real time for path planning; Data preprocessing module, which performs denoising and contrast enhancement on real-time surgical image data; The navigation display module is used to display the surgical path in real time and adjust the surgical path according to the modification instructions.

[0049] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A neurosurgery navigation acceleration processing method, characterized in that: include: Step S100: Obtain historical neurosurgery imaging data, including historical preoperative CT images, historical preoperative MRI images, and historical intraoperative images; perform three-dimensional reconstruction of the historical preoperative CT images and historical preoperative MRI images using the SART algorithm to obtain a historical three-dimensional basic model; annotate the historical imaging data to mark the tissue structure, the position of the surgical instruments, the surgical target position, and the surgical path information; Step S200: adopting a deep learning-based method to train a neurosurgery navigation model using a historical three-dimensional basic model and historical image data, wherein the deep learning method includes using a U-Net image segmentation network to segment tissue structures in the image, using a YOLO target detection network to track the position of surgical instruments, and using a graph neural network for path planning; Step S300: During the neurosurgery operation, the SART algorithm is used to perform three-dimensional reconstruction based on the preoperative CT image and MRI image to obtain a real-time three-dimensional basic model; neurosurgery image data is acquired in real time, and data preprocessing is performed, including denoising and contrast enhancement; Step S400: The neurosurgery navigation model reads the three-dimensional basic model and pre-processed surgical image data in real time, uses the U-Net image segmentation network and the YOLO target detection network to obtain the tissue structure and surgical instrument position in real time, uses the graph neural network for path planning, generates and updates the surgical path in real time, and displays the surgical path in the navigation system.

2. The neurosurgery navigation acceleration processing method according to claim 1, characterized in that: The three-dimensional reconstruction step of the SART algorithm includes: registering the historical preoperative CT image and the historical preoperative MRI image to form a unified coordinate system, and iteratively updating the reconstructed image using the SART algorithm until a preset convergence condition is met.

3. The neurosurgery navigation acceleration processing method according to claim 1, characterized in that: The U-Net image segmentation network structure includes: The first input layer is used to receive real-time image data; The first feature extraction layer uses an encoder for downsampling; Feature fusion layer, which performs feature fusion through skip connections; Upsampling and segmentation layer, using encoder for upsampling; In the first output layer, a 1x1 convolutional layer is used to convert the feature map into a segmentation probability map, and then converted into a binary segmentation result through a sigmoid activation function.

4. The neurosurgery navigation acceleration processing method according to claim 3, characterized in that: The YOLO target detection network structure includes: The second input layer receives real-time image data; The second feature extraction layer uses the Darknet-53 network structure for feature extraction; Feature pyramid construction layer, constructing feature pyramid; The detection head processing layer applies convolutional layers to each multi-scale feature map to generate bounding box predictions and class predictions; In the post-processing layer, non-maximum suppression algorithm is applied to remove overlapping bounding boxes; The second output layer tracks the position of surgical instruments in real time based on continuous detection results.

5. The neurosurgery navigation acceleration processing method according to claim 4, characterized in that: The graph neural network structure includes: The data preprocessing layer builds a graph structure based on the organizational structure segmented by the U-Net network and the surgical instrument location information detected by the YOLO network; The third feature extraction layer uses the CNN extraction method to extract features; Graph convolution layer, which performs convolution operations on the graph structure to update the node feature representation; Path planning layer, which uses reinforcement learning algorithm or search algorithm for path planning; The third output layer outputs the generated surgical path to the output and feedback step of the neurosurgery navigation system.

6. The neurosurgery navigation acceleration processing method according to claim 1, characterized in that: The data preprocessing step includes using a Gaussian filtering method to remove noise in real-time surgical image data, and using an adaptive contrast enhancement method to improve the visibility of tissue structures and surgical instruments in the image.

7. The neurosurgery navigation acceleration processing method according to claim 1, characterized in that: The step S400 specifically includes: The neurosurgery navigation model reads the 3D basic model and surgical image data in real time; Use U-Net image segmentation network to segment image data to obtain tissue structure information; Use the YOLO target detection network to detect the location of surgical instruments; The image segmentation results and target detection results are input into the graph neural network for path planning; Generate surgical paths in real time based on path planning results, and update path planning results immediately when the position of surgical instruments or tissue structure changes; The surgical path is displayed on the 3D base model in the navigation system.

8. A neurosurgery navigation acceleration processing system, characterized in that: include: A data acquisition module, used to acquire historical imaging data and real-time imaging data of neurosurgery operations; The 3D reconstruction module uses the SART algorithm to perform 3D reconstruction based on historical preoperative CT images and MRI images to obtain a historical 3D basic model and a real-time 3D basic model; Deep learning modules, including U-Net image segmentation network, YOLO target detection network and graph neural network, are used to train neurosurgery navigation models and obtain tissue structure and surgical instrument position in real time for path planning; Data preprocessing module, which performs denoising and contrast enhancement on real-time surgical image data; The navigation display module is used to display the surgical path in real time and adjust the surgical path according to the modification instructions.

Images