Anesthesia retardation target positioning method and system based on multi-source data

Abstract

The invention relates to the technical field of computer-aided surgery, in particular to a method and a system for positioning a retardation target spot for anesthesia based on multi-source data. The method comprises the steps that an MRI image of a target area is acquired, and a preoperative three-dimensional structure model is constructed; acquiring an intra-operative US image and performing multi-modal registration on the intra-operative US image and the MRI image; a periodic displacement field of the anatomical structure is extracted based on the continuous US image frame sequence, a dynamic displacement model is constructed, sequential transformation is carried out on the registration image, and a compensation multi-modal image is generated; identifying and segmenting the key anatomical structure through a deep neural network, and obtaining structure mask information; screening a plurality of candidate injection points in the mask range, and selecting an optimal retardation target based on target function scoring; and displaying the retardation target spot and a segmentation structure result in an overlapping manner, matching a preoperative three-dimensional model, constructing a three-dimensional reference model, and realizing intraoperative visual guidance.

Description

Technical Field

[0001] The present invention relates to the field of computer - assisted surgical techniques, and particularly to a method and system for positioning anesthetic block targets based on multi - source data. Background Technique

[0002] During clinical anesthesia, nerve block techniques are widely used in fields such as surgical operations and pain management. To improve the safety and effectiveness of surgeries, doctors often perform nerve block operations with the aid of ultrasound guidance. However, traditional ultrasound guidance methods have problems such as limited image resolution, poor soft - tissue contrast, and the dependence of anatomical structure recognition on the operator's experience, which can easily lead to complications such as inaccurate punctures, improper drug injection, or damage to nerve blood vessels.

[0003] Although current clinical practice has introduced composite techniques such as "ultrasound guidance + nerve stimulator" combined positioning in order to introduce electrophysiological feedback on the basis of visual positioning to enhance the reliability of positioning, there are still many limitations. On the one hand, there are individual differences in nerve sensitivity of different patients to the nerve stimulator, and the feedback signal is not stable; on the other hand, the ultrasound image itself is still limited by the operator's experience and imaging angle, and the structure recognition depends on subjective judgment. In addition, the lack of effective fusion of pre - operative and intra - operative image data makes it difficult to provide stable and consistent spatial reference information, and the dynamic structures such as the diaphragm and pleura shift under respiratory drive, making it more difficult to accurately predict the target point.

[0004] Therefore, a method and system for positioning anesthetic block targets based on multi - source data are proposed. Summary of the Invention

[0005] The present invention provides a method and system for positioning anesthetic block targets based on multi - source data. By combining pre - operative MRI images and intra - operative US images, through techniques such as image registration, dynamic modeling, deep - learning segmentation, and intelligent evaluation, precise recognition of key structures such as nerves, blood vessels, and fascia is achieved, and the optimal block target position is automatically selected. This method can fuse the recognition results with intra - operative images and match them to the pre - operative three - dimensional model to construct a three - dimensional reference model, providing an intuitive and accurate positioning reference for clinical anesthesia and improving the safety and effectiveness of block operations.

[0006] To achieve the above object, the present invention provides the following technical solutions: A method for positioning anesthetic block targets based on multi - source data, comprising: Obtain the MRI image of the target area, perform image segmentation and three - dimensional modeling on the MRI image, extract the boundary information of the anatomical structure, and construct a pre - operative structure model; Obtain the US image of the target area as the intraoperative guidance image. Based on the anatomical structure information between the MRI image and the US image, use the maximization of mutual information to spatially align the MRI image and the US image to generate a registered multimodal image; Based on the displacement field of the periodic motion of the anatomical structure identified in the intraoperative continuous US image frame sequence, construct a dynamic displacement model of the target structure, and perform a temporal transformation on the multimodal registration image to generate a compensated multimodal image; Input the compensated multimodal image into the trained deep neural network model to identify and segment the anatomical structure, and output the structure mask information; Obtain multiple candidate injection points within the segmentation mask, design an objective function to score each point, and select one or more points with the maximum score value as the block target; Overlay the block target and the structure segmentation result on the intraoperative US image, and perform 3D matching of the overlay result to the preoperative structure model to generate a three-dimensional reference model to assist in clinical anesthesia operation to achieve target positioning.

[0007] Furthermore, the steps for constructing the preoperative structure model are: Apply an image enhancement and denoising algorithm to the MRI image to improve the clarity of the soft tissue edges; Based on the deep neural network model, perform semantic segmentation on the MRI image to extract the boundary information of the anatomical structure; Input the segmentation result into a three-dimensional reconstruction algorithm, and through interpolation reconstruction and surface mesh generation, construct a three-dimensional model of the preoperative anatomical structure.

[0008] Furthermore, the steps for generating the registered multimodal image are: Use the MRI image as the reference image and the intraoperative US image as the moving image; Based on the spatial distribution relationship of the key anatomical structures in the MRI image and the US image, calculate the mutual information value between the two images; Adopt a non-rigid registration algorithm based on the maximization of mutual information to perform elastic deformation on the US image, so that the US image is aligned with the MRI image on the key anatomical structures to generate a registered multimodal image.

[0009] Furthermore, the steps for generating the compensated multimodal image are: Based on the continuously acquired intraoperative US image sequence, perform time series modeling on the target area to identify the displacement field of the anatomical structure in multiple frames; Establish a dynamic deformation field of the key anatomical structure for the displacement field through a temporal convolutional neural network; Apply the dynamic deformation field to the registered multimodal image, perform a temporal spatial transformation on the static structure model, and generate a compensated multimodal image that dynamically changes with breathing.

[0010] Further, the calculation formula of the objective function is as follows: ; wherein, represents the comprehensive score of the candidate injection point , represents the minimum Euclidean distance between the candidate point and the target nerve, represents the minimum Euclidean distance between the candidate point and the blood vessel, represents the spatial width between the fascia layers where the candidate point is located, , and are weighting coefficients.

[0011] Further, the steps for generating the three-dimensional reference model are as follows: Convert the structural mask and the block target into a pseudo-color layer, and use the gradient fusion technology to superimpose the pseudo-color layer on the intraoperative US image to obtain a pseudo-color US image; Perform three-dimensional reconstruction on the two-dimensional pseudo-color US image through back-projection reconstruction technology, register it with the three-dimensional preoperative structural model, and generate a three-dimensional reference model.

[0012] The present invention also provides an anesthetic block target positioning system based on multi-source data, including: an MRI image structure extraction module, configured to obtain an MRI image of a target area, perform image segmentation and three-dimensional modeling on the MRI image, extract the boundary information of the anatomical structure, and construct a preoperative structural model; an image registration module, configured to obtain a US image of the target area as an intraoperative guiding image, and based on the anatomical structure information between the MRI image and the US image, perform spatial alignment on the MRI image and the US image by maximizing mutual information to generate a registered multimodal image; an image compensation module, configured to construct a dynamic displacement model of the target structure based on the displacement field of the periodic motion of the anatomical structure identified in the intraoperative continuous US image frame sequence, and perform temporal transformation on the multimodal registered image to generate a compensated multimodal image; a mask acquisition module, configured to input the compensated multimodal image into a trained deep neural network model, identify and segment the anatomical structure, and output the structure mask information; a block target positioning module, configured to obtain multiple candidate injection points within the segmentation mask range, design an objective function to score each point, and select one or more points with the maximum score value as the block target; A three-dimensional model reference module is used to superimpose the blocked target point and the structure segmentation result on the intraoperative US image, and perform 3D matching of the superimposed result to the preoperative structure model to generate a three-dimensional reference model to assist clinical anesthesia operation to achieve target point positioning.

[0013] The beneficial effects of the present invention are as follows: By fusing preoperative MRI images and intraoperative US images, multi-modal structural information acquisition and registration of the patient's target area are realized, overcoming problems such as incomplete information and insufficient recognition accuracy of a single imaging modality. By performing time series modeling and optical flow estimation on the intraoperative continuous US image sequence, the periodic displacement changes of anatomical structures are identified, and a dynamic deformation field is established with the help of a temporal convolutional neural network to effectively compensate for the structural position offset caused by respiratory movement, improving the spatio-temporal accuracy of target point positioning. The trained deep neural network is used to accurately identify and semantically segment key structures (such as nerves, blood vessels, fascia, etc.). On this basis, by designing an objective function for comprehensive evaluation of multiple factors, the blocked target points with the best anatomical safety and anesthesia effect are automatically evaluated and screened. Finally, through image fusion and three-dimensional visualization technology, the structure segmentation result and the blocked target point are superimposed in the intraoperative US image and matched back to the preoperative three-dimensional structure model to construct a reference model to assist intraoperative navigation operations. Description of the Drawings

[0014] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, but do not constitute a limitation to the present invention. In the drawings: Figure 1 is a flowchart of a method for positioning blocked target points for anesthesia based on multi-source data provided by the present invention; Figure 2 is a structural diagram of a system for positioning blocked target points for anesthesia based on multi-source data provided by the present invention. Detailed Embodiments

[0015] The following describes the preferred embodiments of the present invention with reference to the drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0016] A method for positioning blocked target points for anesthesia based on multi-source data, as Figure 1 shown, includes: S100: Obtain the MRI image of the target area, perform image segmentation and three-dimensional modeling on the MRI image, extract the boundary information of anatomical structures, and construct a preoperative structure model; Further, the steps for constructing the preoperative structure model are: Apply an image enhancement and denoising algorithm to the MRI image to improve the clarity of soft tissue edges; Perform semantic segmentation on MRI images based on a deep neural network model to extract the boundary information of anatomical structures; Input the segmentation results into a 3D reconstruction algorithm, and construct a 3D model of the preoperative anatomical structure through interpolation reconstruction and surface mesh generation.

[0017] Specifically, use a 1.5T or 3.0T magnetic resonance imaging system to perform imaging acquisition on the target area of the patient to obtain high-resolution MRI images including soft tissue structures such as nerves, blood vessels, muscles, and fascia. To improve the clarity and segmentation accuracy of the soft tissue boundaries in the images, the following image preprocessing process is adopted: Perform non-local means denoising on the MRI images to reduce background noise interference; Enhance the gray-scale contrast using histogram equalization or an adaptive contrast enhancement algorithm (such as CLAHE) to highlight the soft tissue boundaries; Perform normalization processing (such as z-score normalization) on the enhanced images to adapt to the input requirements of the subsequent neural network model.

[0018] Input the preprocessed MRI images into a pre-trained deep learning model for anatomical structure segmentation. Preferably, the neural network is a multi-channel semantic segmentation network based on the U-Net structure, and the training dataset is sourced from publicly available or clinical MRI images annotated by experts. The network can perform pixel-level segmentation on the following anatomical structures: peripheral nerve bundles (such as the brachial plexus), vascular structures (arteries, veins), fascia layers (intermuscular septum, fascia sac), bony landmarks, and muscle tissues. The multi-channel probability map output by the network is processed through thresholding to generate a segmentation mask, representing the boundary contours of each type of structure.

[0019] Input the above segmentation mask into the 3D reconstruction module. The specific modeling process is as follows: Perform inter-slice interpolation reconstruction on the segmentation layers of each type of structure to enhance the spatial continuity in the Z-axis direction; Apply the Marching Cubes algorithm or the Poisson surface reconstruction algorithm to extract the isosurface and generate a 3D mesh model; Perform smoothing processing and topological optimization on the 3D mesh to improve the model continuity and visualization quality.

[0020] By introducing MRI images for preoperative high-precision 3D modeling, using a deep neural network to achieve automatic segmentation and boundary extraction of key anatomical structures, the recognition accuracy of soft tissues such as nerves, blood vessels, and fascia is effectively improved; combined with interpolation reconstruction and 3D surface mesh generation technologies, a preoperative structural model with continuous space and clear structure is constructed, providing an accurate anatomical basis for subsequent multi-modal image registration and target localization.

[0021] S200: Obtain the US image of the target area as the intraoperative guidance image. Based on the anatomical structure information between the MRI image and the US image, maximize the mutual information to perform spatial alignment on the MRI image and the US image, and generate a registered multimodal image. Further, the steps for generating the registered multimodal image are as follows: Use the MRI image as the reference image and the intraoperative US image as the moving image. Based on the spatial distribution relationship of the key anatomical structures in the MRI image and the US image, calculate the mutual information value between the two images. Adopt a non-rigid registration algorithm based on maximizing mutual information to perform elastic deformation on the US image, so that the US image is aligned with the MRI image on the key anatomical structures, and generate a registered multimodal image.

[0022] Specifically, preprocess the MRI image and the intraoperative US image respectively to improve the stability and accuracy of registration. Apply edge enhancement and smoothing processing to the MRI image to highlight structures such as nerves, blood vessels, and fascia. Perform Speckle denoising and edge enhancement processing on the US image to reduce the interference of speckle noise. Perform unified resolution resampling and size normalization processing on the two images to ensure the consistency of the registration input.

[0023] Based on the position of the intraoperative US probe and the spatial markers (such as bony structures and fascia interfaces) in the MRI image, determine the initial alignment area, and extract the key anatomical structures within the target area (anatomical structures closely related to the block target, such as nerve plexuses in nerve anesthesia, or fascia in fascia space injection) to narrow the scope of registration calculation.

[0024] Select mutual information as the registration evaluation index, construct a joint histogram to calculate the mutual information value to measure the statistical correlation between the US image and the MRI image after transformation, set the maximization of mutual information as the objective function, and globally optimize the spatial transformation parameters of the US image.

[0025] During the registration process, use B-spline free deformation as the non-rigid transformation model, define the US image on the control grid, optimize the positions of the control points, so as to achieve local elastic deformation, and use the gradient descent optimization algorithm to iteratively update the control point parameters to maximize the mutual information value; to avoid overfitting, a smoothing regularization term can be added to maintain the continuity and physiological rationality of the deformation field.

[0026] Overlay and fuse the finally transformed US image and the MRI image to form a registered multimodal image with consistent structural alignment. The registered image will be used for subsequent target localization and structure recognition steps.

[0027] The multimodal image registration method can make full use of the high soft tissue contrast advantage of MRI and the real-time dynamic characteristics of US. Through the elastic alignment strategy driven by mutual information, it can effectively solve the misalignment problem caused by tissue deformation or probe attitude differences between different modal images, improve the anatomical structure matching accuracy, and provide more accurate image guidance support for subsequent nerve block operations.

[0028] S300: Based on the displacement field of the periodic motion of the anatomical structure identified in the intraoperative US image frame sequence, construct a dynamic displacement model of the target structure, and perform a temporal transformation on the multimodal registration image to generate a compensated multimodal image; Further, the steps of generating the compensated multimodal image are as follows: Based on the continuously acquired intraoperative US image sequence, perform time series modeling on the target area to identify the displacement field of the key anatomical structure in multiple frames; Establish a dynamic deformation field of the key anatomical structure for the displacement field through a temporal convolutional neural network; Apply the dynamic deformation field to the registered multimodal image, perform a temporal spatial transformation on the static structure model, and generate a compensated multimodal image that dynamically changes with respiration.

[0029] Specifically, during the anesthesia operation, use an ultrasound imaging device to continuously acquire the intraoperative US image sequence of the target area, and record this image sequence as , where represents the US image of the th frame acquired at time . Perform Lucas optical flow estimation on the key anatomical structure representing the movement driven by respiration in the image sequence, and extract its displacement field in the consecutive frames. For two consecutive US images and , process them through the Lucas optical flow algorithm to obtain the displacement field of the th frame.

[0030] Model and predict the displacement field of the key anatomical structure. Use a temporal convolutional neural network to learn the deformation law of the target structure in the time dimension. The network input is the displacement field sequence of the target structure in consecutive frames. The network output is the predicted deformation field at the future time, that is, the predicted deformation of the target structure at future time points. The network training objective is to minimize the L2 loss between the predicted deformation field and the real displacement field: It can be expressed as: ; where, represents the loss value, and represents the spatial dimension of the displacement field.

[0031] Apply the predicted dynamic deformation field to the static multimodal images obtained from the previous registration, and perform a spatial transformation on each pixel where is the transformed coordinate. The registered images are corrected for deformation through resampling to generate compensated multimodal images at time .

[0032] By introducing the temporal dynamic modeling of anatomical structures and the temporal-spatial compensation of registered images, the influence of tissue position changes caused by intraoperative respiratory movement on the accuracy of block target recognition is effectively solved. By constructing the displacement field of key structures using a continuous ultrasound image sequence, extracting the structural motion law with a temporal convolutional neural network, generating a dynamic deformation field, and then performing a temporal transformation on the static registered images, the actual position of anatomical structures at different respiratory phases can be reflected in real time, improving the timeliness and spatial consistency of image registration, and helping to improve the accuracy and safety of block target positioning.

[0033] S400: Input the compensated multimodal images into the trained deep neural network model to identify and segment the anatomical structures, and output the structural mask information; Specifically, use the compensated multimodal images, that is, the images combining MRI static structure and US dynamic motion compensation information, as the model input. The input image is a single frame or a multi-frame sequence, and the resolution is consistent with the original US image. The deep neural network model can be a medical image segmentation network based on an encoder-decoder structure. In this embodiment, nnU-Net is preferably used. The model is trained on preoperative or intraoperative datasets, and the manually segmented labels of the labeled MRI and US images and their corresponding structures are used as the supervision signals. The output is a structural mask image with multiple channels, each channel representing an anatomical structure (such as nerves, blood vessels, fascia, pleura, etc.), and the value of each pixel represents the probability or classification label of the corresponding position belonging to a certain structure.

[0034] S500: Obtain multiple candidate injection points within the segmentation mask, design an objective function to score each point, and select one or more points with the maximum score value as the block target; Further, the calculation formula of the objective function is: ; where represents the comprehensive score of the candidate injection point , represents the minimum Euclidean distance between the candidate point and the target nerve, Indicates candidate points The minimum Euclidean distance from blood vessels Indicates candidate points The spatial width between the fascia layers where the candidate points are located , and are weighting coefficients, which can be given according to the guidance of experienced physicians or obtained by fitting through labeled historical data

[0035] By obtaining multiple candidate injection points within the segmentation mask and scoring and screening based on the objective function constructed from comprehensive anatomical features, the scientific nature and safety of the selection of block targets are effectively improved. This method can quantitatively evaluate key factors such as the relative position of each candidate point to the nerve, avoiding important structures (such as blood vessels, pleura), and the width of the fascia space, etc., ensuring that the finally selected injection point can reduce the puncture risk while ensuring the anesthetic effect

[0036] S600: Overlay and display the block target and the structure segmentation result on the intraoperative US image, and perform 3D matching of the overlay result to the preoperative structure model to generate a three-dimensional reference model to assist in clinical anesthesia operation for target positioning

[0037] Furthermore, the steps to generate the three-dimensional reference model are as follows Convert the structure mask and the block target into pseudo-color layers, and use the gradient fusion technology to perform image overlay of the pseudo-color layers and the intraoperative US image to obtain a pseudo-color US image Perform three-dimensional reconstruction of the two-dimensional pseudo-color US image through back-projection reconstruction technology, register it to the three-dimensional preoperative structure model, and generate a three-dimensional reference model

[0038] Specifically, different pseudo-color coding is assigned to the block target and each type of recognized anatomical structure (for example: nerves, blood vessels, pleura, fascia, etc.). For example: block target → yellow; blood vessel structure → red; fascia structure → blue; use an image processing library (such as OpenCV, PyTorch Tensor API) to convert the binary image (or probability map) of the structure mask into a semi-transparent color layer, and the transparency of each structure layer can be controlled within the range of 0.3 - 0.6 to facilitate observing the details of the US image. Align the pseudo-color structure layer and the intraoperative US image in the pixel space (which has been completed based on the aforementioned registration and dynamic compensation), and use weighted overlay to achieve gradient fusion to obtain a pseudo-color US image. The implementation method of weighted overlay can be described as ; wherein represents the fused image represents the intraoperative US image Represents a pseudo-color structure layer Represents the layer transparency.

[0039] By using the spatial tracking information of the B-ultrasound probe, the acquisition pose of each frame of pseudo-color US image is recorded, and the position and orientation of each frame of image in the MRI reference coordinate system are obtained. This pose is defined as: , where is the rotation matrix, is the translation vector. By using the position and attitude parameters of each frame of image above, the pixel intensity in the two-dimensional pseudo-color image is back-projected into the three-dimensional space coordinates of the preoperative structure model. A back-projection method based on the voxel space is adopted to project and superimpose the image content of each frame in the three-dimensional voxel grid . The basic principle is as follows: For each frame of pseudo-color ultrasound image, its pixel corresponding spatial point coordinates can be obtained by back-projection through the following formula: ; where, and represent the pixel pitch, represents the image center.

[0040] The above three-dimensional back-projection result is spatially registered with the three-dimensional anatomical model constructed by preoperative MRI (such as based on the B-spline deformation field or Demons algorithm). After registration, the obtained fusion volume is the three-dimensional reference model containing the key structures, block targets and preoperative anatomical comparison information.

[0041] Example 2 In this example, the brachial plexus block is taken as the target, and a block target positioning system for anesthesia based on multi-source data proposed by the present invention is used to accurately locate the block target around the brachial plexus and guide clinical anesthesia operations. The structure of this system is as Figure 2 shown, including: An MRI image structure extraction module, which is used to obtain the MRI image of the target area, perform image segmentation and three-dimensional modeling on the MRI image, extract the boundary information of the anatomical structure, and construct a preoperative structure model; An image registration module, which is used to obtain the US image of the target area as the intraoperative guiding image, and based on the anatomical structure information between the MRI image and the US image, maximize the mutual information to spatially align the MRI image and the US image to generate a registered multimodal image; An image compensation module, which is used to construct a dynamic displacement model of the target structure based on the displacement field of the periodic motion of the anatomical structure identified in the intraoperative continuous US image frame sequence, and perform a temporal transformation on the multimodal registered image to generate a compensated multimodal image; A mask acquisition module, which is used to input a compensated multimodal image into a trained deep neural network model, identify and segment anatomical structures, and output structural mask information; A block target positioning module, which is used to obtain multiple candidate injection points within the segmented mask, design an objective function to score each point, and select one or more points with the largest score value as the block target; A three-dimensional model reference module, which is used to superimpose the block target and the structure segmentation result on the intraoperative US image, and perform 3D matching on the superimposed result to the preoperative structure model to generate a three-dimensional reference model to assist in clinical anesthesia operations to achieve target positioning.

[0042] The specific implementation process is as follows: Perform high-resolution T1-weighted MRI scanning on the patient's shoulder area to obtain a transverse image sequence. Use a deep neural network model with a UNet structure to perform semantic segmentation on the MRI image, identify key structures such as the brachial plexus, axillary artery, subclavian vein, scalenus muscle, and first rib, and use the Marching Cubes algorithm for three-dimensional surface reconstruction to obtain a preoperative three-dimensional model of the anatomical structure.

[0043] In the anesthesia preparation stage, use a high-frequency linear array probe to obtain real-time B-ultrasound images of the target area, and record the image pose through an electromagnetic tracker. Use structural similarity (such as muscle boundaries, blood vessel distributions, etc.) for image registration, with the maximization of mutual information as the registration criterion, and align the MRI image with the current US image by optimizing the rigid transformation parameters to generate a registered multimodal image.

[0044] Collect a continuous sequence of US image frames within a complete respiratory cycle, perform optical flow estimation on the brachial plexus and surrounding muscle tissues, and extract the inter-frame displacement field. Based on these displacement field data, input a temporal convolutional neural network to establish a dynamic deformation field model to describe the spatial trajectory of the brachial plexus moving with respiration. And apply the deformation field to the registered image to generate an image with dynamic compensation for respiratory motion at each time point, improving the temporal consistency of structural registration.

[0045] Input the dynamically compensated image into the DeepLabv3+ semantic segmentation network to output a structural mask, which includes the precise boundary of the brachial plexus region. Extract multiple candidate injection points based on the segmentation result, construct an objective function to calculate the scores of the candidate injection points, and finally select the point with the highest score as the block target.

[0046] Superimpose the segmentation result and the block target position on each frame of the intraoperative US image and present it in a pseudo-color layer. Through the electromagnetic tracking registration pose information and the back-projection reconstruction algorithm, map the pseudo-color US frame sequence to the three-dimensional space, obtain the voxels of the intraoperative pseudo-color image and register them to the preoperative model to generate a three-dimensional reference model.

[0047] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for positioning anesthetic block targets based on multi-source data, characterized in that Including: Obtain the MRI image of the target area, perform image segmentation and 3D modeling on the MRI image, extract the boundary information of the anatomical structure, and construct a preoperative structure model; Obtain the US image of the target area as the intraoperative guidance image. Based on the anatomical structure information between the MRI image and the US image, use mutual information maximization to spatially align the MRI image and the US image to generate a registered multimodal image; Based on the displacement field of the periodic motion of the anatomical structure identified in the intraoperative continuous US image frame sequence, construct a dynamic displacement model of the target structure, and perform a temporal transformation on the multimodal registration image to generate a compensated multimodal image; Input the compensated multimodal image into the trained deep neural network model to identify and segment the anatomical structure, and output the structure mask information; Obtain multiple candidate injection points within the segmentation mask range, design an objective function to score each point, and select one or more points with the maximum score value as the block target; Overlay the block target and the structure segmentation result on the intraoperative US image, and perform 3D matching of the overlay result to the preoperative structure model to generate a three-dimensional reference model to assist in clinical anesthesia operation to achieve target positioning.

2. The anesthetic block target positioning method based on multi-source data according to claim 1, characterized in that The steps to construct the preoperative structure model are: Apply an image enhancement and denoising algorithm to the MRI image to improve the clarity of the soft tissue edge; Based on the deep neural network model, perform semantic segmentation on the MRI image to extract the boundary information of the anatomical structure; Input the segmentation result into a three-dimensional reconstruction algorithm, and through interpolation reconstruction and surface mesh generation, construct a three-dimensional model of the preoperative anatomical structure.

3. A method for positioning anesthetic block targets based on multi-source data according to claim 1, characterized in that The steps to generate the registered multimodal image are: Use the MRI image as the reference image and the intraoperative US image as the moving image; Based on the spatial distribution relationship of the key anatomical structures in the MRI image and the US image, calculate the mutual information value between the two images; Adopt a non-rigid registration algorithm based on mutual information maximization to perform elastic deformation on the US image, so that the US image is aligned with the MRI image on the key anatomical structures to generate a registered multimodal image.

4. The anesthetic block target positioning method based on multi-source data according to claim 1, characterized in that The steps to generate the compensated multimodal image are: Based on the continuously acquired intraoperative US image sequence, perform time series modeling on the target area to identify the displacement field of the anatomical structure in multiple frames; Establish a dynamic deformation field of the key anatomical structure for the displacement field through a temporal convolutional neural network; Apply the dynamic deformation field to the registered multimodal image, perform a temporal spatial transformation on the static structure model, and generate a compensated multimodal image that dynamically changes with breathing.

5. The anesthetic block target positioning method based on multi-source data according to claim 1, wherein The calculation formula of the objective function is: ； Among them, represents the comprehensive score of the candidate injection point , represents the minimum Euclidean distance between the candidate point and the target nerve, represents the minimum Euclidean distance between the candidate point and the blood vessel, represents the spatial width between the fascia layers where the candidate point is located, , and are weighting coefficients.

6. The anesthetic block target positioning method based on multi-source data according to claim 1, characterized in that The steps to generate the three-dimensional reference model are: Convert the structure mask and the block target into a pseudo-color layer, and use a gradient fusion technique to overlay the pseudo-color layer with the intraoperative US image to obtain a pseudo-color US image; Perform three-dimensional reconstruction on the two-dimensional pseudo-color US image through back-projection reconstruction technology, register it to the three-dimensional preoperative structure model, and generate a three-dimensional reference model.

7. An anesthetic block target positioning system based on multi-source data, characterized in that, Including: An MRI image structure extraction module, used to obtain the MRI image of the target area, perform image segmentation and 3D modeling on the MRI image, extract the boundary information of the anatomical structure, and construct a preoperative structure model; An image registration module, which is used to obtain the US image of the target area as the intraoperative guidance image, and based on the anatomical structure information between the MRI image and the US image, maximizes the mutual information to perform spatial alignment on the MRI image and the US image, and generates a registered multimodal image; An image compensation module, which is used to construct a dynamic displacement model of the target structure based on the displacement field of the periodic motion of the anatomical structure identified in the intraoperative continuous US image frame sequence, and perform temporal transformation on the multimodal registration image to generate a compensated multimodal image; A mask acquisition module, which is used to input the compensated multimodal image into a trained deep neural network model to identify and segment the anatomical structure, and output the structure mask information; A block target location module, which is used to obtain multiple candidate injection points within the segmented mask range, design an objective function to score each point, and select one or more points with the largest score value as the block target; A three-dimensional model reference module, which is used to superimpose and display the block target and the structure segmentation result on the intraoperative US image, and perform 3D matching of the superimposed result to the preoperative structure model to generate a three-dimensional reference model to assist clinical anesthesia operation to achieve target location.

Images