A Precision Brain Tumor Resection Path Planning System Integrating Ultrasound and Electromagnetic Tracking

Abstract

This invention discloses a precise brain tumor resection path planning system integrating ultrasound and electromagnetic tracking, relating to the field of medical image processing. The invention includes: preoperative multimodal image segmentation and initial path planning; intraoperative real-time ultrasound image acquisition and feature point extraction; construction of a deformation interpolation model based on tissue type-weighted anisotropic kernel functions, introducing a divergence constraint regularization term; construction of a continuous three-dimensional deformation field within the region of interest using a hierarchical progressive strategy; application of the deformation field to the initial path for dynamic correction; control of the update frequency through an adaptive triggering mechanism, and real-time display of path risks using color coding. This invention achieves dynamic path correction in cases of intraoperative brain drift, contributing to improved accuracy in surgical path planning.

Description

Technical Field

[0001] This invention relates to the field of medical image processing, and more specifically, to a brain tumor precise resection path planning system that integrates ultrasound and electromagnetic tracking. Background Technology

[0002] In brain tumor resection surgery, the preoperatively planned path often becomes invalid due to intraoperative brain drift, leading to a decrease in surgical precision.

[0003] Existing surgical path planning methods primarily rely on preoperative images for static planning, which struggles to address intraoperative tissue deformation. Chinese patent CN105147362A discloses a method for planning surgical incisions and access routes for brain tumors, determining the incision location by calculating the shortest distance from the tumor center to the scalp. However, this method is based on static geometric calculations and does not consider intraoperative tissue displacement. Chinese patent CN110993065A discloses an image-guided keyhole surgery path planning method for brain tumors, using preoperative multimodal images for risk assessment. However, this method still falls under the category of preoperative planning and cannot provide real-time path correction during surgery. None of these methods solve the problem of path deviation caused by intraoperative brain drift, making it difficult to meet the real-time requirements of precision surgery. Therefore, a precise brain tumor resection path planning system integrating ultrasound and electromagnetic tracking is proposed to address the above problems. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a brain tumor precise resection path planning system that integrates ultrasound and electromagnetic tracking, aiming to solve the problem of preoperative planning path failure caused by intraoperative brain drift during brain tumor resection surgery.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A brain tumor precise resection path planning system integrating ultrasound and electromagnetic tracking includes the following steps: S1 acquires preoperative T1-weighted MRI, diffusion tensor imaging, and magnetic resonance angiography data of the patient, segments and reconstructs a three-dimensional model of the brain tumor, blood vessels, ventricles, and white matter fiber tracts, and plans the initial surgical path in the preoperative image space to establish surgical targets and key anatomical structures to be avoided.

[0006] S2 establishes the transformation relationship between the preoperative image coordinate system and the electromagnetic tracking system coordinate system, and calibrates the intraoperative ultrasound probe to obtain the mapping relationship of ultrasound image pixels in the electromagnetic tracking system coordinate system, which is used to accurately locate the intraoperative ultrasound image to a unified spatial coordinate system.

[0007] S3 During the operation, ultrasound image sequences are acquired in real time and the probe spatial pose is recorded synchronously. Based on the probe spatial pose, virtual slices corresponding to each frame of ultrasound image space are obtained by resampling from the preoperative images. A set of common anatomical feature points are extracted from the ultrasound images and corresponding virtual slices through image matching, and the three-dimensional coordinates of each feature point at the current time and its displacement vector relative to the position of the preoperative image are determined, thereby obtaining a sparse displacement vector field describing tissue deformation.

[0008] S4 constructs a continuous three-dimensional deformation field covering the annular region of interest surrounding the tumor based on the extracted feature points and their displacement vectors, which is used to describe the non-uniform deformation of brain tissue during surgery.

[0009] S5 applies the three-dimensional deformation field to the initial surgical path to generate the dynamically corrected current surgical path, and outputs it for surgical navigation display, so that the surgical path can be updated in real time with tissue displacement.

[0010] Furthermore, in step S4, a layered, progressive method is used to construct the continuous three-dimensional deformation field: First, global rigid body displacement is calculated based on skull markers to eliminate the influence of overall head movement. Then, non-rigid deformation estimation is performed only within the annular region of interest extending outward from the tumor boundary, concentrating computational resources on the most complex deformation region. Finally, the deformation field in the annular region of interest is propagated to the peripheral tissues through an elastic decay function to obtain the deformation field of the whole brain, which is used to reduce the amount of computation while ensuring the accuracy of the core region.

[0011] Furthermore, the construction of the continuous three-dimensional deformation field in step S4 also includes: Each feature point is assigned an organization type label, a confidence weight, and a local structure tensor.

[0012] The tissue type label is derived from the preoperative image segmentation results; The confidence weights are based on ultrasound image quality assessment and are used to reduce the impact of low-quality feature points. The local structure tensor is calculated based on gradient information within a neighborhood window of a preset size in the preoperative image and is used to describe the local geometry.

[0013] An tissue-adaptive hybrid interpolation model is adopted. The weighted kernel function is calculated based on the spatial distance between feature points, tissue type, and local geometry. The interpolation coefficients are then solved to obtain the displacement vector of any point in the annular region of interest, so that the interpolation results conform to the biomechanical characteristics of brain tissue.

[0014] Furthermore, the kernel function of the tissue adaptive hybrid interpolation model is a tissue type-weighted anisotropic kernel function, which consists of a stiffness factor determined by the tissue type of the feature points and an anisotropic metric matrix constructed from the local structural tensors of two points. This is used to make the displacement field propagate faster along the direction with stronger tissue continuity and decay faster in the vertical direction.

[0015] Furthermore, when solving for the interpolation coefficients, a constraint condition with zero displacement field divergence is introduced as a regularization term. By solving a system of constrained linear equations, a deformation field that satisfies the incompressibility of the organization is obtained, which is used to avoid generating non-physical local volume changes.

[0016] Furthermore, it also includes an adaptive triggering mechanism: real-time monitoring of the distance between the tip of the surgical instrument and the boundary of the annular region of interest around the tumor, the time interval since the last deformation field update, and the estimated amount of cerebrospinal fluid loss; calculating a comprehensive triggering index based on the distance, time interval, and estimated amount of cerebrospinal fluid loss; and automatically initiating the update process of steps S3 to S5 when the index exceeds a preset threshold, which is used to dynamically control the update frequency according to the surgical progress.

[0017] Furthermore, after constructing the deformation field in step S4, a self-verification step is performed: based on the newly constructed deformation field, the displacement of each feature point is predicted in reverse, and the prediction error is calculated. If the average error or the maximum error exceeds the set threshold, an additional frame of ultrasound image is automatically acquired in the area with a large error, the newly added feature points are extracted, and the deformation field is corrected using an incremental update method to ensure that the accuracy of the deformation field meets the surgical requirements.

[0018] Furthermore, before extracting anatomical feature points in step S3, the positions of three optimal acquisition sections are pre-calculated within the annular region of interest surrounding the tumor based on preoperative images. These optimal acquisition sections are determined by weighted calculation based on the distances from each voxel within the annular region of interest to blood vessels, ventricles, and white matter fiber bundles, ensuring that the feature areas covered by these three sections have the highest feature point distribution density.

[0019] When an update is triggered, the operator is prompted to align the ultrasound probe with the three preset sections in sequence under electromagnetic tracking guidance to acquire images, ensuring that each acquisition yields the ultrasound image with the greatest deformation information.

[0020] Furthermore, in step S5, when the output is used for surgical navigation display, the dynamically corrected current surgical path is superimposed with the updated blood vessel, ventricle, and white matter fiber bundle model, and the distance from each point on the path to the nearest blood vessel, ventricle, and white matter fiber bundle is calculated in real time. The risk distribution on the path is presented in a color-coded manner according to the distance, which is used to provide doctors with a visual reference for the safety of the path.

[0021] Furthermore, the size of the annular region of interest is adaptively determined based on the tumor diameter, and the annular region of interest is divided into a core layer, a transition layer, and an outer layer. Different density feature points and kernel functions with different decay rates are used for deformation interpolation at different layers to optimize computational efficiency while ensuring the accuracy of the core region.

[0022] The technical effects and advantages of this invention are as follows: This invention utilizes intraoperative ultrasound to acquire images in real time and extract anatomical feature points. Combined with an electromagnetic tracking system, the precise spatial location of these feature points is obtained, and their displacement vectors relative to the preoperative state are calculated, providing real-time data input for deformation field construction. Based on this, an interpolation model is constructed using tissue-type weighted anisotropic kernel functions. This model integrates tissue mechanical properties and local geometric structure information, enabling it to more accurately reflect the biomechanical characteristics of brain tissue and improving the interpolation accuracy from sparse feature points to a continuous deformation field.

[0023] This invention introduces a constraint of zero divergence in the displacement field as a regularization term, ensuring that the constructed deformation field satisfies the approximately incompressible physical properties of brain tissue and guaranteeing the physical rationality of the deformation field. Simultaneously, a hierarchical, progressive strategy is employed, concentrating computation on the core region surrounding the tumor, where deformation is most complex, while sparse computation and attenuation propagation are used in the peripheral region. This improves computational efficiency while maintaining accuracy in the core region, enabling the deformation field construction process to be completed within seconds, meeting the real-time requirements of intraoperative surgery.

[0024] This invention applies a constructed deformation field to preoperative surgical path planning, enabling dynamic correction of the surgical path and allowing it to update in real time with tissue displacement. An adaptive triggering mechanism automatically initiates the update process based on instrument position, time interval, and cerebrospinal fluid loss, avoiding unnecessary calculations. The corrected path is overlaid with key structures and color-coded to represent risk distribution, providing physicians with a real-time reference for intuitively assessing path safety. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the overall system workflow of the present invention. Figure 2 This is a flowchart of the hierarchical progressive deformation field construction process of the present invention; Figure 3 This is a branch diagram of the adaptive triggering update mechanism of the present invention; Figure 4 This is a diagram illustrating the ring-shaped region of interest layering processing strategy of the present invention. Detailed Implementation

[0026] 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.

[0027] Example 1 As attached Figures 1 to 4 The system illustrates a precise resection path planning system for brain tumors that integrates ultrasound and electromagnetic tracking. It achieves dynamic correction of the surgical path by using real-time intraoperative ultrasound to sense tissue deformation and combining this with precise electromagnetic tracking positioning. The specific implementation method is as follows: I. System Composition and Hardware Environment This system is built upon existing medical equipment and includes an intraoperative ultrasound diagnostic instrument, an electromagnetic tracking system, and a computer with the system software installed. As a preferred implementation, the ultrasound diagnostic instrument can be a BK Medical 3000 model, a Siemens Acuson series, etc. Electromagnetic tracking systems include NDI Aurora V2 and NDI Polaris series; computer configuration includes an Intel Core i7 processor and 16GB of memory.

[0028] The electromagnetic tracking system includes a magnetic field generator and a 6-DOF positioning sensor mounted on the ultrasonic probe to acquire the probe’s spatial position and orientation in real time.

[0029] The computer connects to the ultrasound diagnostic instrument via an image acquisition card and to the electromagnetic tracking system via a serial interface, simultaneously acquiring ultrasound images and probe pose data. All calculations are performed on the computer, without requiring any hardware modifications to the ultrasound equipment or electromagnetic tracking system.

[0030] II. Preoperative preparation and marking The following preparatory work needs to be done before implementing this method: The first step is to acquire the patient's preoperative imaging data. T1-weighted magnetic resonance imaging (1mm slice thickness, 256×256×192 resolution) and diffusion tensor imaging (DTI) were performed. s / mm 2 (30 directions) and three-dimensional time-flight magnetic resonance angiography.

[0031] All image data were imported into a computer, and three-dimensional surface models of the brain tumor, blood vessels, ventricles, and white matter fiber tracts were reconstructed using medical image segmentation software. The tumor model was used to determine the surgical target, while the models of blood vessels, ventricles, and white matter fiber tracts were used as key anatomical structures to be avoided.

[0032] On a 3D model, an initial surgical path is planned manually by the doctor or automatically by path planning software in the preoperative imaging space. This path enters from a point on the skull surface and terminates at the tumor center or tumor boundary, maintaining a safe distance from all critical structures. As a preferred embodiment, the safe distance is at least 5 mm, but can be adjusted within the range of 3 to 8 mm depending on the specific circumstances.

[0033] The second step is to establish the coordinate system transformation relationship. The magnetic field generator of the electromagnetic tracking system is fixed next to the operating table to ensure that its relative position with the patient's head is stable. Four markers that can be recognized by the electromagnetic tracking system are attached to the patient's scalp, and these markers can also be identified in the preoperative images.

[0034] The coordinates of each marker point in the electromagnetic tracking system coordinate system were collected sequentially using the probe of the electromagnetic tracking system. ( At the same time, mark the corresponding points in the preoperative images. Two sets of corresponding points are obtained.

[0035] Calculate the rigid body transformation matrix using the point pairing algorithm. The input to the point pairing algorithm is two sets of corresponding points. and The output is a rigid body transformation matrix. , making .

[0036] The third step is ultrasonic probe calibration. Prepare a water tank filled with distilled water, and fix a pointed cone-shaped calibration target (a steel ball with a diameter of 0.5 mm) at the bottom of the tank. The position of this target can be accurately measured by the electromagnetic tracking system. Immerse the ultrasonic probe in the water tank so that the ultrasonic image can clearly show the calibration target.

[0037] The electromagnetic tracking system records the sensor pose on the probe and simultaneously acquires an ultrasound image containing the target point.

[0038] Manually mark the pixel coordinates of the target point on the ultrasound image. Based on the known target spatial coordinates Probe sensor pose (rotation matrix) Translation vector ,in For the probe sensor rotation matrix, Given the probe sensor translation vector and image pixel coordinates, solve for the projection matrix. This matrix establishes the pixel coordinates of the ultrasound image. Mapping relationship between the local coordinate system with the probe sensor as the origin: ; By combining the sensor's pose in the electromagnetic tracking system's coordinate system, any ultrasonic image pixel can be converted into a spatial point in the electromagnetic tracking system's coordinate system: .

[0039] III. Intraoperative Real-Time Management Procedures After the surgery begins, perform real-time intraoperative management as follows: S1 Intraoperative data acquisition and feature point extraction.

[0040] The computer acquires ultrasound image sequences in real time and simultaneously records the probe's spatial pose corresponding to each frame. As a preferred implementation, the acquisition rate is set to 15 frames / s, which can be adjusted within the range of 10 to 30 frames / s depending on the probe type.

[0041] For each newly acquired ultrasound image A virtual slice with a completely corresponding spatial location is generated by resampling from preoperative images based on the probe pose. .

[0042] The virtual slice resampling method is as follows: using the plane determined by the probe pose as the slicing plane, trilinear interpolation is performed along this plane on the preoperative 3D image volume data to generate a slice. Grayscale images of the same size and with the same pixel spacing.

[0043] The input for trilinear interpolation is the preoperative three-dimensional volumetric data. and slice plane equation The output is a two-dimensional grayscale image. As a preferred implementation, the image size can be selected as 512×512 pixels, depending on the probe type, with a pixel pitch of 0.2mm.

[0044] For each pair and A feature point detection algorithm is used to extract common anatomical feature points. The principle for selecting feature points is: Significance and reproducibility were observed in both modalities of imaging, such as vascular bifurcation points, ventricular edge inflection points, and tumor boundary protrusions. The specific method employed was as follows: exist Harris corner detection is performed to obtain a candidate point set. , Harris corner detection takes a grayscale image as input. The output is a set of corner coordinates. Harris corner response values ​​are typically 1. In order to determine the magnitude of the response, as a preferred implementation method, a corner response threshold is set. .

[0045] For each candidate point ,exist Centered on the corresponding position in the virtual slice, normalized cross-correlation matching is performed within a search window of a preset size to find the point with the highest matching degree as the corresponding feature point. As a preferred implementation, the search window size is 21×21 pixels (corresponding to an actual range of approximately 4mm), a full search strategy is adopted, and a correlation coefficient threshold of 0.8 is set. Points with a correlation coefficient lower than this value are discarded. This threshold can be adjusted within the range of 0.7 to 0.9 according to the image quality.

[0046] For each pair of successfully matched feature points, record its position in... pixel coordinates Through the probe calibration matrix Calculate the three-dimensional coordinates of the probe in the electromagnetic tracking system coordinate system based on the probe pose. At the same time, based on its Calculate its preoperative three-dimensional coordinates from the pixel coordinates and preoperative image spatial location. .

[0047] The displacement vector of the feature point , The unit is mm. Each time an update is triggered, only feature points extracted from ultrasound images acquired within the current trigger period are used (the trigger period definition is in Part IV). For feature points detected in multiple consecutive frames, deduplication is performed using a spatial distance threshold. If the preoperative coordinates of two feature points If the distance between two points is less than 2mm, they are considered the same feature point, and only one is retained. All feature points extracted in the current triggering cycle are then combined into a set. This serves as the input for step S2. In a preferred implementation, it is expected that 15-30 feature points will be extracted from each frame of the image.

[0048] S2 Deformation field construction. The feature point set extracted in step S1 is used... As input, a continuous three-dimensional deformation field is constructed.

[0049] First, calculate the global rigid body displacement. Using the skull markers as a reference, transform the coordinates of the markers in the preoperative image coordinate system using a transformation matrix. The coordinates are transformed to the electromagnetic tracking system coordinate system and registered with the coordinates of the marker points acquired in real time during the operation. The rigid body transformation matrix caused by the slight movement of the patient's head is then calculated. Translation vector ,in Let be the rigid body transformation rotation matrix. This is the translation vector of the rigid body transformation. Specifically, the iterative nearest-point algorithm is used, with the iteration termination condition set as the change in the transformation matrix being less than... Or the number of iterations reaches 100. This rigid body displacement is applicable to the entire brain, i.e., a rigid body displacement field. ,in These are the coordinates of a point in space.

[0050] Next, define the annular region of interest. With the tumor centroid Centered on the tumor's equivalent radius Determine the radius of the region of interest ,in Based on tumor volume calculate: .Will Defined as satisfying The spatial point set, this region is the core area for deformation estimation.

[0051] Then, attributes are assigned to each feature point. Each feature point in : according to Read the tissue type label from its location in the preoperative image segmentation results. The set of values ​​is {0: tumor, 1: white matter, 2: gray matter, 3: cerebrospinal fluid, 4: blood vessels}; Calculate confidence weights The value ranges from 0 to 1, and the evaluation indicators include image contrast, signal-to-noise ratio, and the angle between the ultrasound incident angle and the tissue interface.

[0052] Image contrast is calculated by taking the local gray-level variance and normalizing it by dividing by the global maximum variance. The signal-to-noise ratio is the ratio of the local region's mean to its standard deviation, normalized to... ,in and These are the minimum and maximum signal-to-noise ratios of all current feature points, respectively. The included angle is calculated by taking the dot product of the probe direction and the surface normal, and the absolute value is directly used as the normalized value.

[0053] The three indicators have equal weights, so they are directly multiplied together to obtain the result. If the product is less than 0.1, the feature point is considered unreliable and is discarded. Calculate the local structure tensor In preoperative imaging, with Centered on a neighborhood window of 5×5×5 voxels, the central difference method is used to calculate the voxels within the window. , , gradient vector of direction The central difference step size is set to a voxel spacing of 1 mm, and the structure tensor matrix is ​​constructed. , It is a 3×3 symmetric positive semi-definite matrix.

[0054] After removing unreliable feature points, update the total number of feature points. And calculate the residual displacement of each feature point. Residual displacement This represents the local deformation after deducting the global rigid body displacement.

[0055] An adaptive hybrid interpolation model is used to solve the non-rigid body displacement field. For Any query point within Its non-rigid displacement vector The weighted interpolation of the residual displacements of all feature points is obtained as follows: ; in This represents the current total number of feature points. The interpolation coefficients (3-dimensional vectors) are to be determined. This is the kernel function.

[0056] Kernel function Defined as an anisotropic kernel function weighted by organization type: ; in The stiffness factor is determined by the tissue type of the feature point and is preset according to the tissue pair list (e.g., 1.0 for tumor-tumor, 0.8 for tumor-white matter, etc.). Let be the anisotropy metric matrix constructed from the two-point local structure tensors. , and All are 3×3 matrices.

[0057] In actual calculations Add a small regularization term ( This ensures reversibility. The essential difference between this kernel function and the traditional isotropic kernel function lies in the introduction of structural mechanical differences and local geometric anisotropy.

[0058] interpolation coefficients This is obtained by solving the following system of linear equations. Due to the kernel function... As a scalar, the three components of the displacement vector can be solved independently, respectively with respect to... , , Establish a system of linear equations using the components: ; in for Matrix, its elements , For all of Composition of components 3D column vector, , similar, Residual displacements at all feature points of Composition of components 3D column vector, , Similarly, the LU decomposition method is used to solve it.

[0059] To satisfy the physical property that brain tissue is approximately incompressible, a constraint of zero divergence in the displacement field is introduced as a regularization term. For balancing parameters.

[0060] Calculate the divergence of the displacement field at the characteristic points. ,in , This indicates that the kernel function is related to the first independent variable. The gradient. Using the sum of squared divergences as a regularization term, we obtain the corrected linear system: ; in Let be the divergence constraint matrix, and its elements be... ; The value ranges from 0.1 to 0.3; in this embodiment, 0.2 is used. The interpolation coefficients are then obtained by solving for them. And thus obtain any point inside Non-rigid displacement Total displacement field .

[0061] for At points outside, an elastic attenuation function is used to propagate the internal deformation field outward. Boundary sphere according to azimuth angle 1° interval, polar angle Sample at 1° intervals to generate discrete point sets For any point ,calculate arrive For all points, the point corresponding to the minimum distance is taken as the nearest projection point. .

[0062] Define the decay function ,in Let be the attenuation space constant, and take . .but displacement .

[0063] S3 Path Correction and Output. This involves adjusting the initial surgical path. All path points on the path are transformed by the deformation field to obtain the corrected current surgical path. : right any point above , .Will The image is overlaid with updated models of blood vessels, ventricles, and white matter fiber tracts, and the distance from each point along the path to the nearest key structure is calculated in real time. ,according to Risks are presented using color coding: mm green, mm yellow, mm red. Used only for navigation display; subsequent triggering decisions are still based on... .

[0064] IV. Adaptive Triggering and Verification Mechanism To improve efficiency, an adaptive triggering mechanism is used to control the deformation field update frequency. The triggering period is defined as the time interval from the start of one trigger to the start of the next. Three parameters are monitored in real time: Surgical instrument tips and Distance of the boundary (mm); Time interval since last update (s); Estimated cerebrospinal fluid loss (mL), estimated by aspirator flow rate and time.

[0065] Comprehensive trigger index The weight , , ,satisfy ,when Exceeding the threshold When (version 2.0) is obtained, the update process will be started automatically.

[0066] After each update, a self-verification process is performed: the displacement of each feature point is predicted in reverse based on the new deformation field, and the error is calculated. The average error was obtained. and maximum error .

[0067] like mm or If the error is mm, then an additional ultrasound image frame is acquired in the region of maximum error, the newly added feature points are extracted, and the Woodbury formula is used to... Perform incremental updates. Each incremental update processes one new point, and this process is repeated until a total of 5 updates or 10 new points have been added. The solution is then completely recalculated.

[0068] V. Pre-calculation of the optimal acquisition section To maximize information gain, preoperatively based on Three optimal acquisition planes are pre-calculated. Discretize the data into a 1mm voxel grid, label candidate feature points (all >2mm from blood vessels, ventricles, and fiber bundles), and calculate their spatial density. .

[0069] For each candidate section (by normal vector) and offset (Definition), calculate the sum of the densities of all voxels on the cross-section, traverse the normal vectors at 10° intervals and offset intervals of 1mm, select the 3 cross-sections with the largest total density and the pairwise included angles >60°, and save the parameters.

[0070] When an update is triggered during surgery, the system guides the surgeon to capture images of these cut surfaces in sequence.

[0071] VI. Layered processing of the circular region of interest Circular Region of Interest Divided into three levels: Core layer : mm; transition layer : mm; outer layer : .

[0072] The layered strategy concentrates computation on the core region with the most complex deformation, while sparse computation is used on the periphery, reducing the computational load to less than one-third of the total computational load. Implementation details: exist Inside, use all Solve for the interpolation coefficients at each feature point. Kernel function cutoff radius mm; exist Inside, use the layer closest to the center. Solving for each feature point , mm; exist Inside, use the layer closest to the center. Solving for each feature point , mm, and stiffness factor Multiply the whole by 0.8.

[0073] The layer center is statically calculated based on preoperative coordinates and is the average of the preoperative coordinates of all feature points within the layer. any point inside First, determine the level to which it belongs, then call the corresponding coefficient set and cutoff radius to calculate. .like exist In addition, the attenuation process is performed as described above.

[0074] Through the above processing, the deformation field can be constructed within 3 seconds, meeting the real-time requirements; In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A brain tumor precise resection path planning system integrating ultrasound and electromagnetic tracking, characterized in that, Includes the following steps: S1 acquires the patient's preoperative T1-weighted MRI, diffusion tensor imaging, and magnetic resonance angiography data, segments and reconstructs a three-dimensional model of the brain tumor, blood vessels, ventricles, and white matter fiber tracts, and plans the initial surgical path in the preoperative image space. S2 Establishes the transformation relationship between the preoperative image coordinate system and the electromagnetic tracking system coordinate system, and calibrates the intraoperative ultrasound probe to obtain the mapping relationship of ultrasound image pixels in the electromagnetic tracking system coordinate system. S3 During the surgery, ultrasound image sequences are acquired in real time and the probe's spatial pose is recorded simultaneously. Based on the probe's spatial pose, virtual slices corresponding to each frame of the ultrasound image are resampled from the preoperative images. Through image matching, a set of common anatomical feature points are extracted from the ultrasound images and the corresponding virtual slices, and the three-dimensional coordinates of each feature point at the current moment and its displacement vector relative to the preoperative image position are determined. S4 Based on the extracted feature points and their displacement vectors, a continuous three-dimensional deformation field covering the annular region of interest surrounding the tumor is constructed. S5 applies the three-dimensional deformation field to the initial surgical path to generate the dynamically corrected current surgical path and outputs it for surgical navigation display.

2. The brain tumor precise resection path planning system integrating ultrasound and electromagnetic tracking according to claim 1, characterized in that, In step S4, a layered, progressive method is used to construct the continuous three-dimensional deformation field: Calculate global rigid body displacement based on skull markers; Non-rigid deformation estimation is performed only within a ring-shaped region of interest extending outward from the tumor boundary; By using an elastic decay function, the deformation field within the annular region of interest is propagated to the surrounding tissues, thus obtaining the deformation field across the entire brain.

3. The brain tumor precise resection path planning system integrating ultrasound and electromagnetic tracking according to claim 1, characterized in that, Step S4, which involves constructing a continuous three-dimensional deformation field, further includes: Each feature point is assigned an organization type label, a confidence weight, and a local structure tensor. The tissue type label is derived from the preoperative image segmentation results, the confidence weight is based on ultrasound image quality assessment, and the local structure tensor is calculated based on gradient information within a neighborhood window of a preset size in the preoperative image. An organization-adaptive hybrid interpolation model is adopted. The weighted kernel function is calculated based on the spatial distance between feature points, the organization type, and the local geometric structure. The interpolation coefficients are then solved to obtain the displacement vector of any point within the annular region of interest.

4. The brain tumor precision resection path planning system integrating ultrasound and electromagnetic tracking according to claim 3, characterized in that, The kernel function of the tissue adaptive hybrid interpolation model is a tissue type-weighted anisotropic kernel function, which consists of a stiffness factor determined by the tissue type of the feature point and an anisotropic metric matrix constructed from the local structure tensors of two points.

5. The brain tumor precise resection path planning system integrating ultrasound and electromagnetic tracking according to claim 3, characterized in that, When solving for the interpolation coefficients, a constraint condition with zero divergence in the displacement field is introduced as a regularization term. By solving the constrained linear equations, the deformation field that satisfies the incompressibility of the structure is obtained.

6. The brain tumor precise resection path planning system integrating ultrasound and electromagnetic tracking according to claim 1, characterized in that, It also includes an adaptive triggering mechanism: real-time monitoring of the distance between the tip of the surgical instrument and the boundary of the annular region of interest around the tumor, the time interval since the last deformation field update, and the estimated amount of cerebrospinal fluid loss; calculating a comprehensive triggering index based on the distance, time interval, and estimated amount of cerebrospinal fluid loss; and automatically starting the update process of steps S3 to S5 when the index exceeds a preset threshold.

7. The brain tumor precise resection path planning system integrating ultrasound and electromagnetic tracking according to claim 1, characterized in that, After constructing the deformation field in step S4, a self-verification step is performed: based on the newly constructed deformation field, the displacement of each feature point is predicted in reverse, the prediction error is calculated, and if the average error or the maximum error exceeds the set threshold, an additional frame of ultrasound image is automatically acquired in the area with a large error, the newly added feature points are extracted, and the deformation field is corrected by incremental update.

8. The brain tumor precision resection path planning system integrating ultrasound and electromagnetic tracking according to claim 1, characterized in that, Before extracting anatomical feature points in step S3, the positions of three optimal acquisition sections are pre-calculated in the annular region of interest around the tumor based on preoperative images. The optimal acquisition sections are determined by weighted calculation based on the distances from each voxel in the annular region of interest to blood vessels, ventricles and white matter fiber bundles, so that the feature areas covered by these three sections have the maximum feature point distribution density. When an update is triggered, the operator is prompted, under electromagnetic tracking guidance, to align the ultrasound probe with the three preset sections in sequence to acquire images.

9. The brain tumor precise resection path planning system integrating ultrasound and electromagnetic tracking according to claim 1, characterized in that, When the output in step S5 is used for surgical navigation display, the dynamically corrected current surgical path is superimposed with the updated blood vessel, ventricle, and white matter fiber bundle model. The distance from each point on the path to the nearest blood vessel, ventricle, and white matter fiber bundle is calculated in real time, and the risk distribution on the path is presented in a color-coded manner according to the distance.

10. The brain tumor precise resection path planning system integrating ultrasound and electromagnetic tracking according to claim 2, characterized in that, The size of the annular region of interest is adaptively determined based on the tumor diameter, and the annular region of interest is divided into a core layer, a transition layer, and a peripheral layer. Different density feature points and kernel functions with different decay rates are used for deformation interpolation at different layers.

Images