CT big data driven puncture ablation path modeling method and system

Abstract

The invention discloses a CT big data driven puncture ablation path modeling method and system, and the method comprises the steps: constructing and marking a historical CT image database, carrying out the registration and fusion of CT image data in different modes, and generating a three-dimensional voxel feature map; performing intelligent segmentation on the three-dimensional voxel feature map to obtain an obstacle mask and an obstacle probability map; generating a preliminary path set by using the virtual light source model, and generating an expansion obstacle model based on the obstacle mask; and performing hard constraint screening on the initial path set by using an expansion obstacle model, and performing optimization in combination with soft constraint and an objective function to obtain a final path. The system comprises a feature map generation module, an intelligent segmentation module, an expansion obstacle model construction module and a path screening module. According to the invention, image segmentation can be carried out by using deep learning, a high-precision obstacle model is constructed, path screening and adjustment are carried out in combination with an optimization algorithm, and the stability, accuracy and automation degree of path calculation are improved.

Description

Technical Field

[0001] The present invention relates to the field of puncture ablation technology, and in particular to a CT big data-driven puncture ablation path modeling method and system. Background Art

[0002] In the field of modern medical imaging analysis and interventional therapy, CT (computed tomography) imaging technology is widely used for lesion detection, surgical planning, and postoperative evaluation. With the accumulation of medical imaging data and the improvement of computer processing capabilities, big data analysis and intelligent decision-making technologies based on CT images have gradually become research hotspots. In particular, in puncture and ablation procedures, accurate path planning is crucial to improving surgical success rates and reducing intraoperative risks. However, traditional path planning methods rely primarily on the physician's experience, which is subject to significant subjectivity and uncertainty. This makes it difficult to fully utilize the high-dimensional information in CT imaging data, limiting the level of intelligent preoperative planning.

[0003] Currently, path planning methods based on computer image processing and artificial intelligence have made considerable progress, with research focusing on CT image segmentation, obstacle detection, and path optimization. Deep learning methods, such as 3D U-Net, have demonstrated strong automation capabilities in medical image segmentation tasks and can be used to identify target areas and potential obstacles. Furthermore, path planning algorithms, such as A*, Dijkstra, and rapidly exploring random trees (RRT), have been used to calculate interventional therapy pathways.

[0004] However, the existing technology still has the following major problems:

[0005] (1) Multimodal data fusion is difficult. CT image data usually includes scan results with different imaging parameters or at different time points. How to effectively align and fuse these data to generate a stable three-dimensional feature map that can be used for path planning remains a challenge.

[0006] (2) The accuracy of obstacle recognition is limited. Existing medical image segmentation algorithms still have accuracy issues under complex anatomical structures, especially in dense tissue or low-contrast areas. It is difficult to accurately segment key areas, affecting the reliability of path planning.

[0007] (3) The intelligence level of path planning is insufficient. Traditional path planning methods are mostly based on fixed rules or heuristic algorithms, lacking adaptive adjustment capabilities, and it is difficult to take into account the safety, feasibility, and optimality of the puncture path.

[0008] (4) Lack of dynamic optimization mechanism. Existing technologies often adopt static path planning schemes, which are difficult to adjust in real time based on intraoperative feedback, affecting puncture accuracy and operational convenience. Summary of the Invention

[0009] In view of this, the purpose of the embodiments of the present invention is to provide a CT big data-driven puncture ablation path modeling method and system, which can use deep learning for image segmentation, build a high-precision obstacle model, and combine optimization algorithms for path screening and adjustment, thereby improving the stability, accuracy and automation of path calculation.

[0010] A CT big data-driven puncture ablation path modeling method includes:

[0011] A historical CT image database is constructed and labeled, and CT image data of different modalities are registered and fused to generate a three-dimensional voxel feature map.

[0012] Intelligent segmentation is performed on the three-dimensional voxel feature map to obtain an obstacle mask and an obstacle probability map.

[0013] A preliminary path set is generated using the virtual light source model, and an expanded obstacle model is generated based on the obstacle mask.

[0014] The expansion obstacle model is used to perform hard constraint screening on the preliminary path set, and then the soft constraints are combined with the objective function to perform optimization to obtain the final path.

[0015] In a preferred embodiment of the present invention, in the above-mentioned CT big data driven puncture ablation path modeling method, the steps of constructing and labeling a historical CT image database, registering and fusing CT image data of different modalities, and generating a three-dimensional voxel feature map include:

[0016] Acquiring CT image data I i , mark each of the CT image data, and collect the marked CT image data to form a historical CT image database Among them, i is the serial number of the image, N is the total number of samples contained in the database, and P i is the puncture path coordinate sequence corresponding to the i-th CT image data, A i is the anatomical structure label corresponding to the i-th CT image data.

[0017] Image registration is performed on the collected CT image data of different modalities to align the data of each modality in three-dimensional space.

[0018] The weighted superposition method is used to fuse the registered modal data to generate a unified three-dimensional voxel feature map. Among them, V(x,y,z) is the three-dimensional voxel feature map value at the spatial coordinate (x,y,z), I k (x, y, z) is the image data of the kth mode at the spatial coordinate (x, y, z), w k is the weight of the kth modality, and K is the total number of modalities involved in the fusion.

[0019] Its technical effects are: by aligning CT image data of different modalities and aligning the data of different modalities in the same three-dimensional coordinate space, it effectively supplements the shortcomings of a single modality, makes the advantages of each modality image complement each other, and improves the accuracy of puncture path planning; by fusing multi-modal image data to generate a unified three-dimensional voxel feature map, it can more intuitively and completely present the patient's anatomical structure and lesion information, which is helpful for preoperative planning and intraoperative guidance.

[0020] In a preferred embodiment of the present invention, in the above-mentioned CT big data driven puncture ablation path modeling method, the intelligent segmentation of the three-dimensional voxel feature map to obtain the obstacle mask and obstacle probability map includes:

[0021] The three-dimensional voxel feature map V(x, y, z) is input into the first-level 3D U-Net model, and the target area mask M is output. t and target probability map P t (v).

[0022] The three-dimensional voxel feature map V (x, y, z) and the target area mask M t Splicing, input the second-level 3D U-Net model, and output the obstacle mask M o and obstacle probability map P o (v).

[0023] In a preferred embodiment of the present invention, in the above-mentioned CT big data driven puncture ablation path modeling method, the first-level 3D U-Net model includes a first encoder, a first decoder and a first output layer.

[0024] The first encoder extracts multi-scale features through multi-layer 3D convolution and pooling.

[0025] The first decoder restores spatial resolution through deconvolution and skip connections.

[0026] The first output layer generates the target area probability map P through the Sigmoid function t (v) and generate the target area mask by thresholding Among them, τ1 is the binarization threshold, 1 means it belongs to the target area, and 0 means it does not belong to the target area.

[0027] In a preferred embodiment of the present invention, in the above-mentioned CT big data driven puncture ablation path modeling method, the second-level 3D U-Net model includes a second encoder, a second decoder and a second output layer.

[0028] The second encoder extracts multi-scale features through multi-layer 3D convolution and pooling, and the input dimension is increased compared to the first encoder.

[0029] The second decoder restores spatial resolution through deconvolution and skip connections, and the input dimension is increased compared to the first decoder.

[0030] The second output layer generates the obstacle probability map P o (v) and generate obstacle mask by thresholding Among them, τ2 is the binarization threshold, 1 means it belongs to the target area, and 0 means it does not belong to the target area.

[0031] Its technical effect is that the traditional single-stage segmentation method may lead to insufficient segmentation accuracy due to the low contrast between the target area (ablation target) and the surrounding anatomical structures (blood vessels, nerves, bones and other obstacles). The first-level 3D U-Net first processes the three-dimensional voxel feature map, focusing on accurately extracting the target area information, obtaining the target area mask and probability map, and ensuring accurate identification of the lesion area. The second-level 3D U-Net combines the target area mask to further learn and segment the obstacle area, which can effectively suppress the mis-segmentation of non-target areas, thereby improving the accuracy of obstacle recognition. This layer-by-layer refinement segmentation method can enhance the network's understanding of complex anatomical structures, reduce misclassification, and improve the safety and feasibility of overall path planning; the method of first segmenting the target area and then segmenting the obstacles enables the obstacle segmentation network to make full use of the target area information, thereby more accurately distinguishing surgical risk areas, such as large blood vessels, bones, and important nerves, ensuring the safety of the puncture path, and effectively improving the recognition ability of obstacles around complex lesions. It is particularly suitable for puncture planning in complex anatomical areas such as the liver, lungs, and prostate; using 3D U-Net can fully utilize the three-dimensional information of CT images, extract anatomical features over a larger range, enhance the spatial consistency of segmentation, make the boundaries of target areas and obstacles more accurate, and avoid deviations in path planning. It is particularly suitable for scenarios that require cross-layer analysis of anatomical structures, such as deep tumor puncture and complex lesion ablation path planning.

[0032] In a preferred embodiment of the present invention, in the above-mentioned CT big data driven puncture ablation path modeling method, generating a preliminary path set using a virtual light source model and generating an expanded obstacle model based on the obstacle mask includes:

[0033] By calculating the target area mask M t The center of mass of the target area is obtained by calculating the coordinates of the target area center (x0, y0, z0), where ∣M t ∣ is the total number of voxels in the target area, x v ,y v ,z v The coordinates of each voxel are respectively set, and the center coordinates of the target area (x0, y0, z0) are set as the "light source point" of the virtual light source.

[0034] Generate a set of radial preliminary paths Wherein, j is the index number of the path, M is the total number of the generated preliminary paths, r j For each path, the parametric equation is Among them, θ j is the polar angle, φ j is the azimuth angle, and t is the path length parameter.

[0035] Each of the paths r j Discretization to obtain point sequence The point spacing is Δt, where p k =(x0+kΔt·sinθ j cosφ j ,y0+kΔt·sinθ j sinφ j ,z0+kΔt·cosθ j ), k is the index number of the discrete point of the path, and N is the total number of discrete points on a single path.

[0036] The obstacle mask M o Perform 3D morphological expansion operation to generate an expanded obstacle model in, is the morphological dilation operator, B(d safe ) is a spherical structural element with a radius of d safe .

[0037] Its technical effects are: determining the "light source point" through the centroid of the target area ensures the accuracy of path planning, avoids directly using external points of the image as the starting point, reduces the computational complexity of the path search, and makes the generated preliminary path more consistent with the anatomical structure characteristics of the target area; controlling the spatial direction of the path by the polar angle and azimuth angle can achieve omnidirectional path coverage, avoid local optimal solutions, improve the diversity of puncture paths, and discretize the continuous path into a series of point sequences, which is convenient for subsequent path optimization calculations and collision detection, and improves computational efficiency; through morphological expansion operations, the obstacle area is appropriately expanded, the safety margin of the puncture path is increased, the path is avoided from being close to important organs, and it can adapt to obstacles of different sizes. Compared with the method with a fixed expansion radius, it has stronger adaptability; since the preliminary path adopts radial uniform distribution, it can ensure balanced coverage of the search space, avoid redundant calculations in dense path areas, reduce computational complexity, and after obstacle expansion processing, paths that do not cross obstacles can be directly screened out during path optimization, reducing the search space for subsequent path optimization and improving computational efficiency.

[0038] In a preferred embodiment of the present invention, in the above-mentioned CT big data driven puncture ablation path modeling method, the expansion obstacle model is used to perform hard constraint screening on the preliminary path set, and then the soft constraints are combined with the objective function for optimization to obtain the final path including:

[0039] Eliminate the paths that intersect with the expanded obstacle model O′, perform hard constraint screening on the preliminary path set, and retain those that meet The path set R′.

[0040] Establish soft constraints, including path coverage Obstacle overlap penalty and path length

[0041] According to the soft constraints, the objective function S(r j )=α·Co(M t ,r j )-β·Ol(O′,r j )-γ·L(r j ), where α is the target area coverage weight, β is the obstacle avoidance weight, and γ is the path length weight.

[0042] Randomly select a path from the path set R' as the initial population, perform crossover and mutation, generate a new path by adjusting the angle, and adjust the generated new path according to the objective function S(r j ), retain the high-scoring paths, iterate until convergence, and output the final path

[0043] The technical benefits of this approach are as follows: any path that intersects an expanded obstacle area is eliminated, preventing damage to critical anatomical structures during the puncture process and improving puncture safety. This reduces the computational burden of invalid paths, avoids redundant calculations for unlikely paths during subsequent optimization, and improves path optimization efficiency. Three key soft constraints are established: path coverage measures the extent to which a path covers the target area, ensuring that the final path fully reaches the target area and improving the therapeutic effect of the puncture. Obstacle overlap penalty encourages the optimization algorithm to avoid critical obstacles, further reducing surgical risk. Path length avoids excessively long paths, reducing surgical time and improving surgical stability and efficiency. Through weight adjustment and an evolutionary optimization algorithm, the optimal path can be dynamically adjusted according to surgical needs, improving the method's applicability and flexibility.

[0044] In a preferred embodiment of the present invention, the above-mentioned CT big data driven puncture ablation path modeling method further includes extracting the skin surface puncture area according to the final path, dynamically adjusting the shape of the skin surface puncture area based on feedback, and re-optimizing the path.

[0045] In a preferred embodiment of the present invention, in the above-mentioned CT big data driven puncture ablation path modeling method, extracting the skin surface puncture area according to the final path, dynamically adjusting the shape of the skin surface puncture area based on feedback, and re-optimizing the path include:

[0046] Determine the skin surface position z=z according to the CT image data skin , z skin is the height coordinate of the skin surface in three-dimensional space.

[0047] Calculate the final path r opt and the skin surface position z=z skin The intersection of the two points gives the skin puncture point S. entry ={(x,y,z skin )|(x,y,z skin )∈r opt}.

[0048] According to the feedback, the new skin surface puncture area is determined to be an elliptical range, and the puncture point is limited to fall within the elliptical range, which is expressed as Among them, (x c ,y c ) are the coordinates of the center of the ellipse, a is the major axis of the ellipse, and b is the minor axis of the ellipse.

[0049] (x entry ,y entry ,z skin ) is set as the initial puncture point. If the initial puncture point exceeds the range of the ellipse, the skin surface puncture point is projected to the ellipse boundary, which is expressed as

[0050] , update the puncture point to

[0051] The updated puncture point As the end point, the target area center coordinates (x0, y0, z0) are the starting point, and the path direction parameters are recalculated. Generate New Path

[0052] For the new path Perform hard constraint screening, and then optimize by combining soft constraints with the objective function to obtain the final optimization path.

[0053] Its technical effects are: traditional puncture path planning only calculates a fixed skin surface puncture point, while this method dynamically adjusts the skin surface area based on feedback, sets it to an elliptical range, allows adjustment within a certain range, adapts to changes in factors such as patient position and skin elasticity during surgery, reduces puncture point limitations, improves surgical flexibility, and helps doctors make adjustments according to intraoperative conditions; sets dynamic projection of skin surface puncture points to avoid puncture points deviating from reasonable areas and reduce errors caused by changes in skin tissue tension; recalculates the path direction, optimizes the surgical operation angle, and ensures the rationality of the path direction; optimizes the path again to improve the safety and accuracy of the final path, avoiding important structures such as blood vessels and nerves; combines intraoperative imaging data to dynamically adjust the puncture area to improve the real-time and intelligent level of path planning.

[0054] A CT big data-driven puncture ablation path modeling system, comprising:

[0055] The feature map generation module is used to build and label the historical CT image database, align and fuse CT image data of different modalities, and generate a three-dimensional voxel feature map.

[0056] The intelligent segmentation module is used to intelligently segment the three-dimensional voxel feature map to obtain an obstacle mask and an obstacle probability map.

[0057] The expanded obstacle model construction module is used to generate a preliminary path set using the virtual light source model and generate an expanded obstacle model based on the obstacle mask.

[0058] The path screening module is used to use the inflated obstacle model to perform hard constraint screening on the preliminary path set, and then optimize it in combination with the soft constraints and the objective function to obtain the final path.

[0059] The beneficial effects of the embodiments of the present invention are:

[0060] This method builds a historical CT image database and accurately registers and fuses multimodal image data to generate a unified 3D voxel feature map, effectively preserving the high-dimensional information in the image. Furthermore, the use of a hierarchical 3D U-Net model for intelligent segmentation accurately identifies the target area and potential obstacles, significantly improving obstacle detection accuracy. The resulting expanded obstacle model can better mitigate potential risks during path planning, ensuring the safety and rationality of the puncture path.

[0061] This invention uses a virtual light source model to generate a set of radial preliminary paths, which makes the preliminary paths more extensive and provides sufficient candidate solutions for subsequent path screening. Combining hard constraint screening and soft constraint optimization strategies, while retaining sufficient feasible paths, an objective function is established (taking into account target area coverage, obstacle avoidance, and path length) for iterative optimization to achieve the optimal puncture path. This not only overcomes the shortcomings of traditional planning, such as fixed paths and inflexible adjustments, but also enables dynamic adaptive adjustments, improving overall planning efficiency and accuracy.

[0062] For the skin puncture area, this invention calculates CT data to determine the skin surface location, extracts the puncture point, and dynamically adjusts the shape of the skin puncture area based on feedback information. The initial puncture point is constrained within a specified elliptical range, and points outside the range are corrected. The path is then regenerated using the updated puncture points. This enables real-time adjustment of the path planning, better adapting to changes in tissue morphology and anatomical structure in complex clinical environments, and improving the safety and reliability of the procedure.

[0063] This invention divides the entire puncture path modeling process into multiple modules, including feature map generation, intelligent segmentation, expansion obstacle model construction, path screening, and path optimization. Each module functions independently yet works collaboratively. This modular design not only simplifies the overall structure and facilitates implementation and maintenance, but also reduces system manufacturing costs and failure rates, making the system more stable, secure, and scalable in practical applications.

[0064] The entire method and system utilizes computers and other devices for information processing. Deep learning and intelligent optimization algorithms automatically complete image processing and path planning, significantly reducing reliance on the physician's subjective experience and the risk of error due to manual operation. This automated process facilitates operation, improves the standardization of puncture path planning, and contributes to the widespread adoption of high-precision puncture technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0066] Figure 1 This is a flow chart of the CT big data driven puncture ablation path modeling method of the present invention. DETAILED DESCRIPTION

[0067] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of 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. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0068] Please refer to Figure 1 The first embodiment of the present invention provides a CT big data-driven puncture ablation path modeling method, which includes: constructing a historical CT image database and marking it, aligning and fusing CT image data of different modalities to generate a three-dimensional voxel feature map; intelligently segmenting the three-dimensional voxel feature map to obtain an obstacle mask and an obstacle probability map; using a virtual light source model to generate a preliminary path set, and generating an expanded obstacle model based on the obstacle mask; using the expanded obstacle model, performing hard constraint screening on the preliminary path set, and then optimizing it in combination with soft constraints and objective functions to obtain a final path.

[0069] In a preferred embodiment of the present invention, in the above-mentioned CT big data driven puncture ablation path modeling method, the construction of a historical CT image database and marking, registration and fusion of CT image data of different modalities, and generation of a three-dimensional voxel feature map include: collecting CT image data I i , mark each of the CT image data, mark the successful puncture path and the anatomical structure related to the image, and collect the marked CT image data to form a historical CT image database Among them, i is the serial number of the image, N is the total number of samples contained in the database, and P i is the puncture path coordinate sequence corresponding to the i-th CT image data, A i is the anatomical structure label corresponding to the i-th CT image data; image registration is performed on the collected CT image data of different modalities, and each modality data is aligned in three-dimensional space; the registered modality data are fused using a weighted superposition method to generate a unified three-dimensional voxel feature map Among them, V(x,y,z) is the three-dimensional voxel feature map value at the spatial coordinate (x,y,z), I k (x, y, z) is the image data of the kth mode at the spatial coordinate (x, y, z), w k is the weight of the kth modality. The weight value is determined by the mutual information maximization method to reflect the contribution of each modality to the final feature map. K is the total number of modalities involved in the fusion.

[0070] Its technical effects are: by aligning CT image data of different modalities and aligning the data of different modalities in the same three-dimensional coordinate space, it effectively supplements the shortcomings of a single modality, makes the advantages of each modality image complement each other, and improves the accuracy of puncture path planning; by fusing multi-modal image data to generate a unified three-dimensional voxel feature map, it can more intuitively and completely present the patient's anatomical structure and lesion information, which is helpful for preoperative planning and intraoperative guidance.

[0071] In a preferred embodiment of the present invention, in the above-mentioned CT big data driven puncture ablation path modeling method, the intelligent segmentation of the three-dimensional voxel feature map to obtain the obstacle mask and obstacle probability map includes: inputting the three-dimensional voxel feature map V (x, y, z) into the first-level 3D U-Net model, and outputting the target area mask M t and target probability map P t (v); The three-dimensional voxel feature map V (x, y, z) and the target area mask M t Splicing, input the second-level 3D U-Net model, and output the obstacle mask M o and obstacle probability map P o (v).

[0072] In a preferred embodiment of the present invention, in the above-mentioned CT big data driven puncture ablation path modeling method, the first-level 3D U-Net model includes a first encoder, a first decoder and a first output layer; the first encoder extracts multi-scale features through multi-layer 3D convolution and pooling; the first decoder restores spatial resolution through deconvolution and jump connection; the first output layer generates a target area probability map P through a Sigmoid function t (v) and generate the target area mask by thresholding Among them, τ1 is the binarization threshold, 1 means it belongs to the target area, and 0 means it does not belong to the target area.

[0073] In a preferred embodiment of the present invention, in the above-mentioned CT big data driven puncture ablation path modeling method, the second-level 3D U-Net model includes a second encoder, a second decoder and a second output layer; the second encoder extracts multi-scale features through multi-layer 3D convolution and pooling, and the input dimension is increased compared to the first encoder; the second decoder restores spatial resolution through deconvolution and jump connection, and the input dimension is increased compared to the first decoder; the second output layer generates an obstacle probability map P o (v) and generate obstacle mask by thresholding Among them, τ2 is the binarization threshold, 1 means it belongs to the target area, and 0 means it does not belong to the target area.

[0074] Its technical effect is that the traditional single-stage segmentation method may lead to insufficient segmentation accuracy due to the low contrast between the target area (ablation target) and the surrounding anatomical structures (blood vessels, nerves, bones and other obstacles). The first-level 3D U-Net first processes the three-dimensional voxel feature map, focusing on accurately extracting the target area information, obtaining the target area mask and probability map, and ensuring accurate identification of the lesion area. The second-level 3D U-Net combines the target area mask to further learn and segment the obstacle area, which can effectively suppress the mis-segmentation of non-target areas, thereby improving the accuracy of obstacle recognition. This layer-by-layer refinement segmentation method can enhance the network's understanding of complex anatomical structures, reduce misclassification, and improve the safety and feasibility of overall path planning; the method of first segmenting the target area and then segmenting the obstacles enables the obstacle segmentation network to make full use of the target area information, thereby more accurately distinguishing surgical risk areas, such as large blood vessels, bones, and important nerves, ensuring the safety of the puncture path, and effectively improving the recognition ability of obstacles around complex lesions. It is particularly suitable for puncture planning in complex anatomical areas such as the liver, lungs, and prostate; using 3D U-Net can fully utilize the three-dimensional information of CT images, extract anatomical features over a larger range, enhance the spatial consistency of segmentation, make the boundaries of target areas and obstacles more accurate, and avoid deviations in path planning. It is particularly suitable for scenarios that require cross-layer analysis of anatomical structures, such as deep tumor puncture and complex lesion ablation path planning.

[0075] In a preferred embodiment of the present invention, in the above-mentioned CT big data driven puncture ablation path modeling method, the use of the virtual light source model to generate a preliminary path set, based on the obstacle mask, generating an expanded obstacle model includes: calculating the target area mask M t The center of mass of the target area is obtained by calculating the coordinates of the target area center (x0, y0, z0), where ∣M t ∣ is the total number of voxels in the target area, x v ,y v ,z v The coordinates of each voxel are respectively set, and the center coordinates of the target area (x0, y0, z0) are set as the "light source point" of the virtual light source; a radial preliminary path set is generated Wherein, j is the index number of the path, M is the total number of the generated preliminary paths, r j For each path, the parametric equation is Among them, θ j is the polar angle, θ j ∈[0,π], used to control the path elevation angle, φ j is the azimuth, φ j ∈[0,2π], used to control the path direction, t is the path length parameter, t∈[0,t max ], t maxis the maximum extension length; each of the paths r j Discretization to obtain point sequence The point spacing is Δt, where p k =(x0+kΔt·sinθ j cosφ j ,y0+kΔt·sinθ j sinφ j ,z0+kΔt·cosθ j ), k is the index number of the discrete point of the path, N is the total number of discrete points of a single path; for the obstacle mask M o Perform 3D morphological expansion operation to generate an expanded obstacle model in, is the morphological dilation operator, B(d safe ) is a spherical structural element with a radius of d safe .

[0076] Its technical effects are: determining the "light source point" through the centroid of the target area ensures the accuracy of path planning, avoids directly using external points of the image as the starting point, reduces the computational complexity of the path search, and makes the generated preliminary path more consistent with the anatomical structure characteristics of the target area; controlling the spatial direction of the path by the polar angle and azimuth angle can achieve omnidirectional path coverage, avoid local optimal solutions, improve the diversity of puncture paths, and discretize the continuous path into a series of point sequences, which is convenient for subsequent path optimization calculations and collision detection, and improves computational efficiency; through morphological expansion operations, the obstacle area is appropriately expanded, the safety margin of the puncture path is increased, the path is avoided from being close to important organs, and it can adapt to obstacles of different sizes. Compared with the method with a fixed expansion radius, it has stronger adaptability; since the preliminary path adopts radial uniform distribution, it can ensure balanced coverage of the search space, avoid redundant calculations in dense path areas, reduce computational complexity, and after obstacle expansion processing, paths that do not cross obstacles can be directly screened out during path optimization, reducing the search space for subsequent path optimization and improving computational efficiency.

[0077] In a preferred embodiment of the present invention, in the above-mentioned CT big data driven puncture ablation path modeling method, the use of the expansion obstacle model to perform hard constraint screening on the preliminary path set, and then combining the soft constraint and the objective function for optimization to obtain the final path includes: eliminating the path that intersects with the expansion obstacle model O′, performing hard constraint screening on the preliminary path set, and retaining the paths that meet the requirements. The path set R′; establish soft constraints, including path coverage Obstacle overlap penalty and path length According to the soft constraints, the objective function S(r j )=α·Co(Mt ,r j )-β·Ol(O′,r j )-γ·L(r j ), where α is the target area coverage weight, β is the obstacle avoidance weight, and γ is the path length weight; a path is randomly selected from the path set R′ as the initial population, crossover and mutation are performed, and a new path is generated by adjusting the angle, and the generated new path is adjusted according to the objective function S(r j ), retain the high-scoring paths, iterate until convergence, and output the final path

[0078] The technical benefits of this approach are as follows: any path that intersects an expanded obstacle area is eliminated, preventing damage to critical anatomical structures during the puncture process and improving puncture safety. This reduces the computational burden of invalid paths, avoids redundant calculations for unlikely paths during subsequent optimization, and improves path optimization efficiency. Three key soft constraints are established: path coverage measures the extent to which a path covers the target area, ensuring that the final path fully reaches the target area and improving the therapeutic effect of the puncture. Obstacle overlap penalty encourages the optimization algorithm to avoid critical obstacles, further reducing surgical risk. Path length avoids excessively long paths, reducing surgical time and improving surgical stability and efficiency. Through weight adjustment and an evolutionary optimization algorithm, the optimal path can be dynamically adjusted according to surgical needs, improving the method's applicability and flexibility.

[0079] In a preferred embodiment of the present invention, the above-mentioned CT big data driven puncture ablation path modeling method further includes extracting the skin surface puncture area according to the final path, dynamically adjusting the shape of the skin surface puncture area based on feedback, and re-optimizing the path.

[0080] In a preferred embodiment of the present invention, in the above-mentioned CT big data driven puncture ablation path modeling method, the extracting of the skin surface puncture area according to the final path, dynamically adjusting the shape of the skin surface puncture area based on feedback, and re-optimizing the path include: determining the skin surface position z=z according to the CT image data skin , z skin is the height coordinate of the skin surface in three-dimensional space; calculate the final path r opt and the skin surface position z=z skin The intersection of the two points gives the skin puncture point S. entry ={(x,y,z skin )|(x,y,z skin )∈r opt According to the feedback, the new skin surface puncture area is determined to be an elliptical range, and the puncture point is limited to fall within the elliptical range, which is expressed as Among them, (x c ,yc ) is the coordinate of the center of the ellipse, a is the major axis of the ellipse, and b is the minor axis of the ellipse; (x entry ,y entry ,z skin ) is set as the initial puncture point. If the initial puncture point exceeds the range of the ellipse, the skin surface puncture point is projected to the ellipse boundary, which is expressed as

[0081] , update the puncture point to The updated puncture point As the end point, the target area center coordinates (x0, y0, z0) are the starting point, and the path direction parameters are recalculated. Generate New Path For the new path Perform hard constraint screening, and then optimize by combining soft constraints with the objective function to obtain the final optimization path.

[0082] Its technical effects are: traditional puncture path planning only calculates a fixed skin surface puncture point, while this method dynamically adjusts the skin surface area based on feedback, sets it to an elliptical range, allows adjustment within a certain range, adapts to changes in factors such as patient position and skin elasticity during surgery, reduces puncture point limitations, improves surgical flexibility, and helps doctors make adjustments according to intraoperative conditions; sets dynamic projection of skin surface puncture points to avoid puncture points deviating from reasonable areas and reduce errors caused by changes in skin tissue tension; recalculates the path direction, optimizes the surgical operation angle, and ensures the rationality of the path direction; optimizes the path again to improve the safety and accuracy of the final path, avoiding important structures such as blood vessels and nerves; combines intraoperative imaging data to dynamically adjust the puncture area to improve the real-time and intelligent level of path planning.

[0083] The second embodiment of the present invention provides a CT big data-driven puncture ablation path modeling system, which includes: a feature map generation module, used to construct and mark a historical CT image database, align and fuse CT image data of different modalities, and generate a three-dimensional voxel feature map; an intelligent segmentation module, used to intelligently segment the three-dimensional voxel feature map to obtain an obstacle mask and an obstacle probability map; an expanded obstacle model construction module, used to generate a preliminary path set using a virtual light source model, and generate an expanded obstacle model based on the obstacle mask; a path screening module, used to use the expanded obstacle model to perform hard constraint screening on the preliminary path set, and then optimize it in combination with soft constraints and objective functions to obtain the final path.

[0084] The computer program product of the CT big data-driven puncture ablation path modeling method and device provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the method in the previous method embodiment. The specific implementation can be found in the method embodiment and will not be repeated here.

[0085] Specifically, the storage medium can be a general storage medium, such as a mobile disk, a hard disk, etc. When the computer program on the storage medium is run, it can execute the above-mentioned CT big data-driven puncture ablation path modeling method, thereby overcoming the defects of traditional methods such as data fusion difficulties, inaccurate obstacle recognition, and inflexible path planning adjustments, and achieving beneficial technical effects of low failure rate, easy operation, safety and reliability, energy saving and environmental protection, providing an efficient and intelligent solution for CT image-driven puncture ablation path planning.

[0086] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0087] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A CT big data driven puncture ablation path modeling method, characterized by: include: Build and label a historical CT image database, register and fuse CT image data of different modalities, and generate a 3D voxel feature map; Intelligently segmenting the three-dimensional voxel feature map to obtain an obstacle mask and an obstacle probability map; generating a preliminary path set using a virtual light source model, and generating an expanded obstacle model based on the obstacle mask; The expansion obstacle model is used to perform hard constraint screening on the preliminary path set, and then the soft constraints are combined with the objective function to perform optimization to obtain the final path.

2. The CT big data driven puncture ablation path modeling method according to claim 1, characterized in that: The steps of constructing and labeling a historical CT image database, registering and fusing CT image data of different modalities, and generating a three-dimensional voxel feature map include: Acquiring CT image data I i , mark each of the CT image data, and collect the marked CT image data to form a historical CT image database Among them, i is the serial number of the image, N is the total number of samples contained in the database, and P i is the puncture path coordinate sequence corresponding to the i-th CT image data, A i is the anatomical structure label corresponding to the i-th CT image data; Perform image registration on the collected CT image data of different modalities and align the data of each modality in three-dimensional space; The weighted superposition method is used to fuse the registered modal data to generate a unified three-dimensional voxel feature map. Among them, V(x,y,z) is the three-dimensional voxel feature map value at the spatial coordinate (x,y,z), I k (x, y, z) is the image data of the kth mode at the spatial coordinate (x, y, z), w k is the weight of the kth modality, and K is the total number of modalities involved in the fusion.

3. The CT big data driven puncture ablation path modeling method according to claim 2, characterized in that: The intelligent segmentation of the three-dimensional voxel feature map to obtain an obstacle mask and an obstacle probability map includes: The three-dimensional voxel feature map V(x, y, z) is input into the first-level 3D U-Net model, and the target area mask M is output. t and target probability map P t (v); The three-dimensional voxel feature map V (x, y, z) and the target area mask M t Splicing, input the second-level 3D U-Net model, and output the obstacle mask M o and obstacle probability map P o (v).

4. The CT big data driven puncture ablation path modeling method according to claim 3, characterized in that: The first-level 3D U-Net model includes a first encoder, a first decoder, and a first output layer; The first encoder extracts multi-scale features through multiple layers of 3D convolution and pooling; The first decoder restores spatial resolution through deconvolution and skip connections; The first output layer generates the target area probability map P through the Sigmoid function t (v) and generate the target area mask by thresholding Among them, τ1 is the binarization threshold, 1 means it belongs to the target area, and 0 means it does not belong to the target area.

5. The CT big data driven puncture ablation path modeling method according to claim 4, characterized in that: The second-level 3D U-Net model includes a second encoder, a second decoder, and a second output layer; The second encoder extracts multi-scale features through multi-layer 3D convolution and pooling, and the input dimension is increased compared to the first encoder; The second decoder restores spatial resolution through deconvolution and skip connections, and the input dimension is increased compared to the first decoder; The second output layer generates the obstacle probability map P o (v) and generate obstacle mask by thresholding Among them, τ2 is the binarization threshold, 1 means it belongs to the target area, and 0 means it does not belong to the target area.

6. The CT big data driven puncture ablation path modeling method according to claim 3, characterized in that: The generating of a preliminary path set by using a virtual light source model and generating an expanded obstacle model based on the obstacle mask includes: By calculating the target area mask M t The center of mass of the target area is obtained by calculating the coordinates of the target area center (x0, y0, z0), where ∣M t ∣ is the total number of voxels in the target area, x v ,y v ,z v The coordinates of each voxel are respectively set as the "light source point" of the virtual light source. Generate a set of radial preliminary paths Wherein, j is the index number of the path, M is the total number of the generated preliminary paths, r j For each path, the parametric equation is Among them, θ j is the polar angle, φ j is the azimuth angle, t is the path length parameter; Each of the paths r j Discretization to obtain point sequence The point spacing is Δt, where p k =(x0+kΔt·sinθ j cosφ j ,y0+kΔt·sinθ j sinφ j ,z0+kΔt·cosθ j ), k is the index number of the discrete point of the path, and N is the total number of discrete points of a single path; The obstacle mask M o Perform 3D morphological expansion operation to generate an expanded obstacle model in, is the morphological dilation operator, B(d safe ) is a spherical structural element with a radius of d safe .

7. The CT big data driven puncture ablation path modeling method according to claim 6, characterized in that: The expansion obstacle model is used to perform hard constraint screening on the preliminary path set, and then the soft constraints and the objective function are combined for optimization to obtain the final path, which includes: Eliminate the paths that intersect with the expanded obstacle model O′, perform hard constraint screening on the preliminary path set, and retain those that meet The path set R′; Establish soft constraints, including path coverage Obstacle overlap penalty and path length According to the soft constraints, the objective function S(r j )=α·Co(M t ,r j )-β·Ol(O′,r j )-γ·L(r j ), where α is the target area coverage weight, β is the obstacle avoidance weight, and γ is the path length weight; Randomly select a path from the path set R' as the initial population, perform crossover and mutation, generate a new path by adjusting the angle, and adjust the generated new path according to the objective function S(r j ), retain the high-scoring paths, iterate until convergence, and output the final path 8. The CT big data driven puncture ablation path modeling method according to claim 7, characterized in that: The method further includes extracting a skin surface puncture area according to the final path, dynamically adjusting the shape of the skin surface puncture area based on feedback, and re-optimizing the path.

9. The CT big data driven puncture ablation path modeling method according to claim 8, characterized in that: Extracting the skin puncture area according to the final path, dynamically adjusting the shape of the skin puncture area based on feedback, and re-optimizing the path include: Determine the skin surface position z=z according to the CT image data skin , z skin is the height coordinate of the skin surface in three-dimensional space; Calculate the final path r opt and the skin surface position z=z skin The intersection of the two points gives the skin puncture point S. entry ={(x,y,z skin )|(x,y,z skin )∈r opt }; According to the feedback, the new skin surface puncture area is determined to be an elliptical range, and the puncture point is limited to fall within the elliptical range, which is expressed as Among them, (x c ,y c ) are the coordinates of the center of the ellipse, a is the major semi-axis of the ellipse, and b is the minor semi-axis of the ellipse; (x entry ,y entry ,z skin ) is set as the initial puncture point. If the initial puncture point exceeds the range of the ellipse, the skin surface puncture point is projected to the ellipse boundary, which is expressed as , update the puncture point to The updated puncture point As the end point, the target area center coordinates (x0, y0, z0) are the starting point, and the path direction parameters are recalculated. Generate New Path For the new path Perform hard constraint screening, and then optimize by combining soft constraints with the objective function to obtain the final optimization path.

10. A CT big data driven puncture ablation path modeling system, characterized by: include: The feature map generation module is used to build and label the historical CT image database, align and fuse CT image data of different modalities, and generate a three-dimensional voxel feature map; An intelligent segmentation module, configured to intelligently segment the three-dimensional voxel feature map to obtain an obstacle mask and an obstacle probability map; an expanded obstacle model construction module, configured to generate a preliminary path set using a virtual light source model, and generate an expanded obstacle model based on the obstacle mask; The path screening module is used to use the inflated obstacle model to perform hard constraint screening on the preliminary path set, and then optimize it in combination with the soft constraints and the objective function to obtain the final path.

Images