Tumor point cloud optimization method and system, tumor ablation planning method and system

Abstract

The application provides a tumor point cloud optimization method and system in tumor ablation planning, which optimizes a tumor point cloud representing a set of three-dimensional voxels of a tumor to obtain an optimized tumor point cloud for tumor ablation planning. The tumor point cloud optimization method comprises the following steps: an acquisition step of acquiring the tumor point cloud and dividing the tumor point cloud into internal points representing a set of three-dimensional voxels inside the tumor and surface points representing a set of three-dimensional voxels on the surface of the tumor; an internal sampling step of down-sampling the internal points to obtain internal sampling points; a surface sampling step of sampling the surface points differently from the internal sampling step to obtain surface sampling points; and an output step of outputting the internal sampling points and the surface sampling points as the optimized tumor point cloud.

Description

Technical Field

[0001] This invention relates to a method and system for optimizing tumor point clouds, as well as a method and system for tumor ablation planning using the tumor point cloud optimization method. Background Technology

[0002] Ablation is a treatment method that uses physical or chemical methods to destroy abnormal tissues or cells to achieve the purpose of treating diseases. Radiofrequency ablation (RFA), microwave ablation (MWA), and cryoablation are currently effective minimally invasive treatments for liver tumors, kidney tumors, and other conditions. To ensure the safety and effectiveness of ablation treatment, precise planning is required to determine the size, shape, and location of the ablation area. Ablation planning is the process of automating the ablation treatment plan using computational technology and medical imaging data. Effective and efficient automated ablation planning targeting the shape of the tumor is crucial for ablation treatment. Therefore, assisting physicians in automatically developing ablation plans is of great significance in ablation therapy.

[0003] Ablation planning typically relies on high-precision patient anatomical data, and point clouds are a common representation of medical anatomical data applicable to tissues such as tumors. A point cloud is a dataset composed of a large number of discrete points that precisely represent the surface or volume of an object in three-dimensional space. Point clouds are usually derived from medical imaging data, such as CT (computed tomography) or MRI (magnetic resonance imaging) scans. The efficiency of ablation planning is significantly negatively correlated with the number of tumor voxels and the point cloud distribution of the tumor; that is, the more tumor voxels and the denser the point cloud distribution, the lower the efficiency of ablation planning.

[0004] Existing technical documents:

[0005] Patent documents

[0006] Patent Document 1: CN105997245A

[0007] Patent Document 2: Japanese Patent Application Publication No. 2010-506621

[0008] Patent Document 3: Japanese Patent Application Publication No. 2016-135177

[0009] Non-patent literature:

[0010] Non-patent document 1: "Point cloud simplification with preserved edge based on normal vector" Huiyan Han, Xie Han, Fusheng Sun, Chunyan Huang (https: / / www.sciencedirect.com / science / article / abs / pii / S0030402615004052) Summary of the Invention

[0011] Problems in the prior art

[0012] However, existing point cloud-based algorithms for automated tumor ablation planning either ignore the shape distribution of the tumor or the impact of the number of tumor point clouds on efficiency.

[0013] For example, Patent Document 1 discloses a method for accurately simulating radiofrequency ablation using ellipsoidal coverage of tumors. This method establishes a relationship between the ablation size and the clustering distance metric using the covariance matrix of the tumor point cloud. Specifically, the covariance matrix is ​​first corrected based on the semi-major and semi-minor axis lengths of the ellipsoid defined by the user. Then, the tumor point cloud data is clustered using the Mahalanobis distance defined by the covariance matrix. Next, the size of the ablation ellipsoid set by the user is used to cover each cluster. If all clusters can be ablated and covered, the number of ablation cycles is output as the minimum value; otherwise, the cluster number is incremented by 1, and the above process is repeated.

[0014] In the method described in Patent Document 1, the clustering of tumor point clouds involves a large number of iterations. Therefore, as shown in Table 1, the inventors' experimental data indicates that the method in Patent Document 1 is more efficient for planning small-volume tumors (e.g., less than 55 ml). However, for large-volume tumors (e.g., more than 67.3 ml), the number of iterations increases significantly, and good efficiency cannot be obtained.

[0015] Table 1 Comparison of iteration counts for the method in Patent Document 1

[0016] serial number Tumor volume (ml) Number of iterations Coverage 1 23.5 89.35 100％ 2 42.6 160.22 100％ 3 55 159.72 100％ 4 67.3 1268.27 100％ 5 81.3 1182.26 100％

[0017] The main reason for the above problem is that larger tumors contain more voxels, thus requiring more time and iterations for clustering. To solve this problem, downsampling can be used to improve the efficiency of clustering.

[0018] However, downsampling introduces new problems. For example, uniform downsampling may miss important information from the point cloud at the edge of the tumor surface, resulting in incomplete ablation coverage of these points. Geometric sampling focuses more on geometric abrupt changes on the surface, avoiding the omissions that may occur with uniform downsampling, but at the cost of uneven sampling results, which will lead to an increase in the number of ablation cycles.

[0019] This invention addresses the problems of the prior art. The purpose of this invention is to provide a tumor point cloud optimization method and system, as well as a tumor ablation planning method and system using this tumor point cloud optimization method, which can optimize the number of tumor point clouds participating in clustering, significantly improving clustering efficiency while ensuring coverage.

[0020] Means for solving technical problems

[0021] According to one aspect of the present invention, a method for optimizing tumor point clouds in tumor ablation planning is provided, which optimizes a tumor point cloud representing a set of three-dimensional voxels of a tumor to obtain an optimized tumor point cloud for tumor ablation planning. The method comprises the following steps: an acquisition step, acquiring the tumor point cloud and dividing it into internal points representing a set of three-dimensional voxels inside the tumor and surface points representing a set of three-dimensional voxels on the tumor surface; an internal sampling step, downsampling the internal points to obtain internal sampling points; a surface sampling step, performing sampling on the surface points differently than that in the internal sampling step to obtain surface sampling points; and an output step, outputting the internal sampling points and the surface sampling points as the optimized tumor point cloud.

[0022] According to another aspect of the present invention, a tumor ablation planning method is provided, which performs tumor ablation planning based on a tumor point cloud representing a set of three-dimensional voxels of a tumor, characterized by comprising the following steps: optimizing the tumor point cloud using the above-described tumor point cloud optimization method to obtain an optimized tumor point cloud; and a planning step of using the optimized tumor point cloud to perform ablation planning for the tumor.

[0023] According to the present invention, the tumor point cloud is divided into internal points and surface points; different sampling methods are used for internal points and surface points to obtain internal sampling points and surface sampling points, and the internal sampling points and surface sampling points are combined as the output of the optimized tumor point cloud. Thus, the present invention can optimize the number of tumor point clouds participating in clustering, and significantly improve the efficiency of clustering while ensuring coverage.

[0024] Alternatively, the present invention can be implemented as a tumor point cloud optimization system and a tumor ablation planning system, or as a computer program product containing a program that enables a computer to execute the above-mentioned tumor point cloud optimization method and the above-mentioned tumor ablation planning method, and a recording medium that records the above-mentioned program. Attached Figure Description

[0025] Figure 1 This is a schematic diagram illustrating the principle of the tumor point cloud optimization method of the present invention.

[0026] Figure 2 This is a flowchart illustrating an example of the tumor point cloud optimization method according to this embodiment.

[0027] Figure 3A and Figure 3B This is a schematic diagram illustrating the specific method of dividing a tumor point cloud into surface points and interior points.

[0028] Figure 4 This is a schematic diagram used to illustrate the rasterization process performed on surface points.

[0029] Figure 5 This is a schematic diagram showing the test results of testing the ablation plan for a liver tumor according to the first embodiment.

[0030] Figure 6 This is a schematic diagram illustrating an example of a user interface for ablation planning according to the present invention.

[0031] Figure 7 This is a flowchart illustrating a tumor point cloud optimization method according to the second embodiment of the present invention.

[0032] Figure 8 This is a flowchart illustrating a tumor point cloud optimization method according to a third embodiment of the present invention.

[0033] Figure 9 This is a block diagram illustrating an example of a tumor point cloud optimization system according to the present invention. Detailed Implementation

[0034] Hereinafter, embodiments of the tumor point cloud optimization method and system of the present invention, and the tumor ablation planning method and system using the tumor point cloud optimization method, will be described in detail with reference to the accompanying drawings. However, the tumor point cloud optimization method and system of the present invention, and the tumor ablation planning method and system using the tumor point cloud optimization method, are not limited to the embodiments shown below. In the following description, the same constituent elements are given common reference numerals and repeated descriptions are omitted.

[0035] (First Embodiment)

[0036] First, refer to Figure 1 The principle of the tumor point cloud optimization method in the first embodiment will be explained.

[0037] Figure 1 This is a schematic diagram illustrating the principle of the tumor point cloud optimization method of the present invention.

[0038] like Figure 1 As shown, the key to the tumor point cloud optimization method of the present invention is to divide the tumor point cloud into internal points representing a set of three-dimensional voxels inside the tumor and surface points representing a set of three-dimensional voxels on the tumor surface, and to perform different downsampling for internal points and surface points.

[0039] Because internal points often overlap during clustering, this embodiment uses a lower sampling density for internal points than for surface points to obtain internal sampling points. Uniform downsampling is employed for internal points in this embodiment. On the other hand, since surface points primarily reflect the shape and distribution of the tumor, a weighted grid voxel method is used for downsampling to obtain surface sampling points, thereby obtaining more key points for clustering processing to ensure coverage. This embodiment uses a combination of internal and surface sampling points to optimize the tumor point cloud for tumor ablation planning.

[0040] Therefore, the tumor point cloud optimization method of this embodiment optimizes the tumor point cloud, which is a set of three-dimensional voxels representing the tumor, to obtain an optimized tumor point cloud for tumor ablation planning. The tumor point cloud optimization method includes the following steps: an acquisition step, in which the tumor point cloud is acquired and divided into internal points representing a set of three-dimensional voxels representing the tumor interior and surface points representing a set of three-dimensional voxels representing the tumor surface; an internal sampling step, in which the internal points are downsampled to obtain internal sampling points; a surface sampling step, in which the surface points are sampled differently than those in the internal sampling step to obtain surface sampling points; and an output step, in which the internal sampling points and the surface sampling points are output as the optimized tumor point cloud.

[0041] The following describes in detail the implementation of the tumor point cloud optimization method of the present invention. Figure 2 This is a flowchart illustrating an example of the tumor point cloud optimization method according to this embodiment.

[0042] like Figure 2 As shown, according to the tumor point cloud optimization method of this embodiment, the first step is to input the acquired tumor point cloud.

[0043] Tumor point clouds consist of a large number of discrete points, each representing the coordinates of the tumor surface or interior. Tumor point clouds can be acquired using techniques known in the art. For example, medical imaging techniques (such as CT and MRI) can be used to acquire three-dimensional tumor data of the patient. The acquired image data can then undergo preprocessing, including denoising, contrast enhancement, and segmentation, to more accurately identify the tumor region. Finally, point cloud data of the tumor region can be extracted from the preprocessed medical image.

[0044] In step 100, the tumor point cloud is divided into interior points and surface points. The specific method for dividing the tumor point cloud into interior and surface points will be described in detail below.

[0045] Next, in step 200, an internal sampling step is performed to uniformly downsample the internal points and obtain internal sampling points.

[0046] Next, in steps 300 to 500, a surface sampling step is performed to sample the surface points in a way that differs from the internal sampling step, thereby obtaining surface sampling points.

[0047] As described above, in this embodiment, taking advantage of the characteristic that surface points mainly reflect the shape and distribution of tumor point clouds, a so-called weighted raster voxel method is used to sample the surface points. Specifically, in step 300, the surface points are first rasterized; then in step 400, the centroid of the surface points in each raster is calculated; and then in step 500, the sampling results of the surface points are obtained according to the weights of the distance of each surface point in the raster from the centroid and the number of surface points in the raster.

[0048] Finally, in step 600, the sampling results of the internal points and the sampling results of the surface points are merged and output as the sampling results of the tumor point cloud.

[0049] The following detailed description is in conjunction with the accompanying drawings. Figure 2 The specific methods for each step in the flowchart.

[0050] Figure 3A and Figure 3B This is a schematic diagram illustrating the specific method of dividing the tumor point cloud into surface points and internal points in step 100.

[0051] like Figure 3A As shown, the 3D bounding box that can contain the entire tumor is first calculated based on the tumor point cloud data.

[0052] Next, the point cloud data within the bounding box is divided into tumor regions and non-tumor regions. This division process can be achieved using known image processing techniques, such as threshold segmentation, region growing, and machine learning classification. These techniques are well known to those skilled in the art and will not be elaborated upon here.

[0053] Next, the regions inside the bounding box are examined on the X, Y, and Z axes respectively. Edge jump points between the obtained tumor and non-tumor regions are detected. The set of these edge jump points is taken as the surface points of the tumor point cloud, and points outside the surface points are taken as the interior points of the tumor point cloud. Edge jump points refer to discontinuities or points with drastic changes in edge detection. Edge jump point detection can be achieved using known and commonly used edge detection algorithms, such as the Sobel operator, the Canny edge detection algorithm, the zero-crossing algorithm, and other edge detection methods based on machine learning. These techniques are well known to those skilled in the art and will not be elaborated upon here.

[0054] Figure 3B The colored region in the image is an example of a tumor region in a slice along the Z-axis, and the black dots are examples of the obtained boundary jump points, i.e., surface points.

[0055] The methods for dividing surface points and interior points are not limited to this. As is known to those skilled in the art, they can also be achieved through other image processing and machine learning algorithms, such as surface reconstruction algorithms, distance-based methods, geometric feature-based methods, point cloud density analysis, normal vector analysis, and region growing algorithms.

[0056] In addition, to obtain a more accurate surface model, the extracted surface points can be smoothed and optimized to reduce the impact of noise and missegmentation or classification.

[0057] The following explains the specific method for performing the internal sampling step in step 200.

[0058] According to this embodiment, uniform downsampling is used to sample internal points. Uniform downsampling is a method to reduce the amount of point cloud data by uniformly selecting a portion of points from the original point cloud.

[0059] Specifically, in this embodiment, the internal points obtained in step 100 are uniformly downsampled using the algorithm shown in Equation 1, thereby obtaining internal sampling points.

[0060]

[0061] In Equation 1, Q is the sampled set of points inside the tumor, p n This represents each internal point in the tumor point cloud. N is the number of points in the internal points, a is the index value of the point in the point cloud, and step is the sampling step size representing the interval between sampling points, where n ≤ N.

[0062] Equation 1 is just one method of uniform downsampling, and the present invention is not limited thereto. Other algorithms well-known to those skilled in the art can be used for uniform downsampling of internal points, such as uniform grid downsampling, spatial pyramid downsampling, minimum distance downsampling, uniform radius downsampling, octree downsampling, etc. The appropriate downsampling method can be selected according to the specific application scenario and the requirements for the point cloud data, which will not be elaborated further here.

[0063] This implementation method reduces the complexity of data processing with a relatively simple algorithm by uniformly downsampling the internal points, while maintaining the overall structure and shape characteristics of the point cloud.

[0064] The following explains the specific method for weighted raster voxel downsampling of surface points in steps 300-500.

[0065] The weighted raster voxel method according to this embodiment is a raster-based weighted downsampling method, including the following steps: Rasterization step 300: using a cubic raster composed of multiple blocks, the surface points are rasterized to obtain blocks corresponding to each point in the surface points as surface blocks, wherein the size of each block corresponds to a point in the tumor point cloud; Centroid calculation step 400: for each cubic raster, the centroid position of each surface block in the cubic raster is calculated; Sampling block selection step 500: for each cubic raster, the distance from each surface block to the calculated centroid position is calculated, and based on the calculated distance and the weight associated with that distance, a predetermined number or less of surface blocks are selected as sampling blocks. The surface points corresponding to the sampling blocks are used as sampling points of the surface points.

[0066] Figure 4 This is a schematic diagram illustrating the rasterization process of surface points in step 300.

[0067] like Figure 4 As shown, the minimum 3D bounding box of the tumor surface points is first calculated, and then the surface of the 3D bounding box is rasterized with a cubic grid of side length S to obtain the set G of cubic grids and the number I of cubic grids.

[0068] Next, check the center G of each cube grid. i K neighborhoods (K = S) 3 -1), to obtain the number L of surface blocks, i.e., the number of blocks containing tumor surface points T in each cubic grid. Figure 4 The right side shows a schematic diagram of a surface block in a cubic grid after the surface of a 3D bounding box is rasterized. The colored blocks in the cubic grid represent the surface points of the tumor voxel points in the block, and these colored blocks are used as surface blocks.

[0069] Then, the centroid M of the surface blocks in each cubic grid is calculated using the following equation 2.i If L = 0, that is, if the cube grid does not contain surface blocks, then the centroid calculation of this cube grid is ignored.

[0070]

[0071] Next, for the cubic grid whose centroid has been calculated, the distance from each surface block in each cubic grid to the calculated centroid is calculated. In this embodiment, the distance is calculated using the Manhattan distance according to the following Equation 3. Where x l y l , z l These are the 3D coordinates of each surface block within the cubic grid, x m y m , z m These are the 3D coordinates of the calculated centroid.

[0072] d l =|x l -x m |+|y l -y m |+|z l -z m | (Equation 3)

[0073] Next, within each cubic grid, the surface blocks are sorted according to their distance from the centroid, and the surface blocks are selected based on the sorting to perform downsampling.

[0074] Specifically, the surface blocks in each cubic grid are downsampled using Equation 4 below. Where P i This represents the result after downsampling the surface patch. (D) j Let J represent the set of surface blocks equidistant from the centroid, and let J be the number of sets of surface blocks at different distances from the centroid. ω j ω represents the selection weight of the surface patch when sampling the surface patch. j It equals 1, otherwise it is 0.

[0075]

[0076] D j The number of surface blocks is not always just one. According to the present invention, when there are multiple surface blocks at the same distance from the center of gravity, the more surface blocks there are, the greater the selection weight is assigned to the surface blocks at the same distance (hereinafter referred to as "alternate surface blocks").

[0077] In this embodiment, the selection weight of candidate surface blocks at the same distance from the centroid is determined by the following algorithm.

[0078] First, based on the direction from the center of gravity towards D... jThe relationship between the vectors derived from the various candidate surface blocks in D (hereinafter referred to as "candidate vectors") will be used to... j Simplified to D' j Specifically, for example, any candidate vector can be used as the basis candidate vector, and D can be performed based on the angle relationship between other candidate vectors and the basis candidate vector. j The process can be simplified as follows: For example, a threshold angle is set for the angle between candidate vectors. If the angle between two candidate vectors is relatively large, exceeding the threshold angle, then both surface patches corresponding to these two candidate vectors are sampled; otherwise, either one is discarded. The above steps are repeated until all candidate vectors have been compared and selected.

[0079] In this embodiment, the aforementioned candidate vector angle threshold can be set according to the step size (step) used in the internal point downsampling. For example, it can be set to be proportional to the step size (step). For instance, when the step size (step) is 3, the angle threshold can be set to 30 degrees, and when the sampling step size (step) is 6, the angle threshold can be adjusted to 60 degrees. By setting the candidate vector angle threshold to be proportional to the step size (step) in the internal point downsampling, the sampling ratio of internal points and external points can be more balanced.

[0080] Therefore, after simplification, Equation 4 above becomes Equation 5:

[0081]

[0082] The surface blocks in each of the I cubic grids are downsampled as described above. Then, the downsampling results of the I cubic grids are accumulated to obtain the sampling results of the surface points shown in Equation 6 below.

[0083]

[0084] Next, in step 600, the combination of the internal sampling points obtained in step 200 and the surface sampling points obtained in step 500 is used as the optimized tumor point cloud output.

[0085] According to a first embodiment of the present invention, the tumor point cloud is divided into internal points and surface points; internal sampling points and surface sampling points are obtained by using different sampling methods for the internal points and surface points, and the combination of internal sampling points and surface sampling points is used as the optimized tumor point cloud output. In particular, the present invention uses grid-based weighted downsampling to sample the surface points. By using the surface sampling points obtained in this way as the optimized tumor point cloud for tumor ablation planning, the present invention can optimize the number of tumor point clouds participating in clustering, and significantly improve the clustering efficiency while ensuring ablation coverage.

[0086] Based on the technical solution of the first embodiment described above, the inventors tested ablation planning for liver tumors and evaluated the test results. The evaluation results were as follows: Figure 5 As shown in the image.

[0087] Figure 5 This is a schematic diagram showing the test results of testing the ablation plan for a liver tumor according to the first embodiment.

[0088] Figure 5 The left side shows the results of ablation planning based on the entire tumor point cloud. Figure 5 The right side shows the result of ablation planning based on the optimized tumor point cloud of the first implementation method. Figure 5 As can be seen, when comparing the results of tumor ablation planning based on the optimized tumor point cloud obtained according to the first embodiment of the present invention with the results of tumor ablation planning based on the entire tumor point cloud, the size and distribution of the ablation spheres are very close. The coverage and clustering efficiency of the present invention are significantly better than those of the prior art.

[0089] The calculation in the first embodiment shows an example of a cubic grid with a side length S = 3, i.e., 3×3, but the side length S of the present invention is obviously not limited to this. The size of the cubic grid can be appropriately adjusted according to the size of the tumor and the size of the minimum 3D bounding box of the tumor surface points calculated in step 300.

[0090] In addition, the sampling step size step for uniform downsampling of points inside the tumor in step 200 can also be set appropriately according to the size of the tumor.

[0091] These settings functions can be integrated, for example, into... Figure 6 The user interface shown is shown. Figure 6 This is a schematic diagram illustrating an example of a user interface for ablation planning according to the present invention. The present invention can also provide users with the function of displaying comparison results when different set values ​​are input through the user interface.

[0092] The tumor point cloud optimization method of this embodiment can be executed in medical image diagnostic devices such as ultrasound diagnostic devices, CT imaging devices, and MRI imaging devices, or it can be executed in a device that exists independently in the form of a workstation.

[0093] (Second Implementation)

[0094] The above describes the first embodiment of the tumor point cloud optimization method of the present invention, but the present invention is not limited thereto. For example, the tumor point cloud optimization method of the first embodiment can also be combined with the traditional manual ablation planning method.

[0095] The second embodiment is a tumor point cloud optimization method that combines the tumor point cloud optimization method of the present invention with the traditional manual ablation planning method. The following describes the method in conjunction with... Figure 7 The second embodiment of the present invention will be described in detail.

[0096] In the description of the second embodiment, the differences from the first embodiment described above will be mainly explained. Furthermore, in the description of the second embodiment, structures identical to those in the first embodiment will be labeled with the same reference numerals and their descriptions will be omitted.

[0097] Figure 7 This is a flowchart illustrating a tumor point cloud optimization method according to a second embodiment of the present invention, wherein steps S10 and S20 enclosed by dashed boxes are newly added steps, and steps 100' and 600' are steps different from those in the first embodiment.

[0098] In step S10, the tumor point cloud optimization method of this embodiment first performs ablation planning on the input tumor point cloud based on manual user operation, and maintains the planning result.

[0099] Specifically, based on the location and direction of the manual ablation plan, the ablation coverage area can be obtained according to the following formula 7:

[0100]

[0101] In the formula, C and C' are the foci of ellipsoidal ablation, which are manually specified by the user. t is the direction of ablation, O is the center of ablation, and a is the semi-major axis.

[0102] Next, in step S20, the ablation plan obtained in step S10 is evaluated to obtain the uncovered area in the planning result, and the points contained in the uncovered area are used as the point cloud that needs to be automatically clustered in subsequent steps.

[0103] Specifically, in step S20, points in the tumor point cloud that are covered by the ablation range planned manually by the user in step S10 are excluded, and the remaining points are used as the point cloud that needs to be automatically clustered.

[0104] Next, in step 100' of this embodiment, only the point cloud that needs to be automatically clustered obtained in step S20 is divided into surface points and interior points, and subsequent steps are performed.

[0105] In addition, the ablation planning in step S10 can generate one ablation range or multiple ablation ranges. In this case, in step S20, the points covered by the multiple ablation ranges are excluded, and the remaining points are used as the point cloud that needs to be automatically clustered.

[0106] That is, the tumor point cloud optimization method of this embodiment also includes the step of the user pre-setting one or more specified ablation ranges in the tumor point cloud by performing pre-planning; in the acquisition step, the tumor point cloud outside the above-mentioned one or more specified ablation ranges is acquired and divided into internal points and surface points.

[0107] In step 600', the points in one or more specified ablation ranges obtained by the user pre-planning in step S10, as well as the internal sampling points obtained in step 200 and the surface sampling points obtained in step 500, are used as the optimized tumor point cloud output.

[0108] The other steps of the tumor point cloud optimization processing in the second embodiment are the same as those in the first embodiment, and detailed descriptions are omitted here.

[0109] The semi-automatic mode of the second implementation makes it easier for users to intervene in the automated ablation plan, thereby improving its practicality in clinical practice. Manually set ablation not only covers areas defined by the physician based on clinical experience, but more importantly, the pre-planned direction and location based on manual ablation aids in directional planning during automated ablation. For example, it allows avoidance of important bones, blood vessels, and organs, which is crucial in ablation planning. This mode can also be used in scenarios where user-pre-performed manual ablation planning is used to plan ablation of points not covered by the automated ablation plan, achieving a further enhanced effect of ablation planning for points not covered by the automated plan.

[0110] According to the second embodiment of the present invention, the tumor point cloud optimization method of the present invention is combined with the traditional manual planning method, thereby achieving the same effect as the first embodiment in ensuring the ablation coverage while significantly improving the clustering efficiency. It can further optimize the number of tumor point clouds participating in clustering, and can assist the direction planning in the automatic ablation planning by manual pre-planning, so as to realize the ablation planning of points that cannot be covered in the automatic ablation planning.

[0111] (Third Implementation)

[0112] The first and second embodiments of the tumor point cloud optimization method of the present invention have been described above, but the present invention is not limited thereto. For example, based on the first and second embodiments, a mode of specifying the number of ablation times can be further added.

[0113] The third embodiment is a method that further adds a specified number of ablation times to the tumor point cloud optimization method of the present invention, which is described below in conjunction with... Figure 8 The third embodiment of the present invention will be described in detail.

[0114] In the description of the third embodiment, the differences from the first and second embodiments described above will be mainly explained. Furthermore, in the description of the third embodiment, structures identical to those in the first and second embodiments will be labeled with the same reference numerals and their descriptions will be omitted.

[0115] Figure 8 This is a flowchart illustrating a tumor point cloud optimization method according to a third embodiment of the present invention, wherein the steps enclosed by the dashed box are the same as those in the first or second embodiment.

[0116] In step S1, the tumor point cloud optimization method of this embodiment first requires the user to specify the number of ablation operations.

[0117] In step 700, clustering is performed based on the number of ablation cycles specified by the user.

[0118] According to this implementation method, if the tumor point cloud can be successfully clustered using the number of ablation cycles specified by the user, the ablation result will be displayed; otherwise, relevant information will be displayed, and the user can adjust the number of ablation cycles and re-cluster the data.

[0119] The other steps of the tumor point cloud optimization processing in the third embodiment are the same as those in the first or second embodiment, and detailed descriptions are omitted here.

[0120] When performing ablation planning, users have the need to consider and compare the results of clustering with different ablation counts by specifying the number of ablation counts. This implementation provides a mode that allows specifying the number of ablation counts to facilitate users in quickly comparing the different effects of different ablation counts.

[0121] According to this embodiment, a mode that allows specifying the number of ablation cycles is provided, thereby achieving the same effect as the first and second embodiments: significantly improving clustering efficiency while ensuring ablation coverage, and optimizing the number of tumor point clouds participating in clustering. In addition, it can provide users with the function of specifying the number of clustering cycles, thereby further improving the ablation coverage through multiple clustering.

[0122] (Other implementation methods)

[0123] In addition to the tumor point cloud optimization methods described in the above embodiments, the present invention can also be implemented as a tumor point cloud optimization system.

[0124] The following describes an embodiment of the tumor point cloud optimization system according to the present invention. The tumor point cloud optimization system of the present invention optimizes a tumor point cloud, which is a set of three-dimensional voxels representing a tumor, to obtain an optimized tumor point cloud for tumor ablation planning.

[0125] Figure 9This is a block diagram illustrating an example of a tumor point cloud optimization system according to the present invention.

[0126] like Figure 9 As shown, the tumor point cloud optimization system 10 of the present invention includes: an acquisition device 1, which acquires a tumor point cloud and divides the tumor point cloud into internal points representing a set of three-dimensional voxels representing the interior of the tumor and surface points representing a set of three-dimensional voxels representing the surface of the tumor; an internal sampling device 2, which downsamples the internal point cloud to obtain internal sampling points; a surface sampling device 3, which performs sampling on the surface points differently than that performed by the internal sampling device 2 to obtain surface sampling points; and an output device 4, which outputs the internal sampling points and the surface sampling points as an optimized tumor point cloud.

[0127] By executing the steps described in the first to third embodiments above, the acquisition device 1, internal sampling device 2, surface sampling device 3, and output device 4 of this embodiment can optimize the number of tumor point clouds participating in clustering, and significantly improve the efficiency of clustering while ensuring the coverage of ablation.

[0128] In addition, the present invention can also be implemented as a method and system for ablation planning using optimized tumor point clouds obtained by tumor point cloud optimization methods or systems.

[0129] That is, the tumor ablation planning method of the present invention performs tumor ablation planning based on a tumor point cloud representing a set of three-dimensional voxels of a tumor, characterized by comprising the following steps: optimizing the tumor point cloud using the above-mentioned tumor point cloud optimization method to obtain an optimized tumor point cloud; and a planning step of using the optimized tumor point cloud to perform tumor ablation planning.

[0130] The tumor ablation planning system of the present invention performs tumor ablation planning based on a tumor point cloud representing a set of three-dimensional voxels of a tumor, characterized in that it comprises: the aforementioned tumor point cloud optimization system for optimizing the tumor point cloud to obtain an optimized tumor point cloud; and a planning device for performing tumor ablation planning using the optimized tumor point cloud.

[0131] In the above embodiments, uniform downsampling is used for internal points, but the present invention is not limited to this, and other downsampling methods other than uniform downsampling can also be used to sample internal points.

[0132] Furthermore, in the above embodiments, the sampling ratio for downsampling internal points can be set according to actual conditions to balance the distribution of the tumor point cloud. In this invention, the sampling rate for downsampling internal points is set according to the side length S of the cubic grid used for weighted downsampling of surface points. For example, it can be set such that the larger the side length S of the cubic grid, the smaller the sampling rate of internal points, that is, the sampling rate of internal points is inversely proportional to S. For example, when using a cubic grid with S=3 for weighted downsampling of surface points, the sampling rate of internal points is preferably 1 / (3*3*3) = 1 / 27. That is, the sampling rate of internal points is preferably P = 1 / S. 3 Experimental results show that this sampling rate for internal points is particularly effective for tumors with a volume of less than 60 ml.

[0133] The image processing, clustering, etc., described in the above embodiments can all be implemented using various methods in the prior art, and detailed descriptions are omitted here.

[0134] This invention can be implemented not only as the aforementioned tumor point cloud optimization method, system, and tumor ablation planning method, system, but also as a computer program product containing a tumor point cloud optimization program and a tumor ablation planning program, or as a medium storing the tumor point cloud optimization program and the tumor ablation planning program.

[0135] The tumor point cloud optimization system and tumor ablation planning system of the present invention can be installed in a medical image diagnostic device, or they can be executed independently. In such cases, the tumor point cloud optimization system and tumor ablation planning system have processing circuits that perform the same processes as described above, and memory that stores programs corresponding to each function and various information. Furthermore, the processing circuit obtains tumor point clouds from a medical image diagnostic device or image storage device, such as an ultrasound diagnostic device, via a network, and performs the aforementioned processing using the obtained tumor point clouds. Here, the processing circuit is a processor that implements the functions corresponding to each program by reading and executing programs from memory.

[0136] The term "processor" used in the above description refers to circuits such as CPUs (Central Processing Units), GPUs (Graphics Processing Units), Application Specific Integrated Circuits (ASICs), and programmable logic devices (e.g., simple programmable logic devices (SPLDs), complex programmable logic devices (CPLDs), and field-programmable gate arrays (FPGAs)). The processor performs its function by reading and executing a program stored in memory. Alternatively, instead of storing the program in memory, the program can be directly loaded into the processor's circuitry. In this case, the processor performs its function by reading and executing the program loaded into the circuitry. Furthermore, the processors in this embodiment are not limited to a single circuit configuration; multiple independent circuits can be combined to form a single processor to perform its function.

[0137] Furthermore, the constituent elements of the devices illustrated in the above embodiments are functional concepts and do not necessarily need to be physically configured as shown in the illustrations. That is, the specific form of the distribution / combination of the devices is not limited to the form shown in the illustrations, and they can be configured, in any unit, functionally or physically, in a distributed / combined manner depending on various loads or usage conditions. Moreover, the processing functions performed by each device can be implemented, in whole or in part, by a CPU and the program parsed and executed by that CPU, or implemented as hardware as interconnect logic.

[0138] Furthermore, the processing method described in the above embodiments can be implemented by executing a pre-prepared processing program by a computer such as a personal computer or workstation. This processing program can be distributed via a network such as the Internet. Alternatively, the processing program can be recorded on a non-transitory recording medium that can be read by a computer, such as a hard disk, floppy disk (FD), CD-ROM, MO, DVD, USB memory, or SD card memory, and executed by the computer after reading it from the non-transitory recording medium.

[0139] Several embodiments of the present invention have been described, but these embodiments are merely illustrative and not intended to limit the scope of the invention. These embodiments can be implemented in a wide variety of other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included within the scope and spirit of the invention, as well as within the scope of the present invention and its equivalents.

Claims

1. A method for optimizing tumor point clouds in tumor ablation planning, comprising optimizing a tumor point cloud, which is a set of three-dimensional voxels representing a tumor, to obtain an optimized tumor point cloud for tumor ablation planning, characterized in that, Includes the following steps: The acquisition step involves acquiring the tumor point cloud and dividing the tumor point cloud into internal points representing a set of three-dimensional voxels inside the tumor and surface points representing a set of three-dimensional voxels on the tumor surface. The internal sampling step involves downsampling the internal points to obtain internal sampling points. The surface sampling step involves sampling the surface points in a manner different from that in the internal sampling step to obtain surface sampling points. as well as In the output step, the internal sampling points and the surface sampling points are output as the optimized tumor point cloud.

2. The tumor point cloud optimization method as described in claim 1, characterized in that, The surface sampling step includes: The rasterization step uses a grid composed of multiple blocks to rasterize the surface points, obtaining blocks corresponding to each point in the surface points, i.e., surface blocks, and the size of the blocks corresponds to the points in the tumor point cloud; The centroid calculation step involves calculating the centroid position of each surface block within a given grid cell. The sampling block selection step involves calculating the distance of each surface block from the centroid position for each grid cell, and selecting a specified number of surface blocks as sampling blocks based on the distance and a distance-related weight. In the output step, the sampling block is used as the surface sampling point.

3. The tumor point cloud optimization method as described in claim 2, characterized in that, The grid is composed of S×S blocks; In the rasterization step, the minimum 3D bounding box of the tumor point cloud is determined, the surface of the minimum 3D bounding box is rasterized using the raster, and the blocks containing tumor surface points in each raster are taken as the surface blocks. In the sampling block selection step, the Manhattan distance is calculated as the distance. In the weighted selection, when there are multiple surface blocks at the same distance, the more surface blocks there are, the greater the selection weight is given to the surface blocks at the same distance. The sampling rate in the internal sampling step is inversely proportional to S.

4. The tumor point cloud optimization method according to any one of claims 1 to 3, characterized in that, It also includes the step of having the user pre-define one or more specified ablation ranges in the tumor point cloud by performing pre-planning; In the acquisition step, one or more tumor point clouds outside the specified ablation range are acquired from the tumor point cloud and divided into internal points and surface points.

5. A tumor ablation planning method, wherein tumor ablation planning is performed based on a tumor point cloud representing a set of three-dimensional voxels of the tumor, characterized in that, Includes the following steps: The tumor point cloud is optimized using the tumor point cloud optimization method according to any one of claims 1 to 4, to obtain the optimized tumor point cloud; and The planning steps involve using the optimized tumor point cloud to plan the ablation of the tumor.

6. The tumor ablation planning method as described in claim 5, characterized in that, In the planning step, the ablation plan is performed by the user pre-specifying the number of ablation attempts.

7. A tumor point cloud optimization system for tumor ablation planning, comprising optimizing a tumor point cloud, which is a set of three-dimensional voxels representing a tumor, to obtain an optimized tumor point cloud for tumor ablation planning, characterized in that, have: The acquisition device acquires the tumor point cloud and divides the tumor point cloud into internal points representing a set of three-dimensional voxels inside the tumor and surface points representing a set of three-dimensional voxels on the tumor surface. An internal sampling device downsamples the internal points to obtain internal sampling points; A surface sampling device performs sampling on the surface points in a manner different from that of the internal sampling device, thereby obtaining surface sampling points; as well as The output device outputs the internal sampling points and the surface sampling points as the optimized tumor point cloud.

8. A tumor ablation planning system, which performs tumor ablation planning based on a tumor point cloud representing a set of three-dimensional voxels of the tumor, characterized in that, have: The tumor point cloud optimization system of claim 7 optimizes the tumor point cloud to obtain an optimized tumor point cloud; and A planning device for ablation planning of the tumor using the optimized tumor point cloud.

Images