Tumor ablation assessment method based on clustering and dynamic weight Monte Carlo sampling

Abstract

The invention relates to the technical field of medical image processing and tumor treatment planning, and discloses a tumor ablation assessment method based on clustering and dynamic weight Monte Carlo sampling, and the method comprises the steps: dividing a tumor, i.e., a healthy region, through spatial clustering, optimizing sample distribution in combination with an error feedback mechanism related to a region density compensation factor, and obtaining a tumor ablation result; introducing a momentum enhanced weight updating strategy, correcting sparse region sampling deviation, adaptively balancing historical errors, rapidly judging a sample membership relationship through an ablation ellipsoid model, and synchronously calculating a tumor ablation coverage rate and a healthy tissue damage rate; the tumor ablation coverage rate and the healthy tissue damage rate can be rapidly, stably and accurately calculated synchronously, and a reliable quantitative basis is provided for real-time planning in an operation and accurate evaluation after the operation.

Description

Technical Field

[0001] This invention relates to the fields of medical image processing and tumor treatment planning technology, and more specifically to a tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling. Background Technology

[0002] In tumor ablation therapy, current efficacy assessments mainly rely on physicians' intuitive judgment of the ablation area through intraoperative observation of medical images (such as CT / MRI-guided images) and indirect evaluation through postoperative image review and patient prognosis. This approach suffers from problems such as poor immediacy, strong subjectivity, and difficulty in precise quantification. To provide a more objective and real-time quantitative assessment, Monte Carlo (MC) methods can calculate spatial distribution through random sampling. However, existing Monte Carlo-based methods typically employ uniform sampling strategies, which are difficult to effectively address the heterogeneity of spatial distribution between tumors and healthy tissues, such as small tumor volume, complex morphology, and wide distribution of healthy tissues. This may lead to insufficient sampling in key areas such as tumor edges, large estimation variance, and slow convergence speed. Especially when the number of sampling points is limited, the accuracy and stability of the assessment results are insufficient. At the same time, there is a lack of efficient collaborative sampling optimization mechanisms for the two closely related but spatially different objectives of ablation coverage and healthy tissue damage rate.

[0003] To address the limitations of the aforementioned clinical assessments and the shortcomings of existing Monte Carlo methods, this invention proposes a tumor ablation assessment method and system based on dynamic weighted Monte Carlo, which can quickly, stably, and more accurately calculate the tumor ablation coverage and healthy tissue damage rate simultaneously, providing a more reliable quantitative basis for intraoperative real-time planning and accurate postoperative assessment. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, this invention provides a tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling, in order to solve the problems existing in the background art.

[0005] This invention provides the following technical solution: a tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling, comprising the following steps:

[0006] Step 1: Obtain voxel data of three-dimensional medical images, distinguish tumor tissue regions from healthy tissue regions using image segmentation algorithms, and calculate volume.

[0007] Step 2: Divide the tumor tissue region into sub-regions based on the spatial clustering algorithm, initialize the sampling weights of each sub-region, and construct a dynamic weight model;

[0008] Step 3: Obtain the weights of each sub-region based on the dynamic weight model and generate a sample set. Combine the ablation parameter model to determine the membership of the samples and obtain the tumor ablation coverage and the degree of damage to healthy tissue.

[0009] Step 4: Dynamically update the weights of each sub-region based on the error feedback mechanism, and optimize the sample distribution by balancing historical errors and current errors through adaptive learning rate and decay factor;

[0010] Step 5: Iterate through steps 2-4 until convergence, output the joint evaluation results of tumor ablation coverage and healthy tissue damage rate, and visualize the ablation boundary.

[0011] Preferably, the volume calculation in step 1 specifically includes:

[0012] Step 11: Obtain the tumor tissue point set and health organization point collection ;

[0013] Step 12: Define the anisotropic scaling factor Converting voxel coordinates to physical space coordinates is represented as: , ; ;in, This is represented as a set of tumor tissue points after being transformed into physical space. Represented as a set of healthy tissue points after being transformed into physical space; They represent axis, shaft and The scaling factor of the axis. This represents element-wise multiplication.

[0014] Step 13: Calculate the tumor volume and the volume of healthy tissue.

[0015] Preferably, the tumor tissue point set is represented as The set of healthy tissue points is represented as ;in, and All are three-dimensional spatial coordinates. , ; The number of voxel points in the tumor tissue region. The number of voxel points in the healthy tissue region; the formula for calculating the tumor volume is expressed as: ;in, The volume of the tumor is represented by the formula for calculating the volume of the healthy tissue: ;in, Indicates the volume of healthy tissue.

[0016] Preferably, step 2, constructing the dynamic weight model, specifically includes:

[0017] Step 21: Perform K-means clustering on the tumor tissue point set and the healthy tissue point set to generate a region label mapping function;

[0018] Step 22: Initialize the tumor region weight vector and healthy region weight vector The tumor region weight vector is represented as: ,in, This represents the weight vector of each subregion of the tumor tissue region. ;satisfy The weight vector of the healthy region is represented as: in, This represents the weight vector of each subregion within the healthy tissue region. ;satisfy .

[0019] Preferably, the region label mapping function of the tumor tissue point set is expressed as:

[0020] ;in, Region labels representing tumor tissue point sets, The number of preset subregions representing the tumor tissue region;

[0021] ;in, The region label represents the set of healthy tissue points. This indicates the number of preset sub-regions within the healthy tissue region.

[0022] Preferably, step 3, generating a sample set and determining the sample membership based on the ablation parameter model, specifically includes:

[0023] Step 31: Generating sub-region weights based on the current tumor tissue region One sampling point:

[0024] ; ;in, This represents the final set of sampling points generated in the tumor tissue region. ,in, Indicates from Randomly selected from One point, Let k represent the set of all points labeled k in the tumor tissue region. Indicates from the first The number of points actually extracted from each sub-region; This represents the floor function; ;

[0025] Step 32, for sample points To determine whether it is located within the ablation ellipsoid, the formula is as follows:

[0026] ;

[0027] in, Indicates an indicator function, These are the coordinates of the ablation ellipsoid center, determined based on the tumor location. , , They represent the ellipsoid in axis, shaft and The semi-axis length in the axial direction is determined by the location and size of the tumor and can be parameterized.

[0028] The coordinates of the center of the ablation ellipsoid are expressed as follows:

[0029] ; ; ;in, Indicates the first The x-axis coordinates of each sampling point; Indicates the first The y-axis coordinates of each sampling point; Indicates the first z-axis coordinates of each sampling point;

[0030] The lengths of each half-shaft are expressed as follows: ; ; ;in, , , Tumor tissue point sets axis, axis, Standard deviation of the axis coordinates This represents the scaling factor. To ensure the ablation effect, the scaling factor takes values ​​within a certain range. ;

[0031] Step 33: Calculate tumor ablation coverage : ;in, Indicates the sampling point currently being judged; Represents the set of sampling points generated from the tumor tissue region. Elements in;

[0032] Calculate the coverage components of each sub-region of the tumor tissue region: ; ;in, Indicates the first A set of sampling points for each tumor subregion;

[0033] Calculate the volume of the ablated tumor : ;

[0034] Step 34: Generate weight vector based on healthy tissue region One sampling point:

[0035] , ;in, This represents the final set of sampling points generated in the healthy tissue region. Let k represent the set of all points labeled k in the healthy tissue region;

[0036] Calculating the degree of damage to healthy tissue : ;in, Indicates the sampling point currently being judged;

[0037] Calculate the damage components of each sub-region within the healthy tissue region. : , ;

[0038] Calculate the volume of damage to healthy tissue : .

[0039] Preferably, step 4, which dynamically updates the weights of each sub-region based on the error feedback mechanism, specifically includes:

[0040] Step 41: Define the regional density measurement method, based on the trace quantization of the regional spatial distribution characteristics of the covariance matrix: ;in, Indicates the first Regional density of each sub-region Represents the matrix trace operation. , The point cloud of tumors represents the first Covariance matrix of each subregion;

[0041] Step 42: Calculate the number of tumor tissue regions. Sub-regions at time step Regional error components , defined as the product of coverage variance and density, using exponential smoothing:

[0042] ;

[0043] in, Indicates the tumor tissue region Sub-regions at time step The regional error components; Indicates the time step. Indicates the error attenuation coefficient;

[0044] Step 43: Update the tumor tissue region weights, which includes two stages: generating the basic weight vector and smoothing the update.

[0045] ;

[0046] in, Represents the weight vector of the tumor tissue region. Indicates the time step Time-healthy tissue region weight vector; , Represents the error vector of the tumor tissue region; This represents the error vector of the tumor tissue subregion R1. Indicates the learning rate. , Indicates the exploration coefficient. .

[0047] Preferably, step 5 further includes:

[0048] The density compensation factor for the damaged area is calculated using the following formula:

[0049] ;in, The density compensation factor represents the damaged area; The first region representing the healthy tissue area The covariance matrix of each subregion Indicates the number of subregions within a healthy tissue region; This represents the truncation function;

[0050] The area error update is expressed by the formula: ;in, , They represent the time steps respectively. , The regional error component of the k-th sub-region of the healthy tissue;

[0051] An independent weight update mechanism is used, represented as follows:

[0052] ;in, Represents the weight vector of healthy tissue regions. Indicates the time step Time-healthy tissue region weight vector; This represents the probability of exploration.

[0053] Preferably, the dynamic sampling framework further includes:

[0054] For regions with high uncertainty ;in, Let R represent the covariance matrix of region R. Indicates a high uncertainty threshold;

[0055] The dynamic sampling framework includes an optimization controller, comprising:

[0056] The learning rate is dynamically adjusted based on the magnitude of weight changes. ;in, Indicates the decay rate. This represents the initial learning rate. Indicates the convergence threshold. , These represent time steps of 1 and 2 respectively. , Weight vector at time;

[0057] and ;in, This is the convergence threshold.

[0058] The technical effects and advantages of this invention are as follows:

[0059] This invention addresses the shortcomings of traditional Monte Carlo methods, such as slow convergence and large variance in heterogeneous spatial distributions, by employing a synergistic mechanism of dynamic weighted Monte Carlo strategy and error feedback mechanism. It achieves precise quantification of both tumor ablation coverage and healthy tissue damage, providing clinicians with accurate and intuitive decision support. Furthermore, it is compatible with ablation models such as ellipsoidal or multi-needle stacking, promoting the evolution of precision medicine from static planning to dynamic evaluation. Attached Figure Description

[0060] Figure 1 This is a flowchart of the tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling of the present invention.

[0061] Figure 2 This is a schematic diagram of a simple liver-tumor-ablation zone case model for the clustering and dynamic weighted Monte Carlo tumor ablation assessment of the present invention.

[0062] Figure 3 Box plot comparing the relative errors of the two algorithms in estimating tumor ablation coverage for small sample points.

[0063] Figure 4 Weight convergence plot for estimating tumor ablation coverage using clustering and dynamic weighted Monte Carlo methods.

[0064] Figure 5The results of estimating tumor ablation coverage using the clustering and dynamic weighted Monte Carlo method are shown in the figure.

[0065] Figure 6 A graph showing the results of estimating the degree of damage in healthy tissue using the clustering and dynamic weighted Monte Carlo method.

[0066] Figure 7 To estimate the convergence plot of tumor ablation coverage using clustering and traditional Monte Carlo methods.

[0067] Figure 8 A convergence plot for estimating the degree of damage in healthy tissue using clustering and traditional Monte Carlo methods.

[0068] Figure 9 Box plot comparing the relative error of tumor ablation coverage in 20 trials with a small sample size under fixed ablation ellipsoid parameters. Detailed Implementation

[0069] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the morphologies of the structures described in the following embodiments are merely illustrative. The tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling involved in the present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Example 1

[0071] like Figure 1 As shown, this invention provides a tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling, comprising the following steps:

[0072] Step 1: Obtain voxel data of three-dimensional medical images, distinguish tumor tissue regions from healthy tissue regions using image segmentation algorithms, and calculate volume.

[0073] Step 2: Divide the tumor tissue region into sub-regions based on the spatial clustering algorithm, initialize the sampling weights of each sub-region, and construct a dynamic weight model;

[0074] Step 3: Obtain the weights of each sub-region based on the dynamic weight model and generate a sample set. Combine the ablation parameter model to determine the membership of the samples and obtain the tumor ablation coverage and the degree of damage to healthy tissue.

[0075] Step 4: Dynamically update the weights of each sub-region based on the error feedback mechanism, and optimize the sample distribution by balancing historical errors and current errors through adaptive learning rate and decay factor;

[0076] Step 5: Iterate through steps 2-4 until convergence, output the joint evaluation results of tumor ablation coverage and healthy tissue damage rate, and visualize the ablation boundary.

[0077] In this embodiment, it should be specifically noted that the volume calculation in step 1 includes:

[0078] Step 11: Obtain the tumor tissue point set and health organization point collection The tumor tissue point set is represented as follows: The set of healthy tissue points is represented as ;in, and All are three-dimensional spatial coordinates. , ; The number of voxel points in the tumor tissue region. The number of voxel points in the healthy tissue region;

[0079] Step 12: Define the anisotropic scaling factor Converting voxel coordinates to physical space coordinates is represented as: , ; ;in, This is represented as a set of tumor tissue points after being transformed into physical space. Represented as a set of healthy tissue points after being transformed into physical space; They represent axis, shaft and The scaling factor of the axis. This represents element-wise multiplication.

[0080] Step 13: Calculate the tumor volume and the volume of healthy tissue;

[0081] The formula for calculating tumor volume is expressed as follows: ;in, The volume of the tumor is represented by the formula for calculating the volume of the healthy tissue: ;in, Indicates the volume of healthy tissue;

[0082] The above calculation process takes into account the anisotropic spatial resolution in medical images, ensuring that the volume calculation results conform to the actual physical dimensions.

[0083] In this embodiment, it should be specifically noted that the construction of the dynamic weight model in step 2 includes:

[0084] Step 21: Perform K-means clustering on the tumor tissue point set and the healthy tissue point set to generate a region label mapping function;

[0085] The region label mapping function for the tumor tissue point set is expressed as follows:

[0086] ;in, Region labels representing tumor tissue point sets, The number of preset subregions representing the tumor tissue region;

[0087] ;in, The region label represents the set of healthy tissue points. Indicates the number of preset subregions of the healthy tissue region;

[0088] Step 22: Initialize the tumor region weight vector and healthy region weight vector The tumor region weight vector is represented as: ,in, This represents the weight vector of each subregion of the tumor tissue region. ;satisfy The weight vector of the healthy region is represented as: in, This represents the weight vector of each subregion within the healthy tissue region. ;satisfy .

[0089] In this embodiment, it should be specifically explained that step 3, which involves generating a sample set and determining the sample membership based on the ablation parameter model, specifically includes:

[0090] Step 31: Generating sub-region weights based on the current tumor tissue region One sampling point:

[0091] ; ;in, This represents the final set of sampling points generated in the tumor tissue region. ,in, Indicates from Randomly selected from One point, Let k represent the set of all points labeled k in the tumor tissue region. Indicates from the first The number of points actually extracted from each sub-region; This represents the floor function; ;

[0092] Step 32, for sample points To determine whether it is located within the ablation ellipsoid, the formula is as follows:

[0093] ;

[0094] in, Indicates an indicator function, These are the coordinates of the ablation ellipsoid center, determined based on the tumor location. , , They represent the ellipsoid in axis, shaft and The semi-axis length in the axial direction is determined by the location and size of the tumor and can be parameterized.

[0095] The coordinates of the center of the ablation ellipsoid are expressed as follows:

[0096] ; ; ;in, Indicates the first sampling points Axis coordinates; Indicates the first sampling points Axis coordinates; Indicates the first sampling points Axis coordinates;

[0097] The lengths of each half-shaft are expressed as follows: ; ; ;in, , , Tumor tissue point sets axis, axis, Standard deviation of the axis coordinates This represents the scaling factor. To ensure the ablation effect, the scaling factor takes values ​​within a certain range. ;

[0098] Step 33: Calculate tumor ablation coverage : ;in, Indicates the sampling point currently being judged; Represents the set of sampling points generated from the tumor tissue region. Elements in;

[0099] Calculate the coverage components of each sub-region of the tumor tissue region: ; ;in, Indicates the first A set of sampling points for each tumor subregion;

[0100] Calculate the volume of the ablated tumor : ;

[0101] Step 34: Generate weight vector based on healthy tissue region One sampling point:

[0102] , ;in, This represents the final set of sampling points generated in the healthy tissue region. Let k represent the set of all points labeled k in the healthy tissue region;

[0103] Calculating the degree of damage to healthy tissue : ;in, Indicates the sampling point currently being judged;

[0104] Calculate the damage components of each sub-region within the healthy tissue region. : , ;

[0105] Calculate the volume of damage to healthy tissue : .

[0106] In this embodiment, it should be specifically explained that step 4, which involves dynamically updating the weights of each sub-region based on the error feedback mechanism, specifically includes:

[0107] Step 41: Define the regional density measurement method, based on the trace quantization of the regional spatial distribution characteristics of the covariance matrix: ;in, Indicates the first Regional density of each sub-region Represents the matrix trace operation. , The point cloud of tumors represents the first Covariance matrix of each subregion;

[0108] Applying range constraints avoids unstable weight updates due to extreme values:

[0109] ;in, This indicates the maximum density in the region. This represents the minimum density of the region; It is a minimum value function. It is a function for maximizing the value;

[0110] Step 42: Calculate the number of tumor tissue regions. Sub-regions at time step Regional error components , defined as the product of coverage variance and density, using exponential smoothing:

[0111] ;

[0112] in, Indicates the tumor tissue region Sub-regions at time step The regional error components; Indicates the time step. Indicates the error attenuation coefficient;

[0113] Step 43: Update the tumor tissue region weights, which includes two stages: generating the basic weight vector and smoothing the update.

[0114] ;

[0115] in, Indicates time step Time-based tumor tissue region weight vector, Indicates time step Time-bound tumor tissue region weight vector; , Represents the error vector of the tumor tissue region; Represents a subregion of tumor tissue The error vector, Indicates the learning rate. , Indicates the exploration coefficient. .

[0116] In this embodiment, it should be specifically noted that the healthy tissue damage rate in step 5 is calculated using the same dynamic sampling framework as when calculating tumor smile coverage. The generation of healthy tissue samples is similar to that of tumor regions, but independent weights are used. Specifically, it includes:

[0117] The density compensation factor for the damaged area is calculated using the following formula:

[0118] ;in, The density compensation factor represents the damaged area; The first region representing the healthy tissue area The covariance matrix of each subregion Indicates the number of subregions within a healthy tissue region; This represents the truncation function;

[0119] The area error update is expressed by the formula: ;in, , They represent the time steps respectively. , The Health Organization's first Regional error components of each sub-region;

[0120] An independent weight update mechanism is used, represented as follows:

[0121] ;in, Represents the weight vector of healthy tissue regions. Indicates the time step Time-healthy tissue region weight vector; Indicates the probability of exploration;

[0122] The weight update for healthy tissue regions follows the same pattern as that for tumor tissue regions, but uses independent learning rates and exploration probabilities. This independent design allows for parameter tuning based on the characteristics of healthy tissues, such as their more dispersed distribution and greater sensitivity to damage, thus avoiding interference from optimization strategies for tumor tissue regions on the evaluation of healthy tissue regions.

[0123] In this embodiment, it should be specifically noted that the dynamic sampling framework further includes:

[0124] For regions with high uncertainty ;in, Let R represent the covariance matrix of region R. It represents a high uncertainty threshold and stores samples from historical high-error regions to accelerate local convergence;

[0125] The dynamic sampling framework includes an optimization controller, comprising:

[0126] The learning rate is dynamically adjusted based on the magnitude of weight changes. ;in, Indicates the decay rate. This represents the initial learning rate. Indicates the convergence threshold. , These represent time steps of 1 and 2 respectively. , Weight vector at time;

[0127] and ;in, This is the convergence threshold;

[0128] The visualization method for the ablation boundary in step 5 includes:

[0129] Based on medical image visualization, a schematic diagram of sample point distribution and an ablation ellipsoid surface model are overlaid to mark the tumor boundary, and parameters such as tumor ablation coverage, healthy tissue damage, tumor ablation volume, and healthy tissue damage volume are output.

[0130] Example 2

[0131] In this embodiment, it should be specifically explained that, as Figure 2 As shown, a simplified geometric model of the liver-tumor-ablation zone was constructed. The liver was modeled as a 200mm × 120mm × 100mm cuboid with a volume of 2.4 × 10⁻⁶. 6 mm³ is equivalent to 2400 mL. The tumor was designed as a sphere with a diameter of 30 mm, centered within the liver parenchyma at a depth of (80 mm, 48 mm, 40 mm), with a volume of approximately 1.4 × 10⁻⁶. 4 The volume is approximately 18765.78 mm³, or 14.1 mL. The ablation zone is designed as an ellipsoid with the tumor center as the focal point, with a major axis (x-axis) of 20 mm, a minor axis (y-axis) of 14 mm, a height (z-axis) of 16 mm, and a volume of approximately 18765.78 mm³, or 18.8 mL. This geometric configuration is to fully verify the feasibility of the algorithm.

[0132] The algorithm implementation includes the following core parameters: total sample size K0=2000, number of iterations T0=20, tumor partition and healthy tissue partition R0=3, and error decay factor. =0.8, learning rate =0.05, Exploration Rate =0.7, weight change threshold =0.01.

[0133] Through iterative optimization, the weights of each sub-region converged, the tumor coverage estimate was Cfinal=97.80%, and the healthy tissue damage estimate was... final = 0.20%.

[0134] Tumor ablation coverage Ctrue_simple and healthy tissue coverage estimated using the classic Monte Carlo method with a sample size M0=100000 With `true_simple` as the true value, `Ctrue_simple = 97.98%`. true_simple=0.20%;

[0135] The relative errors of coverage and damage in this embodiment are RE1 and RE2, respectively, where RE1 = 0.03% and RE2 = 0.00%.

[0136] To further verify the algorithm's performance, a comparative experimental scheme was designed to estimate tumor ablation coverage using the classical Monte Carlo method and the proposed dynamically weighted Monte Carlo algorithm: 20 independent trials were conducted for each of four sample sizes: N={1000, 2000, 3000, 4000}. The true coverage baseline was calculated using the classical Monte Carlo method with M0=100000.

[0137] like Figure 3 Experimental results show that the dynamic weighted Monte Carlo algorithm exhibits significant performance advantages in a simple embodiment: This embodiment demonstrates that the dynamic weighted Monte Carlo method significantly improves evaluation accuracy through intelligent sampling strategy. Under the condition of a small sample size of 1000~4000, the median relative error is lower than that of the traditional classical MC method. This method can significantly improve computational efficiency and accuracy, and can reduce the median relative error by up to 59.09%.

[0138] Example 3

[0139] In this embodiment, medical imaging data of a liver carrying a tumor were loaded for ablation evaluation experiments. The data source was the LiTS (Liver Tumor Segmentation Challenge) dataset, in the form of CT scan images, saved in .nii format. The annotation information used grayscale values ​​of 0, 1, and 2 to represent three categories: background, liver, and tumor. The selected example had a healthy liver volume of 1315.19 mL and a tumor volume of 28.94 mL.

[0140] The algorithm parameters are set as follows: total sample size K1=2000, number of iterations T1=50, tumor partition R1=3, healthy tissue partition R2=4. =0.8、 =0.05, Exploration Rate =0.7, weight change threshold 1=0.01, controlling the scaling factor of the ablation ellipsoid. =2.4. Through iterative optimization, the weights of each sub-region converged, the tumor coverage estimate was Cfinal=95.32%, and the healthy tissue damage estimate was... final = 1.00%.

[0141] To quantify the error, Monte Carlo experiments were conducted to analyze the convergence trends of tumor ablation coverage and healthy tissue damage in unmodified Monte Carlo spatial region estimation with increasing sample size. The convergence value at sample point M was selected as the true value Ctrue for coverage and the true value for damage. If true, the relative error between tumor ablation coverage and healthy tissue damage is obtained:

[0142] ;

[0143] ;

[0144] Reference Figure 4 , is the weight convergence plot when estimating tumor ablation coverage using the clustering and dynamic weight Monte Carlo method;

[0145] Reference Figure 5 The image shows the results of estimating tumor ablation coverage using the clustering and dynamic weighted Monte Carlo method.

[0146] Reference Figure 6 The image shows the results of estimating the degree of damage in healthy tissue using the clustering and dynamic weight Monte Carlo method.

[0147] Reference Figure 7 This is a convergence map for estimating tumor ablation coverage using the traditional Monte Carlo method;

[0148] Reference Figure 8 This is a convergence plot for estimating the degree of damage in healthy tissue using the traditional Monte Carlo method.

[0149] Reference Figure 9 Box plot of relative error for tumor ablation coverage estimation at small sample points under fixed ablation ellipsoid parameters (20 trials).

[0150] In summary, the figure shows that the Monte Carlo method tends to converge after M=20000 when estimating tumor ablation coverage.

[0151] When estimating the degree of damage to healthy tissue, the estimation tends to converge after a sample size of M=40000; the estimate at M1=100000 is taken as the true value of coverage Ctrue=95.45% and the true value of damage. When true = 1.00%, we get:

[0152] REcoverage=0.14%, REdamaged=0.00%;

[0153] The clustering and dynamic weighted Monte Carlo method can control the ablation assessment error to within 0.5% with a small sample size, effectively improving the quantification accuracy of dual targets.

[0154] In this case, to further compare the performance of the proposed dynamic weighted Monte Carlo method and the traditional Monte Carlo method in estimating tumor ablation coverage, 20 experiments were conducted with sample sizes N={1000, 2000, 3000, 4000}, and a box plot of the relative error in tumor ablation coverage estimation was plotted. Figure 7 Compared with the traditional Monte Carlo method, the proposed method can reduce the error distribution range in general with a small sample size, and can reduce the median relative error of tumor ablation coverage by up to 50%.

[0155] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0156] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling, characterized in that: Includes the following steps: Step 1: Obtain voxel data of three-dimensional medical images, distinguish tumor tissue areas from healthy tissue areas using image segmentation algorithms, and calculate volume. Step 2: Divide the tumor tissue region into sub-regions based on the spatial clustering algorithm, initialize the sampling weights of each sub-region, and construct a dynamic weight model; Step 3: Obtain the weights of each sub-region based on the dynamic weight model and generate a sample set. Combine the ablation parameter model to determine the membership of the samples and obtain the tumor ablation coverage and the degree of damage to healthy tissue. Step 4: Dynamically update the weights of each sub-region based on the error feedback mechanism, and optimize the sample distribution by balancing historical errors and current errors through adaptive learning rate and decay factor; Step 5: Iterate through steps 2-4 until convergence, output the joint evaluation results of tumor ablation coverage and healthy tissue damage rate, and visualize the ablation boundary.

2. The tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling according to claim 1, characterized in that: The volume calculation in step 1 specifically includes: Step 11: Obtain the tumor tissue point set and health organization point collection ; Step 12: Define the anisotropic scaling factor Converting voxel coordinates to physical space coordinates is represented as: , ; ;in, This is represented as a set of tumor tissue points after being transformed into physical space. Represented as a set of healthy tissue points after being transformed into physical space; They represent axis, shaft and The scaling factor of the axis. This represents element-wise multiplication. Step 13: Calculate the tumor volume and the volume of healthy tissue.

3. The tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling according to claim 2, characterized in that: The tumor tissue point set is represented as follows: The set of healthy tissue points is represented as ;in, and All are three-dimensional spatial coordinates. , ; The number of voxel points in the tumor tissue region. The number of voxel points in the healthy tissue region; the formula for calculating the tumor volume is expressed as: ;in, The volume of the tumor is represented by the formula for calculating the volume of the healthy tissue: ;in, Indicates the volume of healthy tissue.

4. The tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling according to claim 3, characterized in that: Step 2, which involves constructing the dynamic weight model, specifically includes: Step 21: Perform K-means clustering on the tumor tissue point set and the healthy tissue point set to generate a region label mapping function; Step 22: Initialize the tumor region weight vector and healthy region weight vector The tumor region weight vector is represented as: ,in, This represents the weight vector of each subregion of the tumor tissue region. ;satisfy The weight vector of the healthy region is represented as: in, This represents the weight vector of each subregion within the healthy tissue region. ;satisfy .

5. The tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling according to claim 4, characterized in that: The region label mapping function for the tumor tissue point set is expressed as follows: ;in, Region labels representing tumor tissue point sets, The number of preset subregions representing the tumor tissue region; ;in, The region label represents the set of healthy tissue points. This indicates the number of preset sub-regions within the healthy tissue region.

6. The tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling according to claim 5, characterized in that: Step 3, which involves generating a sample set and determining the membership of samples using the ablation parameter model, specifically includes: Step 31: Generating sub-region weights based on the current tumor tissue region One sampling point: ; ;in, This represents the final set of sampling points generated in the tumor tissue region. ,in, Indicates from Randomly selected One point, Let k represent the set of all points labeled k in the tumor tissue region. Indicates from the first The number of points actually extracted in each sub-region; This represents the floor function; ; Step 32, for sample points To determine whether it is located within the ablation ellipsoid, the formula is as follows: ； in, Indicates an indicator function, These are the coordinates of the ablation ellipsoid center, determined based on the tumor location. , , They represent the ellipsoid in axis, shaft and The semi-axis length in the axial direction is determined by the location and size of the tumor and can be parameterized. The coordinates of the center of the ablation ellipsoid are expressed as follows: ; ; ;in, Indicates the first sampling points Axis coordinates; Indicates the first sampling points Axis coordinates; Indicates the first sampling points Axis coordinates; The lengths of each half-shaft are expressed as follows: ; ; ;in, , , Tumor tissue point sets axis, axis, Standard deviation of the axis coordinates This represents the scaling factor. To ensure the ablation effect, the scaling factor takes values ​​within a certain range. ; Step 33: Calculate tumor ablation coverage : ;in, Indicates the sampling point currently being judged; Represents the set of sampling points generated from the tumor tissue region. Elements in; Calculate the coverage components of each sub-region of the tumor tissue region: ; ;in, Indicates the first A set of sampling points for each tumor subregion; Calculate the volume of the ablated tumor : ; Step 34: Generate weight vector based on healthy tissue region One sampling point: , ;in, This represents the final set of sampling points generated in the healthy tissue region. This indicates that all labels in the healthy organization area are... A set of points; calculating the degree of damage to healthy tissue. : ;in, Indicates the sampling point currently being judged; Calculate the damage components of each sub-region within the healthy tissue region. : , ; Calculate the volume of damage to healthy tissue : .

7. The tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling according to claim 6, characterized in that: Step 4, which involves dynamically updating the weights of each sub-region based on the error feedback mechanism, specifically includes: Step 41: Define the regional density measurement method, based on the trace quantization of the regional spatial distribution characteristics of the covariance matrix: ;in, Indicates the first Regional density of each sub-region Represents the matrix trace operation. , Indicates the first point cloud of the tumor Covariance matrix of each subregion; Step 42: Calculate the number of tumor tissue regions. Sub-regions at time step Regional error components , defined as the product of coverage variance and density, using exponential smoothing: ； in, Indicates the first tumor tissue region Sub-regions at time step The regional error components; Indicates the time step. Indicates the error attenuation coefficient; Step 43: Update the tumor tissue region weights, which includes two stages: generating the basic weight vector and smoothing the update. ； in, Indicates the time step Time-based tumor tissue region weight vector, Indicates the time step Time-bound tumor tissue region weight vector; , Represents the error vector of the tumor tissue region; This represents the error vector of the tumor tissue subregion R1. Indicates the learning rate. , Indicates the exploration coefficient. .

8. The tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling according to claim 7, characterized in that: Step 5 further includes: The density compensation factor for the damaged area is calculated using the following formula: ;in, The density compensation factor representing the damaged area; The first region representing the healthy tissue area The covariance matrix of each subregion Indicates the number of subregions within a healthy tissue region; This represents the truncation function; The area error update is expressed by the formula: ;in, , This indicates that at the time step, respectively , The regional error component of the k-th sub-region of a healthy tissue; An independent weight update mechanism is used, represented as follows: ;in, Indicates the time step Time-healthy tissue region weight vector, Indicates the time step Time-healthy tissue region weight vector; This represents the probability of exploration.

9. The tumor ablation assessment method based on clustering and dynamic weighted Monte Carlo sampling according to claim 8, characterized in that: The dynamic sampling framework further includes: For regions with high uncertainty ;in, Let R represent the covariance matrix of region R. Indicates a high uncertainty threshold; The dynamic sampling framework includes an optimization controller, comprising: The learning rate is dynamically adjusted based on the magnitude of weight changes. ;in, Indicates the decay rate. This represents the initial learning rate. Indicates the convergence threshold. , These represent time steps of 1 and 2 respectively. , Weight vector at time; and ;in, This is the convergence threshold.

Images