Radiotherapy target region accurate sketching method based on super-fraction geometry

Abstract

The invention discloses a radiotherapy target region accurate sketching method based on super-fraction geometry. The method comprises the following steps: S1, generating a preprocessed radiotherapy target region medical image data set; s2, outputting multi-scale super-division geometric feature image data; s3, calculating an edge intensity graph based on the multi-scale super-division geometric feature image data, and generating a prey heat map; s4, constructing a target region segmentation model; s5, extracting an image quality perception factor according to the multi-scale super-division geometric feature image data, taking the prey heat map as a dynamic guide function, embedding the prey heat map into a secretary bird optimization algorithm, updating parameters of the target region segmentation model through combination of global search and local fine tuning in an iteration process, and generating an optimal segmentation parameter set; and S6, using the optimal segmentation parameter set to act on the multi-scale super-division geometric feature image data, executing automatic target region segmentation, and obtaining a radiotherapy target region sketching result. According to the method, the segmentation precision, stability and clinical application value can be remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of medical technology, and in particular to a method for precise delineation of radiotherapy target areas based on super-resolution geometry. Background Technology

[0002] With the continuous development of image segmentation technology and medical image processing methods, precision radiotherapy has placed higher demands on the accuracy and efficiency of target identification and delineation. Accurately delineating tumor tissue boundaries from CT and MRI 3D medical images is a crucial step affecting treatment planning and dose distribution control in tumor radiotherapy. Currently, the widely used radiotherapy target delineation methods in clinical practice are still mainly manual or semi-automatic segmentation, relying on the doctor's subjective judgment of the lesion boundary and simple image processing tools to complete target annotation. While this method is flexible, it has the following significant drawbacks: First, manual operation is time-consuming and labor-intensive, and inefficient when dealing with large amounts of image data; second, due to the limited resolution of imaging equipment, especially in the presence of noise, artifacts, or blurred boundaries, the doctor's subjective judgment is prone to errors, causing target delineation deviations or omissions, thus affecting the subsequent radiotherapy effect.

[0003] In recent years, deep learning-based or image enhancement algorithms have been introduced in automated image segmentation research. However, image enhancement still has certain limitations. On the one hand, due to the spatial resolution and imaging quality of the original image, the model struggles to accurately reproduce the detailed features of tumor edges, especially in tumor regions with complex morphology or strong structural heterogeneity, often exhibiting problems such as blurred boundaries, regional fragmentation, or adhesion. On the other hand, most existing parameter optimization methods use fixed parameters or rely on manual adjustment based on expert experience, lacking adaptability and global search capabilities, making it difficult to maintain the stability and high accuracy of segmentation results under different case and image quality conditions.

[0004] Furthermore, most current optimization strategies have not effectively incorporated image structure feature-driven mechanisms, and optimization algorithms lack the ability to dynamically perceive the target region, often getting trapped in local optima and failing to fully capture the microstructural changes at the tumor boundary.

[0005] In summary, there is an urgent need to propose a new method that integrates multi-scale modeling and intelligent optimization mechanisms to improve the ability to delineate the boundaries of tumor target areas and the scientific nature of radiotherapy pathway planning. Summary of the Invention

[0006] One objective of this invention is to propose a method for precise delineation of radiotherapy target areas based on super-resolution geometry. This invention can significantly improve the accuracy, stability, and clinical application value of segmentation.

[0007] A method for precise delineation of radiotherapy target areas based on super-resolution geometry according to an embodiment of the present invention includes the following steps:

[0008] S1. Obtain radiotherapy target area medical image data of the patient to be treated to form a radiotherapy target area medical image dataset. Perform image preprocessing on the radiotherapy target area medical image dataset to generate a preprocessed radiotherapy target area medical image dataset.

[0009] S2. Input the preprocessed radiotherapy target area medical image dataset into the multi-scale super-resolution geometric modeling module, perform super-resolution reconstruction and geometric feature enhancement for multiple scales, and output multi-scale super-resolution geometric feature image data;

[0010] S3. Calculate edge intensity map based on multi-scale super-resolution geometric feature image data to generate prey heat map;

[0011] S4. Construct a target region segmentation model, and construct a set of segmentation threshold parameters, a set of edge localization parameters, a set of morphological structure adjustment parameters, and a set of multi-scale fusion factors;

[0012] S5. Based on the multi-scale super-resolution geometric feature image data, extract the image quality perception factor, and use the image quality perception factor to adaptively initialize the search radius, exploration step size and perturbation scale of the secretary bird optimization algorithm to obtain the initial secretary bird population parameters. Embed the prey heat map as a dynamic guiding function into the secretary bird optimization algorithm. During the iteration process, update the parameters of the target area segmentation model through global search and local fine-tuning to generate the optimal segmentation parameter set.

[0013] S6. Apply the optimal segmentation parameter set to the multi-scale super-resolution geometric feature image data to perform automatic target segmentation and obtain the radiotherapy target delineation results.

[0014] Optionally, S1 includes the following steps:

[0015] S11. Acquire medical image data of the radiotherapy target area of ​​the patient to be treated, set a uniform acquisition time window, and form a medical image dataset of the radiotherapy target area. :

[0016] ;

[0017] in, Indicates the first Zhang 3D slice image, This represents the spatial coordinates of a 3D slice image in 3D voxel space. This represents the pixel grayscale value corresponding to a 3D slice image. This represents the total number of 3D slice images;

[0018] S12. Perform image preprocessing on the radiotherapy target area medical image dataset to obtain the preprocessed radiotherapy target area medical image dataset. This includes the following preprocessing operations:

[0019] The noise suppression operation employs an isotropic Gaussian filtering method to eliminate artifacts and high-frequency interference during the imaging process, and outputs a noise-suppressed image set.

[0020] Gray-level normalization processing converts the gray-level values ​​of each 3D slice image. Mapping to interval ;

[0021] The structure enhancement process calculates image edge changes based on the local structure tensor and gradient intensity, enhances the edge response of the image in the noise-suppressed image set, and outputs a structure-enhanced image set.

[0022] Optionally, S2 includes the following steps:

[0023] S21. Preprocess the medical image dataset of the radiotherapy target area. Input the multi-scale super-resolution geometric modeling module, based on each 3D slice image Structural heterogeneity level of intermediate radiotherapy target area With the rate of change of edge curvature Constructing asymmetric scale sets Each reconstruction scale factor in the asymmetric scale set Calculated using an adaptive scaling function driven by the radiotherapy target area:

[0024] ;

[0025] in, Indicates the first In the three-dimensional slice image, the first Reconstruction scale factor of a local region of the radiotherapy target area Indicates the first The rate of change in the edge curvature of each region reflects the degree of local geometric mutation at the tumor boundary. This indicates the level of structural heterogeneity between the tumor region and the background tissue in the entire image. This represents the maximum scale control factor. The variation adjustment coefficient, This is the lower limit protection constant;

[0026] S22. Local areas of each radiotherapy target zone According to the corresponding reconstruction scale factor High-fidelity local image patches are constructed by performing curvature-constrained super-resolution reconstruction using a reconstruction function. Reconstruction function Defined as:

[0027] ;

[0028] in, Represents a 3D slice image The Middle A local area of ​​the radiotherapy target zone to be reconstructed. This represents the initial super-resolution image generated using unconstrained interpolation, including bicubic interpolation. To optimize the target image patch, For the first The target curvature distribution function of the tumor boundary in each region, Indicates optimization of target image patch The second derivative, For the local area of ​​the radiotherapy target zone The domain of the integral, is the regularization coefficient for the edge curvature constraint term;

[0029] S23. All high-fidelity local image patches after curvature-constrained super-resolution reconstruction. Geometric alignment and edge continuity stitching are performed to construct a multi-scale geometrically enhanced representation of the complete image. :

[0030]

[0031] During geometric alignment and edge continuity stitching, the continuity of the first derivative and curvature information of adjacent high-fidelity local image blocks at the boundary is maintained, ensuring the structural consistency and integrity of the radiotherapy target area boundary in space;

[0032] S24. Based on multi-scale geometric enhancement representation Constructing a characteristic response map of the enhanced radiotherapy target area :

[0033]

[0034] in, Represents the edge gradient map. Indicators representing fusion mutations between tumor and non-tumor regions in terms of density, texture, and edge direction. It serves as a regulatory factor for the fusion of edge and structural mutation information;

[0035] S25. Output enhanced radiotherapy target area feature response maps for all 3D slice images, and output multi-scale super-resolution geometric feature image data. .

[0036] Optionally, S3 includes the following steps:

[0037] S31. Multi-scale super-resolution geometric feature image data The edge intensity calculation module is input into the image, and it calculates the feature response map of each enhanced radiotherapy target area. Calculate the edge intensity map of the characteristic response map of the enhanced radiotherapy target area. Edge intensity map This represents the gray-level gradient magnitude of each voxel in the three-dimensional spatial direction in the enhanced radiotherapy target region feature response map. This is achieved by simultaneously calculating the enhanced radiotherapy target region feature response map. In space , , The partial derivatives in three directions are obtained by summing the squares of the partial derivatives in three directions and taking the square root. This is used to describe the degree of intensity variation of the radiation target boundary in different directions.

[0038] S32. Edge Intensity Map Normalization is performed to obtain the normalized edge response map. The normalization process involves subtracting the minimum intensity value from the edge intensity map for each voxel in the edge intensity map, and then dividing by the difference between the maximum and minimum values. This maps all edge intensity values ​​to the interval between 0 and 1, resulting in the normalized edge response map. Preserve the relative distribution characteristics of the edges;

[0039] S33. Based on Normalized Edge Response Map Constructing a prey heatmap Prey heatmap This is achieved by normalizing the edge response map. With mask image The mask image is obtained by multiplying the pixel values ​​at corresponding positions point by point. It is a spatial mask that takes a value of 1 only in the candidate region of the radiotherapy target area and a value of 0 in other regions, used to eliminate interference from non-medically relevant areas, prey heatmap Preserve edge response information within the candidate radiotherapy target area;

[0040] S34. Output prey heatmaps of all enhanced radiotherapy target region characteristic response maps. Construct a prey heatmap dataset .

[0041] Optionally, S4 includes constructing a target region segmentation model. The target segmentation model uses the prey heatmap dataset. Based on the input, it includes four core parameter sets: segmentation threshold parameter set, edge localization parameter set, morphological structure adjustment parameter set, and multi-scale fusion factor set, which are used to realize response region extraction, edge fine adjustment, structural continuity optimization, and scale information integration of the radiotherapy target area, respectively.

[0042] Optionally, the set of segmentation threshold parameters is as follows: ,in Indicates the first The initial segmentation threshold used for binarization processing in the heatmap of the prey is used to compare the voxel response intensity in the heatmap with the threshold. Voxels that are greater than or equal to the initial segmentation threshold are marked as candidate regions for the initial radiotherapy target area, while voxels that are less than the initial segmentation threshold are excluded, forming the initial radiotherapy target area mask map.

[0043] The edge positioning parameter set is ,in This represents the edge smoothing adjustment coefficient, used to control the kernel function scale of the prey heatmap edge response map during edge smoothing processing. This represents the edge enhancement weighting factor, used to adjust the degree to which edge intensity enhances the boundary morphology of the initial radiotherapy target area mask map. It adjusts the normalized edge response map using the edge smoothing adjustment coefficient. After smoothing under control, the image is fused with the initial radiotherapy target mask, and edge enhancement weighting factors are incorporated. Adjustments are made to create an edge adjustment mask image, which is used to enhance the target area boundary contour;

[0044] The set of morphological structure adjustment parameters is as follows ,in This represents the radius of the structuring element used for the closing operation, controlling the degree to which small holes in the initial radiotherapy target mask are filled. The radius of the structuring element used for the opening operation is used to eliminate misidentified regions caused by spurious responses. The closing and opening operations are performed sequentially on the edge-adjusted mask to form a structure-adjusted radiotherapy target area mask, making its boundaries continuous, its structure compact, and its shape reasonable.

[0045] The set of multi-scale fusion factors is ,in Indicates the first Zhang 3D slice image in the first Weighting factors at various scales are used to control the participation ratio of morphological modulation results at different scales. A structure-modified radiotherapy target area mask is generated for each 3D slice image at different scales. The radiotherapy target area masks are weighted and summed separately, and then fused according to the weighting factors. Multi-scale information is integrated to form the final radiotherapy target mask map.

[0046] Optionally, S5 includes the following steps:

[0047] S51. Based on multi-scale super-resolution geometric feature image data Extracting the set of image quality perceptual factors Each image quality perceptual factor Based on clarity index Edge density index and structural complexity index Perform weighted fusion;

[0048] Sharpness index This indicates the degree of dispersion of the overall grayscale distribution in the enhanced radiotherapy target area characteristic response map:

[0049] ;

[0050] in, To enhance the characteristic response map of the radiotherapy target area Average gray value, Represents the image voxel space;

[0051] Edge density index Indicates the density of the image boundary response:

[0052] ;

[0053] in, For indicator functions, This is the edge response threshold;

[0054] Structural complexity index Based on local entropy calculation, the texture complexity of heterogeneous structures in the enhanced radiotherapy target region feature response map is represented, and the enhanced radiotherapy target region feature response map is divided into... Sub-block :

[0055] ;

[0056] in, For the first The grayscale value within each sub-block is The probability density distribution;

[0057] S52. Image quality perceptual factor Used for initializing population parameters and setting the search radius in the Secretary Bird optimization algorithm. Exploration Step Size Disturbance scale :

[0058] ;

[0059] in, , , These are the initial default parameters. , , This is the adjustment coefficient;

[0060] S53. Construct the parameter set for the target segmentation model. The optimization objective function is:

[0061] ;

[0062] in, For the first Predictive mask map of Zhang Zengqiang's radiotherapy target area characteristic response map. Its artificial reference image, Dice is the Dice coefficient, and HD is the Hausdorff distance;

[0063] S54. Prey heatmap dataset As a dynamic guiding function, it is embedded in the Secretary Bird optimization algorithm to construct a guiding weight matrix for the segmentation parameters. In the guided target segmentation model, the first The evolution direction and strength of each parameter, where each guiding weight From the prey heatmap in the parameter control area The weighted integral within the range is calculated to obtain:

[0064] ;

[0065] in, Indicates the first Segmentation parameters in Zhang Zengqiang's radiotherapy target area characteristic response map The average intensity of prey response in the controlled area Indicates parameters The corresponding spatial range of action;

[0066] Secretary bird individual in the first In the nth iteration, the th The update definition for each parameter is:

[0067] ;

[0068] in, For the first The secretary bird individual in the first The th iteration One segmentation parameter, For the first The secretary bird individual in the first The th iteration One segmentation parameter, The segmentation parameters are those for the current optimal individual. The perturbation scale. This is standard normally distributed noise;

[0069] S55. Iteratively update the parameter set of the target segmentation model using the Secretary Bird optimization algorithm. To minimize the optimization objective function Using this as the criterion, the optimal splitting parameter set is finally output. .

[0070] Optionally, S6 includes the following steps:

[0071] S61. The optimal segmentation parameter set obtained through the Secretary Bird optimization algorithm is applied to the multi-scale super-resolution geometric feature image dataset. Based on the optimal segmentation parameter set, the automatic segmentation process for the radiotherapy target area is executed sequentially. The segmentation process includes:

[0072] Using the optimal set of segmentation threshold parameters Enhanced response map of the target area for radiotherapy Thresholding is performed to initially extract the response region and form an initial mask image;

[0073] Based on the optimal edge localization parameter set Joint prey heat map The response region is used to construct an edge guiding function, and the normalized edge response map is fused to perform boundary optimization, forming an edge adjustment mask map;

[0074] Based on the optimal morphological structure adjustment parameter set Perform closing and opening operations on the edge adjustment mask image to enhance structural connectivity and noise reduction capabilities;

[0075] Based on the optimal multi-scale fusion factor set The structure-adjusted mask images at each scale are weighted and fused to obtain the final radiotherapy target area mask image. ;

[0076] S63. Output the final set of radiotherapy target mask maps. Finally, based on the structural features and morphological distribution of the radiotherapy target area mask, the radiotherapy target area delineation results in each 3D slice image are classified. The classification rules are as follows:

[0077] Clear target area: edge integrity coefficient ≥ 0.9, region compactness coefficient ≥ 0.85, average prey heatmap response ≥ 0.8;

[0078] Blurred target region: 0.7 ≤ edge integrity coefficient < 0.9, region compactness coefficient ≥ 0.7, average prey heatmap response ∈ [0.5, 0.8);

[0079] Irregular target area: edge integrity coefficient < 0.7 or average prey heatmap response < 0.5;

[0080] S64. Obtain the final radiotherapy target delineation result dataset. And their corresponding category tags.

[0081] The beneficial effects of this invention are:

[0082] (1) In view of the characteristics of the curvature change and structural heterogeneity of the tumor region boundary, the present invention constructs an adaptive scale mapping function driven by the radiotherapy target area, and generates a local asymmetric reconstruction scale factor based on the function. Combined with the curvature-constrained local reconstruction function, the image block is reconstructed with high fidelity to form an enhanced radiotherapy target area feature response map. This effectively breaks through the limitation of traditional multi-scale image reconstruction methods that only rely on fixed scale factors, enabling fine-scale reconstruction of boundary mutation regions and improving the spatial resolution and structural integrity of the tumor edge.

[0083] (2) Based on multi-scale geometric feature images, this invention extracts image quality perception factors and combines three image evaluation dimensions: sharpness, edge density, and structural complexity. It defines quantification formulas to adaptively set the search radius, exploration step size, and perturbation scale of the secretary bird algorithm, thereby realizing the adaptive search capability of the optimization algorithm under different image quality conditions. It introduces prey heatmap as a dynamic guiding function to embed the parameter update path of individual secretary birds and designs parameter guiding weights based on response intensity to drive each segmentation parameter to evolve towards the strong response region of the tumor boundary. This significantly improves the adaptability of the segmentation parameters to the spatial structure of the target area and can stably obtain high-precision segmentation results under different image contrast conditions. Attached Figure Description

[0084] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0085] Figure 1 This is a flowchart of a method for accurately delineating radiotherapy target areas based on super-resolution geometry proposed in this invention. Detailed Implementation

[0086] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0087] refer to Figure 1 A method for precise delineation of radiotherapy target areas based on super-resolution geometry includes the following steps:

[0088] S1. Obtain radiotherapy target area medical image data of the patient to be treated to form a radiotherapy target area medical image dataset. Perform image preprocessing on the radiotherapy target area medical image dataset to generate a preprocessed radiotherapy target area medical image dataset.

[0089] S2. Input the preprocessed radiotherapy target area medical image dataset into the multi-scale super-resolution geometric modeling module, perform super-resolution reconstruction and geometric feature enhancement for multiple scales, and output multi-scale super-resolution geometric feature image data;

[0090] S3. Calculate edge intensity map based on multi-scale super-resolution geometric feature image data to generate prey heat map;

[0091] S4. Construct a target region segmentation model, and construct a set of segmentation threshold parameters, a set of edge localization parameters, a set of morphological structure adjustment parameters, and a set of multi-scale fusion factors;

[0092] S5. Based on the multi-scale super-resolution geometric feature image data, extract the image quality perception factor, and use the image quality perception factor to adaptively initialize the search radius, exploration step size and perturbation scale of the secretary bird optimization algorithm to obtain the initial secretary bird population parameters. Embed the prey heat map as a dynamic guiding function into the secretary bird optimization algorithm. During the iteration process, update the parameters of the target area segmentation model through global search and local fine-tuning to generate the optimal segmentation parameter set.

[0093] S6. Apply the optimal segmentation parameter set to the multi-scale super-resolution geometric feature image data to perform automatic target segmentation and obtain the radiotherapy target delineation results.

[0094] In this embodiment, S1 includes the following steps:

[0095] S11. Acquire medical image data of the radiotherapy target area of ​​the patient to be treated, set a uniform acquisition time window, and form a medical image dataset of the radiotherapy target area. :

[0096] ;

[0097] in, Indicates the first Zhang 3D slice image, This represents the spatial coordinates of a 3D slice image in 3D voxel space. This represents the pixel grayscale value corresponding to a 3D slice image. This represents the total number of 3D slice images;

[0098] S12. Perform image preprocessing on the radiotherapy target area medical image dataset to obtain the preprocessed radiotherapy target area medical image dataset. This includes the following preprocessing operations:

[0099] The noise suppression operation employs an isotropic Gaussian filtering method to eliminate artifacts and high-frequency interference during the imaging process, and outputs a noise-suppressed image set.

[0100] Gray-level normalization processing converts the gray-level values ​​of each 3D slice image. Mapping to interval ;

[0101] The structure enhancement process calculates image edge changes based on the local structure tensor and gradient intensity, enhances the edge response of the image in the noise-suppressed image set, and outputs a structure-enhanced image set.

[0102] In this embodiment, S2 includes the following steps:

[0103] S21. Preprocess the medical image dataset of the radiotherapy target area. Input the multi-scale super-resolution geometric modeling module, based on each 3D slice image Structural heterogeneity level of intermediate radiotherapy target area With the rate of change of edge curvature Constructing asymmetric scale sets Each reconstruction scale factor in the asymmetric scale set Calculated using an adaptive scaling function driven by the radiotherapy target area:

[0104] ;

[0105] in, Indicates the first In the three-dimensional slice image, the first Reconstruction scale factor of a local region of the radiotherapy target area Indicates the first The rate of change in the edge curvature of each region reflects the degree of local geometric mutation at the tumor boundary. This indicates the level of structural heterogeneity between the tumor region and the background tissue in the entire image. This represents the maximum scale control factor. The variation adjustment coefficient, This is the lower limit protection constant;

[0106] S22. Local areas of each radiotherapy target zone According to the corresponding reconstruction scale factor High-fidelity local image patches are constructed by performing curvature-constrained super-resolution reconstruction using a reconstruction function. Reconstruction function Defined as:

[0107] ;

[0108] in, Represents a 3D slice image The Middle A local area of ​​the radiotherapy target zone to be reconstructed. This represents the initial super-resolution image generated using unconstrained interpolation, including bicubic interpolation. To optimize the target image patch, For the first The target curvature distribution function of the tumor boundary in each region, Indicates optimization of target image patch The second derivative, For the local area of ​​the radiotherapy target zone The domain of the integral, is the regularization coefficient for the edge curvature constraint term;

[0109] S23. All high-fidelity local image patches after curvature-constrained super-resolution reconstruction. Geometric alignment and edge continuity stitching are performed to construct a multi-scale geometrically enhanced representation of the complete image. :

[0110]

[0111] During geometric alignment and edge continuity stitching, the continuity of the first derivative and curvature information of adjacent high-fidelity local image blocks at the boundary is maintained, ensuring the structural consistency and integrity of the radiotherapy target area boundary in space;

[0112] S24. Based on multi-scale geometric enhancement representation Constructing a characteristic response map of the enhanced radiotherapy target area :

[0113]

[0114] in, Represents the edge gradient map. Indicators representing fusion mutations between tumor and non-tumor regions in terms of density, texture, and edge direction. It serves as a regulatory factor for the fusion of edge and structural mutation information;

[0115] S25. Output enhanced radiotherapy target area feature response maps for all 3D slice images, and output multi-scale super-resolution geometric feature image data. .

[0116] In this embodiment, S3 includes the following steps:

[0117] S31. Multi-scale super-resolution geometric feature image data The edge intensity calculation module is input into the image, and it calculates the feature response map of each enhanced radiotherapy target area. Calculate the edge intensity map of the characteristic response map of the enhanced radiotherapy target area. Edge intensity map This represents the gray-level gradient magnitude of each voxel in the three-dimensional spatial direction in the enhanced radiotherapy target region feature response map. This is achieved by simultaneously calculating the enhanced radiotherapy target region feature response map. In space , , The partial derivatives in three directions are obtained by summing the squares of the partial derivatives in three directions and taking the square root. This is used to describe the degree of intensity variation of the radiation target boundary in different directions.

[0118] S32. Edge Intensity Map Normalization is performed to obtain the normalized edge response map. The normalization process involves subtracting the minimum intensity value from the edge intensity map for each voxel in the edge intensity map, and then dividing by the difference between the maximum and minimum values. This maps all edge intensity values ​​to the interval between 0 and 1, resulting in the normalized edge response map. Preserve the relative distribution characteristics of the edges;

[0119] S33. Based on Normalized Edge Response Map Constructing a prey heatmap Prey heatmap This is achieved by normalizing the edge response map. With mask image The mask image is obtained by multiplying the pixel values ​​at corresponding positions point by point. It is a spatial mask that takes a value of 1 only in the candidate region of the radiotherapy target area and a value of 0 in other regions, used to eliminate interference from non-medically relevant areas, prey heatmap Preserve edge response information within the candidate radiotherapy target area;

[0120] S34. Output prey heatmaps of all enhanced radiotherapy target region characteristic response maps. Construct a prey heatmap dataset .

[0121] In this embodiment, S4 includes constructing a target segmentation model. The target segmentation model uses the prey heatmap dataset. Based on the input, it includes four core parameter sets: segmentation threshold parameter set, edge localization parameter set, morphological structure adjustment parameter set, and multi-scale fusion factor set, which are used to realize response region extraction, edge fine adjustment, structural continuity optimization, and scale information integration of the radiotherapy target area, respectively.

[0122] In this embodiment, the set of segmentation threshold parameters is as follows: ,in Indicates the first The initial segmentation threshold used for binarization processing in the heatmap of the prey is used to compare the voxel response intensity in the heatmap with the threshold. Voxels that are greater than or equal to the initial segmentation threshold are marked as candidate regions for the initial radiotherapy target area, while voxels that are less than the initial segmentation threshold are excluded, forming the initial radiotherapy target area mask map.

[0123] The edge positioning parameter set is ,in This represents the edge smoothing adjustment coefficient, used to control the kernel function scale of the prey heatmap edge response map during edge smoothing processing. This represents the edge enhancement weighting factor, used to adjust the degree to which edge intensity enhances the boundary morphology of the initial radiotherapy target area mask map. It adjusts the normalized edge response map using the edge smoothing adjustment coefficient. After smoothing under control, the image is fused with the initial radiotherapy target mask, and edge enhancement weighting factors are incorporated. Adjustments are made to create an edge adjustment mask image, which is used to enhance the target area boundary contour;

[0124] The set of morphological structure adjustment parameters is as follows ,in This represents the radius of the structuring element used for the closing operation, controlling the degree to which small holes in the initial radiotherapy target mask are filled. The radius of the structuring element used for the opening operation is used to eliminate misidentified regions caused by spurious responses. The closing and opening operations are performed sequentially on the edge-adjusted mask to form a structure-adjusted radiotherapy target area mask, making its boundaries continuous, its structure compact, and its shape reasonable.

[0125] The set of multi-scale fusion factors is ,in Indicates the first Zhang 3D slice image in the first Weighting factors at various scales are used to control the participation ratio of morphological modulation results at different scales. A structure-modified radiotherapy target area mask is generated for each 3D slice image at different scales. The radiotherapy target area masks are weighted and summed separately, and then fused according to the weighting factors. Multi-scale information is integrated to form the final radiotherapy target mask map.

[0126] In this embodiment, S5 includes the following steps:

[0127] S51. Based on multi-scale super-resolution geometric feature image data Extracting the set of image quality perceptual factors Each image quality perceptual factor Based on clarity index Edge density index and structural complexity index Perform weighted fusion;

[0128] Sharpness index This indicates the degree of dispersion of the overall grayscale distribution in the enhanced radiotherapy target area characteristic response map:

[0129] ;

[0130] in, To enhance the characteristic response map of the radiotherapy target area Average gray value, Represents the image voxel space;

[0131] Edge density index Indicates the density of the image boundary response:

[0132] ;

[0133] in, For indicator functions, This is the edge response threshold;

[0134] Structural complexity index Based on local entropy calculation, the texture complexity of heterogeneous structures in the enhanced radiotherapy target region feature response map is represented, and the enhanced radiotherapy target region feature response map is divided into... Sub-block :

[0135] ;

[0136] in, For the first The grayscale value within each sub-block is The probability density distribution;

[0137] S52. Image quality perceptual factor Used for initializing population parameters and setting the search radius in the Secretary Bird optimization algorithm. Exploration Step Size Disturbance scale :

[0138] ;

[0139] in, , , These are the initial default parameters. , , This is the adjustment coefficient;

[0140] S53. Construct the parameter set for the target segmentation model. The optimization objective function is:

[0141] ;

[0142] in, For the first Predictive mask map of Zhang Zengqiang's radiotherapy target area characteristic response map. Its artificial reference image, Dice is the Dice coefficient, and HD is the Hausdorff distance;

[0143] S54. Prey heatmap dataset As a dynamic guiding function, it is embedded in the Secretary Bird optimization algorithm to construct a guiding weight matrix for the segmentation parameters. In the guided target segmentation model, the first The evolution direction and strength of each parameter, where each guiding weight From the prey heatmap in the parameter control area The weighted integral within the range is calculated to obtain:

[0144] ;

[0145] in, Indicates the first Segmentation parameters in Zhang Zengqiang's radiotherapy target area characteristic response map The average intensity of prey response in the controlled area Indicates parameters The corresponding spatial range of action;

[0146] Secretary bird individual in the first In the nth iteration, the th The update definition for each parameter is:

[0147] ;

[0148] in, For the first The secretary bird individual in the first The th iteration One segmentation parameter, For the first The secretary bird individual in the first The th iteration One segmentation parameter, The segmentation parameters are those for the current optimal individual. The perturbation scale. This is standard normally distributed noise;

[0149] S55. Iteratively update the parameter set of the target segmentation model using the Secretary Bird optimization algorithm. To minimize the optimization objective function Using this as the criterion, the optimal splitting parameter set is finally output. .

[0150] In this embodiment, S6 includes the following steps:

[0151] S61. The optimal segmentation parameter set obtained through the Secretary Bird optimization algorithm is applied to the multi-scale super-resolution geometric feature image dataset. Based on the optimal segmentation parameter set, the automatic segmentation process for the radiotherapy target area is executed sequentially. The segmentation process includes:

[0152] Using the optimal set of segmentation threshold parameters Enhanced response map of the target area for radiotherapy Thresholding is performed to initially extract the response region and form an initial mask image;

[0153] Based on the optimal edge localization parameter set Joint prey heat map The response region is used to construct an edge guiding function, and the normalized edge response map is fused to perform boundary optimization, forming an edge adjustment mask map;

[0154] Based on the optimal morphological structure adjustment parameter set Perform closing and opening operations on the edge adjustment mask image to enhance structural connectivity and noise reduction capabilities;

[0155] Based on the optimal multi-scale fusion factor set The structure-adjusted mask images at each scale are weighted and fused to obtain the final radiotherapy target area mask image. ;

[0156] S63. Output the final set of radiotherapy target mask maps. Finally, based on the structural features and morphological distribution of the radiotherapy target area mask, the radiotherapy target area delineation results in each 3D slice image are classified. The classification rules are as follows:

[0157] Clear target area: edge integrity coefficient ≥ 0.9, region compactness coefficient ≥ 0.85, average prey heatmap response ≥ 0.8;

[0158] Blurred target region: 0.7 ≤ edge integrity coefficient < 0.9, region compactness coefficient ≥ 0.7, average prey heatmap response ∈ [0.5, 0.8);

[0159] Irregular target area: edge integrity coefficient < 0.7 or average prey heatmap response < 0.5;

[0160] S64. Obtain the final radiotherapy target delineation result dataset. And their corresponding category tags.

[0161] Final radiotherapy target mask map The 3D boundary is extracted into an edge set. Calculate the total length of edges in this edge set that can be closed and connected to form a closed boundary, denoted as . Simultaneously, the total edge length of all boundary voxels is calculated and denoted as . .

[0162] The edge integrity coefficient is obtained by comparing the two values ​​mentioned above. It represents the proportion of the closed boundary length to the total boundary length, and the value ranges from 0 to 1. The larger the value, the more complete the edge structure and the more continuous the boundary.

[0163] Based on the final radiotherapy target mask map The total volume of the mask region is calculated by taking the number of voxels contained in the mask and the spatial dimensions of each voxel, and is denoted as . Simultaneously, the spatial area of ​​the mask map boundary is estimated using a 3D reconstruction algorithm, denoted as... ;

[0164] Based on the geometric ratio between the structural volume and boundary area of ​​the radiotherapy target area, the regional compactness coefficient is calculated. This coefficient is used to assess the compactness of the radiotherapy target area structure. A value closer to 1 indicates that the area is closer to an ideal sphere and the structure is more compact; conversely, a value lower than 1 indicates a loose structure or irregular extensions. In the final radiotherapy target area mask map... Within the defined spatial region, extract the corresponding location in the prey heatmap. The average prey heatmap response is obtained by summing the response values ​​of each voxel in the region and dividing the sum by the number of voxels in that region. .

[0165] Average prey heatmap response This represents the average intensity of the location that is prioritized for response by the Secretary Bird optimization algorithm in the radiotherapy target area. Its value ranges from 0 to 1. A higher value indicates that the final delineated target area is supported by a stronger prey heatmap signal, reflecting the consistency between the model and the heatmap guidance in the response.

[0166] Example 1:

[0167] The radiotherapy department imaging workstation at Hospital A received CT image data from a liver cancer patient, numbered P0247. The patient had undergone an enhanced abdominal CT scan the previous day, resulting in 198 three-dimensional slice images. The image size was 512×512×198 pixels, with a voxel resolution of 0.75mm×0.75mm×1.2mm. During the pre-radiotherapy preparation phase, the examining physician initially determined that the patient had a mixed-density space-occupying lesion in the left lobe of the liver, with suspected overlap between the lesion and surrounding hepatic veins, and blurred edges. Conventional segmentation was insufficient to accurately determine its extent.

[0168] The system initiates the processing flow of the method of this invention. First, the system performs image preprocessing on the patient image. The input image dataset is defined as:

[0169] ;

[0170] The image set is generated after performing Gaussian filtering (σ=1.2), grayscale normalization, and structure enhancement. Multiple unstructured edge breakpoints were detected in slice number 104. The maximum value of the structural tensor appeared in the hepatic hilum, indicating that the structure in this region is highly complex.

[0171] The preprocessed image is input into the multi-scale super-resolution geometric modeling module. The system calculates the rate of change of edge curvature of the tumor candidate region in the 104th layer of the image. The value is 0.64, indicating structural heterogeneity. The value is 0.47, and the system executes the mapping function:

[0172] ;

[0173] The values ​​were set as follows: α=2.5, β=1.6, ε=0.05. The minimum reconstruction scale was calculated to be 1.11, and the maximum was 1.93. This parameter directly drives the generation of asymmetric multi-scale image patches, with significant enhancement in the area near the right hepatic vein, increasing the detail recovery rate by approximately 27.4%.

[0174] Edge intensity map After generation, normalization is performed to construct a heatmap. This figure shows that the response values ​​around the edge of the lesion are all higher than 0.78, and the area of ​​the heat concentration region in the upper left quadrant reaches 47 mm², which is significantly higher than the background average heat (0.41). The system marks this area as the critical boundary response area.

[0175] The system uses the heat map response and enhanced image as a basis. Start the target segmentation model construction and initialize the four types of parameters as follows:

[0176] Segmentation threshold =0.48; Edge localization parameters: σ=1.3, θ=0.92; Morphological radius: closing operation radius 3 pixels, opening operation radius 2 pixels; Multi-scale fusion factor λ={0.5,0.3,0.2} corresponds to different super-resolution scales.

[0177] Next, we extract image quality perception factors and sharpness indicators. =124.6, edge density =0.71, local structural complexity =3.18, overall image quality perception factor:

[0178] ;

[0179] The system updates the search radius of individual Secretary Birds accordingly. =1.9, perturbation scale =0.011, dynamic guiding factor =0.81. Then, 30 iterations were performed, with the loss function being:

[0180] ;

[0181] The optimal value was achieved in the 17th round, with Dice=0.926 and Hausdorff distance=4.3mm. The model automatically selected the segmentation result of this round as the final output.

[0182] The system will generate a 3D mask from the final result. The doctor confirmed in the visualization system that the system automatically classified the mask as a "clear target area" because: edge integrity coefficient = 0.91; compactness coefficient = 0.89; and prey heatmap average response = 0.84.

[0183] The system then exports the mask along with the patient information to the radiotherapy planning system Eclipsev16 and synchronizes it to the cloud-based image analysis platform for multi-center comparative studies.

[0184] To verify the overall performance of the system, we selected 34 patients from this batch and compared them using this method with the traditional method. The results are as follows:

[0185] Table 1. Data Comparison Between This Method and Traditional Methods

[0186] Case number method Dice coefficient HD distance Segmentation time (s) P0247 Traditional methods 0.853 8.9 232 Method of the present invention 0.926 4.3 98 P0225 Traditional methods 0.821 10.1 275 Method of the present invention 0.911 5.1 102 P0253 Traditional methods 0.863 7.6 251 Method of the present invention 0.935 4.2 93

[0187] Based on actual feedback, doctors noted in their comparative annotations that "in the complex tumor boundary regions of P0247 and P0253, the automatic delineation results provided by this system are significantly superior to previous experience-based segmentation tools in terms of visual integrity and boundary consistency," and recommended incorporating this system into routine pre-treatment procedures.

[0188] This invention addresses the characteristics of tumor region boundary curvature changes and structural heterogeneity by constructing an adaptive scale mapping function driven by the radiotherapy target area. Based on this function, a local asymmetric reconstruction scale factor is generated. Combined with a curvature-constrained local reconstruction function, high-fidelity reconstruction is performed on image blocks to form an enhanced radiotherapy target area feature response map. This effectively overcomes the limitation of traditional multi-scale image reconstruction methods that rely solely on fixed scale factors, enabling fine-scale reconstruction of boundary mutation regions and improving the spatial resolution and structural integrity of tumor edges.

[0189] This invention extracts image quality perception factors based on multi-scale geometric feature images, and combines three image evaluation dimensions—sharpness, edge density, and structural complexity—to define quantification formulas for adaptively setting the search radius, exploration step size, and perturbation scale of the secretary bird algorithm. This achieves the optimization algorithm's adaptive search capability under different image quality conditions. A prey heatmap is introduced as a dynamic guiding function embedded in the parameter update path of individual secretary birds. Parameter guiding weights based on response intensity are designed to drive each segmentation parameter to evolve towards the strong response region of the tumor boundary, significantly improving the adaptability of the segmentation parameters to the spatial structure of the target area. High-precision segmentation results can be stably obtained under different image contrast conditions.

[0190] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for precise delineation of radiotherapy target areas based on super-resolution geometry, characterized in that, Includes the following steps: S1. Obtain radiotherapy target area medical image data of the patient to be treated to form a radiotherapy target area medical image dataset. Perform image preprocessing on the radiotherapy target area medical image dataset to generate a preprocessed radiotherapy target area medical image dataset. S2. Input the preprocessed radiotherapy target area medical image dataset into the multi-scale super-resolution geometric modeling module, perform super-resolution reconstruction and geometric feature enhancement for multiple scales, and output multi-scale super-resolution geometric feature image data; S3. Calculate edge intensity map based on multi-scale super-resolution geometric feature image data to generate prey heat map; S4. Construct a target region segmentation model, and construct a set of segmentation threshold parameters, a set of edge localization parameters, a set of morphological structure adjustment parameters, and a set of multi-scale fusion factors; S5. Based on the multi-scale super-resolution geometric feature image data, extract the image quality perception factor, and use the image quality perception factor to adaptively initialize the search radius, exploration step size and perturbation scale of the secretary bird optimization algorithm to obtain the initial secretary bird population parameters. Embed the prey heat map as a dynamic guiding function into the secretary bird optimization algorithm. During the iteration process, update the parameters of the target area segmentation model through global search and local fine-tuning to generate the optimal segmentation parameter set. S6. Apply the optimal segmentation parameter set to the multi-scale super-resolution geometric feature image data to perform automatic target segmentation and obtain the radiotherapy target delineation results.

2. The method for precise delineation of radiotherapy target area based on super-resolution geometry according to claim 1, characterized in that, S1 includes the following steps: S11. Acquire medical image data of the radiotherapy target area of ​​the patient to be treated, set a uniform acquisition time window, and form a medical image dataset of the radiotherapy target area. : ； in, Indicates the first Zhang 3D slice image, This represents the spatial coordinates of a 3D slice image in 3D voxel space. This represents the pixel grayscale value corresponding to a 3D slice image. This represents the total number of 3D slice images; S12. Perform image preprocessing on the radiotherapy target area medical image dataset to obtain the preprocessed radiotherapy target area medical image dataset. This includes the following preprocessing operations: The noise suppression operation employs an isotropic Gaussian filtering method to eliminate artifacts and high-frequency interference during the imaging process, and outputs a noise-suppressed image set. Gray-level normalization processing converts the gray-level values ​​of each 3D slice image. Mapping to interval ; The structure enhancement process calculates image edge changes based on the local structure tensor and gradient intensity, enhances the edge response of the image in the noise-suppressed image set, and outputs a structure-enhanced image set.

3. The method for precise delineation of radiotherapy target area based on super-resolution geometry according to claim 1, characterized in that, S2 includes the following steps: S21. Preprocess the medical image dataset of the radiotherapy target area. Input the multi-scale super-resolution geometric modeling module, based on each 3D slice image Structural heterogeneity level of intermediate radiotherapy target area With the rate of change of edge curvature Constructing asymmetric scale sets Each reconstruction scale factor in the asymmetric scale set Calculated using an adaptive scaling function driven by the radiotherapy target area: ； in, Indicates the first In the three-dimensional slice image, the first Reconstruction scale factor of a local region of the radiotherapy target area Indicates the first The rate of change in the edge curvature of each region reflects the degree of local geometric mutation at the tumor boundary. This indicates the level of structural heterogeneity between the tumor region and the background tissue in the entire image. This represents the maximum scale control factor. The variation adjustment coefficient, This is the lower limit protection constant; S22. Local areas of each radiotherapy target zone According to the corresponding reconstruction scale factor High-fidelity local image patches are constructed by performing curvature-constrained super-resolution reconstruction using a reconstruction function. Reconstruction function Defined as: ； in, Represents a 3D slice image The Middle A local area of ​​the radiotherapy target zone to be reconstructed. This represents the initial super-resolution image generated using unconstrained interpolation, including bicubic interpolation. To optimize the target image patch, For the first The target curvature distribution function of the tumor boundary in each region, Indicates optimization of target image patch The second derivative, For the local area of ​​the radiotherapy target zone The domain of the integral, is the regularization coefficient for the edge curvature constraint term; S23. All high-fidelity local image patches after curvature-constrained super-resolution reconstruction. Geometric alignment and edge continuity stitching are performed to construct a multi-scale geometrically enhanced representation of the complete image. ; S24. Based on multi-scale geometric enhancement representation Constructing a characteristic response map of the enhanced radiotherapy target area : in, Represents the edge gradient map. Indicators representing fusion mutations between tumor and non-tumor regions in terms of density, texture, and edge direction. It serves as a regulatory factor for the fusion of edge and structural mutation information; S25. Output enhanced radiotherapy target area feature response maps for all 3D slice images, and output multi-scale super-resolution geometric feature image data. .

4. The method for precise delineation of radiotherapy target area based on super-resolution geometry according to claim 3, characterized in that, S3 includes the following steps: S31. Multi-scale super-resolution geometric feature image data The edge intensity calculation module is used to calculate the feature response map of each enhanced radiotherapy target area. Calculate the edge intensity map of the characteristic response map of the enhanced radiotherapy target area. Edge intensity map This represents the gray-level gradient magnitude of each voxel in the three-dimensional spatial direction of the enhanced radiotherapy target region feature response map. This is achieved by simultaneously calculating the enhanced radiotherapy target region feature response map. In space , , The partial derivatives in three directions are obtained by summing the squares of the partial derivatives in three directions and taking the square root. This is used to describe the degree of intensity variation of the radiation target boundary in different directions. S32. Edge Intensity Map Normalization is performed to obtain the normalized edge response map. The normalization process involves subtracting the minimum intensity value from the edge intensity map for each voxel in the edge intensity map, and then dividing by the difference between the maximum and minimum values. This maps all edge intensity values ​​to the interval between 0 and 1, resulting in the normalized edge response map. Preserve the relative distribution characteristics of the edges; S33. Based on Normalized Edge Response Map Constructing a prey heatmap Prey heatmap This is achieved by normalizing the edge response map. With mask image The mask image is obtained by multiplying the pixel values ​​at corresponding positions point by point. It is a spatial mask that takes a value of 1 only in the candidate region of the radiotherapy target area and a value of 0 in other regions, used to eliminate interference from non-medically relevant areas, prey heatmap Preserve edge response information within the candidate radiotherapy target area; S34. Output prey heatmaps of all enhanced radiotherapy target region characteristic response maps. Construct a prey heatmap dataset .

5. The method for precise delineation of radiotherapy target area based on super-resolution geometry according to claim 4, characterized in that, S4 includes constructing a target segmentation model. The target segmentation model uses the prey heatmap dataset. Based on the input, it includes four core parameter sets: segmentation threshold parameter set, edge localization parameter set, morphological structure adjustment parameter set, and multi-scale fusion factor set, which are used to realize response region extraction, edge fine adjustment, structural continuity optimization, and scale information integration of the radiotherapy target area, respectively.

6. The method for precise delineation of radiotherapy target area based on super-resolution geometry according to claim 5, characterized in that, The set of segmentation threshold parameters is as follows: ,in Indicates the first The initial segmentation threshold used for binarization processing in the heatmap of the prey is used to compare the voxel response intensity in the heatmap with the threshold. Voxels that are greater than or equal to the initial segmentation threshold are marked as candidate regions for the initial radiotherapy target area, while voxels that are less than the initial segmentation threshold are excluded, forming the initial radiotherapy target area mask map. The edge positioning parameter set is ,in This represents the edge smoothing adjustment coefficient, used to control the kernel function scale of the prey heatmap edge response map during edge smoothing processing. This represents the edge enhancement weighting factor, used to adjust the degree to which edge intensity enhances the boundary morphology of the initial radiotherapy target area mask map. It adjusts the normalized edge response map using the edge smoothing adjustment coefficient. After smoothing under control, the image is fused with the initial radiotherapy target mask, and edge enhancement weighting factors are incorporated. Adjustments are made to create an edge adjustment mask image, which is used to enhance the target area boundary contour; The set of morphological structure adjustment parameters is as follows ,in This represents the radius of the structuring element used for the closing operation, controlling the degree to which small holes in the initial radiotherapy target mask are filled. The radius of the structuring element used for the opening operation is used to eliminate misidentified regions caused by spurious responses. The closing and opening operations are performed sequentially on the edge-adjusted mask to form the structure-adjusted radiotherapy target area mask. The set of multi-scale fusion factors is ,in Indicates the first Zhang 3D slice image in the first Weighting factors at various scales are used to control the participation ratio of morphological modulation results at different scales. A structure-modified radiotherapy target area mask is generated for each 3D slice image at different scales. The radiotherapy target area masks are weighted and summed separately, and then fused according to the weighting factors. Multi-scale information is integrated to form the final radiotherapy target mask map.

7. The method for precise delineation of radiotherapy target area based on super-resolution geometry according to claim 5, characterized in that, S5 includes the following steps: S51. Based on multi-scale super-resolution geometric feature image data Extracting the set of image quality perceptual factors Each image quality perceptual factor Based on clarity index Edge density index and structural complexity index Perform weighted fusion; S52. Image quality perceptual factor Used for initializing population parameters and setting the search radius in the Secretary Bird optimization algorithm. Exploration Step Size Disturbance scale ; S53. Construct the parameter set for the target segmentation model. The optimization objective function is: ； in, For the first Predictive mask map of Zhang Zengqiang's radiotherapy target area characteristic response map. Its artificial reference image, Dice is the Dice coefficient, and HD is the Hausdorff distance; S54. Prey heatmap dataset As a dynamic guiding function, it is embedded in the Secretary Bird optimization algorithm to construct a guiding weight matrix for the segmentation parameters. In the guided target segmentation model, the first The evolution direction and strength of each parameter, where each guiding weight From the prey heatmap in the parameter control area The weighted integral within the range is calculated to obtain: ； in, Indicates the first Segmentation parameters in Zhang Zengqiang's radiotherapy target area characteristic response map The average intensity of prey response in the controlled area Indicates parameters The corresponding spatial range of action; Secretary bird individual in the first In the nth iteration, the th The update definition for each parameter is: ； in, For the first The secretary bird individual in the first The th iteration Each segmentation parameter For the first The secretary bird individual in the first The th iteration Each segmentation parameter The segmentation parameters are those for the current optimal individual. The perturbation scale. This is standard normally distributed noise; S55. Iteratively update the parameter set of the target segmentation model using the Secretary Bird optimization algorithm. To minimize the optimization objective function Using this as the criterion, the optimal splitting parameter set is finally output. .

8. The method for precise delineation of radiotherapy target area based on super-resolution geometry according to claim 7, characterized in that, S6 includes the following steps: S61. The optimal segmentation parameter set obtained through the Secretary Bird optimization algorithm is applied to the multi-scale super-resolution geometric feature image dataset. Based on the optimal segmentation parameter set, the automatic segmentation process for the radiotherapy target area is executed sequentially. The segmentation process includes: Using the optimal set of segmentation threshold parameters Enhanced radiotherapy target area characteristic response map Thresholding is performed to initially extract the response region and form an initial mask image; Based on the optimal edge localization parameter set Joint prey heat map The response region is used to construct an edge guiding function, and the normalized edge response map is fused to perform boundary optimization, forming an edge adjustment mask map; Based on the optimal morphological structure adjustment parameter set Perform closing and opening operations on the edge adjustment mask image to enhance structural connectivity and noise reduction capabilities; Based on the optimal multi-scale fusion factor set The structure-adjusted mask images at each scale are weighted and fused to obtain the final radiotherapy target area mask image. ; S63. Output the final set of radiotherapy target mask maps. Finally, based on the structural features and morphological distribution of the radiotherapy target area mask, the radiotherapy target area delineation results in each 3D slice image are classified. The classification rules are as follows: Clear target area: edge integrity coefficient ≥ 0.9, region compactness coefficient ≥ 0.85, average prey heatmap response ≥ 0.8; Blurred target region: 0.7 ≤ edge integrity coefficient < 0.9, region compactness coefficient ≥ 0.7, average prey heatmap response ∈ [0.5, 0.8); Irregular target area: edge integrity coefficient < 0.7 or average prey heatmap response < 0.5; S64. Obtain the final radiotherapy target delineation result dataset. And their corresponding category tags.

Images