Search

CN-122023447-A - Accurate delineating method of radiotherapy target area based on superdivision geometry

CN122023447ACN 122023447 ACN122023447 ACN 122023447ACN-122023447-A

Abstract

The invention discloses a superdivision geometry-based accurate sketching method for a radiotherapy target area, which comprises the steps of S1, generating a preprocessed medical image data set of the radiotherapy target area, S2, outputting multi-scale superdivision geometry characteristic image data, S3, calculating an edge intensity image based on the multi-scale superdivision geometry characteristic image data to generate a target area heat image, S4, constructing a target area segmentation model, S5, extracting an image quality perception factor according to the multi-scale superdivision geometry characteristic image data, embedding the target area heat image as a dynamic guiding function into a secretary bird optimization algorithm, generating an optimal segmentation parameter set by globally searching and locally trimming parameters of the target area segmentation model in an iterative process, S6, using the optimal segmentation parameter set to act on the multi-scale superdivision geometry characteristic image data, and executing automatic segmentation of the target area to obtain a radiotherapy target area sketching result. The invention can obviously improve the precision, stability and clinical application value of segmentation.

Inventors

  • FU DANLI
  • LUO PINGPING
  • YU WEIFEI
  • WANG XIAOQIU

Assignees

  • 丽水市人民医院

Dates

Publication Date
20260512
Application Date
20260210

Claims (8)

  1. 1. A precise delineation method of a radiotherapy target zone based on superdivision geometry is characterized by comprising the following steps: S1, acquiring medical image data of a radiotherapy target area of a patient to be processed to form a medical image data set of the radiotherapy target area, and performing image preprocessing on the medical image data set of the radiotherapy target area to generate a medical image data set of the preprocessed radiotherapy target area; s2, inputting the preprocessed radiotherapy target region medical image dataset into a multi-scale super-resolution geometric modeling module, performing super-resolution reconstruction and geometric feature enhancement aiming at a plurality of scales, and outputting multi-scale super-resolution geometric feature image data; s3, calculating an edge intensity map based on the multi-scale super-resolution geometric feature image data, and generating a hunting heat map; S4, constructing a target region segmentation model, and constructing a segmentation threshold parameter set, an edge positioning parameter set, a morphological structure adjusting parameter set and a multi-scale fusion factor set; S5, extracting image quality perception factors according to the multi-scale super-resolution geometric feature image data, carrying out self-adaptive initialization on the searching radius, the exploration step length and the disturbance scale of a secretary bird optimization algorithm by utilizing the image quality perception factors to obtain initialized secretary bird population parameters, embedding a prey heat map as a dynamic guiding function into the secretary bird optimization algorithm, and updating the parameters of a target region segmentation model through the combination of global searching and local fine tuning in the iterative process to generate an optimal segmentation parameter set; s6, using the optimal segmentation parameter set to act on the multi-scale super-segmentation geometrical feature image data, and executing automatic segmentation of the target area to obtain a radiotherapy target area sketching result.
  2. 2. The method for precisely delineating a radiotherapy target zone based on superdivision geometry according to claim 1, wherein the step S1 comprises the following steps: s11, acquiring medical image data of a radiotherapy target region of a patient to be treated, setting a uniform acquisition time window, and forming a medical image data set of the radiotherapy target region : ; Wherein, the Represent the first The three-dimensional slice image is made, Representing the spatial coordinates of the three-dimensional slice image in three-dimensional voxel space, Representing the corresponding pixel gray value of the three-dimensional slice image, The total number of the three-dimensional slice images is; s12, performing image preprocessing on the medical image data set of the radiotherapy target region to obtain a medical image data set of the preprocessed radiotherapy target region The method comprises the following pretreatment operations: noise suppression operation, namely eliminating artifacts and high-frequency interference in the imaging process by adopting an isotropic Gaussian filtering method, and outputting a noise suppression image set; gray scale normalization processing, which is to normalize gray scale value of each three-dimensional slice image Mapping to intervals ; And (3) structural enhancement processing, namely calculating image edge change based on the local structural tensor and the gradient strength, responding to the image enhancement edges in the noise-suppressed image set, and outputting a structural enhancement image set.
  3. 3. The method for precisely delineating a radiotherapy target zone based on superdivision geometry according to claim 1, wherein the step S2 comprises the following steps: s21, preprocessing medical image data set of radiotherapy target region Inputting a multi-scale super-division geometric modeling module according to each three-dimensional slice image Level of structural heterogeneity in a medium-beam radiotherapy target zone Rate of change with edge curvature Constructing asymmetric scale sets Each reconstructed scale factor in the asymmetric scale set Calculation of an adaptive scale mapping function driven by a radiotherapy target zone: ; Wherein, the Represent the first Sheet of three-dimensional slice image The reconstruction scale factors of local areas of the individual radiotherapy target zone, Represent the first The edge curvature change rate of each region reflects the local geometric mutation degree of the tumor boundary, Represents the level of structural heterogeneity between the tumor region and background tissue in the entire image, Representing the maximum scale control factor(s), In order to adjust the coefficient of variation, Is a lower limit protection constant; s22, for each radiotherapy target zone local area According to the corresponding reconstructed scale factor Performing curvature constrained super-resolution reconstruction operations using reconstruction functions to construct high-fidelity local image blocks Reconstruction function The definition is as follows: ; Wherein, the Representing three-dimensional slice images Middle (f) A local area of the radiotherapy target zone to be reconstructed, Representing an initial super-resolution image generated using unconstrained interpolation, including bicubic interpolation, In order to optimize the target image block, Is the first A target curvature distribution function of the tumor boundary in each region, Representing optimization target image blocks Is used for the first derivative of (c), For local area of radiotherapy target zone Is defined in the integral domain of (a), Regularization coefficients that are edge curvature constraint terms; s23, reconstructing all high-fidelity local image blocks subjected to curvature constraint superdivision Performing geometric alignment and edge continuity stitching to construct a multi-scale geometric enhancement representation of a complete image ; S24, enhancing the representation according to multi-scale geometry Construction of enhanced radiotherapy target zone characteristic response diagram : Wherein, the An edge gradient map is represented and is shown, Indicating fusion mutation indexes of tumor areas and non-tumor areas in the density, texture and edge directions, The regulatory factor is fused with the structural mutation information for the edge; S25, outputting an enhanced radiotherapy target region characteristic response map of all three-dimensional slice images, and outputting multi-scale super-resolution geometric characteristic image data 。
  4. 4. The method for precisely delineating a radiotherapy target zone based on superdivision geometry according to claim 3, wherein S3 comprises the following steps: S31, performing multi-scale super-resolution geometric feature image data Inputting the characteristic response map of the target area of each enhanced radiotherapy into an edge intensity calculation module Computing edge intensity map of enhanced radiotherapy target region characteristic response map Edge intensity map Representing the gray gradient amplitude of each voxel in the enhanced radiotherapy target region characteristic response map in three-dimensional space direction, and simultaneously calculating the enhanced radiotherapy target region characteristic response map In space of 、 、 The three partial derivatives in the three directions are obtained by summing squares of the partial derivatives in the three directions and then squaring the partial derivatives in the three directions, and are used for describing intensity variation degrees of the boundary of the radiotherapy target zone in different directions; s32, edge intensity map Normalization processing is carried out to obtain a normalized edge response graph The normalization process is to subtract the minimum intensity value in the edge intensity map from the edge intensity value of each voxel in the edge intensity map, divide the value by the difference between the maximum value and the minimum value, map all edge intensity values into the interval from 0 to 1, and normalize the edge response map Preserving the relative distribution characteristics of the edges; S33, based on normalized edge response graph Construction of hunting heat map Hunting heat map By normalizing the edge response graph And mask map The pixel values at the corresponding positions are multiplied point by point to obtain a mask map Is a space mask with 1 value only in the target area candidate area and 0 value in other areas, and is used for eliminating the interference of non-medical related areas, and is used for hunting heat map Retaining edge response information in a candidate area of the radiotherapy target zone; s34, outputting a hunting heat map of characteristic response maps of all enhanced radiotherapy target areas Construction of a hunting heat map dataset 。
  5. 5. The method for precisely delineating a radiotherapy target zone based on superdivision geometry according to claim 4, the method is characterized in that S4 comprises the steps of constructing a target region segmentation model Target segmentation model with hunting heat map dataset The method is used for inputting a foundation and comprises four core parameter sets, namely a segmentation threshold parameter set, an edge positioning parameter set, a morphological structure adjusting parameter set and a multi-scale fusion factor set, which are respectively used for realizing the response area extraction, the edge fine adjustment, the structure continuity optimization and the scale information integration of a radiotherapy target region.
  6. 6. The method for precisely delineating a radiotherapy target region based on superdivision geometry according to claim 5, wherein the segmentation threshold parameter set is Wherein Represent the first The initial segmentation threshold value is used for binarization processing in Zhang Liewu heat maps, the initial segmentation threshold value is used for comparing voxel response intensity in the heat maps of the hunting object with the threshold value, voxels which are larger than or equal to the initial segmentation threshold value are marked as initial radiotherapy target region candidate regions, and voxels which are smaller than the initial segmentation threshold value are excluded, so that an initial radiotherapy target region mask map is formed; the edge positioning parameter set is Wherein Represents the edge smoothing adjustment coefficient, is used for controlling the kernel function scale of the edge response diagram of the hunting heat diagram when the edge smoothing treatment is carried out, Representing edge enhancement weight factors for adjusting the enhancement degree of edge intensity to the boundary morphology of the mask image of the initial radiotherapy target region, and smoothing the normalized edge response image at the edge to adjust coefficients After smoothing under control, fusing with the mask map of initial radiotherapy target region, and fusing with edge enhancement weight factors Adjusting to form an edge adjustment mask map for enhancing the boundary profile of the target region; The morphological structure adjusting parameter set is Wherein Representing the radius of the structural element used for the closed operation, controlling the filling degree of the small holes in the mask of the initial radiotherapy target zone, Representing the radius of a structural element used for open operation, removing a false identification area caused by false response, and sequentially executing closed operation and open operation on the edge adjustment mask map to form a radiotherapy target area mask map after structure adjustment; The multiscale fusion factor set is Wherein Represent the first Sheet three-dimensional slice image at the first The weighting factors under the various scales are used for controlling the participation proportion of morphological adjustment results under different scales, each three-dimensional slice image generates a part of radiotherapy target region mask images with structure adjusted under different scales, and the radiotherapy target region mask images are respectively weighted and summed according to the fusion weighting factors And (5) carrying out multi-scale information integration to form a final radiotherapy target region mask map.
  7. 7. The method for precisely delineating a target area for radiotherapy based on superdivision geometry according to claim 5, wherein S5 comprises the following steps: s51, image data are obtained according to multi-scale super-resolution geometric features Extracting image quality perception factor set Each image quality perception factor Based on sharpness index Edge density index And a structural complexity index Carrying out weighted fusion; s52, sensing the image quality factor Be applied to secretary bird optimization algorithm population parameter initialization, set for search radius Search step length Disturbance scale ; S53, constructing a target region segmentation model parameter set Is set to the optimum objective function: ; Wherein, the Is the first Zhang Zengjiang a predictive mask map of a radiotherapy target volume characteristic response map, For the purpose of which reference is made manually to the figures, The Dice is a Dice coefficient, and the HD is a Hausdorff distance; S54, collecting hunting heat map data Embedding the vector as a dynamic guiding function into a secretary bird optimization algorithm, and guiding a weight matrix by constructing a segmentation parameter Guidance of the first target segmentation model The direction and intensity of evolution of each parameter, each guiding weight In the parameter control area by hunting heat map And (3) obtaining weighted integral calculation in the process: ; Wherein, the Represent the first Segmentation parameters in Zhang Zengjiang radiotherapy target region characteristic response diagram The hunting response of the controlled area is of average intensity, Representing parameters Corresponding space action range; The individual secretary bird is at the first stage In the second iteration pair The update of the individual parameters is defined as: ; Wherein, the Is the first Individual secretary bird in the first place The first iteration The number of segmentation parameters is chosen such that, Is the first Individual secretary bird in the first place The first iteration The number of segmentation parameters is chosen such that, For the segmentation parameters in the currently optimal individual, For the dimensions of the perturbation, Noise is normally distributed for standard; S55, iteratively updating a target region segmentation model parameter set through secretary bird optimization algorithm To minimize and optimize the objective function As a criterion, finally outputs the optimal segmentation parameter set 。
  8. 8. The method for precisely delineating a target area for radiotherapy based on superdivision geometry according to claim 7, wherein the step S6 comprises the following steps: S61, acting an optimal segmentation parameter set obtained through iteration of a secretary bird optimization algorithm on a multi-scale super-segmentation geometrical feature image data set, and sequentially executing an automatic segmentation process of a radiotherapy target region based on the optimal segmentation parameter set, wherein the segmentation process comprises the following steps: Using optimal segmentation threshold parameter sets Response map to enhanced radiotherapy target zone characteristics Performing threshold processing, and primarily extracting a response area to form an initial mask map; Positioning parameter set based on optimal edge Joint hunting heat map An edge guide function is constructed in the response area of the mask pattern, and the normalized edge response pattern is fused to perform boundary optimization to form an edge adjustment mask pattern; Adjusting parameter sets based on optimal morphological structure Performing a closed operation and an open operation on the edge adjustment mask map, and enhancing structural connectivity and denoising capability; Based on optimal multiscale fusion factor set Weighting and fusing the mask patterns with the structures adjusted under each scale to obtain a final radiotherapy target region mask pattern ; S63, outputting a final radiotherapy target zone mask pattern set Finally classifying the drawing result of the radiotherapy target zone in each three-dimensional slice image according to the structural characteristics and the morphological distribution of the mask map of the radiotherapy target zone, wherein the classification rule is as follows: The edge integrity coefficient of the clear target area is more than or equal to 0.9, the area compactness coefficient is more than or equal to 0.85, and the response average value of the hunting heat map is more than or equal to 0.8; fuzzy target area 0.7 is less than or equal to edge integrity coefficient <0.9, area compactness coefficient is more than or equal to 0.7, and hunting heat map response average value epsilon [0.5,0.8); irregular target area with edge integrity coefficient <0.7 or hunting heat pattern response mean <0.5; S64, obtaining a final radiotherapy target zone sketching result data set Which corresponds to the class label.

Description

Accurate delineating method of radiotherapy target area based on superdivision geometry Technical Field The invention relates to the technical field of medical treatment, in particular to a precise drawing method of a radiotherapy target zone based on superdivision geometry. Background Along with the continuous development of image segmentation technology and medical image processing methods, accurate radiotherapy provides higher precision and efficiency requirements for target region identification and delineation, and how to accurately delineate tumor tissue boundaries from CT and MRI three-dimensional medical images in tumor radiotherapy is a key link affecting treatment planning and dose distribution control. At present, a radiotherapy target region sketching method widely applied clinically still mainly comprises manual segmentation or semiautomatic segmentation, relies on subjective judgment of doctors on focus boundaries and a simple image processing tool to finish target region marking, and is flexible, but has the obvious defects that firstly, manual operation is time-consuming and labor-consuming, and the efficiency is low when facing large-scale image data; secondly, because the resolution of the imaging equipment is limited, especially under the condition of noise, artifact or boundary blurring, errors are easy to occur in subjective judgment of doctors, and the target area sketching is offset or omitted, so that the follow-up radiotherapy effect is influenced. In the automatic image segmentation research, algorithms based on deep learning or image enhancement are introduced in recent years, but the image enhancement still has a certain limitation, on one hand, due to the spatial resolution and imaging quality of an original image, the model is difficult to accurately restore the detail characteristics of the edge of a tumor, and particularly in a tumor area with complex morphology or strong structural heterogeneity, the problems of boundary blurring, regional splitting or adhesion are often shown. On the other hand, the existing parameter optimization mode mostly adopts fixed parameters or relies on expert experience to manually adjust, has no adaptability and global searching capability, and is difficult to maintain the stability and high precision of the segmentation effect under different cases and image quality conditions. In addition, most of the current optimization strategies do not effectively introduce an image structural feature driving mechanism, the optimization algorithm lacks dynamic perception capability to a target area, is easy to sink into local optimization, and cannot fully capture microstructure changes of tumor boundaries. In view of the foregoing, a new method for integrating multi-scale modeling and intelligent optimization mechanisms is needed to improve the boundary characterization capability of a tumor target area and the scientificity of radiotherapy path planning. Disclosure of Invention The invention aims to provide a precise delineating method of a radiotherapy target region based on superdivision geometry, which can remarkably improve the precision, stability and clinical application value of segmentation. According to the embodiment of the invention, the accurate delineation method of the radiotherapy target zone based on the superdivision geometry comprises the following steps: S1, acquiring medical image data of a radiotherapy target area of a patient to be processed to form a medical image data set of the radiotherapy target area, and performing image preprocessing on the medical image data set of the radiotherapy target area to generate a medical image data set of the preprocessed radiotherapy target area; s2, inputting the preprocessed radiotherapy target region medical image dataset into a multi-scale super-resolution geometric modeling module, performing super-resolution reconstruction and geometric feature enhancement aiming at a plurality of scales, and outputting multi-scale super-resolution geometric feature image data; s3, calculating an edge intensity map based on the multi-scale super-resolution geometric feature image data, and generating a hunting heat map; S4, constructing a target region segmentation model, and constructing a segmentation threshold parameter set, an edge positioning parameter set, a morphological structure adjusting parameter set and a multi-scale fusion factor set; S5, extracting image quality perception factors according to the multi-scale super-resolution geometric feature image data, carrying out self-adaptive initialization on the searching radius, the exploration step length and the disturbance scale of a secretary bird optimization algorithm by utilizing the image quality perception factors to obtain initialized secretary bird population parameters, embedding a prey heat map as a dynamic guiding function into the secretary bird optimization algorithm, and updating the parameters of a target region segmentation model thr