Image dose fusion-based silicone implant region image registration method and device

Abstract

The application discloses a silicone implant region image registration method and device based on image dose fusion, relates to the technical field of medical image processing, and comprises the following steps: S1, implant registration data is collected, and pretreatment is performed on the implant registration data; S2, a gray-dose joint profile line is generated, gray phase peaks and dose phase peaks are extracted, dynamic time warping matching is performed, and gray-dose phase decoupling data is generated; S3, a boundary phase graph is established, and alpha-expansion graph cut optimization is performed according to dose-locked gray peaks, free gray peaks and artifact-pulled substitute values, thereby generating a boundary credible label graph and the gray-dose phase decoupling data; and S4, a gray-dose label registration cost function is constructed, spatial transformation parameters are solved, and a silicone implant region image registration result is output. The problems that, in the prior art, the gray artifact boundary is mis-pulled in silicone implant region image registration, the real boundary of the implant body is unstable in correspondence, and the dose distribution space is large in overlap deviation are solved.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and in particular to a method and apparatus for image registration of silicone implantation areas based on image dose fusion. Background Technology

[0002] With the development of medical imaging equipment, radiotherapy planning systems, and postoperative follow-up technologies, medical imaging data at different time points, imaging modalities, and spatial resolutions are widely used for implantation area observation, radiotherapy dose verification, and treatment outcome evaluation. Image registration technology, which maps reference images and images to be registered to a unified spatial coordinate system, establishes a correspondence between tissue structures, implant boundaries, and dose distribution, and has become a crucial aspect of medical image processing. Existing technologies include research on multi-source image feature point correspondence, cross-modal image enhancement, and medical image registration model training.

[0003] For example, application CN114365189A discloses an image registration apparatus, an image generation system, an image registration method, and an image registration program. The image registration apparatus is communicatively connected to a range sensor (10) and a camera (20). The range sensor generates a reflected light image containing distance information by sensing reflected light reflected from an object through light illumination by a light-receiving element, and generates a background light image in the same coordinate system as the reflected light image by sensing background light relative to the reflected light by the light-receiving element. The camera generates a camera image by detecting incident light from the outside by a camera element. The image registration apparatus includes: an image acquisition unit (41) that acquires a reflected light image, a background light image, and a camera image; and an image processing unit (42) that performs image registration of the reflected light image and the camera image in the same coordinate system as the background light image by determining the correspondence between feature points of the background light image and feature points of the camera image.

[0004] For example, application CN118279364B discloses a method for registering MRI images and CBCT images, including: S1, labeling the vertebral segments in the CBCT images and their paired CT and MRI images, and segmenting them to obtain corresponding vertebral segment images; S2, obtaining several sets of paired CBCT and CT vertebral segment images through S1, performing registration, and using the CT and CBCT vertebral segment images as a sample set to train an enhancement inference model; S3, inputting the CBCT vertebral segment images processed by S1 into the enhancement inference model for enhancement, and using the enhancement result and its paired MRI vertebral segment images as training samples to train the registration inference model; S4, inputting the CBCT and MRI vertebral segment images to be registered after processing by S1 into the registration inference model to obtain the registration result.

[0005] However, during image registration of the silicone implant area, differences in silicone material interfaces, metal markers, local hardened tissue, and imaging conditions can easily lead to bright rings, dark bands, and local signal loss on the outer side of the implant, causing the gray-level abrupt change peak to no longer stably coincide with the actual outer boundary of the implant. Existing registration schemes typically rely on gray-level mutual information, feature point correspondence, segmented contours, or structural similarity after cross-modal enhancement for spatial transformation solutions, failing to incorporate the normal misalignment relationship between the gray-level abrupt change peak and the dose phase peak, artifact traction costs, and reliable boundary labels into the registration constraints. When artifact abrupt change bands are mistakenly used as the outer boundary of the implant for registration, the floating image is easily pulled by the incorrect boundary, resulting in outward expansion, inward contraction, and local offset of the silicone implant boundary after registration, thereby affecting the spatial correspondence between the implant area and the surrounding soft tissue and dose distribution.

[0006] Therefore, in order to address the above problems, there is an urgent need for a method and device for image registration of silicone implantation areas based on image dose fusion. Summary of the Invention

[0007] To address the problem of grayscale artifact boundary mistrapping in existing silicone implantation area image registration, which leads to unstable correspondence between the implant's true boundary and significant deviations in dose distribution spatial overlay, this invention provides a silicone implantation area image registration method and apparatus based on image dose fusion. The technical solution is as follows: On the one hand, an image registration method for silicone implantation regions based on image dose fusion is provided, the method comprising: S1 constructs a unified coordinate system for implantation registration using the reference image coordinate space, collects implantation registration data, and performs grayscale denoising, grayscale scale correction, and spatial sampling alignment processing on the implantation registration data, outputting the preprocessed implantation registration data. S2, based on the preprocessed implantation registration data, construct a reference image marker surface and a floating image marker surface, generate gray agent joint profile lines along the outward normal of the reference image marker surface and the floating image marker surface respectively, extract the gray-scale phase peak and the dose phase peak in the gray agent joint profile lines respectively, and perform dynamic time warping matching on the gray-scale phase peak and the dose phase peak to generate gray agent phase misalignment data containing dose-locked gray-scale peak, free gray-scale peak and artifact traction cost value; S3. Based on the phase misalignment data of the gray agent, a boundary phase map is established, and α-expansion map cut optimization is performed according to the dose-locked gray peak, free gray peak and artifact traction cost to generate a boundary reliable label map and gray agent phase decoupling data. S4. Based on gray agent phase misalignment data, gray agent phase decoupling data and boundary reliable label map, a gray agent label registration cost function is constructed according to stable boundary nodes, artifact traction nodes and gray scale failure nodes. The spatial transformation parameters are solved and the image registration results of the silicone implantation area are output.

[0008] Further, the specific steps for constructing a unified coordinate system for implantation registration using the reference image coordinate space, acquiring implantation registration data, and performing grayscale denoising, grayscale scale correction, and spatial sampling alignment on the implantation registration data are as follows: A unified coordinate system for implantation registration is constructed using the reference image coordinate space. The center point of the voxel at the lower left corner of the first slice of the reference image is set as the origin. The X-axis is along the direction of increasing column number of the reference image, the Y-axis is along the direction of increasing row number of the reference image, and the Z-axis is along the direction of increasing slice number of the reference image. A one-to-one transformation relationship between the voxel coordinate values ​​of the reference image and the DICOM patient coordinates is established based on the plane voxel spacing value, slice thickness value, and orientation matrix of the reference image. The floating image voxel coordinate values ​​and dose voxel coordinate values ​​are mapped to the unified coordinate system for implantation registration according to the same transformation relationship. Implantation registration data of the silicone implantation area is acquired, including the reference image. The preprocessing steps for implant registration data include: orientation matrix, floating image orientation matrix, dose distribution orientation matrix, reference image voxel gray value, floating image voxel gray value, reference image voxel coordinate value, floating image voxel coordinate value, dose voxel coordinate value, implant surface marker coordinate value, metal marker center coordinate value, reference image planar voxel spacing value, floating image planar voxel spacing value, reference image layer thickness value, floating image layer thickness value, unit voxel dose value, dose voxel spacing value, and dose voxel layer thickness value. For the acquired implant registration data, a nonlocal mean filtering algorithm is used to perform voxel gray-level noise suppression and boundary structure preservation. A histogram matching algorithm is used to perform cross-time gray-level scale correction on the image gray-level distribution in the implant registration data. A trilinear interpolation algorithm is used to unify the spatial sampling interval and align the coordinate sampling positions of the implant registration data, outputting the preprocessed implant registration data.

[0009] Further, based on the preprocessed implant registration data, a reference image marker surface and a floating image marker surface are constructed. The specific steps for generating gray-media joint profile lines along the outward normals of the reference image marker surface and the floating image marker surface are as follows: Read the preprocessed implant registration data and group the coordinate values ​​of the marker points on the implant surface according to the source markers of the reference image and the floating image; perform a three-dimensional convex hull algorithm and Laplacian surface smoothing on each group of marker point coordinate values ​​on the implant surface to obtain the reference image marker surface and the floating image marker surface; generate surface sampling points on the corresponding marker surfaces according to the voxel spacing values ​​of the reference image plane and the floating image plane; calculate the outward normal vector of each surface sampling point and extend the preset normal sampling length along the outward normal vector to the inner and outer sides of the marker surface to generate gray-media joint profile lines; wherein, the sampling interval of the gray-media joint profile lines is determined by the smaller value between the corresponding image plane voxel spacing value and the corresponding image layer thickness value.

[0010] Further, the specific steps for extracting the gray-scale phase peaks and dose phase peaks from the gray-dose joint profile are as follows: Read the voxel gray-scale values ​​and unit voxel dose values ​​of the reference image on the gray-dose joint profile corresponding to the reference image; read the voxel gray-scale values ​​and unit voxel dose values ​​of the floating image on the gray-dose joint profile corresponding to the floating image, forming gray-scale profile sequences and dose profile sequences respectively; perform one-dimensional wavelet modulus maxima detection on the gray-scale profile sequences to obtain a set of gray-scale phase peaks; perform one-dimensional second-order difference zero-crossing detection on the dose profile sequences to obtain a set of dose phase peaks; convert each gray-scale phase peak and each dose phase peak into a normal distance value relative to the marked surface.

[0011] Further, the specific steps for performing dynamic time warping matching on grayscale phase peaks and dose phase peaks to generate gray-phase misalignment data containing dose-locked grayscale peaks, free grayscale peaks, and artifact traction cost values ​​are as follows: Perform dynamic time warping matching on the set of grayscale phase peaks and the set of dose phase peaks in the same gray-phase joint profile. Use the normal distance sequence of the grayscale phase peaks as the first matching sequence and the normal distance sequence of the dose phase peaks as the second matching sequence. Calculate the minimum matching path and generate matching pairs between grayscale phase peaks and dose phase peaks based on the minimum matching path. Calculate the normal distance difference between the grayscale phase peak and the dose phase peak in each matching pair. Mark the grayscale phase peak corresponding to the same dose phase peak with the smallest normal distance difference as the dose-locked grayscale peak. Calculate the normal distance between each remaining grayscale phase peak and the nearest dose phase peak. Mark the remaining grayscale phase peaks whose normal distance is greater than the phase distance threshold and whose grayscale change amplitude is greater than the median grayscale change amplitude of all grayscale phase peaks in the same gray-phase joint profile. For each free grayscale peak, the shortest distance from the combined grayscale profile to the center coordinates of the metal marker is calculated. The shortest distance and the influence radius of the metal marker are input into a Gaussian attenuation function to obtain the metal perturbation attenuation coefficient, which decreases as the shortest distance increases. The influence radius of the metal marker is determined by the distance from the outer edge of the metal artifact in the historical silicone implantation area image to the center coordinates of the metal marker. For each free grayscale peak, the normal distance difference between the free grayscale peak and the dose-locked grayscale peak is calculated. The normal distance difference, the grayscale change amplitude of the free grayscale peak, and the metal perturbation attenuation coefficient are multiplied to obtain the artifact traction cost. For each dose-locked grayscale peak, the normal distance difference between the dose-locked grayscale peak and the corresponding dose phase peak is calculated to obtain the dose-locking deviation value. The surface sampling point sequence and outer normal vector, the combined grayscale profile, the dose phase peak, the dose-locked grayscale peak, the free grayscale peak, the dose-locking deviation value, and the artifact traction cost are bound according to the implantation registration unified coordinate system to generate grayscale phase misalignment data.

[0012] Further, the specific steps for establishing a boundary phase map based on gray agent phase misalignment data are as follows: Read the gray agent phase misalignment data, establish a boundary phase map according to the spatial connection relationship between surface sampling points, take each surface sampling point as a graph node, and take the connection relationship between adjacent surface sampling points as graph edges; set candidate labels for each graph node, including dose-locked gray peak label, free gray peak label, and marked surface backoff label; take the dose-locked deviation value as the unary cost of the dose-locked gray peak label, take the reciprocal normalized value of the sum of the artifact traction cost and the preset minimum constant as the unary cost of the free gray peak label, take the difference in normal distance from the marked surface sampling point to the dose phase peak within the same gray agent joint profile as the unary cost of the marked surface backoff label, and take the abrupt change in normal distance corresponding to the difference in labels of adjacent graph nodes as the binary smoothing cost.

[0013] Further, based on the dose-locked grayscale peak, free grayscale peak, and artifact traction cost, the specific steps for performing α-expanded graph cut optimization to generate the boundary confidence label map and gray agent phase decoupling data are as follows: Perform α-expanded graph cut optimization on the boundary phase map, selecting the candidate label combination with the minimum total cost from all graph nodes; when a graph node is assigned as a dose-locked grayscale peak label, the corresponding dose-locked grayscale peak coordinates are used as the confidence boundary point; when a graph node is assigned as a free grayscale peak label, the artifact traction back-off distance is moved in the opposite direction along the outward normal vector of the surface, and the back-off coordinates are used as the confidence boundary point. The artifact traction back-off distance is equal to the normal distance difference between the free grayscale peak and the dose-locked grayscale peak within the same gray agent joint profile line; when a graph node is assigned as a label surface back-off label, the corresponding... The dose-phase peak coordinates on the gray-mask joint profile are used as reliable boundary points. All reliable boundary points are connected according to the spatial connection relationship between surface sampling points to generate reliable boundary phase surfaces for the reference image and floating image respectively. Graph nodes assigned as free gray-scale peak labels are marked as artifact traction nodes, graph nodes assigned as surface back-off labels are marked as gray-scale failure nodes, and graph nodes assigned as dose-locked gray-scale peak labels are marked as stable boundary nodes. A boundary reliable label map is generated based on the stable boundary nodes, artifact traction nodes, and gray-scale failure nodes. The reliable boundary phase surface of the reference image, the reliable boundary phase surface of the floating image, the boundary reliable label map, and the artifact traction back-off distance are bound according to the implantation registration unified coordinate system to generate gray-mask phase decoupling data.

[0014] Furthermore, based on gray agent phase misalignment data, gray agent phase decoupling data, and boundary credibility label map, the specific steps for constructing the gray agent label registration cost function according to stable boundary nodes, artifact traction nodes, and gray-scale failure nodes are as follows: Read the gray agent phase misalignment data and gray agent phase decoupling data; use the stable boundary nodes in the boundary credibility label map as phase-locking nodes, the artifact traction nodes as traction retraction nodes, and the gray-scale failure nodes as dose substitution nodes; establish the same-label correspondence between the reference image node and the floating image node according to the unified coordinate system for implantation registration and the surface sampling point sequence number; using the spatial transformation parameter as the variable to be solved, calculate the coordinate difference between the reference image node and the floating image node after transformation by the spatial transformation parameter for the phase-locking node, obtain the phase-locking registration residual, and use the sum of the dose-locking deviation value and the preset minimum constant... The reciprocal normalized value is used as the phase-locking residual weight. For the traction back-off node, the floating image node is moved backward along the surface outward normal vector by the artifact traction back-off distance. Then, the coordinate difference between the reference image node and the back-off floating image node after spatial transformation parameter transformation is calculated to obtain the traction back-off registration residual. The reciprocal normalized value of the sum of the artifact traction cost and the preset minimum constant is used as the traction back-off residual weight. For the dose substitution node, the coordinate difference between the dose phase peak of the reference image and the dose phase peak of the floating image after spatial transformation parameter transformation is calculated to obtain the dose substitution registration residual. The product of the phase-locking registration residual and the phase-locking residual weight, the product of the traction back-off registration residual and the traction back-off residual weight, and the product of the dose substitution registration residual and the dose substitution residual weight are normalized according to the number of nodes and summed to obtain the gray label registration cost function.

[0015] Further, the specific steps for solving the spatial transformation parameters and outputting the image registration results for the silicone implantation region are as follows: The Levonburg-Marquardt algorithm is used to iteratively solve the gray label registration cost function to obtain the spatial transformation parameters of the floating image relative to the reference image; during the iterative solution process, when the product of the traction backtracking registration residual and the traction backtracking residual weight is greater than the product of the phase-locked registration residual and the phase-locked residual weight, the traction backtracking residual weight is reduced; when the dose substitution registration residual is less than the traction backtracking registration residual, the dose substitution residual weight is multiplied by a weight enhancement coefficient to obtain the updated dose substitution residual weight; based on the updated traction backtracking residual weight and the updated dose substitution residual weight... The residual weights are iterated until the change in spatial transformation parameters obtained from two adjacent iterations is less than the transformation convergence threshold. The voxel coordinates, voxel gray values, dose voxel coordinates, and unit voxel dose values ​​of the floating image are read. Based on the spatial transformation parameters, the voxel coordinates and dose voxel coordinates of the floating image are mapped to the unified coordinate system for implantation registration. The voxel gray values ​​and unit voxel dose values ​​of the registered floating image are generated using a trilinear interpolation algorithm. The voxel gray values ​​of the reference image, the voxel gray values ​​of the registered floating image, the unit voxel dose values ​​of the registered image, the reliable boundary phase surface and the reliable boundary label map of the reference image and the floating image are bound together, and the silicone implantation area image registration result is output.

[0016] On the other hand, an image registration device for silicone implantation regions based on image dose fusion is provided. This system is applied to an image registration method for silicone implantation regions based on image dose fusion. The device includes: The registration data processing module is used to construct a unified coordinate system for implantation registration based on the reference image coordinate space, collect implantation registration data, and perform grayscale denoising, grayscale scale correction and spatial sampling alignment processing on the implantation registration data, and output the preprocessed implantation registration data. The gray agent phase recognition module is used to construct a reference image marker surface and a floating image marker surface based on the preprocessed implantation registration data. It generates gray agent joint profile lines along the outward normal of the reference image marker surface and the floating image marker surface, respectively. It extracts the gray-scale phase peak and the dose phase peak in the gray agent joint profile lines, and performs dynamic time warping matching on the gray-scale phase peak and the dose phase peak to generate gray agent phase misalignment data containing dose-locked gray-scale peaks, free gray-scale peaks and artifact traction cost values. The credible boundary optimization module is used to establish a boundary phase map based on gray agent phase misalignment data, and to perform α-expansion graph cut optimization based on dose-locked gray peaks, free gray peaks and artifact traction costs to generate a boundary credible label map and gray agent phase decoupling data. The phase-constrained registration module is used to construct a gray agent label registration cost function based on gray agent phase misalignment data, gray agent phase decoupling data, and boundary reliable label map, according to stable boundary nodes, artifact traction nodes, and gray-scale failure nodes, solve for spatial transformation parameters, and output the image registration result of the silicone implantation area.

[0017] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: (1) The present invention generates a gray-coated joint profile line along the outer normal of the reference image marker surface and the floating image marker surface, and extracts the gray phase peak and the dose phase peak respectively. The dose-locked gray peak and the free gray peak are identified by dynamic time warping matching, so that the gray abrupt changes formed by bright rings, dark bands and local signal loss are no longer directly used as the real boundary to participate in registration, thereby improving the positioning stability of the real boundary of the silicone implant.

[0018] (2) The present invention calculates the artifact traction cost based on the free gray peak, dose-locked gray peak and the center coordinate value of the metal marker point, and converts the artifact traction cost into the label selection cost in the boundary phase map, so that the abnormal gray peak caused by the metal marker point, material interface and imaging difference can be suppressed in the global optimization process, and the floating image is reduced by the artifact boundary.

[0019] (3) The present invention uses the boundary phase map to take the surface sampling points as graph nodes, the connection relationship between adjacent surface sampling points as graph edges, and uses α-extended graph cut optimization to select reliable boundary points, so that the dose-locked gray peak, free gray peak and marked surface back-off results are selected within the same optimization framework, avoiding boundary breakage or local jump caused by independent judgment point by point.

[0020] (4) The present invention constructs a gray agent label registration cost function based on the stable boundary node, artifact traction node and gray scale failure node in the boundary credibility label map, so that the stable boundary, artifact regression boundary and dose substitution boundary participate in the solution of spatial transformation parameters in different residual forms, thereby improving the spatial superposition consistency of the reference image, floating image and dose distribution around the silicone implantation area. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of an image registration method for silicone implantation areas based on image dose fusion; Figure 2This is a structural diagram of a silicone implantation area image registration device based on image dose fusion; Figure 3 This is a schematic diagram of the boundary phase map construction; Figure 4 This is a phase misalignment decoupling diagram of the gray material in the silicone implantation area. Detailed Implementation

[0023] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0024] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0025] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0026] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0027] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0028] This invention provides an image registration method for silicone implantation regions based on image dose fusion, such as... Figure 1 The flowchart shown is for an image registration method for silicone implantation areas based on image dose fusion. The processing flow of this method may include the following steps: S1 constructs a unified coordinate system for implantation registration using the reference image coordinate space, collects implantation registration data, and performs grayscale denoising, grayscale scale correction, and spatial sampling alignment processing on the implantation registration data, outputting the preprocessed implantation registration data. S2, based on the preprocessed implantation registration data, construct a reference image marker surface and a floating image marker surface, generate gray agent joint profile lines along the outward normal of the reference image marker surface and the floating image marker surface respectively, extract the gray-scale phase peak and the dose phase peak in the gray agent joint profile lines respectively, and perform dynamic time warping matching on the gray-scale phase peak and the dose phase peak to generate gray agent phase misalignment data containing dose-locked gray-scale peak, free gray-scale peak and artifact traction cost value; S3. Based on the phase misalignment data of the gray agent, a boundary phase map is established, and α-expansion map cut optimization is performed according to the dose-locked gray peak, free gray peak and artifact traction cost to generate a boundary reliable label map and gray agent phase decoupling data. S4. Based on gray agent phase misalignment data, gray agent phase decoupling data and boundary reliable label map, a gray agent label registration cost function is constructed according to stable boundary nodes, artifact traction nodes and gray scale failure nodes. The spatial transformation parameters are solved and the image registration results of the silicone implantation area are output.

[0029] Optionally, the specific steps for constructing a unified coordinate system for implantation registration using the reference image coordinate space, collecting implantation registration data, and performing grayscale denoising, grayscale scale correction, and spatial sampling alignment on the implantation registration data are as follows: Construct a unified coordinate system for implantation registration using the reference image coordinate space, setting the center point of the voxel at the lower left corner of the first slice of the reference image as the origin, with the X-axis along the increasing direction of the column number of the reference image, the Y-axis along the increasing direction of the row number of the reference image, and the Z-axis along the increasing direction of the slice number of the reference image; read the reference image image position parameters, reference image image orientation parameters, reference image planar voxel spacing values, and reference image slice thickness values ​​from the reference image file header information, and based on the row direction cosine vector and column direction cosine vector in the reference image image orientation parameters... Calculate the reference image tomographic direction vector, and form the reference image orientation matrix by combining the row direction cosine vector, column direction cosine vector, and tomographic direction vector. Read the floating image position parameters, floating image orientation parameters, floating image planar voxel spacing values, and floating image slice thickness values ​​from the floating image file header information. Calculate the floating image tomographic direction vector based on the row and column direction cosine vectors in the floating image orientation parameters, and form the floating image orientation matrix by combining the row direction cosine vector, column direction cosine vector, and tomographic direction vector. Read the dose distribution start position parameters, dose distribution orientation parameters, dose voxel spacing values, and dose voxel slice thickness values ​​from the dose distribution file header information. Calculate the floating image tomographic direction vector based on the row direction cosine vector in the dose distribution orientation parameters. The dose distribution tomographic direction vector is calculated using the column direction cosine vector, and the dose distribution direction matrix is ​​formed by the row direction cosine vector, column direction cosine vector, and tomographic direction vector. The column, row, and tomographic indices of the reference image voxels are used as reference image voxel index coordinates. These coordinates are then multiplied by the reference image planar voxel spacing and slice thickness, respectively, and rotated using the reference image direction matrix and superimposed with the reference image position parameters to obtain the DICOM patient coordinates corresponding to the reference image voxels. This establishes a one-to-one conversion relationship between the reference image voxel index coordinates and the DICOM patient coordinates. The column, row, and tomographic indices of the floating image voxels are used as floating image voxel index coordinates. The voxel index coordinates are multiplied by the voxel spacing and slice thickness of the floating image plane, respectively, and then rotated by the floating image orientation matrix and superimposed with the floating image position parameters to obtain the DICOM patient coordinates corresponding to the voxels. This establishes a one-to-one conversion relationship between the voxel index coordinates and the DICOM patient coordinates. The column number, row number, and tomographic number of the dose voxel are used as the dose voxel index coordinates. The dose voxel index coordinates are multiplied by the voxel spacing and slice thickness, respectively, and then rotated by the dose distribution orientation matrix and superimposed with the dose distribution start position parameters to obtain the DICOM patient coordinates corresponding to the voxels. This establishes a one-to-one conversion relationship between the dose voxel index coordinates and the DICOM patient coordinates.Using the DICOM patient coordinates corresponding to the center point of the voxel at the lower left corner of the first slice of the reference image as the unified coordinate origin, any DICOM patient coordinate is subtracted from the unified coordinate origin and then projected onto the X, Y, and Z axes corresponding to the orientation matrix of the reference image to obtain the three-dimensional coordinates in the unified coordinate system for implantation registration. This ensures that the reference image voxel coordinates, floating image voxel coordinates, and dose voxel coordinates are all expressed as three-dimensional coordinates in mm units. The reference image voxel coordinates are used as the reference coordinates for subsequent surface sampling, gray-dose combined profile generation, and spatial transformation parameter solving. Implantation registration data for the silicone implantation area is collected, specifically by voxel-by-voxel from the reference image voxel matrix. The reference image voxel grayscale values ​​are read, and the reference image voxel coordinate values ​​are obtained according to the transformation order between the reference image voxel index coordinates, DICOM patient coordinates, and the implantation registration unified coordinate system. The floating image voxel grayscale values ​​are read voxel by voxel from the floating image voxel matrix, and the floating image voxel coordinate values ​​are obtained according to the transformation order between the floating image voxel index coordinates, DICOM patient coordinates, and the implantation registration unified coordinate system. The unit voxel dose values ​​are read voxel by voxel from the three-dimensional dose matrix, and the dose voxel coordinate values ​​are obtained according to the transformation order between the dose voxel index coordinates, DICOM patient coordinates, and the implantation registration unified coordinate system. The reference image and the floating image... In the image, the coordinates of boundary points are read along the visible boundary of the silicone implant. These boundary point coordinates are formed from at least one of manually labeled point coordinates and automatically segmented boundary point coordinates. The read manually labeled point coordinates and automatically segmented boundary point coordinates are then converted according to the conversion order between the voxel index coordinates of the corresponding image, the DICOM patient coordinates, and the implant registration unified coordinate system to obtain the coordinate values ​​of the marker points on the implant surface. In the reference image and the floating image, the high grayscale threshold is determined by the grayscale distribution of metal marker points in the historical silicone implantation area image. Specifically, the 5th percentile value of the voxel grayscale value within the manually labeled area of ​​the metal marker points is taken, and all voxels within the high grayscale connected area are read. The voxel coordinate values ​​are calculated, and the arithmetic mean of all voxel coordinate values ​​is performed to obtain the center coordinate value of the metal marker point. The reference image voxel gray value, floating image voxel gray value, reference image voxel coordinate value, floating image voxel coordinate value, dose voxel coordinate value, implant surface marker point coordinate value, metal marker point center coordinate value, reference image planar voxel spacing value, floating image planar voxel spacing value, reference image layer thickness value, floating image layer thickness value, unit voxel dose value, dose voxel spacing value, dose voxel layer thickness value, reference image orientation matrix, floating image orientation matrix, and dose distribution orientation matrix are bound according to the unified coordinate system for implantation registration to form implantation registration data.Among them, the reference image orientation matrix, the floating image orientation matrix, and the dose distribution orientation matrix are derived from the image orientation parameters and tomographic orientation vectors in the corresponding file header information, respectively. The reference image position parameters, the floating image position parameters, and the dose distribution start position parameters are used to determine the spatial starting point of various data in the DICOM patient coordinate system. The reference image plane voxel spacing value, the floating image plane voxel spacing value, the reference image slice thickness value, the floating image slice thickness value, the dose voxel spacing value, and the dose voxel slice thickness value are used to determine the actual sampling scale of various voxels in the DICOM patient coordinate system. The implantation registration unified coordinate system is used to unify the reference image, the floating image, and the dose distribution to the same millimeter coordinate reference. For the acquired implantation registration data, a nonlocal mean filtering algorithm was used to perform voxel gray-level noise suppression and boundary structure preservation. Specifically, voxel blocks similar to the current voxel's gray-level neighborhood were searched within the 3D neighborhood, and the gray-level values ​​of the reference image voxels and the floating image voxels were weighted and averaged according to the voxel block similarity to smooth out random noise while preserving the gray-level abrupt changes at the boundary of the silicone implant. A histogram matching algorithm was used to perform cross-time gray-level scale correction processing on the image gray-level distribution in the implantation registration data. Specifically, the cumulative gray-level distribution of the reference image voxels was used as the target distribution, and the gray-level values ​​of the floating image voxels were mapped to the gray-level scale of the reference image to ensure that the gray-level distribution at different acquisition times was corrected. The grayscale range, grayscale peak position, and grayscale contrast relationship are kept consistent. A trilinear interpolation algorithm is used to unify the spatial sampling interval and align the coordinate sampling position of the implantation registration data. Specifically, the reference image voxel coordinates in the unified implantation registration coordinate system are used as the target sampling position. Eight adjacent voxels surrounding the target sampling position are determined in the three-dimensional voxel grid formed by the floating image voxel coordinates and the dose voxel coordinates. The interpolation weight is calculated based on the distance ratio from the target sampling position to the center point of the eight adjacent voxels. The grayscale value and unit voxel dose value of the floating image voxel are weighted and interpolated to obtain the grayscale value and dose value of the floating image consistent with the sampling position of the reference image. The preprocessed implantation registration data is then output.

[0030] In this implementation scheme, by unifying the reference image, floating image, and dose distribution into the same implantation registration coordinate system, and by making the grayscale information, coordinate information, dose information, coordinate values ​​of the implant surface markers, and center coordinate values ​​of the metal markers in the implantation registration data form a consistent data benchmark, it can provide a consistent data foundation in terms of scale, position, and grayscale comparability for subsequent construction of reference image marker surfaces and floating image marker surfaces, generation of gray-dose joint profile lines, and matching of grayscale phase peaks and dose phase peaks. This reduces phase misjudgment caused by differences in sampling interval, grayscale scale, and coordinate benchmark, and enables the true boundary of the silicone implantation area, the location of artifact interference, and the location of dose phase to be stably correlated within the same spatial frame.

[0031] Optionally, the specific steps for constructing a reference image marker surface and a floating image marker surface based on the preprocessed implant registration data, and generating gray-coated joint profile lines along the outward normal of the reference image marker surface and the floating image marker surface respectively, are as follows: Read the preprocessed implant registration data, group the coordinate values ​​of the implant surface marker points according to the source markers of the reference image and the floating image; the source marker is determined by the image source corresponding to the coordinate values ​​of the implant surface marker points, the coordinate values ​​of the implant surface marker points originating from the reference image are written into the reference image point set, the coordinate values ​​of the implant surface marker points originating from the floating image are written into the floating image point set, and duplicate marker points in the same image point set whose coordinate distance is less than the duplicate deletion threshold are deleted; the duplicate deletion threshold is one-tenth of the smaller value between the plane voxel spacing of the reference image and the plane voxel spacing of the floating image; through the source... The markers are grouped so that the reference image marker surface and the floating image marker surface are generated from the coordinate values ​​of the marker points on the implant surface in their respective image spaces, avoiding the mixing of boundary points at different acquisition times that could cause the surface shape to be prematurely affected by the registration results. For each group of implant surface marker point coordinate values, a 3D convex hull algorithm and Laplacian surface smoothing are sequentially applied to obtain the reference image marker surface and the floating image marker surface. The 3D convex hull algorithm is used to form the outer envelope of the bounding point set based on the coordinate values ​​of the implant surface marker points, transforming discrete marker points into closed surfaces with computable external normal vectors. Laplacian surface smoothing is used to reduce the discrete jumps of manually labeled points and the jagged jumps of automatic boundary points, ensuring the continuity of the external normal vector direction of adjacent surface sampling points. The preset smoothing step size of Laplacian surface smoothing ranges from 0.10 times to 0 times the average side length of adjacent vertices of the same marker surface.The smoothing factor is 30x, with a preset smoothing iteration count of 3 to 8. The preset smoothing step size and the preset smoothing iteration count are determined from historical silicone implantation area annotation images. Specifically, within the range of values, candidate smoothing step sizes and candidate smoothing iteration counts are traversed, and the candidate combination that minimizes the average distance between the marked surface and the boundary points of the artificially verified implant is selected as the preset smoothing step size and preset smoothing iteration count. Surface sampling points are generated on the corresponding marked surfaces according to the reference image planar voxel spacing value and the floating image planar voxel spacing value, respectively. The surface sampling step size of the reference image marked surface is taken from the reference image planar voxel spacing value, and the surface sampling step size of the floating image marked surface is taken from the floating image planar voxel spacing value. The surface sampling points are generated according to the circumferential and axial order on the marked surfaces. The sampling point sequence number ensures that the sampling points in the reference image marker surface and the floating image marker surface, corresponding in positional order, can be used for subsequent establishment of the label correspondence. The technical principle of generating surface sampling points based on the planar voxel spacing value is to ensure that the surface sampling density matches the actual spatial resolution of the image, preventing overly dense sampling from introducing repetitive grayscale responses and underly sparse sampling from missing local artifact abrupt change zones. The outer normal vector of each surface sampling point is calculated, and a preset normal sampling length is extended along the outer normal vector towards the inner and outer sides of the marker surface to generate a gray-to-gray joint profile line. The outer normal vector is determined by the normal direction of the local surface where the surface sampling point is located, and corrected to the outer direction by pointing from the geometric center of the marker surface to the surface sampling point. Gray-to-gray content is generated along the outer normal vector. The technical principle of the combined profile line lies in the fact that the outer boundary of the silicone implant, the gray-scale artifact abrupt change band, and the dose-phase transition band are mainly misaligned along the normal direction of the boundary. After projecting the gray-scale change position and the dose change position onto the same normal distance reference, the degree of misalignment between the gray-scale phase peak and the dose-phase peak can be directly compared. The preset normal sampling length is 3mm to 8mm, which is determined by the maximum width of the artifact abrupt change band and the maximum width of the dose-phase transition band in the historical silicone implantation area image. Specifically, the larger of the maximum width of the artifact abrupt change band and the maximum width of the dose-phase transition band is taken, and then a 1mm safety margin is added, with the final value limited to the range of 3mm to 8mm. The sampling interval of the gray-phase combined profile line is determined by the corresponding image. The sampling interval is determined by the smaller of the planar voxel spacing value and the corresponding image layer thickness value. Specifically, the sampling interval for the gray-to-analyte joint profile line corresponding to the reference image is the smaller of the planar voxel spacing value and the reference image layer thickness value. Similarly, the sampling interval for the gray-to-analyte joint profile line corresponding to the floating image is the smaller of the planar voxel spacing value and the floating image layer thickness value. Profile sampling points are generated sequentially from the inner endpoint of the gray-to-analyte joint profile line according to the sampling interval, up to the outer endpoint. The technical principle of using a smaller sampling interval is to ensure that the sampling density on the normal profile is not lower than the minimum spatial resolution of the image, guaranteeing that narrow gray-level abrupt changes and dose phase changes can be captured by the phase peak detection process, and ensuring that the gray-level phase peak and the dose phase peak have the same normal distance reference.

[0032] In this implementation, by converting the coordinate values ​​of the marker points on the implant surface into a reference image marker surface and a floating image marker surface, and forming a gray-dose joint profile line with surface sampling point numbers and a unified normal distance reference on the two types of marker surfaces, the gray-scale phase peak and the dose phase peak can be compared on the same boundary normal path. This provides stable spatial support for subsequent identification of dose-locked gray-scale peaks, free gray-scale peaks, and calculation of artifact traction cost, and reduces the boundary phase judgment deviation caused by surface marker point discrepancy error, surface sawtooth error, and sampling scale difference.

[0033] Optionally, the specific steps for extracting the grayscale phase peak and dose phase peak from the gray-dose combined profile line are as follows: Read the voxel grayscale value and unit voxel dose value of the reference image on the gray-dose combined profile line corresponding to the reference image; read the voxel grayscale value and unit voxel dose value of the floating image on the gray-dose combined profile line corresponding to the floating image, forming a grayscale profile sequence and a dose profile sequence respectively; specifically, according to the order of the gray-dose combined profile line from the inner endpoint to the outer endpoint of the marked surface, record the normal sampling sequence number and the signed normal distance value relative to the marked surface for each profile sampling point, wherein the sampling point located inside the marked surface... The profile sampling points correspond to negative normal distance values, the profile sampling points located outside the marked surface correspond to positive normal distance values, and the surface sampling points located on the marked surface correspond to zero normal distance values. When reading the voxel gray values ​​of the reference image, the voxel gray values ​​of the floating image, and the unit voxel dose values, the coordinates of the profile sampling points on the gray-dil combined profile line corresponding to the reference image are mapped to a three-dimensional grid composed of the voxel coordinate values ​​of the reference image, and trilinear interpolation is used to read the voxel gray values ​​of the reference image at the profile sampling points. The coordinates of the profile sampling points on the gray-dil combined profile line corresponding to the floating image are mapped to a three-dimensional grid composed of the voxel coordinate values ​​of the floating image. The system employs trilinear interpolation to read the voxel grayscale values ​​of the floating image at the profile sampling points. The dose distribution corresponding to the unit voxel dose value is bound to the reference image using the DICOM patient coordinates as a common spatial reference. The coordinates of the profile sampling points on the gray-dose joint profile lines corresponding to the reference image and the gray-dose joint profile lines corresponding to the floating image are mapped to a three-dimensional grid composed of dose voxel coordinate values. Trilinear interpolation is then used to read the unit voxel dose value at the profile sampling points, forming a dose profile sequence on the reference image side and a dose profile sequence on the floating image side, respectively. This ensures that each grayscale profile sequence and its corresponding dose profile sequence have the same normal sampling. Sequence number; Perform one-dimensional wavelet modulus maxima detection on the grayscale profile sequence to obtain a set of grayscale phase peaks; Specifically, use cubic B-spline wavelets to perform one-dimensional continuous wavelet transform on the grayscale profile sequence, calculate the wavelet coefficient modulus at each normal sampling sequence number, and mark the sampling points where the wavelet coefficient modulus is greater than the grayscale peak detection threshold and achieves a local maximum within the range of adjacent sampling sequences as grayscale phase peaks; The grayscale peak detection threshold is taken as 1.5 to 3.0 times the median of the wavelet coefficient modulus of the same grayscale profile sequence, and the grayscale peak detection threshold is determined by historical silicone implantation area images, specifically within the range of 1.5 to 3.0 times as 0.Candidate multiples are iterated at one-time intervals, and the candidate multiple that results in the highest detection rate of gray-level abrupt change positions during manual verification is selected as the gray-level peak detection threshold multiple. The technical principle of one-dimensional wavelet mode maxima detection is that gray-level abrupt changes formed at the interface of silicone material material manifest as local high gradient changes on the normal profile. Wavelet mode maxima can enhance the abrupt change position and suppress slow gray-level fluctuations, so that the gray-level phase peak corresponds to the normal position where the gray-level change is most concentrated. One-dimensional second-order difference zero-crossing detection is performed on the dose profile sequence to obtain the dose phase peak set. Specifically, the first-order difference and second-order difference of the dose profile sequence are calculated according to the normal sampling sequence number. The sampling position where the sign of the second-order difference switches between positive and negative is determined as the candidate zero-crossing position. The absolute value of the first-order difference is read on both sides of the candidate zero-crossing position. When the absolute value of the first-order difference is greater than the dose phase detection threshold, the candidate zero-crossing position is marked as the dose phase peak. The dose phase detection threshold is taken as 1.2 to 2.5 times the median of the absolute value of the first-order difference of the same dose profile sequence. The dose phase detection threshold is determined by historical dose distribution data. Specifically, the candidate multiples are iterated within a range of 1.2 to 2.5 times at 0.1 times intervals, and the candidate multiple that minimizes the deviation of the center position of the dose transition zone during manual verification is selected as the dose phase detection threshold multiple. The technical principle of one-dimensional second-order differential zero-crossing detection is that the dose distribution in the vicinity of the implant boundary usually exhibits a continuous transition, and the position where the dose change rate reaches its peak corresponds to the second-order differential sign change position. Therefore, the dose phase peak can be located at the normal position where the dose gradient change is most significant. Each gray-level phase peak and each dose phase peak are converted into a normal distance value relative to the marked surface. Specifically, the normal sampling sequence number of the gray-level phase peak and the dose phase peak is read, and the signed distance from the corresponding profile sampling point to the surface sampling point is used as the normal distance value. This normal distance value is then bound to the surface sampling point number, the gray-dose joint profile line number, the gray-level change amplitude of the gray-level phase peak, and the dose change amplitude of the dose phase peak. This allows subsequent dynamic time warping matching to compare the positional offset between the gray-level phase peak and the dose phase peak under the same normal distance reference.

[0034] In this implementation scheme, by forming grayscale profile sequences and dose profile sequences on the same gray agent joint profile line, and uniformly converting the grayscale phase peak set and the dose phase peak set into normal distance values ​​relative to the marked surface, it is possible to form a comparable phase relationship between the grayscale abrupt change position and the dose gradient change position on the same normal scale. This provides a stable data foundation for subsequent dynamic time warping matching, dose-locked grayscale peak screening, and free grayscale peak identification, and reduces the phase peak positioning deviation caused by grayscale noise, dose gradual change, and inconsistent sampling sequence numbers.

[0035] Optionally, the specific steps for performing dynamic time warping matching on grayscale phase peaks and dose phase peaks to generate gray phase misalignment data containing dose-locked grayscale peaks, free grayscale peaks, and artifact traction cost values ​​are as follows: Perform dynamic time warping matching on the set of grayscale phase peaks and the set of dose phase peaks in the same gray-dose combined profile. Use the normal distance sequence of the grayscale phase peaks as the first matching sequence and the normal distance sequence of the dose phase peaks as the second matching sequence. Calculate the minimum matching path and generate matching pairs between grayscale phase peaks and dose phase peaks based on the minimum matching path. Specifically, first arrange the set of grayscale phase peaks and the set of dose phase peaks in ascending order of normal distance values. Then, use the absolute difference in normal distance between any grayscale phase peak and any dose phase peak as the single-step matching cost. Accumulate the single-step matching cost using the dynamic time warping algorithm and select the path with the minimum accumulated matching cost as the minimum matching path. The technical principle of using dynamic time warping matching is that the number of peaks at grayscale abrupt change positions and dose gradient change positions in the same gray-dose combined profile may be inconsistent. The dynamic time warping algorithm... It can maintain normal order constraints even when the number of peaks is inconsistent, avoiding the misclassification of outward-shifting artifact peaks as true boundary peaks by simple nearest neighbor matching; in each matching pair, the normal distance difference between the gray-level phase peak and the dose phase peak is calculated, and the gray-level phase peak with the smallest normal distance difference corresponding to the same dose phase peak is marked as the dose-locked gray-level peak; where the dose-locked gray-level peak represents the gray-level abrupt peak that is closest to the dose phase peak in normal position, and serves as a candidate position for the true boundary gray-level response; the relationship between each remaining gray-level phase peak and the nearest dose phase peak is calculated. The normal distance of the peak is used to identify the remaining gray-level phase peaks whose normal distance is greater than the phase distance threshold and whose gray-level change amplitude is greater than the median of the gray-level change amplitude of all gray-level phase peaks within the same gray agent joint profile line. These remaining gray-level phase peaks are marked as free gray-level peaks. The gray-level change amplitude is the larger of the absolute values ​​of the gray-level differences between adjacent sampling points on both sides of the normal sampling number of the gray-level phase peak. The phase distance threshold is set to 0.50 mm to 1.50 mm and is determined by historical silicone implantation area images. Specifically, it is set to 0 within the range of 0.50 mm to 1.50 mm.Candidate thresholds are iterated at 10mm intervals, and the candidate threshold that achieves the highest accuracy in manually verifying artifact peak identification is selected as the phase distance threshold. The technical principle of this determination method is that the gray-level phase peak corresponding to the true boundary is usually close to the dose phase peak, and the bright rings and dark bands formed by artifacts will create strong gray-level abrupt changes on the outside of the normal direction. Therefore, using both the normal distance and the gray-level change amplitude can distinguish between the true boundary response and the artifact boundary response. The shortest distance from each gray-dose joint profile line to the center coordinate value of the metal marker point is calculated, and the shortest distance and the influence radius of the metal marker point are input into the Gaussian decay. The metal disturbance attenuation coefficient, which decreases as the shortest distance increases, is obtained by subtracting the function. The influence radius of the metal marker is determined by the distance from the outer edge of the metal artifact in the historical silicone implantation area image to the center coordinates of the metal marker. Specifically, the gray-coated joint profile is discretized into a set of profile sampling points. The Euclidean distance from the center coordinates of the metal marker to each profile sampling point is calculated, and the minimum Euclidean distance is taken as the shortest distance. The shortest distance and the influence radius of the metal marker are substituted into the metal disturbance Gaussian attenuation function to obtain the metal disturbance attenuation coefficient. The metal disturbance Gaussian attenuation function is: ;in, This represents the metal disturbance attenuation coefficient corresponding to the i-th ash agent joint profile line. This represents the shortest distance from the i-th ash-coated profile line to the center coordinates of the metal marker point. The influence radius of the metal marker is denoted as 5mm to 15mm. This influence radius is determined by historical silicone implantation area images, which include at least 20 sets of manually reviewed images. Each set of historical silicone implantation area images contains the center coordinates of the metal marker, the outer edge coordinates of the metal artifact, the image acquisition device number, the scanning protocol marker, and the voxel spacing value. When determining the influence radius of the metal marker, historical silicone implantation area images with the same image acquisition device number and the same scanning protocol marker as the current registration task are prioritized. If the number of corresponding historical silicone implantation area images is less than 20 sets, images with the same image modality and a voxel spacing difference not exceeding 0 are supplemented. Images of historical silicone implantation areas (5mm thick) were collected. For each set of historical silicone implantation area images, the Euclidean distance from the center coordinates of the metal marker to the center coordinates of the metal artifact was calculated, starting from the center coordinates of the metal marker. The distance values ​​of the historical artifacts were then sorted in ascending order, and the 95th percentile value was taken as the influence radius of the candidate metal marker. When the influence radius of the candidate metal marker was less than 5mm, 5mm was taken as the influence radius of the metal marker. When the influence radius of the candidate metal marker was greater than 15mm, 15mm was taken as the influence radius of the metal marker. When the influence radius of the candidate metal marker was within the range of 5mm to 15mm, the influence radius of the candidate metal marker was taken as the influence radius of the metal marker. By filtering historical silicone implantation area images according to image acquisition device number, scanning protocol marker, and voxel spacing value, and using the 95th percentile value instead of the mean or maximum value to determine the influence radius of metal markers, the excessive amplification of parameters by individual extreme artifact samples can be reduced, while preserving most of the outer edge coverage of metal artifacts. This allows the influence radius of metal markers to adapt to differences in implantation location among different patients and imaging scale differences among different devices. The technical principle of the Gaussian attenuation function is that the local signal anomaly caused by metal markers decreases with increasing spatial distance. Therefore, free gray-level peaks near the center coordinates of metal markers have a higher risk of mis-trapping during registration. For each free gray-level peak, the same... The normal distance difference between the free grayscale peak and the dose-locked grayscale peak within a combined grayscale profile is calculated, and this difference is divided by the normalized upper limit of the phase distance normalization to obtain the normalized normal offset value. The normalized upper limit of the phase distance normalization is twice the phase distance threshold, and is set to 1 when the normalized normal offset value is greater than 1. The grayscale variation amplitude of the free grayscale peak is divided by the maximum grayscale variation amplitude of all grayscale phase peaks within the same combined grayscale profile to obtain the normalized grayscale abrupt change value, which is set to 1 when it is greater than 1. The normalized normal offset value, the normalized grayscale abrupt change value, and the metal perturbation attenuation coefficient are multiplied to obtain the artifact traction cost. The artifact traction cost is calculated using the following formula: ,in, This represents the artifact dragging cost corresponding to the j-th free grayscale peak in the i-th grayscale composite profile. This represents the normalized normal offset value. Represents the normalized grayscale abrupt change value. This represents the metal disturbance attenuation coefficient. , and All values ​​are dimensionless and range from 0 to 1. The value is dimensionless and ranges from 0 to 1. The larger the artifact traction cost, the greater the outward shift of the free grayscale peak relative to the dose-locked grayscale peak, the stronger the grayscale abrupt change, and the closer it is to the center coordinates of the metal marker. In subsequent boundary phase map optimization, it is necessary to reduce the probability that the free grayscale peak will be selected as a reliable boundary point. The technical principle of this product operation is that the normalized normal offset value is used to characterize the degree of deviation of the free grayscale peak from the dose-locked grayscale peak, the normalized grayscale abrupt change value is used to characterize the intensity of the grayscale abrupt change corresponding to the free grayscale peak, and the metal perturbation attenuation coefficient is used to characterize the spatial possibility of the free grayscale peak being affected by the metal marker. When all three take large values, the free grayscale peak is more in line with the characteristics of the outward shift artifact traction boundary. By performing the product operation under dimensionless conditions, the distortion of the artifact traction cost value caused by the difference in normal distance and grayscale change amplitude due to different unit scales can be avoided. For each dose-locked grayscale peak, the difference in normal distance between the dose-locked grayscale peak and the corresponding dose phase peak is calculated to obtain the dose-locking deviation value. This dose-locking deviation value characterizes the degree of residual misalignment between the dose-locked grayscale peak and the dose phase peak, and is subsequently used to construct the univariate cost of the dose-locked grayscale peak label. The surface sampling point number, external normal vector, gray-matter joint profile line, dose phase peak, dose-locked grayscale peak, free grayscale peak, dose-locking deviation value, and artifact traction cost are bound according to the unified coordinate system for implantation registration to generate gray-matter phase misalignment data. During binding, the surface sampling point number is used as an index, and the external normal vector, gray-matter joint profile line number, dose phase peak normal distance value, dose-locked grayscale peak normal distance value, free grayscale peak normal distance value, dose-locking deviation value, and artifact traction cost value corresponding to the same surface sampling point are written into the same data record, enabling the subsequent boundary phase map to directly read the candidate boundary position and candidate label cost on the same surface sampling point.

[0036] In this implementation scheme, by sequentially constraining and matching the grayscale phase peak set and the dose phase peak set within the same gray-matrix joint profile, and by uniformly binding the dose-locked grayscale peak, free grayscale peak, dose-locked deviation value, and artifact traction cost value into gray-matrix phase misalignment data, the true boundary grayscale response, artifact mutation response, and dose phase response can be distinguished within the same normal distance frame. This allows the subsequent boundary phase map to no longer solely rely on the grayscale mutation intensity when selecting reliable boundary points, thereby reducing the traction effect of outward bright rings, dark bands, and metallic perturbations on the registration of silicone implant boundaries and improving the stability of reliable boundary label determination and spatial transformation parameter solution.

[0037] Optionally, the specific steps for establishing a boundary phase map based on gray agent phase misalignment data are as follows: Figure 3As shown, gray phase misalignment data is read, and a boundary phase map is established according to the spatial connection relationship between surface sampling points. Each surface sampling point is used as a graph node, and the connection relationship between adjacent surface sampling points is used as a graph edge. The spatial connection relationship between surface sampling points is determined by the adjacency relationship of triangular patches on the reference image marker surface and the floating image marker surface. Two surface sampling points sharing the same triangular patch edge and having consecutive sampling point numbers are identified as adjacent surface sampling points, and graph edges are established between the corresponding two graph nodes to ensure the boundary phase map maintains spatial continuity on the implanted external surface. Candidate labels are set for each graph node, including dose-locked gray peak labels, free gray peak labels, and marker surface rollback labels. The dose-locked gray peak label indicates that the current graph node uses the dose-locked gray peak as a reliable boundary candidate position, while the free gray peak label indicates that the current graph node has an artifact traction risk and needs to be rolled back in subsequent optimizations. The process involves several steps. First, the marked surface backtracking label does not fix the reliable boundary point to the marked surface. Instead, it represents the backtracking from the grayscale phase peak candidate position to the dose phase peak position on the same gray-drug joint profile line. The dose-locking deviation value is divided by the phase distance threshold and truncated to the 0-1 range to obtain the univariate cost of the dose-locked grayscale peak label. Second, the reciprocal of the sum of the artifact traction cost and the preset minimum constant is taken, and the reciprocal values ​​for all graph nodes are subjected to maximum-minimum normalization to obtain the univariate cost of the free grayscale peak label. Third, the normal distance difference between the marked surface sampling point and the dose phase peak within the same gray-drug joint profile line is divided by the preset normal sampling length and truncated to the 0-1 range to obtain the univariate cost of the marked surface backtracking label. Fourth, the absolute value of the normal distance difference between two adjacent graph nodes at their corresponding reliable boundary candidate positions under the candidate label is divided by the preset normal sampling length and truncated to the 0-1 range to obtain the binary smoothing cost. The preset minimum constant ranges from 0.000001 to 0.0001, the preset minimum constant is determined by the smallest non-zero order of magnitude of the artifact traction cost value, specifically one percent of the smallest non-zero value of the artifact traction cost value in the historical silicone implantation area image; before reciprocal normalization, the artifact traction cost value of each image node is first added to the preset minimum constant, and then the reciprocal of the addition result is taken to obtain the inverse value of the free grayscale peak; the maximum and minimum inverse values ​​of the free grayscale peaks of all image nodes are read, and the minimum inverse value of the free grayscale peaks of the current image node is subtracted from the minimum inverse value. After the value is calculated, it is divided by the difference between the maximum and minimum reciprocal values ​​to obtain the unary cost of the free grayscale peak label in the interval of 0 to 1. When the maximum reciprocal value is equal to the minimum reciprocal value, the unary cost of the free grayscale peak label is recorded as 0. Since the free grayscale peak reciprocal value is smaller when the artifact traction cost is larger, the graph nodes with larger artifact traction costs are more likely to be assigned as free grayscale peak labels when the graph cut is minimized. The difference in normal distance from the sampling point of the marked surface to the dose phase peak is the normal distance of the dose phase peak within the same gray agent joint profile. The absolute difference between the distance value and the zero normal distance value, this univariate cost, is used to measure the displacement from the marked surface back to the dose phase peak when grayscale phase information fails. The normal distance abrupt change corresponding to the difference in labels between adjacent graph nodes is the absolute value of the normal distance difference between the candidate positions of the corresponding credible boundaries under the candidate labels of the candidate nodes. When the candidate label is a dose-locked grayscale peak label, the credible boundary candidate position is taken as the normal distance value of the dose-locked grayscale peak. When the candidate label is a free grayscale peak label, the credible boundary candidate position is taken as the normal distance value of the free grayscale peak minus the artifact-induced back-off distance. When the candidate label is a marked surface back-off label, the credible boundary candidate position is taken as the normal distance value of the dose phase peak. The normal distance abrupt change is divided by the preset normal sampling length and truncated to the interval of 0 to 1 to obtain the binary smoothing cost, so that the boundary phase map simultaneously constrains the boundary continuity between adjacent surface sampling points when selecting local credible boundary points, avoiding isolated jumps in the credible boundary phase surface caused by a single strong grayscale abrupt change peak.

[0038] In this implementation scheme, by converting surface sampling points into graph nodes in the boundary phase map, and incorporating dose lock deviation, artifact traction cost, dose phase peak normal distance, and normal distance abrupt change between adjacent graph nodes into the label cost constraint, the selection of reliable boundary points can be simultaneously constrained by gray phase consistency, artifact traction risk, and boundary space continuity. This reduces the local interference of a single abnormal gray peak on the boundary phase judgment, and makes the reliable label map of the boundary generated by subsequent α-expansion graph cut optimization more stably reflect the true boundary position of the silicone implant.

[0039] Optionally, the specific steps for generating the boundary confidence label map and gray phase decoupling data by performing α-expanded graph cut optimization based on the dose-locked gray peak, free gray peak, and artifact traction cost are as follows: Perform α-expanded graph cut optimization on the boundary phase map, selecting the candidate label combination with the minimum total cost from all graph nodes; specifically, using the dose-locked gray peak label, free gray peak label, and marker surface backoff label as expandable labels, and recording the target label in each iteration as the α label, determining whether each graph node should retain its current candidate label or switch to the α label, and calculating the total cost before and after the switch based on the univariate cost and the binary smoothing cost; when switching to the α label can reduce the total cost of the boundary phase map, update the candidate label of the corresponding graph node to the α label. The process continues until all candidate labels have been traversed and the total cost of the boundary phase graph no longer decreases, yielding the candidate label combination with the minimum total cost. The technical principle of α-extended graph cut optimization lies in placing the local candidate boundary selection of each graph node and the boundary continuity constraint between adjacent graph nodes into the same energy function, so that the reliable boundary point is not independently determined by a single gray-dose joint profile line, but jointly determined in the adjacent spatial relationship on the implanted external surface. When a graph node is assigned as a dose-locked gray-scale peak label, the corresponding dose-locked gray-scale peak coordinates are used as reliable boundary points. Among them, the dose-locked gray-scale peak coordinates are calculated from the dose-locked gray-scale peak normal distance value, the surface sampling point coordinates, and the external normal vector. Specifically, starting from the surface sampling point coordinates, the dose is moved along the external normal vector direction. The normal distance value of the locked grayscale peak is used to obtain the coordinates of the dose-locked grayscale peak. When a graph node is assigned as a free grayscale peak label, the artifact traction back-off distance is moved in the opposite direction of the surface outward normal vector, and the coordinates after back-off are used as the reliable boundary point. The artifact traction back-off distance is the non-negative difference between the normal distance value of the free grayscale peak and the normal distance value of the dose-locked grayscale peak within the same grayscale agent joint profile, and is truncated using the maximum back-off length. The maximum back-off length is twice the phase distance threshold. Specifically, the coordinates of the free grayscale peak are first calculated using the normal distance value of the free grayscale peak, the coordinates of the surface sampling point, and the outward normal vector. Then, the truncated artifact traction back-off distance is moved in the opposite direction of the outward normal vector so that the reliable boundary point returns from the outward grayscale abrupt change position to the vicinity of the dose-locked grayscale peak. When the normal distance of the free grayscale peak is not greater than the normal distance of the dose-locked grayscale peak, the artifact traction back-off distance is set to 0. The technical principle of this back-off processing is that the free grayscale peak is usually formed by bright rings, dark bands and metallic disturbances at the interface of silicone material. Directly using the free grayscale peak will cause the boundary to shift outward, and the reverse back-off can weaken the influence of artifact traction on the reliable boundary point. When the graph node is assigned as a marker surface back-off label, the dose phase peak coordinates on the corresponding gray-to-gray-coated joint profile line are used as reliable boundary points. The dose phase peak coordinates are calculated from the normal distance of the dose phase peak, the coordinates of the surface sampling point, and the external normal vector. Specifically, the dose phase peak coordinates are obtained by moving the normal distance of the dose phase peak along the direction of the external normal vector, starting from the coordinates of the surface sampling point.The technical principle of using dose phase peak coordinates is that when the gray-level phase peak is missing or severely interfered with by artifacts, the dose phase peak can still reflect the transition position of the dose gradient in the vicinity of the implant, thus providing alternative constraints for credible boundary points. All credible boundary points are connected according to the spatial connection relationship between surface sampling points, generating a credible boundary phase surface for the reference image and a credible boundary phase surface for the floating image. Specifically, the triangular patch connection relationship between surface sampling points is read from the reference image marker surface and the floating image marker surface, and the same triangular patch connection relationship is mapped to the credible boundary points to form a triangular mesh surface composed of credible boundary points. The triangular mesh surface corresponding to the reference image is the credible boundary phase surface of the reference image, and the triangular mesh surface corresponding to the floating image is the credible boundary phase surface of the floating image. Graph nodes assigned as free gray-level peak labels are marked as artifact traction nodes, graph nodes assigned as marker surface regression labels are marked as gray-level failure nodes, and graph nodes assigned as dose-locked gray-level peak labels are marked as stable boundary nodes. Based on the stable boundary nodes, Artifact-driven nodes and grayscale failure nodes generate boundary credibility label maps. Specifically, using the surface sampling point number as the graph node index, stable boundary nodes, artifact-driven nodes, and grayscale failure nodes are written into the credibility label field of their respective graph nodes. The coordinates of the credibility boundary points, the normal distance values ​​of the credibility boundary points, and the connection relationships of the graph nodes are also written into the boundary credibility label map, enabling the boundary credibility label map to simultaneously represent the location and origin of the credibility boundary points. The reference image credibility boundary phase surface, the floating image credibility boundary phase surface, the boundary credibility label map, and the artifact-driven back-off distance are bound according to the unified coordinate system for implantation registration, generating grayscale phase decoupling data. During binding, the surface sampling point number is used as the index, and the credibility boundary points in the reference image credibility boundary phase surface, the credibility boundary points in the floating image credibility boundary phase surface, the node labels in the boundary credibility label map, and the artifact-driven back-off distance are written into the same data record. This allows the subsequent grayscale label registration cost function to read the corresponding registration constraint data according to the stable boundary nodes, artifact-driven nodes, and grayscale failure nodes.

[0040] In this implementation scheme, the α-expanded graph cut optimization unifies the candidate locations of the reliable boundaries corresponding to the dose-locked grayscale peak, the free grayscale peak, and the dose phase peak into the boundary phase map for global selection. This enables the reliable boundary phase surface of the reference image, the reliable boundary phase surface of the floating image, and the reliable boundary label map to form a continuous and consistent boundary expression under the same unified coordinate system for implantation registration. This avoids the local artifact mutation peaks from individually determining the location of reliable boundary points, allows artifact-pulling nodes to be corrected by the artifact-pulling back distance, and enables grayscale failure nodes to maintain the continuity of boundary constraints through the dose phase peak. This provides registration constraint data with clear sources, stable locations, and distinguishable labels for the subsequent gray agent label registration cost function.

[0041] Table 1. Example Data Table of Ash Agent Phase Misalignment Decoupling

[0042] In this embodiment, the number of surface sampling points is 12, the normal distance unit is mm, and the normal distance sampling range of the gray-to-concentrate joint profile line is 0.60 mm to 2.90 mm. Table 1 is an example data table of gray-to-concentrate phase misalignment decoupling, used to illustrate the correspondence between gray-scale phase peak, dose phase peak, artifact traction cost, and reliable boundary point. Table 1 specifically lists the normal distance of the reference image gray-scale phase peak, the normal distance of the reference image dose phase peak, the normal distance of the floating image gray-scale phase peak, the normal distance of the floating image dose phase peak, the shortest distance of the metallic marker point, the dose lock deviation value, the artifact traction cost, the normal distance of the graph node label, and the reliable boundary point for the 12 surface sampling points. Among them, the normal distance of the floating image gray-level phase peak of surface sampling point 1 is 1.35 mm, the normal distance of the floating image dose phase peak is 1.18 mm, the artifact dragging cost is 0.08, the graph node label is stable boundary node, and the corresponding reliable boundary point normal distance is 1.35 mm; the normal distance of the floating image gray-level phase peak of surface sampling point 3 is 2.35 mm, the normal distance of the floating image dose phase peak is 1.24 mm, the artifact dragging cost is 0.76, the graph node label is artifact dragging node, and the corresponding reliable boundary point normal distance is 1.35 mm. The distance between the floating image gray-scale phase peak and the dose phase peak of surface sampling point 4 is 1.24 mm. The normal distance between these two points is 2.60 mm, and the normal distance is 1.29 mm. The artifact traction cost is 0.94, and the node is labeled as an artifact traction node. The corresponding reliable boundary point normal distance is 1.29 mm. No effective floating image gray-scale phase peak was detected at surface sampling point 6. The normal distance between the floating image dose phase peak and the peak is 1.10 mm. The node is labeled as a gray-scale failure node, and the corresponding reliable boundary point normal distance is 1.10 mm. As shown in Table 1, surface sampling points with high artifact traction cost do not directly use the outward-shifted gray-scale phase peak as reliable boundary points. Gray-scale failure nodes determine reliable boundary points through the dose phase peak, while stable boundary nodes maintain the gray-scale phase peak as a reliable boundary point.

[0043] like Figure 4 As shown, the horizontal axis represents the sampling point number on the surface, and the vertical axis represents the normal distance relative to the marked surface. The background color blocks represent the grayscale phase response intensity, and the dose phase response is represented by contour lines. In the figure, circular nodes represent stable boundary nodes, cross-shaped nodes represent artifact traction nodes, square nodes represent grayscale failure nodes, and triangles represent the grayscale phase peak and dose phase peak of the floating image. Trustworthy boundary phase surface intercepts connect all trustworthy boundary points. Figure 4It is evident that high grayscale phase response values ​​do not always correspond to reliable boundary positions, and there is a visual misalignment relationship between the outward-shifted grayscale phase peak, the dose phase peak, and reliable boundary points. The reliable boundary label map distinguishes local grayscale abrupt change locations into stable boundary nodes, artifact-induced nodes, and grayscale failure nodes, and displays the corrected continuous boundary through the reliable boundary phase surface intercept. Figure 4 This invention illustrates how the process of generating grayscale artifact boundary traction and grayscale failure location is achieved through gray-element joint profile lines, grayscale phase peak and dose phase peak matching, and boundary credibility label map generation.

[0044] Optionally, based on gray agent phase misalignment data, gray agent phase decoupling data, and boundary credibility label map, the specific steps for constructing the gray agent label registration cost function according to stable boundary nodes, artifact traction nodes, and gray-scale failure nodes are as follows: Read the gray agent phase misalignment data and gray agent phase decoupling data; use the stable boundary nodes in the boundary credibility label map as phase-locking nodes, the artifact traction nodes as traction-back nodes, and the gray-scale failure nodes as dose replacement nodes; establish the same label correspondence between the reference image nodes and the floating image nodes according to the unified coordinate system for implantation registration and the surface sampling point number; specifically, using the surface sampling point number as an index, on the reliable boundary phase surface of the reference image and the floating... In the image's reliable boundary phase surface, reliable boundary points with the same reliable label are read. When the reliable labels of the reference image node and the floating image node are consistent, and their surface sampling point numbers are also consistent, the reference image node and the floating image node are written into the same label correspondence relationship. The technical principle of the same label correspondence relationship is that it ensures that the spatial transformation parameter solution establishes constraints only between nodes with the same boundary source, avoiding direct correspondence between stable boundary nodes and artifact-driven nodes, which would lead to confusion in the meaning of the residuals. Using the spatial transformation parameters as the variables to be solved, the coordinate difference between the reference image node and the floating image node transformed by the spatial transformation parameters is calculated for the phase-locked node, obtaining the phase-locked registration residual, and dose-locked. The reciprocal normalized value of the sum of the dose-locking deviation and the preset minimum constant is used as the phase-locking residual weight. The spatial transformation parameters include 3D translation, 3D rotation, and scale parameters. The scale parameter ranges from 0.95 to 1.05. The preset minimum constant ranges from 0.000001 to 0.0001, and is determined by the smallest non-zero order of magnitude of the dose-locking deviation, specifically one-hundredth of the smallest non-zero dose-locking deviation in historical silicone implantation area images. The phase-locking registration residual is a 3D Euclidean distance residual, used to constrain the stable boundary position where the grayscale phase peak and the dose phase peak are aligned. The smaller the dose-locking deviation, the larger the phase-locking residual weight, ensuring a stable boundary. Nodes are subject to stronger constraints in solving the spatial transformation parameters. For traction-back nodes, the floating image node is moved backward along the surface outward normal vector by the artifact traction-back distance. Then, the coordinate difference between the reference image node and the retraceable floating image node after spatial transformation is calculated to obtain the traction-back registration residual. The reciprocal normalized value of the sum of the artifact traction cost and the preset minimum constant is used as the weight of the traction-back residual. The retraceable floating image node is obtained by subtracting the product of the floating image node's corresponding outward normal vector and the artifact traction-back distance from the floating image node's coordinates. The traction-back registration residual is a three-dimensional Euclidean distance residual. The preset minimum constant ranges from 0.000001 to 0.0001, the preset minimum constant is determined by the smallest non-zero order of magnitude of the artifact traction cost value, specifically one percent of the smallest non-zero value of the artifact traction cost value in historical silicone implantation area images; the technical principle of traction back-registration residuals is that the outward shift of grayscale abrupt change corresponding to the artifact traction node is not directly used as a floating image node for registration, but is first backed back along the outward normal vector to the vicinity of the dose-locked grayscale peak, and then participates in the solution of spatial transformation parameters. The larger the artifact traction cost value, the smaller the weight of the traction back-registration residual, thus reducing the pulling effect of strong artifact positions on spatial transformation parameters; for dose substitution nodes, the dose phase peak of the reference image and the value transformed by spatial transformation parameters are calculated. The coordinate difference between the dose phase peaks of the subsequent floating image is used to obtain the dose substitution registration residual. This residual is a three-dimensional Euclidean distance residual, used to replace the gray-level phase peak at gray-level failure nodes in the spatial transformation parameter solution, preventing boundary constraint interruption due to local gray-level loss. The product of the phase-locked registration residual and its weight, the product of the traction-back registration residual and its weight, and the product of the dose substitution registration residual and its weight are normalized according to the number of nodes and summed to obtain the gray label registration cost function. Specifically, this involves counting the number of phase-locked nodes, the number of traction-back nodes, and... The dose substitution value is calculated by summing the products of the phase-lock registration residuals and their weights for all phase-locked nodes, and then dividing by the number of phase-locked nodes. Similarly, the normalized dose substitution value is calculated by summing the products of the traction-back registration residuals and their weights for all traction-back nodes, and then dividing by the number of traction-back nodes. Likewise, the normalized dose substitution value is calculated by summing the products of the dose substitution registration residuals and their weights for all dose substitution nodes, and then dividing by the number of dose substitution nodes. Finally, the normalized dose substitution value is calculated by combining the normalized dose substitution value, the normalized dose substitution value, and the dose substitution value. The normalized cost value is summed to obtain the gray agent label registration cost function. When the number of phase-locked nodes, traction-back nodes, or dose substitution nodes is zero, the corresponding normalized cost value is recorded as zero and does not participate in the weight update of that type of node. The technical principle of the gray agent label registration cost function is to assign different residual forms and different weights to nodes from different reliable sources based on the boundary reliable label map. This ensures that the stable boundary position, artifact back-back position, and dose substitution position can still jointly constrain the spatial transformation parameters of the floating image relative to the reference image even when the number of nodes is unbalanced, avoiding the spatial transformation parameter solution being biased towards a single boundary source due to an excessive number of nodes of a certain type.

[0045] In this implementation scheme, by incorporating stable boundary nodes, artifact traction nodes, and grayscale failure nodes into the grayscale label registration cost function, and by making the phase-locked registration residual, traction-back registration residual, and dose-substitution registration residual jointly constrain the spatial transformation parameters under the same implantation registration unified coordinate system, the registration process can simultaneously utilize stable boundary positions, artifact back-back positions, and dose-phase substitution positions. This reduces the transformation deviation caused by strong artifact nodes pulling on the floating image, avoids local constraint gaps caused by grayscale failure nodes, and ensures that the silicone implantation area image registration results maintain higher consistency in the spatial superposition relationship of the implant's true boundary, surrounding soft tissue position, and unit voxel dose value.

[0046] Optionally, the specific steps for solving the spatial transformation parameters and outputting the silicone implantation region image registration results are as follows: The Levenberg-Marquardt algorithm is used to iteratively solve the gray label registration cost function to obtain the spatial transformation parameters of the floating image relative to the reference image. These spatial transformation parameters include three-dimensional translation parameters, three-dimensional rotation parameters, and scale parameters. The three-dimensional rotation parameters are expressed in radians. The initial three-dimensional translation parameter is determined by the coordinate difference between the geometric center of the reliable boundary phase surface of the reference image and the geometric center of the reliable boundary phase surface of the floating image. The initial three-dimensional rotation parameter is set to 0, and the initial scale parameter is set to 1, so that the iterative solution starts from the center alignment state of the two reliable boundary phase surfaces. The Levenberg-Marquardt algorithm reads the gray material in each iteration. The label registration cost function is calculated using the Jacobian matrix of the spatial transformation parameters. The proportions of the Gaussian-Newton descent direction and the gradient descent direction are adjusted according to the damping factor. The damping factor is decreased when the gray agent label registration cost function decreases and increased when it does not, thus maintaining stability in the spatial transformation parameter solution even with nonlinear residual changes. During the iterative solution process, when the product of the traction-back registration residual and its weight is greater than the product of the phase-locked registration residual and its weight, the traction-back residual weight is multiplied by a weight attenuation coefficient to obtain the updated traction-back residual weight. The weight attenuation coefficient ranges from 0.60 to 0.90 and is determined by historical silicone implantation area images, specifically within the range of 0. Candidate weight attenuation coefficients are iterated at 0.05 intervals within the range of 0.60 to 0.90. The candidate weight attenuation coefficient that minimizes the average registration error of the manually verified implant boundary is selected as the weight attenuation coefficient, and the lower limit of the traction backtracking residual weight is 0.20 times the initial traction backtracking residual weight. When the dose substitution registration residual is less than the traction backtracking registration residual, the dose substitution residual weight is multiplied by the weight enhancement coefficient to obtain the updated dose substitution residual weight. The weight enhancement coefficient is between 1.10 and 1.50, and is determined by historical silicone implantation area images. Specifically, candidate weight enhancement coefficients are iterated at 0.05 intervals within the range of 1.10 to 1.50, and the candidate weight enhancement coefficient that minimizes the spatial superposition error of the unit voxel dose value is selected. The strong coefficient is used as the weight enhancement coefficient, and the upper limit of the dose substitution residual weight is 2.00 times the initial dose substitution residual weight. Based on the updated traction backtracking residual weight and the updated dose substitution residual weight, the gray label registration cost function is iteratively solved. The technical principle of this weight adjustment is that the boundary position corresponding to the traction backtracking node may still have residual artifact effects. When the proportion of traction backtracking registration residual is too large, the traction backtracking residual weight is reduced, which can reduce the pull of outward gray-level abrupt change on the spatial transformation parameters. When the dose substitution registration residual is small, the dose substitution residual weight is increased, which can use dose phase constraints to supplement the boundary registration basis at the gray-level failure position. This continues until the change in spatial transformation parameters obtained from two adjacent iterations is less than the transformation convergence threshold.The spatial transformation parameter changes are the square root of the normalized sum of the squares of the changes in 3D translation parameters, 3D rotation parameters, and scale parameters. The transformation convergence threshold is between 0.0001 and 0.001, determined by the voxel spacing of the reference image plane, the voxel spacing of the floating image plane, the layer thickness of the reference image, and the layer thickness of the floating image. Specifically, it is taken as one-thousandth to one-hundredth of the minimum value among these three values. The floating image voxel coordinates, voxel grayscale values, dose voxel coordinates, and unit voxel dose values ​​are read, and then the spatial transformation parameters are used to determine the convergence threshold. The method maps the floating image voxel coordinates and dose voxel coordinates to a unified coordinate system for implantation registration. Trilinear interpolation resampling is then performed on the floating image voxel grayscale values ​​and unit voxel dose values ​​to generate registered floating image voxel grayscale values ​​and registered unit voxel dose values. The registered unit voxel dose value is the dose voxel value obtained after spatial transformation and resampling of the existing unit voxel dose value; no recalculation of the unit voxel dose value is performed. Specifically, each reference image voxel coordinate value is used as the target sampling coordinate. Based on the inverse transformation of the spatial transformation parameters, the corresponding sampling coordinates are located in the three-dimensional grid formed by the floating image voxel coordinate values, and the bounding coordinates of the corresponding sampling coordinates are read. The grayscale values ​​of the eight floating image voxels are used as targets. Interpolation weights are calculated based on the distance ratios from the corresponding sampling coordinates to the center points of the eight floating image voxels to obtain the registered floating image voxel grayscale values. Simultaneously, the corresponding dose sampling coordinates are located in a 3D grid composed of dose voxel coordinate values ​​using the inverse transformation of spatial transformation parameters. The dose values ​​of the eight unit voxels surrounding the corresponding dose sampling coordinates are read. Interpolation weights are calculated based on the distance ratios from the corresponding dose sampling coordinates to the center points of the eight dose voxels. The eight unit voxel dose values ​​are then weighted and summed to obtain the registered unit voxel dose value. This process is a spatial resampling process of the dose volume in the implantation registration unified coordinate system. The registered unit voxel dose value... The voxel dose values ​​retain the physical meaning of the original dose distribution. The technical principle of the trilinear interpolation algorithm is that the voxel coordinate values ​​of the floating image and the dose voxel coordinate values, after being mapped by the spatial transformation parameters, usually fall at the non-integer coordinate positions of the voxel grid of the reference image. By weighting the eight neighboring voxels, the gray values ​​and dose values ​​corresponding to the voxel coordinate values ​​of the reference image can be obtained while maintaining spatial continuity. The reference image voxel gray values, the registered floating image voxel gray values, the registered unit voxel dose values, the reliable boundary phase surface of the reference image, the reliable boundary phase surface of the floating image after spatial transformation parameters, and the reliable boundary label map are bound together to output the image registration result of the silicone implantation area.During binding, the reference image voxel coordinates and surface sampling point numbers are used as indices. The grayscale values ​​of the reference image voxels corresponding to the same spatial location, the grayscale values ​​of the registered floating image voxels, the dose value per unit voxel after registration, the reliable boundary phase surface of the reference image, the reliable boundary phase surface of the floating image after spatial transformation parameters, and the reliable boundary label map are written into the same registration result record. This allows the output silicone implantation area image registration result to simultaneously express the grayscale registration relationship, dose superposition relationship, reliable boundary location, and reliable boundary label.

[0047] In this implementation scheme, by constraining the spatial transformation parameters with the gray agent label registration cost function, and binding the registered floating image voxel gray value, the registered unit voxel dose value, the reference image reliable boundary phase surface, the floating image reliable boundary phase surface transformed by the spatial transformation parameters, and the boundary reliable label map to the same registration result record, the image registration result of the silicone implantation area can simultaneously maintain the consistent expression of gray position, dose position, and reliable boundary position, reduce the interference of artifact traction nodes on the mapping of floating image voxel coordinate values, enhance the spatial superposition reliability of unit voxel dose values ​​at gray failure nodes, and thus improve the registration stability of the reference image and the floating image near the real boundary of the silicone implant.

[0048] like Figure 2 As shown, another aspect of the present invention provides an image registration apparatus for silicone implantation regions based on image dose fusion. This apparatus is applied to an image registration method for silicone implantation regions based on image dose fusion. The apparatus includes: The registration data processing module is used to construct a unified coordinate system for implantation registration based on the reference image coordinate space, collect implantation registration data, and perform grayscale denoising, grayscale scale correction and spatial sampling alignment processing on the implantation registration data, and output the preprocessed implantation registration data. The gray agent phase recognition module is used to construct a reference image marker surface and a floating image marker surface based on the preprocessed implantation registration data. It generates gray agent joint profile lines along the outward normal of the reference image marker surface and the floating image marker surface, respectively. It extracts the gray-scale phase peak and the dose phase peak in the gray agent joint profile lines, and performs dynamic time warping matching on the gray-scale phase peak and the dose phase peak to generate gray agent phase misalignment data containing dose-locked gray-scale peaks, free gray-scale peaks and artifact traction cost values. The credible boundary optimization module is used to establish a boundary phase map based on gray agent phase misalignment data, and to perform α-expansion graph cut optimization based on dose-locked gray peaks, free gray peaks and artifact traction costs to generate a boundary credible label map and gray agent phase decoupling data. The phase-constrained registration module is used to construct a gray agent label registration cost function based on gray agent phase misalignment data, gray agent phase decoupling data, and boundary reliable label map, according to stable boundary nodes, artifact traction nodes, and gray-scale failure nodes, solve for spatial transformation parameters, and output the image registration result of the silicone implantation area.

[0049] In this implementation scheme, by connecting the implantation registration data, gray agent phase misalignment data, gray agent phase decoupling data, and silicone implantation area image registration results in the processing order, the implantation registration unified coordinate system, gray agent joint profile line, boundary phase map, boundary confidence label map, and gray agent label registration cost function can form a continuous data transfer relationship. This allows gray-scale phase peaks, dose phase peaks, artifact traction nodes, gray-scale failure nodes, and stable boundary nodes in the silicone implantation area to be identified, screened, and used for registration constraints in a unified process. This improves the spatial correspondence consistency between the reference image, floating image, and unit voxel dose value, and reduces the interference of gray-scale artifact boundaries on the solution of spatial transformation parameters.

[0050] The following points need to be explained: (1) The accompanying drawings of the embodiments of the present invention only involve the structures involved in the embodiments of the present invention. Other structures can refer to the general design.

[0051] (2) For clarity, the thickness of layers or regions is enlarged or reduced in the drawings used to describe embodiments of the invention, i.e., these drawings are not drawn to scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being “above” or “below” another element, the element may be “directly” located “above” or “below” the other element or there may be intermediate elements.

[0052] (3) Where there is no conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

[0053] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for image registration of silicone implantation areas based on image dose fusion, characterized in that, The method includes: S1 constructs a unified coordinate system for implantation registration using the reference image coordinate space, collects implantation registration data, and performs grayscale denoising, grayscale scale correction, and spatial sampling alignment processing on the implantation registration data, outputting the preprocessed implantation registration data. S2, based on the preprocessed implantation registration data, construct a reference image marker surface and a floating image marker surface, generate gray agent joint profile lines along the outward normal of the reference image marker surface and the floating image marker surface respectively, extract the gray-scale phase peak and the dose phase peak in the gray agent joint profile lines respectively, and perform dynamic time warping matching on the gray-scale phase peak and the dose phase peak to generate gray agent phase misalignment data containing dose-locked gray-scale peak, free gray-scale peak and artifact traction cost value; S3. Based on the phase misalignment data of the gray agent, a boundary phase map is established, and α-expansion map cut optimization is performed according to the dose-locked gray peak, free gray peak and artifact traction cost to generate a boundary reliable label map and gray agent phase decoupling data. S4. Based on gray agent phase misalignment data, gray agent phase decoupling data and boundary reliable label map, a gray agent label registration cost function is constructed according to stable boundary nodes, artifact traction nodes and gray scale failure nodes. The spatial transformation parameters are solved and the image registration results of the silicone implantation area are output.

2. The method for image registration of silicone implantation areas based on image dose fusion according to claim 1, characterized in that, The specific steps for constructing a unified coordinate system for implantation registration using the reference image coordinate space, acquiring implantation registration data, and performing grayscale denoising, grayscale scale correction, and spatial sampling alignment on the implantation registration data are as follows: A unified coordinate system for implantation registration was constructed using the reference image coordinate space. The center point of the voxel at the lower left corner of the first slice of the reference image was set as the origin of the coordinate system. The X-axis was along the direction of increasing column number of the reference image, the Y-axis was along the direction of increasing row number of the reference image, and the Z-axis was along the direction of increasing slice number of the reference image. A one-to-one transformation relationship between the voxel coordinate values ​​of the reference image and the DICOM patient coordinates was established based on the voxel spacing value of the reference image plane, the slice thickness value of the reference image, and the direction matrix of the reference image. The floating image voxel coordinate values ​​and the dose voxel coordinate values ​​were mapped to the unified coordinate system for implantation registration according to the same transformation relationship. Implantation registration data for the silicone implantation area was collected. The implantation registration data included the reference image orientation matrix, the floating image orientation matrix, the dose distribution orientation matrix, the reference image voxel gray value, the floating image voxel gray value, the reference image voxel coordinate value, the floating image voxel coordinate value, the dose voxel coordinate value, the coordinate value of the implant surface marker point, the coordinate value of the center of the metal marker point, the reference image planar voxel spacing value, the floating image planar voxel spacing value, the reference image layer thickness value, the floating image layer thickness value, the unit voxel dose value, the dose voxel spacing value, and the dose voxel layer thickness value. For the acquired implantation registration data, a nonlocal mean filtering algorithm is used to perform voxel gray-level noise suppression and boundary structure preservation; a histogram matching algorithm is used to perform cross-time gray-level scale correction on the image gray-level distribution in the implantation registration data; and a trilinear interpolation algorithm is used to perform spatial sampling interval unification and coordinate sampling position alignment on the implantation registration data, outputting the preprocessed implantation registration data.

3. The method for image registration of silicone implantation areas based on image dose fusion according to claim 2, characterized in that, The specific steps for constructing reference image marker surfaces and floating image marker surfaces based on preprocessed implantation registration data, and generating gray agent joint profile lines along the outward normals of the reference image marker surfaces and floating image marker surfaces respectively, are as follows: The preprocessed implant registration data is read, and the coordinate values ​​of the implant surface markers are grouped according to the source markers of the reference image and the floating image. For each group of implant surface marker coordinate values, a 3D convex hull algorithm and Laplacian surface smoothing are sequentially applied to obtain the reference image marker surface and the floating image marker surface. Surface sampling points are generated on the corresponding marker surfaces according to the voxel spacing values ​​of the reference image and the floating image, respectively. The outer normal vector of each surface sampling point is calculated, and a preset normal sampling length is extended along the outer normal vector towards the inner and outer sides of the marker surface to generate a gray-solution joint profile line. The sampling interval of the gray-solution joint profile line is determined by the smaller value between the corresponding image voxel spacing value and the corresponding image layer thickness value.

4. The method for image registration of silicone implantation areas based on image dose fusion according to claim 3, characterized in that, The specific steps for extracting the grayscale phase peak and dose phase peak from the combined ash profile are as follows: The voxel grayscale values ​​and unit voxel dose values ​​of the reference image are read on the gray-drug joint profile line corresponding to the reference image, and the voxel grayscale values ​​and unit voxel dose values ​​of the floating image are read on the gray-drug joint profile line corresponding to the floating image, forming grayscale profile sequences and dose profile sequences respectively; one-dimensional wavelet mode maxima detection is performed on the grayscale profile sequence to obtain a set of grayscale phase peaks; one-dimensional second-order difference zero-crossing detection is performed on the dose profile sequence to obtain a set of dose phase peaks; each grayscale phase peak and each dose phase peak are converted into a normal distance value relative to the marked surface.

5. The method for image registration of silicone implantation areas based on image dose fusion according to claim 4, characterized in that, The specific steps for performing dynamic time warping matching on the grayscale phase peak and the dose phase peak to generate grayscale phase misalignment data containing dose-locked grayscale peaks, free grayscale peaks, and artifact traction cost values ​​are as follows: Dynamic time warping matching is performed on the set of gray-scale phase peaks and the set of dose phase peaks in the same ash agent joint profile. The normal distance sequence of the gray-scale phase peaks is used as the first matching sequence, and the normal distance sequence of the dose phase peaks is used as the second matching sequence. The minimum matching path is calculated, and matching pairs of gray-scale phase peaks and dose phase peaks are generated according to the minimum matching path. In each matching pair, the difference in normal distance between the grayscale phase peak and the dose phase peak is calculated. The grayscale phase peak that corresponds to the same dose phase peak and has the smallest difference in normal distance is marked as the dose-locked grayscale peak. The normal distance between each remaining grayscale phase peak and the nearest dose phase peak is calculated. The remaining grayscale phase peaks that have a normal distance greater than the phase distance threshold and a grayscale change amplitude greater than the median grayscale change amplitude of all grayscale phase peaks in the same grayscale phase peak joint profile are marked as free grayscale peaks. The shortest distance from each gray agent joint profile line to the center coordinates of the metal marker is calculated. The shortest distance and the influence radius of the metal marker are input into a Gaussian attenuation function to obtain the metal perturbation attenuation coefficient, which decreases as the shortest distance increases. The influence radius of the metal marker is determined by the distance from the outer edge of the metal artifact in the historical silicone implantation area image to the center coordinates of the metal marker. For each free grayscale peak, the normal distance difference between the free grayscale peak and the dose-locked grayscale peak is calculated. The normal distance difference, the grayscale change amplitude of the free grayscale peak, and the metal perturbation attenuation coefficient are multiplied to obtain the artifact traction cost. For each dose-locked grayscale peak, the normal distance difference between the dose-locked grayscale peak and the corresponding dose phase peak is calculated to obtain the dose-locking deviation value. The sampling point number of the curved surface, the external normal vector, the gray agent joint profile line, the dose phase peak, the dose-locked gray peak, the free gray peak, the dose-locked deviation value, and the artifact traction cost are bound according to the unified coordinate system of implantation registration to generate gray agent phase misalignment data.

6. The method for image registration of silicone implantation areas based on image dose fusion according to claim 5, characterized in that, The specific steps for establishing the boundary phase map based on the gray agent phase misalignment data are as follows: Read the phase misalignment data of the ash agent, establish a boundary phase map according to the spatial connection relationship between the surface sampling points, take each surface sampling point as a graph node, and take the connection relationship between adjacent surface sampling points as a graph edge; For each graph node, candidate labels are set, including dose-locked grayscale peak labels, free grayscale peak labels, and marker surface backoff labels; The dose-locking deviation value is used as the unary cost of the dose-locking grayscale peak label, the inverse normalized value of the sum of the artifact traction cost and the preset minimum constant is used as the unary cost of the free grayscale peak label, the difference in normal distance from the sampling point of the marked surface within the same gray agent joint profile to the dose phase peak is used as the unary cost of the marked surface backtracking label, and the abrupt change in normal distance corresponding to the difference in labels of adjacent graph nodes is used as the binary smoothing cost.

7. The method for image registration of silicone implantation areas based on image dose fusion according to claim 6, characterized in that, The specific steps for performing α-expansion graph cut optimization based on dose-locked grayscale peaks, free grayscale peaks, and artifact-induced cost to generate boundary confidence label maps and grayscale phase decoupling data are as follows: An α-expanded graph cut optimization is performed on the boundary phase map, and the candidate label combination with the minimum total cost is selected from all graph nodes. When a graph node is assigned as a dose-locked grayscale peak label, the corresponding dose-locked grayscale peak coordinates are used as a reliable boundary point. When a graph node is assigned as a free grayscale peak label, the artifact traction back-off distance is moved in the opposite direction along the outward normal vector of the surface, and the back-off coordinates are used as a reliable boundary point. The artifact traction back-off distance is equal to the difference in normal distance between the free grayscale peak and the dose-locked grayscale peak within the same grayscale agent joint profile. When a graph node is assigned as a label for a surface rollback, the dose phase peak coordinates on the corresponding ash-coated joint profile line are used as reliable boundary points. Connect all reliable boundary points according to the spatial connection relationship between the surface sampling points, and generate the reliable boundary phase surface of the reference image and the reliable boundary phase surface of the floating image respectively; mark the graph nodes that are assigned as free gray-scale peak labels as artifact traction nodes, mark the graph nodes that are assigned as surface regression labels as gray-scale failure nodes, and mark the graph nodes that are assigned as dose-locked gray-scale peak labels as stable boundary nodes. Generate a boundary credibility label map based on stable boundary nodes, artifact traction nodes, and grayscale failure nodes; The reference image reliable boundary phase surface, the floating image reliable boundary phase surface, the boundary reliable label map, and the artifact traction back-off distance are bound according to the unified coordinate system of implantation registration to generate gray agent phase decoupling data.

8. The method for image registration of silicone implantation areas based on image dose fusion according to claim 7, characterized in that, The specific steps for constructing the gray agent label registration cost function based on gray agent phase misalignment data, gray agent phase decoupling data, and boundary reliable label map, according to stable boundary nodes, artifact traction nodes, and gray-scale failure nodes, are as follows: Read gray agent phase misalignment data and gray agent phase decoupling data, take stable boundary nodes in the boundary credibility label map as phase locking nodes, take artifact traction nodes as traction back nodes, take gray scale failure nodes as dose replacement nodes, and establish the same label correspondence between reference image nodes and floating image nodes according to the implantation registration unified coordinate system and surface sampling point sequence number. Using the spatial transformation parameters as the variables to be solved, the coordinate difference between the reference image node and the floating image node transformed by the spatial transformation parameters is calculated for the phase-locked node to obtain the phase-locked registration residual. The normalized value of the inverse of the sum of the dose-locking deviation and the preset minimum constant is used as the weight of the phase-locked residual. For the traction-back node, the floating image node is moved backward along the out-of-plane normal vector of the surface by the artifact traction-back distance, and then the coordinate difference between the reference image node and the retraceable floating image node transformed by the spatial transformation parameters is calculated to obtain the traction-back registration residual. The reciprocal normalized value of the sum of the artifact traction cost and the preset minimum constant is used as the traction backoff residual weight. For dose substitution nodes, the coordinate difference between the dose phase peak of the reference image and the dose phase peak of the floating image after spatial transformation parameter transformation is calculated to obtain the dose substitution registration residual. The product of the phase-locked registration residual and the phase-locked residual weight, the product of the traction backoff registration residual and the traction backoff residual weight, and the product of the dose substitution registration residual and the dose substitution residual weight are normalized according to the number of nodes and summed to obtain the gray label registration cost function.

9. The method for image registration of silicone implantation areas based on image dose fusion according to claim 8, characterized in that, The specific steps for solving the spatial transformation parameters and outputting the image registration results of the silicone implantation region are as follows: The Levonburg-Marquardt algorithm is used to iteratively solve the gray label registration cost function to obtain the spatial transformation parameters of the floating image relative to the reference image. During the iterative solution process, when the product of the traction back-registration residual and the traction back-registration residual weight is greater than the product of the phase-locked registration residual and the phase-locked residual weight, the traction back-registration residual weight is reduced. When the dose substitution registration residual is less than the traction back-registration residual, the dose substitution residual weight is multiplied by the weight enhancement coefficient to obtain the updated dose substitution residual weight. The iteration continues based on the updated traction back-registration residual weight and the updated dose substitution residual weight until the change in spatial transformation parameters obtained from two adjacent iterations is less than the transformation convergence threshold. Read the floating image voxel coordinate values, floating image voxel gray values, dose voxel coordinate values ​​and unit voxel dose values. Based on the spatial transformation parameters, map the floating image voxel coordinate values ​​and dose voxel coordinate values ​​to the implantation registration unified coordinate system. Use the trilinear interpolation algorithm to generate the registered floating image voxel gray values ​​and registered unit voxel dose values. The reference image voxel gray values, the registered floating image voxel gray values, the registered unit voxel dose values, the reliable boundary phase planes and boundary reliable label maps of the reference image and the floating image are bound together to output the silicone implantation area image registration results.

10. An image registration device for silicone implantation regions based on image dose fusion, wherein the device applies the image registration method for silicone implantation regions based on image dose fusion as described in any one of claims 1-9, characterized in that, The device includes: The registration data processing module is used to construct a unified coordinate system for implantation registration based on the reference image coordinate space, collect implantation registration data, and perform grayscale denoising, grayscale scale correction and spatial sampling alignment processing on the implantation registration data, and output the preprocessed implantation registration data. The gray agent phase recognition module is used to construct a reference image marker surface and a floating image marker surface based on the preprocessed implantation registration data. It generates gray agent joint profile lines along the outward normal of the reference image marker surface and the floating image marker surface, respectively. It extracts the gray-scale phase peak and the dose phase peak in the gray agent joint profile lines, and performs dynamic time warping matching on the gray-scale phase peak and the dose phase peak to generate gray agent phase misalignment data containing dose-locked gray-scale peaks, free gray-scale peaks and artifact traction cost values. The credible boundary optimization module is used to establish a boundary phase map based on gray agent phase misalignment data, and to perform α-expansion graph cut optimization based on dose-locked gray peaks, free gray peaks and artifact traction costs to generate a boundary credible label map and gray agent phase decoupling data. The phase-constrained registration module is used to construct a gray agent label registration cost function based on gray agent phase misalignment data, gray agent phase decoupling data, and boundary reliable label map, according to stable boundary nodes, artifact traction nodes, and gray-scale failure nodes, solve for spatial transformation parameters, and output the image registration result of the silicone implantation area.

Images