A shock wave therapy instrument target three-dimensional reconstruction system based on voxel morphology

Abstract

The present application relates to the technical field of three-dimensional reconstruction, and discloses a target area three-dimensional reconstruction system of a shock wave therapeutic instrument based on voxel morphology, which comprises the following steps: generating resampling body data through resampling; separating a bone mask and a candidate calcification mask; extracting the morphological skeleton of a target calcification body; generating a weight field and calculating a segregation amount; generating a core target area; generating a final effective target area; and outputting the final effective target area, the target area volume, the target area center and an equivalent ellipsoid. The present application constructs a three-dimensional reconstruction process based on voxel morphology, separates bone and calcification tissue by using a thickness attribute open operation, combines the propagation characteristics of a shock wave in biological tissue with local anatomical structure, quantifies the action intensity difference of different regions inside a calcification body, extracts a target area range matched with the action distribution of a shock wave, and finally outputs a voxel-level target area and an equivalent geometric target area, so as to adapt to the calling requirements of a shock wave therapeutic instrument and provide three-dimensional data support more in line with the actual action distribution for treatment planning.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional reconstruction technology, and more specifically, to a three-dimensional reconstruction system for the target area of ​​a shockwave therapy device based on voxel morphology. Background Technology

[0002] Rotator cuff calcification is a common cause of shoulder pain and limited function, and is often treated conservatively with shockwave therapy. For this type of treatment, the location, extent, and depth of the treatment area directly affect pre-treatment planning and focus adjustment during treatment. Current practices largely rely on two-dimensional slice observation, empirical surface marking, real-time ultrasound observation, or overall localization of high-density deposition areas. The basic idea is to first locate the calcification foci, and then use the entire calcification or its geometric center as a treatment reference. This method is still applicable to regular, mass-like lesions, but its applicability is significantly limited in the rotator cuff attachment area, especially in the region near the greater tuberosity of the humerus.

[0003] This is because rotator cuff calcifications are often located close to the cortical bone and may exhibit lobulated, spiculated, thin-layered shell-like structures, fine bridge connections, and multi-branched extensions. Simultaneously, both bone tissue and calcifications appear at high gray levels in images, easily converging during thresholding. Relying solely on a single gray-level threshold, fixed-scale morphological preprocessing, or a maximum connected region selection strategy can easily result in the bone tissue, main calcification, and migration pathways being mixed into a single target. Furthermore, the treatment position and imaging position are not entirely consistent; changes in position further alter the spatial relationships between the bone surface, soft tissue, and calcifications, making it difficult for static slice coordinates and simple depth measurement methods to stably correspond to the actual target area before treatment.

[0004] Furthermore, existing technologies typically lack a three-dimensional reconstruction mechanism that can simultaneously consider unified spatial metrics, bone calcification separation, object thickness, bone proximity, and local bone surface morphology. Consequently, it is difficult to distinguish effective regions worthy of treatment planning from the entire calcified deposit. Especially when the voxels of the bone lateral surface, bone surface curvature, and calcified shell thickness all work together, relying solely on the geometric center, circumscribed range, or ordinary segmentation contours is insufficient to fully represent the truly planned three-dimensional target area. Therefore, it is necessary to propose a voxel-based morphology-based three-dimensional target area reconstruction system for shockwave therapy devices to generate an effective target area output for pre-treatment planning. Summary of the Invention

[0005] This invention provides a three-dimensional reconstruction system for the target area of ​​a shockwave therapy device based on voxel morphology, which solves the technical problems mentioned in the background art.

[0006] This invention provides a three-dimensional reconstruction system for the target area of ​​a shockwave therapy device based on voxel morphology, comprising: The first module performs resampling based on the same voxel side length of the original 3D volume data to generate resampled volume data; The second module extracts the mineralized joint mask from the resampled volume data, calculates the thick core scale according to the inner distance field, and uses the thick core scale to perform a thickness attribute opening operation to separate the bone mask and the candidate calcification mask. The third module decomposes the candidate calcification mask into connected domains, selects the target calcification body according to the volume of the connected domain and the average bone distance to the bone mask, and extracts the morphological skeleton of the target calcification body to obtain the median thickness. The fourth module calculates the depth of the voxel shell, bone distance, and positive curvature of the bone surface within the target calcified body. It then generates a weighted field by combining the median thickness and calculates the volume average value of the weighted field to obtain the segregation amount. The fifth module calculates the seed threshold based on the weight field to extract the seed set, determines the geodesic dilatation steps based on the median thickness and the amount of segregation, and performs geodesic dilatation on the seed set within the target calcification to generate the core target area. The sixth module calculates the closing radius based on the median thickness and the amount of segregation. It then uses the closing radius to perform a closing operation on the core target area to generate an intermediate target area. Finally, it performs morphological reconstruction on the intermediate target area using the target calcification as a constraint to generate the final effective target area. The seventh module uses the final effective target area to obtain the target volume and target center, extracts the equivalent ellipsoid based on the covariance matrix of the internal voxel coordinates and the target center, and outputs the final effective target area, target volume, target center and equivalent ellipsoid.

[0007] The beneficial effects of this invention are as follows: This invention unifies the length scale of all geometric operations through isovoxel resampling, avoiding three-dimensional morphological distortion caused by the anisotropy of the original voxels. It employs thickness attribute opening operations to separate bone and calcified tissue, solving the segmentation problem caused by the same grayscale value in the bone-adhering region. The scale parameters of all morphological operations are intrinsically determined by the characteristics of the target calcification itself, eliminating the need for manually preset thresholds and improving the processing adaptability of calcifications of different morphologies. By constructing curvature-weighted near-bone shell segregation features, it distinguishes the differences in shock wave intensity in different regions within the calcification, making the target area more closely match the actual shock wave action area. It uses a combination of constrained geodesic dilatation and constrained morphological reconstruction to repair the target area boundary while preventing the target area from intruding into normal tissue or being mistakenly included in migratory calcification pathways. The final output voxel target area, center of action, equivalent ellipsoid, and volume data can be directly integrated into the treatment planning process of shock wave therapy devices, reducing manual adjustments and providing a target area reference that better matches anatomical and physical characteristics for clinical treatment. Attached Figure Description

[0008] Figure 1 This is a calculation flowchart of a three-dimensional reconstruction system for the target area of ​​a shock wave therapy device based on voxel morphology, according to the present invention. Detailed Implementation

[0009] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0010] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" indicate that the element or object preceding the term encompasses the elements or objects listed following the term and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0011] like Figure 1 As shown, a three-dimensional reconstruction system for the target area of ​​a shockwave therapy device based on voxel morphology includes: The first module performs resampling based on the same voxel side length of the original 3D volume data to generate resampled volume data; The second module extracts the mineralized joint mask from the resampled volume data, calculates the thick core scale according to the inner distance field, and uses the thick core scale to perform a thickness attribute opening operation to separate the bone mask and the candidate calcification mask. The third module decomposes the candidate calcification mask into connected domains, selects the target calcification body according to the volume of the connected domain and the average bone distance to the bone mask, and extracts the morphological skeleton of the target calcification body to obtain the median thickness. The fourth module calculates the depth of the voxel shell, bone distance, and positive curvature of the bone surface within the target calcified body. It then generates a weighted field by combining the median thickness and calculates the volume average value of the weighted field to obtain the segregation amount. The fifth module calculates the seed threshold based on the weight field to extract the seed set, determines the geodesic dilatation steps based on the median thickness and the amount of segregation, and performs geodesic dilatation on the seed set within the target calcification to generate the core target area. The sixth module calculates the closing radius based on the median thickness and the amount of segregation. It then uses the closing radius to perform a closing operation on the core target area to generate an intermediate target area. Finally, it performs morphological reconstruction on the intermediate target area using the target calcification as a constraint to generate the final effective target area. The seventh module uses the final effective target area to obtain the target volume and target center, extracts the equivalent ellipsoid based on the covariance matrix of the internal voxel coordinates and the target center, and outputs the final effective target area, target volume, target center and equivalent ellipsoid.

[0012] In one embodiment of the present invention, resampling is performed according to the same voxel side length of the original three-dimensional volume data to generate resampled volume data, including: The formula for calculating the side length of a voxel is as follows: ; In the formula, For equal voxel side lengths, The voxel spacing in the first direction represents the original 3D volume data. The voxel spacing in the second direction represents the original 3D volume data. The voxel spacing in the third direction of the original 3D volume data; The formula for calculating the resampled volume data is as follows: ; In the formula, This represents the value of the resampled volume data at the voxel location. Voxel position, For resampling operators, This is the original three-dimensional volume data.

[0013] It should be noted that the original 3D volumetric data is a set of 3D grayscale volumetric data input to the system before treatment. It carries the spatial grayscale information of calcifications, bone tissue, and surrounding soft tissue. This data can be obtained through shoulder computed tomography (CT) scanning or by importing 3D volumetric data files output from an imaging workstation. The voxel spacing in the first direction is the sampling interval of the original 3D volumetric data in the first direction, used to determine the physical distance between the centers of adjacent voxels in that direction. It can be obtained by reading the spatial resolution field in the header of the original imaging file. The voxel spacing in the second direction is the sampling interval of the original 3D volumetric data in the second direction, used to determine the physical distance between the centers of adjacent voxels in that direction. It can be obtained by reading the spatial resolution field in the header of the original imaging file. The voxel spacing in the third direction is the sampling interval of the original 3D volumetric data in the third direction, usually corresponding to the sampling interval in the slice thickness direction or the scanning advance direction. It can be obtained by reading the slice spacing or slice thickness field in the header of the original imaging file. The isotropic voxel edge length is the target edge length used to uniformly convert the original anisotropic voxels into an isotropic voxel mesh, serving as a unified length benchmark for subsequent calculations of thickness, distance, volume, and structural element radius. Resampled volumetric data is standardized 3D volumetric data formed by reconstructing the original 3D volumetric data using equal-voxel methods. It serves as a unified input for subsequent thresholding, distance field calculation, skeleton extraction, and target reconstruction. Voxel positions are discrete position indices within the 3D volumetric data grid, used to mark the spatial location of a particular voxel in the 3D array. The resampling operator is a computational operator that maps the original 3D volumetric data onto a target regular grid using equal-voxel side lengths, used to generate resampled volumetric data based on the original grayscale values.

[0014] It should be noted that the specific implementation of interpolation resampling is as follows: Using the physical coordinate system of the original 3D volume data as a reference, while keeping the bounding box of the original volume data unchanged, a regular target mesh is established in 3D space with equal voxel side lengths. After back-mapping the center coordinates of each target voxel to the original data coordinates, trilinear interpolation is used to calculate the grayscale value. When the target point falls outside the boundary of the original volume data, the nearest boundary voxel copy method is used to avoid edge holes. The mapping method from original coordinates to resampling coordinates is as follows: First, the voxel spacing in three directions and the image origin information are read. Then, based on the physical coordinates of the target mesh voxel center, its continuous coordinate position in the original mesh is calculated. Subsequently, the eight nearest original voxels corresponding to this continuous coordinate position are used in the interpolation calculation, thus ensuring a one-to-one correspondence between the resampling result and the original physical space. The resampling boundary processing and grayscale fidelity strategy is as follows: For coordinates exceeding the range of the original volume data, boundary copying or mirror extension is used. The interpolated grayscale value retains the dynamic range allowed by the original data type, preventing abnormal negative values, oversaturation values, or edge breaks during the resampling process. The method for determining the origin and coverage area of ​​the resampling target grid is as follows: the smallest physical coordinate of the original data bounding box is used as the starting point of the target grid, and the largest physical coordinate of the original data bounding box is used as the ending point of the target grid. Then, the target grid size is generated by rounding up according to the equal voxel side length, ensuring that the data completely covers the original scan area after resampling.

[0015] Specifically, because the sampling intervals of the original 3D volume data are usually inconsistent in the three directions, directly calculating thickness, distance, and volume on the original mesh would introduce additional scaling errors in the layer thickness direction. Therefore, the cube root of the product of the voxel spacing in the three directions is used as the equal voxel side length to establish a unified length scale from the perspectives of preserving the overall volume and compromising the three-dimensional scale. For example, when the layer thickness is significantly greater than the intra-layer pixel spacing, this processing can prevent the target calcification from being falsely elongated in the layer thickness direction. In addition, subsequent thresholding, morphological opening, skeleton extraction, and geodesic dilatation all rely on regular voxel meshes, so interpolation resampling with equal voxel side lengths is required to ensure that each voxel has a consistent physical scale in the three directions. In this way, when a spherical structural element with a radius of 1 voxel is used for subsequent calculations, its physical meaning remains consistent in the three directions.

[0016] In one embodiment of the present invention, a mineralized joint mask is extracted from the resampled volume data, a thick core scale is calculated based on the inner distance field, and a thickness attribute opening operation is performed using the thick core scale to separate the bone mask and candidate calcification masks, including: The calculation formula for the combined mask for extracting mineralized bodies is as follows: ; ; In the formula, For mineralized composite mask, Voxel position, This represents the value of the resampled volume data at the voxel location. For the threshold, For Otsu's algorithm operator, For resampled volume data; The formula for calculating the interior distance field is as follows: ; In the formula, This represents the value of the interior distance field at the voxel location. For voxel locations that do not belong to the mineralized composite mask, Let be the three-dimensional spatial coordinates of the voxel position. The three-dimensional spatial coordinates of the voxel positions that do not belong to the mineralized composite mask. Euclidean distance; The formula for calculating the core thickness is as follows: ; In the formula, For the thick core scale, Extraction operator for the upper quartile; The formula for calculating the separation of the bone mask and the candidate calcification mask is as follows: ; ; In the formula, For bone masking, For the opening operator of the thickness attribute, As a candidate calcification mask, It is a set difference operator.

[0017] It should be noted that the threshold is the gray-level boundary value used to distinguish resampled volume data into mineralized and non-mineralized regions, and is used to generate the mineralized joint mask. The Otsu algorithm operator is a threshold selection operator that automatically calculates the foreground and background segmentation thresholds based on the gray-level histogram, aiming to ensure that the two types of gray levels after segmentation have good statistical distinguishability. The mineralized joint mask is a binary set formed by voxel positions in the resampled volume data whose gray-level values ​​are greater than or equal to the threshold, used to simultaneously represent bone tissue and high-density calcified regions. Voxel positions outside the mineralized joint mask refer to voxel positions that do not belong to the mineralized joint mask, used as the search objects for the shortest distance in the inner distance field. The three-dimensional spatial coordinates of the voxel positions are the coordinate expressions after mapping the discrete voxel positions to physical space, used to perform Euclidean distance calculations. The three-dimensional spatial coordinates of the voxel positions outside the mineralized joint mask are the coordinate expressions of the voxels outside the mineralized joint mask in physical space, used together with the internal voxel coordinates to form the endpoints for distance calculation. Euclidean distance is the straight-line distance between two points in three-dimensional space, used to define geometric quantities such as interior distance field, bone distance, shell depth, and local radius.

[0018] It should be noted that the inner distance field is a distance distribution field assigned to each voxel within the mineralized composite mask, based on its shortest distance to the nearest location outside the composite mask, used to characterize the local thickness of different regions. The upper quartile position is a statistical location parameter used to extract the representative scale of the thicker portion from the numerical distribution of the inner distance field; a preferred value of 0.75 is used, meaning this position preserves the representativeness of the thick region while reducing the impact of extreme large values ​​on scale estimation. The upper quartile extraction operator is a statistical operator that extracts values ​​located at the 75th percentile after sorting a set of values ​​from smallest to largest, used to obtain the thick core scale from the inner distance field. The thick core scale is a scale parameter reflecting the typical radius of the thicker region within the mineralized composite mask, used as a screening threshold for the thickness attribute opening operation. The thickness attribute opening operation operator is a morphological screening operator that retains or removes voxel structures based on whether the radius of the local maximum inscribed sphere reaches the specified thick core scale, used to preferentially retain thick and continuous bone structures. The bone mask is the set of thick structures retained from the mineralized composite mask after the thickness attribute opening operation, used to characterize the bone tissue region. The candidate calcification mask is the set of remaining high-density regions obtained after removing the bone mask from the mineralization joint mask, and is used as input for target calcification selection. The set difference operator is a set operation tool that removes the overlapping portion of one set from another, and is used to subtract the bone mask from the mineralization joint mask.

[0019] It should be noted that the Otsu algorithm constructs the input histogram as follows: first, grayscale range clipping is performed on the resampled volume data to remove outliers that significantly exceed the effective range of the tissue; then, a histogram of all voxels is calculated using a fixed number of grayscale bins; finally, a threshold search to maximize inter-class variance is performed on this histogram. For data with obvious metal artifacts, artifact regions can be masked before histogram calculation. The internal distance field is calculated as follows: the mineralized joint mask is transformed into a three-dimensional binary volume, and a three-dimensional Euclidean distance transformation is performed on the foreground voxels to obtain the shortest physical distance from each foreground voxel to the nearest background voxel; to ensure distance accuracy, the distance calculation should be based on the physical length of the equal-voxel grid, rather than simply accumulating by the number of index steps. The upper quartile extraction method is as follows: first, collect the internal distance field values ​​of all voxels inside the mineralized joint mask, then sort them in ascending order, and determine the target position by multiplying the sample number by 0.75; when the target position falls between two adjacent samples, the upper quartile is obtained by linear interpolation; when the number of internal voxels is too small, the average value of adjacent positions can be used directly as an approximate result.

[0020] It should be noted that the thickness attribute opening operation is implemented as follows: first, the maximum inscribed sphere radius of each voxel is calculated within the mineralized joint mask; then, only voxels with a maximum inscribed sphere radius not less than the thick core scale and connected to adjacent thick core regions are retained; finally, connectivity repair is performed on the retained regions to form the bone mask. This method emphasizes local thickness attributes, rather than the traditional opening operation with a fixed structuring element radius. The boundary determination method between the bone mask and the candidate calcification mask in the attached region is as follows: when a thick core region and a thin layer extension region exist simultaneously within the same connected high-density structure, the continuous thick core body and its stable thickness continuation are classified into the bone mask, while thin layer protrusions, burrs, and separated fragments below the thick core threshold are retained in the candidate calcification mask; for regions with ambiguous boundaries, an additional round of refinement segmentation based on the distance field gradient can be added. The method for handling isolated noise and holes in the joint mask is as follows: after Otsu threshold segmentation, independent noise with a number of voxels less than the preset minimum number of voxels is removed first, and then local hole filling is performed on small holes that are obviously formed by sampling noise, so as to avoid these discrete anomalies from interfering with the calculation of the inner distance field and the upper quartile.

[0021] Specifically, this invention addresses three-dimensional volumetric data where bone tissue and calcifications may simultaneously exhibit high grayscale values. Therefore, the threshold cannot rely on manually fixed values ​​but should be adaptively determined based on the overall grayscale distribution of the current data. Thus, the Otsu algorithm is introduced to automatically calculate the threshold; this maintains good threshold adaptability across different scanning devices, doses, and patient data. Because this invention does not directly distinguish between bone and calcification at this stage, but rather captures all high-density mineralized structures uniformly, the set of voxel locations with grayscale values ​​greater than or equal to the threshold is used as a combined mask for the mineralized body. This avoids missing true lesion structures due to bone-calcification adhesion in the early stages. Furthermore, because bone and calcification may have similar intensities but often differ in local thickness, it is necessary to calculate the shortest Euclidean distance from the outside of the mask to each voxel within the combined mask, forming an inner distance field. Locations with larger inner distance fields are usually closer to thicker core regions, while locations with smaller inner distance fields are usually closer to thinner layers, edges, or fine bridges.

[0022] Specifically, because maxima are easily affected by occasional thick regions and discrete noise, and the median may be too conservative, the upper quartile of the distribution of the inner distance field is used as the thick core scale, which can stably characterize the typical level of thick regions; for example, the main cortical bone region often maintains a distance value significantly greater than that of thin calcification at this statistical position. Since conventional fixed structural element opening operations are easily affected by object orientation and local irregularities, thickness attribute opening operations are used, with the maximum inscribed sphere radius as the criterion for structure screening; this preserves thick and continuous bone tissue while suppressing thinner, more fragmented, or more irregular calcification deposition. Furthermore, this invention requires specific analysis of calcification bodies in subsequent steps, so after obtaining the bone mask, the bone mask must be removed from the mineralized body combined mask to obtain candidate calcification masks; this ensures that subsequent connected component screening mainly focuses on possible calcification deposition, rather than mixing in bone tissue again.

[0023] In one embodiment of the present invention, the candidate calcification mask is decomposed into connected regions, and target calcifications are selected based on the volume of the connected regions and the average bone distance to the bone mask. The morphological framework of the target calcifications is extracted to obtain the median thickness, including: The formula for calculating the volume of a connected component is as follows: ; In the formula, For the volume of the connected component, For connected components, The number of connected voxels. The side length is equal to that of a voxel; The formula for calculating the average bone distance is as follows: ; In the formula, The average bone distance, For connected components Any voxel location within, The boundary of the bone mask, The voxel position on the bone mask boundary, Let the three-dimensional spatial coordinates of the voxels inside the connected domain be denoted as . The three-dimensional spatial coordinates of the voxel positions on the bone mask boundary; The calculation formula for selecting the target calcification is as follows: ; ; In the formula, Choose a score for the connected component. For the target calcification, As a candidate calcification mask, To mask the candidate calcification The set of all connected components obtained from the decomposition. This indicates that each connected component in the set is traversed. Select an operator for the maximum value; The formula for extracting the morphological skeleton is as follows: ; In the formula, For morphological skeleton, Corrosion level, This indicates all corrosion levels. The union of the resulting skeleton residue sets is taken. For morphological structural elements, For the erosion operation operator, For set difference operators, This is the opening operation operator; The formula for calculating the median thickness is as follows: ; ; In the formula, For local radius, This refers to the location of skeletal voxels in a morphological skeleton. For the target calcification boundary, The voxel position on the boundary of the target calcification. The three-dimensional spatial coordinates of the skeleton voxel positions are: Let be the three-dimensional spatial coordinates of the voxel position on the boundary of the target calcification. The median thickness, The median extraction operator.

[0024] It should be noted that a connected component is an independent three-dimensional voxel block that is reachable from each other under a selected adjacency rule within a candidate calcification mask, used to represent a candidate calcification structural unit. The connected component set is the set of all connected components obtained after decomposing the candidate calcification mask, used as a candidate library for screening target calcifications. The number of connected component voxels is the number of voxels contained within a connected component, used in volume calculations. The connected component volume is the product of the number of connected component voxels and the cube of the side length of an equal voxel, used to measure the spatial scale of the candidate structure. The bone mask boundary is the set of boundary voxels between the bone mask and its external space, used as a benchmark for calculating the average bone distance and bone distance. The average bone distance is the average of the shortest Euclidean distances from all voxels within a connected component to the bone mask boundary, used to characterize the overall bone adhesion of the connected component. The connected component selection score is a score that encodes both the connected component volume and bone adhesion, used to select the target calcification that best fits the treatment scenario from multiple connected components. The target calcification is the connected component that should be the primary target for subsequent target reconstruction; it is the direct object of the weight field, the partial clustering, and the final target generation. The maximum value selection operator is the selection operator that chooses the object corresponding to the maximum score from the candidate score set, and is used to determine the target calcification.

[0025] It should be noted that the morphological skeleton is a discrete central support within the target calcification that maintains topological connectivity and reflects the orientation of local centers, used to estimate the local radius and median thickness. The erosion level is a marker of the number of iterations performed when repeatedly performing erosion operations on the target calcification, used to describe different contraction stages in the skeleton extraction process. The morphological structuring element is a local neighborhood template used in morphological erosion and opening operations, used to control the morphological response range during skeleton extraction. The erosion operator is a morphological operator that causes the target calcification boundary to contract inward, used to progressively reduce outer voxels. The opening operator is a morphological operator that erodes first and then expands, used to remove protrusions smaller than the structuring element. The union operator is a set operation tool that merges the skeleton residue sets obtained from multiple stages into a total skeleton. The skeleton voxel position is the discrete voxel position of a specific skeleton point in the morphological skeleton, used to calculate the local radius. The target calcification boundary is the set of boundary voxels between the target calcification and its external space, used to calculate the local radius and shell depth. The local radius is the minimum Euclidean distance from a given skeleton voxel to the boundary of the target calcification, representing half the local thickness of the object at that skeleton point. The median extraction operator is a statistical operator that extracts the median representative value from a set of sorted values, used to improve the robustness of thickness estimation to extreme spikes and anomalous branches. Constant 2 is a scale constant that converts the local radius to the corresponding local thickness; that is, the thickness is geometrically equal to twice the corresponding radius. The median thickness is the representative thickness of the target calcification obtained by multiplying the median of all local radii by constant 2, used for subsequent calculations of the weight field, geodesic dilatation radius, and closure radius.

[0026] It should be noted that the adjacency criterion for connected components is as follows: in 3D voxel space, a 26-neighborhood is used as the default definition of connectivity; any two foreground voxels that touch on a face, edge, or vertex are considered connected. If the data noise is strong and there are many false connections with thin bridges, it can be changed to an 18-neighborhood, which will not be elaborated here. The method for extracting the bone mask boundary is as follows: first, perform a single spherical structuring element erosion on the bone mask, and then subtract the erosion result from the original bone mask to obtain a boundary set consisting only of the outermost bone voxels; this boundary set is then used for subsequent shortest Euclidean distance calculation. The handling method for connected component selection scores with the same value is as follows: prioritize the connected component with the smaller average bone distance; if the average bone distance is still the same, prioritize the connected component with the larger volume; if both are still the same, select the connected component with the larger contact area with the bone mask; this rule can avoid object ambiguity in the case of parallel selection.

[0027] It should be noted that the shape, scale progression strategy, and termination condition of the morphological structural elements are as follows: A unit spherical structural element is used as the basic structural element. The number of erosion layers is increased progressively from the initial erosion level until the target calcification is completely eroded away. Each layer undergoes erosion and opening difference extraction, and the results of each layer are merged into the overall skeleton. The handling of empty or excessive skeleton branches is as follows: When the skeleton is empty, the degeneracy uses the set of local maxima of the Euclidean distance transformation of the target calcification as a substitute skeleton; when there are too many skeleton branches, a small branch pruning is performed first, retaining only the main trunk and major branches with a length not less than the preset minimum skeleton length, and then the local radius is calculated. The robustness handling of the median thickness when the number of skeleton points is too small is as follows: When the number of effective skeleton points is less than 3, the median is no longer relied upon alone. Instead, the thickness of an equal-volume sphere corresponding to the target calcification volume is also referenced, and the smaller value between the two is taken as a conservative thickness estimate to avoid abnormal thickness amplification due to a very small number of skeleton points.

[0028] It should be noted that bone distance refers to the shortest Euclidean distance from the voxel position inside the target calcification to the boundary of the bone mask, used to represent the spatial proximity of the voxel position inside the target calcification to the bone surface; mean bone distance refers to the arithmetic mean of the shortest Euclidean distances from all voxel positions inside a connected region to the boundary of the bone mask, used to represent the average spatial proximity of the entire connected region relative to the boundary of the bone mask. The boundary of the bone mask is the set of boundary voxels between the bone mask and its external space; both bone distance and mean bone distance are calculated based on the physical length corresponding to the side length of equal voxels using Euclidean distance calculation.

[0029] Specifically, the candidate calcification mask may contain multiple discrete high-density blocks. Therefore, it must first be decomposed into multiple independent connected three-dimensional voxel blocks before comparing which connected region best matches the target calcification in the shoulder treatment scenario. Without this step, noise blocks or migrating fragments far from the bone surface may be mixed with the actual lesion. Because the number of voxels alone cannot reflect the true physical scale of the object, the volume of the connected region is calculated by multiplying the number of voxels in the connected region by the cube of the side length of the same voxel. The volume obtained in this way can maintain a consistent physical meaning under different scanning resolutions. In addition, this invention focuses on treatment objects close to the bone surface, so in addition to volume, the average bone distance must also be introduced. The smaller the average shortest distance from all voxels inside the connected region to the boundary of the bone mask, the closer the connected region is to the bone surface, and the more likely it is to be the main calcification near the rotator cuff attachment area.

[0030] Specifically, the outer edges of irregular calcifications are often rough and contain local fine bridges, making it difficult to stably characterize the object's scale using the overall circumscribed box or average thickness. Therefore, a combination of erosion and opening operations is used to extract the morphological skeleton. The skeleton preserves the internal center orientation of the object, making it suitable for local radius statistics. The minimum Euclidean distance from a skeleton point to the boundary can represent the maximum inscribed sphere radius at that location; therefore, this minimum distance is defined as the local radius. The set of local radii can statistically reflect the object's thickness distribution well. Furthermore, since the local radii may contain extreme values ​​caused by fine bridges, burrs, or local branches, the median is used instead of the average, and then multiplied by a constant two to obtain the median thickness. The representative thickness obtained in this way is more stable and more suitable as the basis for subsequent adaptive scale parameters.

[0031] In one embodiment of the present invention, the voxel shell depth, bone distance, and positive curvature of the bone surface within the target calcified body are calculated, and a weighted field is generated by combining the median thickness. The volume average value of the weighted field is then calculated to obtain the segregation amount, including: The formula for calculating the shell depth is as follows: ; In the formula, The depth of the shell. The location of voxels within the target calcified body. For the target calcification boundary, The voxel position on the boundary of the target calcification. Voxel location within the target calcified body Three-dimensional spatial coordinates, The three-dimensional spatial coordinates of the voxel positions on the boundary of the target calcification; The formula for calculating bone distance is as follows: ; In the formula, For the septum, As the boundary of the bone mask, The voxel position on the bone mask boundary, The three-dimensional spatial coordinates of the voxel positions on the bone mask boundary; The formula for calculating the positive curvature of the bone surface is as follows: ; In the formula, For the positive curvature of the bone surface, For local mean curvature, Maximum value extraction operator; The formula for generating the weight field is as follows: ; In the formula, The value of the weight field at the voxel location. For the natural exponential function operator, The median thickness, voxel position The positive curvature of the bone surface at the nearest bone surface point. Curvature adjustment factor; The formula for calculating the segregation amount is as follows: ; In the formula, For segregation amount, For the target calcification volume, For the target calcification, Target calcification Any voxel location within, Indicates to Sum the positions of all internal voxels. voxel position The weight field value at that location, The side length is equal to that of a voxel.

[0032] It should be noted that shell depth is the shortest Euclidean distance from a voxel within the target calcification to its boundary, indicating whether the voxel is located on the surface or deep within the object. Bone distance is the shortest Euclidean distance from a voxel within the target calcification to its bone mask boundary, indicating the spatial proximity of the voxel to the bone surface. Positive bone surface curvature is the positive portion of the local average curvature at the point on the bone mask boundary closest to the target voxel, used to emphasize the spatial position in front of a convex bone face. Local average curvature is a quantification of the curvature of the local surface at the bone mask boundary, used to characterize whether the bone surface is flat, convex, or concave near a certain point. The maximum value extraction operator is a tool that takes the larger of two candidate values, used to truncate negative values ​​in the local average curvature to zero. The natural exponential function operator is an exponential mapping tool with the natural constant as its base, used to convert shell depth and bone distance into weighting factors that continuously decay with increasing distance. The weighted field is a voxel distribution field that assigns relative importance to each voxel within the target calcification, used to indicate which voxels are closer to the effective area to be reconstructed in this invention. The constant is the basic bias of the curvature adjustment factor, meaning that the basic weight must be retained even when the positive curvature of the bone surface is zero, without compressing the term to zero. The concentration factor is the volume average of the weighted field within the target calcification, used to characterize whether high-weighted voxels are more concentrated in areas that are close to the bone, on the surface, and located in front of the protruding bone.

[0033] It should be noted that the method for finding the nearest point and handling multiple solutions at the bone mask boundary is as follows: First, a three-dimensional nearest neighbor query structure is established for the bone mask boundary, and the nearest bone surface point is searched sequentially for the target voxel. When there are multiple nearest bone surface points with the same distance, the point with the smaller absolute value of the local average curvature is selected as the stable representative point. The method for calculating the local average curvature is as follows: First, the bone surface point cloud is extracted from the voxels of the bone mask boundary, and then a quadratic surface is fitted using the local neighborhood. The average curvature of the point is obtained based on the principal curvature of the fitted surface. To reduce the voxel step effect, a slight smoothing can be performed on the bone surface before fitting. This can obtain a curvature approximation suitable for discrete volume data. The implementation method for the positive curvature neighborhood scale of the bone surface is as follows: Taking the nearest bone surface point as the center, a neighborhood with a radius of 2 to 4 equal voxel side lengths is selected for surface fitting, and a default window of 3 equal voxel side lengths is preferred. This range can cover a sufficient number of surface points without smoothing out local bone surface protrusions due to an excessively large neighborhood. The strategy for stabilizing the weighted field is as follows: when the median thickness is too small, causing the exponential term to decay too quickly, a minimum lower limit is first set for the median thickness; when the positive curvature of the bone surface is abnormally large, an upper limit is set for the curvature adjustment factor to prevent individual curvature peaks from raising the weighted field to an unreasonable level; through the bidirectional constraints of the lower and upper limits, the subsequent cohesion calculation can be stabilized. The cohesion degradation handling method when the target calcification volume is too small or the number of voxels is too few is as follows: when the number of voxels in the target calcification is lower than the preset minimum number of voxels, the use of the volume-averaged cohesion as an independent scale indicator is suspended, and instead, the normalized weighted sum and the bone distance quantile statistics are used together to form an alternative cohesion score to avoid statistical fluctuations caused by a very small number of voxels.

[0034] Specifically, because this invention focuses not on the average properties of the entire calcified body, but rather on whether the target voxel is located in the surface region that is truly likely to bear the main shock wave action, the first step is to calculate the shortest distance from each internal voxel to the boundary of the target calcified body, forming the shell depth. The smaller the shell depth, the closer the voxel is to the outer layer, and the more likely it is to belong to the action shell. Since bone proximity is one of the decisive factors in the scenario of this invention, it is also necessary to calculate the shortest distance from each internal voxel to the boundary of the bone mask, forming the bone distance. The smaller the bone distance, the closer the voxel is to the bone surface, and the more easily it is affected by bone surface reflection and gap geometry constraints. In addition, different bone surface morphologies have different effects on local energy distribution and spatial constraints, so we cannot only look at the bone distance, but also at the local average curvature at the nearest bone surface point, and only retain its positive value as the positive curvature of the bone surface. The implication of this is to highlight the target voxel in front of the convex bone surface, rather than giving the concave and convex areas the same weight.

[0035] Specifically, since both shell depth and bone distance should exhibit a gradually decreasing effect with increasing distance, the negatives of their ratios to median thickness are fed into the natural exponential function to construct two smooth attenuation terms. This preserves the continuous trend while avoiding abrupt changes caused by using a hard threshold. Because the effect of bone surface curvature needs to be adjusted beyond the basic distance weights, a curvature adjustment factor is constructed by adding a constant to the product of median thickness and positive bone surface curvature. This allows voxels located in front of more convex bone surfaces to receive additional emphasis while maintaining the basic weights. Furthermore, since subsequent geodesic dilatation requires an overall scale index and cannot rely solely on individual voxel weights, the weight field of all voxels within the target calcification is volume-averaged to obtain a concentration. A larger concentration indicates that high-weight voxels are more concentrated in the bone lateral shell region that this invention aims to reconstruct preferentially.

[0036] In one embodiment of the present invention, a seed set is extracted based on a seed threshold calculated using a weighted field, the number of geodesic dilatation steps is determined by the median thickness and the amount of segregation, and geodesic dilatation is performed on the seed set within the target calcification to generate a core target region, including: The formula for calculating the seed threshold and extracting the seed set is as follows: ; ; In the formula, Seed threshold Indicates to Sum the positions of all internal voxels. For the target calcification, For the weight field at the voxel position The value at that location, For seed set; The formula for determining the number of geodesic dilatation steps is as follows: ; ; In the formula, For geodesic expansion radius, The median thickness, For segregation amount, To determine the number of geodesic dilatation steps, For equal voxel side lengths, This is the floor operator; The formula for generating the core target region is as follows: ; In the formula, As the core target area, This is a geodesic dilatation operator that uses the target calcification as a boundary mask constraint and performs a certain number of geodesic dilatation steps.

[0037] It should be noted that the seed threshold is an adaptive threshold obtained based on the statistical results of all weight field values ​​within the target calcification, used to screen out high-weight seed voxels in the weight field. The seed set is the set of all voxel positions within the target calcification whose weight field values ​​are greater than or equal to the seed threshold, used as the starting region for geodesic dilatation. The geodesic dilatation radius is the growth radius determined by the median thickness and the degree of segregation, used to control the physical scale of the seed set's expansion. The rounding up operator is an arithmetic tool that rounds up to the nearest integer when converting continuous radii to discrete voxel steps, used to ensure that the actual dilatation steps are not less than the target physical radius. The geodesic dilatation step count is the discrete number of iterations obtained by dividing the geodesic dilatation radius by the equal voxel side length, used to control the actual number of growth rounds of geodesic dilatation. The geodesic dilatation operation operator is a morphological operator that performs local expansion under the constraint of the target calcification boundary mask, used to expand the seed set without exceeding the target calcification's range. The core target region is a high-weight main region formed by geodesic dilatation of the seed set inside the target calcification, which is used as input for subsequent closing operations and morphological reconstruction.

[0038] It should be noted that the anti-degeneration processing method when the sum of the weighted fields is zero is as follows: when the sum of all weighted field values ​​inside the target calcification is zero or close to zero, first check whether the median thickness and bone distance calculations are abnormal; if the object does not have an effective high-weight region, then directly use the single element or the smallest connected high-weight cluster corresponding to the maximum weighted field as the degeneration seed set, and set the seed threshold to the maximum weighted field value. The method of geodesic dilatation neighborhood structure and structuring element form is as follows: on the three-dimensional isovoxel grid, a unit spherical structuring element or a 26-neighborhood is used as the neighborhood template for one dilatation. After each dilatation, it is immediately intersected with the target calcification mask to ensure that the result never goes out of bounds. The merging strategy between multiple discrete seed components is as follows: first retain all seed components that meet the threshold to participate in geodesic dilatation; if multiple completely separated connected components appear in the final core target area, then the component with the largest volume and the smallest average bone distance is retained first, and the remaining components are only merged when the minimum distance from the principal component does not exceed one isovoxel side length. The method for trimming and iterating voxels that cross the boundary under the boundary mask constraint is as follows: in each round of geodesic dilatation, candidate dilatation results are generated first, and then all voxels that do not belong to the target calcification are immediately deleted, and then the next round of iteration is entered; by trimming in round by round instead of uniform trimming in the last round, bridging errors caused by cross-boundary propagation can be avoided.

[0039] Specifically, because voxels truly worthy of being seeds in the weighted field should simultaneously possess weights significantly higher than the average level, and cannot be simply relied upon based on fixed percentiles, a seed threshold is constructed using the ratio of the sum of squares of the weighted field to the first-order sum of the weighted field. This threshold naturally favors high-weight regions, resulting in more significant suppression of low-weight long-tailed portions. This invention does not require a single high-weight point, but rather a high-weight starting region capable of stable subsequent growth. Therefore, all voxels with weighted field values ​​greater than or equal to the seed threshold are extracted from within the target calcification to form a seed set. This seed set is typically located in high-weight regions adhering to the bone surface. Furthermore, since the seed set only covers the most prominent high-weight regions and cannot represent the complete target area, the geodesic expansion radius must be determined based on the object's geometric scale and degree of clustering. Multiplying the median thickness by the degree of clustering allows thick, significantly clustered objects to achieve more complete expansion, while thin, dispersed objects automatically shrink their expansion range.

[0040] In one embodiment of the present invention, the closing radius is calculated from the median thickness and the amount of segregation; the closing radius is used to perform a closing operation on the core target region to generate an intermediate target region; and morphological reconstruction is performed on the intermediate target region using the target calcification as a constraint to generate the final effective target region, including: The formula for calculating the radius of the closed operation is as follows: ; In the formula, The radius of the closed operation. The median thickness, This refers to the amount of segregation; The formula for generating the intermediate target region is as follows: ; In the formula, For the intermediate target area, As the core target area, This is a morphological dilation operator. For morphological erosion operations, A spherical structural element constructed using the closed operation radius; The formula for calculating the final effective target area is as follows: ; In the formula, For the final effective target area, This is a morphological reconstruction operator that uses the target calcification as a reconstruction constraint mask. Target calcification.

[0041] It should be noted that the closing radius is a structural repair scale adaptively calculated based on the median thickness and segregation, used to control the boundary completion intensity of the core target area. The spherical structural element is a three-dimensional local neighborhood template constructed using the closing radius as the scale, used to perform morphological expansion and erosion. The morphological expansion operator is a morphological operator that expands the core target area outwards into its neighborhood, used to fill small cracks and connect near-distance boundary gaps. The intermediate target area is a temporary target area obtained after the closing operation of the core target area, used as a marker starting point for morphological reconstruction. The morphological reconstruction operator is a morphological operator that performs restricted propagation within the target calcification constraint mask, used to limit the repair results of the intermediate target area within the range of legal objects. The final effective target area is the terminal target area obtained by morphological reconstruction of the intermediate target area under the constraint of the target calcification, and is the main spatial object directly output by the therapeutic instrument in this invention.

[0042] It should be noted that the construction method of discretizing the spherical structural element onto the voxel mesh is as follows: the discrete radius is obtained by dividing the physical length corresponding to the closed operation radius by the side length of the voxel. Then, taking the central voxel of the structural element as the origin, all voxels whose Euclidean distance to the center does not exceed the discrete radius are collected as members of the spherical structural element; the structural element obtained in this way is consistent with the physical radius. The method of converting the closed operation radius to the discrete voxel radius is as follows: first calculate the continuous radius, then use rounding or up to convert it to an integer voxel radius, and set a minimum radius lower limit of not less than 1 voxel; if the continuous radius is less than half a voxel, at least a spherical structural element of 1 voxel is used to ensure that the closed operation has a practical repair effect. The adjacency criterion, iterative convergence condition, and stopping condition of morphological reconstruction are as follows: the 26-neighborhood is used as the propagation adjacency rule, the intermediate target area is used as the label set, and the target calcification is used as the constraint mask. In each iteration, only newly added voxels are allowed to simultaneously satisfy the condition of being adjacent to the current label and belonging to the constraint mask. The iteration stops when no new voxels are added after one iteration. The pruning rules for intermediate target regions that cross boundaries or touch multiple discrete branches are as follows: all voxels that do not belong to the target calcification constraint mask are immediately deleted; for intermediate target regions that touch multiple discrete branches at the same time, only the propagation results connected to the principal component of the core target region are retained, and the remaining independent propagation parts are removed before reconstruction begins to prevent unreasonable expansion of the target region.

[0043] Specifically, because the core target area typically only covers the high-weight main body, and there may be small gaps and burrs at the edges, a closing operation radius related to the object thickness and the degree of cohesion is needed. Using the median thickness divided by the sum of the cohesion and a constant, the closing operation radius can be made more convergent when the cohesion is stronger, and moderately wider when the cohesion is weaker. Since this invention deals with three-dimensional voxel objects, the structural element should be a three-dimensional sphere, not a two-dimensional planar structural element. Spherical structural elements can maintain a relatively consistent geometric completion effect in all directions, suitable for repairing small holes, short cracks, and locally rough boundaries. The closing operation consists of expansion followed by erosion, which can fill small gaps at the boundaries and small holes inside without significantly expanding the overall area of ​​the object. Therefore, applying this operation to the core target area can yield a more complete intermediate target area. Furthermore, since simple closing operations may still cross the boundaries of real objects and incorrectly include migration branches or neighboring noise, it is also necessary to use the target calcification as a reconstruction constraint mask to perform morphological reconstruction on the intermediate target area. Through this restricted reconstruction, all repaired or propagated voxels are confined within the target calcification, thus forming the final effective target area.

[0044] In one embodiment of the present invention, the target volume and target center are obtained using the final effective target area, and an equivalent ellipsoid is extracted based on the covariance matrix of the internal voxel coordinates and the target center. The final effective target area, target volume, target center, and equivalent ellipsoid are then combined and output, including: The formulas for calculating the target volume and the target center are as follows: ; ; In the formula, For target volume, This represents the total number of voxels contained within the final effective target area. For the final effective target area, For equal voxel side lengths, Center of the target area The voxel locations within the final effective target area. Internal voxel coordinates; The formula for calculating the covariance matrix is ​​as follows: ; In the formula, Let covariance matrix be the variance matrix. This is the position offset vector. For matrix transpose operator, This indicates the final effective target area. Sum the positions of all internal voxels; The formula for extracting the equivalent ellipsoid is as follows: ; ; In the formula, The semi-axis length of the equivalent ellipsoid. These are the eigenvalues ​​of the covariance matrix. This is a positive integer identifier with values ​​of 1, 2, and 3. For an equivalent ellipsoid, For the coordinates of a point in three-dimensional space, The direction matrix, It is the transpose of the direction matrix. Construct operators for diagonal matrices; The merged output is as follows: ; In the formula, To quantify the 3D reconstruction result set.

[0045] Specifically, because pre-treatment localization requires not only the object's size but also its overall position, the average coordinates of all voxels within the final effective target area are used as the target area center; this center reflects the overall spatial distribution of the object better than the simple circumscribed box center. Since the spatial extension direction and dispersion of the final effective target area cannot be expressed solely by volume and center, a position offset vector relative to the center is constructed for each internal voxel, and the cross product results are averaged to form a covariance matrix. The covariance matrix can simultaneously reflect the object's distribution and directional coupling relationships in three directions. Furthermore, since the eigenvalues ​​and eigenvectors of the covariance matrix represent the dispersion intensity and the principal direction itself, respectively, the principal axis orientation system of the object can be established by obtaining the three eigenvalues ​​and their corresponding three orthogonal eigenvectors; this orientation system is used to subsequently construct the equivalent ellipsoid.

[0046] Specifically, the output of this invention requires not only a true voxel mask but also a compact geometric representation that is easy for devices to read. Therefore, the three eigenvalues ​​are multiplied by a constant five and then squared to obtain the three semi-axis lengths of the equivalent ellipsoid. The resulting ellipsoid has a consistent spatial distribution with the original target area in a second-order statistical sense. Furthermore, the equivalent ellipsoid must simultaneously contain three elements: orientation, position, and scale. Therefore, a quadratic inequality of spatial coordinates is constructed using the orientation matrix, the target center, and the three semi-axis lengths to define the ellipsoid boundary. Any three-dimensional point that satisfies this inequality is considered a point inside the equivalent ellipsoid.

[0047] It should be noted that the total number of voxels contained within the final effective target area is the count of all foreground voxels in the final effective target area, serving as the basis for volume, center, and covariance statistics. The target area volume is the product of the total number of voxels within the final effective target area and the cube of the side length of an equal voxel, used to quantify the size of the final effective target area in physical space. The target area center is the average position of the coordinates of all voxels within the final effective target area, used to characterize the overall spatial center of the final effective target area. The position offset vector is the displacement vector of a certain internal voxel coordinate relative to the target area center, used to construct the covariance matrix. The covariance matrix is ​​a second-order statistical matrix obtained by averaging the outer products of all position offset vectors, used to describe the dispersion of the final effective target area in the three principal directions. The matrix transpose operator is a linear algebraic operation tool that converts column vectors into row vectors or flips a matrix along its main diagonal, used to calculate the outer product of position offset vectors and the quadratic form of the equivalent ellipsoid. Eigenvalues ​​are the discrete intensity quantities corresponding to the covariance matrix in each principal direction, used to derive the lengths of the three semi-axes of the equivalent ellipsoid. Positive integers of 1, 2, and 3 are used to distinguish the three eigenvalues ​​and their corresponding three semi-axis directions. Orthogonal eigenvectors are three direction vectors that correspond to and are orthogonal to the eigenvalues ​​of the covariance matrix, used to construct the direction matrix. The direction matrix is ​​a matrix composed of the three orthogonal eigenvectors arranged column-wise or row-wise, used to determine the orientation of the equivalent ellipsoid in three-dimensional space. Constant 5 is the proportionality constant that maps the eigenvalues ​​of the covariance matrix to the semi-axis lengths of the equivalent ellipsoid of a uniform entity; that is, the standard correspondence between the second moment of the uniform ellipsoid and the semi-axis length. The semi-axis length is the scale parameter of the equivalent ellipsoid in the three principal directions, used to determine the major, median, and minor axes of the equivalent ellipsoid. The diagonal matrix construction operator is an operator that fills the main diagonal with the squares of the reciprocals of the three semi-axis lengths to form a quadratic matrix, used to construct the equivalent ellipsoid determination formula. The equivalent ellipsoid is an ellipsoid that matches the final effective target area in a second-order statistical sense, used to provide a more compact and easily accessible geometric representation for equipment. The quantized 3D reconstruction result set is a set of results that are uniformly encapsulated from the final effective target area, target volume, target center, and equivalent ellipsoid, and is used as the final output of the system.

[0048] It should be noted that the coordinate system and unit reference used for the covariance matrix are as follows: a unified physical coordinate system after equal voxel resampling is adopted, with millimeters as the unit of length, and the physical origin corresponding to the origin of the resampling grid as the coordinate zero point; all internal voxel coordinates should be transformed to this unified coordinate system before participating in the calculation of center and covariance. The method for unifying the eigenvalue sorting and orthogonal eigenvector directions is as follows: eigenvalues ​​are first sorted from largest to smallest, so that the first eigenvalue corresponds to the major axis, the second eigenvalue to the median axis, and the third eigenvalue to the minor axis; for eigenvectors with opposite signs but equivalent directions, the orientation with the smallest angle to the main positive direction of the patient coordinate system is used as the unified orientation to avoid random flipping of the output direction for different cases. The method for setting the constant five is as follows: the final effective target area is regarded as an equivalent entity with uniform density, and its covariance matrix principal axis eigenvalues ​​satisfy a fixed proportional relationship with the square of the ellipsoidal semi-axis. Using the constant five can restore the second-order statistical distribution to the corresponding geometric semi-axis length.

[0049] It should be noted that the equivalent ellipsoid boundary discretization extraction method is as follows: first, a continuous ellipsoid criterion is established based on the direction matrix and semi-axis length; then, the criterion is checked point by point within the local 3D mesh covering the ellipsoid to extract the set of voxels inside the ellipsoid. If only parameter output is required, the center, direction matrix, and semi-axis length can be directly output without discretization into voxels. The degenerate handling method for singular covariance matrices caused by an insufficient number of effective target voxels is as follows: when the total number of internal voxels is less than 3 or the rank of the covariance matrix is ​​insufficient, a degenerate ellipsoid based on voxel bounding boxes is used to replace the covariance ellipsoid, or the minimum bounding sphere is used as the alternative output to ensure the completeness of the result set. The output data format and interface structure of the quantified 3D reconstruction result set are as follows: the final effective target area is output as a 3D binary mask file, the target volume and target center are output as a structured parameter file, and the equivalent ellipsoid is output as three parameters: center, direction matrix and semi-axis length. At the same time, the case identifier, coordinate system description and timestamp are recorded in the result set for easy reading by the treatment instrument and workstation.

[0050] Specifically, in actual deployment, the present invention is preferentially deployed on the image processing workstation supporting the shock wave therapy apparatus, and can also be deployed on the bypass computing node of the hospital image server. In actual use, three-dimensional image acquisition of the patient's shoulder is completed before treatment, and it is recommended to cover the greater tuberosity of the humerus, the attachment area of the supraspinatus tendon, the subacromial space and adjacent soft tissue areas. After acquisition is completed, the original three-dimensional volume data is imported into the system, and the system automatically reads the voxel spacing, image origin and grayscale volume data in three directions, and completes isometric voxel resampling first. After resampling is completed, the system sequentially performs mineralized body combined mask extraction, bone mask and candidate calcification mask separation, target calcified body screening, morphological skeleton extraction, median thickness calculation, weight field generation, deviation agglomeration calculation, seed set extraction, geodesic dilation, closing operation, morphological reconstruction and equivalent ellipsoid fitting. The entire processing flow can be completed automatically after a single case is loaded, and doctors can also be allowed to perform manual confirmation between candidate results.

[0051] It should be noted that in the data acquisition link, the original three-dimensional volume data can come from conventional shoulder computed tomography, or from other imaging devices that can output standard three-dimensional volume data. For the application scenario of the present invention, it is more suitable to select scan data that can clearly present bone tissue and high-density calcification deposits. After the system reads the data, it does not require the operator to manually delineate the entire lesion, and only needs to confirm that the processing area is within the relevant anatomical range of the shoulder. Then the system completes target object recognition by using grayscale threshold, adaptive thickness separation and connected domain scoring. Since adaptive quantities such as isovoxel side length, median thickness and deviation agglomeration are used in the process, frequent adjustment of manual thresholds is not required between different cases, which is conducive to maintaining consistent processing caliber in continuous work of multiple cases.

[0052] It should be noted that the final output result of the present invention is not a single segmentation map, but a quantitative result set that can be directly used for pre-treatment planning. The result set includes at least the final effective target area, target area volume, target area center and equivalent ellipsoid. The final effective target area can be superimposed and displayed on the original image in the form of a three-dimensional binary mask, which is used to visually display the suggested treatment area. The target volume is used to reflect the spatial scale of the suggested treatment area. The center of the target area is used to give the reference position of the focal point of the therapy apparatus. The equivalent ellipsoid describes the general spatial range of the target area with the center position, three main directions and three semi-axis lengths, which is convenient for the equipment to perform visualization superposition and path planning. For example, the original voxel spacing of a certain case is 0.35 mm, 0.35 mm and 1.00 mm respectively, the isovoxel side length calculated by the system is 0.49 mm, the final effective target volume is 138.6 cubic millimeters, the coordinates of the target center are 42.3 mm, 61.8 mm and 27.4 mm, and the three semi-axis lengths of the equivalent ellipsoid are 5.2 mm, 3.4 mm and 2.8 mm respectively. Doctors can confirm the focal depth, target range and orientation of the treatment area before treatment accordingly.

[0053] It should be noted that when used in conjunction with a shockwave therapy device, the workstation can send the target area center to the positioning module as a focal reference point, send the equivalent ellipsoid to the display module as a three-dimensional reference boundary, and send the final effective target area to the overlay display module for case verification. In this way, pre-treatment positioning no longer relies solely on a single high-brightness point on a two-dimensional slice, nor solely on empirically marked areas on the body surface, but can combine three-dimensional voxel morphological results for a holistic interpretation of the target area. For cases with lateral bone adhesion, irregular outer edges, migration pathways, and local fine bridges, this deployment method can reduce the bias caused by mechanically equating the entire calcified lesion with the treatment target area. The practical application value of this invention mainly lies in its ability to further transform high-density calcification deposits on imaging into effective target areas for treatment planning, reducing the impact of empirical thresholds on the stability of results, and simultaneously outputting the true voxel target area and the equivalent geometric target area, facilitating clinical verification, equipment recall, case follow-up comparison, and treatment effect evaluation, which will not be elaborated upon here.

[0054] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A three-dimensional reconstruction system for the target area of ​​a shockwave therapy device based on voxel morphology, characterized in that, include: The first module performs resampling based on the same voxel side length of the original 3D volume data to generate resampled voxel data; The second module extracts the mineralized joint mask from the resampled voxel data, calculates the thick core scale according to the inner distance field, and uses the thick core scale to perform a thickness attribute opening operation to separate the bone mask and the candidate calcification mask. The third module decomposes the candidate calcification mask into connected domains, selects the target calcification body according to the volume of the connected domain and the average bone distance to the bone mask, and extracts the morphological skeleton of the target calcification body to obtain the median thickness. The fourth module calculates the depth of the voxel shell, bone distance, and positive curvature of the bone surface within the target calcified body. It then generates a weighted field by combining the median thickness and calculates the volume average value of the weighted field to obtain the segregation amount. The fourth module performs the following operations: The formula for calculating the depth of the shell is as follows: ; In the formula, The depth of the shell. The location of voxels within the target calcified body. For the target calcification boundary, The voxel position on the boundary of the target calcification. Voxel location within the target calcified body Three-dimensional spatial coordinates, The three-dimensional spatial coordinates of the voxel positions on the boundary of the target calcification; The formula for calculating bone distance is as follows: ; In the formula, For the septum, As the boundary of the bone mask, The voxel position on the bone mask boundary, The three-dimensional spatial coordinates of the voxel positions on the bone mask boundary; The formula for calculating the positive curvature of the bone surface is as follows: ; In the formula, For the positive curvature of the bone surface, For local mean curvature, Maximum value extraction operator; The formula for generating the weight field is as follows: ； In the formula, The value of the weight field at the voxel location. For the natural exponential function operator, The median thickness, voxel position The positive curvature of the bone surface at the nearest bone surface point. Curvature adjustment factor; The formula for calculating the segregation amount is as follows: ; In the formula, For segregation amount, For the target calcification volume, For the target calcification, Target calcification Any voxel location within, Indicates to Sum the positions of all internal voxels. voxel position The weight field value at that location, The side length is equal to that of a voxel; The fifth module calculates the seed threshold based on the weight field to extract the seed set, determines the geodesic dilatation steps based on the median thickness and the amount of segregation, and performs geodesic dilatation on the seed set within the target calcification to generate the core target area. The sixth module calculates the closing radius based on the median thickness and the amount of segregation. It then uses the closing radius to perform a closing operation on the core target area to generate an intermediate target area. Finally, it performs morphological reconstruction on the intermediate target area using the target calcification as a constraint to generate the final effective target area. The seventh module uses the final effective target area to obtain the target volume and target center, extracts the equivalent ellipsoid based on the covariance matrix of the internal voxel coordinates and the target center, and outputs the final effective target area, target volume, target center and equivalent ellipsoid.

2. The three-dimensional reconstruction system for the target area of ​​a shockwave therapy device based on voxel morphology according to claim 1, characterized in that, Obtain the voxel spacing in three directions of the original 3D volume data; Calculate the cube root of the product of the voxel spacing in the three directions to obtain the side length of the equal voxel. Resampled voxel data is obtained by interpolating and resampling the original 3D volume data according to the same voxel side length.

3. The three-dimensional reconstruction system for the target area of ​​a shockwave therapy device based on voxel morphology according to claim 1, characterized in that, The threshold for resampled voxel data was adaptively calculated using the Otsu algorithm. A mineralization joint mask is obtained by extracting the set of voxel locations with gray values ​​greater than or equal to a threshold from the resampled voxel data. The inner distance field is obtained by calculating the shortest Euclidean distance from the voxel position inside the mineralized joint mask to the outside of the mineralized joint mask; The thick core scale is obtained by extracting the upper quartile of the distribution of the inner distance field values; Using the thick core scale as the parameter of the maximum inscribed sphere radius, a thickness attribute opening operation is performed on the mineralized body joint mask to obtain the bone mask. The bone mask is then removed from the mineralized body joint mask to obtain candidate calcification masks.

4. The three-dimensional reconstruction system for the target area of ​​a shockwave therapy device based on voxel morphology according to claim 1, characterized in that, The candidate calcification mask is decomposed into multiple independent and connected three-dimensional voxel blocks, forming a connected domain; The volume of a connected region is obtained by multiplying the number of connected voxels contained in the connected region by the cube of the side length of each voxel. The average bone distance is obtained by calculating the average of the shortest Euclidean distances from the voxels inside the connected domain to the boundary of the bone mask. The connected component selection score is obtained by calculating the ratio of the cube root of the volume of the connected component to the sum of the average bone distance and the side length of the isovoxel. Among multiple connected components, the connected component with the highest score is selected and extracted as the target calcification. Morphological frameworks were extracted by performing erosion and opening operations on the target calcification using morphological structural elements. The local radius is obtained by calculating the minimum Euclidean distance from the skeletal voxel position in the morphological skeleton to the boundary of the target calcification. Extract the median of all local radius values, and multiply the median by a constant two to obtain the median thickness.

5. The three-dimensional reconstruction system for the target area of ​​a shockwave therapy device based on voxel morphology according to claim 1, characterized in that, Calculate the sum of squares of the weighted field values ​​at all voxel positions inside the target calcification, calculate the direct summation of the weighted field values ​​at all voxel positions inside the target calcification, and divide the sum of squares by the direct summation to obtain the seed threshold. The seed set is obtained by extracting the set of voxel positions with weight field values ​​greater than or equal to the seed threshold inside the target calcification. The geodesic expansion radius is obtained by multiplying the median thickness by the segregation rate. The number of geodesic dilatation steps is obtained by dividing the geodesic dilatation radius by the side length of the equal voxel and taking the nearest integer in the positive infinity direction. Using the target calcification as a boundary mask constraint, perform morphological geodesic dilatation on the seed set according to the number of geodesic dilatation steps to generate the core target region.

6. The three-dimensional reconstruction system for the target area of ​​a shockwave therapy device based on voxel morphology according to claim 1, characterized in that, Add the segregation amount to a constant to obtain a numerical sum, and divide the median thickness by the numerical sum to obtain the closed operation radius; A spherical structural element is constructed using the closed operation radius. The spherical structural element is then used to perform morphological expansion and morphological erosion operations on the core target area to generate an intermediate target area. Using the target calcification as a reconstruction constraint mask, and starting from the intermediate target area, morphological reconstruction operations are performed inside the reconstruction constraint mask to generate the final effective target area.

7. The three-dimensional reconstruction system for the target area of ​​a shockwave therapy device based on voxel morphology according to claim 1, characterized in that, The total number of voxels contained within the final effective target area is counted, and the target volume is obtained by multiplying the total number of voxels by the cube of the side length of the voxels. The coordinates of all internal voxels within the final effective target area are summed, and the result of the summation is divided by the total number of voxels to obtain the center of the target area. The position offset vector is obtained by subtracting the target center from the internal voxel coordinates. The position offset vector is then multiplied by the transpose of the position offset vector. The covariance matrix is ​​obtained by taking the arithmetic mean of all the matrix outer product results within the final effective target area.

8. The three-dimensional reconstruction system for the target area of ​​a shockwave therapy device based on voxel morphology according to claim 7, characterized in that, Calculate the three eigenvalues ​​of the covariance matrix, and obtain the three orthogonal eigenvectors corresponding to the three eigenvalues ​​to form the direction matrix; The lengths of the three semi-axes of the equivalent ellipsoid are obtained by multiplying the three eigenvalues ​​by a constant of five and then calculating the square root. By combining the direction matrix, the target center, and the lengths of the three semi-axis, a quadratic form inequality of spatial coordinates is constructed to define the boundary and extract the equivalent ellipsoid. The final effective target area, target volume, target center, and equivalent ellipsoid are combined and packaged to obtain a quantitative three-dimensional reconstruction result set, which is then merged and output.

Images