Urolithiasis medical image segmentation method based on regional connectivity

Abstract

This invention discloses a medical image segmentation method for urinary stones based on region connectivity, belonging to the field of medical image processing technology, to solve the problems of high-density bone interference and topological fragmentation in stone segmentation. First, a piecewise weighted mapping function is constructed to adaptively adjust the window width and perform anisotropic filtering on the image, generating a bone suppression feature map. Then, a maximum stable extremum region algorithm combined with anatomical priors is used to extract high-confidence seed points. A fast traversal algorithm is executed using a constructed gradient resistance field to detect geodesic temporal field abrupt changes to define boundaries, generating a simply connected mask. Finally, three-dimensional dense morphology and texture consistency features are fused for weighted discrimination to remove vascular calcification artifacts. This invention ensures the topological connectivity and high accuracy of the segmentation results.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to a method for medical image segmentation of urinary stones based on regional connectivity. Background Technology

[0002] Urinary tract stones are a common disease in urology. For patients requiring surgical intervention, accurate preoperative assessment of the stone's volume, shape, and location is crucial for developing a surgical plan. Currently, non-contrast computed tomography (CT) is the preferred imaging method for diagnosing urinary tract stones. However, in clinical practice, doctors primarily rely on visual observation of two-dimensional tomographic slices to estimate the stone burden. This method struggles to visually construct the three-dimensional spatial distribution of the stones and cannot quantify their specific morphological parameters. Therefore, utilizing computer vision technology to automate the segmentation of urinary tract stones has significant clinical value in assisting doctors in planning puncture pathways and assessing risks during percutaneous nephrolithotomy or ureteroscopic lithotripsy.

[0003] Existing methods for urinary calculi segmentation suffer from the following substantial drawbacks: Traditional methods based on grayscale thresholding or region growing heavily rely on image contrast. Because the spine, pelvis, and other skeletal structures in abdominal images have a high CT value overlap with calculi, and noise and artifacts generated during scanning disrupt image edges, algorithms struggle to distinguish between high-density skeletal backgrounds and calculi, leading to frequent missegmentation. Furthermore, existing deep learning methods based on pixel-level classification often neglect the overall topological structure and regional connectivity constraints of the target. When dealing with small or poorly defined calculi, they tend to predict a complete calculus as multiple discrete fragments or generate isolated false positives in non-calculi areas. This lack of spatial continuity in segmentation results in significant errors in subsequent calculations of calculi volume and maximum diameter, failing to meet clinical requirements for surgical precision. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a medical image segmentation method for urinary stones based on regional connectivity, thus solving the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a medical image segmentation method for urinary calculi based on regional connectivity, comprising the following steps: S1. Acquiring the original abdominal CT image sequence, calculating the global CT value histogram and local sliding window statistical features, analyzing the distribution positions of bone high-density peaks and soft tissue low-density peaks within the local sliding window, constructing a piecewise weighted mapping function, applying the piecewise weighted mapping function to adaptively adjust the window width and window level and perform anisotropic diffusion filtering on the original abdominal CT image sequence, generating a bone suppression-calculi enhancement feature map; S2. Inputting the bone suppression-calculi enhancement feature map into the maximum stable extreme value region algorithm to detect gray-level consistency regions, combining the preset anatomical size range of calculi to remove areas exceeding the limit and noise spots, and extracting the set of calculi seed points located within the calculi; S3. Calculating the gradient of the bone suppression-calculi enhancement feature map. Amplitude: Construct a wavefront propagation resistance field based on gradient amplitude, set the stone seed point set as the wavefront evolution source point, execute the fast traversal algorithm to calculate the geodesic arrival time field from the pixel point in the image domain to the wavefront evolution source point, detect the local maximum abrupt change position of the geodesic arrival time field, define the local maximum abrupt change position as the stone boundary, and extract the candidate stone three-dimensional mask with a simply connected topology based on the stone boundary; S4. Map the candidate stone three-dimensional mask to three-dimensional space to calculate sphericity and compactness morphological features, remove the cancellous bone region with compactness below the preset threshold, extract the candidate stone three-dimensional mask coverage area on the original abdominal CT image sequence, calculate the gray-level co-occurrence matrix and extract texture consistency features, perform weighted fusion analysis on the compactness morphological features and texture consistency features, distinguish isotropic stone texture and anisotropic vascular calcification texture, and output the urinary stone segmentation result.

[0006] Further, the following process was used to acquire the original abdominal CT image sequence, calculate the global CT value histogram and local sliding window statistical features, analyze the distribution positions of the high-density peaks of bone and the low-density peaks of soft tissue within the local sliding window, and construct a piecewise weighted mapping function: The process involved traversing the pixels of the original abdominal CT image sequence, statistically analyzing the global gray-level distribution frequency, and extracting the pixel gray-level probability density curves within the local sliding window; identifying the first peak corresponding to the soft tissue matrix and the second peak corresponding to the bone structure in the gray-level probability density curves, calculating the right inflection point of the first peak as the low-threshold cutoff point, and calculating the left inflection point of the second peak as the high-threshold cutoff point; establishing a linear stretching interval between the low-threshold cutoff point and the high-threshold cutoff point, establishing a zero-weight suppression interval below the low-threshold cutoff point, and establishing a logarithmic decay interval above the high-threshold cutoff point; and combining the linear stretching interval, the zero-weight suppression interval, and the logarithmic decay interval to construct a nonlinear piecewise weighted mapping function.

[0007] Furthermore, the specific process of applying a piecewise weighted mapping function to adaptively adjust the window width and level and perform anisotropic diffusion filtering on the original abdominal CT image sequence to generate a bone suppression-stone enhancement feature map is as follows: Input the pixel gray values ​​of the original abdominal CT image sequence to a nonlinear piecewise weighted mapping function, output the mapped gray values, calculate the gradient magnitude vector of the mapped gray values ​​in the three-dimensional spatial neighborhood, construct an exponential diffusion transmission coefficient function with the gradient magnitude vector as the independent variable; set a constraint relationship that high gradient regions correspond to low diffusion transmission coefficients and low gradient regions correspond to high diffusion transmission coefficients, perform iterative smoothing updates on the mapped gray values ​​through the exponential diffusion transmission coefficient function, retain high gradient edge information and smooth low gradient internal textures, and output the bone suppression-stone enhancement feature map.

[0008] Further, step S2 includes the following steps: setting a grayscale threshold sequence, performing binarization processing on the bone inhibition-stone enhancement feature map step by step according to the grayscale threshold sequence, and extracting connected components at different grayscale threshold levels; tracking the area change rate of connected components between adjacent grayscale threshold levels, and selecting regions with an area change rate lower than a preset stable threshold as candidate subsets of the maximum stable extreme value region; calculating the aspect ratio and solidity of the three-dimensional bounding rectangle of the candidate subsets of the maximum stable extreme value region, comparing the aspect ratio of the three-dimensional bounding rectangle with the upper limit of the human spine anatomical size, comparing the solidity with the lower limit of vascular calcification morphology, eliminating interference regions that meet the upper limit of the spine anatomical size or are lower than the lower limit of vascular calcification morphology, and retaining the centroids of regions that conform to the spatial distribution characteristics of urinary system stones to form a set of stone seed points.

[0009] Furthermore, the gradient magnitude of the bone suppression-stone enhancement feature map is calculated, and a wavefront propagation drag field is constructed based on the gradient magnitude. The set of stone seed points is set as the wavefront evolution source point. The specific process of executing the fast travel algorithm to calculate the geodesic arrival time field from the pixel point in the image domain to the wavefront evolution source point is as follows: The bone suppression-stone enhancement feature map is convolved by a three-dimensional difference operator to calculate the gradient components of the pixel points in the spatial coordinate axis direction, and the gradient magnitude matrix is ​​synthesized; a wavefront propagation velocity function inversely proportional to the gradient magnitude matrix is ​​constructed, and the reciprocal of the wavefront propagation velocity function is taken to generate a high gradient. Regions with high resistance and low gradients correspond to wavefront propagation resistance fields with low resistance. The set of stone seed points is initialized as frozen nodes at time zero, and the remaining pixels in the image domain are initialized as far-end nodes at time infinity. A narrowband min-heap structure is established to maintain the wavefront extension edge. Following the principle of minimum arrival time priority in the narrowband min-heap structure, the discretized Eikonal equation is solved iteratively. The arrival time of the geodesics of the neighboring narrowband nodes is updated using the known frozen nodes until the calculation of the arrival time of the geodesics of the pixels in the preset calculation neighborhood centered on the set of stone seed points is completed.

[0010] Furthermore, the process of detecting the abrupt changes in local maxima of the geodesic arrival time field and defining these abrupt changes as the stone boundary, and then extracting a candidate stone 3D mask with a simply connected topology based on the stone boundary, is as follows: The first derivative of the geodesic arrival time field is calculated to generate a time gradient field reflecting the rate of change of wavefront propagation velocity. A modulus jump ridge is searched in the time gradient field to locate the stagnation point of wavefront propagation, and the position of the modulus jump ridge is marked as the energy propagation stagnation boundary. Region growth is performed outward from the stone seed point set until the energy propagation stagnation boundary is reached, at which point filling stops and the contour is closed, generating an initial segmented region. A 3D morphological closing operation is performed on the initial segmented region to fill internal micro-holes and smooth edge burrs. The voxel topological connectivity within the region is confirmed, and the candidate stone 3D mask is output.

[0011] Furthermore, the sphericity and density morphological features of the candidate stone 3D mask are calculated in 3D space, and cancellous bone regions with density below a preset threshold are removed. The specific process of cropping the candidate stone 3D mask coverage area from the original abdominal CT image sequence, calculating the gray-level co-occurrence matrix, and extracting texture consistency features is as follows: The total number of voxels inside the candidate stone 3D mask is counted to calculate the volume; the surface area of ​​the candidate stone 3D mask is calculated by extracting triangular meshes; the sphericity is calculated based on the geometric ratio of volume to surface area; the spatial distribution covariance matrix of the candidate stone 3D mask is calculated, and the... The covariance matrix eigenvalues ​​are solved, and the density morphology features are calculated based on the ratio of the minimum to the maximum eigenvalues. A lower limit threshold for cancellous bone density is set, and non-stone regions with density morphology features below the lower limit threshold are eliminated. The gray-level co-occurrence matrix is ​​defined to calculate the step size and direction parameters. The projection area of ​​the candidate stone 3D mask on the original abdominal CT image sequence is traversed, and the joint probability distribution of pixel gray levels is statistically analyzed. The second moment of the angle is extracted from the main diagonal as the energy feature. The entropy feature is calculated by the logarithm of the probability distribution. The energy feature and the entropy feature are combined to form the texture consistency feature.

[0012] Furthermore, the specific process of performing weighted fusion analysis on dense morphological features and texture consistency features to distinguish between isotropic stone textures and anisotropic vascular calcification textures, and outputting the urinary stone segmentation results is as follows: Range normalization is performed on dense morphological features and texture consistency features to map feature values ​​to a unified dimensionless interval. Morphological feature weight coefficients and texture feature weight coefficients are set. Linear weighted summation is performed on the normalized dense morphological features and texture consistency features to generate a stone probability score. A true positive threshold is set, and the stone probability score is compared with the true positive threshold. Regions with stone probability scores higher than the true positive threshold are retained and identified as isotropic stone texture regions. Regions with stone probability scores lower than the true positive threshold are removed and identified as anisotropic vascular calcification texture regions. The urinary stone segmentation results are then output.

[0013] The present invention has the following beneficial effects:

[0014] (1) A medical image segmentation method for urinary stones based on regional connectivity solves the problem of inaccurate localization caused by skeletal background interference and noise through adaptive enhancement based on scene feature perception and extraction of high-confidence seed points. Specifically, this invention first constructs a piecewise weighted mapping function by calculating the global histogram and local window statistical features, and adaptively adjusts the window width and window level for the image features of different patients and performs anisotropic diffusion filtering. This process effectively smooths the internal texture and suppresses the high-brightness interference of the bones while preserving the sharpness of the stone edges, providing a high-contrast feature map for subsequent processing. On this basis, combined with the maximum stable extremum region algorithm and anatomical size prior, it can accurately locate the region located only inside the stone as a seed point in the complex abdominal background. This strategy of enhancement before localization avoids a large number of false positive noise points introduced by the traditional method due to full-image search, and provides reliable physical anchors for subsequent accurate segmentation.

[0015] (2) The medical image segmentation method for urinary stones based on regional connectivity solves the problems of topological breakage and easily confused calcification artifacts in the stone segmentation results of the prior art by using wavefront propagation evolution and multidimensional feature joint verification. Specifically, this invention constructs a wavefront propagation resistance field based on gradient amplitude and uses the fast travel algorithm to calculate the geodesic arrival time field, and uses the physical property of energy propagation encountering abrupt changes at the edge of the stone to define the boundary. This method forces the segmentation results to have a simply connected topological structure, effectively preventing the segmentation region from being broken or fractured due to edge blurring. Furthermore, by calculating the three-dimensional morphological density and gray-level co-occurrence matrix texture features of the candidate region and performing weighted fusion analysis, it is possible to effectively distinguish isotropic true stones from anisotropic vascular calcification or cancellous bone artifacts from two dimensions: spatial structure and material texture, thereby achieving high-precision and high-specificity urinary stone segmentation.

[0016] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0017] Figure 1 This is a flowchart of the medical image segmentation method for urinary stones based on regional connectivity according to the present invention. Detailed Implementation

[0018] This application's embodiments address the technical problems of existing technologies, such as difficulty in distinguishing stones against complex skeletal backgrounds and image fragmentation due to lack of topological connectivity in segmentation results, through a medical image segmentation method for urinary stones based on regional connectivity.

[0019] The overall approach of the scheme in this application embodiment is as follows: First, a piecewise weighted mapping function is constructed using local statistical features to adaptively enhance the original CT image, thereby increasing the contrast between the stone and the bone background. Second, a maximum stable extremum region algorithm is used to extract high-confidence seed points inside the stone on the enhanced feature map. Next, using the seed point as the source point, a fast traversal algorithm is executed based on the gradient resistance field. The stone boundary is determined by detecting gradient abrupt changes in the geodesic arrival time field, generating a candidate mask with simple connectivity. Finally, the candidate mask is distinguished as genuine or fake by combining the compactness features of three-dimensional morphology and the consistency features of two-dimensional texture, eliminating interference such as vascular calcification, and outputting the final segmentation result.

[0020] Please see Figure 1 This invention provides a technical solution: a medical image segmentation method for urinary calculi based on regional connectivity, comprising the following steps: S1. Acquiring the original abdominal CT image sequence, calculating the global CT value histogram and local sliding window statistical features, analyzing the distribution positions of bone high-density peaks and soft tissue low-density peaks within the local sliding window, constructing a piecewise weighted mapping function, applying the piecewise weighted mapping function to the original abdominal CT image sequence for adaptive window width and window level adjustment and anisotropic diffusion filtering, generating a bone suppression-calculi enhancement feature map; S2. Inputting the bone suppression-calculi enhancement feature map into the maximum stable extreme value region algorithm to detect gray-level consistency regions, combining a preset anatomical size range for calculi to remove areas exceeding the limit and noise spots, and extracting a set of calculi seed points located within the calculi; S3. Calculating the gradient magnitude of the bone suppression-calculi enhancement feature map, based on... A wavefront propagation resistance field is constructed based on the gradient amplitude. The set of stone seed points is set as the wavefront evolution source point. The fast traversal algorithm is executed to calculate the geodesic arrival time field from the pixel point in the image domain to the wavefront evolution source point. The local maximum abrupt change position of the geodesic arrival time field is detected and defined as the stone boundary. A candidate stone 3D mask with a simply connected topology is extracted based on the stone boundary. S4. The candidate stone 3D mask is mapped to 3D space to calculate the sphericity and compactness morphological features. The cancellous bone region with a compactness lower than the preset threshold is removed. The area covered by the candidate stone 3D mask is extracted from the original abdominal CT image sequence. The gray-level co-occurrence matrix is ​​calculated and the texture consistency feature is extracted. The compactness morphological feature and the texture consistency feature are weighted and fused to distinguish between isotropic stone texture and anisotropic vascular calcification texture, and the urinary stone segmentation result is output.

[0021] In this implementation scheme, step S1 is mainly used for medical image preprocessing and feature enhancement. By calculating global and local sliding window statistical features, the drift of CT values ​​under different scanning environments can be perceived, and a piecewise weighted mapping function is constructed to achieve adaptive contrast stretching for the stone target. Among them, anisotropic diffusion filtering is a nonlinear smoothing technique. Its core mechanism is to dynamically adjust the smoothing degree according to the changes in image gradient. Strong smoothing is performed in flat areas to remove noise, while smoothing is stopped in edge areas with drastic gradient changes to preserve details. Combined with this filtering technique, the interference of high-density bone background can be effectively suppressed, while the edge sharpness of small stones can be enhanced, providing high-quality feature maps for subsequent accurate segmentation. Step S2 is mainly used for the localization of the stone target and the extraction of initial seed points. This step uses the maximum stable extremum region algorithm to search for regions with gray-level consistency on the bone suppression-stone enhancement feature map. The maximum stable extremum region refers to a connected region whose shape and area remain relatively stable under a certain range of gray-level threshold changes. Such regions usually correspond to anatomical structures with physical entities. By combining a preset range of anatomical sizes for stones, the system can automatically remove excessively large spinal bone interference and excessively small random noise spots, thereby accurately extracting the set of pixels whose spatial location is strictly within the stone as the seed point set, solving the problem of traditional methods struggling to automatically initialize in complex backgrounds. Step S3 is mainly used for accurate contour evolution and segmentation based on regional connectivity. In this process, the fast travel algorithm is used to simulate the physical process of wavefront diffusion from the seed point to the surrounding area, while the wavefront propagation resistance field is a virtual medium constructed based on the image gradient magnitude; the larger the image gradient, the greater the resistance. The geodesic arrival time field reflects the minimum cumulative cost required for the wavefront to propagate from the source point to any point in the image, and can reflect the topological structure of the image content. By detecting the abrupt change locations of local maxima in the arrival time field, i.e., the boundaries where wavefront propagation is blocked and stagnant, the physical contour of the stone can be naturally defined. This step ensures that the extracted candidate stone 3D mask is topologically simply connected, avoiding the fragmented or discrete segmentation results common in deep learning methods. Step S4 is mainly used for the identification of the authenticity of candidate targets and the output of the final segmentation result. By mapping a two-dimensional mask back to three-dimensional space, sphericity and density morphological features are calculated. Density reflects the compactness of the material distribution within the target, used to identify and remove cancellous bone regions with a sponge-like porous structure. Simultaneously, texture consistency features are extracted using a gray-level co-occurrence matrix (GLCM), a method that describes texture by statistically analyzing the spatial correlation of pixel gray levels. Through weighted fusion analysis, this step effectively distinguishes true stones with isotropic texture features from vascular calcification artifacts with anisotropic or linear texture features, thus outputting high-precision, high-specificity urinary stone segmentation results.

[0022] Specifically, the process of acquiring the original abdominal CT image sequence, calculating the global CT value histogram and local sliding window statistical features, analyzing the distribution positions of high-density peaks in the bone and low-density peaks in the soft tissue within the local sliding window, and constructing a piecewise weighted mapping function is as follows: Traverse the pixels of the original abdominal CT image sequence, statistically analyze the global gray-level distribution frequency, and extract the pixel gray-level probability density curve within the local sliding window; identify the first peak corresponding to the soft tissue matrix and the second peak corresponding to the bone structure in the gray-level probability density curve, calculate the right inflection point of the first peak as the low-threshold cutoff point, and calculate the left inflection point of the second peak as the high-threshold cutoff point; establish a linear stretching interval between the low-threshold cutoff point and the high-threshold cutoff point, establish a zero-weight suppression interval below the low-threshold cutoff point, and establish a logarithmic decay interval above the high-threshold cutoff point; combine the linear stretching interval, the zero-weight suppression interval, and the logarithmic decay interval to construct a nonlinear piecewise weighted mapping function.

[0023] In this implementation scheme, the core objective of this step is to automatically identify and enhance the region of interest (ROI) for stones from a complex abdominal anatomical background by utilizing the statistical distribution characteristics of image data, while suppressing soft tissue background and bright bone interference. First, a probability density curve is constructed within a local sliding window by traversing image pixels. This curve reflects the gray-level aggregation of different tissues within the local area. After identifying the first peak representing the soft tissue matrix and the second peak representing the bone structure, the inflection points of these two peaks are used as the benchmark for adaptive segmentation, enabling standardized gray-level interval division across different patients or scanning devices. Based on this, a nonlinear piecewise weighted mapping function is constructed to implement differentiated enhancement strategies for different intervals: below the low threshold cutoff point, the gray-level interval is considered useless soft tissue or background noise and is suppressed; between the two cutoff points, the gray-level interval is considered where stones may exist and is linearly stretched to increase contrast; above the high threshold cutoff point, the gray-level interval is considered dense bone and logarithmic attenuation is used to prevent overexposure and halo artifacts. The nonlinear piecewise weighted mapping function can be constructed and calculated using the following formula: The meanings of the parameters in the above formula are explained as follows: : Represents the mapped pixel grayscale output value; : Represents the grayscale value of the input pixel in the original abdominal CT image sequence; : Indicates the low threshold cutoff point corresponding to the inflection point on the right side of the first peak, used to distinguish soft tissue from potential stone areas; : Indicates the high threshold cutoff point corresponding to the inflection point on the left side of the second peak, used to distinguish potential stone areas from dense bone; This represents the gain coefficient for the linear stretching range, and its value can be set to 255 to map to the standard grayscale space. The compression factor represents the logarithmic decay interval and is used to control the degree of brightness suppression in the bright areas. : Represents the natural constant, used to ensure the validity of the domain of the logarithmic function; This represents the brightness compensation constant at the high threshold, ensuring the continuity of the mapping function at the segmentation points.

[0024] Specifically, the process of applying a piecewise weighted mapping function to adaptively adjust the window width and level and perform anisotropic diffusion filtering on the original abdominal CT image sequence to generate a bone suppression-stone enhancement feature map is as follows: Input the pixel gray values ​​of the original abdominal CT image sequence to a nonlinear piecewise weighted mapping function, output the mapped gray values, calculate the gradient magnitude vector of the mapped gray values ​​in the three-dimensional spatial neighborhood, construct an exponential diffusion transmission coefficient function with the gradient magnitude vector as the independent variable; set a constraint relationship between high gradient regions and low diffusion transmission coefficients, and perform iterative smoothing updates on the mapped gray values ​​through the exponential diffusion transmission coefficient function, retaining high gradient edge information and smoothing low gradient internal textures, and outputting a bone suppression-stone enhancement feature map.

[0025] In this implementation, the mapped grayscale value is used as input, and the spatial texture of the image is further processed using anisotropic diffusion filtering technology. Traditional linear filtering (such as Gaussian filtering) blurs edges while denoising, while the anisotropic diffusion strategy used in this embodiment focuses on constructing a diffusion coefficient that depends on the local gradient of the image. After calculating the gradient magnitude vector in the three-dimensional spatial neighborhood, this vector reflects whether the current pixel is in a flat region or an edge region. By constructing an exponential diffusion conduction coefficient function, an inverse constraint relationship is established: when the gradient magnitude is large (corresponding to the edge of a stone or the boundary of a bone), the diffusion conduction coefficient approaches zero, thereby blocking heat flow diffusion and protecting the edges from being blurred; when the gradient magnitude is small (corresponding to the inside of a stone or uniform soft tissue), the diffusion conduction coefficient is large, allowing heat flow to diffuse smoothly, thereby removing noise. This process is achieved through iterative smooth updates, and the specific iterative calculation formula is shown below: The meanings of the parameters in the above formula are explained as follows: : indicates the first After the next iteration update, the pixels The grayscale value, i.e. the pixel value in the final output bone suppression-stone enhancement feature map; Indicates the first In the next iteration, the pixel The initial state is the mapped gray value output by the nonlinear piecewise weighted mapping function; This indicates the time step of the iteration, used to control the magnitude of each smooth update; q represents the three-dimensional neighborhood set of pixel p; q represents the neighborhood set. Adjacent pixels; : Represents the gradient magnitude vector between pixel p and its neighboring pixel q at the k-th iteration; K: Represents the gradient sensitivity threshold parameter, which determines at what gradient intensity the diffusion process stops. This parameter can be the median of the gradient of the mapped image. : Represents an exponential function used to construct the diffusion coefficient, which is inversely proportional to the gradient intensity. Through the above iterative calculation, it is possible to effectively smooth the internal texture and eliminate quantum noise caused by CT scans while preserving the sharp edge features of the stone.

[0026] Specifically, step S2 includes the following steps: setting a grayscale threshold sequence, performing binarization processing on the bone inhibition-stone enhancement feature map step by step according to the grayscale threshold sequence, and extracting connected components at different grayscale threshold levels; tracking the area change rate of connected components between adjacent grayscale threshold levels, and selecting regions with an area change rate lower than a preset stable threshold as candidate subsets of the maximum stable extreme value region; calculating the aspect ratio and solidity of the three-dimensional bounding rectangle of the candidate subsets of the maximum stable extreme value region, comparing the aspect ratio of the three-dimensional bounding rectangle with the upper limit of the human spine anatomical size, comparing the solidity with the lower limit of vascular calcification morphology, eliminating interference regions that meet the upper limit of the spine anatomical size or are lower than the lower limit of vascular calcification morphology, and retaining the centroids of regions that conform to the spatial distribution characteristics of urinary system stones to form a set of stone seed points.

[0027] In this implementation scheme, specifically, step S2 aims to accurately screen physically stable stone targets from a complex anatomical background, excluding noise and non-stone tissue. First, the system sets a series of incremental grayscale threshold sequences, equivalent to continuously raising the water level in a grayscale topographic map. The bone suppression-stone enhancement feature map is binarized and truncated level by level according to this sequence. At each threshold level, pixels with grayscale values ​​higher than the threshold are clustered into connected components. To determine whether these connected components represent real anatomical structures rather than random noise, their evolutionary behavior between adjacent threshold levels needs to be tracked. Real stones, due to their high density and relatively clear boundaries, maintain a relatively stable connected component area within a certain threshold range, while noise spots or soft tissue with blurred edges exhibit drastic morphological fluctuations with threshold changes. To quantify this stability, the area change rate of the connected components needs to be calculated, and a candidate subset satisfying the maximum stable extreme value region condition is selected. The stability judgment is based on the following formula: The meanings of the parameters in the above formula are explained as follows: : Indicates the current grayscale threshold level Down-connected components The stability metric is such that a smaller value indicates a more stable region. : Indicates a grayscale threshold level of The connected component regions extracted at that time; This refers to an operator that calculates the total number of pixels or volume contained in a connected component. This indicates the index of the grayscale threshold level currently being traversed; : Indicates the grayscale threshold step size used for stability assessment, which determines the sensitivity of stability detection; : Indicates a higher grayscale threshold level The corresponding connected component region; Indicates a lower grayscale threshold level The corresponding connected component region. By setting a preset stability threshold, only the connected component region is retained. Regions with values ​​below this threshold are selected as candidate subsets of the maximum stable extreme value region, effectively filtering out unstable background noise. After obtaining the candidate subset, to further eliminate interfering tissues with similar calcification but different morphology (such as slender vertebral bone structures or irregularly shaped vascular calcifications), it is necessary to calculate the three-dimensional geometric features of the candidate regions. Stones typically exhibit spherical or ellipsoidal shapes, while vertebral bones are elongated, and vascular calcifications often present as discrete structures with low solidity. Therefore, morphological screening is performed by calculating the aspect ratio and solidity of the three-dimensional bounding rectangle. The specific feature calculation formula is as follows: The meanings of the parameters in the above formula are explained as follows: The aspect ratio of the three-dimensional bounding rectangle representing the candidate region is used to measure the elongation of the target in space; This represents the maximum projection span of the candidate region along the X-axis in three-dimensional space. This represents the maximum projection span of the candidate region along the Y-axis in three-dimensional space; This represents the maximum projection span of the candidate region along the Z-axis in three-dimensional space. This represents the function that takes the maximum value. This represents a function that takes the minimum value. It indicates the solidity of the candidate region, used to measure the fullness of the target's interior and the irregularity of its edges; This represents the total volume of voxels actually contained within the candidate region. This represents the volume of the 3D convex hull corresponding to the candidate region. Based on the above calculation results, the calculated aspect ratio is compared with the upper limit of the anatomical dimensions of the human spine. If the aspect ratio is too large, it is identified as the spine or ribs and discarded. The solidity is compared with the lower limit of the vascular calcification morphology. If the solidity is too small, it is identified as loose calcification spots and discarded. Finally, only regions whose features all conform to the spatial distribution characteristics of urinary system stones are retained, and the geometric centroids of these regions are extracted to form a set of stone seed points, providing a reliable localization benchmark for subsequent accurate segmentation.

[0028] Specifically, the gradient magnitude of the bone suppression-stone enhancement feature map is calculated, a wavefront propagation drag field is constructed based on the gradient magnitude, and the set of stone seed points is set as the wavefront evolution source point. The specific process of calculating the geodesic arrival time field from the pixel point in the image domain to the wavefront evolution source point using the fast travel algorithm is as follows: The bone suppression-stone enhancement feature map is convolved by a three-dimensional difference operator to calculate the gradient components of the pixel points in the spatial coordinate axis directions, and the gradient magnitude matrix is ​​synthesized; a wavefront propagation velocity function inversely proportional to the gradient magnitude matrix is ​​constructed, and the reciprocal of the wavefront propagation velocity function is taken to generate a high gradient. Regions with high resistance and low gradients correspond to wavefront propagation resistance fields with low resistance. The set of stone seed points is initialized as frozen nodes at time zero, and the remaining pixels in the image domain are initialized as far-end nodes at time infinity. A narrowband min-heap structure is established to maintain the wavefront extension edge. Following the principle of minimum arrival time priority in the narrowband min-heap structure, the discretized Eikonal equation is solved iteratively. The arrival time of the geodesics of the neighboring narrowband nodes is updated using the known frozen nodes until the calculation of the arrival time of the geodesics of the pixels in the preset calculation neighborhood centered on the set of stone seed points is completed.

[0029] In this implementation scheme, the core of this step lies in establishing a wavefront propagation model based on physical optics, transforming the image segmentation problem into a shortest path finding problem on a Riemannian manifold. First, a three-dimensional difference operator is used to convolve the bone suppression-stone enhancement feature map to extract the three-dimensional spatial gradient information of the image. Positions with larger gradient magnitudes represent more significant edges (i.e., stone boundaries), while those with smaller gradients represent flat interior regions. To control the wavefront propagation behavior, a wavefront propagation velocity function inversely proportional to the gradient magnitude is constructed. This means that the wavefront propagates quickly inside the stone, but its velocity drops sharply when encountering high-gradient regions at the stone's edge, thus forming a high-resistance barrier. Subsequently, a fast-margin algorithm is used to solve the Eikonal equation. This algorithm utilizes narrowband technology and a min-heap data structure to efficiently update grid nodes in ascending order of arrival time, simulating the wavefront's outward diffusion from the seed point. To accurately describe this physical propagation process, the geodesic arrival time of each pixel is solved using the following discretized equation: The meanings of the parameters in the above formula are explained as follows: The cumulative cost required to propagate from the source to that point; The dimension indices represent the three-dimensional space, corresponding to the x, y, and z coordinate axes, respectively. Indicates the first Each dimension and pixel The geodesic arrival times of adjacent pixels with known arrival times are usually selected in the calculation to satisfy the upwind pattern condition. This represents the spatial step size of the image grid, which is usually set to the physical spacing between pixels; : Represents a pixel The wavefront propagation velocity function value at a given location is constructed from the gradient magnitude of the feature map, for example... ,in Let p be the gradient magnitude of the feature map at point p. This is a constant used to control the sensitivity to velocity decay. By solving this equation, the geodesic arrival time distribution across the entire field is obtained, which contains the geometric and topological information of the stones.

[0030] Specifically, the process of detecting the abrupt changes in local maxima of geodesics arriving at the time field, defining these abrupt changes as the stone boundary, and extracting candidate 3D masks with a simply connected topology based on the stone boundary is as follows: The first derivative of the geodesics arriving at the time field is calculated to generate a time gradient field reflecting the rate of change of wavefront propagation velocity. Modulus jump ridges are searched in the time gradient field to locate the points where wavefront propagation is blocked or stalled. The locations of these modulus jump ridges are marked as energy propagation hindrance boundaries. Region growth is performed outward from the stone seed point set until the energy propagation hindrance boundary is reached. Filling is stopped and the contour is closed to generate the initial segmented region. A 3D morphological closing operation is performed on the initial segmented region to fill internal micro-holes and smooth edge burrs. The voxel topological connectivity within the region is confirmed, and the candidate 3D mask is output.

[0031] In this implementation, this step aims to accurately resolve discrete anatomical boundaries from a continuous arrival time field. Because the wavefront moves extremely slowly through the high-resistance region at the edge of the stone, the arrival time value experiences a sharp jump at this point. Therefore, the first derivative of the arrival time field (i.e., the time gradient) forms a ridge-like local maximum at the boundary. The system calculates the time gradient field and searches for its modulus jump ridges. These ridges physically correspond to interfaces where energy propagation is impeded, i.e., the true physical boundary of the stone. Subsequently, confined region growth, i.e., a flooding process, is performed centered on the seed point. Growth stops when the growth front touches the aforementioned impediment boundary, thus naturally forming a closed initial region. Finally, to ensure the integrity of the medical anatomical structure, eliminate potential micropores due to uneven internal texture, and smooth the burr noise at the segmentation edges, a three-dimensional morphological closing operation is performed to ensure that the output mask is a simply connected solid in topological space. The mathematical expressions for the above morphological correction and mask generation are shown below: The meanings of the parameters in the above formula are explained as follows: : Represents the final output 3D mask of candidate stones with a simply connected topology; : Represents the initial segmented region generated after region growth and truncation at the energy propagation blocking boundary; This represents a morphological dilation operator used to fill tiny gaps inside a target and connect broken edges; This represents a morphological erosion operator used to restore the original size of a target after expansion and smooth burrs on the surface of an object. Indicates radius as A three-dimensional spherical structuring element that defines the scope of morphological operations, its radius... The value can be set according to the spatial resolution of the image, typically 1 to 2 voxels. This operation forces a constraint on the topological connectivity of the segmentation results, preventing unnatural discrete fragments from appearing in the stone region.

[0032] Specifically, the process involves mapping the candidate stone's 3D mask onto 3D space to calculate sphericity and density morphological features, removing cancellous bone regions with density below a preset threshold, cropping the candidate stone's 3D mask-covered area from the original abdominal CT image sequence, calculating the gray-level co-occurrence matrix, and extracting texture consistency features. The specific steps are as follows: 1) Calculate the volume by counting the total number of voxels within the candidate stone's 3D mask; 2) Extract the triangular mesh from the candidate stone's 3D mask surface to calculate the surface area; 3) Calculate the sphericity based on the geometric ratio of volume to surface area; 4) Calculate the spatial distribution covariance matrix of the candidate stone's 3D mask; and 5) Solve for... The covariance matrix eigenvalues ​​are solved, and the density morphology features are calculated based on the ratio of the minimum to the maximum eigenvalues. A lower limit threshold for cancellous bone density is set, and non-stone regions with density morphology features below the lower limit threshold are eliminated. The gray-level co-occurrence matrix is ​​defined to calculate the step size and direction parameters. The projection area of ​​the candidate stone 3D mask on the original abdominal CT image sequence is traversed, and the joint probability distribution of pixel gray levels is statistically analyzed. The second moment of the angle is extracted from the main diagonal as the energy feature. The entropy feature is calculated by the logarithm of the probability distribution. The energy feature and the entropy feature are combined to form the texture consistency feature.

[0033] In this implementation scheme, this step aims to verify the authenticity of candidate targets from two levels: geometric morphology and internal texture, through multi-dimensional feature quantification. First, regarding geometric morphology, considering that urinary stones typically present a spherical or ellipsoidal physical appearance, while bone spurs or irregular calcifications often have rough surfaces or flat structures, a dimensionless index of sphericity is introduced. This index, based on the isoperimetric theorem, measures the degree to which an object approximates an ideal sphere by calculating the ratio of its volume to its surface area. Subsequently, to assess the compactness of the target's distribution in three-dimensional space, the covariance matrix of the voxel coordinates of the candidate region is calculated, and its eigenvalues ​​are solved. The ratio of these eigenvalues ​​is used to define compactness. If the eigenvalues ​​in the three principal directions differ significantly, the object is elongated (like a blood vessel); otherwise, it is blocky. Based on this, a lower threshold for cancellous bone compactness is set, effectively eliminating non-stone tissues that are connected but loosely structured. The calculation formulas for the above geometric morphological features are shown below: The meanings of the parameters in the above formula are explained as follows: : Represents the calculated sphericity characteristic value, which usually ranges from 0 to 1. The closer the value is to 1, the closer the target is to a sphere. : Represents the physical volume obtained by converting the total number of voxels inside the three-dimensional mask of candidate stones; : Represents the surface area calculated after extracting the triangular mesh on the mask surface using the moving cube algorithm; : Represents the calculated compactness morphological characteristics, used to distinguish between blocky structures and linear or sheet-like structures; : Represents the smallest eigenvalue of the spatial distribution covariance matrix of the candidate region, and represents the dispersion of the target in the narrowest direction of the distribution; : Represents the largest eigenvalue of the spatial distribution covariance matrix of the candidate region, representing the dispersion of the target along the principal axis. After completing the initial morphological screening, the gray-level co-occurrence matrix is ​​further used to analyze the texture patterns inside the target. The gray-level co-occurrence matrix describes the joint probability distribution of the simultaneous occurrence of gray values ​​of two pixels at a certain distance in an image. Stones are usually hard and uniform in texture, exhibiting high energy and low entropy; while cancellous bone or vascular calcification, due to the presence of bone trabeculae or blood flow artifacts, often have messy textures, exhibiting low energy and high entropy. The formula for calculating the texture consistency feature based on the gray-level co-occurrence matrix is ​​as follows: The meanings of the parameters in the above formula are explained as follows: : Represents the second moment of the angle, i.e., the texture energy feature, which reflects the uniformity of the gray-level distribution and the coarseness of the texture in the image; : Represents texture entropy features, reflecting the complexity or disorder of texture information in an image. Represents the row index of the gray-level co-occurrence matrix, corresponding to the gray level of the reference pixel; Represents the column index of the gray-level co-occurrence matrix, corresponding to the gray level of the neighboring pixels; : represents the th element in the normalized gray-level co-occurrence matrix. Line 1 The element values ​​of a column represent the probability of a pixel pair with a specific grayscale combination appearing. By combining high-energy and low-entropy features, it is possible to accurately capture the highly consistent crystalline structure texture inside the stone.

[0034] Specifically, the process of performing weighted fusion analysis on dense morphological features and texture consistency features to distinguish between isotropic stone textures and anisotropic vascular calcification textures, and outputting the urinary stone segmentation results is as follows: Range normalization is performed on the dense morphological features and texture consistency features to map the feature values ​​to a unified dimensionless interval. Morphological feature weight coefficients and texture feature weight coefficients are set. Linear weighted summation is performed on the normalized dense morphological features and texture consistency features to generate a stone probability score. A true positive threshold is set, and the stone probability score is compared with the true positive threshold. Regions with stone probability scores higher than the true positive threshold are retained and identified as isotropic stone texture regions. Regions with stone probability scores lower than the true positive threshold are removed and identified as anisotropic vascular calcification texture regions. The urinary stone segmentation results are then output.

[0035] In this implementation scheme, this step is the decision-making terminal of the entire algorithm, aiming to map the heterogeneous feature space into a unified probability space. Since the dense morphological features (between 0 and 1) and texture entropy features (logarithmic, with a large range) calculated in the preceding steps differ significantly in numerical magnitude and physical dimensions, direct fusion would lead to the larger feature dominating the decision. Therefore, range normalization must be performed first to map all features to a unified dimensionless interval. Subsequently, a linear weighted summation strategy is used to generate the final stone probability score, which integrates the double-reset confidence of the target stone's shape (morphological item) and texture (texture item). Finally, by setting a true positive threshold as the boundary for binary classification, only regions with scores higher than this threshold are retained, thereby achieving the final elimination of anisotropic vascular calcifications (usually with low scores). The mathematical model of this weighted fusion and decision-making process is expressed as follows: The meanings of the parameters in the above formula are explained as follows: : Represents the final generated stone probability score, used to quantify the confidence level that the candidate region is a true positive stone; : This represents the set morphological feature weighting coefficient. Based on anatomical experience, this value can be increased if more emphasis is placed on the regularity of the shape. : Represents the compactness morphological characteristic value of the current region obtained from the previous steps; , : These represent the statistical minimum and maximum values ​​of the compactness feature in the sample set, respectively, used for normalization boundary delineation; : Represents the set texture feature weight coefficient, used to adjust the proportion of internal texture features in the decision-making process; : Represents the comprehensive texture consistency feature value, which can be synthesized by weighted sum of the inverse of the energy feature and the entropy feature; , : These represent the extreme value boundaries of texture features in the statistical samples. By comparison... Compared with the preset true positive threshold This ensured the accuracy of the segmentation results in both morphological and pathological terms.

[0036] In summary, this application has at least the following effects:

[0037] A medical image segmentation method for urinary stones based on regional connectivity effectively overcomes the masking interference of high-density skeletal backgrounds by constructing a piecewise weighted mapping function to achieve adaptive enhancement and maximum stable extremum region localization. Utilizing a wavefront propagation evolution mechanism based on gradient resistance fields, it forcibly constrains the single-connected topology of the segmentation results in physical space, completely solving the problem of fragmented and discrete stone segmentation caused by edge blurring in existing deep learning methods. Furthermore, by combining the fusion and verification of three-dimensional dense morphology and texture consistency features, it accurately removes high-similarity artifacts such as vascular calcification, thus achieving high-precision, continuous, and automatic segmentation of small stones against complex abdominal anatomical backgrounds, significantly improving the accuracy and clinical practical value of preoperative diagnosis and treatment planning in urology.

[0038] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0039] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0040] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.

[0041] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0042] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0043] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for segmentation of urinary stone medical images based on region connectivity, characterized in that, Includes the following steps: S1. Obtain the original abdominal CT image sequence, calculate the global CT value histogram and local sliding window statistical features, analyze the distribution positions of bone high density peaks and soft tissue low density peaks within the local sliding window, construct a piecewise weighted mapping function, apply the piecewise weighted mapping function to the original abdominal CT image sequence for adaptive window width and window level adjustment and anisotropic diffusion filtering, and generate a bone inhibition-stone enhancement feature map. S2. Input the bone suppression-stone enhancement feature map into the maximum stable extreme value region algorithm to detect gray-level consistent regions, combine the preset stone anatomical size range to remove areas exceeding the limit and noise spots, and extract the set of stone seed points located inside the stone. S3. Calculate the gradient magnitude of the bone suppression-stone enhancement feature map, construct a wavefront propagation resistance field based on the gradient magnitude, set the stone seed point set as the wavefront evolution source point, execute the fast travel algorithm to calculate the geodesic arrival time field from the pixel point in the image domain to the wavefront evolution source point, detect the local maximum abrupt change position of the geodesic arrival time field, define the local maximum abrupt change position as the stone boundary, and extract the candidate stone three-dimensional mask with a simply connected topology based on the stone boundary; S4. Map the candidate stone 3D mask onto 3D space to calculate sphericity and density morphological features, remove cancellous bone areas with density below a preset threshold, crop the candidate stone 3D mask coverage area on the original abdominal CT image sequence, calculate the gray-level co-occurrence matrix and extract texture consistency features, perform weighted fusion analysis on the density morphological features and texture consistency features, distinguish isotropic stone textures from anisotropic vascular calcification textures, and output the urinary stone segmentation results.

2. The region connectivity based medical image segmentation method for urinary stone according to claim 1, characterized in that: The specific process of acquiring raw abdominal CT image sequences, calculating global CT value histograms and local sliding window statistical characteristics, analyzing the distribution locations of bone high-density peaks and soft tissue low-density peaks within local sliding windows, and constructing a piecewise weighted mapping function is as follows: The pixels of the original abdominal CT image sequence are traversed, the global gray-level distribution frequency is statistically analyzed, and the pixel gray-level probability density curve within the local sliding window is extracted. Identify the first peak corresponding to the soft tissue matrix and the second peak corresponding to the bone structure in the gray-scale probability density curve, calculate the right inflection point of the first peak as the low threshold cutoff point, and calculate the left inflection point of the second peak as the high threshold cutoff point. A linear stretching interval is established between the low threshold cutoff point and the high threshold cutoff point. A zero-weight suppression interval is established below the low threshold cutoff point, and a logarithmic decay interval is established above the high threshold cutoff point. The linear stretching interval, the zero-weight suppression interval, and the logarithmic decay interval are combined to form a nonlinear piecewise weighted mapping function.

3. The region connectivity based medical image segmentation method for urinary stone according to claim 2, characterized in that: The specific process of applying a piecewise weighted mapping function to adaptively adjust the window width and level and perform anisotropic diffusion filtering on the original abdominal CT image sequence to generate a bone suppression-stone enhancement feature map is as follows: Input the pixel gray values ​​of the original abdominal CT image sequence to a nonlinear piecewise weighted mapping function, output the mapped gray values, calculate the gradient magnitude vector of the mapped gray values ​​in the three-dimensional spatial neighborhood, and construct an exponential diffusion conduction coefficient function with the gradient magnitude vector as the independent variable. A constraint relationship is set between high gradient regions and low diffusion conduction coefficients, and low gradient regions and high diffusion conduction coefficients. The mapped gray values ​​are iteratively and smoothly updated by an exponential diffusion conduction coefficient function. High gradient edge information is preserved and low gradient internal texture is smoothed, and a skeleton suppression-stone enhancement feature map is output.

4. The region connectivity based medical image segmentation method for urinary stone according to claim 1, characterized in that: Step S2 includes the following steps: A grayscale threshold sequence is set, and the skeleton inhibition-stone enhancement feature map is binarized step by step according to the grayscale threshold sequence to extract the connected components under different grayscale threshold levels. Track the area change rate of connected components between adjacent grayscale threshold levels, and select regions with an area change rate lower than a preset stability threshold as candidate subsets of the maximum stable extreme value region. Calculate the aspect ratio and solidity of the three-dimensional bounding rectangle of the candidate subset of the maximum stable extreme value region. Compare the aspect ratio of the three-dimensional bounding rectangle with the upper limit of the human spine anatomical size and the solidity with the lower limit of the vascular calcification morphology. Eliminate interference regions that meet the upper limit of the spine anatomical size or are lower than the lower limit of the vascular calcification morphology. Retain the centroids of regions that conform to the spatial distribution characteristics of urinary system stones to form a set of stone seed points.

5. The region connectivity based medical image segmentation method for urinary stone according to claim 1, wherein: The specific process of calculating the gradient magnitude of the bone suppression-stone enhancement feature map, constructing a wavefront propagation resistance field based on the gradient magnitude, setting the stone seed point set as the wavefront evolution source point, and executing the fast travel algorithm to calculate the geodesic arrival time field from the pixel point in the image domain to the wavefront evolution source point is as follows: By convolving the bone suppression-stone enhancement feature map with a three-dimensional differential operator, the gradient components of the pixels in the spatial coordinate axis direction are calculated, and the gradient magnitude matrix is ​​synthesized. Construct a wavefront propagation velocity function that is inversely proportional to the gradient magnitude matrix. Take the reciprocal of the wavefront propagation velocity function to generate a wavefront propagation drag field that corresponds to high resistance in high gradient regions and low resistance in low gradient regions. Initialize the stone seed point set as the zero-time frozen node, initialize the remaining pixels in the image domain as the infinite time far end node, and establish a narrowband minimum heap structure to maintain the wavefront extension edge. Following the principle of prioritizing the minimum arrival time of nodes in the narrowband min-heap structure, the discretized Eikonal equation is solved iteratively. The arrival time of the geodesic of the neighboring narrowband nodes is updated using the known frozen nodes until the calculation of the arrival time of the geodesic of the pixels in the preset calculation neighborhood centered on the stone seed point set is completed.

6. The region connectivity based medical image segmentation method of urinary stones according to claim 1, wherein: The specific process of detecting the abrupt changes in local maxima in the geodesic arrival time field, defining these abrupt changes as the stone boundaries, and extracting a 3D mask of candidate stones with a simply connected topology based on these stone boundaries is as follows: Calculate the first derivative of the geodesic arrival time field, generate a time gradient field reflecting the rate of change of wavefront propagation velocity, search for modulus jump ridges in the time gradient field, locate the position where wavefront propagation is blocked and stagnant, and mark the position of the modulus jump ridge as the energy propagation hindrance boundary. The region grows outward from the set of stone seed points until it reaches the energy propagation blockage boundary, then stops filling and closes the contour to generate the initial segmented region. The initial segmented region is then subjected to a three-dimensional morphological closing operation to fill the internal micro-holes and smooth the edge burrs. The voxel topological connectivity within the region is confirmed, and the candidate stone three-dimensional mask is output.

7. The region connectivity based medical image segmentation method of urinary stones according to claim 1, wherein: The specific process of mapping the candidate stone 3D mask onto 3D space to calculate sphericity and density morphological features, removing cancellous bone regions with density below a preset threshold, cropping the candidate stone 3D mask coverage area from the original abdominal CT image sequence, calculating the gray-level co-occurrence matrix, and extracting texture consistency features is as follows: The volume is calculated by counting the total number of voxels inside the 3D mask of candidate stones, the surface area is calculated by extracting the triangular mesh on the surface of the 3D mask of candidate stones, and the sphericity is calculated based on the geometric ratio of volume to surface area. Calculate the covariance matrix of the three-dimensional mask spatial distribution of candidate stones, solve for the eigenvalues ​​of the covariance matrix, and calculate the compact morphological characteristics based on the ratio of the minimum eigenvalue to the maximum eigenvalue. Set a lower limit threshold for cancellous bone density and remove non-stone areas whose density morphology characteristics are lower than the lower limit threshold for cancellous bone density; Define the gray-level co-occurrence matrix to calculate the step size and direction parameters, traverse the projection area of ​​the candidate stone 3D mask on the original abdominal CT image sequence, statistically analyze the joint probability distribution of pixel gray levels, extract the second moment of the angle from the main diagonal as the energy feature, calculate the entropy feature through the logarithm of the probability distribution, and combine the energy feature and the entropy feature to form the texture consistency feature.

8. The region connectivity based medical image segmentation method of urinary stones according to claim 1, wherein: The specific process of performing weighted fusion analysis of dense morphological features and texture consistency features to distinguish between isotropic stone textures and anisotropic vascular calcification textures, and outputting the urinary stone segmentation results is as follows: Range normalization is performed on the dense morphological features and texture consistency features to map the feature values ​​to a unified dimensionless interval. The morphological feature weight coefficient and texture feature weight coefficient are set. The normalized dense morphological features and texture consistency features are then linearly weighted and summed to generate a stone probability score. Set a true positive threshold, compare the stone probability score with the true positive threshold, retain the area with the stone probability score higher than the true positive threshold and identify it as an isotropic stone texture area, remove the area with the stone probability score lower than the true positive threshold and identify it as anisotropic vascular calcification texture area, and output the urinary stone segmentation result.

Images