Renal scoring method and device for kidney tumor

Abstract

The invention discloses a Renal scoring method and device for kidney tumors, and belongs to the field of medical image processing. The method comprises the steps that segmentation data are acquired, and the segmentation data comprise data of kidneys, tumors and blood vessels; anatomical analysis is carried out based on the segmentation data, the kidney principal axis, the renal pelvis position and the ureter are determined, and a natural coordinate system of the kidney is established; calculating the maximum diameter of the tumor based on the tumor voxel, and determining an R score according to the maximum diameter of the tumor; constructing a kidney convex hull based on the kidney parenchyma coordinates, and determining an E score based on the proportion of tumor voxels inside and outside the kidney convex hull; taking the ureter as a set system, determining the minimum distance from the tumor voxel to the set system, and determining an N score according to the minimum distance; constructing a natural interface based on the pelvis position, and determining an A score based on the natural interface; detecting polar lines based on the renal door depression, and determining an L score according to the polar lines. The method provided by the invention eliminates the subjective judgment difference of the scorer.

Description

Technical Field

[0001] This invention relates to the fields of medical image processing, computer-aided diagnosis, and three-dimensional image analysis, and in particular to a Renal scoring method and apparatus for kidney tumors. Background Technology

[0002] To standardize the assessment of the anatomical complexity of renal tumors, Kutikov and Uzzo proposed the RENAL Nephrometry Score in 2009. This system has been widely used in clinical practice and has become a standard tool for assessing the difficulty of partial nephrectomy.

[0003] The RENAL scoring system consists of five components:

[0004] (1) R (Radius) - Maximum diameter of tumor (1 point: ≤4 cm; 2 points: >4 cm but <7 cm; 3 points: ≥7 cm);

[0005] (2) E (Exophytic / Endophytic) - the exophytic / endophytic characteristics of the tumor (1 point: ≥50% exophytic; 2 points: <50% exophytic; 3 points: completely endophytic);

[0006] (3) N (Nearness) - The closest distance between the tumor and the renal collecting system or renal sinus (1 point: ≥7 mm; 2 points: >4 mm but <7 mm; 3 points: ≤4 mm);

[0007] (4) A (Anterior / Posterior) - Tumor location (anterior / posterior) (a: tumor located in front of the kidney; p: tumor located behind the kidney; x: tumor location uncertain or crosses the coronal plane);

[0008] (5) L (Location) - The position of the tumor relative to the renal poles (1 point: completely outside the poles (upper or lower pole); 2 points: tumor <50% crossing the poles; 3 points: tumor >50% crossing the poles or completely between the poles or crossing the midline of the kidney).

[0009] The suffix 'h' indicates that the tumor is in contact with the main renal artery or main renal vein.

[0010] Complexity levels: Low complexity: 4-6 points; Medium complexity: 7-9 points; High complexity: 10-12 points.

[0011] Currently, RENAL scoring is mainly divided into manual scoring and automatic / semi-automatic scoring.

[0012] The following problems exist with manual scoring:

[0013] (1) Subjectivity and variability: The consistency among different raters is only 60-85%; there are also differences in the ratings of the same rater at different times; in particular, the judgment of L rating (polar position) is the most subjective;

[0014] (2) Time-consuming and labor-intensive: A complete assessment of a case takes 15-30 minutes; it requires repeated measurements on multiple sections; and it requires manual marking and calculation.

[0015] (3) Experience dependence: accurate scoring requires extensive anatomical knowledge; the learning curve for beginners is steep; training costs are high;

[0016] (4) Limitations of two-dimensional evaluation: Traditional evaluation is conducted on two-dimensional slices; it is difficult to accurately grasp the three-dimensional spatial relationship; and key information is easily missed.

[0017] The existing automatic / semi-automatic scoring has the following problems:

[0018] (1) Inaccurate polarity localization: Most methods simply divide the long axis of the kidney into three equal parts; without considering the actual anatomical structure of the kidney; and there is a large difference between the actual anatomical structure and the judgment of clinicians;

[0019] (2) Memory and performance issues: memory overflow when calculating the maximum diameter of large tumors; inability to process high-resolution data; excessively long computation time;

[0020] (3) Functional limitations: It does not support the assessment of multiple tumors; it requires a lot of manual intervention; and it lacks complete visualization.

[0021] (4) Ambiguous definition of the collecting system: There is no unified definition of "collecting system"; it is difficult to identify the renal pelvis, renal calyces and ureter; it affects the accuracy of N score;

[0022] (5) Lack of anatomical basis: Ignoring important anatomical landmarks such as the renal hilum and renal sinus; oversimplifying the determination of the anterior-posterior direction; not considering the natural shape of the kidney. Summary of the Invention

[0023] The purpose of this invention is to overcome the shortcomings of the prior art and provide a Renal scoring method and device for kidney tumors, thereby improving the objectivity and accuracy of Renal scoring.

[0024] This invention is achieved through the following technical solution:

[0025] The first aspect of this invention discloses a Renal scoring method for kidney tumors, comprising:

[0026] Acquire segmentation data, which includes data on the kidneys, tumors, and blood vessels;

[0027] Anatomical analysis based on segmented data was performed to determine the main axis of the kidney, the location of the renal pelvis and ureter, and to establish the natural coordinate system of the kidney.

[0028] The maximum diameter of the tumor is calculated based on tumor voxels, and the R score is determined based on the maximum diameter of the tumor.

[0029] The renal convex hull was constructed based on renal parenchymal coordinates, and the E score was determined based on the proportion of tumor voxels inside and outside the renal convex hull.

[0030] Using the ureter as a collecting system, the minimum distance from the tumor voxel to the collecting system was determined, and the N score was determined based on the minimum distance from the tumor voxel to the collecting system.

[0031] Construct a natural interface based on the location of the renal pelvis, and determine the A score based on this natural interface;

[0032] Based on the detection of the polar lines in the renal hilum, the L score is determined according to the polar lines.

[0033] Furthermore, anatomical analysis was performed based on the segmented data to determine the renal axis, the location of the renal pelvis and ureter, and to establish the natural coordinate system of the kidney, including:

[0034] Based on the segmented data, the renal parenchyma coordinates are extracted and converted into physical coordinates. The PCA analysis method is used to calculate the directions of the three principal components: the first principal component corresponds to the long axis of the kidney, the second principal component corresponds to the short axis of the kidney, and the third principal component corresponds to the deep axis of the kidney. The sign of the Z component of the first principal component vector is determined. If it is negative, the first principal component vector is reversed.

[0035] The renal pelvis is located based on the concave region on the surface of the kidney identified by convex hull analysis; the renal pelvis is located based on the centroid of the medullary cone; the renal pelvis is located based on the position of the renal artery and renal vein entering the kidney; the renal pelvis locations determined by the three methods are weighted and fused to obtain the final renal pelvis location.

[0036] With the location of the renal pelvis as the center, search for ureteral branches within a preset radius; calculate the geometric distance from each ureteral branch to the renal pelvis and the number of voxels for each ureteral branch; calculate the score of the ureteral branch based on the geometric distance from the ureteral branch to the renal pelvis and the number of voxels for each ureteral branch; and determine the ureteral branch with the highest score as the target ureter.

[0037] The kidney's long axis is defined as the Z-axis, its width axis as the X-axis, and its depth axis as the Y-axis, thus establishing a natural coordinate system for the kidney. The origin of this natural coordinate system is the geometric center of the kidney.

[0038] Furthermore, the maximum diameter of the tumor is calculated based on tumor voxels, including:

[0039] Detecting the number of tumor voxels;

[0040] If the number of tumor voxels is less than the first preset value, the Euclidean distance between all point pairs in the tumor voxels is calculated, and the largest Euclidean distance is determined as the diameter of the tumor.

[0041] If the number of tumor voxels is greater than or equal to the first preset value and less than the second preset value, then the three-dimensional convex hull of the tumor is calculated; when the number of vertices of the three-dimensional convex hull is less than or equal to the third preset value, the Euclidean distance between all pairs of vertices in the convex hull is calculated, and the largest Euclidean distance is determined as the diameter of the tumor; when the number of vertices of the three-dimensional convex hull is greater than the third preset value, the number of vertices in the convex hull of the third preset value is randomly sampled, the Euclidean distance between all pairs of sampled vertices in the convex hull is calculated, and the largest Euclidean distance is determined as the diameter of the tumor.

[0042] If the number of tumor voxels is greater than or equal to the second preset value, the three-dimensional convex hull of the tumor is calculated. When the number of vertices of the three-dimensional convex hull is less than or equal to the third preset value, the Euclidean distance between all pairs of vertices in the convex hull is calculated, and the largest Euclidean distance is determined as the diameter of the tumor. When the number of vertices of the three-dimensional convex hull is greater than the third preset value, multiple rounds of random sampling are performed on all tumor voxels (e.g., 3 rounds of sampling). In each round, a fourth preset value number of tumor voxels are sampled, and the Euclidean distance between all pairs of vertices in the sampled tumor voxels is calculated. The largest Euclidean distance is determined as the diameter of the tumor.

[0043] Furthermore, a renal convexity is constructed based on renal parenchymal coordinates, and an E score is determined based on the proportion of tumor voxels inside and outside the renal convexity, including:

[0044] Extract the coordinates of the renal parenchyma;

[0045] Construct the renal convex hull based on renal parenchyma coordinates;

[0046] Delaunay triangulation was used to determine the position of each tumor voxel relative to the renal bulge.

[0047] Count the number of tumor voxels inside and outside the renal bulge;

[0048] The E score is determined based on the proportion of tumor voxels inside and outside the renal bulge.

[0049] If the number of renal parenchyma voxels is less than or equal to the fifth preset value, then the renal parenchyma convex hull is constructed based on all points; if the number of renal parenchyma voxels is greater than the fifth preset value, then the fifth preset value number of points are randomly sampled, and then the renal parenchyma convex hull is constructed based on the sampled points.

[0050] Furthermore, considering the ureter as a collecting system, the minimum distance from the tumor voxel to the collecting system is determined, and the N score is determined based on this minimum distance, including:

[0051] Treating the ureter as a collection system, create a collection system mask;

[0052] Calculate the Euclidean distance from each tumor voxel to the ensemble system;

[0053] Determine the minimum distance from tumor voxels to the assemblies;

[0054] The N score is determined based on the minimum distance from the tumor voxel to the collection system.

[0055] Furthermore, a natural interface based on the location of the renal pelvis is constructed, and the A score is determined based on this natural interface, including:

[0056] Obtain the kidney's main axis and center;

[0057] Calculate the vector of the renal pelvis relative to the center of the kidney, and take the cross product of the vertical component of this vector with the principal axis of the kidney to obtain the interface normal vector;

[0058] The natural interface is defined as a plane passing through the center of the kidney, with its normal vector perpendicular to the main axis of the kidney and the direction of the renal pelvis.

[0059] Calculate the directed distance from the tumor voxel to the natural boundary;

[0060] Statistical analysis of the ratio of positive to negative values ​​in directed distances;

[0061] The A score is determined based on the ratio of positive to negative values.

[0062] Furthermore, based on the detection of polarity in the renal hilum, the L score is determined according to the polarity, including:

[0063] Morphological operations are used to fill the internal cavity of the kidney, extract the coordinate points on the kidney surface, construct a convex hull based on the coordinate points on the kidney surface, search for depression points, the depression points are inside the convex hull but not inside the filled kidney, and the renal hilum region is determined based on the depression points;

[0064] Calculate the direction vector from each depression point to the renal pelvis, calculate the angle between the direction vector and the inward direction vector of the depression point, and retain the depression points corresponding to the angle being less than the eleventh preset value.

[0065] Calculate the distance from each indentation point to the center of the kidney and the depth of the indentation, and retain the indentation points whose distance from the center of the kidney and the depth of the indentation are within a preset range;

[0066] The K value is determined based on the set of renal hilar depression points. The K-nearest neighbor local density of each point in the set of renal hilar depression points is calculated. The median of the K-nearest neighbor local density of all points is taken as the threshold. Points with a density greater than the threshold are clustered by DBSCAN. The largest cluster is the core region of the renal hilum.

[0067] Project the depression point of the core area of ​​the renal hilum onto the main axis of the kidney. Using the Z coordinate of the renal pelvis as the boundary, divide the depression point into upper and lower groups according to the position of the renal pelvis. Determine the upper and lower pole lines according to the percentile of the Z coordinate of the depression points in the upper and lower groups, and constrain the distance between the upper and lower pole lines within a preset range.

[0068] The L score is calculated based on the positional relationship between the tumor and the superior and inferior poles.

[0069] Furthermore, the Renal scoring method also includes:

[0070] Perform format conversion and standardization on the segmented data;

[0071] Unify the coordinate system of the segmented data;

[0072] Separate the left and right kidney data from the segmented data;

[0073] The test determines whether there are multiple tumors; if so, the tumors are separated.

[0074] The process includes detecting whether there are multiple tumors. If there are multiple tumors, the tumors are separated, including: separating tumors using connected components; filtering out tumors with a volume smaller than a tenth preset value; calculating the number of overlapping voxels between the tumor and the left and right kidneys respectively; if the proportion of overlapping voxels between the tumor and a certain kidney is greater than 50%, then the kidney is marked as the kidney of the tumor; and sorting the tumors in descending order of volume.

[0075] Furthermore, the Renal scoring method also includes:

[0076] Voxel data of kidney, tumor, and blood vessel are downsampled, isosurfaces are extracted using the Marching Cubes algorithm to generate triangular meshes, coordinate transformations are performed on the triangular meshes to restore the physical scale, and three-dimensional surface models of kidney, tumor, and blood vessel are constructed.

[0077] Render 3D surface models of kidneys, tumors, and blood vessels, overlay and display heatmaps of depressions, add epipolar markers, and generate an interactive 3D scene.

[0078] The second aspect of this embodiment discloses a Renal scoring device for kidney tumors, comprising:

[0079] The data acquisition module is used to acquire segmented data, which includes data on kidneys, tumors, and blood vessels.

[0080] The anatomical analysis module is used to perform anatomical analysis based on segmented data, determine the main axis of the kidney, the location of the renal pelvis and ureter, and establish the natural coordinate system of the kidney;

[0081] The R-score module is used to calculate the maximum diameter of the tumor based on tumor voxels and determine the R-score based on the maximum diameter of the tumor.

[0082] The E-score module is used to construct the renal convexity based on renal parenchyma coordinates and to determine the E-score based on the proportion of tumor voxels inside and outside the renal convexity.

[0083] The N-score module is used to treat the ureter as a collecting system, determine the minimum distance from the tumor voxel to the collecting system, and determine the N-score based on the minimum distance from the tumor voxel to the collecting system.

[0084] The A-score module is used to construct a natural interface based on the location of the renal pelvis, and to determine the A-score based on this natural interface.

[0085] The L-score module is used to detect the polar lines based on the renal hilum indentation and determine the L-score based on the polar lines.

[0086] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0087] (1) The method of the present invention eliminates the subjective judgment difference of the raters and provides repeatable and verifiable rating results;

[0088] (2) The method of the present invention reduces the single case scoring time from 20-30 minutes to 30-60 seconds, and can automatically generate reports without manual processing;

[0089] (3) Traditional manual scoring requires doctors to repeatedly measure and judge each case, while this invention transforms all scoring steps (anatomical analysis, five-item scoring calculation) into automated algorithms, realizing a fully automated RENAL scoring process. The processing flow of each case is completely independent, and multiple case data can be processed in parallel at the same time.

[0090] (3) The method of the present invention is based on three-dimensional voxel data of real anatomical structure for spatial analysis. It uses 3D analysis to replace 2D assessment, avoids information loss of two-dimensional measurement, provides more comprehensive information, is more in line with clinical cognition, and improves measurement accuracy.

[0091] (4) The method of the present invention supports the processing of ultra-large tumors (>300cm³), automatically processes multiple tumors, and is compatible with incomplete segmentation data (such as missing blood vessels).

[0092] (5) The method of the present invention can assist in the formulation of surgical plans, objectively assess surgical risks, facilitate the standardization of multi-center studies, and help train young doctors. Attached Figure Description

[0093] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0094] Figure 1 This is a flowchart of the Renal scoring method in this invention. Detailed Implementation

[0095] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0096] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0097] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "inner," and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the accompanying drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0098] In the description of this invention, unless otherwise explicitly specified and limited, the term "connection" or similar designation indicating a connection between components should be interpreted broadly. For example, it can refer to a fixed connection, a detachable connection, or an integral part; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can refer to the internal communication between two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0099] like Figure 1 As shown in the figure, this embodiment discloses a Renal scoring method and device for kidney tumors.

[0100] The first aspect of this embodiment discloses a Renal scoring method for kidney tumors, such as... Figure 1 As shown, the Renal scoring method includes steps S100 to S700.

[0101] Step 100. Obtain segmentation data, which includes data on the kidneys, tumors, and blood vessels.

[0102] The format of the segmented data is DICOM / NIFTI.

[0103] The segmented data includes data on kidneys, tumors, blood vessels, etc.

[0104] Step S200. Perform anatomical analysis based on the segmented data to determine the kidney's main axis, renal pelvis location, and ureter, and establish the kidney's natural coordinate system.

[0105] In some embodiments of this example, principal component analysis (PCA) is used to calculate the renal principal axis.

[0106] The principal axis of the kidney was calculated using PCA analysis, including: extracting renal parenchymal coordinates from segmented data and converting them into physical coordinates; calculating the directions of three principal components using PCA analysis: the first principal component corresponds to the long axis of the kidney (which is the principal axis), the second principal component corresponds to the width axis, and the third principal component corresponds to the depth axis; and determining the sign of the Z-component of the first principal component vector, if negative, reversing the first principal component vector. The first principal component obtained by PCA analysis is a three-dimensional vector [x, y, z], and its Z-component refers to the third component value of this three-dimensional vector. Determining the sign of the Z-component is to ensure that the long axis points upwards (consistent with medical conventions).

[0107] The renal parenchyma coordinates are the three-dimensional position coordinates of all voxels labeled as kidney tissue in the segmented data. The segmented data is a three-dimensional matrix, with each voxel having corresponding index coordinates [i,j,k]. The renal parenchyma region has specific label values ​​in the segmentation result. These position indices are extracted and then converted into actual physical coordinates in millimeters based on the voxel spacing parameters of the medical image.

[0108] In some embodiments of this example, the renal parenchyma coordinates are extracted based on the segmented data and converted into physical coordinates, including: obtaining index coordinates by traversing the voxels in the segmentation mask whose values ​​are kidney labels, and multiplying the index coordinates by the voxel spacing (e.g., [1.0, 1.0, 3.0] mm) to convert them into physical coordinates.

[0109] In this embodiment, PCA analysis yields three mutually orthogonal principal component directions: the first principal component corresponds to the direction of greatest data variation, i.e., the long axis of the kidney (vertical direction); the second principal component corresponds to the wide axis of the kidney (horizontal direction); and the third principal component corresponds to the deep axis of the kidney (anteroposterior direction). Medical convention dictates that the long axis of the kidney should point from the lower pole to the upper pole (positive Z-axis direction). However, PCA analysis can only determine the axis direction but not the pointing direction. In this embodiment, by determining the sign of the Z-component of the first principal component vector, if it is negative, the vector is reversed (multiplied by -1) to ensure the Z-component is positive, thus guaranteeing that the long axis points upwards and unifying the coordinate system orientation for all cases.

[0110] In some embodiments of this example, the method for determining the location of the renal pelvis includes: identifying a concave region on the surface of the kidney based on convex hull analysis, and determining the location of the renal pelvis based on the concave region; identifying the medullary cone, and determining the location of the renal pelvis based on the centroid of the medullary cone; determining the location of the renal pelvis based on the location where the renal artery and renal vein enter the kidney; and weightedly fusing the renal pelvis locations determined by the three methods to obtain the final renal pelvis location.

[0111] The renal pelvis, located inside the renal hilum, is where urine collects. The hilar indentation is the natural opening of the renal pelvis to the outside of the kidney. The location of the renal pelvis can be determined by locating the largest indentation on the surface of the kidney. The medullary pyramids radiate around the renal pelvis, with their centroids pointing towards the center of the pelvis. The geometric center of the medullary pyramids can be used to locate the interior of the renal pelvis. The renal artery and renal vein enter at the renal hilum, with their entrances located near the renal pelvis. The location of the renal hilum can be confirmed by identifying the points of entry of these vessels.

[0112] This embodiment determines the anatomical location of the renal pelvis from different angles, such as the concave area on the surface of the kidney, the centroid of the medullary cone, and the positions where the renal artery and renal vein enter the kidney. Then, the renal pelvis locations determined by multiple methods are weighted and fused to obtain the final renal pelvis location. The accuracy of renal pelvis location is improved by cross-validating multiple methods, and a more robust renal pelvis location result is obtained through weighted fusion.

[0113] In some embodiments of this example, the weights of the renal pelvis location determined by indentation detection, medullary analysis, and vascular localization decrease sequentially.

[0114] In this embodiment, the renal pelvis location determined by the three methods is weighted and fused, including: linearly combining the three-dimensional coordinates of the renal pelvis location obtained by the three methods according to their weights. For example, the weight ratios of the three methods—indentation detection, medullary pyramid analysis, and vascular renal hilum localization—are 0.6, 0.25, and 0.15, respectively. Therefore, the final location = 0.6 × indentation method location + 0.25 × medullary method location + 0.15 × vascular method location.

[0115] In some embodiments of this example, the method for determining the ureter includes: searching for ureteral branches within a preset radius centered on the location of the renal pelvis; calculating the geometric distance from each ureteral branch to the renal pelvis and the number of voxels for each ureteral branch; calculating a score for each ureteral branch based on the geometric distance from the ureteral branch to the renal pelvis and the number of voxels for each ureteral branch; and determining the ureteral branch with the highest score as the target ureter.

[0116] The scoring formula for ureteral branches is: Ureteral branch score = Number of voxels of ureteral branches / (Geometric distance from ureteral branch to renal pelvis + 1). The addition of 1 in the scoring formula avoids division by zero.

[0117] In these implementations, when multiple ureteral branches exist, the correct target ureter can be automatically selected.

[0118] In some embodiments of this example, establishing a natural coordinate system for the kidney includes: defining the major axis of the kidney as the Z-axis, the width axis of the kidney as the X-axis, and the depth axis of the kidney as the Y-axis, thereby establishing a natural coordinate system for the kidney, wherein the origin of the natural coordinate system for the kidney is the geometric center of the kidney.

[0119] These implementations establish a natural coordinate system for the kidney based on its own morphological characteristics, serving as an anatomical reference system and eliminating the influence of patient position and individual differences in the kidneys. The natural axis of the kidney is automatically located using PCA analysis, ensuring that all subsequent spatial judgments (anterior-posterior, superior-inferior, internal-external) are based on the kidney's own anatomical structure rather than external coordinates. This natural coordinate system is used throughout the entire scoring process, ensuring the anatomical accuracy of the scoring. This is one of the advantages of the method in this embodiment compared to existing technologies.

[0120] Step S300. Calculate the maximum diameter of the tumor based on tumor voxels, and determine the R score based on the maximum diameter of the tumor.

[0121] In some embodiments of this example, the maximum diameter of the tumor is calculated based on the number of tumor voxels, including steps S310 to S340.

[0122] Step S310. Detect the number of tumor voxels.

[0123] Step S320. If the number of tumor voxels is less than the first preset value, calculate the Euclidean distance between all point pairs in the tumor voxels and determine the largest Euclidean distance as the diameter of the tumor.

[0124] Step S330. If the number of tumor voxels is greater than or equal to the first preset value and less than the second preset value, calculate the three-dimensional convex hull of the tumor; when the number of vertices of the three-dimensional convex hull is less than or equal to the third preset value, calculate the Euclidean distance between all pairs of vertices in the convex hull and determine the largest Euclidean distance as the diameter of the tumor; when the number of vertices of the three-dimensional convex hull is greater than the third preset value, randomly sample the number of vertices in the convex hull of the third preset value, calculate the Euclidean distance between all pairs of vertices in the sampled convex hull and determine the largest Euclidean distance as the diameter of the tumor.

[0125] Step S340. If the number of tumor voxels is greater than or equal to the second preset value, calculate the three-dimensional convex hull of the tumor; when the number of vertices of the three-dimensional convex hull is less than or equal to the third preset value, calculate the Euclidean distance between all point pairs of the convex hull vertices, and determine the largest Euclidean distance as the diameter of the tumor; when the number of vertices of the three-dimensional convex hull is greater than the third preset value, perform multiple rounds of random sampling on all tumor voxels (e.g., perform 3 rounds of sampling), sample a fourth preset value number of tumor voxels in each round, calculate the Euclidean distance between all point pairs of the sampled tumor voxels, and determine the largest Euclidean distance as the diameter of the tumor.

[0126] For example, the first preset value is 3000, the second preset value is 50000, the third preset value is 2000, and the fourth preset value is 10000.

[0127] In this embodiment, for tumors with 50,000 or more voxels, a convex hull is first constructed. If the number of vertices in the convex hull is less than or equal to 2,000, the convex hull method is used. If the number of vertices in the convex hull is greater than 2,000 (indicating that the tumor shape is highly irregular, such as multi-lobed or branched), the Monte Carlo method is used to approximate the true maximum diameter through multiple random samplings. This ensures the accuracy of the results while controlling memory usage. The distance matrix for 2,000 vertices requires approximately 16MB of memory, which is an acceptable upper limit. The computational complexity of convex hull construction itself is O(nlogn), so even if it fails, it will not cause serious performance loss. This adaptive method does not simply divide based on the voxel count threshold, but dynamically selects the optimal algorithm according to the actual geometric complexity of the tumor, balancing accuracy and efficiency and ensuring robust handling of tumors of various morphologies.

[0128] In these implementations, for cases where the number of tumor voxels is less than a first preset value, the distance between all point pairs is directly calculated, resulting in low computational complexity and 100% accuracy. For cases where the number of tumor voxels is greater than or equal to the first preset value but less than the second preset value, convex hull vertices are used to replace the original points for calculation. Since there are usually only a few hundred convex hull vertices, the computational load is significantly reduced. For cases where the number of tumor voxels is greater than the second preset value, an adaptive selection of convex hull calculation or intelligent sampling method is used to ensure controllable memory usage.

[0129] These embodiments optimize memory by reducing the number of points involved in distance calculation, using convex hull vertices or sampled points instead of all voxel points, reducing the storage requirement of the O(n²) distance matrix to O(m²), where m << n. The capital O symbol in O() is a mathematical notation representing algorithm complexity. O(n²) and O(m²) represent the time complexity of the algorithm, and n and m represent the number of data points. Calculating the distances between all pairs of points requires storing an n×n distance matrix. Through the convex hull algorithm, 50,000 voxels may only require a few hundred convex hull vertices, and through sampling, only 10,000 sampled points need to be calculated, which significantly reduces memory occupancy and thus effectively optimizes memory. For example, for a 10K voxel tumor: 760MB → 200MB (3.8-fold optimization), for a 50K voxel tumor: 19GB → 300MB (63-fold optimization), for a 100K voxel tumor: 81GB → 400MB (202-fold optimization).

[0130] Step S400. Construct a kidney convex hull based on the renal parenchyma coordinates and determine the E score based on the proportion of tumor voxels inside and outside the kidney convex hull.

[0131] In some embodiments of this embodiment, constructing a kidney convex hull based on the renal parenchyma coordinates and determining the E score based on the proportion of tumor voxels inside and outside the kidney convex hull includes steps S410 to S450.

[0132] Step S410. Extract the renal parenchyma coordinates.

[0133] Step S420. Construct a kidney convex hull based on the renal parenchyma coordinates.

[0134] In some embodiments of this embodiment, constructing a kidney convex hull based on the renal parenchyma coordinates includes: if the number of renal parenchyma voxels is less than or equal to the fifth preset value, construct a renal parenchyma convex hull based on all points; if the number of renal parenchyma voxels is greater than the fifth preset value, randomly sample the number of points of the fifth preset value, and then construct a renal parenchyma convex hull based on the sampled points.

[0135] For example, the fifth preset value is 15,000.

[0136] Step S430. Determine the position of each tumor voxel relative to the kidney convex hull.

[0137] In some embodiments of this embodiment, determining the position of each tumor voxel relative to the kidney convex hull includes: using Delaunay triangulation to determine whether each tumor voxel is inside or outside the kidney convex hull.

[0138] Step S440. Count the number of tumor voxels inside and outside the kidney convex hull.

[0139] Step S450. Determine the E score based on the proportion of tumor voxels inside and outside the renal bulge.

[0140] In some embodiments of this example, the E score is determined based on the proportion of tumor voxels located inside and outside the renal bulge, including: if the proportion of tumor voxels located outside the renal bulge is greater than or equal to 50% of the total tumor voxels, the E score is 1; if the proportion of tumor voxels located outside the renal bulge is greater than 0 and less than 50% of the total tumor voxels, the E score is 2; if the proportion of tumor voxels located outside the renal bulge is equal to 0% of the total tumor voxels, the E score is 3.

[0141] These implementations are based on rigorous geometric theory and can accurately process kidneys of any shape. They are computationally efficient and suitable for real-time applications.

[0142] Step S500. Using the ureter as a collecting system, determine the minimum distance from the tumor voxel to the collecting system, and determine the N score based on the minimum distance from the tumor voxel to the collecting system.

[0143] In some embodiments of this example, the ureter is used as a collecting system, the minimum distance from the tumor voxel to the collecting system is determined, and the N score is determined based on the minimum distance from the tumor voxel to the collecting system, including steps S510 to S540.

[0144] Step S510. Treat the ureter as a collection system and create a collection system mask.

[0145] Step S520. Calculate the Euclidean distance from each tumor voxel to the ensemble system.

[0146] Step S530. Determine the minimum distance from the tumor voxel to the collection system.

[0147] Step S540. Determine the N score based on the minimum distance from the tumor voxel to the collection system.

[0148] In some embodiments of this example, the N score is determined based on the minimum distance from the tumor voxel to the collection system, including: if the minimum distance from the tumor voxel to the collection system is greater than or equal to a sixth preset value, then the N score is 1; if the minimum distance from the tumor voxel to the collection system is greater than or equal to a seventh preset value and less than a sixth preset value, then the N score is 2; if the minimum distance from the tumor voxel to the collection system is less than a seventh preset value, then the N score is 3.

[0149] In some embodiments of this example, if the minimum distance from the tumor voxel to the collection system is less than an eighth preset value, it is considered that the contact collection system has been established.

[0150] The traditional RENAL scoring system assigns a score of 3 to tumors ≤4mm. However, in clinical practice, whether a tumor touches the collecting system directly affects whether repair of the collecting system is necessary, significantly altering the surgical difficulty and complication risk. Touching the collecting system (distance <1mm) has special clinical significance, indicating that the tumor may have invaded the collecting system, influencing the choice of surgical approach, and requires special note in the report.

[0151] For example, the sixth preset value is 7mm, the seventh preset value is 4mm, and the eighth preset value is 1mm.

[0152] Traditional methods typically identify complex structures such as the renal pelvis and calyces for N-score calculations. However, these structures have blurred boundaries and are difficult to identify on images, leading to complex and unstable algorithms. This embodiment uses only the ureter as the collection system. The ureter is directly and continuously connected to the renal pelvis and calyces; measuring the distance to the ureter is equivalent to measuring the distance to the entire collection system, avoiding the identification of complex internal structures. This approach is characterized by ease of identification and segmentation, direct connection to the collection system, and clear clinical significance. The collection system is a connected whole; the shortest distance from the tumor to any part reflects surgical risk. Choosing the most easily identifiable ureter as a representative simplifies the algorithm while maintaining clinical significance.

[0153] The methods for calculating N-scores in these implementations have the following advantages: low computational complexity, which can efficiently process large amounts of data; only the minimum distance value is saved during the calculation process, and the complete distance array from each tumor voxel to the set system is not stored, which reduces the storage requirement from O(n) to O(1); the N-score only needs the minimum distance value, and the minimum value is dynamically updated during traversal calculation, without the need to save all intermediate results, which greatly saves memory.

[0154] Step S600. Construct a natural interface based on the location of the renal pelvis, and determine the A score based on this natural interface.

[0155] In some embodiments of this example, a natural interface based on the location of the renal pelvis is constructed, and an A score is determined based on the natural interface, including steps S610 to S660.

[0156] Step S610. Obtain the kidney's main axis and kidney center.

[0157] Step S620. Calculate the vector of the renal pelvis relative to the center of the kidney, and take the cross product of the vertical component of the vector with the principal axis of the kidney to obtain the interface normal vector.

[0158] In some embodiments of this example, when the renal pelvis coincides with the center of the kidney or the renal pelvis vector is parallel to the main axis of the kidney, the effective normal vector cannot be calculated by cross product, and the default Y-axis is used as the interface normal vector.

[0159] Step S630. Determine the natural interface, which is a plane passing through the center of the kidney, with its normal vector perpendicular to the main axis of the kidney and the direction of the renal pelvis.

[0160] Step S640. Calculate the directed distance from the tumor voxel to the natural interface.

[0161] Step S650. Calculate the ratio of positive to negative values ​​in the directed distance.

[0162] Step S660. Determine the A score based on the ratio of positive to negative values.

[0163] In some embodiments of this example, the A score is determined based on the ratio of positive to negative values, including: if the ratio of positive values ​​is greater than a ninth preset value, the A score is a (indicating that the tumor is located in front of the kidney); if the ratio of negative values ​​is greater than the ninth preset value, the A score is p (indicating that the tumor is located behind the kidney); if both the ratio of positive and negative values ​​are less than or equal to the ninth preset value, the A score is x (indicating that the tumor crosses the kidney or its location is uncertain); if the tumor's Z-axis span is greater than 70% of the kidney's span, the determination of its anterior and posterior position becomes meaningless, and the A score is x (uncertain).

[0164] Methods for determining whether the tumor's Z-axis span is greater than 70% of that of the kidney include: calculating the ratio of the difference between the maximum and minimum values ​​of the tumor voxel Z-coordinates to the difference between the maximum and minimum values ​​of the kidney voxel Z-coordinates. If the ratio is >0.7, then the tumor's Z-axis span is greater than 70% of that of the kidney.

[0165] Traditional methods use a simple coronal plane (Y=0 plane) to divide the tumor anteriorly and posteriorly, without considering the natural tilt angle of the kidney and individual differences. This leads to inconsistencies between the judgment and the clinician's anatomically based assessment, potentially misjudging an anterior tumor as a posterior one, affecting the selection of the surgical approach. These implementations construct a natural dividing line that considers the natural tilt of the kidney and the position of the renal pelvis, which is more consistent with renal anatomy than a simple coronal plane. When the renal pelvis is anterior, the dividing line tilts accordingly, more accurately distinguishing the anterior and posterior positions of the tumor. These implementations provide a quantitative anterior-posterior distribution ratio based on actual anatomical landmarks (such as the renal hilum, renal pelvis, renal vessels, etc.), rather than an artificially defined geometric plane, making them more clinically significant than arbitrary geometric divisions.

[0166] Step S700. Detect the polar lines based on the renal hilum indentation and determine the L score according to the polar lines.

[0167] In some embodiments of this example, the L score is determined based on the detection of the polar line in the renal hilum indentation, including steps S710 to S750.

[0168] Step S710. Locate the renal hilum region by detecting the renal hilum indentation.

[0169] In some embodiments of this example, the renal hilum region is located by detecting renal hilum depressions, including: filling the internal cavity of the kidney using morphological operations; extracting coordinate points on the kidney surface; constructing a convex hull based on the coordinate points on the kidney surface; searching for depression points that are within the convex hull but not within the filled kidney; and determining the renal hilum region based on the depression points.

[0170] The kidney surface coordinates refer to the coordinates of the outermost voxel in the three-dimensional kidney voxel matrix, which has at least one exposed face. Through morphological erosion operations or boundary detection algorithms, voxel points adjacent to the background in the kidney mask are found, and these points constitute the surface of the kidney.

[0171] In these implementations, the internal cavity of the kidney is first filled with morphological closure operations, then surface points are extracted, and a three-dimensional convex hull is constructed for these surface points to subsequently locate concave areas.

[0172] Step S720. Calculate the direction vector from each depression point to the renal pelvis, calculate the angle between the direction vector and the inward direction vector of the depression point, and retain the depression points corresponding to the angle being less than the eleventh preset value.

[0173] The inward direction of a depression refers to the direction from the surface of the convex hull towards the inside.

[0174] For example, the eleventh preset value is 60°.

[0175] In these implementations, orientation verification ensures that the detected depression is indeed the renal hilum and not another depression. Points with consistent orientation (e.g., the angle between the direction vector from the depression point to the renal pelvis and the inward concave direction vector of the depression point is <60°) are retained as candidate points for the renal hilum; points with inconsistent orientation are discarded and do not participate in subsequent polarimetric calculations. There may be multiple depressions on the surface of the kidney, but only the depression pointing towards the renal pelvis is the true renal hilum. Orientation verification filters out noise and improves the accuracy of polarimetric localization.

[0176] Step S730. Calculate the distance from each depression point to the center of the kidney and the depression depth, and retain the depression points whose distance from the center of the kidney and the depression depth are within the preset range.

[0177] In these implementations, the distance from each depression point to the center of the kidney and the depression depth are calculated. Points with a distance within a reasonable range (30%-70% of the kidney radius) and a depression depth >2mm are retained. Surface noise and non-hilar depressions are excluded through geometric constraints to ensure that the points participating in the polarimetric calculation are all real hilar region points.

[0178] Step S740. Extract the core region of the renal hilum by density clustering.

[0179] In some embodiments of this example, the core region of the renal hilum is extracted by density clustering, including: determining a K value based on the set of renal hilum depression points, where K refers to the square root of the number of points; calculating the K nearest neighbor local density of each point in the set of renal hilum depression points; taking the median of the K nearest neighbor local densities of all points as a threshold; and clustering points with a density greater than the threshold using DBSCAN, with the largest cluster being the core region of the renal hilum.

[0180] In these implementations, the core region is extracted using density clustering, discrete noise points are removed, and the central region of the renal hilum is retained. The renal hilum depressions may be scattered; density clustering finds the most concentrated region as the renal hilum core, which more accurately represents the location of the renal hilum for polar line localization.

[0181] Step S750. Determine the polar position based on the projection of the renal hilum core region onto the renal main axis.

[0182] In some embodiments of this example, the polar line position is determined based on the projection of the renal hilum core region onto the renal main axis, including: projecting the concave points of the renal hilum core region onto the renal main axis; dividing the concave points into upper and lower groups based on the Z-coordinate of the renal pelvis, wherein concave points with Z-coordinates greater than those of the renal pelvis are in the upper group, and concave points with Z-coordinates less than those of the renal pelvis are in the lower group; determining the upper and lower polar lines based on the percentiles of the Z-coordinates of the concave points in the upper and lower groups, for example, the upper polar line position is taken as the 5th percentile of the Z-coordinates of the upper group points, with 95% of the points in the upper group above this position, ensuring that the polar line is close to the lower edge of the renal hilum; the lower polar line position is taken as the 95th percentile of the Z-coordinates of the lower group points, with 95% of the points in the lower group below this position, ensuring that the polar line is close to the upper edge of the renal hilum; and constraining the polar line spacing to be 15% to 45% of the renal length.

[0183] After calculating the initial polarity, check if the distance between the polarities is within 15%-45% of the kidney length. If it is less than 15%, expand proportionally towards both ends, for example, by multiplying by 1.5, until it reaches 15%; if it is greater than 45%, contract proportionally towards the center, for example, by multiplying by 0.8, until it reaches 45%. During adjustment, maintain the central position and symmetrically expand or contract towards both ends to ensure a reasonable polarity (e.g., within 15%-45% of the kidney length). A polarity that is too small will classify most tumors as crossing the polarities (3 points), while a polarity that is too large will lose its distinguishing significance. The 15%-45% constraint is based on clinical experience and anatomical principles.

[0184] Step S760. Calculate the L score based on the positional relationship between the tumor and the superior and inferior poles.

[0185] In this embodiment, the process proceeds from rough indentation points to verified valid points, then to the clustered core region, and finally to projection to obtain accurate epipolar lines. Each step optimizes the result of the previous step, thus improving the accuracy of the obtained epipolar lines.

[0186] In some embodiments of this example, determining the L score based on the polar lines includes: calculating the projection range of the tumor on the main axis of the kidney; calculating the midline position, where midline position = (superior polar line Z-coordinate + inferior polar line Z-coordinate) / 2; if the tumor is completely below the inferior polar line or completely above the superior polar line, the L score is 1; if the tumor crosses the midline, the L score is 3. When the tumor is not completely outside the polar lines and does not cross the midline, the proportion of the tumor volume inside the polar lines is calculated; if the proportion is greater than or equal to 50%, the L score is 3; otherwise, the L score is 2.

[0187] Traditional methods simply trisect the long axis of the kidney, neglecting its actual anatomical structure, leading to significant discrepancies with clinicians' judgments. This embodiment proposes a novel algorithm based on renal hilum indentation detection, offering the following advantages: robustness due to its reliance on the actual renal hilum structure rather than arbitrary geometric divisions; reduced noise impact through statistical methods and density clustering; dynamic adjustment based on the specific morphology of each kidney; and a multiple validation mechanism achieving a confidence level of 0.98.

[0188] In some embodiments of this example, the segmented data is preprocessed after it is acquired.

[0189] In some embodiments of this example, the segmented data is preprocessed, including: format conversion and standardization of the segmented data; unification of the coordinate system of the segmented data; separation of left and right kidney data in the segmented data; detection of whether there are multiple tumors, and if there are multiple tumors, separation of the multiple tumors.

[0190] In some embodiments of this example, the detection of whether there are multiple tumors, and if there are multiple tumors, the separation of multiple tumors, including: separating tumors using connected components; filtering out tumors with a volume smaller than a tenth preset value; determining the kidney to which the tumor belongs; and sorting the tumors in descending order of volume.

[0191] For example, the tenth preset value is 0.5cm. 3 .

[0192] In some embodiments of this example, filtering out tumors with a volume smaller than the tenth preset value includes: traversing each tumor region, extracting the mask of the current tumor, calculating the tumor volume based on the tumor mask, and the formula for calculating the tumor volume is: tumor volume = number of mask voxels × product of voxel spacing / 1000, and filtering out tumors with a volume smaller than the tenth preset value.

[0193] In some embodiments of this example, determining the kidney to which the tumor belongs includes: calculating the number of overlapping voxels between the tumor and the left and right kidneys, respectively. If the percentage of overlapping voxels between the tumor and a certain kidney is greater than 50%, then that kidney is marked as the kidney to which the tumor belongs. For example, if the percentage of overlapping voxels between the tumor and the left kidney is greater than 50%, then the tumor belongs to the left kidney. If the percentage of overlapping voxels between the tumor and both the left and right kidneys is less than or equal to 50%, then the tumor is an extrarenal tumor and is ignored.

[0194] In these implementations, for patients with multiple renal tumors, the RENAL score of each tumor can be automatically detected and evaluated separately, generating a list of tumor analysis results. The list of tumor analysis results includes a unique tumor identifier, tumor volume, the kidney, and the RENAL score results.

[0195] In this embodiment, all distance calculations are based on physical coordinates, and the unit is millimeters (mm).

[0196] In some embodiments of this example, the Renal scoring method further includes: constructing a three-dimensional surface model of the kidney, tumor, and blood vessels to display the anatomical structure; and visualizing geometric markers such as the convex hull contour, renal hilum depression, and polar plane to aid in understanding the scoring, thereby helping to verify and interpret the scoring results.

[0197] The methods for constructing 3D surface models include: downsampling voxel data, extracting isosurfaces to generate triangular meshes using the Marching Cubes algorithm, and performing coordinate transformations on the triangular meshes to restore the physical scale. In these implementations, the sampling rate is dynamically adjusted according to the data size to balance visual quality and performance, accurately restoring the physical coordinates after downsampling, and avoiding the need to process the original high-resolution data.

[0198] The visualization process includes: rendering the kidney and tumor meshes, overlaying a heatmap of depressions, adding epipolar markers, and generating an interactive 3D scene.

[0199] In these implementations, the modeling process converts discrete voxels into continuous surfaces for easy rendering. During visualization, the basic organ model, scoring-related markers (polar lines, depressions, convex hulls, etc.), interactive controls, etc., are displayed in layers to help doctors understand the scoring criteria.

[0200] In some embodiments of this example, the renal hilum depression region is visualized by: calculating the distance from each point to the center of the depression as the depression depth; creating a scatter plot of depth mapping; and color mapping the points based on the depression depth, with the color mapping being a gradient from light blue to dark blue.

[0201] For example, the minimum depth corresponds to light blue (RGB: 220, 220, 255), the middle value corresponds to cornflower blue (RGB: 100, 149, 237), and the maximum value corresponds to dark blue (RGB: 25, 25, 112).

[0202] In these implementations, depth information is intuitively expressed through color gradients; a medical-grade color scheme is adopted, which conforms to clinical practice; and interactive 3D display supports observation from any angle.

[0203] In some embodiments of this example, the representation method of convex hull includes: calculating the area of ​​each triangular facet, determining the number of sampling points on the triangular facet based on the area of ​​the triangular facet, for example, the larger the area of ​​the triangular facet, the more sampling points there are; uniformly sampling on the triangular facet according to the determined number of sampling points to form a point cloud dataset for constructing the convex hull.

[0204] In these implementations, a point cloud dataset that better reflects the details of the original geometry is created by adaptively sampling each face of the triangular mesh according to the area ratio. Larger facets generate more points, thereby improving the point cloud density and detail representation in these areas.

[0205] In these implementations, point cloud visualization is used, which is softer than wireframes and does not interfere with the observation of the main structure; adaptive sampling density is used, with denser sampling of large areas; and a semi-transparent effect is provided while maintaining the convex hull shape.

[0206] Single tumor case implementation

[0207] Case information:

[0208] Patient: BaiXinqi

[0209] Diagnosis: Right lower pole tumor

[0210] Data: Enhanced CT scan, slice thickness 1mm

[0211] Processing flow:

[0212] 1. Data Loading and Preprocessing

[0213] Loading NIfTI format segmented data:

[0214] - RenalOccupation.nii.gz (Tumor)

[0215] - Collection.nii.gz (Medullary tissue)

[0216] - RenalArtery.nii.gz (Renal Artery)

[0217] - Ureter.nii.gz (Ureter) 7

[0218] Voxel spacing: [1.0, 1.0, 3.0] mm

[0219] 2. Anatomical Analysis

[0220] Kidney separation: identified as right kidney

[0221] Spindle calculation: Direction [0.102, -0.123, 0.987], Length 142mm

[0222] Renal pelvis location: [165.3, 203.7, 285.4]

[0223] Ureteral selection: Right ureter, volume 2.3 cm³

[0224] 3. RENAL Score Calculation

[0225] R score: Tumor diameter 52mm → 2 points; E score: Exophytic percentage 35% → 2 points

[0226] N score: 8.2 mm from the ureter → 1 point 4 A score: 85% anterior → a

[0227] L-score: Completely at the lower end → 1 point 6

[0228] Final score: 2-2-1-a-1

[0229] Complexity: Low (out of 6 points).

[0230] 4. Result Output

[0231] Generate 3D interactive visualizations

[0232] Output clinical reports

[0233] Save detailed analysis data

[0234] Multiple tumor cases implemented

[0235] Case information:

[0236] Patient: MultiTumorCase

[0237] Diagnosis: Multiple renal tumors (3 in total) Treatment outcome:

[0238] Tumor test results:

[0239] Tumor 1: 15.3 cm³ → Right kidney (92% overlap)

[0240] Tumor 2: 8.7 cm³ → Left kidney (88% overlap)

[0241] Tumor 3: 3.2 cm³ → Extrarenal (15% overlap) [Ignore] 5

[0242] Independent rating results:

[0243] Tumor 1 (Right Kidney): 3-2-2-p-2 (Total Score 9, Medium Complexity)

[0244] Tumor 2 (Left Kidney): 2-1-1-a-1 (Total score 5, low complexity).

[0245] Case studies of treating very large tumors

[0246] Case information:

[0247] Patient: BaiYulu

[0248] Tumor voxels: 104,383

[0249] Tumor volume: 313.15 cm³ Algorithm selection and performance:

[0250] Tumor voxel count: 104,383 → Select hybrid strategy algorithm

[0251] Number of vertices in the convex hull: 387 (268 times dimensionality reduction)

[0252] Peak memory usage: 392 MB

[0253] Calculation time: 5.1 seconds

[0254] Maximum diameter: 142.3 mm.

[0255] The second aspect of this embodiment discloses a Renal scoring device for kidney tumors, including a data acquisition module, an anatomical analysis module, an R scoring module, an E scoring module, an N scoring module, an A scoring module, and an L scoring module.

[0256] The data acquisition module is used to acquire segmented data, which includes data on the kidneys, tumors, and blood vessels.

[0257] The anatomical analysis module is used to perform anatomical analysis based on segmented data, determine the main axis of the kidney, the location of the renal pelvis and ureter, and establish the natural coordinate system of the kidney.

[0258] The R-score module is used to calculate the maximum diameter of the tumor based on tumor voxels, and to determine the R-score based on the maximum diameter of the tumor.

[0259] The E-score module is used to construct the renal convexity based on the renal parenchyma coordinates and to determine the E-score based on the proportion of tumor voxels inside and outside the renal convexity.

[0260] The N-score module is used to treat the ureter as a collection system, determine the minimum distance from the tumor voxel to the collection system, and determine the N-score based on the minimum distance from the tumor voxel to the collection system.

[0261] The A-score module is used to construct a natural interface based on the location of the renal pelvis, and to determine the A-score based on this natural interface.

[0262] The L-score module is used to detect the polar lines based on the renal hilum indentation and determine the L-score based on the polar lines.

[0263] It should be noted that those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system or device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0264] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A Renal scoring method for renal tumors, characterized in that, include: Acquire segmentation data, which includes data on the kidneys, tumors, and blood vessels; Anatomical analysis based on segmented data was performed to determine the main axis of the kidney, the location of the renal pelvis and ureter, and to establish the natural coordinate system of the kidney. The maximum diameter of the tumor is calculated based on tumor voxels, and the R score is determined based on the maximum diameter of the tumor. The renal convex hull was constructed based on renal parenchymal coordinates, and the E score was determined based on the proportion of tumor voxels inside and outside the renal convex hull. Using the ureter as a collecting system, the minimum distance from the tumor voxel to the collecting system was determined, and the N score was determined based on the minimum distance from the tumor voxel to the collecting system. Construct a natural interface based on the location of the renal pelvis, and determine the A score based on this natural interface; Based on the detection of the polar lines in the renal hilum, the L score is determined according to the polar lines.

2. The Renal scoring method for renal tumors according to claim 1, characterized in that, Anatomical analysis based on segmented data was performed to determine the renal axis, renal pelvis location, and ureter, and to establish the renal's natural coordinate system, including: Based on the segmented data, the renal parenchyma coordinates are extracted and converted into physical coordinates. The PCA analysis method is used to calculate the directions of the three principal components: the first principal component corresponds to the long axis of the kidney, the second principal component corresponds to the short axis of the kidney, and the third principal component corresponds to the deep axis of the kidney. The sign of the Z component of the first principal component vector is determined. If it is negative, the first principal component vector is reversed. The renal pelvis is located based on the concave region on the surface of the kidney identified by convex hull analysis; the renal pelvis is located based on the centroid of the medullary cone; the renal pelvis is located based on the position of the renal artery and renal vein entering the kidney; the renal pelvis locations determined by the three methods are weighted and fused to obtain the final renal pelvis location. With the location of the renal pelvis as the center, search for ureteral branches within a preset radius; calculate the geometric distance from each ureteral branch to the renal pelvis and the number of voxels for each ureteral branch; calculate the score of the ureteral branch based on the geometric distance from the ureteral branch to the renal pelvis and the number of voxels for each ureteral branch; and determine the ureteral branch with the highest score as the target ureter. The kidney's long axis is defined as the Z-axis, its width axis as the X-axis, and its depth axis as the Y-axis, thus establishing a natural coordinate system for the kidney. The origin of this natural coordinate system is the geometric center of the kidney.

3. The Renal scoring method for renal tumors according to claim 1, characterized in that, The maximum diameter of a tumor is calculated based on tumor voxels, including: Detecting the number of tumor voxels; If the number of tumor voxels is less than the first preset value, the Euclidean distance between all point pairs in the tumor voxels is calculated, and the largest Euclidean distance is determined as the diameter of the tumor. If the number of tumor voxels is greater than or equal to the first preset value and less than the second preset value, then the three-dimensional convex hull of the tumor is calculated; when the number of vertices of the three-dimensional convex hull is less than or equal to the third preset value, the Euclidean distance between all pairs of vertices in the convex hull is calculated, and the largest Euclidean distance is determined as the diameter of the tumor; when the number of vertices of the three-dimensional convex hull is greater than the third preset value, the number of vertices in the convex hull of the third preset value is randomly sampled, the Euclidean distance between all pairs of sampled vertices in the convex hull is calculated, and the largest Euclidean distance is determined as the diameter of the tumor. If the number of tumor voxels is greater than or equal to the second preset value, the three-dimensional convex hull of the tumor is calculated. When the number of vertices of the three-dimensional convex hull is less than or equal to the third preset value, the Euclidean distance between all pairs of vertices in the convex hull is calculated, and the largest Euclidean distance is determined as the diameter of the tumor. When the number of vertices of the three-dimensional convex hull is greater than the third preset value, multiple rounds of random sampling are performed on all tumor voxels (e.g., 3 rounds of sampling). In each round, a fourth preset value number of tumor voxels are sampled, and the Euclidean distance between all pairs of vertices in the sampled tumor voxels is calculated. The largest Euclidean distance is determined as the diameter of the tumor.

4. The Renal scoring method for renal tumors according to claim 1, characterized in that, The renal convexity was constructed based on renal parenchymal coordinates, and the E score was determined based on the proportion of tumor voxels inside and outside the renal convexity, including: Extract the coordinates of the renal parenchyma; Construct the renal convex hull based on renal parenchyma coordinates; Delaunay triangulation was used to determine the position of each tumor voxel relative to the renal bulge. Count the number of tumor voxels inside and outside the renal bulge; The E score is determined based on the proportion of tumor voxels inside and outside the renal bulge. If the number of renal parenchyma voxels is less than or equal to the fifth preset value, then the renal parenchyma convex hull is constructed based on all points; if the number of renal parenchyma voxels is greater than the fifth preset value, then the fifth preset value number of points are randomly sampled, and then the renal parenchyma convex hull is constructed based on the sampled points.

5. The Renal scoring method for renal tumors according to claim 1, characterized in that, Using the ureter as a collecting system, the minimum distance from the tumor voxel to the collecting system is determined, and the N score is calculated based on this minimum distance, including: Treating the ureter as a collection system, create a collection system mask; Calculate the Euclidean distance from each tumor voxel to the ensemble system; Determine the minimum distance from tumor voxels to the assemblies; The N score is determined based on the minimum distance from the tumor voxel to the collection system.

6. The Renal scoring method for renal tumors according to claim 1, characterized in that, A natural interface based on the location of the renal pelvis is constructed, and the A score is determined based on this natural interface, including: Obtain the kidney's main axis and center; Calculate the vector of the renal pelvis relative to the center of the kidney, and take the cross product of the vertical component of this vector with the principal axis of the kidney to obtain the interface normal vector; The natural interface is defined as a plane passing through the center of the kidney, with its normal vector perpendicular to the main axis of the kidney and the direction of the renal pelvis. Calculate the directed distance from the tumor voxel to the natural boundary; Statistical analysis of the ratio of positive to negative values ​​in directed distances; The A score is determined based on the ratio of positive to negative values.

7. The Renal scoring method for renal tumors according to claim 1, characterized in that, Based on the renal hilum indentation, the polarity is detected, and the L score is determined according to the polarity, including: Morphological operations are used to fill the internal cavity of the kidney, extract the coordinate points on the kidney surface, construct a convex hull based on the coordinate points on the kidney surface, search for depression points, the depression points are inside the convex hull but not inside the filled kidney, and the renal hilum region is determined based on the depression points; Calculate the direction vector from each depression point to the renal pelvis, calculate the angle between the direction vector and the inward direction vector of the depression point, and retain the depression points corresponding to the angle being less than the eleventh preset value. Calculate the distance from each indentation point to the center of the kidney and the depth of the indentation, and retain the indentation points whose distance from the center of the kidney and the depth of the indentation are within a preset range; The K value is determined based on the set of renal hilar depression points. The K-nearest neighbor local density of each point in the set of renal hilar depression points is calculated. The median of the K-nearest neighbor local density of all points is taken as the threshold. Points with a density greater than the threshold are clustered by DBSCAN. The largest cluster is the core region of the renal hilum. Project the depression point of the core area of ​​the renal hilum onto the main axis of the kidney. Using the Z coordinate of the renal pelvis as the boundary, divide the depression point into upper and lower groups according to the position of the renal pelvis. Determine the upper and lower pole lines according to the percentile of the Z coordinate of the depression points in the upper and lower groups, and constrain the distance between the upper and lower pole lines within a preset range. The L score is calculated based on the positional relationship between the tumor and the superior and inferior poles.

8. The Renal scoring method for renal tumors according to claim 1, characterized in that, The Renal scoring method also includes: Perform format conversion and standardization on the segmented data; Unify the coordinate system of the segmented data; Separate the left and right kidney data from the segmented data; The test determines whether there are multiple tumors; if so, the tumors are separated. The process includes detecting whether there are multiple tumors. If there are multiple tumors, the tumors are separated, including: separating tumors using connected components; filtering out tumors with a volume smaller than a tenth preset value; calculating the number of overlapping voxels between the tumor and the left and right kidneys respectively; if the proportion of overlapping voxels between the tumor and a certain kidney is greater than 50%, then the kidney is marked as the kidney of the tumor; and sorting the tumors in descending order of volume.

9. The Renal scoring method for renal tumors according to claim 1, characterized in that, The Renal scoring method also includes: Voxel data of kidney, tumor, and blood vessel are downsampled, isosurfaces are extracted using the Marching Cubes algorithm to generate triangular meshes, coordinate transformations are performed on the triangular meshes to restore the physical scale, and three-dimensional surface models of kidney, tumor, and blood vessel are constructed. Render 3D surface models of kidneys, tumors, and blood vessels, overlay and display heatmaps of depressions, add epipolar markers, and generate an interactive 3D scene.

10. A Renal scoring device for kidney tumors, characterized in that, include: The data acquisition module is used to acquire segmented data, which includes data on kidneys, tumors, and blood vessels. The anatomical analysis module is used to perform anatomical analysis based on segmented data, determine the main axis of the kidney, the location of the renal pelvis and ureter, and establish the natural coordinate system of the kidney; The R-score module is used to calculate the maximum diameter of the tumor based on tumor voxels and determine the R-score based on the maximum diameter of the tumor. The E-score module is used to construct the renal convexity based on renal parenchyma coordinates and to determine the E-score based on the proportion of tumor voxels inside and outside the renal convexity. The N-score module is used to treat the ureter as a collecting system, determine the minimum distance from the tumor voxel to the collecting system, and determine the N-score based on the minimum distance from the tumor voxel to the collecting system. The A-score module is used to construct a natural interface based on the location of the renal pelvis, and to determine the A-score based on this natural interface. The L-score module is used to detect the polar lines based on the renal hilum indentation and determine the L-score based on the polar lines.

Images