Breast tumor growth trend prediction method and system based on big data

Abstract

The invention relates to the technical field of tumor prediction, in particular to a breast tumor growth trend prediction method and system based on big data, and the method comprises the following steps: obtaining a breast voxel sequence mark mutation region, carrying out the clustering recognition of an extension path, analyzing a density trend screening direction, positioning a diffusion hot spot region, and carrying out the modeling to generate a three-dimensional growth dynamic map. According to the method, through mutation point identification and multidirectional intersection analysis, the spatial identification capability of density variation of the mammary gland tissue is enhanced, a structural path with a stable growth trend is accurately extracted in combination with direction consistency clustering, a forward screening mechanism is introduced to perform density difference trend analysis, directivity expression in the direction of a path terminal is improved, and the accuracy of the density variation of the mammary gland tissue is improved. A diffusion hot spot area is positioned through direction included angle consistency calculation, the space focusing capacity of the diffusion trend is enhanced, a dynamic trend map is formed through three-dimensional track construction, visualization, continuity and resolution of the directional expansion trend of the breast tumor are enhanced, and prediction precision and space expression capacity are improved.

Description

Technical Field

[0001] This invention relates to the field of tumor prediction technology, and in particular to a method and system for predicting the growth trend of breast tumors based on big data. Background Technology

[0002] The field of tumor prediction technology encompasses research on the identification, analysis, and prediction of human tumor development using multi-source information such as clinical data, bioinformatics data, and medical image data. Its core content involves using interdisciplinary methods such as medical informatics, biostatistics, and artificial intelligence to model and infer the development and progression of tumors, assisting doctors in judging disease progression trends and formulating intervention plans. This covers the collection, processing, and analysis of tumor-related data, particularly in common tumor types such as breast, lung, and liver tumors, where comprehensive modeling of historical data and dynamic clinical characteristics enables the prediction of future disease progression trends.

[0003] Among them, the big data-based method for predicting the growth trend of breast tumors refers to using historical information such as structured and unstructured clinical data, imaging data, and follow-up records of breast tumor patients. By establishing time-series feature extraction rules and combining them with breast tumor growth-related indicators such as tumor size, morphological boundary changes, and density evolution, a time-dependent feature sequence dataset is constructed. Then, a sample association strategy is used to divide the tumor status into stages, and a prediction model is established using statistical learning methods to estimate the spatial volume change trend of breast tumors in the future. This is mainly accomplished by relying on data archiving and integration, feature parameter selection rules, and growth curve modeling based on time-series data.

[0004] Current methods for predicting breast tumor growth trends lack a dynamic perception mechanism for structural mutations in the spatial distribution of tumors, failing to reflect microscopic density anomalies along growth paths at the structural level. There is a lack of systematic screening methods for spatial continuity and directional consistency, relying solely on volume-level overall fitting of time-series variables, resulting in high fuzziness and uncertainty in directional judgment. The lack of directional stability detection at the terminals of potential expansion paths limits path extension representation to boundary positions, making it difficult to support spatial reconstruction of the true growth trajectory. Insufficient spatial directional consistency analysis capabilities prevent the construction of hotspot regions formed by multiple overlapping directions, restricting the directional extraction of diffusion trends. In the trend modeling stage, the inability to generate continuous dynamic structures leads to incomplete representation of the spatial trajectory of the growth evolution process, resulting in directional deviations and modeling gaps in the understanding of the true development trend of breast tumors. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a method and system for predicting the growth trend of breast tumors based on big data.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for predicting the growth trend of breast tumors based on big data, comprising the following steps:

[0007] S1: Obtain continuous structural voxel sequences in the three-dimensional structure of the breast, detect spatial mutation locations based on density gradient change points and mark multi-directional intersection regions to obtain a set of density gradient mutation points;

[0008] S2: Based on the set of density gradient mutation points, cluster the spatial proximity relationship between adjacent units to identify continuous structures with consistent orientation, mark potential growth paths, and obtain boundary expansion direction segments.

[0009] S3: Extract tumor density distribution from breast tumor regions, analyze density difference trends in potential expansion directions, screen directional segments with positive expansion trends, and obtain a set of stable path terminal directions;

[0010] S4: Based on the boundary expansion direction segment and the stable path terminal direction set, perform direction consistency analysis, select direction vectors within a preset angle threshold, identify direction intersection hotspot areas, locate potential diffusion directions in the tumor region, and obtain a set of tumor growth prediction vectors.

[0011] S5: Based on the tumor growth prediction vector set, locate the spatial trajectory path, perform trend continuity modeling within the complete region, and obtain a three-dimensional dynamic map of breast tumor growth trend prediction.

[0012] As a further aspect of the present invention, the density gradient mutation point set includes spatial variation extreme points, directional overlap intersection points, and density jump coordinate points; the boundary expansion direction segment includes stable vector segments, continuous boundary distribution, and potential migration paths; the stable path terminal direction set includes terminal vector directions, positive growth segments, and density trend directions; the tumor growth prediction vector set includes a consistent direction set, spatial hotspot regions, and diffusion trend vectors; and the three-dimensional dynamic map of breast tumor growth trend prediction includes a path trajectory network, a trend flow map, and a continuous growth prediction distribution map.

[0013] As a further aspect of the present invention, the specific steps for obtaining the set of density gradient abrupt change points are as follows:

[0014] S111: Obtain continuous structural voxel sequences in each spatial direction of the three-dimensional structure of the breast, calculate the density difference between each pair, identify the index position of the difference mutation point in the sequence, perform spatial coordinate position mapping, and generate a set of directional mutation coordinate points.

[0015] S112: Based on the set of coordinate points for abrupt changes in direction, compare the positions of the coordinate points for abrupt changes in different directions in the three-dimensional coordinate system, filter out the abrupt changes in different directions, and determine whether they constitute an intersection relationship in the three-dimensional space, thereby generating a set of intersection coordinate points.

[0016] S113: Based on the set of intersection coordinate points, reorder the coordinate values ​​of all intersection points, determine whether the spatial distance between adjacent points is continuous, filter the set of continuous region points, and obtain the set of density gradient abrupt change points.

[0017] As a further aspect of the present invention, the specific steps for obtaining the boundary extension direction segment are as follows:

[0018] S211: Based on the set of density gradient mutation points, obtain the three-dimensional coordinate values ​​between each point, determine whether the point pairs constitute a spatial proximity relationship, perform connection and classification operations on the point pairs that satisfy the proximity relationship, and obtain a set of spatial aggregated points.

[0019] S212: Based on the spatial aggregation point set, obtain the density difference sequence between voxels in each aggregation unit, calculate the change trend of density value with spatial coordinates, filter the aggregation voxel sequence that meets the unidirectional condition, and generate directional continuous point segments.

[0020] S213: Based on the directional continuity point segments, calculate the difference in the directional vector change angle between adjacent voxel segments, filter the continuous voxel regions, determine the spatial range between the start and end segments in the continuous regions that meet the conditions, and obtain the boundary extension direction segment.

[0021] As a further aspect of the present invention, the specific steps for obtaining the stable path terminal direction set are as follows:

[0022] S311: Extract voxel density values ​​within the three-dimensional structure from the breast tumor region, obtain the spatial coordinates between adjacent voxels, construct a continuous voxel chain with spatial consistency and density similarity, and obtain a set of structural connectivity paths;

[0023] S312: Based on the set of structural connected paths, detect the density value change sequence of continuous voxels in the spatial direction adjacent to the terminal position of each path, analyze the consistency of density change direction, and obtain the terminal trend path segment;

[0024] S313: Based on the terminal trend path segment, and obtain the density difference between continuous voxels and the spatial step distance in the corresponding vector direction, calculate the density change rate on each directional vector, filter out directional segments with positive density change rate values, mark them as path extension trend directions, and obtain a stable path terminal direction set.

[0025] As a further aspect of the present invention, the density change rate is expressed by the formula:

[0026]

[0027] Calculations are performed, in which, This represents the rate of change of density in the directional segment. This represents the density value of the i-th voxel in the k-th terminal trend path segment. Represents the density value of the next adjacent voxel. This represents the spatial step distance between the i-th pair of adjacent voxels in the k-th path. This represents the average value of all voxel density values ​​in the k-th path. Let n represent the average step distance of all steps along the k-th path. k This represents the number of voxel pairs involved in the calculation in the k-th path segment.

[0028] As a further aspect of the present invention, the specific steps for obtaining the tumor growth prediction vector set are as follows:

[0029] S411: Based on the boundary extension direction segment and the stable path terminal direction set, calculate the angle between vector pairs, filter the direction vector combinations with angle values ​​less than a set angle threshold, and obtain a set of direction consistency vectors.

[0030] S412: Based on the set of direction consistency vectors, extract the spatial coordinate points corresponding to each vector combination and perform three-dimensional mesh division, and count the frequency of the direction vector in each mesh cell, extract the center position coordinates, and obtain the set of coordinate points in the direction concentration area.

[0031] S413: Based on the set of coordinate points in the direction concentration area, extract the vector directions that appear most frequently in the direction vector group corresponding to each coordinate point, calculate the average difference of the angle values ​​of the vectors in the same region, and filter the vector directions whose average difference is lower than the vector consistency fluctuation threshold. Mark them as potential spread paths of the tumor region according to the region to obtain a set of tumor growth prediction vectors.

[0032] As a further aspect of the present invention, the average difference is expressed by the formula:

[0033]

[0034] Perform the calculation, where D r This represents the average angle consistency value. Let G represent the unit form of the g-th direction vector in the r-th region, where G represents the number of direction vectors in region r. q represents the result of the dot product between the g-th and j-th vectors. g This represents the number of times the g-th direction vector appears within the region. This represents the average number of occurrences of all directional vectors within the region.

[0035] As a further aspect of the present invention, the specific steps for obtaining the three-dimensional dynamic atlas for predicting the growth trend of breast tumors are as follows:

[0036] S511: Based on the tumor growth prediction vector set and combined with the spatial coordinate set of three-dimensional structural voxels, locate continuously connected spatial voxel units in the three-dimensional voxel network according to the vector direction, integrate the continuously connected voxel paths into a linear trajectory structure, and obtain the spatial trajectory path set.

[0037] S512: Based on the spatial trajectory path set, extract the three-dimensional coordinate distribution in the voxel sequence connected by each path, calculate the path curvature change rate, merge paths in the same direction and reorder the voxel node numbers to obtain continuous spatial trajectory segments.

[0038] S513: Based on the continuous spatial trajectory segments, extract the density values, spatial step lengths and temporal change information of the nodes in each path segment, construct a density change time series and superimpose the spatial location index, establish the temporal evolution trend of voxel density in the path order, and obtain a three-dimensional dynamic map for predicting the growth trend of breast tumors.

[0039] A breast tumor growth trend prediction system based on big data includes:

[0040] The structural mutation detection module acquires the three-dimensional structure voxel sequence of the breast, calculates the density gradient values ​​of adjacent voxels, detects spatial mutation locations, marks spatial intersection regions, and obtains a set of density gradient mutation points.

[0041] The spatial orientation filtering module, based on the set of density gradient mutation points, compares the spatial proximity and directional angle values ​​between mutation points, clusters and extracts stable orientation segments, and obtains boundary expansion orientation segments.

[0042] The path trend extraction module analyzes the density change trend within the connected path based on the voxel density data of the tumor region, extracts the positive expansion segment, and obtains the set of stable path terminal directions.

[0043] The directional intersection analysis module calculates the vector angle between the boundary extension direction segment and the stable path terminal direction set, filters direction pairs with an angle less than a consistency threshold, statistically locates intersection hotspots, and obtains a set of tumor growth prediction vectors based on the frequency of the boundary extension direction segment and the stable path terminal direction set.

[0044] The three-dimensional trajectory modeling module determines the connection path and directional stability of vector segments based on the tumor growth prediction vector set, constructs a continuous spatial trajectory, and generates a three-dimensional dynamic map of breast tumor growth trend prediction.

[0045] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0046] In this invention, mutation point identification and multi-directional intersection analysis enhance the spatial recognition ability of breast tissue density variations. Combined with directional consistency clustering, structural paths with stable growth trends are accurately extracted. A positive screening mechanism is introduced to analyze density difference trends, improving the directional expression of path terminal directions. The consistency calculation of directional angles locates diffusion hotspots, strengthening the spatial focusing ability of diffusion trends. A three-dimensional trajectory is constructed to form a dynamic trend map, realizing a closed-loop expression from structural mutation perception to spatial evolution modeling. Overall, it enhances the visualization, continuity, and resolution of the directional expansion trend of breast tumors, improving prediction accuracy and spatial expression ability. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the steps of the present invention;

[0048] Figure 2 This is a flowchart illustrating the process of obtaining the set of density gradient mutation points in this invention.

[0049] Figure 3 This is a flowchart illustrating the process of obtaining the boundary extension direction segment of the present invention.

[0050] Figure 4 This is a flowchart illustrating the process of obtaining the stable path terminal direction set according to the present invention.

[0051] Figure 5 This is a flowchart illustrating the process of obtaining the tumor growth prediction vector set in this invention.

[0052] Figure 6 This is a flowchart illustrating the process of obtaining a three-dimensional dynamic atlas for predicting the growth trend of breast tumors according to the present invention.

[0053] Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0055] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the 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, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0056] Please see Figure 1 A method for predicting the growth trend of breast tumors based on big data includes the following steps:

[0057] S1: Obtain continuous structural voxel sequences in each spatial direction of the three-dimensional structure of the breast. Based on the density gradient change points of adjacent units in the sequence (the mutation locations detected by calculating the density difference between adjacent voxels, reflecting the areas of significant changes in tissue density), detect spatial mutation locations (spatial coordinate points in three-dimensional space where the tumor density characteristics change significantly) and mark them in the multi-directional intersection area in the three-dimensional coordinate system (the overlapping area of ​​analysis results in different directions in the three-dimensional coordinate system). Combine the spatial intersection point set to form a continuous distribution point set, and obtain the density gradient mutation point set.

[0058] S2: Based on the set of density gradient mutation points, cluster the spatial proximity relationship between adjacent units (based on Euclidean distance or topological connection) to extract the density difference trend in the outer region of the continuous structure, identify the continuous structure with directional consistency, screen the stable direction vector field region (the continuous region where the change amplitude of the spatial vector direction is less than the set threshold) and mark it as a potential growth path to obtain the boundary expansion direction segment.

[0059] S3: Extract tumor density distribution from breast tumor regions, construct structural connectivity paths (connected regions established based on density similarity and spatial continuity), identify stable expansion trends at the terminals of each path, and filter out direction segments with positive expansion trends by detecting density difference trends in potential expansion directions of terminal regions to obtain a set of stable path terminal directions.

[0060] S4: Based on the boundary expansion direction segment and the stable path terminal direction set, perform directional consistency analysis on the spatial distribution relationship of the two types of directions, select directional vectors within the preset angle threshold (angle tolerance parameter for whether the direction is consistent, usually set to 15°-30°) and sort them according to the frequency of occurrence, identify the directional intersection hotspot area (spatial area where the results of multiple directional analysis sets are significantly overlapping), locate the potential diffusion direction in the tumor area, and obtain the tumor growth prediction vector set;

[0061] S5: Based on the directions marked in the tumor growth prediction vector set, locate the spatial trajectory path in the three-dimensional structure, construct the growth trajectory according to the spatial continuity distribution of the direction segments, perform trend continuity modeling within the complete region, and obtain a three-dimensional dynamic map of breast tumor growth trend prediction.

[0062] The density gradient mutation point set includes spatial variation extreme points, directional overlap intersection points, and density jump coordinate points. The boundary expansion direction segment includes stable vector segments, continuous boundary distribution, and potential migration paths. The stable path terminal direction set includes terminal vector directions, positive growth segments, and density trend directions. The tumor growth prediction vector set includes a consistent direction set, spatial hotspot regions, and diffusion trend vectors. The three-dimensional dynamic map of breast tumor growth trend prediction includes a path trajectory network, trend flow map, and continuous growth prediction distribution map.

[0063] Please see Figure 2 The specific steps of S1 are as follows:

[0064] S111: Obtain continuous structural voxel sequences in each spatial direction of the three-dimensional structure of the breast, calculate the density difference between each pair, identify the index position of the difference mutation point in the sequence, and map the position to the spatial coordinate point in the corresponding direction in the three-dimensional coordinate system to generate a set of directional mutation coordinate points.

[0065] To obtain continuous voxel sequences in each spatial direction of the three-dimensional structure of the breast, three-dimensional reconstruction must first be performed based on the original DICOM data. A voxel raster matrix is ​​constructed to obtain continuously arranged voxel sequences in each direction (X, Y, Z). For each sequence, the corresponding gray value of each voxel is obtained as a density index. The gray value is directly mapped from the pixel intensity. According to the formula ρ(i) = G(i), where ρ(i) is the density value of the i-th voxel and G(i) is its gray intensity value, the density difference Δρ(i) = |ρ(i+1) - ρ(i)| between any two adjacent voxels i and i+1 is calculated sequentially, forming a density difference sequence. Then, each Δρ(i) in this difference sequence is evaluated. If the mutation threshold Tρ is exceeded, and Δρ(i)≥Tρ, the point is marked as a mutation point. The Tρ value is set as a combination of the standard deviation σρ and the mean μρ, such as Tρ=μρ+1.5σρ. For example, if μρ=20 and σρ=5 in a directional sequence, then Tρ=27.5. When the difference of a point is 30, it is marked as a mutation point, and its index position i is recorded. The three-dimensional coordinates (x, y, z) are obtained by directional mapping transformation of the index in the original voxel matrix. For example, if the current direction is the X-axis, and the i-th voxel is located at (10+i, 25, 30), then its corresponding coordinates are (10+i, 25, 30). Finally, all coordinate points that meet the mutation conditions are added to the directional mutation coordinate point set.

[0066] S112: Based on the set of coordinate points of directional change, compare the positions of the coordinate points of change in different directions in the three-dimensional coordinate system, filter out the directional change points that are adjacent in space, and determine whether they form an intersection relationship in the three-dimensional space to generate a set of intersection coordinate points.

[0067] Based on the coordinate set of abrupt change points, pairwise comparisons are performed on the coordinate sets of abrupt change points in the X, Y, and Z directions, using the Cartesian distance formula. Calculate the spatial distance between adjacent abrupt change points. When the distance d is less than the spatial proximity threshold Td, they are determined to be adjacent abrupt change points. The Td value can be set according to the scanning accuracy and the actual spatial size of the tissue. For example, under the condition of 0.5mm resolution, Td = 1.5mm is set, which means that a maximum of 3 voxels are allowed to form an adjacent relationship in physical space. Then, these adjacent abrupt change points are further judged to determine whether they form an intersection relationship. That is, it is judged whether the abrupt change points of any two of the three directions are simultaneously adjacent to a point in the third direction. For example, the abrupt change points in the X direction are (20, 30, 40), the abrupt change points in the Y direction are (21, 30, 39), and the abrupt change points in the Z direction are (20, 31, 41). Since the distance between each pair of the three is less than 1.5mm, they are determined to form an intersection relationship. The intersection coordinates (20.33, 30.33, 40) are obtained by averaging the intersection positions of the three points and adding them to the intersection coordinate point set.

[0068] S113: Based on the set of intersection coordinate points, the coordinate values ​​of all intersection points are reordered according to the coordinate axis direction, and a continuity threshold is set to determine whether the spatial distance between adjacent points is continuous. The set of continuous region points is filtered to obtain the set of density gradient abrupt change points.

[0069] Based on the set of intersection coordinate points, all intersection points are sorted in ascending order of X, Y, and Z coordinate values. For each direction's coordinate sequence after sorting, the directional distance Δd(i) = coord(i+1) - coord(i) between adjacent coordinate points is calculated point by point. A continuity threshold Tc is set to judge Δd(i). When Δd(i) ≤ Tc, it is determined to be a continuous point. The Tc value is set with reference to the density of adjacent structures in breast tissue. For example, Tc = 1 mm. Based on this, a window method is used to traverse all coordinate points. A coherent region is constructed. If the number of consecutive points in a window exceeds the set lower limit Nmin, then all points in the window are considered a coherent region. Nmin can be 10, indicating that at least 10 consecutive points form a region. For example, in the X direction, if the coordinates are [100, 100.8, 101.7, 102.5, 105.6], then Δd is [0.8, 0.9, 0.8, 3.1], and the first 3 have a spacing of less than Tc = 1, forming a continuous region. The coordinates of this region are extracted and added to the density gradient abrupt change point set.

[0070] Please see Figure 3 The specific steps of S2 are as follows:

[0071] S211: Based on the set of density gradient mutation points, obtain the three-dimensional coordinate values ​​between each point, determine whether the point pairs constitute a spatial proximity relationship according to Euclidean distance and topological connection rules, perform connection and classification operations on the point pairs that satisfy the proximity relationship, and obtain a spatial aggregation point set;

[0072] Based on the density gradient mutation point set, the three-dimensional coordinate values ​​of all mutation points are first extracted. Let each point P... k The corresponding spatial coordinates are (x k y k , z k ), traverse any two points (P) in the set of mutation points i P j ), respectively call its coordinate values ​​to calculate the Euclidean distance. The result is then compared with the spatial proximity threshold Td. When d(i,j)≤Td, P is considered to be P. i With P j The points are considered to be adjacent to each other. The value of Td is set with reference to the average spacing between voxels in adjacent tissues. When the scanning accuracy is 0.5 mm, Td can be 1.5 mm, which means that spatial connections are allowed between 3 voxels. For example, P1 = (10.0, 20.0, 30.0) and P2 = (10.5, 20.3, 30.2). The calculated value of d(1,2)≈0.61 mm is less than Td, so it is considered to be adjacent. Next, according to the topological connection rules, the adjacency list of each point is constructed. Points with mutual adjacency are classified, that is, all points that are directly or indirectly connected are grouped into the same cluster. A layer-by-layer search method is used to recursively search for its neighboring points for each point and add them to the same category. The search continues based on the newly added points until no more new points are added to the current category. For example, if P1 is adjacent to P2 and P2 is adjacent to P3, then P1, P2, and P3 are grouped into one category. This process is repeated for all point sets, and finally multiple spatial cluster point sets are obtained.

[0073] S212: Based on the spatial aggregation point set, obtain the density difference sequence between voxels in each aggregation unit, calculate the change trend of density value with spatial coordinates according to the voxel arrangement order, determine whether the continuous change direction maintains unidirectional consistency, filter the aggregation voxel sequence that meets the unidirectional condition, and generate directional continuity point segments.

[0074] Based on the spatial aggregation point set, for each aggregation unit, a sequence is constructed according to the voxel points in their arrangement order in three-dimensional space. Based on the density value ρ(i) corresponding to each voxel coordinate, the density difference sequence Δρ(i) = ρ(i+1) - ρ(i) is calculated. A density change curve is formed according to the voxel numbering order, and the directionality of the density change trend is analyzed. The sign sequence of Δρ(i) is calculated, and it is determined whether it remains continuously positive or negative. When the sign sequence is continuously positive or continuously negative within a certain segment, that segment is considered... For a unidirectional consistent segment, a consistency threshold Lmin is set, which represents the minimum length of consecutive identical signs. For example, Lmin = 5 means that at least 5 consecutive density rising or falling points constitute a consistent segment. For example, if the density value sequence is [22, 24, 27, 30, 33, 35, 34], then Δρ = [+2, +3, +3, +3, +2, -1]. The first 5 terms are positive, satisfying Lmin = 5, thus forming a directional continuous segment. Its voxel index is used to construct point segments, and the corresponding coordinate points are extracted to finally form a set of directional continuous point segments.

[0075] S213: Based on the directional continuity point segments, calculate the difference in the directional vector change angle between adjacent voxel segments, filter the continuous voxel regions, determine the spatial range between the start and end segments in the continuous regions that meet the conditions, and obtain the boundary extension direction segments.

[0076] Based on the directional continuity of the point segments, adjacent point segments D are processed one by one. i With D i+1 Calculate their respective direction vectors. The direction vector V is represented as the coordinate difference vector between the end point and the beginning point of the segment, V = (x... e -x s y e -y s , z e -z s Then, according to the formula for the angle between vectors, cosθ=(V i ·V i+1 ) / (|V i ||V i+1 |) Calculate the angle of change in direction θ i Using threshold θ m Set the criteria for determining continuous regions, θ m For the angular offset tolerance, if θ i ≤θ m If θ is the direction segment, then it is determined to be a continuous direction segment. m A value not exceeding 30° indicates that a segment's directional deflection not exceeding this angle can be considered continuous. For example, if D... i =[(10,10,10),(15,15,15)],D i+1 = [(15, 15, 15), (20, 15, 15)], then V i= (5, 5, 5), V i+1 = (5, 0, 0), calculate θ≈54.74°, which is greater than θ m =30°, does not constitute a continuous region, conversely if D i+1 If the endpoint is (20, 20, 20), then V i+1 = (5, 5, 5), θ = 0°, which satisfies the condition. Then, the coordinates of all voxel points in the starting and ending segments of the continuous region are extracted, and the coordinates of the minimum and maximum boundary points are calculated. These are defined as the spatial boundary of the continuous region. The vector direction is then extended by a certain range d to expand the boundary direction segment. For example, d is taken as 3mm. Spatial coordinate extension is performed on the original direction vector to obtain the boundary expansion direction segment.

[0077] Please see Figure 4 The specific steps of S3 are as follows:

[0078] S311: Extract voxel density values ​​within the three-dimensional structure from the breast tumor region, obtain the spatial coordinates between adjacent voxels, construct a continuous voxel chain with spatial consistency and density similarity, and obtain a set of structural connectivity paths;

[0079] To extract voxel density values ​​within a three-dimensional structure of a breast tumor region, firstly, a set of all voxels in the tumor region in three-dimensional space is extracted based on medical imaging data. The grayscale value of each voxel is then used as the density parameter ρ(i), and its corresponding spatial position (x, y, y) in the three-dimensional coordinate system is indexed. i y i , z i Construct a voxel position relationship matrix, then compare the positions of all voxels pairwise to determine if there is a direct adjacency relationship. The adjacency condition is set as the Manhattan distance d between two voxels. m (i, j) = |x i -x j |+|y i -y j |+|z i -z j| = 1 indicates a connection is formed within the six-neighborhood. For voxel pairs satisfying the adjacency relationship, they are further screened according to whether their density difference Δρ(i, j)=|ρ(i)-ρ(j)| is less than the similarity threshold Tρ. This threshold is set according to the density fluctuation range of the tumor tissue. If the average density μρ is 40 and the standard deviation σρ is 5, then Tρ = 2σρ = 10, indicating that if the density difference is less than 10, the densities are considered similar. For example, voxels A and B are (ρ = 42, x = 10, y = 20, z = 30) and (ρ = 47, x = 10, y = 21, z = 30) respectively. They satisfy the adjacency relationship and Δρ = 5 < Tρ, forming a density-continuous voxel pair. On this basis, starting from any voxel, a depth-first traversal method is used to connect all voxels that meet the adjacency and density conditions in turn, forming a continuous voxel chain. The same operation is performed on all starting points to obtain the global structure connectivity path set.

[0080] S312: Based on the structure connectivity path set, extract the spatial positions and density values of the voxels at the ends of each path, detect the density value change sequences of the continuous voxels in the spatial directions adjacent to the terminal positions of each path, analyze the consistency of the density change directions, screen the paths with monotonically increasing or decreasing trends, and obtain the terminal trend path segments;

[0081] Based on the structure connectivity path set, traverse each path P k , extract the position coordinates (x e , y e , z e ) and its density ρ e of the voxel points at its end. According to the six-directional spatial expansion rule (±X, ±Y, ±Z), access its adjacent voxels respectively. For the voxel points continuously existing in each direction, read their density values in turn to construct a density sequence [ρ e+1 , ρ e+2 ,..., ρ e+n . Calculate the density difference sequence Δρ n = ρ e+n - ρ e+n-1 , extract the sign of its difference to form a sign sequence S = [sign(Δρ1), sign(Δρ2),..., sign(Δρ n-1 ). Set the minimum length of the analysis segment Lmin = 5, screen the directions with the number of consecutive positive or negative signs not less than Lmin, and judge them as monotonous trend directions. For example, the density of the path end point is 35, and the densities of the adjacent points are [37, 39, 42, 44, 47]. Δρ = [+2, +2, +3, +2]. All signs are positive and the number is 4. If the starting point is included, there are a total of 5 points, meeting the Lmin condition, and it is determined as a monotonically increasing trend path. Mark the voxel points of this segment as terminal trend path segments, and repeat the process for all path ends. Finally, obtain the set of all terminal trend path segments.

[0082] S313: Based on the terminal trend path segment, extract its corresponding direction vector, and obtain the density difference between continuous voxels and the spatial step distance in the corresponding vector direction. Calculate the density change rate on each direction vector, filter the direction segments with positive density change rate values, mark them as path extension trend directions, and obtain the stable path terminal direction set.

[0083] The rate of change of density is expressed by the formula:

[0084]

[0085] Calculations are performed, in which, This represents the rate of change of density in the directional segment. This represents the density value of the i-th voxel in the k-th terminal trend path segment. This represents the density value of its next adjacent voxel. This represents the spatial step distance between the i-th pair of adjacent voxels in the k-th path. This represents the average value of all voxel density values ​​in the k-th path. Let n represent the average step distance of all steps along the k-th path. k This represents the number of voxel pairs involved in the calculation in the k-th path segment.

[0086] In three-dimensional imaging data of breast tumors, the gray values ​​of each voxel in the tumor region are obtained through MRI scanning equipment and quantified into density values. Gray values ​​in 8-bit grayscale images typically range from 0 to 255. Breast tumor regions, after image annotation and region segmentation, show stable density values ​​concentrated in the interval [30, 50]. Spatial step distance. It is obtained by calculating the difference in voxel coordinates, which satisfies the adjacency definition of Manhattan distance of 1, and the maximum change value is 2.

[0087] Taking the k-th path segment as an example, which contains 5 voxels, the data is as follows:

[0088]

[0089] Calculate the average density:

[0090]

[0091] Calculate the average step distance:

[0092]

[0093] Calculate the numerator:

[0094]

[0095] Calculate the first term in the denominator:

[0096]

[0097]

[0098] Calculate the second term in the denominator:

[0099]

[0100] Total of denominators:

[0101] 7.29 + 1.5 + 1 = 9.79;

[0102] The final calculated density change rate value is:

[0103]

[0104] The results indicate that the density change rate of the path segment in the direction of density growth trend is 1.225, suggesting a certain degree of continuity and consistency in the density change of the path in spatial direction. In the three-dimensional structural analysis of breast tumors, paths with higher density change rate values ​​correspond to the growth direction or spread trend of the tumor. Therefore, directional segments with positive density change rate values ​​can be marked as the path extension trend direction to obtain a set of stable path terminal directions.

[0105] The density change rate reflects the intensity and continuity of density expansion in the three-dimensional path segment of breast tumors in spatial direction. The higher the value, the more continuous and significant the density value of the path segment in the corresponding direction shows an increasing or decreasing trend with spatial advancement, accompanied by small density fluctuations and spatial structural disturbances. Therefore, it can be regarded as an indicator that the tumor tissue has a stable structural extension or growth trend along this direction. Conversely, a low density change rate indicates that the density change in the path is irregular or the spatial step is uneven, and the structure is discrete or inconsistent, lacking stable extension direction characteristics. As a comprehensive measure of structural and density information, this index can be used to screen tumor tissue connectivity directions with continuity and consistency, thereby helping to identify potential tissue evolution paths.

[0106] formula The computational logic reflects a joint assessment of the density variation trend and spatial relationship within the path segment. Its molecular part calculates the density difference between each pair of continuum voxels. Corresponding spatial step distance The summation of the products reflects the cumulative intensity of density changes in spatial directions. The multiplication operation gives the density change value a higher weight over a longer spatial span, thus more accurately reflecting the density extension trend along the path. The absolute value is then taken to ensure that the evaluation is of the intensity of change rather than the positive or negative direction. The denominator first sums the squares of the deviations of the density values ​​of all voxels from the mean and then takes the square root to form the dispersion index of density change. The square root operation is used to restore the original units for easy comparison with the numerator. The second term is the sum of the absolute values ​​of the deviations of each spatial step distance from its mean, reflecting the spatial balance of the path structure. If the step distance fluctuates greatly, this term will be larger. The overall denominator is normalized by adding the changes in the two directions and adding 1 (to prevent the denominator from being zero and to perform minimum normalization). Finally, the ratio of the numerator to the denominator corresponds the intensity of density change in the direction to the degree of density and structural fluctuation of the path, thus reflecting whether the path exhibits a continuous, stable and obvious density expansion trend in space.

[0107] Please see Figure 5 The specific steps of S4 are as follows:

[0108] S411: Based on the boundary extension direction segment and the stable path terminal direction set, extract the three-dimensional coordinate starting point and direction unit vector of each direction vector, calculate the angle between vector pairs, compare it with the set direction consistency angle threshold, filter the direction vector combination within the angle value less than the set angle threshold, and obtain the direction consistency vector set.

[0109] Based on the boundary extension direction segment and the stable path terminal direction set, the starting three-dimensional coordinates (x0, y0, z0) and the direction unit vector V = (vx, vy, vz) of each direction segment are extracted. The unit vector is obtained by normalizing the starting vector after subtracting the ending vector from the beginning vector of the direction segment. Next, for any two vector pairs (V) in the direction segments i V j ) Calculate the included angle using the formula cosθ=(V i ·V j ) / (||V i ||||V j Since the vector has been normalized, it simplifies to cosθ=V i ·V j The calculated θ is consistent with the set direction threshold θ. t Comparison, θ t The maximum permissible deviation angle is typically set to θ. t=15°, that is, to determine if cosθ≥cos15°≈0.9659 is a consistent direction combination. For example, if V1=(0.577, 0.577, 0.577) and V2=(0.6, 0.6, 0.529), then the dot product is approximately 0.997, which satisfies the consistency condition. This vector combination is added to the set of vectors with consistent directions. The above calculation is performed on all vector combinations, and finally all the included angles less than θ are obtained. t A combination of directional consistency vectors.

[0110] S412: Based on the set of direction-consistent vectors, extract the spatial coordinate points corresponding to each vector combination, divide each coordinate point into a three-dimensional grid according to its spatial location, count the frequency of the direction vector in each grid cell, extract the center position coordinates, and obtain the set of coordinate points in the direction-concentrated region.

[0111] Based on a set of direction-consistent vectors, for each pair of vectors, the starting coordinates (x, y, z) of the corresponding direction segment are extracted. All starting coordinates are then divided into a 3D mesh using a fixed step size δ. Let δ = 2 mm, then each coordinate point is located to a specific mesh cell (G). x G γ G z Simultaneously, a direction vector statistics table is established for each grid, recording all direction vectors within that grid in the corresponding table entries. If the number of direction vectors in a certain grid exceeds a frequency threshold N, the result is considered valid. t Then it will participate in subsequent statistics, N t The data density is set to 5, meaning that a region is considered a concentrated region of direction only when there are at least 5 vectors with consistent direction within the same grid. For example, if there are 8 direction vectors in a grid cell (10, 15, 12), which meets the frequency requirement, the average value of the coordinate points corresponding to these vectors is calculated as the grid center point. For example, if the coordinate set is [(20, 30, 40), ... (25, 34, 43)], the calculated center is (22.5, 32, 41.5), which is added to the coordinate point set of the concentrated region of direction.

[0112] S413: Based on the set of coordinate points in the direction concentration area, extract the vector directions that appear most frequently in the direction vector group corresponding to each coordinate point, calculate the average difference of the angle values ​​of the vectors in the same region, and filter the vector directions whose average difference is lower than the vector consistency fluctuation threshold. Mark them as potential spread paths of the tumor region according to the region to obtain the tumor growth prediction vector set.

[0113] The average difference is calculated using the following formula:

[0114]

[0115] Perform the calculation, where D r This represents the average angle consistency value. Let G represent the unit form of the g-th direction vector in the r-th region, where G represents the number of direction vectors in region r. q represents the result of the dot product between the g-th and j-th vectors. g This represents the number of times the g-th direction vector appears within the region. This represents the average number of occurrences of all directional vectors within the region.

[0116] In determining the spatial diffusion direction of tumors, the unit vector group within the direction concentration region is used to measure spatial consistency. The number is obtained by statistically analyzing the voxel direction segments reconstructed after imaging. Taking the number of direction vectors in a certain region as G=4, the unit direction vectors calculated after acquisition are as follows:

[0117]

[0118] The above values ​​have been normalized from the coordinate vectors from the starting point to the ending point. Directional frequencies are recorded statistically through a 3D mesh and are denoted as:

[0119] q1 = 6;

[0120] q2 = 5;

[0121] q3 = 7;

[0122] q4 = 6;

[0123] Calculate the average frequency:

[0124]

[0125] Calculate the directional consistency term, which is the average of the dot products between the vectors:

[0126]

[0127]

[0128] The sum of the dot products of all combinations is:

[0129] ∑=0.9975+0.9853+1.0+0.9741+0.9975+0.9853=5.9397;

[0130] Substitute into the formula to calculate the first term:

[0131]

[0132] Calculate the frequency equalization term:

[0133]

[0134] Finally, combining the two items, we get:

[0135] Dr =0.494975 + 1.4142 = 1.909175;

[0136] The results indicate that the average angle between the directional vectors in this region is small and the frequency difference is low, indicating that it belongs to a region with high directional consistency. The calculation results reflect that the directional structure here is coherent and the spatial trend is stable, which can be used as a candidate direction for predicting tumor growth paths.

[0137] The average angle consistency value represents the combined characteristics of spatial convergence and distribution stability among all directional vectors within the same directional set region. This value comprehensively evaluates the closeness of the angles between directional vectors and their frequency balance within the local region. The lower the value, the smaller the angular deviation between directions in the region, the closer the directions are, and the more even the frequency distribution of each vector is, indicating a more consistent overall spatial structure. A higher value means greater directional dispersion or stronger frequency fluctuations, reflecting a lack of a unified diffusion trend in the region. Therefore, this index can be used to identify directional sets in tumor regions that have spatial coherence and the potential for frequent expansion.

[0138] Formula D r The computational logic consists of two parts. The first is the average absolute value of the dot product between all pairs of directional vectors within the region. This part is achieved by performing the dot product on all different vector combinations and taking the absolute value, so that vector combinations with smaller angles between them contribute more. This is then normalized and divided by the total number of combinations to eliminate the influence of sample size, and is used to measure the overall directional consistency. The second is the square root of the total deviation between the directional frequency and the regional average frequency. This part is achieved by calculating the absolute deviation between the frequency of each vector and the average value, summing them, and finally taking the square root to quantify the frequency distribution dispersion to a unified scale, reflecting whether the vector occurrence distribution is concentrated. The two parts are combined by addition to reflect the spatial consistency of the direction itself and comprehensively evaluate the stability of the direction's occurrence within the region, ensuring that the evaluation results consider both the directional trend and its density within the region.

[0139] Please see Figure 6 The specific steps of S5 are as follows:

[0140] S511: Based on the directions marked in the tumor growth prediction vector set, extract the starting position and spatial unit vector direction of each vector, combine the spatial coordinate set of three-dimensional structural voxels, and gradually locate the continuously connected spatial voxel units in the three-dimensional voxel network according to the vector direction, and integrate the continuously connected voxel paths into a linear trajectory structure to obtain the spatial trajectory path set.

[0141] Based on the directions marked in the tumor growth prediction vector set, the starting point coordinates (x0, y0, z0) and unit direction vector V = (vx, vy, vz) of each vector are extracted. Using the known voxel coordinate set in the 3D structured voxel mesh, positioning is performed point-by-point along the V direction from the starting point with a fixed spatial step size Δd. At each step, the current position coordinates (x0, y0, z0, vz) are... i ,yi,z i Adding Δd·V forms a new target point (x) i+1 y i+1 , z i+1 The process searches the 3D voxel grid for the existence of a voxel corresponding to a given point. If it exists, it is recorded as a continuous connection point, and the search continues in the same direction from that point until the continuous search fails or the maximum path length Nmax is reached. Δd is set according to the image resolution, typically 1 mm. Nmax is set to 20 steps, indicating that the maximum path length does not exceed 20 mm. For example, for vectors... Starting from point (10, 10, 10), the first step is to obtain (10.58, 10.58, 10.58). Check if there is a voxel under this approximate coordinate. If there is, continue to the next point. After repeating the operation, a path segment composed of multiple continuous voxels is formed. Each path records the voxel number in chronological order to form a spatial linear trajectory structure. All prediction vectors are processed in sequence, and finally all the formed continuous path segments are integrated to construct a spatial trajectory path set.

[0142] S512: Based on the spatial trajectory path set, extract the three-dimensional coordinate distribution in the voxel sequence connected by each path, calculate the path curvature change rate according to the direction change angle between adjacent coordinates, merge paths in the same direction and reorder the voxel node numbers to obtain continuous spatial trajectory segments.

[0143] Based on the spatial trajectory path set, the three-dimensional coordinate sequence of voxel points contained in each path is extracted sequentially [(x1, y1, z1), (x2, y2, z2), ..., (x...]. n y n , z n )], calculate the directional change angle θ(i) between adjacent points = arccos[(V i ·V i+1 ) / (||V i ||||V i+1 ||)], where V i =(x i+1 -x i y i+1 -y i , z i+1 -z i), θ(i) represents the angle between the current segment and the next segment, then calculate the rate of change of all angles in each path Δθ / Δs, where Δs is the distance between two points, representing the rate of change of the path curvature, and let the curvature change threshold C be... t =15° / mm, when the rate of change of the included angle in a continuous path is less than C t When a segment is identified as being in the same direction, the path segments that meet the conditions are merged and their voxel numbers are updated according to the three-dimensional coordinates. The renumbering method is to arrange them in ascending order of the three-axis coordinates from left to right, from top to bottom, and from front to back. For example, if the path segments are [(10, 10, 10), (11, 10, 10), (12, 10, 10)] and [(13, 10, 10), (14, 10, 10)], and their included angle is 0°, the length of the merged path segment becomes 5, and the voxel numbers are updated to V1 to V5. Finally, all path segments that meet the curvature stability condition are extracted to form a continuous spatial trajectory segment.

[0144] S513: Based on continuous spatial trajectory segments, extract the density values, spatial step size and temporal change information of nodes in each path segment, construct a density change time series and superimpose spatial location index, establish the temporal evolution trend of voxel density according to the path order, and obtain a three-dimensional dynamic map of breast tumor growth trend prediction.

[0145] Based on the continuous spatial trajectory segments, the density value ρ(i) and spatial coordinates (x, y) of the voxel nodes within each segment are read sequentially. i y i , z i ), and call the imaging time point t i Construct a time series T = [t1, t2, ..., t] n [, combining the path node order with its corresponding spatial step size] Calculate the density change value Δρ(i) = ρ(i+1) - ρ(i) for each node at the corresponding time, forming the density time series R = [ρ1, ρ2, ..., ρ n ], superimposed with its spatial index information S = [(x1, y1, z1), (x2, y2, z2), ...], and on this basis, construct a multivariate time evolution matrix E, where E(i) = [x i y i , z i , t i , ρ iThe E matrix reflects the trend of voxel density evolution over time and space in the tumor path. For example, on a certain path, t = [1, 2, 3], ρ = [30, 35, 39], and the coordinates are [(10, 10, 10), (11, 10, 10), (12, 10, 10)]. The E matrix is ​​constructed as E(1) = [10, 10, 10, 1, 30], E(2) = [11, 10, 10, 2, 35], and E(3) = [12, 10, 10, 3, 39]. This matrix describes the evolution process of density increasing over time in the spatial trajectory. Finally, the E matrices of all trajectory segments are integrated to construct a three-dimensional dynamic map for predicting the growth trend of breast tumors.

[0146] Please see Figure 7 A breast tumor growth trend prediction system based on big data includes:

[0147] The structural mutation detection module acquires the three-dimensional structure voxel sequence of the breast, calculates the density gradient values ​​of adjacent voxels, detects spatial mutation locations, marks spatial intersection regions, and obtains a set of density gradient mutation points.

[0148] The spatial orientation filtering module is based on the set of density gradient mutation points. It compares the spatial proximity and orientation angle between mutation points, clusters and extracts stable orientation segments, and obtains boundary expansion orientation segments.

[0149] The path trend extraction module analyzes the density change trend within the connected path based on the voxel density data of the tumor region, extracts the positive expansion segment, and obtains the set of stable path terminal directions.

[0150] The directional intersection analysis module calculates the vector angle based on the boundary expansion directional segment and the stable path terminal directional set, filters directional pairs with an angle less than a consistency threshold, statistically locates intersection hotspots, and obtains a set of tumor growth prediction vectors.

[0151] The 3D trajectory modeling module is based on a set of tumor growth prediction vectors. It determines the connection path and directional stability of vector segments, constructs a continuous spatial trajectory, and generates a 3D dynamic map of breast tumor growth trend prediction.

[0152] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for predicting the growth trend of breast tumors based on big data, characterized in that, Includes the following steps: S1: Obtain continuous structural voxel sequences in the three-dimensional structure of the breast, detect spatial mutation locations based on density gradient change points and mark multi-directional intersection regions to obtain a set of density gradient mutation points; S2: Based on the set of density gradient mutation points, cluster the spatial proximity relationship between adjacent units to identify continuous structures with consistent orientation, mark potential growth paths, and obtain boundary expansion direction segments. S3: Extract tumor density distribution from breast tumor regions, analyze density difference trends in potential expansion directions, screen directional segments with positive expansion trends, and obtain a set of stable path terminal directions; S4: Based on the boundary expansion direction segment and the stable path terminal direction set, perform direction consistency analysis, select direction vectors within a preset angle threshold, identify direction intersection hotspot areas, locate potential diffusion directions in the tumor region, and obtain a set of tumor growth prediction vectors. S5: Based on the tumor growth prediction vector set, locate the spatial trajectory path, perform trend continuity modeling within the complete region, and obtain a three-dimensional dynamic map of breast tumor growth trend prediction.

2. The method for predicting breast tumor growth trends based on big data according to claim 1, characterized in that, The density gradient mutation point set includes spatial variation extreme points, directional overlap intersection points, and density jump coordinate points. The boundary expansion direction segment includes stable vector segments, continuous boundary distribution, and potential migration paths. The stable path terminal direction set includes terminal vector directions, positive growth segments, and density trend directions. The tumor growth prediction vector set includes a consistent direction set, spatial hotspot regions, and diffusion trend vectors. The three-dimensional dynamic map of breast tumor growth trend prediction includes a path trajectory network, a trend flow map, and a continuous growth prediction distribution map.

3. The method for predicting breast tumor growth trends based on big data according to claim 1, characterized in that, The specific steps for obtaining the set of density gradient abrupt change points are as follows: S111: Obtain continuous structural voxel sequences in each spatial direction of the three-dimensional structure of the breast, calculate the density difference between each pair, identify the index position of the difference mutation point in the sequence, perform spatial coordinate position mapping, and generate a set of directional mutation coordinate points. S112: Based on the set of coordinate points for abrupt changes in direction, compare the positions of the coordinate points for abrupt changes in different directions in the three-dimensional coordinate system, filter out the abrupt changes in different directions, and determine whether they constitute an intersection relationship in the three-dimensional space, thereby generating a set of intersection coordinate points. S113: Based on the set of intersection coordinate points, reorder the coordinate values ​​of all intersection points, determine whether the spatial distance between adjacent points is continuous, filter the set of continuous region points, and obtain the set of density gradient abrupt change points.

4. The method for predicting breast tumor growth trends based on big data according to claim 1, characterized in that, The specific steps for obtaining the boundary extension direction segment are as follows: S211: Based on the set of density gradient abrupt change points, obtain the three-dimensional coordinate values ​​between each point, determine whether the point pairs constitute a spatial proximity relationship, perform connection and classification operations on the point pairs that satisfy the proximity relationship, and obtain a spatial aggregation point set; S212: Based on the spatial aggregation point set, obtain the density difference sequence between voxels in each aggregation unit, calculate the change trend of density value with spatial coordinates, filter the aggregation voxel sequence that meets the unidirectional condition, and generate directional continuous point segments. S213: Based on the directional continuity point segment, calculate the difference in the directional vector change angle between adjacent voxel segments, filter the continuous voxel region, determine the spatial range between the start and end segments in the continuous region that meets the conditions, and obtain the boundary extension direction segment.

5. The method for predicting breast tumor growth trends based on big data according to claim 1, characterized in that, The specific steps for obtaining the stable path terminal direction set are as follows: S311: Extract voxel density values ​​within the three-dimensional structure from the breast tumor region, obtain the spatial coordinates between adjacent voxels, construct a continuous voxel chain with spatial consistency and density similarity, and obtain the set of structural connectivity paths; S312: Based on the set of structural connected paths, detect the density value change sequence of continuous voxels in the spatial direction adjacent to the terminal position of each path, analyze the consistency of density change direction, and obtain the terminal trend path segment; S313: Based on the terminal trend path segment, and obtain the density difference between continuous voxels and the spatial step distance in the corresponding vector direction, calculate the density change rate on each directional vector, filter out directional segments with positive density change rate values, mark them as path extension trend directions, and obtain a stable path terminal direction set.

6. The method for predicting breast tumor growth trends based on big data according to claim 5, characterized in that, The density change rate is expressed by the formula: Calculations are performed, in which, This represents the rate of change of density in the directional segment. This represents the density value of the i-th voxel in the k-th terminal trend path segment. Represents the density value of the next adjacent voxel. This represents the spatial step distance between the i-th pair of adjacent voxels in the k-th path. This represents the average value of all voxel density values ​​in the k-th path. Let n represent the average step distance of all steps along the k-th path. k This represents the number of voxel pairs involved in the calculation in the k-th path segment.

7. The method for predicting breast tumor growth trends based on big data according to claim 1, characterized in that, The specific steps for obtaining the tumor growth prediction vector set are as follows: S411: Based on the boundary extension direction segment and the stable path terminal direction set, calculate the angle between vector pairs, filter the direction vector combinations with angle values ​​less than a set angle threshold, and obtain a set of direction consistency vectors. S412: Based on the set of direction consistency vectors, extract the spatial coordinate points corresponding to each vector combination and perform three-dimensional mesh division, and count the frequency of the direction vector in each mesh cell, extract the center position coordinates, and obtain the set of coordinate points in the direction concentration area. S413: Based on the set of coordinate points in the direction concentration area, extract the vector directions that appear most frequently in the direction vector group corresponding to each coordinate point, calculate the average difference of the angle values ​​of the vectors in the same region, and filter the vector directions whose average difference is lower than the vector consistency fluctuation threshold. Mark them as potential spread paths of the tumor region according to the region to obtain a set of tumor growth prediction vectors.

8. The method for predicting breast tumor growth trends based on big data according to claim 7, characterized in that, The average difference is expressed by the formula: Perform the calculation, where D r This represents the average angle consistency value. Let G represent the unit form of the g-th direction vector in the r-th region, where G represents the number of direction vectors in region r. q represents the result of the dot product between the g-th and j-th vectors. g This represents the number of times the g-th direction vector appears within the region. This represents the average number of occurrences of all directional vectors within the region.

9. The method for predicting breast tumor growth trends based on big data according to claim 1, characterized in that, The specific steps for obtaining the three-dimensional dynamic atlas for predicting the growth trend of breast tumors are as follows: S511: Based on the tumor growth prediction vector set and combined with the spatial coordinate set of three-dimensional structural voxels, locate continuously connected spatial voxel units in the three-dimensional voxel network according to the vector direction, integrate the continuously connected voxel paths into a linear trajectory structure, and obtain the spatial trajectory path set. S512: Based on the spatial trajectory path set, extract the three-dimensional coordinate distribution in the voxel sequence connected by each path, calculate the path curvature change rate, merge paths in the same direction and reorder the voxel node numbers to obtain continuous spatial trajectory segments. S513: Based on the continuous spatial trajectory segments, extract the density values, spatial step lengths and temporal change information of the nodes in each path segment, construct a density change time series and superimpose the spatial location index, establish the temporal evolution trend of voxel density in the path order, and obtain a three-dimensional dynamic map for predicting the growth trend of breast tumors.

10. A breast tumor growth trend prediction system based on big data, characterized in that, The system is used to implement the breast tumor growth trend prediction method based on big data as described in any one of claims 1-9, and the system comprises: The structural mutation detection module acquires the three-dimensional structure voxel sequence of the breast, calculates the density gradient values ​​of adjacent voxels, detects spatial mutation locations, marks spatial intersection regions, and obtains a set of density gradient mutation points. The spatial orientation filtering module, based on the set of density gradient mutation points, compares the spatial proximity and directional angle values ​​between mutation points, clusters and extracts stable orientation segments, and obtains boundary expansion orientation segments. The path trend extraction module analyzes the density change trend within the connected path based on the voxel density data of the tumor region, extracts the positive expansion segment, and obtains the set of stable path terminal directions. The directional intersection analysis module calculates the vector angle between the boundary extension direction segment and the stable path terminal direction set, filters direction pairs with an angle less than a consistency threshold, statistically locates intersection hotspots, and obtains a set of tumor growth prediction vectors based on the frequency of the boundary extension direction segment and the stable path terminal direction set. The three-dimensional trajectory modeling module determines the connection path and directional stability of vector segments based on the tumor growth prediction vector set, constructs a continuous spatial trajectory, and generates a three-dimensional dynamic map of breast tumor growth trend prediction.

Images