Tumor partition recognition method and system based on multi-scale image analysis

Abstract

The present invention relates to the field of medical image analysis technology, specifically to a tumor partition identification method and system based on multi-scale image analysis, comprising the following steps: based on an input medical tumor partition image, defining the gray value distribution range of a pixel block, dividing the pixel block group according to a fixed size, calculating the neighborhood gray value of the pixel block, and determining the distribution change set by accumulating the gray value difference of the pixel block group, and generating a multi-scale gray distribution map. In the present invention, by calculating the gray value distribution and the gray value difference of the pixel block group, the image detail recognition of the tumor area is enhanced, the tumor boundary resolution capability is improved, the texture feature difference of the boundary area is analyzed, the gray gradient difference and the connection characteristics are quantified, the geometric characteristic analysis accuracy is improved, and the spatial geometry expansion data and texture characteristics are combined to improve the partition accuracy and the accuracy of image analysis, effectively avoid the misidentification of the boundary fuzzy area, and provide a more refined tumor heterogeneity research basis.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image analysis, and in particular to a tumor partition recognition method and system based on multi-scale image analysis. Background Art

[0002] The field of medical image analysis technology involves the use of algorithms and visual processing techniques to analyze image data generated by various medical imaging devices (such as CT, MRI, and ultrasound). The core goal of this field is to assist medical research by improving the interpretability of image data. Medical image analysis is often used for image reconstruction, image enhancement, visual content extraction and classification, and automatic annotation of image content. This technology not only increases the efficiency of extracting information from medical images, but also improves the management and analysis process of image data, supporting medical research and educational applications.

[0003] Among them, the tumor zoning identification method refers to the use of image analysis technology to distinguish and mark different areas in tumor images. The areas have different biological characteristics and clinical significance. This method focuses on accurately distinguishing different parts of the tumor from medical images, such as the tumor core, proliferative area and necrotic area. This distinction is of great significance for studying the growth pattern and development stage of the tumor, especially in tumor-related biomedical research and education. It can provide an in-depth understanding of tumor heterogeneity. The main uses of this method include supporting medical research and exploring the biological behavior of tumors and their variability through precise image analysis.

[0004] Tumor zoning and identification methods in the prior art rely on a rough division of the tumor area and ignore the subtle biological differences within the area to a certain extent. Most of the existing methods are based on simple threshold segmentation and single-scale image processing, which limits the accuracy of tumor zoning, especially when dealing with fuzzy tumor boundaries or multiple heterogeneous regions, which easily leads to oversimplified segmentation results. Due to the lack of in-depth analysis of the complex relationship between grayscale differences, texture features and geometric characteristics, the existing technology cannot effectively identify and distinguish detailed information such as hyperplastic areas and necrotic areas in tumors. This deficiency directly affects the study of tumor behavior and clinical decision support. For example, traditional methods confuse the tumor core with the hyperplastic area, resulting in misjudgment of the tumor growth pattern. The existing technology is difficult to fully utilize the spatial information of the image, and the accurate calibration of boundaries and partitions is insufficient, which limits the in-depth analysis of tumor zoning and the breadth of tumor heterogeneity research. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings existing in the prior art and to propose a tumor partition identification method and system based on multi-scale image analysis.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a tumor partition identification method based on multi-scale image analysis, comprising the following steps:

[0007] S1: Based on the input medical tumor partition image, the gray value distribution range of the pixel block is defined, the pixel block group is divided into fixed sizes, the neighborhood gray value of the pixel block is calculated, and the distribution change set is determined by accumulating the gray value difference of the pixel block group, and a multi-scale gray distribution map is generated;

[0008] S2: Based on the multi-scale grayscale distribution map, extract the boundary area of ​​the pixel block, analyze the texture feature difference of the boundary area, select the area of ​​the connection relationship in the neighborhood, obtain the boundary distribution characteristic data, quantify the connection relationship and the grayscale gradient difference, analyze the boundary area and the connection characteristics, and generate a boundary connection weight table;

[0009] S3: Based on the boundary connection weight table, extract the geometric characteristics within the weight area, analyze the boundary perimeter, area and edge continuity, compare the geometric characteristic values, generate a candidate geometric feature matrix, take the candidate geometric feature matrix as the center, analyze the spatial expansion characteristics of the neighborhood, and obtain spatial geometric expansion data;

[0010] S4: Based on the spatial geometric expansion data, combined with the texture characteristics of the expansion area and the candidate area, the feature distribution of the boundary area is analyzed, the boundary line of each partition is calibrated, the pixel blocks within the tumor partition are extracted, and the shape and distribution of the pixel blocks are statistically characterized to obtain a tumor area partition identification feature value table.

[0011] As a further solution of the present invention, the step of obtaining the multi-scale grayscale distribution map is specifically as follows:

[0012] S111: Based on the input medical tumor partition image, define the grayscale value distribution range of the pixel blocks in the medical tumor partition image, divide the image into multiple pixel block groups according to a fixed size, each pixel block is composed of continuous pixel points, and obtain an initial pixel block group distribution array;

[0013] S112: Perform grayscale value analysis on each pixel block group in the initial pixel block group distribution array, calculate and accumulate grayscale differences between adjacent pixel block groups, using the formula:

[0014] ;

[0015] Perform normalization and measure the grayscale changes to obtain a grayscale difference accumulation array;

[0016] in, Representative The gray value of a pixel block, represents the average gray value of the neighborhood, and Represent the maximum and minimum grayscale values, respectively. represents the normalized sum of grayscale changes in a pixel block group, Indicates the total number of pixel blocks;

[0017] S113: Analyze the distribution changes between pixel block groups using the information recorded in the grayscale difference accumulation array, determine the distribution change set according to the statistical data of the grayscale difference, and obtain a multi-scale grayscale distribution map.

[0018] As a further solution of the present invention, the step of acquiring the boundary distribution characteristic data is specifically as follows:

[0019] S211: Analyze the multi-scale grayscale distribution map, extract the boundary area according to the grayscale gradient change rate and contrast change of the pixel block, compare the grayscale gradient difference of adjacent pixel blocks, filter the high gradient area by the gradient difference threshold, and obtain preliminary boundary area characteristic data;

[0020] S212: extracting neighborhood connection features from the preliminary boundary region characteristic data using the formula:

[0021] ;

[0022] Analyze the connection strength of each pixel pair and set a threshold based on the statistical distribution of the connection strength to generate a local connection area in the neighborhood;

[0023] in, and Represents the gray value of the pixel in the neighborhood. Represents the coordinates of the corresponding pixel point, represents the neighborhood pixel set, is the fine-tuning factor, Represents the connection weight between pixels;

[0024] S213: Utilizing the local connected area in the neighborhood, texture feature analysis is performed to compare texture feature differences in differentiated boundary areas, evaluate the integrity of the boundary areas, and obtain boundary distribution characteristic data.

[0025] As a further solution of the present invention, the step of obtaining the boundary connection weight table is specifically as follows:

[0026] S221: Based on the boundary distribution characteristic data, perform image pixel division, extract boundary areas and non-boundary areas in the image, perform pixel gray value classification, distinguish differential areas by threshold values, mark boundary areas, and obtain boundary area data;

[0027] S222: Based on the boundary area data, performing difference analysis on the grayscale values ​​of the pixel pairs in the boundary area, comparing the grayscale differences of adjacent pixels, and quantizing the differences to obtain quantized grayscale gradient difference data;

[0028] S223: Based on the quantized grayscale gradient difference data, a boundary area connection relationship analysis is performed, by traversing the connection points, quantizing the grayscale difference of each connection, and assigning a weight to each connection according to the grayscale gradient difference, to generate a boundary connection weight table.

[0029] As a further solution of the present invention, the step of obtaining the candidate geometric feature matrix is ​​specifically as follows:

[0030] S311: extracting each boundary point in the region based on the boundary connection weight table, traversing the connection status of each pixel, recording the connection weights between adjacent regions, determining the connection strength, and sorting the connection points to obtain the perimeter data of the tumor partition boundary;

[0031] S312: Based on the perimeter data of the tumor partition boundary, in each boundary area, measure the real-time length of the boundary, identify the boundary position, and obtain boundary area data through pixel overlay analysis;

[0032] S313: Based on the boundary area data, the continuity of the edge is analyzed in each area, the break points appearing in each area are detected, and the edge continuity analysis is performed by comparing the connection status of the boundary points, and the boundary perimeter data, boundary area data and edge continuity data are compared to generate a candidate geometric feature matrix.

[0033] As a further solution of the present invention, the step of acquiring the spatial geometric extension data is specifically as follows:

[0034] S321: Calculate the neighborhood average value of each geometric feature according to the candidate geometric feature matrix, filter the neighborhood range associated with the central feature, determine whether each neighborhood feature deviates from the central feature, and generate a candidate neighborhood geometric feature set;

[0035] S322: Based on the candidate neighborhood geometric feature set, analyze the spatial extension data of each neighborhood feature, analyze the neighborhood feature extension and the center feature, and use the formula:

[0036] ;

[0037] Calculate the spatial expansion characteristic coefficient and generate the spatial expansion coefficient distribution result;

[0038] in, represents the spatial expansion characteristic coefficient, is the neighborhood eigenvalue, is the central eigenvalue, is the weight parameter, is the number of neighborhood features;

[0039] S323: According to the spatial expansion coefficient distribution result, a threshold is set to extract the spatial expansion area, the geometric characteristics of the area are aggregated, and the global relationship with the candidate geometric features is analyzed to obtain the spatial geometric expansion data.

[0040] As a further solution of the present invention, the step of obtaining the tumor region partition identification feature value table is specifically as follows:

[0041] S411: Analyzing the texture characteristics of the extended area and the candidate area based on the spatial geometric expansion data, extracting the characteristic distribution of the boundary area, using the data for calibrating the tumor boundary line, and establishing a boundary line feature data set;

[0042] S412: extracting pixel blocks corresponding to the tumor boundary from the image data set using the boundary line feature data set, performing statistical analysis on the shape and distribution of the pixel blocks, wherein the statistical analysis results include the brightness mean, brightness variance and shape parameters of the regional pixel blocks, and obtaining shape and distribution statistical data;

[0043] S413: According to the shape and distribution statistics, capture the tumor shape and distribution data, analyze the geometric attributes and position of each data point, extract boundary clarity, compactness and direction characteristics, and use the formula:

[0044] ;

[0045] Generate a tumor region partition identification feature value table;

[0046] in, Represents the identification feature value of the tumor area, is the average brightness of the pixel block in the region, is the brightness variance, is the shape parameter, and It is the adjustment factor determined based on data sensitivity analysis.

[0047] A tumor partition recognition system based on multi-scale image analysis, wherein the tumor partition recognition system based on multi-scale image analysis is used to execute the above-mentioned tumor partition recognition method based on multi-scale image analysis, and the system comprises:

[0048] The grayscale distribution analysis module divides the pixel block groups into preset sizes based on the input medical tumor partition image, extracts the grayscale value in each pixel block, counts and classifies the pixel blocks with differentiated grayscale ranges, performs grayscale distribution statistical analysis, and generates a multi-scale grayscale distribution map;

[0049] The boundary texture analysis module extracts the boundary area in the tumor partition image based on the multi-scale grayscale distribution map, identifies the texture features in the boundary area, compares the grayscale difference of adjacent pixels using grayscale gradient calculation, and quantifies and classifies them according to the gradient value to generate a boundary connection weight table;

[0050] The geometric feature acquisition module extracts geometric features within the boundary area based on the boundary connection weight table, including perimeter, area, and edge connectivity, uses perimeter-to-area ratio and edge curvature indicators to perform morphological analysis on the area, compares the geometric features of the area, and combines the spatial expansion information to obtain spatial geometric expansion data;

[0051] The spatial geometry expansion feature analysis module analyzes the expansion ratio and distribution characteristics of each region based on the spatial geometry expansion data, identifies the spatial expansion characteristics of the differentiated regions, performs spatial expansion feature analysis, and obtains regional expansion ratio data;

[0052] The spatial zoning delineation module analyzes the morphological changes and spatial distribution characteristics within the region based on the regional expansion ratio data, identifies the spatial distribution patterns of differentiated regions in combination with the morphological characteristics of the boundaries, performs regional refinement division based on the identification results, and generates a numerical table of tumor regional zoning identification characteristics.

[0053] Compared with the prior art, the advantages and positive effects of the present invention are:

[0054] In the present invention, by performing detailed grayscale value distribution calculation and accumulation of grayscale differences of pixel block groups on the medical tumor partition image, not only the image detail recognition inside the tumor area is enhanced, but also the boundary resolution capability of different tumor areas is improved. Further analysis of the texture feature differences in the boundary area, combined with the neighborhood connection relationship, quantification of the grayscale gradient difference and connection characteristics, makes the geometric characteristic analysis of the boundary area more accurate, especially the perimeter, area and edge continuity of the boundary, providing rich data support for the subsequent construction of the geometric feature matrix, by combining the spatial geometry expansion data and texture characteristics, accurately calibrating the boundary line of the tumor partition, and performing statistical analysis on the pixel blocks in the partition, enhancing the partition accuracy of the tumor area, improving the meticulousness and accuracy of image analysis, effectively avoiding the problem of misidentification and missed judgment of the fuzzy boundary area in the traditional method, and providing a more refined basis for tumor heterogeneity research. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0056] Figure 2 It is a flow chart of the multi-scale grayscale distribution diagram in the present invention;

[0057] Figure 3It is a flow chart of the boundary distribution characteristic data in the present invention;

[0058] Figure 4 It is a flow chart of the boundary connection weight table in the present invention;

[0059] Figure 5 It is a flow chart of the candidate geometric feature matrix in the present invention;

[0060] Figure 6 It is a flow chart of spatial geometry expansion data in the present invention;

[0061] Figure 7 This is a flow chart of the tumor region partition identification feature value table in the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0063] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0064] Embodiment 1:

[0065] See also Figure 1 The present invention provides a technical solution: a tumor partition identification method based on multi-scale image analysis, comprising the following steps:

[0066] S1: Based on the input medical tumor partition image, the gray value distribution range of the pixel block is defined, the pixel block group is divided into fixed sizes, the neighborhood gray value of the pixel block is calculated, and the distribution change set is determined by accumulating the gray value difference of the pixel block group, and a multi-scale gray distribution map is generated;

[0067] S2: Based on the multi-scale grayscale distribution map, the boundary area of ​​the pixel block is extracted, the texture feature difference of the boundary area is analyzed, the area of ​​the connection relationship in the neighborhood is selected, the boundary distribution characteristic data is obtained, the connection relationship and the grayscale gradient difference are quantified, the boundary area and the connection characteristics are analyzed, and the boundary connection weight table is generated;

[0068] S3: Based on the boundary connection weight table, the geometric characteristics within the weight area are extracted, the boundary perimeter, area and edge continuity are analyzed, the geometric characteristic values ​​are compared, and a candidate geometric feature matrix is ​​generated. The candidate geometric feature matrix is ​​used as the center to analyze the spatial expansion characteristics of the neighborhood and obtain spatial geometric expansion data;

[0069] S4: Based on the spatial geometric expansion data, combined with the texture characteristics of the expanded area and the candidate area, the feature distribution of the boundary area is analyzed, the boundary line of each partition is calibrated, the pixel blocks within the tumor partition are extracted, the shape and distribution of the pixel blocks are statistically analyzed, and the feature value table of tumor area partition identification is obtained.

[0070] The multi-scale grayscale distribution map includes the grayscale difference of pixel blocks, the accumulation of grayscale difference of pixel block groups, and the distribution change set. The boundary distribution characteristic data includes the boundary distribution density, boundary distribution form, and boundary distribution law. The boundary connection weight table includes the texture feature difference of the boundary area, the grayscale gradient difference of the connection area, and the boundary connection relationship. The candidate geometric feature matrix includes the boundary perimeter, boundary area, and edge continuity. The spatial geometry expansion data includes the texture characteristics of the expansion area, the geometric form of the candidate area, and the boundary feature distribution. The tumor area partition recognition feature value table includes the pixel block shape, pixel block distribution, and the boundary line characteristics within the partition.

[0071] See also Figure 2 , the specific steps for obtaining the multi-scale grayscale distribution map are:

[0072] S111: Based on the input medical tumor partition image, define the grayscale value distribution range of the pixel blocks in the medical tumor partition image, divide the image into multiple pixel block groups according to a fixed size, each pixel block is composed of continuous pixel points, and obtain an initial pixel block group distribution array;

[0073] In the analysis of medical tumor zoning images, the grayscale value distribution range of each pixel block is first defined. The key to the process is to extract effective information from the original image. The grayscale value distribution range is defined based on the statistical characteristics of the image, such as the maximum and minimum values ​​of the pixel intensity. Based on the statistical characteristics, the image is divided according to the preset pixel block size. Each pixel block includes a fixed number of pixels, and the pixels together constitute a pixel block group. By performing statistical analysis on the pixel block group, the initial pixel block group distribution array can be obtained. This array is the basis for subsequent analysis. By analyzing each element in the array one by one, the processing process can be accurately controlled to ensure that each step of the operation is based on the actual image data, providing a solid foundation for subsequent image processing and analysis.

[0074] S112: Perform grayscale value analysis on each pixel block group in the initial pixel block group distribution array, calculate and accumulate grayscale differences between adjacent pixel block groups, using the formula:

[0075] ;

[0076] Perform normalization and measure the grayscale changes to obtain a grayscale difference accumulation array;

[0077] in, Representative The gray value of a pixel block, represents the average gray value of the neighborhood, and Represent the maximum and minimum grayscale values, respectively. represents the normalized sum of grayscale changes in a pixel block group, Indicates the total number of pixel blocks;

[0078] The benefit of the formula is that by normalizing the difference between the grayscale value of each pixel block and the average grayscale value of the neighborhood, the grayscale changes between different pixel blocks can be effectively measured, which is especially important for analyzing the texture and structural information of the image. The normalization process ensures that no matter how bright the original image is, the calculation result can reflect the actual grayscale changes.

[0079] Assume that there is a pixel block group containing 10 pixel blocks, the grayscale values ​​of each pixel block are {200, 180, 220, 210, 230, 240, 260, 250, 240, 230}, the average grayscale value of the neighborhood is 225, and the maximum and minimum grayscale values ​​are 260 and 180 respectively. According to the formula, the absolute value of the difference between the grayscale value of each pixel block and the average grayscale value is first calculated, and then divided by the grayscale range (260-180=80), and the result is:

[0080] ;

[0081] The result shows that in this pixel block group, the normalized sum of grayscale changes is 4.375, which indicates that compared with the average grayscale, the grayscale change in this area is larger and is a characteristic area in the image, such as a tumor area or an important medical image feature.

[0082] S113: using the information recorded in the grayscale difference accumulation array, analyzing the distribution changes between pixel block groups, determining the distribution change set according to the statistical data of the grayscale differences, and obtaining a multi-scale grayscale distribution map;

[0083] The grayscale change data recorded by the grayscale difference accumulation array is used to analyze the grayscale differences between each pixel block group. This array serves as the basis for analysis, and contains the grayscale value comparison results of each pixel block and its neighborhood. The data can be used to effectively identify heterogeneous areas in the image. These areas are related to pathological changes, such as tumor growth areas. By counting grayscale differences, multi-scale grayscale distribution analysis of the image can be performed to generate a multi-scale grayscale distribution map. This map can show the grayscale distribution of the image at different scales and provide doctors with more information.

[0084] See also Figure 3 , the specific steps for obtaining boundary distribution characteristic data are:

[0085] S211: Analyze the multi-scale grayscale distribution map, extract the boundary area according to the grayscale gradient change rate and contrast change of the pixel block, compare the grayscale gradient difference of adjacent pixel blocks, screen the high gradient area through the gradient difference threshold, and obtain preliminary boundary area characteristic data;

[0086] When analyzing the multi-scale grayscale distribution map, it is first necessary to preprocess the image data, including noise reduction and contrast enhancement, so as to more clearly identify the boundaries between different pixel blocks. By calculating the grayscale gradient change rate of each pixel block, the edge area in the image can be accurately determined. The area shows a high contrast change. The extraction of the boundary area is carried out by comparing the grayscale gradient difference between adjacent pixel blocks. A grayscale gradient difference threshold is set. When the grayscale gradient difference between adjacent pixel blocks is greater than this threshold, it is identified as a boundary pixel. This not only improves the accuracy of boundary detection, but also reduces false positive boundary identification and obtains preliminary boundary area characteristic data.

[0087] S212: extracting neighborhood connection features from preliminary boundary region feature data using the formula:

[0088] ;

[0089] Analyze the connection strength of each pixel pair and set a threshold based on the statistical distribution of the connection strength to generate a local connection area in the neighborhood;

[0090] in, and Represents the gray value of the pixel in the neighborhood. Represents the coordinates of the corresponding pixel point, represents the neighborhood pixel set, is the fine-tuning factor, Represents the connection weight between pixels;

[0091] The benefit of the formula is that it allows weighting based on the actual physical distance between pixels and the gray value difference, thereby more accurately reflecting the neighborhood connection strength in the image. This is particularly important in boundary detection and texture analysis because it helps distinguish pixels that are only adjacent in grayscale but not spatially related.

[0092] Suppose there is a set of neighboring pixels, where the grayscale values ​​of pixel p and pixel q are and , the coordinates are and , then the difference in the formula , the sum of the squares of the distances is Pick To prevent division by zero errors, substitute the formula into the calculation to get , which indicates that there is a strong connection between pixel p and pixel q;

[0093] This result shows that the connection weights calculated by The value is close to 20, which is a high connection strength, indicating that the pair of pixels has a strong visual connectivity, which helps to more accurately select those pixels that are significantly correlated in both space and grayscale when generating local connected regions in the neighborhood, thereby improving the accuracy of boundary detection in image processing.

[0094] S213: using the local connected area in the neighborhood to perform texture feature analysis, compare the texture feature differences of the differentiated boundary areas, evaluate the integrity of the boundary areas, and obtain boundary distribution characteristic data;

[0095] Using the data obtained from the local connected areas in the neighborhood, detailed texture feature analysis is required, including calculating the texture directionality, uniformity and contrast of each area. By comparing the texture features of different boundary areas, the integrity and accuracy of each boundary area can be evaluated. The process involves a large amount of pixel data analysis, including calculating the grayscale mean, variance and high-order statistical characteristics of the pixels in each area. The calculation results will provide a scientific basis for determining the effective boundary areas in the image, ensuring that the boundary information in the image is accurately captured through the texture difference analysis at the detail level, and obtaining the boundary distribution characteristic data, which will directly affect the success rate of image processing tasks, such as boundary detection, image segmentation, etc.

[0096] See also Figure 4 , the specific steps for obtaining the boundary connection weight table are:

[0097] S221: Based on the boundary distribution characteristic data, perform image pixel division, extract boundary areas and non-boundary areas in the image, perform pixel gray value classification, distinguish differential areas by threshold values, mark boundary areas, and obtain boundary area data;

[0098] A fixed-size partitioning strategy is adopted to divide the entire image into several rectangular areas at equal intervals. The number of pixels in each area remains consistent. The distribution of pixel grayscale values ​​in each partition is calculated one by one, and the grayscale values ​​are sorted according to the standard deviation of the grayscale value change to determine the initial range of the boundary area. When extracting the boundary area and non-boundary area in the image, the pixel points in each partition are traversed to calculate the grayscale value difference between each pixel point and the surrounding pixels. The pixels whose difference value exceeds the preset threshold are marked as boundary pixels. When performing pixel grayscale value classification, the image grayscale value range is divided into multiple intervals, and the number of pixels in each interval is calculated respectively. The distribution ratio is statistically calculated, and the differentiated areas are obtained and distinguished by threshold calculation. In the boundary area marking, the binary image is used to generate a marking matrix, and the boundary pixels are marked as 1 and the remaining pixels are marked as 0. The marking matrix is ​​used for further processing to obtain the boundary area data.

[0099] S222: Based on the boundary area data, performing difference analysis on the grayscale values ​​of the pixels in the boundary area, comparing the grayscale differences of adjacent pixels, and quantizing the differences to obtain quantized grayscale gradient difference data;

[0100] Select the pixels in each boundary area and their adjacent pixels in turn, calculate the grayscale value difference between the two, and generate a data matrix containing the difference. When comparing the grayscale difference of adjacent pixels, set a step range and pair the pixels in the horizontal, vertical and diagonal directions in turn. Each time a set of grayscale differences is generated, and the distribution characteristics of all paired differences are counted. The statistical results of the grayscale differences are compared and analyzed with the mean and standard deviation within the area. When quantifying the differences, the levels are divided according to the actual range of the grayscale differences, and quantized grayscale gradient difference data is generated according to the level distribution. A specific weight is assigned to each gradient difference through a mapping relationship for subsequent analysis.

[0101] S223: Based on the quantized grayscale gradient difference data, a boundary area connection relationship analysis is performed, by traversing the connection points, quantizing the grayscale difference of each connection, and assigning a weight to each connection according to the grayscale gradient difference, to generate a boundary connection weight table;

[0102] Generate a connection matrix for the pixel points in the boundary area according to the geometric adjacency relationship, traverse the connection points one by one, count the pixel grayscale values ​​involved in each connection point, and calculate the grayscale difference of the pixels at both ends of the connection point. When quantizing the grayscale difference of each connection point, further generate hierarchical weight values ​​according to the quantization rule of grayscale gradient difference. When traversing all connection points, recursively calculate the grayscale difference of adjacent connection points layer by layer according to the geometric position of the connection points, assign weights to each connection, generate a boundary connection weight table, and use the assigned weight values ​​to construct a comprehensive table including connection point coordinates, connection weights and grayscale gradient differences to provide data support for subsequent image boundary optimization and analysis.

[0103] See also Figure 5 , the specific steps for obtaining the candidate geometric feature matrix are:

[0104] S311: extracting each boundary point in the region based on the boundary connection weight table, traversing the connection status of each pixel, recording the connection weights between adjacent regions, determining the connection strength, and sorting the connection points to obtain the perimeter data of the tumor partition boundary;

[0105] When extracting each boundary point in the tumor partition, the boundary connection weight table is scanned point by point, and the specific pixel position of each boundary point is located in combination with the partition boundary characteristics in the tumor image, and the connection weight between it and the adjacent partition is recorded. When traversing the connection status of each pixel, the connection relationship of the boundary points in the tumor partition is represented by the two-dimensional adjacency matrix, the difference in connection weights between tumor partitions is counted, and high-weight boundary connection points are extracted. The difference areas between partitions are analyzed in detail. When recording the connection weights between adjacent areas, the pixel connections across tumor partitions are calculated, focusing on the areas with significant changes in grayscale values. The key point data reflecting the partition boundary characteristics are extracted. When judging the connection strength, the strong and weak connection points are marked according to the weight value. After sorting the connection points, the tumor boundary shape characteristics are further analyzed. Finally, the perimeter data of the tumor partition boundary is obtained, laying the foundation for the analysis of tumor morphological characteristics.

[0106] S312: Based on the perimeter data of the tumor partition boundary, in each boundary area, measure the real-time length of the boundary, identify the boundary position, and obtain boundary area data through pixel overlay analysis;

[0107] When measuring the real-time length of the boundary in each tumor partition, the geometric path of the partition boundary is calculated pixel by pixel, combined with the pixel scale of the tumor image, to accurately measure the total length of the partition boundary. When identifying the boundary position, the boundary points of the tumor partition are mapped to the two-dimensional image coordinate system to generate the trajectory line of the tumor boundary points. The change characteristics of the boundary position are determined by coordinate difference calculation. When obtaining the boundary area data through pixel overlay analysis, the number of all pixels inside and on the boundary of each tumor partition is counted, and the partition area is calculated according to the pixel size. The accuracy of the area data is verified by combining the real scale in the image. During the analysis process, the grayscale difference between the internal tissue of the tumor and the surrounding area is further distinguished to ensure that the boundary area data can accurately reflect the morphological characteristics of the tumor partition and provide support for subsequent lesion analysis.

[0108] S313: Based on the boundary area data, edge continuity is analyzed in each region, breakpoints appearing in each region are detected, and edge continuity analysis is performed by comparing the connection status of boundary points, and boundary perimeter data, boundary area data, and edge continuity data are compared to generate a candidate geometric feature matrix;

[0109] The connection of the partition boundary points is checked according to the pixels, the positions of the breakpoints in the tumor partition boundary are marked, and the number and distribution characteristics of the breakpoints are counted. When detecting the breakpoints in each tumor partition, the distance between adjacent boundary points is calculated to determine whether the gap between the breakpoints exceeds a specific threshold, and the coordinates of the break area exceeding the threshold are recorded. When the tumor edge continuity is analyzed by comparing the boundary point connection status, the breakpoint information of each boundary is comprehensively analyzed with the tumor partition connection status to quantify the integrity of the boundary. When comparing the boundary perimeter data, boundary area data and edge continuity data, a candidate geometric feature matrix is ​​generated, which includes the partition boundary point coordinates, connection weights, breakpoint positions and grayscale difference information, to provide data support for further analysis of tumor lesion morphology and partition characteristics.

[0110] See also Figure 6 , the specific steps for obtaining spatial geometry extension data are:

[0111] S321: Calculate the neighborhood average value of each geometric feature according to the candidate geometric feature matrix, filter the neighborhood range associated with the central feature, determine whether each neighborhood feature deviates from the central feature, and generate a candidate neighborhood geometric feature set;

[0112] When analyzing the candidate geometric feature matrix, the focus is on understanding the feature changes of the central feature and its neighborhood, determining the position of the central feature, which is identified and calibrated by the matrix coordinate system, and then defining the neighborhood range, which is achieved through a preset spatial radius. The radius determines which neighborhood features will be included, and each neighborhood feature is collected one by one and compared with the central feature. The relationship between the two is evaluated by calculating the absolute difference between each neighborhood feature and the central feature, including adding all neighborhood feature values ​​and then dividing by the number of features to obtain the average neighborhood feature value. This average is compared with the central feature value to determine which neighborhood features are significantly different from the center. The significance is defined by a set threshold. If the difference exceeds this threshold, it is considered that the neighborhood feature deviates from the central feature and should be marked and eliminated in subsequent analysis. The remaining neighborhood features will be saved as a candidate neighborhood geometric feature set for further analysis.

[0113] S322: Based on the candidate neighborhood geometric feature set, analyze the spatial extension data of each neighborhood feature, analyze the neighborhood feature extension and the center feature, and use the formula:

[0114] ;

[0115] Calculate the spatial expansion characteristic coefficient and generate the spatial expansion coefficient distribution result;

[0116] in, represents the spatial expansion characteristic coefficient, is the neighborhood eigenvalue, is the central eigenvalue, is the weight parameter, is the number of neighborhood features;

[0117] The formula is useful in that the weight parameter The influence of different neighborhood features is weighted to avoid error deviation caused by uneven distribution of eigenvalues, while optimizing the accuracy of spatial extension feature calculation;

[0118] Obtaining the center feature through the candidate neighborhood geometric feature set and neighborhood eigenvalues ,in , ;

[0119] Determine the weight parameters , the weighted product of the weight and the absolute difference is calculated as ;

[0120] The summation gives the numerator: , the denominator is the weight and ;

[0121] Calculate the expansion factor ,The result shows that the extended correlation degree between the central feature and the neighborhood feature is 0.4615, which can be further used for significant region analysis.

[0122] S323: extracting the spatial extension region by setting a threshold according to the spatial extension coefficient distribution result, aggregating the geometric characteristics of the region, analyzing the global relationship with the candidate geometric features, and obtaining the spatial geometric extension data;

[0123] A significance threshold is set to identify which areas have significant spatial expansion and should be paid attention to. By comparing with the preset threshold, feature areas with expansion coefficients greater than or equal to the threshold are selected. This involves iterating the data set. Each iteration evaluates the expansion coefficient of each neighborhood feature and records the feature areas that exceed the threshold. The areas are considered to be of special significance for understanding spatial geometric expansion. Significant feature areas will be grouped and aggregated according to their positions in space. The process involves not only simple numerical comparisons, but also requires analysis of the spatial relationship of features, such as the relative position with the central feature. The aggregated data of the significant areas are combined with the global geometric features to calculate the overall geometric expansion characteristics. Comprehensive analysis helps to understand how spatial features unfold in a larger geometric context and obtain spatial geometric expansion data.

[0124] See also Figure 7 , the specific steps for obtaining the tumor region partition identification feature value table are:

[0125] S411: Based on the spatial geometric expansion data, the texture characteristics of the expansion area and the candidate area are analyzed, the characteristic distribution of the boundary area is extracted, the data is used for the calibration of the tumor boundary line, and the boundary line feature data set is established;

[0126] First, the extended area and candidate area are extracted from the high-resolution image, and the texture characteristics of the area are analyzed. The process requires calling the image segmentation tool to mark the target area, and using the feature extraction module to obtain the texture parameters inside the area, grayscale mean, grayscale variance, texture contrast and other indicators. The texture parameters are then partitioned, and the texture characteristics of each partition are summarized as an independent vector representation. The geometric extension characteristics and texture parameters of the region are combined, and the boundary area is identified by comparing its similarity with the standard model. The feature distribution in the boundary area is refined, and all partitions are marked as specific boundary line data to support the precise positioning of subsequent analysis and generate a boundary line feature data set.

[0127] S412: extracting pixel blocks corresponding to the tumor boundary from the image data set using the boundary line feature data set, performing statistical analysis on the shape and distribution of the pixel blocks, wherein the statistical analysis results include the brightness mean, brightness variance and shape parameters of the regional pixel blocks, and obtaining shape and distribution statistical data;

[0128] The corresponding pixel blocks are further extracted from the original image, and the boundary line data is called to perform spatial cropping on the entire image. The pixels in each cropped area are aggregated into the target object, and then statistical analysis is performed on the target pixel blocks. The brightness mean, brightness variance and shape parameters of the pixel blocks are calculated respectively. For example, the shape factor is fitted through the geometric boundary of the pixel block, and the aspect ratio and perimeter area ratio are calculated to quantify its shape characteristics. An association matrix is ​​established for all pixel block data, and the brightness and shape data are combined into a unified feature set. The features not only include basic brightness and variance, but also combine the shape parameters of the pixel blocks to form composite features, obtain shape and distribution statistical data, and provide a data basis for subsequent feature extraction.

[0129] S413: Based on shape and distribution statistics, capture tumor shape and distribution data, analyze the geometric properties and position of each data point, extract boundary clarity, compactness and direction features, and use the formula:

[0130] ;

[0131] Generate a tumor region partition identification feature value table;

[0132] in, Represents the identification feature value of the tumor area, is the average brightness of the pixel block in the region, is the brightness variance, is the shape parameter, and It is the adjustment factor determined based on data sensitivity analysis;

[0133] The benefit of the formula is that it integrates three types of features: brightness mean, brightness variance, and shape parameter ratio, and adjusts the coefficients and The flexibility of the algorithm can adapt to the feature distribution requirements of different types of images and improve the comprehensiveness of feature representation;

[0134] Through statistical analysis, we set , obtained from the extracted pixel block brightness mean data, , is the brightness variance, calculated by statistics of the entire image, , represents the shape parameter, measures the complexity of the pixel block, and adjusts the coefficient and , determined by regional characteristics analysis;

[0135] Calculation process:

[0136] ;

[0137] The results show that the calculated recognition feature value of the tumor area is 124. The comprehensive brightness and shape characteristics are the quantitative basis for tumor identification in subsequent zoning diagnosis, and a tumor area zoning recognition feature value table is obtained.

[0138] The tumor partition recognition system based on multi-scale image analysis is used to execute the above-mentioned tumor partition recognition method based on multi-scale image analysis, and the system includes:

[0139] The grayscale distribution analysis module divides the pixel block groups into preset sizes based on the input medical tumor partition image, extracts the grayscale value in each pixel block, counts and classifies the pixel blocks with differentiated grayscale ranges, performs grayscale distribution statistical analysis, and generates a multi-scale grayscale distribution map;

[0140] The boundary texture analysis module extracts the boundary area in the tumor partition image based on the multi-scale grayscale distribution map, identifies the texture features in the boundary area, compares the grayscale difference of adjacent pixels using grayscale gradient calculation, and quantifies and classifies them according to the gradient value to generate a boundary connection weight table;

[0141] The geometric feature acquisition module extracts the geometric features of the boundary area based on the boundary connection weight table, including perimeter, area, and edge connectivity. It uses the perimeter-area ratio and edge curvature indicators to perform morphological analysis on the area, compare the geometric features of the area, and combine the spatial expansion information to obtain spatial geometric expansion data.

[0142] The spatial geometry expansion feature analysis module analyzes the expansion ratio and distribution characteristics of each region based on the spatial geometry expansion data, identifies the spatial expansion characteristics of differentiated regions, performs spatial expansion feature analysis, and obtains regional expansion ratio data;

[0143] The spatial zoning delineation module analyzes the morphological changes and spatial distribution characteristics within the region based on the regional expansion ratio data, combines the morphological characteristics of the boundaries, identifies the spatial distribution patterns of differentiated regions, and divides the regions into finer areas based on the identification results to generate a numerical table of tumor region zoning identification characteristics.

[0144] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A tumor partition identification method based on multi-scale image analysis, characterized in that: The following steps are involved: Based on the input medical tumor partition image, the gray value distribution range of the pixel block is defined, the pixel block group is divided into fixed sizes, the neighborhood gray value of the pixel block is calculated, and the distribution change set is determined by accumulating the gray difference of the pixel block group, and a multi-scale gray distribution map is generated; Based on the multi-scale grayscale distribution map, the boundary area of ​​the pixel block is extracted, the texture feature difference of the boundary area is analyzed, the area of ​​the connection relationship in the neighborhood is selected, the boundary distribution characteristic data is obtained, the connection relationship and the grayscale gradient difference are quantified, the boundary area and the connection characteristics are analyzed, and a boundary connection weight table is generated; The boundary distribution characteristic data includes boundary distribution density, boundary distribution form, and boundary distribution law; Based on the boundary connection weight table, the geometric characteristics within the weight area are extracted, the boundary perimeter, area and edge continuity are analyzed, the geometric characteristic values ​​are compared, and a candidate geometric feature matrix is ​​generated. The candidate geometric feature matrix is ​​used as the center to analyze the spatial expansion characteristics of the neighborhood and obtain spatial geometric expansion data; The spatial geometric expansion data includes the texture characteristics of the expansion area, the geometric shape of the candidate area, and the boundary feature distribution; Based on the spatial geometric expansion data, combined with the texture characteristics of the expanded area and the candidate area, the feature distribution of the boundary area is analyzed, the boundary line of each partition is calibrated, the pixel blocks within the tumor partition are extracted, the shape and distribution of the pixel blocks are statistically characterized, and a tumor area partition identification feature value table is obtained.

2. The tumor partition identification method based on multi-scale image analysis according to claim 1, characterized in that: The steps of obtaining the multi-scale grayscale distribution map are specifically as follows: Based on the input medical tumor partition image, the grayscale value distribution range of the pixel blocks in the medical tumor partition image is defined, the image is divided into a plurality of pixel block groups according to a fixed size, each pixel block is composed of continuous pixel points, and an initial pixel block group distribution array is obtained; Perform grayscale value analysis on each pixel block group in the initial pixel block group distribution array, calculate and accumulate grayscale differences between adjacent pixel block groups, using the formula: ； Perform normalization and measure the grayscale changes to obtain a grayscale difference accumulation array; in, Representative The gray value of a pixel block, represents the average gray value of the neighborhood, and Represent the maximum and minimum grayscale values, respectively. represents the normalized sum of grayscale changes in a pixel block group, Indicates the total number of pixel blocks; The information recorded in the grayscale difference accumulation array is used to analyze the distribution changes between pixel block groups, and the distribution change set is determined according to the statistical data of the grayscale differences to obtain a multi-scale grayscale distribution map.

3. The method for tumor partition identification based on multi-scale image analysis according to claim 2, characterized in that: The step of obtaining the boundary distribution characteristic data is specifically as follows: Analyze the multi-scale grayscale distribution map, extract the boundary area according to the grayscale gradient change rate and contrast change of the pixel block, compare the grayscale gradient difference of adjacent pixel blocks, screen the high gradient area through the gradient difference threshold, and obtain preliminary boundary area characteristic data; The neighborhood connection features are extracted from the preliminary boundary area characteristic data using the formula: ； Analyze the connection strength of each pixel pair and set a threshold based on the statistical distribution of the connection strength to generate a local connection area in the neighborhood; in, and Represents the gray value of the pixel in the neighborhood. Represents the coordinates of the corresponding pixel point, represents the neighborhood pixel set, is the fine-tuning factor, Represents the connection weight between pixels; The texture feature analysis is performed using the local connected area in the neighborhood, the texture feature differences of the differentiated boundary areas are compared, the integrity of the boundary areas is evaluated, and the boundary distribution characteristic data is obtained.

4. The method for tumor partition identification based on multi-scale image analysis according to claim 3, characterized in that: The steps for obtaining the boundary connection weight table are specifically as follows: Based on the boundary distribution characteristic data, image pixel division is performed to extract boundary areas and non-boundary areas in the image, pixel grayscale value classification is performed, differential areas are distinguished by threshold values, and boundary area marking is performed to obtain boundary area data; Based on the boundary area data, performing difference analysis on the grayscale values ​​of the pixel pairs in the boundary area, comparing the grayscale differences of adjacent pixels, and quantizing the differences to obtain quantized grayscale gradient difference data; Based on the quantized grayscale gradient difference data, the boundary area connection relationship analysis is performed, by traversing the connection points, quantizing the grayscale difference of each connection, and assigning a weight to each connection according to the grayscale gradient difference to generate a boundary connection weight table.

5. The method for tumor partition identification based on multi-scale image analysis according to claim 4, characterized in that: The steps of obtaining the candidate geometric feature matrix are specifically as follows: Based on the boundary connection weight table, each boundary point in the region is extracted, the connection status of each pixel is traversed, the connection weights between adjacent regions are recorded, the connection strength is determined, and the connection points are sorted to obtain the perimeter data of the tumor partition boundary; Based on the perimeter data of the tumor partition boundary, in each boundary area, the real-time length of the boundary is measured, the boundary position is identified, and the boundary area data is obtained through pixel overlay analysis; Based on the boundary area data, the continuity of the edge is analyzed in each area, the break points appearing in each area are detected, and the edge continuity analysis is performed by comparing the connection status of the boundary points. The boundary perimeter data, boundary area data and edge continuity data are compared to generate a candidate geometric feature matrix.

6. The method for tumor partition identification based on multi-scale image analysis according to claim 5, characterized in that: The steps for obtaining the spatial geometry extension data are specifically as follows: According to the candidate geometric feature matrix, the neighborhood average value of each geometric feature is calculated, the neighborhood range associated with the central feature is screened, and it is determined whether each neighborhood feature deviates from the central feature to generate a candidate neighborhood geometric feature set; Based on the candidate neighborhood geometric feature set, the spatial extension data of each neighborhood feature is analyzed, and the neighborhood feature extension and the center feature are analyzed using the formula: ； Calculate the spatial expansion characteristic coefficient and generate the spatial expansion coefficient distribution result; in, represents the spatial expansion characteristic coefficient, is the neighborhood eigenvalue, is the central eigenvalue, is the weight parameter, is the number of neighborhood features; According to the spatial expansion coefficient distribution result, the spatial expansion area is extracted by setting a threshold, the geometric characteristics of the area are aggregated, and the global relationship with the candidate geometric features is analyzed to obtain the spatial geometric expansion data.

7. The method for tumor partition identification based on multi-scale image analysis according to claim 6, characterized in that: The steps for obtaining the tumor region partition identification feature value table are specifically as follows: Based on the spatial geometric expansion data, analyzing the texture characteristics of the expansion area and the candidate area, extracting the characteristic distribution of the boundary area, using the data for calibrating the tumor boundary line, and establishing a boundary line feature data set; Utilizing the boundary line feature data set, extracting pixel blocks corresponding to the tumor boundary from the image data set, performing statistical analysis on the shape and distribution of the pixel blocks, wherein the statistical analysis results include the brightness mean, brightness variance and shape parameters of the regional pixel blocks, and obtaining shape and distribution statistical data; Based on the shape and distribution statistics, the tumor shape and distribution data are captured, the geometric properties and position of each data point are analyzed, and the boundary clarity, compactness and direction characteristics are extracted using the formula: ； Generate a tumor region partition identification feature value table; in, Represents the identification feature value of the tumor area, is the average brightness of the pixel block in the region, is the brightness variance, is the shape parameter, and It is the adjustment factor determined based on data sensitivity analysis.

8. A tumor partition recognition system based on multi-scale image analysis, characterized in that: According to any one of claims 1 to 7, the method for tumor partition identification based on multi-scale image analysis comprises: The grayscale distribution analysis module divides the pixel block groups into preset sizes based on the input medical tumor partition image, extracts the grayscale value in each pixel block, counts and classifies the pixel blocks with differentiated grayscale ranges, performs grayscale distribution statistical analysis, and generates a multi-scale grayscale distribution map; The boundary texture analysis module extracts the boundary area in the tumor partition image based on the multi-scale grayscale distribution map, identifies the texture features in the boundary area, compares the grayscale difference of adjacent pixels using grayscale gradient calculation, and quantifies and classifies them according to the gradient value to generate a boundary connection weight table; The geometric feature acquisition module extracts geometric features within the boundary area based on the boundary connection weight table, including perimeter, area, and edge connectivity, uses perimeter-to-area ratio and edge curvature indicators to perform morphological analysis on the area, compares the geometric features of the area, and combines the spatial expansion information to obtain spatial geometric expansion data; The spatial geometry expansion feature analysis module analyzes the expansion ratio and distribution characteristics of each region based on the spatial geometry expansion data, identifies the spatial expansion characteristics of the differentiated regions, performs spatial expansion feature analysis, and obtains regional expansion ratio data; The spatial zoning delineation module analyzes the morphological changes and spatial distribution characteristics within the region based on the regional expansion ratio data, identifies the spatial distribution patterns of differentiated regions in combination with the morphological characteristics of the boundaries, performs regional refinement division based on the identification results, and generates a numerical table of tumor regional zoning identification characteristics.

Images