Medical image processing method for tumor partition recognition

Abstract

The invention relates to the technical field of image segmentation, in particular to a medical image processing method for tumor partition recognition, which comprises the following steps: acquiring a multi-modal medical image, extracting gray features and adjusting image gray, recognizing tissue boundary and metabolism features, grouping and generating tumor boundary response partitions, extracting heterogeneous nodes and classifying and labeling. Dividing a core area, an infiltration area and a necrosis area, and generating a tumor functional structure chart. According to the method, the gray frequency and the edge trend of the multi-modal image are unified, the structure contrast is enhanced, the consistency of each modal image in space and gray level is ensured, the information fusion integrity is improved, the gray, texture and metabolism characteristics are combined, the density mutation and metabolism enhancement region is identified, and the boundary clear contour is constructed; boundary partitions are established through response direction consistency, the cross-modal discrimination ability is enhanced, texture and metabolic characteristics are fused to classify heterogeneous nodes, a core area, an infiltration area and a necrosis area are divided, and functional partitions with clear expression and clear structures are generated.

Description

Technical Field

[0001] The present invention relates to the technical field of image segmentation, and in particular to a medical image processing method for tumor partition identification. Background Art

[0002] The field of image segmentation technology includes the processing and analysis of various image types such as medical images, satellite images, and industrial inspection images, aiming to identify and divide regions of specific significance from the original image. In medical image analysis, image segmentation technology is mainly used to identify organ structures, lesion areas, and their boundary information, providing important references for clinical diagnosis, treatment planning, and efficacy evaluation. The core content of this technical field includes image preprocessing, feature extraction, region classification, and boundary recognition. The main methods include threshold-based segmentation, region-based segmentation, graph theory-based segmentation, and deep learning-driven segmentation technology. In recent years, with the development of deep convolutional neural networks, the application of image segmentation in medical image processing has been significantly improved, but it still faces challenges in complex lesion morphology, multimodal data fusion, and recognition of fuzzy boundary areas.

[0003] Among them, the medical image processing method for tumor zoning identification refers to a processing flow based on multimodal medical image input (including CT and MRI, etc.), which introduces image registration, image fusion and feature extraction operations to perform multi-scale segmentation and regional identification of tumor areas in the image. This method mainly covers the use of image registration to unify the spatial reference system of images of different modalities, the use of image fusion to enhance the complementary characteristics between the modalities, and the extraction of detailed features such as edge texture and density distribution, combined with a neural network model to complete the automatic identification and zoning of tumor areas. In particular, for small lesions and areas with unclear boundaries, structure-sensitive filtering and multi-scale feature encoding methods are used to improve the ability to express the complexity of tumor morphology, thereby completing the identification and division of different pathological areas.

[0004] In medical image processing, existing technologies typically rely on a single image modality for structural recognition. Although multimodal input is introduced during processing, the lack of coordination between grayscale and edge trends leads to gaps in spatial and information integration between different modalities, compromising the integrity of subsequent feature extraction. For lesion identification, existing methods primarily rely on contour edge and intensity feature extraction, making it difficult to address complex regions formed by texture discontinuities, density mutations, and metabolic discontinuities, leading to biased boundary identification. During the functional delineation phase, region identification is often limited to a single metric, such as density or metabolism, which fails to reflect the heterogeneity of tumor tissue along the functional dimension. This is particularly true in regions where multidimensional features coexist, leading to functional misclassification or omission. For example, existing systems may mistakenly classify low-metabolism but high-density areas as non-lesion regions, reducing the accuracy of image interpretation. The lack of a comprehensive processing pathway tailored to the response trends of different modalities and structural heterogeneity has become a core bottleneck restricting existing technologies in processing complex tumor expression patterns. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a medical image processing method for tumor partition identification.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a medical image processing method for tumor zoning identification, comprising the following steps: S1: Acquire computed tomography images, tissue structure images from magnetic resonance imaging, fluid-sensitive images, and positron emission tomography images, extract the grayscale distribution characteristics of the structural regions in the images, adjust the grayscale of the images based on the grayscale frequency and edge change trend, and generate a grayscale consistent structural image set; S2: Based on the tissue distribution information extracted from the grayscale consistency structure image set, identifying density mutations, texture interruptions, and metabolic enhancement locations in the edge area, judging the coherence and difference of the spatial contours, and generating an initial contour map of the tumor area boundary; S3: Calling edge information extracted from the initial contour map of the tumor region boundary, determining metabolic activity boundaries in the positron emission tomography image, identifying structural density boundaries in the computed tomography image, grouping boundaries according to direction differences and consistency of response characteristics, and generating a tumor boundary response partition set; S4: Based on the tumor boundary response partition set, the texture and metabolic characteristics of the tumor nodes are extracted in the atlas, the feature expressions are pattern-classified and regional characteristics are annotated to generate a heterogeneous node distribution map of the tumor tissue atlas.

[0007] As a further solution of the present invention, the grayscale consistency structural image set includes a grayscale frequency distribution feature set and an edge change trend feature set; the initial contour map of the tumor area boundary includes a density mutation point set, a texture interruption area set, and a metabolic enhancement position set; the tumor boundary response partition set includes a metabolic activity boundary group, a structural density boundary group, and a direction difference response group; the tumor tissue atlas heterogeneous node distribution map includes a texture feature pattern set, a metabolic feature pattern set, and a regional characteristic annotation set.

[0008] As a further solution of the present invention, the specific steps of S1 are: S101: After acquiring a computed tomography image, a tissue structure image from a magnetic resonance imaging (MRI), a fluid-sensitive image, and a positron emission tomography (PET) image, grayscale value information of layered regions in the image is retrieved, and grayscale frequency features are extracted based on the spatial distribution of pixels and grayscale gradient changes to generate grayscale distribution frequency value intervals; S102: extracting grayscale change trends at the edges of the structural regions according to the grayscale distribution frequency value intervals, classifying and combining change trend information based on grayscale differences and spatial position relationships between adjacent pixels, and generating grayscale change edge gradient segments; S103: Based on the grayscale change edge gradient segment, extract a pixel set with similar grayscale change characteristics, make a judgment based on the grayscale difference and the boundary gradient, screen the pixel set that meets the smoothing condition, complete the grayscale mapping process, and generate a grayscale consistency structure image set.

[0009] As a further solution of the present invention, the specific steps of S2 are: S201: Obtain pixel intensity distribution and neighborhood gradient information in the grayscale consistency structure image set, combine the edge area positioning result, perform difference judgment on the grayscale change and gradient direction difference within the edge, identify the grayscale mutation segment and determine whether it is a boundary abnormal area, and generate the edge grayscale mutation distribution value; S202: extracting the regional texture distribution direction and boundary spacing based on the edge grayscale mutation distribution value, identifying the texture boundary interruption area in combination with the grayscale roughness value, and performing boundary structure fitting on the interruption position to generate the texture boundary distribution fracture value; S203: Calling the texture boundary distribution fracture value, collecting the metabolic signal enhancement value and spatial signal density of the corresponding position, determining the correspondence between the spatial distribution of the metabolic enhancement area and the texture fracture area, establishing a spatial mapping relationship of the tumor edge contour in the image, and obtaining an initial contour map of the tumor area boundary.

[0010] As a further solution of the present invention, the specific steps of S3 are: S301: Retrieving edge information extracted from the initial contour map of the tumor region boundary and locating it in the positron emission image, calculating a metabolic activity feature value based on the relationship between the grayscale value of the corresponding pixel and the metabolic activity threshold, selecting a location set of metabolic activity display areas, and generating a metabolic activity edge coordinate set; S302: extracting structural density information of corresponding coordinates in the computed tomography image according to the metabolic activity edge coordinate set, screening boundary points according to a density threshold, and integrating the screening results based on consistency of spatial positions to generate edge groups corresponding to structural density.

[0011] S303: Based on the direction characteristics and edge response consistency of the pixel points in the edge group corresponding to the structural density, a set of pixel points meeting the difference characteristics is screened, and boundary classification and grouping are performed to generate a tumor boundary response partition set.

[0012] As a further embodiment of the present invention, the specific calculation formula for calculating the metabolic activity characteristic value is: ; in, Represents the coordinates in the positron emission image The pixel gray value, represents the metabolic activity threshold, The initial contour of the tumor boundary is represented in The spatial gradient amplitude at Represents the weight adjustment coefficient of spatial gradient and grayscale difference, stands for metabolic threshold morphological correction factor.

[0013] As a further solution of the present invention, the specific steps of S4 are: S401: Based on the tumor boundary response partition set, extract the image grayscale and metabolic activity information of each partition, construct corresponding texture and metabolic feature parameters, integrate the two types of parameters in the same spatial dimension, calculate the fusion feature value, and generate the partition joint feature value; S402: calling the partition joint feature value, demarcating the interval range of texture and metabolic features, performing feature classification operations on the difference partitions, constructing feature cluster labels, and generating texture metabolic cluster numerical features; S403: Mapping the clustering results to the original atlas according to the texture metabolism clustering numerical features, assigning cluster numbers to the node identifiers and assigning image identification areas to generate a heterogeneous node atlas distribution map.

[0014] As a further solution of the present invention, the specific calculation formula for calculating the fusion feature value is: ; in, represents the normalized value of the texture feature of the i-th partition, represents the normalized value of metabolic characteristics of the i-th partition, is the cube root stability coefficient, Represents the amplitude of the high-frequency component of the wavelet transform in the k-th dimension of the texture feature, Represents the k-th dimension of metabolic characteristics in the frequency domain energy density value, Represents the peak value of the texture feature space autocorrelation function, represents the integral value of the metabolic feature space power spectrum, is the product smoothing factor.

[0015] As a further embodiment of the present invention, the method further comprises: S5: calling the heterogeneous node distribution map of the tumor tissue atlas, dividing it into core area, infiltration area and necrotic area according to the difference in node characteristics, completing image annotation in the form of spatial structure, and generating a tumor functional zonation structure map; The tumor functional zonation structure diagram includes core area space marking, infiltration area space marking, and necrotic area space marking.

[0016] As a further solution of the present invention, the specific steps of S5 are: S501: Based on the heterogeneous node distribution map of the tumor tissue atlas, the morphological index, spatial coordinates and number of neighbors of the node are obtained, and a joint judgment is made according to the spatial aggregation characteristics and the morphological change trend to generate an initial partition value of the partition structure; S502: calling a node set of a difference region in the initial partition value of the partition structure, extracting its morphological index content, comparing the morphological changes of adjacent nodes based on the difference between regions, and generating a morphological boundary offset strength coefficient; S503: Selecting regional nodes with prominent changes according to the offset of the boundary nodes in the morphological boundary offset intensity coefficient, integrating spatial position and regional information, and generating a tumor functional zoning structure diagram.

[0017] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by unifying the grayscale frequency and edge trend of multimodal images, enhancing structural contrast, ensuring the consistency of images of each modality at the spatial and grayscale levels, and improving the integrity of information fusion, the grayscale, texture and metabolic characteristics are combined to identify density mutations and metabolic enhancement areas, construct a clear boundary contour, establish boundary partitions through response direction consistency, enhance cross-modal discrimination capabilities, fuse texture and metabolic features to classify heterogeneous nodes, divide the core area, infiltration area and necrotic area, and generate functional partitions with clear expression and structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of the steps of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present 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 only used to explain the present invention and are not intended to limit the present invention.

[0020] 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, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0021] See also Figure 1 , a medical image processing method for tumor partition identification, comprising the following steps: S1: Acquire computed tomography images, tissue structure images from magnetic resonance imaging, fluid-sensitive images, and positron emission tomography images, extract the grayscale distribution characteristics of the structural regions in the images, adjust the grayscale of the images based on the grayscale frequency and edge change trend, and generate a grayscale consistent structural image set; S2: Based on the tissue distribution information extracted from the grayscale consistency structure image set, density mutations, texture interruptions, and metabolic enhancement locations are identified in the edge area, the coherence and difference of the spatial contours are determined, and the initial contour map of the tumor area boundary is generated; S3: Call the edge information extracted from the initial contour map of the tumor region boundary, determine the metabolic activity boundary in the positron emission tomography image, identify the structural density boundary in the computed tomography image, group the boundaries according to the difference in direction and the consistency of the response characteristics, and generate a tumor boundary response partition set; S4: Based on the tumor boundary response partition set, the texture and metabolic features of the tumor nodes are extracted from the atlas, the feature expressions are classified into patterns and regional characteristics are annotated to generate a heterogeneous node distribution map of the tumor tissue atlas; S5: Call the heterogeneous node distribution map of the tumor tissue atlas, divide it into core area, infiltration area and necrotic area according to the difference of node characteristics, complete the image annotation in the form of spatial structure, and generate the tumor functional zoning structure map.

[0022] The grayscale consistency structural image set includes a grayscale frequency distribution feature set and an edge change trend feature set. The initial contour map of the tumor region boundary includes a density mutation point set, a texture interruption region set, and a metabolic enhancement position set. The tumor boundary response partition set includes a metabolic activity boundary group, a structural density boundary group, and a trend difference response group. The heterogeneous node distribution map of the tumor tissue atlas includes a texture feature pattern set, a metabolic feature pattern set, and a regional characteristic annotation set. The tumor functional zoning structure map includes core area spatial annotation, infiltration area spatial annotation, and necrotic area spatial annotation. Grayscale frequency distribution feature set explanation: This feature set is based on statistical analysis of the distribution of pixel grayscale values ​​in multimodal images. It constructs frequency segments (slow change, medium change, and sudden change) according to the grayscale gradient changes. It is used to reflect the overall grayscale stability and the delicacy of the internal structure of the image. It is the basic input for subsequent edge recognition and texture clustering. Explanation of edge change trend feature set: This feature set detects the grayscale change speed (rate of change) of the edge area of ​​the structure in the image to form different gradient segments (slow change, medium change, and sudden change edges). It is used to identify the direction of the tissue structure boundary and its sudden change trend, which is the key to the subsequent generation of contour maps and boundary response partitions. Explanation of the density mutation point set: By detecting the location of grayscale value mutations and evaluating their concentration in the edge area, a set of density mutation points is formed, which is one of the components of the "initial contour map" and is used to identify significant change points in the potential tumor boundary; Texture break area set explanation: This set detects discontinuities in the texture distribution direction in the image (such as sudden changes in texture direction, abnormal texture roughness) and fits the boundaries to identify texture structure fracture areas, which is used to assist in judging structural heterogeneity and spatial splitting phenomena. The metabolic enhancement location set combines the signal intensity (metabolic value) and signal density in the PET image to identify areas of high metabolic signal concentration. After comparing with the texture break area, a spatial map is constructed to generate a complete edge contour. Metabolic Activity Boundary Group Explanation: By calculating metabolic activity characteristic values ​​(such as SUV), the coordinate set of the metabolic high response area is extracted from the initial contour map to form the metabolic activity edge, which is classified into the "response partition set" for subdividing regional differences.

[0023] Explanation of the structure density boundary group: Judging the density of tissue at the boundary by the structural density value in the CT image and forming a group of edge points with clear structural features is an important part of spatial consistency judgment and boundary matching. Explanation of direction difference response group: Based on the consistency analysis of directional gradient and edge response value, pixels with different directional performance are classified and grouped to form important subgroups of boundary response partition, which is used to enhance the stability of boundary mapping between modalities; Explanation of texture feature pattern set: Texture features such as grayscale run matrix, wavelet high frequency coefficient, autocorrelation function peak, etc. are extracted from this set, forming one of the main feature dimensions of texture clustering and the basis for heterogeneous node classification; Metabolic feature pattern set explanation: Metabolic features include SULpeak, metabolic frequency domain energy density, etc., and texture features are integrated for clustering to reflect the functional activity differences in different regions; Explanation of regional feature annotation set: The cluster numbering results are mapped back to the original atlas, and each image node is marked with a feature type identifier (such as a combination of high texture and low metabolism), which is ultimately used to express the heterogeneous atlas; Explanation of core area space annotation, infiltration area space annotation, and necrotic area space annotation: By comprehensively considering the morphological indicators, spatial aggregation degree and structural offset intensity of the nodes, they are divided into three major functional area types and spatially explicitly annotated in the image, which is the final output of the tumor functional structure map.

[0024] The specific steps of S1 are: S101: After acquiring a computed tomography image, a tissue structure image from a magnetic resonance imaging (MRI), a fluid-sensitive image, and a positron emission tomography (PET) image, grayscale value information of layered regions in the image is retrieved, and grayscale frequency features are extracted based on the spatial distribution of pixels and grayscale gradient changes to generate grayscale distribution frequency value intervals; After acquiring computed tomography images, magnetic resonance imaging tissue structure images, liquid sensitive images and positron emission images, it is necessary to read the image matrix data according to the image type, load the image in DICOM format and uniformly convert it into a single-channel grayscale image format, then extract the grayscale value of the pixel point in the image and its corresponding spatial coordinates, build a pixel space distribution index according to the row and column positions, calculate the grayscale difference between each pixel and other pixels in its neighborhood in turn, record the grayscale gradient change, and divide it into three sections: slow change, medium change and sudden change. The slow change area is the grayscale difference below 10, the medium change area is 11 to 30, and the sudden change area is 31 or above, count the number of pixels appearing in each segment, and convert it proportionally to the total number of pixels in the image to obtain the frequency percentage. For example, the slowly changing area accounts for 52%, the moderately changing area accounts for 30%, and the sudden change area accounts for 18%. The frequencies corresponding to these three segments are integrated to generate a grayscale frequency distribution vector, and the complete grayscale distribution frequency value interval is formed by combining the intervals. This interval can reflect the grayscale change trend of the entire image. For example, in magnetic resonance images, sudden change areas are often formed at the tissue boundaries, and slowly changing areas are mostly formed inside normal tissues. The distribution result can be used as the grayscale feature basis for subsequent partition judgment, and finally a grayscale distribution frequency value interval is generated.

[0025] S102: Extracting grayscale change trends at the edges of the structural regions based on the grayscale distribution frequency value intervals, classifying and combining change trend information based on grayscale differences and spatial position relationships between adjacent pixels, and generating grayscale change edge gradient segments; The grayscale change trend of structural edges in images is extracted based on the grayscale distribution frequency value interval. First, a set of pixel points on the edge of the structural contour is selected. The grayscale value change amplitude between two adjacent pixels is detected in order of their arrangement. The grayscale change rate is calculated based on the spatial distance between the corresponding position coordinates. The grayscale change rate is then distinguished into slowly changing edges, moderately changing edges, and suddenly changing edges based on the change rate. For example, a grayscale change rate less than 0.5 is classified as a slowly changing segment, 0.5 to 1.5 as a moderately changing segment, and greater than 1.5 as a suddenly changing segment. Each type of change trend is labeled and numbered. Then, regions are grouped according to the consecutive pixel positions. The change rate ranges within each group are merged to obtain a unified grayscale change feature. Edge segments with similar grayscale gradient trends are identified and integrated as a complete edge gradient segment. For example, if the change rate of more than 30 consecutive pixels on the tumor boundary in an image is concentrated between 1.2 and 2.3, it is classified as a medium-high change segment. This segmentation can be used to determine the regional structural complexity and boundary strength, and ultimately generate grayscale change edge gradient segments.

[0026] S103: Based on the grayscale change edge gradient segment, extract a set of pixels with similar grayscale change characteristics, make a judgment based on the grayscale difference and the boundary gradient, select a set of pixels that meet the smoothing condition, complete the grayscale mapping process, and generate a grayscale consistent structure image set; Based on the grayscale change edge gradient segment, a set of pixels with similar grayscale change characteristics is extracted, and pixel regions with consistent change attributes are identified and classified. First, the pixels in each gradient segment are grouped, and the grayscale average and variation degree of the pixels in each group are counted respectively. The grayscale change amplitude is calculated as the difference reference standard. If the grayscale standard deviation in a group does not exceed 20% of the average value, it is judged as a grayscale distribution stable region. Then, the difference between the maximum and minimum edge gradients is obtained. If the difference is less than the boundary stability threshold of 25, the region meets both the grayscale difference and boundary stability conditions and is marked as a grayscale consistency region. These pixels are reassigned a uniform grayscale value, and a mapping operation is performed and the spatial position of the region in the image is recorded. For example, the average grayscale of the pixels in a region is 135, the standard deviation is 21, and the grayscale change ratio is 15.6%, which meets the judgment criteria. The edge change difference is 22, which is less than the set threshold, so it is included in the mapping range. The mapping results of all regions that meet the conditions are merged and output, and finally a grayscale consistency structure image set is generated.

[0027] The specific steps of S2 are: S201: Obtain pixel intensity distribution and neighborhood gradient information in the grayscale consistency structure image set, combine the edge area positioning results, perform difference judgment on the grayscale change and gradient direction difference within the edge, identify the grayscale mutation segment and determine whether it is a boundary abnormal area, and generate the edge grayscale mutation distribution value; First, traverse each pixel in the image, collect its grayscale value and extract the grayscale difference change direction of the adjacent area. The grayscale difference direction information is obtained by comparing the grayscale in the horizontal and vertical directions to form the gradient direction and amplitude corresponding to each pixel. Then, combined with the located edge area, select a set of pixels within a certain range inside the edge line, extract the grayscale value and gradient direction of the pixels in the set, and perform difference operations on them with the corresponding data in the adjacent pixel area to construct a difference data matrix. The difference judgment standard is based on the grayscale change setting of the median and standard deviation of the image grayscale distribution. The judgment threshold value is used as the benchmark for identifying mutation points. If the difference in grayscale or gradient direction of a pixel exceeds the judgment threshold, it is marked as a grayscale mutation point. All mutation points are grouped into a set and compared with the original edge positioning result to determine their concentration within the edge. When the area occupied by mutation points in a certain region exceeds 1.5% of the total pixel area, it is considered that there is a structural abnormality in the region. Finally, the ratio of the number of mutation points in the abnormal region to the area of ​​the region where they are located is counted to obtain the grayscale mutation distribution density used to describe the edge change trend, which is used as the edge grayscale mutation distribution value.

[0028] S202: extracting the regional texture distribution direction and boundary spacing based on the edge grayscale mutation distribution value, identifying the texture boundary interruption area in combination with the grayscale roughness value, and performing boundary structure fitting on the interruption position to generate the texture boundary distribution fracture value; According to the edge grayscale mutation distribution value, texture direction information is collected for the boundary nodes in the abnormal area. The texture direction is obtained by the most concentrated arrangement direction in the image grayscale arrangement. The nearest vertical distance between this direction and the edge in the image is used to evaluate the degree of correlation between the texture structure and the boundary. If the texture direction changes significantly in different areas and the boundary spacing is less than 3 pixels, it is preliminarily identified as a direction jump area. At the same time, the grayscale roughness of the area is collected. The roughness is measured by the degree of fluctuation of the grayscale value of the local area. When the roughness of a certain area is more than 15% higher than the set benchmark value of the average roughness of the image, it is determined that there is a discontinuity in its texture continuity. The boundary structure of this interruption position is fitted with continuous points, and the boundary fitting error reflects the degree of boundary continuity. If the error value exceeds the boundary fitting benchmark limit, the position is determined to be a texture break point. All break points are sorted along the boundary and their number is counted. The proportion of them to the entire boundary length is the texture boundary distribution breakage value, which is used to reflect the degree of discontinuity of the texture structure distribution in the image.

[0029] S203: Calling the texture boundary distribution fracture value, collecting the metabolic signal enhancement value and spatial signal density of the corresponding position, determining the correspondence between the spatial distribution of the metabolic enhancement area and the texture fracture area, establishing a spatial mapping relationship of the tumor edge contour in the image, and obtaining an initial contour map of the tumor area boundary; The texture boundary distribution fracture value is called to determine the location area corresponding to the boundary breakpoint set. The metabolic signal enhancement value and spatial signal density of this location are extracted from the image multi-channel. The metabolic enhancement value is obtained by the brightness contrast before and after enhancement. The signal density is measured by the number of signal points per unit area. When the enhancement value is more than 20% higher than the image average enhancement reference value and the signal density exceeds 0.7, it is identified as a high metabolic accumulation area. These metabolic enhancement areas are then spatially compared with the aforementioned fracture areas, and the degree of overlap is calculated. When the overlapping area ratio exceeds 60%, it is identified as a spatial matching area. The boundary points of the spatial matching area are summarized, and their two-dimensional coordinate positions and the belonging area numbers are recorded. The spatial relationship set of the boundary points is constructed, and the boundary points are connected to form the edge of a complete closed area. Finally, the initial contour map of the tumor area boundary is generated for subsequent regional contour analysis and partitioning.

[0030] The specific steps of S3 are: S301: Retrieving edge information extracted from the initial contour map of the tumor region boundary and locating it in the positron emission image, calculating metabolic activity feature values ​​based on the relationship between the grayscale value of the corresponding pixel and the metabolic activity threshold, selecting a location set of metabolic activity-displaying regions, and generating a metabolic activity edge coordinate set; The specific calculation formula for calculating the metabolic activity characteristic value is: ; in, Represents the coordinates in the positron emission image The pixel gray value, represents the metabolic activity threshold, The initial contour of the tumor boundary is represented in The spatial gradient amplitude at Represents the weight adjustment coefficient of spatial gradient and grayscale difference, represents the metabolic threshold morphological correction factor; According to the clinical PET-CT imaging data specification, the gray value of the positron emission tomography image The maximum SUV value of a tumor region was 6.2, and the metabolic activity threshold was quantified by the standardized uptake value (SUV). The clinical standard of SUV ≥ 2.5 was adopted; Spatial gradient amplitude After calculation by the Sobel operator, the gradient amplitude measured at a certain edge point is 3.8; Weight adjustment coefficient According to the covariance matrix calculation of gradient amplitude and grayscale difference, when the standard deviation of gradient amplitude is 1.2 and the standard deviation of grayscale difference is 2.1, the eigenvalue decomposition of the covariance matrix is ; Morphological correction factor By calculating the fractal dimension of tumor morphology, when the fractal dimension of a case is D=1.78, ; Substitute the parameters: , numerator , the denominator ; final ; This value exceeds the preset threshold of 3.0, and the coordinate point is determined to belong to the edge area of ​​metabolic activity and is included in the coordinate set.

[0031] S302: extracting structural density information of corresponding coordinates in the computed tomography image based on the metabolic activity edge coordinate set, screening boundary points based on a density threshold, and integrating the screening results based on spatial position consistency to generate edge groups corresponding to structural density; Based on the metabolic activity edge coordinate set, the corresponding positions in the computed tomography image are located, and the structural density values ​​of these coordinate points are extracted. The structural density is expressed as the grayscale intensity of each pixel in the CT image, and the corresponding value range is usually between -1000 and 300 H. Based on experience, the structural density threshold is set to 80 H. The density values ​​of all coordinate points are screened point by point. When the density value of a pixel point exceeds this threshold, it is considered to be a boundary point with clear structural characteristics. After screening, spatial position consistency judgment is introduced. For each point that meets the density requirement, if at least 5 of its 8 adjacent points also meet the density condition, the point is retained. If not, it is eliminated to eliminate isolated points or image noise interference. For example, if the density of a pixel point is 95 H and there are 6 points in the neighborhood with density values ​​higher than 80 H, the point is considered a valid boundary point. The set of valid boundary points constructed in this way constitutes the edge group corresponding to the structural density.

[0032] S303: Based on the directional characteristics and edge response consistency of the pixel points in the edge group corresponding to the structural density, a set of pixel points that meet the difference characteristics is selected, and boundary classification and grouping are performed to generate a tumor boundary response partition set; Based on the directional characteristics and edge response consistency of each pixel in the edge group corresponding to the structural density, each point is preliminarily classified according to the direction of grayscale change. Directional information is extracted based on the horizontal and vertical grayscale differences of each point in the image. The directional angle is determined by the ratio of the grayscale changes in the two directions. Different angle intervals are classified into different types. For example, if the directional change angle of a point is close to 45 degrees, it is classified as a diagonal edge point. After all pixels are classified, their edge response consistency is compared. The judgment method is to compare whether the directional difference of neighboring pixels is less than the set angle range and whether the difference of their response values ​​is less than the set amplitude. The directional difference is set to 20 degrees, and the response value change is controlled within 15 units. Pixels that meet the conditions are retained and assigned to the same group. For example, if the directional angle of a point is 60 degrees, and most of the neighboring points have angles between 50 and 70 degrees and similar response values, the point is assigned to the same boundary segment. Finally, all points are classified based on the principles of consistent direction and similar response. They are integrated into multiple regional boundary response segments and spliced ​​and integrated based on spatial continuity to obtain a set of tumor boundary response partitions.

[0033] The specific steps of S4 are: S401: Based on the tumor boundary response partition set, extract the image grayscale and metabolic activity information of each partition, construct corresponding texture and metabolic feature parameters, integrate the two types of parameters in the same spatial dimension, calculate the fusion feature value, and generate the partition joint feature value; The specific calculation formula for calculating the fusion eigenvalue is: ; in, represents the normalized value of the texture feature of the i-th partition (calculated based on the long run dominance of the grayscale run length matrix), represents the normalized value of the metabolic signature of the i-th partition (calculated based on SULpeak and tumor / liver ratio), is the cube root stability coefficient (the value is 1×10⁻ 4 ), Represents the amplitude of the high-frequency component of the wavelet transform in the k-th dimension of the texture feature, Represents the k-th dimension of metabolic characteristics in the frequency domain energy density value, Represents the peak value of the texture feature space autocorrelation function, represents the integral value of the metabolic feature space power spectrum, is the product smoothing factor (the value is the inverse of the spatial resolution); Texture feature normalization value The long run advantage is calculated by the grayscale run length matrix, and the formula is: ,in is the long stroke emphasis factor (measured value range is 0.5-2.0, take 1.2), is the grayscale non-uniformity coefficient (measured value range is 0.1-0.9, take 0.3); Metabolic characteristic normalized values The formula is calculated by SULpeak and tumor / liver ratio: ,in is the maximum standardized uptake value of the tumor in PET images (measured value 3.5), is the average liver uptake value (measured value 2.0); Cube root stability coefficient According to the CT image resolution setting, the value is 1×10⁻ 4 (Calculated based on 0.5mm voxel size). Amplitude of high-frequency component of texture feature space wavelet transform Obtained through Daubechies wavelet basis function decomposition, the amplitude of the k-th dimension component is 3.8 (measured range 2.0-5.0);

[0034] Frequency domain energy density value of metabolic feature space The energy density in the kth dimension was calculated to be 4.2 (measured range 3.0-6.0) by Fourier transform power spectrum integration; Texture feature space autocorrelation function peak Calculated by the maximum value of the autocorrelation function, taking 2.5 (range 1.0-3.0); Metabolic feature space power spectrum integral value By calculating the energy accumulation in the frequency domain, we take 1.8 (range 1.0-2.5); Multiplicative smoothing factor According to the MRI spatial resolution setting of 0.7 mm, the value is 1.428 (1 / 0.7); Substituting into the formula, we get: the numerator of the first term is 1.2×0.923=1.108, and the denominator is , the first result is 0.699; The numerator of the second term is 3.8×4.2=15.96, and the denominator is 2.5×1.8+1.428=4.5+1.428=5.928. The result of the second term is 2.691. final The results indicate that the joint eigenvalue is determined by the product of texture and metabolic features and the spatial coupling term. A value greater than 1.0 reflects a significant synergistic effect of the features. The value of 3.39 needs to be compared with the clinical threshold of 2.5 to determine the degree of tumor heterogeneity.

[0035] S402: calling the partition joint feature value, demarcating the interval range of texture and metabolic features, performing feature classification operations on the difference partitions, constructing feature cluster labels, and generating texture metabolic cluster numerical features; The partitioned joint feature values ​​are used to classify the texture and metabolic features into intervals, with thresholds set for each parameter based on their numerical distribution. For example, texture parameter grayscale contrast values ​​between 0.4 and 1.2 are divided into three levels based on frequency distribution, corresponding to low, medium, and high values. Metabolic parameter coefficients of variation between 0.05 and 0.25 are also divided into three intervals. Based on this, each node is classified based on its combination of values ​​in the two parameter categories. For example, if a node's grayscale contrast falls within the median interval and its coefficient of variation also falls within the median interval, it is classified as "medium texture + medium metabolism." After the combined classification is completed, cluster labels are assigned to each combination type. The number of clusters is set, and node clustering is performed. By comparing the differences in node features point by point, a cluster distribution is formed. Upon completion, each node is labeled with a unified cluster number. Each cluster number stores the average texture value and average metabolic parameter value, along with their corresponding deviation and stability. This constructs a unified feature label structure for use in graph mapping, ultimately generating texture and metabolic cluster numerical features.

[0036] S403: Mapping the clustering results to the original atlas based on the texture metabolism clustering numerical features, assigning cluster numbers to the node identifiers and assigning image identification areas to generate a heterogeneous node atlas distribution map; Based on the numerical characteristics of texture metabolism clustering, the cluster label of each node is mapped to the original atlas. Based on the positional coordinates of each node in the original atlas, a cluster number field is added as an identification basis. This number is recorded as an additional attribute of the node, alongside the original structure. A clear mapping relationship is established between the atlas coordinates of the node and the cluster number. For example, if node A has a spatial position of (x1, y1) and a cluster number of C1, it is labeled as (x1, y1, C1). Subsequently, during the image generation phase, spatial regions of the image are assigned values ​​based on the node cluster number. Image annotation uses number coding or graphical color coding to distinguish regions. Node regions are identified by cluster number within the original atlas coordinate system. After complete mapping, each node in the atlas is assigned a clear cluster label, forming a clearly classified atlas distribution map for the entire image space. Cluster information is embedded in the structure while preserving the original atlas spatial relationships, ultimately generating a heterogeneous node atlas distribution map.

[0037] The specific steps of S5 are: S501: Based on the heterogeneous node distribution map of the tumor tissue atlas, the morphological index, spatial coordinates and number of neighbors of the node are obtained, and the initial partition value of the partition structure is generated by making a joint judgment based on the spatial aggregation characteristics and morphological change trends; First, the basic parameters of each node in the image are extracted, including spatial coordinates, morphological characteristics, and the number of neighbors. Image reading can be completed using image processing tools. The spatial coordinates of the node are set based on the pixel position of the cell nucleus center, and the number of neighbors is obtained by counting the adjacent nodes within a certain radius. Morphological characteristics can be constructed using indicators such as area, circularity, and aspect ratio. After recording the parameters of each node, the spatial aggregation state needs to be classified. The degree of aggregation is determined by the number of adjacent nodes per unit area. The judgment range is set as below a certain density value, marked as sparse, above another density value, marked as dense, and the intermediate range is marked as moderately aggregated. The morphological change trend is analyzed by analyzing the difference in morphological feature vectors between nodes. The difference intensity between all nodes can be calculated and then sorted by numerical value to determine the degree of change significance. Then, fixed weights are assigned to the spatial aggregation index and the morphological difference index to form a comprehensive node score. The scoring range is set as above a certain value, marked as the core area, the intermediate range, marked as the infiltrated area, and below a certain value, marked as the necrotic area. The division of each category can be quantitatively determined by statistical distribution. After the scoring is completed, the correspondence between node and region type is established to generate the initial partitioning value of the partition structure.

[0038] S502: Calling the node set of the difference area in the initial partition value of the partition structure, extracting its morphological index content, comparing the morphological changes of adjacent nodes based on the difference between the regions, and generating a morphological boundary offset strength coefficient; The node sets in different regions of the initial partition value of the partition structure are called to extract their morphological parameters, including the average area of ​​the nucleus and the difference in cytoplasmic texture. The boundary of each nucleus is identified by the image contour analysis method, and the area within the pixel contour is obtained to calculate the average value. The cytoplasmic texture information can be used to measure the degree of grayscale change according to the grayscale distribution law of the image and record it as a heterogeneity index. The morphological parameters of adjacent nodes in each region are then compared in pairs. The degree of difference is expressed as the ratio of the corresponding parameter difference between the two nodes to the reference value. The larger the difference, the stronger the morphological change. After sorting out the morphological offset degree between all nodes, the difference weight is assigned according to the region where the node is located. For example, the weight of the core area node is larger, and the weight of the necrotic area is smaller. All difference values ​​are matched with the corresponding weights and summarized to calculate the average offset degree of the region. The offset degree is used as the boundary change indicator of the region, and the morphological boundary offset intensity coefficient is further generated.

[0039] S503: Selecting regional nodes with significant changes based on the offset of boundary nodes in the morphological boundary offset intensity coefficient, integrating spatial position and regional information, and generating a tumor functional zoning structure diagram; Based on the boundary node values ​​recorded in the morphological boundary offset intensity coefficient, nodes whose offset intensity values ​​exceed the upper limit of the overall distribution are identified. The upper limit value can be set to the 80th percentile node value in the offset intensity distribution. Nodes exceeding this value are judged as significantly offset nodes. The spatial coordinate information of these nodes and the regional identification of their locations are collected, and position and type pairing data are constructed. The data are organized into a structured matrix to record the locations and regional affiliations of all significant nodes. The matrix content is converted into an image annotation format, and labels are drawn at the corresponding coordinates in the atlas. Color or shape marks are distinguished according to the regional type to complete the visual expression of the regional range, and finally a tumor functional zoning structure diagram with a spatial structural hierarchy is formed.

[0040] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A medical image processing method for tumor partition identification, characterized in that: The following steps are involved: Acquire computed tomography images, tissue structure images from magnetic resonance imaging, fluid-sensitive images, and positron emission tomography images, extract the grayscale distribution characteristics of the structural areas in the images, adjust the grayscale of the images based on the grayscale frequency and edge change trend, and generate a grayscale consistent structural image set; Based on the tissue distribution information extracted from the grayscale consistency structure image set, density mutations, texture interruptions, and metabolic enhancement locations are identified in the edge area, the coherence and difference of the spatial contours are determined, and an initial contour map of the tumor area boundary is generated; Invoking edge information extracted from the initial contour map of the tumor region boundary, determining metabolic activity boundaries in the positron emission tomography image, identifying structural density boundaries in the computed tomography image, grouping boundaries according to direction differences and consistency of response characteristics, and generating a tumor boundary response partition set; Based on the tumor boundary response partition set, the texture and metabolic characteristics of the tumor nodes are extracted in the atlas, the feature expressions are pattern-classified and the regional characteristics are annotated to generate a heterogeneous node distribution map of the tumor tissue atlas.

2. The medical image processing method for tumor partition identification according to claim 1, characterized in that: The grayscale consistency structural image set includes a grayscale frequency distribution feature set and an edge change trend feature set; the tumor area boundary initial contour map includes a density mutation point set, a texture interruption area set, and a metabolic enhancement position set; the tumor boundary response partition set includes a metabolic activity boundary group, a structural density boundary group, and a direction difference response group; the tumor tissue atlas heterogeneous node distribution map includes a texture feature pattern set, a metabolic feature pattern set, and a regional characteristic annotation set.

3. The medical image processing method for tumor partition identification according to claim 1, characterized in that: The specific steps of S1 are: S101: After acquiring a computed tomography image, a tissue structure image from a magnetic resonance imaging (MRI), a fluid-sensitive image, and a positron emission tomography (PET) image, grayscale value information of layered regions in the image is retrieved, and grayscale frequency features are extracted based on the spatial distribution of pixels and grayscale gradient changes to generate grayscale distribution frequency value intervals; S102: extracting grayscale change trends at the edges of the structural regions according to the grayscale distribution frequency value intervals, classifying and combining change trend information based on grayscale differences and spatial position relationships between adjacent pixels, and generating grayscale change edge gradient segments; S103: Based on the grayscale change edge gradient segment, extract a pixel set with similar grayscale change characteristics, make a judgment based on the grayscale difference and the boundary gradient, screen the pixel set that meets the smoothing condition, complete the grayscale mapping process, and generate a grayscale consistency structure image set.

4. The medical image processing method for tumor partition identification according to claim 3, characterized in that: The specific steps of S2 are: S201: Obtain pixel intensity distribution and neighborhood gradient information in the grayscale consistency structure image set, combine the edge area positioning result, perform difference judgment on the grayscale change and gradient direction difference within the edge, identify the grayscale mutation segment and determine whether it is a boundary abnormal area, and generate the edge grayscale mutation distribution value; S202: extracting the regional texture distribution direction and boundary spacing based on the edge grayscale mutation distribution value, identifying the texture boundary interruption area in combination with the grayscale roughness value, and performing boundary structure fitting on the interruption position to generate the texture boundary distribution fracture value; S203: Calling the texture boundary distribution fracture value, collecting the metabolic signal enhancement value and spatial signal density of the corresponding position, determining the correspondence between the spatial distribution of the metabolic enhancement area and the texture fracture area, establishing a spatial mapping relationship of the tumor edge contour in the image, and obtaining an initial contour map of the tumor area boundary.

5. The medical image processing method for tumor partition identification according to claim 4, characterized in that: The specific steps of S3 are: S301: Retrieving edge information extracted from the initial contour map of the tumor region boundary and locating it in the positron emission image, calculating a metabolic activity feature value based on the relationship between the grayscale value of the corresponding pixel and the metabolic activity threshold, selecting a location set of metabolic activity display areas, and generating a metabolic activity edge coordinate set; S302: extracting structural density information of corresponding coordinates in the computed tomography image according to the metabolic activity edge coordinate set, screening boundary points according to a density threshold, and integrating the screening results based on consistency of spatial positions to generate edge groups corresponding to structural density; S303: Based on the direction characteristics and edge response consistency of the pixel points in the edge group corresponding to the structural density, a set of pixel points meeting the difference characteristics is screened, and boundary classification and grouping are performed to generate a tumor boundary response partition set.

6. The medical image processing method for tumor partition identification according to claim 5, characterized in that: The specific calculation formula for calculating the metabolic activity characteristic value is: ； in, Represents the coordinates in the positron emission image The pixel gray value, represents the metabolic activity threshold, The initial contour of the tumor boundary is represented in The spatial gradient amplitude at Represents the weight adjustment coefficient of spatial gradient and grayscale difference, stands for metabolic threshold morphological correction factor.

7. The medical image processing method for tumor partition identification according to claim 5, characterized in that: The specific steps of S4 are: S401: Based on the tumor boundary response partition set, extract the image grayscale and metabolic activity information of each partition, construct corresponding texture and metabolic feature parameters, integrate the two types of parameters in the same spatial dimension, calculate the fusion feature value, and generate the partition joint feature value; S402: calling the partition joint feature value, demarcating the interval range of texture and metabolic features, performing feature classification operations on the difference partitions, constructing feature cluster labels, and generating texture metabolic cluster numerical features; S403: Mapping the clustering results to the original atlas according to the texture metabolism clustering numerical features, assigning cluster numbers to the node identifiers and assigning image identification areas to generate a heterogeneous node atlas distribution map.

8. The medical image processing method for tumor partition identification according to claim 7, characterized in that: The specific calculation formula for calculating the fusion eigenvalue is: ； in, represents the normalized value of the texture feature of the i-th partition, represents the normalized value of metabolic characteristics of the i-th partition, is the cube root stability coefficient, Represents the amplitude of the high-frequency component of the wavelet transform in the k-th dimension of the texture feature, Represents the k-th dimension of metabolic characteristics in the frequency domain energy density value, Represents the peak value of the texture feature space autocorrelation function, represents the integral value of the metabolic feature space power spectrum, is the product smoothing factor.

9. The medical image processing method for tumor partition identification according to claim 1, characterized in that: The method further comprises: S5: calling the heterogeneous node distribution map of the tumor tissue atlas, dividing it into core area, infiltration area and necrotic area according to the difference in node characteristics, completing image annotation in the form of spatial structure, and generating a tumor functional zonation structure map; The tumor functional zonation structure diagram includes core area space marking, infiltration area space marking, and necrotic area space marking.

10. The medical image processing method for tumor partition identification according to claim 9, characterized in that: The specific steps of S5 are: S501: Based on the heterogeneous node distribution map of the tumor tissue atlas, the morphological index, spatial coordinates and number of neighbors of the node are obtained, and a joint judgment is made according to the spatial aggregation characteristics and the morphological change trend to generate an initial partition value of the partition structure; S502: calling a node set of a difference region in the initial partition value of the partition structure, extracting its morphological index content, comparing the morphological changes of adjacent nodes based on the difference between regions, and generating a morphological boundary offset strength coefficient; S503: Selecting regional nodes with prominent changes according to the offset of the boundary nodes in the morphological boundary offset intensity coefficient, integrating spatial position and regional information, and generating a tumor functional zoning structure diagram.

Images