A method for differentiating postoperative recurrence and radiation injury images of brain glioma

Abstract

The application discloses a brain glioma postoperative recurrence and radiation injury image identification method, and belongs to the technical field of medical image analysis, which comprises the following steps: multi-modal image data standardization; microvessel metabolism feature reconstruction; radiation injury structure analysis; and joint constraint image identification. Local perfusion response features, diffusion restriction features and reinforced coupling features are constructed through cross-modal voxel mapping and neighborhood statistical analysis to form microvessel metabolism representation data; meanwhile, radiation injury structure representation data are constructed through hierarchical structure region division, necrosis cavity distribution modeling, boundary continuity analysis and peripheral tissue disturbance modeling, etc.; the application can depict the microvessel metabolism state and the structure injury state of the lesion region at the same time, and improves the accuracy and stability of brain glioma postoperative recurrence and radiation injury image identification.

Description

Technical Field

[0001] This invention relates to the field of medical image analysis technology, specifically to a method for identifying postoperative recurrence and radiation damage in glioma images. Background Technology

[0002] The image differentiation method for postoperative recurrence and radiation damage of glioma refers to a medical image analysis method that systematically analyzes multimodal medical imaging data obtained during the postoperative follow-up of glioma patients, determines the pathological nature of abnormal imaging manifestations in the surgical area, and thus distinguishes whether the lesion is a tumor recurrence or tissue damage caused by radiotherapy.

[0003] The main function of this type of method is to provide clinicians with auxiliary judgment by comprehensively analyzing information such as changes in imaging signals, tissue structure, and functional metabolism in the lesion area, thereby improving the accuracy of postoperative follow-up assessment. In practical applications, accurately distinguishing between tumor recurrence and radiation damage can help doctors develop more reasonable subsequent treatment strategies, such as adjusting radiotherapy plans, performing reoperation, or implementing targeted drug therapy, thereby improving patient treatment outcomes and reducing unnecessary medical interventions.

[0004] Currently, most clinical imaging analysis methods rely on single-modality imaging features or simple statistical indicators for judgment, such as using enhanced MRI signal changes, diffusion-weighted signals, or perfusion indicators for auxiliary analysis. However, since different imaging modalities reflect different physiological mechanisms, such as blood perfusion, tissue diffusion, and structural morphological changes, without a unified feature integration mechanism, it is often difficult to comprehensively depict the true pathological state of the lesion area. Furthermore, existing methods typically do not pay sufficient attention to the structural damage characteristics of the lesion, making it difficult to systematically describe the hierarchical relationship between necrotic structures, boundary disruption, and surrounding tissue disturbance, thus easily leading to misdiagnosis in complex cases.

[0005] Therefore, how to integrate the microvascular metabolic activity characteristics and structural damage characteristics of the lesion area based on multimodal medical imaging data, and construct a stable and reliable joint discrimination mechanism to accurately distinguish between postoperative recurrence of glioma and radiation damage has become a technical problem that urgently needs to be solved in the field of medical imaging-assisted diagnosis. Summary of the Invention

[0006] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a method for identifying postoperative recurrence and radiation damage in gliomas using images. The technical solution adopted by this invention is as follows: This invention provides a method for identifying postoperative recurrence and radiation damage in gliomas using images, which includes the following steps:

[0007] Step S1: Standardize multimodal image data;

[0008] Step S2: Reconstruction of microvascular metabolic characteristics;

[0009] Step S3: Radiation damage structure analysis;

[0010] Step S4: Joint constraint image identification.

[0011] Furthermore, in step S1, the multimodal image data is standardized to standardize the multimodal medical images of patients after glioma surgery. Specifically, multimodal medical image data during the postoperative follow-up stage is acquired, and spatial registration, voxel scale unification, grayscale intensity normalization, and intracranial structural region limitation are performed on the multimodal medical image data. Based on the anatomical structure of the surgical area, the candidate lesion region is located and cropped to obtain standardized multimodal image data.

[0012] Further, in step S2, the microvascular metabolic feature reconstruction is used to reconstruct composite image features reflecting the microvascular perfusion state, tissue diffusion state, and local metabolic activity state of the lesion area based on standardized image representation. Specifically, based on the standardized multimodal image data, a cross-modal microvascular metabolic reconstruction improvement method is used to perform cross-modal voxel mapping and feature association analysis on the lesion area and the lesion edge area. By constructing the relationship between local perfusion response distribution, diffusion abnormality distribution, and enhancement response change, joint feature modeling is performed on the functional information in different modal images. Based on the local neighborhood statistical results, microvascular perfusion activity characteristics, diffusion restriction characteristics, and metabolic heterogeneity distribution characteristics are formed to obtain microvascular metabolic characterization data.

[0013] The improved method for cross-modal microvascular metabolic remodeling includes the following steps:

[0014] Step S21: Construction of cross-modal voxel correspondence in lesion region, used to establish voxel-level correspondence between different modal images of lesion region under a unified spatial coordinate system. Specifically, based on the standardized multimodal image data, the signal response of each voxel position in the candidate lesion region in different modal images is indexed and mapped, and the lesion is divided into internal lesion region and edge lesion region according to the spatial structure of the lesion region, thereby forming a voxel-level correspondence between different modal images and obtaining cross-modal voxel mapping data;

[0015] Step S22: Local microvascular perfusion response modeling, used to characterize the activity level of microvascular blood flow perfusion and its spatial distribution differences within the lesion area. Specifically, based on the cross-modal voxel mapping data, statistical analysis is performed on the perfusion modal response values ​​of each voxel neighborhood within the lesion area. By calculating the local perfusion intensity and perfusion fluctuation, a local perfusion response index is constructed to obtain local perfusion response characteristic data that characterizes the activity level of microvascular perfusion in the lesion area.

[0016] Step S23: Modeling of diffusion-restricted state, used to characterize the abnormal state of water molecule diffusion caused by changes in tissue cell density in the lesion area. Specifically, based on the cross-modal voxel mapping data, statistical analysis is performed on the diffusion modal response values ​​of each voxel neighborhood in the lesion area. A diffusion-restricted index is constructed by calculating the local diffusion intensity and the degree of diffusion change, and diffusion-restricted characteristic data characterizing the degree of diffusion restriction in the lesion tissue is obtained.

[0017] Step S24: Feature deviation analysis, used to construct the synergistic relationship between enhanced response features, microvascular perfusion features and diffusion-restricted features, and to identify the degree of deviation of each voxel from the synergistic relationship. Specifically, it constructs a cross-modal functional feature model based on the enhanced modal response value, the local perfusion response feature data and the diffusion-restricted feature data, and calculates the feature deviation index according to the synergistic change relationship between different modal features to obtain enhanced feature data that characterizes the joint change relationship between enhanced response and perfusion and diffusion features.

[0018] Step S25: Reconstruction of metabolic heterogeneity between the lesion interior and exterior regions, used to characterize the differences between the lesion interior region and the lesion periphery region in terms of microvascular perfusion, diffusion status and enhancement characteristics. Specifically, regional statistical analysis is performed on the perfusion response characteristics, diffusion restriction characteristics and enhancement characteristics of the lesion interior region and the lesion periphery region, and metabolic heterogeneity index is constructed based on the degree of difference between the two types of regions to obtain microvascular metabolic heterogeneity index data reflecting the differences in metabolic status of different regions within the lesion.

[0019] Step S26: Generation of microvascular metabolic characterization data, used to comprehensively organize the perfusion response characteristics, diffusion restriction characteristics, enhancement characteristics, and metabolic heterogeneity indicators of the lesion area. Specifically, the local perfusion response characteristic data, diffusion restriction characteristic data, enhancement characteristic data, and microvascular metabolic heterogeneity indicator data are summarized and constructed to obtain microvascular metabolic characterization data used to characterize the microvascular perfusion state, diffusion restriction state, and metabolic activity state of the lesion area.

[0020] The microvascular metabolic characterization data specifically includes: local perfusion response characteristic data, diffusion restriction characteristic data, enhancement characteristic data, and microvascular metabolic heterogeneity index data; the local perfusion response characteristic data is used to characterize the microvascular blood flow perfusion level and its spatial distribution differences within the lesion area; the diffusion restriction characteristic data is used to characterize the abnormal diffusion state caused by changes in lesion tissue cell density; the enhancement characteristic data is used to characterize the combined changes in the enhancement response and perfusion and diffusion characteristics; and the microvascular metabolic heterogeneity index data is used to characterize the degree of difference in blood supply and metabolic activity in different regions within the lesion.

[0021] Further, in step S3, the radiation damage structure analysis is used to identify the structural damage features related to radiation damage in the postoperative lesion area and its surrounding brain tissue. Specifically, based on the standardized multimodal imaging data, a joint analysis is performed on the internal structural morphology of the lesion area, the continuity of the lesion boundary, and the disturbance of the surrounding brain tissue structure. An improved method of hierarchical damage structure topology analysis is adopted. By identifying the distribution pattern of necrotic cavities, the integrity of the edge contour, and the edema diffusion pattern, a damage structure feature expression reflecting the degree of tissue structure damage and the hierarchical relationship is constructed. The structural damage of the lesion area is described according to the degree of structural abnormality to obtain radiation damage structure characterization data.

[0022] The improved method for topology analysis of hierarchical damage structures includes the following steps:

[0023] Step S31: Structural hierarchical region division, used to establish the basis for structural hierarchical analysis within the lesion region and its surrounding brain tissue. Specifically, based on the standardized multimodal image data, spatial morphological analysis is performed on the candidate lesion region, and the lesion is divided into the internal lesion region, the edge region of the lesion region, and the surrounding tissue region according to the spatial structural relationship of the lesion region. The structural hierarchical region mapping relationship of the lesion is constructed to obtain the structural hierarchical region mapping data.

[0024] Step S32: Necrotic cavity distribution modeling, used to identify the spatial distribution characteristics of low-activity necrotic tissue in the lesion area. Specifically, based on the structural hierarchical region mapping data, the image signal intensity of voxels in the lesion area and its neighborhood changes are analyzed. By identifying the local signal abnormal reduction area and its connected structure, the spatial location and morphological distribution of necrotic cavities are modeled and described to obtain necrotic cavity distribution characteristic data.

[0025] Step S33: Boundary continuity analysis, used to characterize the integrity of the lesion boundary structure and the local interruption, specifically, based on the structural hierarchical region mapping data, statistical analysis is performed on the image gradient changes, edge morphology continuity and local structural changes of the lesion edge region, and boundary continuity index is constructed to quantitatively express the integrity of the lesion boundary contour, thereby obtaining boundary continuity feature data;

[0026] Step S34: Peripheral tissue disturbance modeling, used to characterize the structural disturbance and edema spread of brain tissue around the lesion under radiation damage. Specifically, based on the structural hierarchical region mapping data, statistical analysis is performed on the changes in image signals and spatial distribution characteristics in the peripheral tissue region. By identifying the abnormal signal diffusion area and its disturbance degree, the damage to the peripheral brain tissue structure is modeled and described to obtain the peripheral tissue disturbance characteristic data.

[0027] Step S35: Structural topology propagation modeling, used to characterize the structural damage propagation relationship between the internal region of the lesion, the edge region of the lesion, and the surrounding tissue region. Specifically, based on the distribution feature data of the necrotic cavity, the boundary continuity feature data, and the disturbance feature data of the surrounding tissue, a topological association model between the structural levels of the lesion is constructed. By analyzing the spatial transmission relationship between structural abnormalities in different level regions, the structural damage propagation intensity is modeled and expressed to obtain structural level association feature data.

[0028] Step S36: Generation of radiation damage structural characterization data, used to comprehensively organize and express the structural damage characteristics of the lesion area. Specifically, it involves summarizing and constructing the distribution characteristic data of the necrotic cavity, the boundary continuity characteristic data, the surrounding tissue disturbance characteristic data, and the structural hierarchy association characteristic data to obtain radiation damage structural characterization data used to characterize the degree of structural damage and structural hierarchy relationship of the lesion area.

[0029] The radiation damage structural characterization data specifically includes: necrotic cavity distribution characteristics data, boundary continuity characteristics data, surrounding tissue disturbance characteristics data, and structural hierarchy correlation characteristics data;

[0030] The necrotic cavity distribution feature data is used to characterize the spatial distribution of low-activity necrotic areas within the lesion; the boundary continuity feature data is used to characterize the integrity of the lesion boundary contour and local interruption; the surrounding tissue disturbance feature data is used to characterize the edema and infiltration state and structural damage degree of the brain tissue surrounding the lesion; and the structural hierarchy association feature data is used to characterize the morphological transmission relationship and topological association relationship between the internal area, the edge area, and the surrounding tissue area of ​​the lesion.

[0031] Further, in step S4, the joint constraint image identification is used to identify postoperative lesions as recurrence or radiation damage. Specifically, it maps the microvascular metabolic characterization data and the radiation damage structural characterization data to a unified discrimination space, performs correlation analysis on the functional activity state and structural damage state of the lesion area, constructs functional activity consistency constraints and structural damage consistency constraints, jointly screens candidate discrimination results, and suppresses discrimination results that do not meet the cross-mechanism consistency conditions, outputs the discrimination result that the lesion belongs to postoperative recurrence or radiation damage of glioma, and obtains joint identification result data.

[0032] The beneficial effects achieved by the present invention using the above solution are as follows:

[0033] (1) In view of the existing methods for differentiating postoperative recurrence of glioma from radiation injury, which rely solely on a single image modality or simple image features for discrimination, resulting in unstable spatial correspondence and inconsistent grayscale between different modal images, thus affecting the accuracy of subsequent feature analysis, this solution creatively adopts a multimodal image data standardization processing mechanism. Through spatial registration, voxel scale unification, grayscale intensity normalization, and intracranial structural region limitation, different modal medical images are uniformly expressed. Combined with the surgical area structural information, the candidate lesion region is located and locally cropped, thereby forming standardized multimodal image data, thus providing a reliable data foundation for subsequent cross-modal feature reconstruction.

[0034] (2) In view of the technical problem that existing glioma postoperative recurrence discrimination technology has difficulty in simultaneously depicting the synergistic changes between the microvascular perfusion state, diffusion abnormality state and enhancement response in the lesion area, resulting in insufficient ability to identify the microvascular metabolic activity characteristics corresponding to tumor recurrence, this solution creatively adopts a cross-modal microvascular metabolic feature reconstruction mechanism. By constructing a cross-modal voxel mapping relationship and jointly modeling the perfusion modal response, diffusion modal response and enhancement modal response, local perfusion response characteristics, diffusion restriction characteristics and enhancement characteristics are formed. Furthermore, metabolic heterogeneity indicators are constructed through the difference analysis of the inside and outside regions of the lesion, so as to more accurately depict the microvascular metabolic activity characteristics of the tumor recurrence area.

[0035] (3) In view of the technical problem that existing radiation injury image recognition methods are difficult to systematically depict the hierarchical relationship between the necrotic structure inside the lesion, the degree of destruction of the lesion boundary and the disturbance of the surrounding brain tissue, resulting in incomplete expression of radiation injury structural features, this solution creatively adopts a hierarchical damage structure topology analysis mechanism. By establishing the structural hierarchical regional mapping relationship of the lesion internal area, edge area and surrounding tissue area, and jointly modeling the distribution of necrotic cavity, boundary continuity and surrounding tissue disturbance, the spatial transmission relationship of structural damage between different hierarchical regions is further depicted through structural topology propagation analysis, thereby forming radiation injury structural characterization data. Through structural hierarchical topology modeling, the damage mode of this type of structure can be completely described. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating a method for identifying postoperative recurrence and radiation damage in gliomas according to the present invention.

[0037] Figure 2 This is a schematic diagram of the process for reconstructing microvascular metabolic features in step S2.

[0038] Figure 3 This is a schematic diagram of the process for analyzing the radiation damage structure in step S3.

[0039] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0040] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0041] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0042] Example 1, see Figure 1 The present invention provides a method for identifying postoperative recurrence and radiation damage in gliomas, the method comprising the following steps:

[0043] Step S1: Standardize multimodal image data;

[0044] Step S2: Reconstruction of microvascular metabolic characteristics;

[0045] Step S3: Radiation damage structure analysis;

[0046] Step S4: Joint constraint image identification.

[0047] By performing the above operations, this solution addresses the technical problem in existing methods for differentiating postoperative recurrence of glioma from radiation injury, which relies solely on a single image modality or simple image features, leading to unstable spatial correspondences and inconsistent grayscale scales between different modalities, thus affecting the accuracy of subsequent feature analysis. This solution creatively employs a multimodal image data standardization mechanism. Through spatial registration, voxel scale unification, grayscale intensity normalization, and intracranial structural region limitation, it unifies the representation of different modalities of medical images. Furthermore, it combines surgical area structural information to locate and locally crop candidate lesion regions, thereby forming standardized multimodal image data. For example, in actual clinical follow-up images, enhanced T1 and T2-FLAIR images acquired at different times or with different devices often exhibit resolution differences and spatial offsets. This solution, by unifying the spatial coordinate system and voxel scale, establishes a stable correspondence between voxel positions of the same lesion region in various modalities, thus providing a reliable data foundation for subsequent cross-modal feature reconstruction.

[0048] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the multimodal image data is standardized to standardize the multimodal medical images of patients after glioma surgery. Specifically, multimodal medical image data during the postoperative follow-up stage is obtained, and spatial registration, voxel scale unification, gray intensity normalization, and intracranial structural region limitation are performed on the multimodal medical image data. The candidate lesion region is located and cropped according to the anatomical structure of the surgical area to obtain standardized multimodal image data.

[0049] This embodiment describes the specific implementation of the multimodal image data standardization described in step S1. In this embodiment, follow-up image data of patients after glioma surgery is used as input. By performing unified and standardized processing on multimodal medical images, a standardized image representation that can be used for subsequent feature reconstruction and structural analysis is obtained.

[0050] In this embodiment, the multimodal medical imaging data originates from the magnetic resonance imaging examination results during the patient's postoperative follow-up phase, and may specifically include one or more of enhanced T1-weighted magnetic resonance images, T2-FLAIR images, diffusion-weighted images, and perfusion-weighted images; the above imaging data is acquired through the hospital's image storage and transmission system and read in a digital imaging and communication standard format; to ensure the consistency of subsequent analysis, it is necessary to first perform unified preprocessing on the images of different modalities;

[0051] First, spatial registration processing is performed on the multimodal medical image data. Specifically, using the enhanced T1-weighted image as the reference modality, other modal images are mapped to the spatial coordinate system of the reference modality through rigid registration or affine registration. In some embodiments, the registration process can employ a registration strategy based on maximizing mutual information, or a similarity measurement method based on normalized cross-correlation, to iteratively optimize and solve for spatial transformation parameters. Rigid registration is used to eliminate translation and rotation errors, while affine registration is further used to compensate for scale changes and shear deformation. Preferably, the number of iterations in the registration process can be set to 50–200, and the convergence threshold is set to 10. -4 ~10 -6 Through the above processing, the images of different modalities are made to form a consistent structural correspondence in spatial position; after spatial registration processing, the anatomical structures in each modality image can maintain a consistent position under a unified coordinate system, thereby forming the modal spatially aligned image data.

[0052] Subsequently, voxel scale unification processing is performed on the spatially registered images. Specifically, based on a preset voxel size standard, each modal image is resampled to ensure that different modal images maintain consistent voxel resolution. In some embodiments, the resampling process may employ trilinear interpolation or B-spline interpolation-based resampling. Preferably, the unified voxel size can be set between 0.8mm×0.8mm×0.8mm and 1.5mm×1.5mm×1.5mm. Through voxel scale unification processing, the subsequently extracted image features can be compared in spatial scale, and scale deviations caused by different scanning devices and scanning protocols can be reduced.

[0053] After voxel-scale uniform processing, grayscale intensity normalization is performed on the images. Specifically, the grayscale value range of each modality image is uniformly mapped so that different modalities of images express tissue signal changes at a uniform intensity scale. In some embodiments, the grayscale normalization can adopt a linear normalization method based on the global intensity range or a standardization method based on the statistical distribution of brain tissue regions. Preferably, the grayscale values ​​of each modality image can be mapped to the 0-1 interval, or converted into a standard distribution with a mean of 0 and a standard deviation of 1. More preferably, abnormally high or low intensity values ​​can be truncated, with the truncation range set to the quantile intervals of the first 1% to 5% and the last 1% to 5% to reduce the influence of extreme values ​​on the grayscale distribution. The grayscale normalized image data is formed after grayscale normalization.

[0054] Furthermore, to avoid interference from non-brain tissues such as the skull and scalp in subsequent analysis, intracranial structural region limitation processing is performed on the normalized images. Specifically, the skull and scalp regions are removed using brain tissue segmentation methods, retaining only the intracranial brain tissue regions. In some embodiments, the brain tissue segmentation can be performed using a method combining threshold segmentation and morphological processing, or automatically segmented using a pre-trained brain region segmentation model (e.g., a medical image segmentation model based on the U-Net structure). Preferably, by performing connected component filtering and boundary smoothing on the segmentation results, the largest connected brain tissue regions are retained, thereby obtaining image data limited to the intracranial anatomical structure range.

[0055] After completing the above processing, the candidate lesion areas are located based on the anatomical structure information of the surgical area. Specifically, the suspected lesion areas are automatically or semi-automatically located based on the spatial distribution of areas with abnormal enhancement, edema, and abnormal signals. In some embodiments, abnormal areas can be initially screened by setting a threshold range for enhanced modality intensity (preferably 1.2 to 2.5 times the mean of normal brain tissue) and a threshold range for high FLAIR signal (preferably 1.5 to 3.0 times the mean of background), and the candidate areas are aggregated by combining spatial connectivity constraints; thereby forming the lesion candidate area location data.

[0056] After obtaining the lesion candidate region location data, local region cropping is performed with the candidate region as the center. Specifically, based on a preset region size, a local image block containing the lesion region and its neighboring tissues is extracted from the standardized image. In some embodiments, the size of the local image block can be set between 64×64×64 voxels and 192×192×192 voxels, preferably 96×96×96 or 128×128×128 voxels. A certain range (preferably 5 to 20 voxels) can be extended outside the candidate region boundary to retain surrounding tissue information. The local lesion image block data is obtained after cropping.

[0057] Through the above processing flow, this embodiment finally obtains the standardized multimodal image data. The standardized multimodal image data specifically includes modal spatially aligned image data, grayscale normalized image data, lesion candidate region localization data, and local lesion image block data. Specifically, the modal spatially aligned image data is used to characterize the spatial correspondence of different modal images in a unified coordinate system; the grayscale normalized image data is used to characterize the tissue response characteristics of each modality image at a unified intensity scale; the lesion candidate region localization data is used to characterize the spatial location range of postoperative abnormal enhancement areas, edema areas, or necrosis areas; and the local lesion image block data is used as direct input for subsequent microvascular metabolic feature reconstruction and radiation damage structure analysis.

[0058] Example 3, see Figure 1 , Figure 2 This embodiment is based on the above embodiment. In step S2, the microvascular metabolic feature reconstruction is used to reconstruct composite image features reflecting the microvascular perfusion state, tissue diffusion state, and local metabolic activity state of the lesion area on the basis of standardized image expression. Specifically, based on the standardized multimodal image data, a cross-modal microvascular metabolic reconstruction improvement method is used to perform cross-modal voxel mapping and feature association analysis on the lesion area and the lesion edge area. By constructing the relationship between local perfusion response distribution, diffusion abnormality distribution, and enhancement response change, joint feature modeling is performed on the functional information in different modal images. Based on the local neighborhood statistical results, microvascular perfusion activity characteristics, diffusion restriction characteristics, and metabolic heterogeneity distribution characteristics are formed to obtain microvascular metabolic characterization data.

[0059] This embodiment describes the specific implementation of the microvascular metabolic feature reconstruction described in step S2. In this embodiment, the standardized multimodal image data obtained in step S1 is used as input. Under a unified spatial coordinate system, the enhanced modal response, perfusion modal response, and diffusion modal response of the lesion candidate region are jointly characterized at the voxel level. The enhanced modality is preferably an enhanced T1-weighted image, the perfusion modality is preferably a perfusion-weighted image or a relative cerebral blood volume parameter map calculated from it, and the diffusion modality is preferably a diffusion-weighted image or an apparent diffusion coefficient map. To ensure dimensional consistency of different modal features during the joint analysis process, it is preferable to first perform regional standardization on the response values ​​of each modality to convert them into comparable dimensionless feature values. In some implementations, z-score standardization or robust normalization based on the 5%–95% quantile interval within the lesion candidate region can be used.

[0060] The improved method for cross-modal microvascular metabolic remodeling includes the following steps:

[0061] Step S21: Construction of cross-modal voxel correspondence in lesion region, used to establish voxel-level correspondence between different modal images of lesion region under a unified spatial coordinate system. Specifically, based on the standardized multimodal image data, the signal response of each voxel position in the candidate lesion region in different modal images is indexed and mapped, and the lesion is divided into internal lesion region and edge lesion region according to the spatial structure of the lesion region, thereby forming a voxel-level correspondence between different modal images and obtaining cross-modal voxel mapping data;

[0062] In this embodiment, the candidate lesion region can be represented as a set of regions, and any voxel is denoted as v. i Where i is the voxel index number; for any voxel v i The response values ​​in the enhancement mode, injection mode, and diffusion mode are denoted as E(v). i ), P(v i ) and D(v i This allows for the construction of voxel-level cross-modal response vectors, calculated using the following formula:

[0063] ;

[0064] In the formula, It is cross-modal voxel mapping data, E(v i P(v) is the response value under the augmented mode. i D(v) is the response value under the infusion mode. i ) is the response value under the diffusion mode;

[0065] Further preferably, to distinguish between the internal region and the edge region of the lesion, morphological erosion and boundary extraction operations can be performed based on the candidate region mask; in some embodiments, the radius of the eroded structural element is preferably 2 to 5 voxels, the region retained after erosion is defined as the internal region of the lesion, and the difference between the original candidate region and the internal region is defined as the edge region of the lesion; when the lesion volume is small, the width of the edge region can also preferably be set to 1 to 3 voxels to avoid statistical distortion caused by an excessively large edge region ratio; after the above processing, the cross-modal voxel mapping data is obtained, which specifically includes voxel coordinate index data, cross-modal response vector data, and region level identification data;

[0066] Step S22: Local microvascular perfusion response modeling, used to characterize the activity level of microvascular blood flow perfusion and its spatial distribution differences within the lesion area. Specifically, based on the cross-modal voxel mapping data, statistical analysis is performed on the perfusion modal response values ​​of each voxel neighborhood within the lesion area. By calculating the local perfusion intensity and perfusion fluctuation, a local perfusion response index is constructed to obtain local perfusion response characteristic data that characterizes the activity level of microvascular perfusion in the lesion area.

[0067] In this embodiment, voxel v i Construct a local neighborhood N around the center i The local neighborhood is preferably a 3×3×3 voxel, a 5×5×5 voxel, or a spherical neighborhood with a radius of 2mm to 5mm; the local perfusion intensity can be characterized by the mean perfusion response within the neighborhood, and the degree of local perfusion fluctuation can be characterized by the standard deviation or coefficient of variation of the perfusion response within the neighborhood.

[0068] Preferably, the local perfusion response index (LPI) is... i It can be defined as:

[0069] ;

[0070] In the formula, It is a local perfusion response index. This is the weight of the mean of the perfusion response, preferably between 0.55 and 0.80. It is the neighborhood N i Mean of perfusion response within, This is the weight of the standard deviation of the perfusion response, preferably ranging from 0.20 to 0.45. It is the neighborhood N i The standard deviation of the infusion response within;

[0071] Furthermore, to reduce local noise interference, the perfusion response map can be smoothed before neighborhood statistics; Gaussian smoothing is preferred, and the standard deviation of the smoothing kernel is preferably 0.6 to 1.5 voxels; after the above processing, the local perfusion response feature data is obtained, which may include a local perfusion mean map, a local perfusion fluctuation map, and a local perfusion response index map;

[0072] Step S23: Modeling of diffusion-restricted state, used to characterize the abnormal state of water molecule diffusion caused by changes in tissue cell density in the lesion area. Specifically, based on the cross-modal voxel mapping data, statistical analysis is performed on the diffusion modal response values ​​of each voxel neighborhood in the lesion area. A diffusion-restricted index is constructed by calculating the local diffusion intensity and the degree of diffusion change, and diffusion-restricted characteristic data characterizing the degree of diffusion restriction in the lesion tissue is obtained.

[0073] In this embodiment, the diffusion-limited state is preferably characterized by a combination of low diffusion value features and spatial aggregation features; when the diffusion mode uses the apparent diffusion coefficient diagram, a lower diffusion response usually corresponds to a more obvious diffusion-limited state; in order to unify the direction of the index, it is preferable to first perform a reverse mapping on the diffusion response so that a larger value indicates a higher degree of diffusion limitation.

[0074] Preferably, the diffusion restriction index DRI(v) i It can be defined as:

[0075] ;

[0076] In the formula, It is a diffusion-limited index. This is the diffusion response weight, selected from 0.60 to 0.85. It is the normalized diffusion response value. The weight for local changes is preferably set between 0.15 and 0.40. It is the neighborhood N i The degree of change in the diffusion response within;

[0077] In some implementations, threshold conditions can be set for diffusion-restricted areas. For example, voxels with diffusion restriction indices higher than 1.10 to 1.30 times the average value of lesion areas can be initially marked as highly suspected diffusion-restricted voxels, and discrete noise areas with volumes smaller than 10 to 30 voxels can be screened out in combination with connected domains.

[0078] Step S24: Feature deviation analysis, used to construct the synergistic relationship between enhanced response features, microvascular perfusion features and diffusion-restricted features, and to identify the degree of deviation of each voxel from the synergistic relationship. Specifically, it constructs a cross-modal functional feature model based on the enhanced modal response value, the local perfusion response feature data and the diffusion-restricted feature data, and calculates the feature deviation index according to the synergistic change relationship between different modal features to obtain enhanced feature data that characterizes the joint change relationship between enhanced response and perfusion and diffusion features.

[0079] In this embodiment, to characterize the synergistic relationship between enhanced response and perfusion / diffusion limitation, a cross-modal functional synergistic prediction model can be constructed. In some embodiments, the cross-modal functional synergistic prediction model can employ a linear regression model, a ridge regression model, or a lightweight multilayer perceptron model. When emphasizing the engineering reproducibility in the pre-review text, a linear combination model is preferred, which can be expressed as follows:

[0080] ;

[0081] In the formula, It is a theoretically enhanced response predicted by the infusion response and the diffusion-confined state. It is a response bias term. The infusion weight is preferably 0.35 to 0.70. The diffusion weight is preferably 0.20 to 0.55.

[0082] Based on this, the characteristic deviation index FDI(v) can be further calculated. i ), defined as:

[0083] ;

[0084] In the formula, It is the characteristic deviation index;

[0085] Furthermore, it can be determined according to FDI(v) i The size of FDI (v) identifies co-enhanced voxels and deviated voxels; for example, when FDI (v) i When the FDI (v) is less than 0.8 to 1.0 times the mean deviation index of the lesion area, it can be determined that the current voxel satisfies a strong synergistic relationship; when the FDI (v) is less than 0.8 to 1.0 times the mean deviation index of the lesion area, it can be determined that the current voxel satisfies a strong synergistic relationship; i When the deviation index of the lesion area is 1.2 to 1.8 times higher than the mean, the current voxel can be identified as a voxel with significant characteristic deviation.

[0086] Through the above processing, the enhanced feature data can be obtained, which may include a theoretical enhanced response map, a feature deviation index map, and a collaborative enhanced identifier map;

[0087] Step S25: Reconstruction of metabolic heterogeneity between the lesion interior and exterior regions, used to characterize the differences between the lesion interior region and the lesion periphery region in terms of microvascular perfusion, diffusion status and enhancement characteristics. Specifically, regional statistical analysis is performed on the perfusion response characteristics, diffusion restriction characteristics and enhancement characteristics of the lesion interior region and the lesion periphery region, and metabolic heterogeneity index is constructed based on the degree of difference between the two types of regions to obtain microvascular metabolic heterogeneity index data reflecting the differences in metabolic status of different regions within the lesion.

[0088] In this embodiment, the internal region of the lesion is denoted as... The edge area of ​​the lesion is The mean, standard deviation, and proportion of high-value voxels of the local perfusion response index, diffusion restriction index, and characteristic deviation index in the two types of regions were calculated respectively.

[0089] Preferably, the metabolic heterogeneity index MHI can be constructed as follows:

[0090] ;

[0091] In the formula, It is the weight of the mean of the local perfusion response index, preferably between 0.30 and 0.45. and These are the mean local perfusion response indices for the internal and peripheral regions of the lesion, respectively. It is the mean weight of the diffusion-limiting index, preferably ranging from 0.25 to 0.40. and These are the mean diffusion restriction indices for the internal and peripheral regions of the lesion, respectively. This is the weight of the feature deviation from the mean, preferably between 0.20 and 0.35. and These are the mean characteristic deviation indices of the internal region and the peripheral region of the lesion, respectively.

[0092] In some implementations, a high-value voxel ratio difference item can be added as a supplementary indicator. For example, the proportion of voxels in the two types of regions that exceed the regional mean plus 0.5 to 1.0 times the standard deviation can be counted separately and used as a supplementary descriptive parameter for heterogeneity.

[0093] Step S26: Generation of microvascular metabolic characterization data, used to comprehensively organize the perfusion response characteristics, diffusion restriction characteristics, enhancement characteristics, and metabolic heterogeneity indicators of the lesion area. Specifically, the local perfusion response characteristic data, diffusion restriction characteristic data, enhancement characteristic data, and microvascular metabolic heterogeneity indicator data are summarized and constructed to obtain microvascular metabolic characterization data used to characterize the microvascular perfusion state, diffusion restriction state, and metabolic activity state of the lesion area.

[0094] In this embodiment, the local perfusion response index, diffusion restriction index, characteristic deviation index, and metabolic heterogeneity index can be organized into a comprehensive characterization vector in a unified format; preferably, for any voxel v i Its voxel-level comprehensive representation vector can be defined as:

[0095] ;

[0096] In the formula, It is a voxel-level comprehensive representation vector. It is a region-level identifier variable used to mark whether the current voxel belongs to the internal region of the lesion or the edge region of the lesion; at the same time, it can generate region-level summary features for the entire region within the lesion range, including the mean LPI value, mean DRI value, mean FDI value, MHI value, and the corresponding proportion of high-value voxels.

[0097] In some embodiments, the microvascular metabolic characterization data can be stored in a multi-field structured data format, including: voxel coordinate field, region category field, perfusion response field, diffusion restriction field, enhancement deviation field, and heterogeneity statistics field, so as to be uniformly associated with the radiation damage structural characterization data output in step S3.

[0098] The microvascular metabolic characterization data specifically includes: local perfusion response characteristic data, diffusion restriction characteristic data, enhancement characteristic data, and microvascular metabolic heterogeneity index data; the local perfusion response characteristic data is used to characterize the microvascular blood flow perfusion level and its spatial distribution differences within the lesion area; the diffusion restriction characteristic data is used to characterize the abnormal diffusion state caused by changes in lesion tissue cell density; the enhancement characteristic data is used to characterize the combined changes in the enhancement response and perfusion and diffusion characteristics; and the microvascular metabolic heterogeneity index data is used to characterize the degree of difference in blood supply and metabolic activity in different regions within the lesion.

[0099] To facilitate the explanation of the engineering implementation of this step, the following is an example of parameter values: the local neighborhood N_i is preferably a 5×5×5 voxel; in the perfusion response index... , In the diffusion restriction index , In cross-modal functional co-prediction models , Among the indicators of metabolic heterogeneity , , ;

[0100] It should be understood that the above parameters are only preferred examples. Under different scanning protocols, different lesion volumes, and different image quality conditions, they can be adjusted within the aforementioned preferred range, but this does not affect the implementation logic of cross-modal microvascular metabolic feature reconstruction described in this embodiment.

[0101] By performing the above operations, this solution addresses the technical problem in existing glioma postoperative recurrence discrimination techniques: the inability to simultaneously characterize the synergistic changes among microvascular perfusion status, abnormal diffusion status, and enhancement response in the lesion area, resulting in insufficient ability to identify the microvascular metabolic activity characteristics corresponding to tumor recurrence. This solution creatively employs a cross-modal microvascular metabolic feature reconstruction mechanism. By constructing a cross-modal voxel mapping relationship and jointly modeling the perfusion modal response, diffusion modal response, and enhancement modal response, local perfusion response characteristics, diffusion restriction characteristics, and enhancement characteristics are formed. Furthermore, metabolic heterogeneity indicators are constructed through regional differences within and outside the lesion. For example, in the tumor recurrence area, there is often a synergistic change in perfusion enhancement, diffusion restriction, and enhancement response. This solution can identify such synergistic change patterns through feature deviation analysis, thereby more accurately characterizing the microvascular metabolic activity characteristics of the tumor recurrence area.

[0102] Example 4, see Figure 1 , Figure 3This embodiment is based on the above embodiment. In step S3, the radiation damage structure analysis is used to identify the structural destruction features related to radiation damage in the lesion area and its surrounding brain tissue after surgery. Specifically, based on the standardized multimodal image data, the internal structural morphology of the lesion area, the continuity of the lesion boundary, and the disturbance of the surrounding brain tissue structure are jointly analyzed. An improved method of hierarchical damage structure topology analysis is adopted. By identifying the distribution pattern of necrotic cavities, the integrity of the edge contour, and the edema diffusion pattern, a damage structure feature expression reflecting the degree of tissue structure destruction and the hierarchical relationship is constructed. The structural damage of the lesion area is described according to the degree of structural abnormality to obtain radiation damage structure characterization data.

[0103] The improved method for topology analysis of hierarchical damage structures includes the following steps:

[0104] Step S31: Structural hierarchical region division, used to establish the basis for structural hierarchical analysis within the lesion region and its surrounding brain tissue. Specifically, based on the standardized multimodal image data, spatial morphological analysis is performed on the candidate lesion region, and the lesion is divided into the internal lesion region, the edge region of the lesion region, and the surrounding tissue region according to the spatial structural relationship of the lesion region. The structural hierarchical region mapping relationship of the lesion is constructed to obtain the structural hierarchical region mapping data.

[0105] In this embodiment, the candidate lesion region can be denoted as... The areas within the lesion that are highly suspected of necrosis or low activity, i.e., the internal areas of the lesion, are denoted as... The transition area of ​​the lesion outline, i.e., the edge area of ​​the lesion, is denoted as The surrounding tissue area outside the lesion affected by structural disturbance is denoted as ;in, , and Together they constitute the structural hierarchy analysis region;

[0106] Preferably, morphological erosion, boundary extraction and expansion operations can be performed based on the lesion candidate area mask to generate corresponding hierarchical areas; in some embodiments, the erosion radius of the internal area is preferably 2 to 5 voxels, the width of the edge area is preferably 1 to 4 voxels, and the expansion width of the surrounding tissue area is preferably 3 to 10 voxels or 2 mm to 8 mm.

[0107] Furthermore, any voxel v i Assign a hierarchical identifier variable R(v) i ), where R(v i R(v) = 1 represents the internal region of the lesion. i R(v) = 2 represents the edge region of the lesion. i=3 indicates the surrounding tissue area; thus forming a hierarchical mapping relationship of the lesion structure; after the above processing, the hierarchical mapping data of the structure is obtained, which specifically includes the hierarchical mask data of the region, the voxel-level index data, and the hierarchical adjacency relationship data;

[0108] Step S32: Necrotic cavity distribution modeling, used to identify the spatial distribution characteristics of low-activity necrotic tissue in the lesion area. Specifically, based on the structural hierarchical region mapping data, the image signal intensity of voxels in the lesion area and its neighborhood changes are analyzed. By identifying the local signal abnormal reduction area and its connected structure, the spatial location and morphological distribution of necrotic cavities are modeled and described to obtain necrotic cavity distribution characteristic data.

[0109] In this embodiment, necrotic cavities can preferably be identified based on low-enhancement, low-response regions within the lesion in enhanced T1-weighted images; for any voxel v within the lesion i The necrosis candidate response value can be constructed by combining local low signal intensity and neighborhood consistency; preferably, the necrosis cavity response index NCI(v) can be defined. i )for:

[0110] ;

[0111] In the formula, It is the response index of the necrotic cavity. This enhances the modal response weights, preferably ranging from 0.55 to 0.80. It is the normalized enhanced modal response value. These are local gradient weights, preferably ranging from 0.20 to 0.45. It is the normalized result of the local gradient magnitude;

[0112] Furthermore, it can Voxels with values ​​above the necrosis determination threshold are initially marked as necrosis candidate voxels; the necrosis determination threshold is preferably set between 0.50 and 0.75; after performing three-dimensional connected component analysis on the candidate voxels, small discrete regions with fewer than 20 to 80 voxels can be screened out to preserve necrotic cavity structures with spatial coherence.

[0113] Step S33: Boundary continuity analysis, used to characterize the integrity of the lesion boundary structure and the local interruption, specifically, based on the structural hierarchical region mapping data, statistical analysis is performed on the image gradient changes, edge morphology continuity and local structural changes of the lesion edge region, and boundary continuity index is constructed to quantitatively express the integrity of the lesion boundary contour, thereby obtaining boundary continuity feature data;

[0114] In this embodiment, it can be done in the lesion edge area. Extract the boundary voxel set B from the inner boundary voxels and perform joint analysis on the gradient strength, orientation consistency and adjacency connectivity of the boundary voxels;

[0115] Preferably, the boundary continuity index (BCI) can be defined as:

[0116] ;

[0117] In the formula, BCI is the boundary continuity index. This is the average gradient strength weight, preferably between 0.30 and 0.50. It is the average gradient intensity within the boundary voxel set B. This is the boundary direction consistency weight, preferably between 0.20 and 0.35. It is the boundary direction consistency coefficient. This is the boundary connectivity weight, preferably between 0.25 and 0.40. It is the boundary connectivity coefficient;

[0118] Step S34: Peripheral tissue disturbance modeling, used to characterize the structural disturbance and edema spread of brain tissue around the lesion under radiation damage. Specifically, based on the structural hierarchical region mapping data, statistical analysis is performed on the changes in image signals and spatial distribution characteristics in the peripheral tissue region. By identifying the abnormal signal diffusion area and its disturbance degree, the damage to the peripheral brain tissue structure is modeled and described to obtain the peripheral tissue disturbance characteristic data.

[0119] In this embodiment, it is preferable to identify the extent of disturbance in the surrounding tissue based on the high signal diffusion region in the T2-FLAIR image, and to characterize the degree of edema infiltration and structural disturbance by combining its spatial distance relationship with the lesion edge region; for any voxel v in the surrounding tissue region i The Peripheral Organization Disturbance Index (PDI) can be defined. i )for:

[0120] ;

[0121] In the formula, It is the disturbance index of surrounding organizations. This is the weight of the abnormal signal, preferably between 0.55 and 0.80. It is the normalized response value of the abnormal signal from the surrounding tissue. This is the distance attenuation weight, preferably between 0.20 and 0.45. It is the shortest distance from the voxel to the edge of the lesion. The distance attenuation parameter is preferably 2mm to 6mm or 2 to 5 voxels.

[0122] Preferably, it can be Voxels exceeding the perturbation threshold are marked as perturbed voxels, with the perturbation threshold preferably ranging from 0.45 to 0.70. Continuous diffusion regions are extracted using connected component analysis and region growing. For the extracted peripheral abnormal regions, their area, average thickness, maximum diffusion distance, and distribution ratio around the lesion can be further statistically analyzed to form peripheral tissue perturbation characteristic data. Through the above processing, the peripheral tissue perturbation characteristic data is obtained.

[0123] Step S35: Structural topology propagation modeling, used to characterize the structural damage propagation relationship between the internal region of the lesion, the edge region of the lesion, and the surrounding tissue region. Specifically, based on the distribution feature data of the necrotic cavity, the boundary continuity feature data, and the disturbance feature data of the surrounding tissue, a topological association model between the structural levels of the lesion is constructed. By analyzing the spatial transmission relationship between structural abnormalities in different level regions, the structural damage propagation intensity is modeled and expressed to obtain structural level association feature data.

[0124] In this embodiment, the internal area of ​​the lesion can be... lesion margin area and surrounding organizational areas Abstracted as a three-layer structure of node collections A set of topological edges between layers is established, wherein the structural state of each node is characterized by the strength of the necrotic cavity, the level of boundary continuity, and the degree of disturbance of the surrounding tissue. Preferably, the structural state quantities of the three-layer region can be defined as follows:

[0125] ;

[0126] In the formula, It is a quantity representing the structural state of the internal region of the lesion. It is the mean necrosis cavity response index of the internal region of the lesion. It is a quantity representing the structural state of the lesion's edge region. Overall, it refers to the degree of boundary damage. It is a quantity representing the structural state of the surrounding organizational region. It is the average disturbance index of the surrounding organizational area;

[0127] Furthermore, the propagation strength between layers can be constructed, and the calculation formula is as follows:

[0128] ;

[0129] In the formula, This refers to the intensity of inter-level transmission; a and b represent the corresponding internal regions of the lesion. lesion margin area and surrounding organizational areas Hierarchical index, The weights are state-driven, and are preferably set between 0.35 and 0.55. and It is the structural state quantity corresponding to the hierarchical index. This refers to the contact relationship weight, which is preferably set between 0.20 and 0.35. It is the ratio of contact area between regions. The weight is for continuity, and is preferably between 0.20 and 0.35. It is the spatial continuity coefficient between regions;

[0130] Based on this, the overall structural damage propagation index (STI) can be further defined as:

[0131] ;

[0132] In the formula, This refers to the weights of the hierarchical propagation paths corresponding to the internal and peripheral regions of the lesion, preferably ranging from 0.30 to 0.45. It refers to the inter-layer propagation intensity corresponding to the internal region and the peripheral region of the lesion. This refers to the weight of the hierarchical transmission path corresponding to the lesion's edge area and surrounding tissue areas, preferably ranging from 0.35 to 0.50. It refers to the inter-layer transmission intensity corresponding to the lesion edge area and the surrounding tissue area. This refers to the weight of the hierarchical transmission path corresponding to the internal region of the lesion and the surrounding tissue region, preferably ranging from 0.35 to 0.50. It is the inter-level propagation intensity between the internal region of the lesion and the surrounding tissue region, used to describe the cross-boundary coupling effect from the necrotic core to the peripheral abnormal area under severe structural damage conditions;

[0133] More preferably, cases with an STI higher than the significant transmission threshold can be identified as having significant structural transmission characteristics, and the significant transmission threshold is preferably set between 0.50 and 0.75.

[0134] Step S36: Generation of radiation damage structural characterization data, used to comprehensively organize and express the structural damage characteristics of the lesion area. Specifically, it involves summarizing and constructing the distribution characteristic data of the necrotic cavity, the boundary continuity characteristic data, the surrounding tissue disturbance characteristic data, and the structural hierarchy association characteristic data to obtain radiation damage structural characterization data used to characterize the degree of structural damage and structural hierarchy relationship of the lesion area.

[0135] In this embodiment, the necrotic cavity response index, boundary continuity index, surrounding tissue disturbance index, and structural damage propagation index can be organized into a regional-level comprehensive structural characterization vector in a unified format.

[0136] Preferably, the radiation damage structure representation vector of the lesion region can be expressed as:

[0137] ;

[0138] In the formula, S is the structural characterization data of radiation damage;

[0139] The radiation damage structural characterization data specifically includes: necrotic cavity distribution characteristics data, boundary continuity characteristics data, surrounding tissue disturbance characteristics data, and structural hierarchy correlation characteristics data;

[0140] The necrotic cavity distribution feature data is used to characterize the spatial distribution of low-activity necrotic areas within the lesion; the boundary continuity feature data is used to characterize the integrity of the lesion boundary contour and local interruption; the surrounding tissue disturbance feature data is used to characterize the edema and infiltration state and structural damage degree of the brain tissue surrounding the lesion; and the structural hierarchy association feature data is used to characterize the morphological transmission relationship and topological association relationship between the internal area, the edge area, and the surrounding tissue area of ​​the lesion.

[0141] To facilitate the explanation of the engineering implementation of this step, the following is an example of a set of preferred parameter values: [The text abruptly ends here, likely due to an incomplete sentence or missing information.] , The necrosis determination threshold is set at 0.62; the boundary continuity index... , , ; Peripheral organizational disturbance index , Distance attenuation parameter The disturbance determination threshold is set to 0.58; in the structural topology propagation model , , In the overall transmission index , , The threshold for determining significant propagation is set to 0.60;

[0142] It should be understood that the above parameters are only preferred examples. They can be adjusted within the aforementioned preferred range under different image resolutions, lesion volumes, and lesion boundary complexities, but this does not affect the implementation logic of the improved hierarchical damage structure topology analysis method described in this embodiment.

[0143] By performing the above operations, this solution addresses the technical problem in existing radiation injury image recognition methods that struggle to systematically characterize the hierarchical relationships between necrotic structures within lesions, the degree of lesion boundary destruction, and the disturbance of surrounding brain tissue, resulting in incomplete representation of radiation injury structural features. This solution creatively employs a hierarchical damage structure topology analysis mechanism. It establishes a hierarchical regional mapping relationship between the lesion's internal, peripheral, and surrounding tissue regions, and jointly models the distribution of necrotic cavities, boundary continuity, and surrounding tissue disturbance. Furthermore, it uses structural topology propagation analysis to characterize the spatial transmission relationship of structural damage between different hierarchical regions, thereby forming radiation injury structural characterization data. For example, in radiation injury cases, common lesions often exhibit significant internal necrotic cavities, disrupted boundary structures, and diffuse edema in surrounding tissues. This solution, through hierarchical topology modeling, can comprehensively describe this type of structural damage pattern.

[0144] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the joint constraint image identification is used to identify and determine whether the postoperative lesion is recurrent or caused by radiation damage. Specifically, the microvascular metabolic characterization data and the radiation damage structural characterization data are mapped to a unified discrimination space. The functional activity state and structural damage state of the lesion area are correlated and analyzed. By constructing functional activity consistency constraints and structural damage consistency constraints, the candidate discrimination results are jointly screened, and the discrimination results that do not meet the cross-mechanism consistency conditions are suppressed. The discrimination result that the lesion belongs to the postoperative recurrence or radiation damage of glioma is output, and the joint identification result data is obtained.

[0145] This embodiment uses joint analysis of microvascular metabolic characterization data and radiation damage structural characterization data to differentiate between postoperative lesions and radiation damage, thereby obtaining stable lesion classification results.

[0146] In this embodiment, the microvascular metabolic characterization data output in step S2 and the radiation damage structural characterization data output in step S3 are first obtained, and the two types of feature data are processed in a unified format. Specifically, the local perfusion response feature data, diffusion restriction feature data, enhancement feature data and microvascular metabolic heterogeneity index data in the microvascular metabolic characterization data are normalized with the necrotic cavity distribution feature data, boundary continuity feature data, surrounding tissue disturbance feature data and structural hierarchy association feature data in the radiation damage structural characterization data, and then combined according to a unified feature vector format to form a comprehensive discriminative feature vector of the lesion area.

[0147] Furthermore, the comprehensive discriminative feature vector preferably includes two parts: a functional activity feature sub-vector and a structural damage feature sub-vector. The functional activity feature sub-vector can be composed of the mean local perfusion response index, the mean diffusion restriction index, the mean enhancement feature, and microvascular metabolic heterogeneity indicators. The structural damage feature sub-vector can be composed of necrosis cavity degree indicators, boundary continuity indicators, surrounding tissue disturbance indicators, and structural damage propagation indicators. In some embodiments, the dimension of the comprehensive discriminative feature vector can be set to 8 to 16 dimensions, preferably 8 or 12 dimensions.

[0148] In some preferred embodiments, the unified feature vector format can be organized in the form of a fixed-length array, a structured field table, or a model input tensor. When a lightweight classification model is used for auxiliary discrimination, the comprehensive discrimination feature vector can be further organized into a one-dimensional feature array and input into a logistic regression model, a support vector machine model, or a multilayer perceptron model. The multilayer perceptron model preferably adopts a 2-4 layer fully connected layer structure, and the number of neurons in each layer is preferably 16-64.

[0149] After constructing the feature vectors, the functional activity features and structural damage features are jointly discriminated. Specifically, firstly, the functional activity score of the lesion is calculated based on local perfusion response features, diffusion restriction features, and enhancement features, which characterizes the microvascular perfusion activity and metabolic activity status of the lesion area; simultaneously, the structural damage score of the lesion is calculated based on the distribution features of necrotic cavities, boundary continuity features, and surrounding tissue disturbance features, which characterizes the degree of structural damage and damage diffusion status of the lesion area; by simultaneously calculating the functional activity score and the structural damage score, a two-dimensional discriminant index reflecting the functional and structural status of the lesion area can be formed.

[0150] In some implementations, the functional activity score can be constructed using a weighted summation method, that is, by combining the relative contributions of local perfusion response characteristics, diffusion restriction characteristics, and enhancement characteristics.

[0151] Furthermore, the structural damage score can be obtained by using a similar weighted integration method, which comprehensively characterizes the radiation damage manifestation of the lesion based on the distribution characteristics of necrotic cavities, boundary continuity characteristics, surrounding tissue disturbance characteristics, and structural hierarchical correlation characteristics.

[0152] In some implementations, to avoid the excessive influence of fluctuations in a single indicator on the final score, each feature can be truncated or smoothed before scoring; preferably, feature values ​​that exceed the mean ± 2 standard deviations can be truncated, or local abnormal fluctuations can be weakened by moving average, exponential smoothing, or other methods.

[0153] In some implementations, the functional activity score and structural damage score can be obtained by explicit rule calculation or by outputting the corresponding category score from an existing classification model. When using a logistic regression model, functional activity-related features can be used as one set of inputs to output a functional activity tendency score, and structural damage-related features can be used as another set of inputs to output a structural damage tendency score. When using a support vector machine model, a radial basis function kernel function is preferred, with the penalty parameter preferably set to 0.5–5.0 and the kernel function parameter preferably set to 0.01–0.20.

[0154] Furthermore, functional activity consistency constraints and structural damage consistency constraints are introduced in the candidate discrimination process. The functional activity consistency constraint is used to determine whether the perfusion level, diffusion restriction degree, and enhancement response of the lesion area exhibit functional activity characteristics consistent with tumor recurrence. The structural damage consistency constraint is used to determine whether the lesion area has obvious necrotic cavities, boundary structure destruction, or surrounding tissue disturbance, etc., structural features consistent with radiation damage. When the functional activity score is significantly higher than the structural damage score, the candidate result is tended to be judged as postoperative recurrence; when the structural damage score is significantly higher than the functional activity score, the candidate result is tended to be judged as radiation damage. When the two scores are close, microvascular metabolic heterogeneity indicators and structural hierarchical correlation characteristics are combined for further auxiliary judgment to reduce misjudgment in borderline cases.

[0155] After obtaining the final discrimination result, the credibility of the discrimination result is evaluated; specifically, the category confidence is calculated based on the degree of difference between the functional activity score and the structural damage score, and the current discrimination result is recorded as mainly derived from the functional activity feature or the structural damage feature, thereby forming the discrimination basis identification information;

[0156] Through the above processing, this embodiment finally obtains the joint identification result data; the joint identification result data specifically includes lesion category determination data, category confidence data, joint constraint consistency assessment data, and discrimination criterion identification data; wherein, the lesion category determination data is used to characterize the determination result that the target lesion belongs to postoperative recurrence of glioma or radiation injury; the category confidence data is used to characterize the credibility of the corresponding determination result; the joint constraint consistency assessment data is used to characterize the degree of support between microvascular metabolic characteristics and radiation injury structural characteristics for the current determination result; the discrimination criterion identification data is used to characterize whether the source of the characteristic that plays a dominant role in this identification is abnormal functional activity or abnormal structural damage.

[0157] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0158] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0159] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for identifying postoperative recurrence and radiation injury in gliomas using images, characterized in that: The method includes the following steps: Step S1: Standardize the multimodal image data to obtain standardized multimodal image data; Step S2: Microvascular metabolic feature reconstruction. Based on the standardized multimodal image data, the signal responses of each voxel position within the lesion candidate region in different modal images are indexed and mapped. The lesion is divided into an internal lesion region and a lesion edge region according to the spatial structure of the lesion region, forming a voxel-level correspondence between different modal images. Statistical analysis is performed on the perfusion modal response values ​​of each voxel neighborhood within the lesion region to construct a local perfusion response index, obtaining local perfusion response feature data. Statistical analysis is also performed on the diffusion modal response values ​​of each voxel neighborhood within the lesion region to construct a diffusion restriction index, obtaining diffusion restriction feature data. A feature deviation index is calculated based on the enhancement modal response values, the local perfusion response feature data, and the diffusion restriction feature data to obtain enhancement feature data. Regional statistical analysis is performed on the perfusion response features, diffusion restriction features, and enhancement features of the internal lesion region and the lesion edge region, respectively. A metabolic heterogeneity index is constructed based on the degree of difference between the internal lesion region and the lesion edge region, obtaining microvascular metabolic characterization data. Step S3: Radiation damage structure analysis. Based on the standardized multimodal image data, spatial morphology analysis is performed on the candidate lesion regions. According to the spatial structural relationship of the lesion regions, the lesions are divided into internal lesion regions, edge lesion regions, and surrounding tissue regions. The image signal intensity of voxels in the internal lesion regions and their neighborhood changes are analyzed to identify areas of abnormal signal reduction and their connectivity structures, obtaining necrotic cavity distribution characteristic data. The image gradient changes, edge morphological continuity, and local structural changes in the edge lesion regions are statistically analyzed to construct boundary continuity indicators, obtaining boundary continuity characteristic data. The image signal changes and spatial distribution characteristics in the surrounding tissue regions are statistically analyzed to identify abnormal signal diffusion areas and their disturbance degree, obtaining surrounding tissue disturbance characteristic data. Based on the necrotic cavity distribution characteristic data, the boundary continuity characteristic data, and the surrounding tissue disturbance characteristic data, the spatial transmission relationship of structural anomalies between different hierarchical regions is analyzed to obtain structural hierarchy correlation characteristic data, and the radiation damage structure characterization data is summarized. Step S4: Joint constraint image identification. The microvascular metabolic characterization data and the radiation damage structural characterization data are used as joint discrimination inputs. The functional activity state and structural damage state of the lesion area are correlated and analyzed to construct functional activity consistency constraints and structural damage consistency constraints. The candidate discrimination results are jointly screened, and the discrimination results that do not meet the cross-mechanism consistency conditions are suppressed. The discrimination result that the lesion belongs to the postoperative recurrence of glioma or radiation damage is output to obtain the joint identification result data. The functional activity consistency constraint is used to determine whether the perfusion level, diffusion restriction degree and enhancement response of the lesion area show functional activity characteristics consistent with tumor recurrence; the structural damage consistency constraint is used to determine whether there are structural features consistent with radiation damage, such as necrotic cavities, boundary structure destruction or surrounding tissue disturbance, in the lesion area.

2. The method for identifying postoperative recurrence and radiation injury of glioma according to claim 1, characterized in that: In step S1, the multimodal image data is standardized, and multimodal medical image data during the postoperative follow-up stage is obtained. Spatial registration processing, voxel scale unification processing, gray intensity normalization processing, and intracranial structural region limitation processing are performed on the multimodal medical image data. Based on the anatomical structure of the surgical area, the candidate lesion region is located and cropped to obtain standardized multimodal image data.

3. The method for identifying postoperative recurrence and radiation injury of glioma according to claim 2, characterized in that: In step S2, the microvascular metabolic characterization data specifically includes: local perfusion response characteristic data, diffusion restriction characteristic data, enhancement characteristic data, and microvascular metabolic heterogeneity index data; the local perfusion response characteristic data is used to characterize the microvascular blood flow perfusion level and its spatial distribution differences within the lesion area; the diffusion restriction characteristic data is used to characterize the abnormal diffusion state caused by changes in lesion tissue cell density; the enhancement characteristic data is used to characterize the combined change relationship between the enhancement response and perfusion and diffusion characteristics; and the microvascular metabolic heterogeneity index data is used to characterize the degree of difference in blood supply and metabolic activity in different regions within the lesion.

4. The method for identifying postoperative recurrence and radiation injury of glioma according to claim 3, characterized in that: The radiation damage structural characterization data specifically includes: necrotic cavity distribution characteristics data, boundary continuity characteristics data, surrounding tissue disturbance characteristics data, and structural hierarchy correlation characteristics data; The necrotic cavity distribution feature data is used to characterize the spatial distribution of low-activity necrotic areas within the lesion; the boundary continuity feature data is used to characterize the integrity of the lesion boundary contour and local interruption; the surrounding tissue disturbance feature data is used to characterize the edema and infiltration state and structural damage degree of the brain tissue surrounding the lesion; and the structural hierarchy association feature data is used to characterize the morphological transmission relationship and topological association relationship between the internal area, the edge area, and the surrounding tissue area of ​​the lesion.

Images